blob: 48791d4fa8620f5ef491fac9ba5ba011bc229406 [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Johannes Weiner1575e682015-04-14 15:44:51 -070017 * Native page reclaim
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
22 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080023 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
27 *
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
32 */
33
Johannes Weiner3e32cb22014-12-10 15:42:31 -080034#include <linux/page_counter.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080035#include <linux/memcontrol.h>
36#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080037#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080038#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080039#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080040#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080041#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080043#include <linux/bit_spinlock.h>
44#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070045#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040046#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080047#include <linux/mutex.h>
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -070048#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070049#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080051#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080052#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080053#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050054#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080055#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080056#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080057#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070058#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070059#include <linux/mm_inline.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080060#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080061#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070062#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070063#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050064#include <linux/file.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080065#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000066#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070067#include <net/ip.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +000068#include <net/tcp_memcontrol.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080069#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080070
Balbir Singh8697d332008-02-07 00:13:59 -080071#include <asm/uaccess.h>
72
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070073#include <trace/events/vmscan.h>
74
Tejun Heo073219e2014-02-08 10:36:58 -050075struct cgroup_subsys memory_cgrp_subsys __read_mostly;
76EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080077
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070078#define MEM_CGROUP_RECLAIM_RETRIES 5
Kirill A. Shutemov6bbda352012-05-29 15:06:55 -070079static struct mem_cgroup *root_mem_cgroup __read_mostly;
Tejun Heodbee2272015-05-22 17:13:20 -040080struct cgroup_subsys_state *mem_cgroup_root_css __read_mostly;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080081
Johannes Weiner21afa382015-02-11 15:26:36 -080082/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070083#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080084int do_swap_account __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080085#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070086#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080087#endif
88
Johannes Weineraf7c4b02012-05-29 15:07:08 -070089static const char * const mem_cgroup_stat_names[] = {
90 "cache",
91 "rss",
David Rientjesb070e652013-05-07 16:18:09 -070092 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070093 "mapped_file",
Greg Thelen7c9d3ff2015-05-22 17:13:16 -040094 "dirty",
Sha Zhengju3ea67d02013-09-12 15:13:53 -070095 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -070096 "swap",
97};
98
Johannes Weineraf7c4b02012-05-29 15:07:08 -070099static const char * const mem_cgroup_events_names[] = {
100 "pgpgin",
101 "pgpgout",
102 "pgfault",
103 "pgmajfault",
104};
105
Sha Zhengju58cf1882013-02-22 16:32:05 -0800106static const char * const mem_cgroup_lru_names[] = {
107 "inactive_anon",
108 "active_anon",
109 "inactive_file",
110 "active_file",
111 "unevictable",
112};
113
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700114#define THRESHOLDS_EVENTS_TARGET 128
115#define SOFTLIMIT_EVENTS_TARGET 1024
116#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700117
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700118/*
119 * Cgroups above their limits are maintained in a RB-Tree, independent of
120 * their hierarchy representation
121 */
122
123struct mem_cgroup_tree_per_zone {
124 struct rb_root rb_root;
125 spinlock_t lock;
126};
127
128struct mem_cgroup_tree_per_node {
129 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
130};
131
132struct mem_cgroup_tree {
133 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
134};
135
136static struct mem_cgroup_tree soft_limit_tree __read_mostly;
137
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700138/* for OOM */
139struct mem_cgroup_eventfd_list {
140 struct list_head list;
141 struct eventfd_ctx *eventfd;
142};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800143
Tejun Heo79bd9812013-11-22 18:20:42 -0500144/*
145 * cgroup_event represents events which userspace want to receive.
146 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500147struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500148 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500149 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500150 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500151 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500152 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500153 * eventfd to signal userspace about the event.
154 */
155 struct eventfd_ctx *eventfd;
156 /*
157 * Each of these stored in a list by the cgroup.
158 */
159 struct list_head list;
160 /*
Tejun Heofba94802013-11-22 18:20:43 -0500161 * register_event() callback will be used to add new userspace
162 * waiter for changes related to this event. Use eventfd_signal()
163 * on eventfd to send notification to userspace.
164 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500165 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500166 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500167 /*
168 * unregister_event() callback will be called when userspace closes
169 * the eventfd or on cgroup removing. This callback must be set,
170 * if you want provide notification functionality.
171 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500172 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500173 struct eventfd_ctx *eventfd);
174 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500175 * All fields below needed to unregister event when
176 * userspace closes eventfd.
177 */
178 poll_table pt;
179 wait_queue_head_t *wqh;
180 wait_queue_t wait;
181 struct work_struct remove;
182};
183
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700184static void mem_cgroup_threshold(struct mem_cgroup *memcg);
185static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800186
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800187/* Stuffs for move charges at task migration. */
188/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800189 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800190 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800191#define MOVE_ANON 0x1U
192#define MOVE_FILE 0x2U
193#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800194
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800195/* "mc" and its members are protected by cgroup_mutex */
196static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800197 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800198 struct mem_cgroup *from;
199 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800200 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800201 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800202 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800203 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800204 struct task_struct *moving_task; /* a task moving charges */
205 wait_queue_head_t waitq; /* a waitq for other context */
206} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700207 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800208 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
209};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800210
Balbir Singh4e416952009-09-23 15:56:39 -0700211/*
212 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
213 * limit reclaim to prevent infinite loops, if they ever occur.
214 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700215#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700216#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700217
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800218enum charge_type {
219 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700220 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800221 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700222 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700223 NR_CHARGE_TYPE,
224};
225
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800226/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800227enum res_type {
228 _MEM,
229 _MEMSWAP,
230 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800231 _KMEM,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800232};
233
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700234#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
235#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800236#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700237/* Used for OOM nofiier */
238#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800239
Balbir Singh75822b42009-09-23 15:56:38 -0700240/*
Glauber Costa09998212013-02-22 16:34:55 -0800241 * The memcg_create_mutex will be held whenever a new cgroup is created.
242 * As a consequence, any change that needs to protect against new child cgroups
243 * appearing has to hold it as well.
244 */
245static DEFINE_MUTEX(memcg_create_mutex);
246
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700247/* Some nice accessors for the vmpressure. */
248struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
249{
250 if (!memcg)
251 memcg = root_mem_cgroup;
252 return &memcg->vmpressure;
253}
254
255struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
256{
257 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
258}
259
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700260static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
261{
262 return (memcg == root_mem_cgroup);
263}
264
Li Zefan4219b2d2013-09-23 16:56:29 +0800265/*
266 * We restrict the id in the range of [1, 65535], so it can fit into
267 * an unsigned short.
268 */
269#define MEM_CGROUP_ID_MAX USHRT_MAX
270
Li Zefan34c00c32013-09-23 16:56:01 +0800271static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
272{
Tejun Heo15a4c832014-05-04 15:09:14 -0400273 return memcg->css.id;
Li Zefan34c00c32013-09-23 16:56:01 +0800274}
275
Vladimir Davydovadbe4272015-04-15 16:13:00 -0700276/*
277 * A helper function to get mem_cgroup from ID. must be called under
278 * rcu_read_lock(). The caller is responsible for calling
279 * css_tryget_online() if the mem_cgroup is used for charging. (dropping
280 * refcnt from swap can be called against removed memcg.)
281 */
Li Zefan34c00c32013-09-23 16:56:01 +0800282static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
283{
284 struct cgroup_subsys_state *css;
285
Tejun Heo7d699dd2014-05-04 15:09:13 -0400286 css = css_from_id(id, &memory_cgrp_subsys);
Li Zefan34c00c32013-09-23 16:56:01 +0800287 return mem_cgroup_from_css(css);
288}
289
Glauber Costae1aab162011-12-11 21:47:03 +0000290/* Writing them here to avoid exposing memcg's inner layout */
Michal Hocko4bd2c1e2012-10-08 16:33:10 -0700291#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
Glauber Costae1aab162011-12-11 21:47:03 +0000292
Glauber Costae1aab162011-12-11 21:47:03 +0000293void sock_update_memcg(struct sock *sk)
294{
Glauber Costa376be5f2012-01-20 04:57:14 +0000295 if (mem_cgroup_sockets_enabled) {
Glauber Costae1aab162011-12-11 21:47:03 +0000296 struct mem_cgroup *memcg;
Glauber Costa3f134612012-05-29 15:07:11 -0700297 struct cg_proto *cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000298
299 BUG_ON(!sk->sk_prot->proto_cgroup);
300
Glauber Costaf3f511e2012-01-05 20:16:39 +0000301 /* Socket cloning can throw us here with sk_cgrp already
302 * filled. It won't however, necessarily happen from
303 * process context. So the test for root memcg given
304 * the current task's memcg won't help us in this case.
305 *
306 * Respecting the original socket's memcg is a better
307 * decision in this case.
308 */
309 if (sk->sk_cgrp) {
310 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
Li Zefan5347e5a2013-07-08 16:00:30 -0700311 css_get(&sk->sk_cgrp->memcg->css);
Glauber Costaf3f511e2012-01-05 20:16:39 +0000312 return;
313 }
314
Glauber Costae1aab162011-12-11 21:47:03 +0000315 rcu_read_lock();
316 memcg = mem_cgroup_from_task(current);
Glauber Costa3f134612012-05-29 15:07:11 -0700317 cg_proto = sk->sk_prot->proto_cgroup(memcg);
Michal Hocko6b34b852015-09-08 15:01:16 -0700318 if (cg_proto && test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags) &&
Tejun Heoec903c02014-05-13 12:11:01 -0400319 css_tryget_online(&memcg->css)) {
Glauber Costa3f134612012-05-29 15:07:11 -0700320 sk->sk_cgrp = cg_proto;
Glauber Costae1aab162011-12-11 21:47:03 +0000321 }
322 rcu_read_unlock();
323 }
324}
325EXPORT_SYMBOL(sock_update_memcg);
326
327void sock_release_memcg(struct sock *sk)
328{
Glauber Costa376be5f2012-01-20 04:57:14 +0000329 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
Glauber Costae1aab162011-12-11 21:47:03 +0000330 struct mem_cgroup *memcg;
331 WARN_ON(!sk->sk_cgrp->memcg);
332 memcg = sk->sk_cgrp->memcg;
Li Zefan5347e5a2013-07-08 16:00:30 -0700333 css_put(&sk->sk_cgrp->memcg->css);
Glauber Costae1aab162011-12-11 21:47:03 +0000334 }
335}
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000336
337struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
338{
339 if (!memcg || mem_cgroup_is_root(memcg))
340 return NULL;
341
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700342 return &memcg->tcp_mem;
Glauber Costad1a4c0b2011-12-11 21:47:04 +0000343}
344EXPORT_SYMBOL(tcp_proto_cgroup);
Glauber Costae1aab162011-12-11 21:47:03 +0000345
Glauber Costa3f134612012-05-29 15:07:11 -0700346#endif
347
Glauber Costaa8964b92012-12-18 14:22:09 -0800348#ifdef CONFIG_MEMCG_KMEM
Glauber Costa55007d82012-12-18 14:22:38 -0800349/*
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800350 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800351 * The main reason for not using cgroup id for this:
352 * this works better in sparse environments, where we have a lot of memcgs,
353 * but only a few kmem-limited. Or also, if we have, for instance, 200
354 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
355 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800356 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800357 * The current size of the caches array is stored in memcg_nr_cache_ids. It
358 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800359 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800360static DEFINE_IDA(memcg_cache_ida);
361int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800362
Vladimir Davydov05257a12015-02-12 14:59:01 -0800363/* Protects memcg_nr_cache_ids */
364static DECLARE_RWSEM(memcg_cache_ids_sem);
365
366void memcg_get_cache_ids(void)
367{
368 down_read(&memcg_cache_ids_sem);
369}
370
371void memcg_put_cache_ids(void)
372{
373 up_read(&memcg_cache_ids_sem);
374}
375
Glauber Costa55007d82012-12-18 14:22:38 -0800376/*
377 * MIN_SIZE is different than 1, because we would like to avoid going through
378 * the alloc/free process all the time. In a small machine, 4 kmem-limited
379 * cgroups is a reasonable guess. In the future, it could be a parameter or
380 * tunable, but that is strictly not necessary.
381 *
Li Zefanb8627832013-09-23 16:56:47 +0800382 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800383 * this constant directly from cgroup, but it is understandable that this is
384 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800385 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800386 * increase ours as well if it increases.
387 */
388#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800389#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800390
Glauber Costad7f25f82012-12-18 14:22:40 -0800391/*
392 * A lot of the calls to the cache allocation functions are expected to be
393 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
394 * conditional to this static branch, we'll have to allow modules that does
395 * kmem_cache_alloc and the such to see this symbol as well
396 */
Glauber Costaa8964b92012-12-18 14:22:09 -0800397struct static_key memcg_kmem_enabled_key;
Glauber Costad7f25f82012-12-18 14:22:40 -0800398EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800399
Glauber Costaa8964b92012-12-18 14:22:09 -0800400#endif /* CONFIG_MEMCG_KMEM */
401
Balbir Singhf64c3f52009-09-23 15:56:37 -0700402static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700403mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700404{
Jianyu Zhane2318752014-06-06 14:38:20 -0700405 int nid = zone_to_nid(zone);
406 int zid = zone_idx(zone);
407
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700408 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700409}
410
Tejun Heoe4487942015-05-27 20:00:02 -0400411/**
412 * mem_cgroup_css_from_page - css of the memcg associated with a page
413 * @page: page of interest
414 *
415 * If memcg is bound to the default hierarchy, css of the memcg associated
416 * with @page is returned. The returned css remains associated with @page
417 * until it is released.
418 *
419 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
420 * is returned.
421 *
422 * XXX: The above description of behavior on the default hierarchy isn't
423 * strictly true yet as replace_page_cache_page() can modify the
424 * association before @page is released even on the default hierarchy;
425 * however, the current and planned usages don't mix the the two functions
426 * and replace_page_cache_page() will soon be updated to make the invariant
427 * actually true.
428 */
429struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
430{
431 struct mem_cgroup *memcg;
432
433 rcu_read_lock();
434
435 memcg = page->mem_cgroup;
436
Tejun Heoded8e212015-09-18 11:56:28 -0400437 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoe4487942015-05-27 20:00:02 -0400438 memcg = root_mem_cgroup;
439
440 rcu_read_unlock();
441 return &memcg->css;
442}
443
Balbir Singhf64c3f52009-09-23 15:56:37 -0700444static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700445mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700446{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700447 int nid = page_to_nid(page);
448 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700449
Jianyu Zhane2318752014-06-06 14:38:20 -0700450 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700451}
452
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700453static struct mem_cgroup_tree_per_zone *
454soft_limit_tree_node_zone(int nid, int zid)
455{
456 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
457}
458
459static struct mem_cgroup_tree_per_zone *
460soft_limit_tree_from_page(struct page *page)
461{
462 int nid = page_to_nid(page);
463 int zid = page_zonenum(page);
464
465 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
466}
467
Johannes Weinercf2c8122014-06-06 14:38:21 -0700468static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz,
469 struct mem_cgroup_tree_per_zone *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800470 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700471{
472 struct rb_node **p = &mctz->rb_root.rb_node;
473 struct rb_node *parent = NULL;
474 struct mem_cgroup_per_zone *mz_node;
475
476 if (mz->on_tree)
477 return;
478
479 mz->usage_in_excess = new_usage_in_excess;
480 if (!mz->usage_in_excess)
481 return;
482 while (*p) {
483 parent = *p;
484 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
485 tree_node);
486 if (mz->usage_in_excess < mz_node->usage_in_excess)
487 p = &(*p)->rb_left;
488 /*
489 * We can't avoid mem cgroups that are over their soft
490 * limit by the same amount
491 */
492 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
493 p = &(*p)->rb_right;
494 }
495 rb_link_node(&mz->tree_node, parent, p);
496 rb_insert_color(&mz->tree_node, &mctz->rb_root);
497 mz->on_tree = true;
498}
499
Johannes Weinercf2c8122014-06-06 14:38:21 -0700500static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
501 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700502{
503 if (!mz->on_tree)
504 return;
505 rb_erase(&mz->tree_node, &mctz->rb_root);
506 mz->on_tree = false;
507}
508
Johannes Weinercf2c8122014-06-06 14:38:21 -0700509static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
510 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700511{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700512 unsigned long flags;
513
514 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700515 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700516 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700517}
518
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800519static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
520{
521 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700522 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800523 unsigned long excess = 0;
524
525 if (nr_pages > soft_limit)
526 excess = nr_pages - soft_limit;
527
528 return excess;
529}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700530
531static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
532{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800533 unsigned long excess;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700534 struct mem_cgroup_per_zone *mz;
535 struct mem_cgroup_tree_per_zone *mctz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700536
Jianyu Zhane2318752014-06-06 14:38:20 -0700537 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700538 /*
539 * Necessary to update all ancestors when hierarchy is used.
540 * because their event counter is not touched.
541 */
542 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Jianyu Zhane2318752014-06-06 14:38:20 -0700543 mz = mem_cgroup_page_zoneinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800544 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700545 /*
546 * We have to update the tree if mz is on RB-tree or
547 * mem is over its softlimit.
548 */
549 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700550 unsigned long flags;
551
552 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700553 /* if on-tree, remove it */
554 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700555 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700556 /*
557 * Insert again. mz->usage_in_excess will be updated.
558 * If excess is 0, no tree ops.
559 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700560 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700561 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700562 }
563 }
564}
565
566static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
567{
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700568 struct mem_cgroup_tree_per_zone *mctz;
Jianyu Zhane2318752014-06-06 14:38:20 -0700569 struct mem_cgroup_per_zone *mz;
570 int nid, zid;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700571
Jianyu Zhane2318752014-06-06 14:38:20 -0700572 for_each_node(nid) {
573 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
574 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
575 mctz = soft_limit_tree_node_zone(nid, zid);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700576 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700577 }
578 }
579}
580
581static struct mem_cgroup_per_zone *
582__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
583{
584 struct rb_node *rightmost = NULL;
585 struct mem_cgroup_per_zone *mz;
586
587retry:
588 mz = NULL;
589 rightmost = rb_last(&mctz->rb_root);
590 if (!rightmost)
591 goto done; /* Nothing to reclaim from */
592
593 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
594 /*
595 * Remove the node now but someone else can add it back,
596 * we will to add it back at the end of reclaim to its correct
597 * position in the tree.
598 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700599 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800600 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400601 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700602 goto retry;
603done:
604 return mz;
605}
606
607static struct mem_cgroup_per_zone *
608mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
609{
610 struct mem_cgroup_per_zone *mz;
611
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700612 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700613 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700614 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700615 return mz;
616}
617
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700618/*
619 * Implementation Note: reading percpu statistics for memcg.
620 *
621 * Both of vmstat[] and percpu_counter has threshold and do periodic
622 * synchronization to implement "quick" read. There are trade-off between
623 * reading cost and precision of value. Then, we may have a chance to implement
624 * a periodic synchronizion of counter in memcg's counter.
625 *
626 * But this _read() function is used for user interface now. The user accounts
627 * memory usage by memory cgroup and he _always_ requires exact value because
628 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
629 * have to visit all online cpus and make sum. So, for now, unnecessary
630 * synchronization is not implemented. (just implemented for cpu hotplug)
631 *
632 * If there are kernel internal actions which can make use of some not-exact
633 * value, and reading all cpu value can be performance bottleneck in some
634 * common workload, threashold and synchonization as vmstat[] should be
635 * implemented.
636 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700637static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700638 enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800639{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700640 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800641 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800642
Tejun Heo11aa0bf2015-05-22 18:23:18 -0400643 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700644 val += per_cpu(memcg->stat->count[idx], cpu);
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800645 return val;
646}
647
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700648static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700649 enum mem_cgroup_events_index idx)
650{
651 unsigned long val = 0;
652 int cpu;
653
Tejun Heo11aa0bf2015-05-22 18:23:18 -0400654 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700655 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700656 return val;
657}
658
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700659static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700660 struct page *page,
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700661 int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800662{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700663 /*
664 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
665 * counted as CACHE even if it's on ANON LRU.
666 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700667 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700668 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700669 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800670 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700671 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700672 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700673
David Rientjesb070e652013-05-07 16:18:09 -0700674 if (PageTransHuge(page))
675 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
676 nr_pages);
677
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800678 /* pagein of a big page is an event. So, ignore page size */
679 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700680 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800681 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700682 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800683 nr_pages = -nr_pages; /* for event */
684 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800685
Johannes Weiner13114712012-05-29 15:07:07 -0700686 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800687}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800688
Jianyu Zhane2318752014-06-06 14:38:20 -0700689static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
690 int nid,
691 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700692{
Jianyu Zhane2318752014-06-06 14:38:20 -0700693 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700694 int zid;
695
Jianyu Zhane2318752014-06-06 14:38:20 -0700696 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700697
Jianyu Zhane2318752014-06-06 14:38:20 -0700698 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
699 struct mem_cgroup_per_zone *mz;
700 enum lru_list lru;
701
702 for_each_lru(lru) {
703 if (!(BIT(lru) & lru_mask))
704 continue;
705 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
706 nr += mz->lru_size[lru];
707 }
708 }
709 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700710}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700711
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700712static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700713 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800714{
Jianyu Zhane2318752014-06-06 14:38:20 -0700715 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700716 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800717
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800718 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700719 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
720 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800721}
722
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800723static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
724 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800725{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700726 unsigned long val, next;
727
Johannes Weiner13114712012-05-29 15:07:07 -0700728 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700729 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700730 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800731 if ((long)next - (long)val < 0) {
732 switch (target) {
733 case MEM_CGROUP_TARGET_THRESH:
734 next = val + THRESHOLDS_EVENTS_TARGET;
735 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700736 case MEM_CGROUP_TARGET_SOFTLIMIT:
737 next = val + SOFTLIMIT_EVENTS_TARGET;
738 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800739 case MEM_CGROUP_TARGET_NUMAINFO:
740 next = val + NUMAINFO_EVENTS_TARGET;
741 break;
742 default:
743 break;
744 }
745 __this_cpu_write(memcg->stat->targets[target], next);
746 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700747 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800748 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800749}
750
751/*
752 * Check events in order.
753 *
754 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700755static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800756{
757 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800758 if (unlikely(mem_cgroup_event_ratelimit(memcg,
759 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700760 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800761 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800762
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700763 do_softlimit = mem_cgroup_event_ratelimit(memcg,
764 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700765#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800766 do_numainfo = mem_cgroup_event_ratelimit(memcg,
767 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700768#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800769 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700770 if (unlikely(do_softlimit))
771 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800772#if MAX_NUMNODES > 1
773 if (unlikely(do_numainfo))
774 atomic_inc(&memcg->numainfo_events);
775#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700776 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800777}
778
Balbir Singhcf475ad2008-04-29 01:00:16 -0700779struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800780{
Balbir Singh31a78f22008-09-28 23:09:31 +0100781 /*
782 * mm_update_next_owner() may clear mm->owner to NULL
783 * if it races with swapoff, page migration, etc.
784 * So this can be called with p == NULL.
785 */
786 if (unlikely(!p))
787 return NULL;
788
Tejun Heo073219e2014-02-08 10:36:58 -0500789 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800790}
Michal Hocko1e977cc2015-09-08 15:01:02 -0700791EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800792
Johannes Weinerdf381972014-04-07 15:37:43 -0700793static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800794{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700795 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700796
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800797 rcu_read_lock();
798 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700799 /*
800 * Page cache insertions can happen withou an
801 * actual mm context, e.g. during disk probing
802 * on boot, loopback IO, acct() writes etc.
803 */
804 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700805 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700806 else {
807 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
808 if (unlikely(!memcg))
809 memcg = root_mem_cgroup;
810 }
Tejun Heoec903c02014-05-13 12:11:01 -0400811 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800812 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700813 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800814}
815
Johannes Weiner56600482012-01-12 17:17:59 -0800816/**
817 * mem_cgroup_iter - iterate over memory cgroup hierarchy
818 * @root: hierarchy root
819 * @prev: previously returned memcg, NULL on first invocation
820 * @reclaim: cookie for shared reclaim walks, NULL for full walks
821 *
822 * Returns references to children of the hierarchy below @root, or
823 * @root itself, or %NULL after a full round-trip.
824 *
825 * Caller must pass the return value in @prev on subsequent
826 * invocations for reference counting, or use mem_cgroup_iter_break()
827 * to cancel a hierarchy walk before the round-trip is complete.
828 *
829 * Reclaimers can specify a zone and a priority level in @reclaim to
830 * divide up the memcgs in the hierarchy among all concurrent
831 * reclaimers operating on the same zone and priority.
832 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700833struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800834 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700835 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700836{
Michal Hocko1e977cc2015-09-08 15:01:02 -0700837 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800838 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800839 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800840 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700841
Andrew Morton694fbc02013-09-24 15:27:37 -0700842 if (mem_cgroup_disabled())
843 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800844
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700845 if (!root)
846 root = root_mem_cgroup;
847
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800848 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800849 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800850
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800851 if (!root->use_hierarchy && root != root_mem_cgroup) {
852 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800853 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700854 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800855 }
856
Michal Hocko542f85f2013-04-29 15:07:15 -0700857 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800858
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800859 if (reclaim) {
860 struct mem_cgroup_per_zone *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800861
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800862 mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone);
863 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700864
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800865 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700866 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800867
868 do {
Jason Low4db0c3c2015-04-15 16:14:08 -0700869 pos = READ_ONCE(iter->position);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800870 /*
871 * A racing update may change the position and
872 * put the last reference, hence css_tryget(),
873 * or retry to see the updated position.
874 */
875 } while (pos && !css_tryget(&pos->css));
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800876 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800877
878 if (pos)
879 css = &pos->css;
880
881 for (;;) {
882 css = css_next_descendant_pre(css, &root->css);
883 if (!css) {
884 /*
885 * Reclaimers share the hierarchy walk, and a
886 * new one might jump in right at the end of
887 * the hierarchy - make sure they see at least
888 * one group and restart from the beginning.
889 */
890 if (!prev)
891 continue;
892 break;
893 }
894
895 /*
896 * Verify the css and acquire a reference. The root
897 * is provided by the caller, so we know it's alive
898 * and kicking, and don't take an extra reference.
899 */
900 memcg = mem_cgroup_from_css(css);
901
902 if (css == &root->css)
903 break;
904
Johannes Weinerb2052562014-12-10 15:42:48 -0800905 if (css_tryget(css)) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800906 /*
907 * Make sure the memcg is initialized:
908 * mem_cgroup_css_online() orders the the
909 * initialization against setting the flag.
910 */
911 if (smp_load_acquire(&memcg->initialized))
912 break;
913
914 css_put(css);
915 }
916
917 memcg = NULL;
918 }
919
920 if (reclaim) {
921 if (cmpxchg(&iter->position, pos, memcg) == pos) {
922 if (memcg)
923 css_get(&memcg->css);
924 if (pos)
925 css_put(&pos->css);
926 }
927
928 /*
929 * pairs with css_tryget when dereferencing iter->position
930 * above.
931 */
932 if (pos)
933 css_put(&pos->css);
934
935 if (!memcg)
936 iter->generation++;
937 else if (!prev)
938 reclaim->generation = iter->generation;
939 }
940
Michal Hocko542f85f2013-04-29 15:07:15 -0700941out_unlock:
942 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800943out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700944 if (prev && prev != root)
945 css_put(&prev->css);
946
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800947 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700948}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800949
Johannes Weiner56600482012-01-12 17:17:59 -0800950/**
951 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
952 * @root: hierarchy root
953 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
954 */
955void mem_cgroup_iter_break(struct mem_cgroup *root,
956 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800957{
958 if (!root)
959 root = root_mem_cgroup;
960 if (prev && prev != root)
961 css_put(&prev->css);
962}
963
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700964/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800965 * Iteration constructs for visiting all cgroups (under a tree). If
966 * loops are exited prematurely (break), mem_cgroup_iter_break() must
967 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700968 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800969#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800970 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800971 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800972 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700973
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800974#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800975 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800976 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800977 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700978
Johannes Weiner925b7672012-01-12 17:18:15 -0800979/**
980 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
981 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700982 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -0800983 *
984 * Returns the lru list vector holding pages for the given @zone and
985 * @mem. This can be the global zone lruvec, if the memory controller
986 * is disabled.
987 */
988struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
989 struct mem_cgroup *memcg)
990{
991 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800992 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -0800993
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800994 if (mem_cgroup_disabled()) {
995 lruvec = &zone->lruvec;
996 goto out;
997 }
Johannes Weiner925b7672012-01-12 17:18:15 -0800998
Jianyu Zhane2318752014-06-06 14:38:20 -0700999 mz = mem_cgroup_zone_zoneinfo(memcg, zone);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001000 lruvec = &mz->lruvec;
1001out:
1002 /*
1003 * Since a node can be onlined after the mem_cgroup was created,
1004 * we have to be prepared to initialize lruvec->zone here;
1005 * and if offlined then reonlined, we need to reinitialize it.
1006 */
1007 if (unlikely(lruvec->zone != zone))
1008 lruvec->zone = zone;
1009 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001010}
1011
Johannes Weiner925b7672012-01-12 17:18:15 -08001012/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001013 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -08001014 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001015 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001016 *
1017 * This function is only safe when following the LRU page isolation
1018 * and putback protocol: the LRU lock must be held, and the page must
1019 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -07001020 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001021struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001022{
1023 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001024 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001025 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001026
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001027 if (mem_cgroup_disabled()) {
1028 lruvec = &zone->lruvec;
1029 goto out;
1030 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001031
Johannes Weiner1306a852014-12-10 15:44:52 -08001032 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001033 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001034 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -08001035 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -08001036 */
Johannes Weiner29833312014-12-10 15:44:02 -08001037 if (!memcg)
1038 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001039
Jianyu Zhane2318752014-06-06 14:38:20 -07001040 mz = mem_cgroup_page_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001041 lruvec = &mz->lruvec;
1042out:
1043 /*
1044 * Since a node can be onlined after the mem_cgroup was created,
1045 * we have to be prepared to initialize lruvec->zone here;
1046 * and if offlined then reonlined, we need to reinitialize it.
1047 */
1048 if (unlikely(lruvec->zone != zone))
1049 lruvec->zone = zone;
1050 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001051}
1052
1053/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001054 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1055 * @lruvec: mem_cgroup per zone lru vector
1056 * @lru: index of lru list the page is sitting on
1057 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001058 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001059 * This function must be called when a page is added to or removed from an
1060 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001061 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001062void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1063 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001064{
1065 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001066 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001067
1068 if (mem_cgroup_disabled())
1069 return;
1070
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001071 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1072 lru_size = mz->lru_size + lru;
1073 *lru_size += nr_pages;
1074 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001075}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001076
Johannes Weiner2314b422014-12-10 15:44:33 -08001077bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001078{
Johannes Weiner2314b422014-12-10 15:44:33 -08001079 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001080 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001081 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001082
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001083 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001084 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001085 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001086 task_unlock(p);
1087 } else {
1088 /*
1089 * All threads may have already detached their mm's, but the oom
1090 * killer still needs to detect if they have already been oom
1091 * killed to prevent needlessly killing additional tasks.
1092 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001093 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001094 task_memcg = mem_cgroup_from_task(task);
1095 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001096 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001097 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001098 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1099 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001100 return ret;
1101}
1102
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001103#define mem_cgroup_from_counter(counter, member) \
Balbir Singh6d61ef42009-01-07 18:08:06 -08001104 container_of(counter, struct mem_cgroup, member)
1105
Johannes Weiner19942822011-02-01 15:52:43 -08001106/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001107 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001108 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001109 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001110 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001111 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001112 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001113static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001114{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001115 unsigned long margin = 0;
1116 unsigned long count;
1117 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001118
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001119 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001120 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001121 if (count < limit)
1122 margin = limit - count;
1123
1124 if (do_swap_account) {
1125 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001126 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001127 if (count <= limit)
1128 margin = min(margin, limit - count);
1129 }
1130
1131 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001132}
1133
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001134/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001135 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001136 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001137 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1138 * moving cgroups. This is for waiting at high-memory pressure
1139 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001140 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001141static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001142{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001143 struct mem_cgroup *from;
1144 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001145 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001146 /*
1147 * Unlike task_move routines, we access mc.to, mc.from not under
1148 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1149 */
1150 spin_lock(&mc.lock);
1151 from = mc.from;
1152 to = mc.to;
1153 if (!from)
1154 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001155
Johannes Weiner2314b422014-12-10 15:44:33 -08001156 ret = mem_cgroup_is_descendant(from, memcg) ||
1157 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001158unlock:
1159 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001160 return ret;
1161}
1162
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001163static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001164{
1165 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001166 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001167 DEFINE_WAIT(wait);
1168 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1169 /* moving charge context might have finished. */
1170 if (mc.moving_task)
1171 schedule();
1172 finish_wait(&mc.waitq, &wait);
1173 return true;
1174 }
1175 }
1176 return false;
1177}
1178
Sha Zhengju58cf1882013-02-22 16:32:05 -08001179#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001180/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001181 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001182 * @memcg: The memory cgroup that went over limit
1183 * @p: Task that is going to be killed
1184 *
1185 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1186 * enabled
1187 */
1188void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1189{
Tejun Heoe61734c2014-02-12 09:29:50 -05001190 /* oom_info_lock ensures that parallel ooms do not interleave */
Michal Hocko08088cb2014-02-25 15:01:44 -08001191 static DEFINE_MUTEX(oom_info_lock);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001192 struct mem_cgroup *iter;
1193 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001194
Michal Hocko08088cb2014-02-25 15:01:44 -08001195 mutex_lock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001196 rcu_read_lock();
1197
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001198 if (p) {
1199 pr_info("Task in ");
1200 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1201 pr_cont(" killed as a result of limit of ");
1202 } else {
1203 pr_info("Memory limit reached of cgroup ");
1204 }
1205
Tejun Heoe61734c2014-02-12 09:29:50 -05001206 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001207 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001208
Balbir Singhe2224322009-04-02 16:57:39 -07001209 rcu_read_unlock();
1210
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001211 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1212 K((u64)page_counter_read(&memcg->memory)),
1213 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1214 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1215 K((u64)page_counter_read(&memcg->memsw)),
1216 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1217 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1218 K((u64)page_counter_read(&memcg->kmem)),
1219 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001220
1221 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001222 pr_info("Memory cgroup stats for ");
1223 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001224 pr_cont(":");
1225
1226 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1227 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1228 continue;
1229 pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1230 K(mem_cgroup_read_stat(iter, i)));
1231 }
1232
1233 for (i = 0; i < NR_LRU_LISTS; i++)
1234 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1235 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1236
1237 pr_cont("\n");
1238 }
Michal Hocko08088cb2014-02-25 15:01:44 -08001239 mutex_unlock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001240}
1241
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001242/*
1243 * This function returns the number of memcg under hierarchy tree. Returns
1244 * 1(self count) if no children.
1245 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001246static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001247{
1248 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001249 struct mem_cgroup *iter;
1250
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001251 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001252 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001253 return num;
1254}
1255
Balbir Singh6d61ef42009-01-07 18:08:06 -08001256/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001257 * Return the memory (and swap, if configured) limit for a memcg.
1258 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001259static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001260{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001261 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001262
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001263 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001264 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001265 unsigned long memsw_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001266
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001267 memsw_limit = memcg->memsw.limit;
1268 limit = min(limit + total_swap_pages, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001269 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001270 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001271}
1272
David Rientjes19965462012-12-11 16:00:26 -08001273static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1274 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001275{
1276 struct mem_cgroup *iter;
1277 unsigned long chosen_points = 0;
1278 unsigned long totalpages;
1279 unsigned int points = 0;
1280 struct task_struct *chosen = NULL;
1281
David Rientjes876aafb2012-07-31 16:43:48 -07001282 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001283 * If current has a pending SIGKILL or is exiting, then automatically
1284 * select it. The goal is to allow it to allocate so that it may
1285 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001286 */
Oleg Nesterovd003f372014-12-12 16:56:24 -08001287 if (fatal_signal_pending(current) || task_will_free_mem(current)) {
Michal Hocko49550b62015-02-11 15:26:12 -08001288 mark_tsk_oom_victim(current);
David Rientjes876aafb2012-07-31 16:43:48 -07001289 return;
1290 }
1291
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001292 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL, memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001293 totalpages = mem_cgroup_get_limit(memcg) ? : 1;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001294 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001295 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001296 struct task_struct *task;
1297
Tejun Heo72ec7022013-08-08 20:11:26 -04001298 css_task_iter_start(&iter->css, &it);
1299 while ((task = css_task_iter_next(&it))) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001300 switch (oom_scan_process_thread(task, totalpages, NULL,
1301 false)) {
1302 case OOM_SCAN_SELECT:
1303 if (chosen)
1304 put_task_struct(chosen);
1305 chosen = task;
1306 chosen_points = ULONG_MAX;
1307 get_task_struct(chosen);
1308 /* fall through */
1309 case OOM_SCAN_CONTINUE:
1310 continue;
1311 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001312 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001313 mem_cgroup_iter_break(memcg, iter);
1314 if (chosen)
1315 put_task_struct(chosen);
1316 return;
1317 case OOM_SCAN_OK:
1318 break;
1319 };
1320 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001321 if (!points || points < chosen_points)
1322 continue;
1323 /* Prefer thread group leaders for display purposes */
1324 if (points == chosen_points &&
1325 thread_group_leader(chosen))
1326 continue;
1327
1328 if (chosen)
1329 put_task_struct(chosen);
1330 chosen = task;
1331 chosen_points = points;
1332 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001333 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001334 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001335 }
1336
1337 if (!chosen)
1338 return;
1339 points = chosen_points * 1000 / totalpages;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001340 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1341 NULL, "Memory cgroup out of memory");
David Rientjes9cbb78b2012-07-31 16:43:44 -07001342}
1343
Michele Curtiae6e71d2014-12-12 16:56:35 -08001344#if MAX_NUMNODES > 1
1345
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001346/**
1347 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001348 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001349 * @nid: the node ID to be checked.
1350 * @noswap : specify true here if the user wants flle only information.
1351 *
1352 * This function returns whether the specified memcg contains any
1353 * reclaimable pages on a node. Returns true if there are any reclaimable
1354 * pages in the node.
1355 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001356static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001357 int nid, bool noswap)
1358{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001359 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001360 return true;
1361 if (noswap || !total_swap_pages)
1362 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001363 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001364 return true;
1365 return false;
1366
1367}
Ying Han889976d2011-05-26 16:25:33 -07001368
1369/*
1370 * Always updating the nodemask is not very good - even if we have an empty
1371 * list or the wrong list here, we can start from some node and traverse all
1372 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1373 *
1374 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001375static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001376{
1377 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001378 /*
1379 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1380 * pagein/pageout changes since the last update.
1381 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001382 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001383 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001384 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001385 return;
1386
Ying Han889976d2011-05-26 16:25:33 -07001387 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001388 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001389
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001390 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001391
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001392 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1393 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001394 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001395
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001396 atomic_set(&memcg->numainfo_events, 0);
1397 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001398}
1399
1400/*
1401 * Selecting a node where we start reclaim from. Because what we need is just
1402 * reducing usage counter, start from anywhere is O,K. Considering
1403 * memory reclaim from current node, there are pros. and cons.
1404 *
1405 * Freeing memory from current node means freeing memory from a node which
1406 * we'll use or we've used. So, it may make LRU bad. And if several threads
1407 * hit limits, it will see a contention on a node. But freeing from remote
1408 * node means more costs for memory reclaim because of memory latency.
1409 *
1410 * Now, we use round-robin. Better algorithm is welcomed.
1411 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001412int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001413{
1414 int node;
1415
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001416 mem_cgroup_may_update_nodemask(memcg);
1417 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001418
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001419 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001420 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001421 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001422 /*
1423 * We call this when we hit limit, not when pages are added to LRU.
1424 * No LRU may hold pages because all pages are UNEVICTABLE or
1425 * memcg is too small and all pages are not on LRU. In that case,
1426 * we use curret node.
1427 */
1428 if (unlikely(node == MAX_NUMNODES))
1429 node = numa_node_id();
1430
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001431 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001432 return node;
1433}
Ying Han889976d2011-05-26 16:25:33 -07001434#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001435int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001436{
1437 return 0;
1438}
1439#endif
1440
Andrew Morton0608f432013-09-24 15:27:41 -07001441static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1442 struct zone *zone,
1443 gfp_t gfp_mask,
1444 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001445{
Andrew Morton0608f432013-09-24 15:27:41 -07001446 struct mem_cgroup *victim = NULL;
1447 int total = 0;
1448 int loop = 0;
1449 unsigned long excess;
1450 unsigned long nr_scanned;
1451 struct mem_cgroup_reclaim_cookie reclaim = {
1452 .zone = zone,
1453 .priority = 0,
1454 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001455
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001456 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001457
Andrew Morton0608f432013-09-24 15:27:41 -07001458 while (1) {
1459 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1460 if (!victim) {
1461 loop++;
1462 if (loop >= 2) {
1463 /*
1464 * If we have not been able to reclaim
1465 * anything, it might because there are
1466 * no reclaimable pages under this hierarchy
1467 */
1468 if (!total)
1469 break;
1470 /*
1471 * We want to do more targeted reclaim.
1472 * excess >> 2 is not to excessive so as to
1473 * reclaim too much, nor too less that we keep
1474 * coming back to reclaim from this cgroup
1475 */
1476 if (total >= (excess >> 2) ||
1477 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1478 break;
1479 }
1480 continue;
1481 }
Andrew Morton0608f432013-09-24 15:27:41 -07001482 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1483 zone, &nr_scanned);
1484 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001485 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001486 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001487 }
Andrew Morton0608f432013-09-24 15:27:41 -07001488 mem_cgroup_iter_break(root_memcg, victim);
1489 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001490}
1491
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001492#ifdef CONFIG_LOCKDEP
1493static struct lockdep_map memcg_oom_lock_dep_map = {
1494 .name = "memcg_oom_lock",
1495};
1496#endif
1497
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001498static DEFINE_SPINLOCK(memcg_oom_lock);
1499
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001500/*
1501 * Check OOM-Killer is already running under our hierarchy.
1502 * If someone is running, return false.
1503 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001504static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001505{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001506 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001507
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001508 spin_lock(&memcg_oom_lock);
1509
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001510 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001511 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001512 /*
1513 * this subtree of our hierarchy is already locked
1514 * so we cannot give a lock.
1515 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001516 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001517 mem_cgroup_iter_break(memcg, iter);
1518 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001519 } else
1520 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001521 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001522
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001523 if (failed) {
1524 /*
1525 * OK, we failed to lock the whole subtree so we have
1526 * to clean up what we set up to the failing subtree
1527 */
1528 for_each_mem_cgroup_tree(iter, memcg) {
1529 if (iter == failed) {
1530 mem_cgroup_iter_break(memcg, iter);
1531 break;
1532 }
1533 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001534 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001535 } else
1536 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001537
1538 spin_unlock(&memcg_oom_lock);
1539
1540 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001541}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001542
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001543static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001544{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001545 struct mem_cgroup *iter;
1546
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001547 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001548 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001549 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001550 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001551 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001552}
1553
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001554static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001555{
1556 struct mem_cgroup *iter;
1557
Tejun Heo47088812015-06-24 16:58:23 -07001558 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001559 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heo47088812015-06-24 16:58:23 -07001560 iter->under_oom++;
1561 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001562}
1563
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001564static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001565{
1566 struct mem_cgroup *iter;
1567
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001568 /*
1569 * When a new child is created while the hierarchy is under oom,
Tejun Heo47088812015-06-24 16:58:23 -07001570 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001571 */
Tejun Heo47088812015-06-24 16:58:23 -07001572 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001573 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heo47088812015-06-24 16:58:23 -07001574 if (iter->under_oom > 0)
1575 iter->under_oom--;
1576 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001577}
1578
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001579static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1580
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001581struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001582 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001583 wait_queue_t wait;
1584};
1585
1586static int memcg_oom_wake_function(wait_queue_t *wait,
1587 unsigned mode, int sync, void *arg)
1588{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001589 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1590 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001591 struct oom_wait_info *oom_wait_info;
1592
1593 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001594 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001595
Johannes Weiner2314b422014-12-10 15:44:33 -08001596 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1597 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001598 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001599 return autoremove_wake_function(wait, mode, sync, arg);
1600}
1601
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001602static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001603{
Tejun Heo47088812015-06-24 16:58:23 -07001604 /*
1605 * For the following lockless ->under_oom test, the only required
1606 * guarantee is that it must see the state asserted by an OOM when
1607 * this function is called as a result of userland actions
1608 * triggered by the notification of the OOM. This is trivially
1609 * achieved by invoking mem_cgroup_mark_under_oom() before
1610 * triggering notification.
1611 */
1612 if (memcg && memcg->under_oom)
Tejun Heo55c3f9b2015-06-24 16:58:21 -07001613 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001614}
1615
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001616static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001617{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001618 if (!current->memcg_oom.may_oom)
1619 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001620 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001621 * We are in the middle of the charge context here, so we
1622 * don't want to block when potentially sitting on a callstack
1623 * that holds all kinds of filesystem and mm locks.
1624 *
1625 * Also, the caller may handle a failed allocation gracefully
1626 * (like optional page cache readahead) and so an OOM killer
1627 * invocation might not even be necessary.
1628 *
1629 * That's why we don't do anything here except remember the
1630 * OOM context and then deal with it at the end of the page
1631 * fault when the stack is unwound, the locks are released,
1632 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001633 */
Johannes Weiner49426422013-10-16 13:46:59 -07001634 css_get(&memcg->css);
1635 current->memcg_oom.memcg = memcg;
1636 current->memcg_oom.gfp_mask = mask;
1637 current->memcg_oom.order = order;
1638}
1639
1640/**
1641 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1642 * @handle: actually kill/wait or just clean up the OOM state
1643 *
1644 * This has to be called at the end of a page fault if the memcg OOM
1645 * handler was enabled.
1646 *
1647 * Memcg supports userspace OOM handling where failed allocations must
1648 * sleep on a waitqueue until the userspace task resolves the
1649 * situation. Sleeping directly in the charge context with all kinds
1650 * of locks held is not a good idea, instead we remember an OOM state
1651 * in the task and mem_cgroup_oom_synchronize() has to be called at
1652 * the end of the page fault to complete the OOM handling.
1653 *
1654 * Returns %true if an ongoing memcg OOM situation was detected and
1655 * completed, %false otherwise.
1656 */
1657bool mem_cgroup_oom_synchronize(bool handle)
1658{
1659 struct mem_cgroup *memcg = current->memcg_oom.memcg;
1660 struct oom_wait_info owait;
1661 bool locked;
1662
1663 /* OOM is global, do not handle */
1664 if (!memcg)
1665 return false;
1666
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001667 if (!handle || oom_killer_disabled)
Johannes Weiner49426422013-10-16 13:46:59 -07001668 goto cleanup;
1669
1670 owait.memcg = memcg;
1671 owait.wait.flags = 0;
1672 owait.wait.func = memcg_oom_wake_function;
1673 owait.wait.private = current;
1674 INIT_LIST_HEAD(&owait.wait.task_list);
1675
1676 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001677 mem_cgroup_mark_under_oom(memcg);
1678
1679 locked = mem_cgroup_oom_trylock(memcg);
1680
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001681 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001682 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001683
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001684 if (locked && !memcg->oom_kill_disable) {
1685 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001686 finish_wait(&memcg_oom_waitq, &owait.wait);
1687 mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask,
1688 current->memcg_oom.order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001689 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001690 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001691 mem_cgroup_unmark_under_oom(memcg);
1692 finish_wait(&memcg_oom_waitq, &owait.wait);
1693 }
1694
1695 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001696 mem_cgroup_oom_unlock(memcg);
1697 /*
1698 * There is no guarantee that an OOM-lock contender
1699 * sees the wakeups triggered by the OOM kill
1700 * uncharges. Wake any sleepers explicitely.
1701 */
1702 memcg_oom_recover(memcg);
1703 }
Johannes Weiner49426422013-10-16 13:46:59 -07001704cleanup:
1705 current->memcg_oom.memcg = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001706 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001707 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001708}
1709
Johannes Weinerd7365e72014-10-29 14:50:48 -07001710/**
1711 * mem_cgroup_begin_page_stat - begin a page state statistics transaction
1712 * @page: page that is going to change accounted state
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001713 *
Johannes Weinerd7365e72014-10-29 14:50:48 -07001714 * This function must mark the beginning of an accounted page state
1715 * change to prevent double accounting when the page is concurrently
1716 * being moved to another memcg:
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001717 *
Johannes Weiner6de22612015-02-11 15:25:01 -08001718 * memcg = mem_cgroup_begin_page_stat(page);
Johannes Weinerd7365e72014-10-29 14:50:48 -07001719 * if (TestClearPageState(page))
1720 * mem_cgroup_update_page_stat(memcg, state, -1);
Johannes Weiner6de22612015-02-11 15:25:01 -08001721 * mem_cgroup_end_page_stat(memcg);
Balbir Singhd69b0422009-06-17 16:26:34 -07001722 */
Johannes Weiner6de22612015-02-11 15:25:01 -08001723struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001724{
1725 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001726 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001727
Johannes Weiner6de22612015-02-11 15:25:01 -08001728 /*
1729 * The RCU lock is held throughout the transaction. The fast
1730 * path can get away without acquiring the memcg->move_lock
1731 * because page moving starts with an RCU grace period.
1732 *
1733 * The RCU lock also protects the memcg from being freed when
1734 * the page state that is going to change is the only thing
1735 * preventing the page from being uncharged.
1736 * E.g. end-writeback clearing PageWriteback(), which allows
1737 * migration to go ahead and uncharge the page before the
1738 * account transaction might be complete.
1739 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001740 rcu_read_lock();
1741
1742 if (mem_cgroup_disabled())
1743 return NULL;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001744again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001745 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001746 if (unlikely(!memcg))
Johannes Weinerd7365e72014-10-29 14:50:48 -07001747 return NULL;
1748
Qiang Huangbdcbb652014-06-04 16:08:21 -07001749 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weinerd7365e72014-10-29 14:50:48 -07001750 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001751
Johannes Weiner6de22612015-02-11 15:25:01 -08001752 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001753 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001754 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001755 goto again;
1756 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001757
1758 /*
1759 * When charge migration first begins, we can have locked and
1760 * unlocked page stat updates happening concurrently. Track
1761 * the task who has the lock for mem_cgroup_end_page_stat().
1762 */
1763 memcg->move_lock_task = current;
1764 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001765
1766 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001767}
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04001768EXPORT_SYMBOL(mem_cgroup_begin_page_stat);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001769
Johannes Weinerd7365e72014-10-29 14:50:48 -07001770/**
1771 * mem_cgroup_end_page_stat - finish a page state statistics transaction
1772 * @memcg: the memcg that was accounted against
Johannes Weinerd7365e72014-10-29 14:50:48 -07001773 */
Johannes Weiner6de22612015-02-11 15:25:01 -08001774void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001775{
Johannes Weiner6de22612015-02-11 15:25:01 -08001776 if (memcg && memcg->move_lock_task == current) {
1777 unsigned long flags = memcg->move_lock_flags;
1778
1779 memcg->move_lock_task = NULL;
1780 memcg->move_lock_flags = 0;
1781
1782 spin_unlock_irqrestore(&memcg->move_lock, flags);
1783 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001784
Johannes Weinerd7365e72014-10-29 14:50:48 -07001785 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001786}
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04001787EXPORT_SYMBOL(mem_cgroup_end_page_stat);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001788
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001789/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001790 * size of first charge trial. "32" comes from vmscan.c's magic value.
1791 * TODO: maybe necessary to use big numbers in big irons.
1792 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001793#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001794struct memcg_stock_pcp {
1795 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001796 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001797 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001798 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001799#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001800};
1801static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001802static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001803
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001804/**
1805 * consume_stock: Try to consume stocked charge on this cpu.
1806 * @memcg: memcg to consume from.
1807 * @nr_pages: how many pages to charge.
1808 *
1809 * The charges will only happen if @memcg matches the current cpu's memcg
1810 * stock, and at least @nr_pages are available in that stock. Failure to
1811 * service an allocation will refill the stock.
1812 *
1813 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001814 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001815static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001816{
1817 struct memcg_stock_pcp *stock;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001818 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001819
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001820 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001821 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001822
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001823 stock = &get_cpu_var(memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001824 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001825 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001826 ret = true;
1827 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001828 put_cpu_var(memcg_stock);
1829 return ret;
1830}
1831
1832/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001833 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001834 */
1835static void drain_stock(struct memcg_stock_pcp *stock)
1836{
1837 struct mem_cgroup *old = stock->cached;
1838
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001839 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001840 page_counter_uncharge(&old->memory, stock->nr_pages);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001841 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001842 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001843 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001844 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001845 }
1846 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001847}
1848
1849/*
1850 * This must be called under preempt disabled or must be called by
1851 * a thread which is pinned to local cpu.
1852 */
1853static void drain_local_stock(struct work_struct *dummy)
1854{
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001855 struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001856 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001857 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001858}
1859
1860/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001861 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001862 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001863 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001864static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001865{
1866 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1867
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001868 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001869 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001870 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001871 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001872 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001873 put_cpu_var(memcg_stock);
1874}
1875
1876/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001877 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001878 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001879 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001880static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001881{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001882 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001883
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001884 /* If someone's already draining, avoid adding running more workers. */
1885 if (!mutex_trylock(&percpu_charge_mutex))
1886 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001887 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001888 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001889 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001890 for_each_online_cpu(cpu) {
1891 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001892 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001893
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001894 memcg = stock->cached;
1895 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001896 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001897 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001898 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001899 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1900 if (cpu == curcpu)
1901 drain_local_stock(&stock->work);
1902 else
1903 schedule_work_on(cpu, &stock->work);
1904 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001905 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001906 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001907 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001908 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001909}
1910
Paul Gortmaker0db06282013-06-19 14:53:51 -04001911static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001912 unsigned long action,
1913 void *hcpu)
1914{
1915 int cpu = (unsigned long)hcpu;
1916 struct memcg_stock_pcp *stock;
1917
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001918 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001919 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001920
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07001921 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001922 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001923
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001924 stock = &per_cpu(memcg_stock, cpu);
1925 drain_stock(stock);
1926 return NOTIFY_OK;
1927}
1928
Johannes Weiner00501b52014-08-08 14:19:20 -07001929static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1930 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001931{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001932 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001933 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001934 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001935 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001936 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001937 bool may_swap = true;
1938 bool drained = false;
Johannes Weiner05b84302014-08-06 16:05:59 -07001939 int ret = 0;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001940
Johannes Weinerce00a962014-09-05 08:43:57 -04001941 if (mem_cgroup_is_root(memcg))
1942 goto done;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001943retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001944 if (consume_stock(memcg, nr_pages))
1945 goto done;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001946
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001947 if (!do_swap_account ||
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001948 !page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1949 if (!page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001950 goto done_restock;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001951 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001952 page_counter_uncharge(&memcg->memsw, batch);
1953 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001954 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001955 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001956 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001957 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001958
Johannes Weiner6539cc02014-08-06 16:05:42 -07001959 if (batch > nr_pages) {
1960 batch = nr_pages;
1961 goto retry;
1962 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001963
Johannes Weiner06b078f2014-08-06 16:05:44 -07001964 /*
1965 * Unlike in global OOM situations, memcg is not in a physical
1966 * memory shortage. Allow dying and OOM-killed tasks to
1967 * bypass the last charges so that they can exit quickly and
1968 * free their memory.
1969 */
1970 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1971 fatal_signal_pending(current) ||
1972 current->flags & PF_EXITING))
1973 goto bypass;
1974
1975 if (unlikely(task_in_memcg_oom(current)))
1976 goto nomem;
1977
Johannes Weiner6539cc02014-08-06 16:05:42 -07001978 if (!(gfp_mask & __GFP_WAIT))
1979 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001980
Johannes Weiner241994ed2015-02-11 15:26:06 -08001981 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
1982
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001983 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1984 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07001985
Johannes Weiner61e02c72014-08-06 16:08:16 -07001986 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07001987 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07001988
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001989 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001990 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001991 drained = true;
1992 goto retry;
1993 }
1994
Johannes Weiner28c34c22014-08-06 16:05:47 -07001995 if (gfp_mask & __GFP_NORETRY)
1996 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001997 /*
1998 * Even though the limit is exceeded at this point, reclaim
1999 * may have been able to free some pages. Retry the charge
2000 * before killing the task.
2001 *
2002 * Only for regular pages, though: huge pages are rather
2003 * unlikely to succeed so close to the limit, and we fall back
2004 * to regular pages anyway in case of failure.
2005 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002006 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002007 goto retry;
2008 /*
2009 * At task move, charge accounts can be doubly counted. So, it's
2010 * better to wait until the end of task_move if something is going on.
2011 */
2012 if (mem_cgroup_wait_acct_move(mem_over_limit))
2013 goto retry;
2014
Johannes Weiner9b130612014-08-06 16:05:51 -07002015 if (nr_retries--)
2016 goto retry;
2017
Johannes Weiner06b078f2014-08-06 16:05:44 -07002018 if (gfp_mask & __GFP_NOFAIL)
2019 goto bypass;
2020
Johannes Weiner6539cc02014-08-06 16:05:42 -07002021 if (fatal_signal_pending(current))
2022 goto bypass;
2023
Johannes Weiner241994ed2015-02-11 15:26:06 -08002024 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
2025
Johannes Weiner61e02c72014-08-06 16:08:16 -07002026 mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002027nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002028 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002029 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002030bypass:
Johannes Weinerce00a962014-09-05 08:43:57 -04002031 return -EINTR;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002032
2033done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002034 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002035 if (batch > nr_pages)
2036 refill_stock(memcg, batch - nr_pages);
Vladimir Davydov7d638092015-06-10 11:14:46 -07002037 if (!(gfp_mask & __GFP_WAIT))
2038 goto done;
Johannes Weiner241994ed2015-02-11 15:26:06 -08002039 /*
2040 * If the hierarchy is above the normal consumption range,
2041 * make the charging task trim their excess contribution.
2042 */
2043 do {
2044 if (page_counter_read(&memcg->memory) <= memcg->high)
2045 continue;
2046 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
2047 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
2048 } while ((memcg = parent_mem_cgroup(memcg)));
Johannes Weiner6539cc02014-08-06 16:05:42 -07002049done:
Johannes Weiner05b84302014-08-06 16:05:59 -07002050 return ret;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002051}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002052
Johannes Weiner00501b52014-08-08 14:19:20 -07002053static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002054{
Johannes Weinerce00a962014-09-05 08:43:57 -04002055 if (mem_cgroup_is_root(memcg))
2056 return;
2057
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002058 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner05b84302014-08-06 16:05:59 -07002059 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002060 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002061
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002062 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002063}
2064
2065/*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002066 * try_get_mem_cgroup_from_page - look up page's memcg association
2067 * @page: the page
2068 *
2069 * Look up, get a css reference, and return the memcg that owns @page.
2070 *
2071 * The page must be locked to prevent racing with swap-in and page
2072 * cache charges. If coming from an unlocked page table, the caller
2073 * must ensure the page is on the LRU or this can race with charging.
2074 */
Wu Fengguange42d9d52009-12-16 12:19:59 +01002075struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002076{
Johannes Weiner29833312014-12-10 15:44:02 -08002077 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002078 unsigned short id;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002079 swp_entry_t ent;
2080
Sasha Levin309381fea2014-01-23 15:52:54 -08002081 VM_BUG_ON_PAGE(!PageLocked(page), page);
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002082
Johannes Weiner1306a852014-12-10 15:44:52 -08002083 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08002084 if (memcg) {
2085 if (!css_tryget_online(&memcg->css))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002086 memcg = NULL;
Wu Fengguange42d9d52009-12-16 12:19:59 +01002087 } else if (PageSwapCache(page)) {
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002088 ent.val = page_private(page);
Bob Liu9fb4b7c2012-01-12 17:18:48 -08002089 id = lookup_swap_cgroup_id(ent);
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002090 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07002091 memcg = mem_cgroup_from_id(id);
Tejun Heoec903c02014-05-13 12:11:01 -04002092 if (memcg && !css_tryget_online(&memcg->css))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002093 memcg = NULL;
KAMEZAWA Hiroyukia3b2d692009-04-02 16:57:45 -07002094 rcu_read_unlock();
Daisuke Nishimura3c776e62009-04-02 16:57:43 -07002095 }
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002096 return memcg;
KAMEZAWA Hiroyukib5a84312009-01-07 18:08:35 -08002097}
2098
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002099static void lock_page_lru(struct page *page, int *isolated)
2100{
2101 struct zone *zone = page_zone(page);
2102
2103 spin_lock_irq(&zone->lru_lock);
2104 if (PageLRU(page)) {
2105 struct lruvec *lruvec;
2106
2107 lruvec = mem_cgroup_page_lruvec(page, zone);
2108 ClearPageLRU(page);
2109 del_page_from_lru_list(page, lruvec, page_lru(page));
2110 *isolated = 1;
2111 } else
2112 *isolated = 0;
2113}
2114
2115static void unlock_page_lru(struct page *page, int isolated)
2116{
2117 struct zone *zone = page_zone(page);
2118
2119 if (isolated) {
2120 struct lruvec *lruvec;
2121
2122 lruvec = mem_cgroup_page_lruvec(page, zone);
2123 VM_BUG_ON_PAGE(PageLRU(page), page);
2124 SetPageLRU(page);
2125 add_page_to_lru_list(page, lruvec, page_lru(page));
2126 }
2127 spin_unlock_irq(&zone->lru_lock);
2128}
2129
Johannes Weiner00501b52014-08-08 14:19:20 -07002130static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002131 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002132{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002133 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002134
Johannes Weiner1306a852014-12-10 15:44:52 -08002135 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002136
2137 /*
2138 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2139 * may already be on some other mem_cgroup's LRU. Take care of it.
2140 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002141 if (lrucare)
2142 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002143
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002144 /*
2145 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002146 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002147 *
2148 * - the page is uncharged
2149 *
2150 * - the page is off-LRU
2151 *
2152 * - an anonymous fault has exclusive page access, except for
2153 * a locked page table
2154 *
2155 * - a page cache insertion, a swapin fault, or a migration
2156 * have the page locked
2157 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002158 page->mem_cgroup = memcg;
Hugh Dickins3be91272008-02-07 00:14:19 -08002159
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002160 if (lrucare)
2161 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002162}
2163
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002164#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovdbf22eb2015-02-10 14:11:41 -08002165int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp,
2166 unsigned long nr_pages)
Glauber Costa749c5412012-12-18 14:23:01 -08002167{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002168 struct page_counter *counter;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002169 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002170
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002171 ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter);
2172 if (ret < 0)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002173 return ret;
2174
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002175 ret = try_charge(memcg, gfp, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002176 if (ret == -EINTR) {
2177 /*
Johannes Weiner00501b52014-08-08 14:19:20 -07002178 * try_charge() chose to bypass to root due to OOM kill or
2179 * fatal signal. Since our only options are to either fail
2180 * the allocation or charge it to this cgroup, do it as a
2181 * temporary condition. But we can't fail. From a kmem/slab
2182 * perspective, the cache has already been selected, by
2183 * mem_cgroup_kmem_get_cache(), so it is too late to change
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002184 * our minds.
2185 *
2186 * This condition will only trigger if the task entered
Johannes Weiner00501b52014-08-08 14:19:20 -07002187 * memcg_charge_kmem in a sane state, but was OOM-killed
2188 * during try_charge() above. Tasks that were already dying
2189 * when the allocation triggers should have been already
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002190 * directed to the root cgroup in memcontrol.h
2191 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002192 page_counter_charge(&memcg->memory, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002193 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002194 page_counter_charge(&memcg->memsw, nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002195 css_get_many(&memcg->css, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002196 ret = 0;
2197 } else if (ret)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002198 page_counter_uncharge(&memcg->kmem, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002199
2200 return ret;
2201}
2202
Vladimir Davydovdbf22eb2015-02-10 14:11:41 -08002203void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002204{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002205 page_counter_uncharge(&memcg->memory, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002206 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002207 page_counter_uncharge(&memcg->memsw, nr_pages);
Glauber Costa7de37682012-12-18 14:22:07 -08002208
Johannes Weiner64f21992014-12-10 15:42:45 -08002209 page_counter_uncharge(&memcg->kmem, nr_pages);
Glauber Costa7de37682012-12-18 14:22:07 -08002210
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002211 css_put_many(&memcg->css, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002212}
2213
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002214static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002215{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002216 int id, size;
2217 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002218
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002219 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002220 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2221 if (id < 0)
2222 return id;
2223
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002224 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002225 return id;
2226
2227 /*
2228 * There's no space for the new id in memcg_caches arrays,
2229 * so we have to grow them.
2230 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002231 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002232
2233 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002234 if (size < MEMCG_CACHES_MIN_SIZE)
2235 size = MEMCG_CACHES_MIN_SIZE;
2236 else if (size > MEMCG_CACHES_MAX_SIZE)
2237 size = MEMCG_CACHES_MAX_SIZE;
2238
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002239 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002240 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002241 err = memcg_update_all_list_lrus(size);
2242 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002243 memcg_nr_cache_ids = size;
2244
2245 up_write(&memcg_cache_ids_sem);
2246
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002247 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002248 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002249 return err;
2250 }
2251 return id;
2252}
2253
2254static void memcg_free_cache_id(int id)
2255{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002256 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002257}
2258
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002259struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002260 struct mem_cgroup *memcg;
2261 struct kmem_cache *cachep;
2262 struct work_struct work;
2263};
2264
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002265static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002266{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002267 struct memcg_kmem_cache_create_work *cw =
2268 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002269 struct mem_cgroup *memcg = cw->memcg;
2270 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002271
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002272 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002273
Vladimir Davydov5722d092014-04-07 15:39:24 -07002274 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002275 kfree(cw);
2276}
2277
2278/*
2279 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002280 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002281static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2282 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002283{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002284 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002285
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002286 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002287 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002288 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002289
2290 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002291
2292 cw->memcg = memcg;
2293 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002294 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002295
Glauber Costad7f25f82012-12-18 14:22:40 -08002296 schedule_work(&cw->work);
2297}
2298
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002299static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2300 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002301{
2302 /*
2303 * We need to stop accounting when we kmalloc, because if the
2304 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002305 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002306 *
2307 * However, it is better to enclose the whole function. Depending on
2308 * the debugging options enabled, INIT_WORK(), for instance, can
2309 * trigger an allocation. This too, will make us recurse. Because at
2310 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2311 * the safest choice is to do it like this, wrapping the whole function.
2312 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002313 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002314 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002315 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002316}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002317
Glauber Costad7f25f82012-12-18 14:22:40 -08002318/*
2319 * Return the kmem_cache we're supposed to use for a slab allocation.
2320 * We try to use the current memcg's version of the cache.
2321 *
2322 * If the cache does not exist yet, if we are the first user of it,
2323 * we either create it immediately, if possible, or create it asynchronously
2324 * in a workqueue.
2325 * In the latter case, we will let the current allocation go through with
2326 * the original cache.
2327 *
2328 * Can't be called in interrupt context or from kernel threads.
2329 * This function needs to be called with rcu_read_lock() held.
2330 */
Zhang Zhen056b7cc2014-12-12 16:55:38 -08002331struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002332{
2333 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002334 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002335 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002336
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002337 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002338
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002339 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002340 return cachep;
2341
Vladimir Davydov8135be52014-12-12 16:56:38 -08002342 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002343 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002344 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002345 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002346
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002347 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002348 if (likely(memcg_cachep))
2349 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002350
2351 /*
2352 * If we are in a safe context (can wait, and not in interrupt
2353 * context), we could be be predictable and return right away.
2354 * This would guarantee that the allocation being performed
2355 * already belongs in the new cache.
2356 *
2357 * However, there are some clashes that can arrive from locking.
2358 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002359 * memcg_create_kmem_cache, this means no further allocation
2360 * could happen with the slab_mutex held. So it's better to
2361 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002362 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002363 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002364out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002365 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002366 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002367}
Glauber Costad7f25f82012-12-18 14:22:40 -08002368
Vladimir Davydov8135be52014-12-12 16:56:38 -08002369void __memcg_kmem_put_cache(struct kmem_cache *cachep)
2370{
2371 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002372 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002373}
2374
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002375/*
2376 * We need to verify if the allocation against current->mm->owner's memcg is
2377 * possible for the given order. But the page is not allocated yet, so we'll
2378 * need a further commit step to do the final arrangements.
2379 *
2380 * It is possible for the task to switch cgroups in this mean time, so at
2381 * commit time, we can't rely on task conversion any longer. We'll then use
2382 * the handle argument to return to the caller which cgroup we should commit
2383 * against. We could also return the memcg directly and avoid the pointer
2384 * passing, but a boolean return value gives better semantics considering
2385 * the compiled-out case as well.
2386 *
2387 * Returning true means the allocation is possible.
2388 */
2389bool
2390__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
2391{
2392 struct mem_cgroup *memcg;
2393 int ret;
2394
2395 *_memcg = NULL;
Glauber Costa6d42c232013-07-08 16:00:00 -07002396
Johannes Weinerdf381972014-04-07 15:37:43 -07002397 memcg = get_mem_cgroup_from_mm(current->mm);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002398
Vladimir Davydovcf2b8fb2014-10-09 15:28:59 -07002399 if (!memcg_kmem_is_active(memcg)) {
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002400 css_put(&memcg->css);
2401 return true;
2402 }
2403
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002404 ret = memcg_charge_kmem(memcg, gfp, 1 << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002405 if (!ret)
2406 *_memcg = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002407
2408 css_put(&memcg->css);
2409 return (ret == 0);
2410}
2411
2412void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
2413 int order)
2414{
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002415 VM_BUG_ON(mem_cgroup_is_root(memcg));
2416
2417 /* The page allocation failed. Revert */
2418 if (!page) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002419 memcg_uncharge_kmem(memcg, 1 << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002420 return;
2421 }
Johannes Weiner1306a852014-12-10 15:44:52 -08002422 page->mem_cgroup = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002423}
2424
2425void __memcg_kmem_uncharge_pages(struct page *page, int order)
2426{
Johannes Weiner1306a852014-12-10 15:44:52 -08002427 struct mem_cgroup *memcg = page->mem_cgroup;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002428
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002429 if (!memcg)
2430 return;
2431
Sasha Levin309381fea2014-01-23 15:52:54 -08002432 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002433
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002434 memcg_uncharge_kmem(memcg, 1 << order);
Johannes Weiner1306a852014-12-10 15:44:52 -08002435 page->mem_cgroup = NULL;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002436}
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002437
2438struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr)
2439{
2440 struct mem_cgroup *memcg = NULL;
2441 struct kmem_cache *cachep;
2442 struct page *page;
2443
2444 page = virt_to_head_page(ptr);
2445 if (PageSlab(page)) {
2446 cachep = page->slab_cache;
2447 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002448 memcg = cachep->memcg_params.memcg;
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002449 } else
2450 /* page allocated by alloc_kmem_pages */
2451 memcg = page->mem_cgroup;
2452
2453 return memcg;
2454}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002455#endif /* CONFIG_MEMCG_KMEM */
2456
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002457#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2458
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002459/*
2460 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002461 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2462 * charge/uncharge will be never happen and move_account() is done under
2463 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002464 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002465void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002466{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002467 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002468
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002469 if (mem_cgroup_disabled())
2470 return;
David Rientjesb070e652013-05-07 16:18:09 -07002471
Johannes Weiner29833312014-12-10 15:44:02 -08002472 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002473 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002474
Johannes Weiner1306a852014-12-10 15:44:52 -08002475 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002476 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002477}
Hugh Dickins12d27102012-01-12 17:19:52 -08002478#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002479
Andrew Mortonc255a452012-07-31 16:43:02 -07002480#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002481static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2482 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002483{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002484 int val = (charge) ? 1 : -1;
2485 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002486}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002487
2488/**
2489 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2490 * @entry: swap entry to be moved
2491 * @from: mem_cgroup which the entry is moved from
2492 * @to: mem_cgroup which the entry is moved to
2493 *
2494 * It succeeds only when the swap_cgroup's record for this entry is the same
2495 * as the mem_cgroup's id of @from.
2496 *
2497 * Returns 0 on success, -EINVAL on failure.
2498 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002499 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002500 * both res and memsw, and called css_get().
2501 */
2502static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002503 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002504{
2505 unsigned short old_id, new_id;
2506
Li Zefan34c00c32013-09-23 16:56:01 +08002507 old_id = mem_cgroup_id(from);
2508 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002509
2510 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002511 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002512 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002513 return 0;
2514 }
2515 return -EINVAL;
2516}
2517#else
2518static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002519 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002520{
2521 return -EINVAL;
2522}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002523#endif
2524
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002525static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002526
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002527static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002528 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002529{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002530 unsigned long curusage;
2531 unsigned long oldusage;
2532 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002533 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002534 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002535
2536 /*
2537 * For keeping hierarchical_reclaim simple, how long we should retry
2538 * is depends on callers. We set our retry-count to be function
2539 * of # of children which we should visit in this loop.
2540 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002541 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2542 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002543
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002544 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002545
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002546 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002547 if (signal_pending(current)) {
2548 ret = -EINTR;
2549 break;
2550 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002551
2552 mutex_lock(&memcg_limit_mutex);
2553 if (limit > memcg->memsw.limit) {
2554 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002555 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002556 break;
2557 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002558 if (limit > memcg->memory.limit)
2559 enlarge = true;
2560 ret = page_counter_limit(&memcg->memory, limit);
2561 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002562
2563 if (!ret)
2564 break;
2565
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002566 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2567
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002568 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002569 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002570 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002571 retry_count--;
2572 else
2573 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002574 } while (retry_count);
2575
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002576 if (!ret && enlarge)
2577 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002578
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002579 return ret;
2580}
2581
Li Zefan338c8432009-06-17 16:27:15 -07002582static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002583 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002584{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002585 unsigned long curusage;
2586 unsigned long oldusage;
2587 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002588 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002589 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002590
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002591 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002592 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2593 mem_cgroup_count_children(memcg);
2594
2595 oldusage = page_counter_read(&memcg->memsw);
2596
2597 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002598 if (signal_pending(current)) {
2599 ret = -EINTR;
2600 break;
2601 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002602
2603 mutex_lock(&memcg_limit_mutex);
2604 if (limit < memcg->memory.limit) {
2605 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002606 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002607 break;
2608 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002609 if (limit > memcg->memsw.limit)
2610 enlarge = true;
2611 ret = page_counter_limit(&memcg->memsw, limit);
2612 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002613
2614 if (!ret)
2615 break;
2616
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002617 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2618
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002619 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002620 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002621 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002622 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002623 else
2624 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002625 } while (retry_count);
2626
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002627 if (!ret && enlarge)
2628 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002629
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002630 return ret;
2631}
2632
Andrew Morton0608f432013-09-24 15:27:41 -07002633unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2634 gfp_t gfp_mask,
2635 unsigned long *total_scanned)
2636{
2637 unsigned long nr_reclaimed = 0;
2638 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2639 unsigned long reclaimed;
2640 int loop = 0;
2641 struct mem_cgroup_tree_per_zone *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002642 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002643 unsigned long nr_scanned;
2644
2645 if (order > 0)
2646 return 0;
2647
2648 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
2649 /*
2650 * This loop can run a while, specially if mem_cgroup's continuously
2651 * keep exceeding their soft limit and putting the system under
2652 * pressure
2653 */
2654 do {
2655 if (next_mz)
2656 mz = next_mz;
2657 else
2658 mz = mem_cgroup_largest_soft_limit_node(mctz);
2659 if (!mz)
2660 break;
2661
2662 nr_scanned = 0;
2663 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
2664 gfp_mask, &nr_scanned);
2665 nr_reclaimed += reclaimed;
2666 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002667 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002668 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002669
2670 /*
2671 * If we failed to reclaim anything from this memory cgroup
2672 * it is time to move on to the next cgroup
2673 */
2674 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002675 if (!reclaimed)
2676 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2677
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002678 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002679 /*
2680 * One school of thought says that we should not add
2681 * back the node to the tree if reclaim returns 0.
2682 * But our reclaim could return 0, simply because due
2683 * to priority we are exposing a smaller subset of
2684 * memory to reclaim from. Consider this as a longer
2685 * term TODO.
2686 */
2687 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002688 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002689 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002690 css_put(&mz->memcg->css);
2691 loop++;
2692 /*
2693 * Could not reclaim anything and there are no more
2694 * mem cgroups to try or we seem to be looping without
2695 * reclaiming anything.
2696 */
2697 if (!nr_reclaimed &&
2698 (next_mz == NULL ||
2699 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2700 break;
2701 } while (!nr_reclaimed);
2702 if (next_mz)
2703 css_put(&next_mz->memcg->css);
2704 return nr_reclaimed;
2705}
2706
Tejun Heoea280e72014-05-16 13:22:48 -04002707/*
2708 * Test whether @memcg has children, dead or alive. Note that this
2709 * function doesn't care whether @memcg has use_hierarchy enabled and
2710 * returns %true if there are child csses according to the cgroup
2711 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2712 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002713static inline bool memcg_has_children(struct mem_cgroup *memcg)
2714{
Tejun Heoea280e72014-05-16 13:22:48 -04002715 bool ret;
2716
Johannes Weiner696ac172013-10-31 16:34:15 -07002717 /*
Tejun Heoea280e72014-05-16 13:22:48 -04002718 * The lock does not prevent addition or deletion of children, but
2719 * it prevents a new child from being initialized based on this
2720 * parent in css_online(), so it's enough to decide whether
2721 * hierarchically inherited attributes can still be changed or not.
Johannes Weiner696ac172013-10-31 16:34:15 -07002722 */
Tejun Heoea280e72014-05-16 13:22:48 -04002723 lockdep_assert_held(&memcg_create_mutex);
2724
2725 rcu_read_lock();
2726 ret = css_next_child(NULL, &memcg->css);
2727 rcu_read_unlock();
2728 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002729}
2730
2731/*
Michal Hockoc26251f2012-10-26 13:37:28 +02002732 * Reclaims as many pages from the given memcg as possible and moves
2733 * the rest to the parent.
2734 *
2735 * Caller is responsible for holding css reference for memcg.
2736 */
2737static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2738{
2739 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002740
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002741 /* we call try-to-free pages for make this cgroup empty */
2742 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002743 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002744 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002745 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002746
Michal Hockoc26251f2012-10-26 13:37:28 +02002747 if (signal_pending(current))
2748 return -EINTR;
2749
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002750 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2751 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002752 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002753 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002754 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002755 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002756 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002757
2758 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002759
2760 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002761}
2762
Tejun Heo6770c642014-05-13 12:16:21 -04002763static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2764 char *buf, size_t nbytes,
2765 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002766{
Tejun Heo6770c642014-05-13 12:16:21 -04002767 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002768
Michal Hockod8423012012-10-26 13:37:29 +02002769 if (mem_cgroup_is_root(memcg))
2770 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002771 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002772}
2773
Tejun Heo182446d2013-08-08 20:11:24 -04002774static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2775 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002776{
Tejun Heo182446d2013-08-08 20:11:24 -04002777 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002778}
2779
Tejun Heo182446d2013-08-08 20:11:24 -04002780static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2781 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002782{
2783 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002784 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002785 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002786
Glauber Costa09998212013-02-22 16:34:55 -08002787 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07002788
2789 if (memcg->use_hierarchy == val)
2790 goto out;
2791
Balbir Singh18f59ea2009-01-07 18:08:07 -08002792 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002793 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002794 * in the child subtrees. If it is unset, then the change can
2795 * occur, provided the current cgroup has no children.
2796 *
2797 * For the root cgroup, parent_mem is NULL, we allow value to be
2798 * set if there are no children.
2799 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002800 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002801 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002802 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002803 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002804 else
2805 retval = -EBUSY;
2806 } else
2807 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002808
2809out:
Glauber Costa09998212013-02-22 16:34:55 -08002810 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002811
2812 return retval;
2813}
2814
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002815static unsigned long tree_stat(struct mem_cgroup *memcg,
2816 enum mem_cgroup_stat_index idx)
Johannes Weinerce00a962014-09-05 08:43:57 -04002817{
2818 struct mem_cgroup *iter;
2819 long val = 0;
2820
2821 /* Per-cpu values can be negative, use a signed accumulator */
2822 for_each_mem_cgroup_tree(iter, memcg)
2823 val += mem_cgroup_read_stat(iter, idx);
2824
2825 if (val < 0) /* race ? */
2826 val = 0;
2827 return val;
2828}
2829
2830static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
2831{
2832 u64 val;
2833
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002834 if (mem_cgroup_is_root(memcg)) {
2835 val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE);
2836 val += tree_stat(memcg, MEM_CGROUP_STAT_RSS);
2837 if (swap)
2838 val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP);
2839 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002840 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002841 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002842 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002843 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002844 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002845 return val << PAGE_SHIFT;
2846}
2847
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002848enum {
2849 RES_USAGE,
2850 RES_LIMIT,
2851 RES_MAX_USAGE,
2852 RES_FAILCNT,
2853 RES_SOFT_LIMIT,
2854};
Johannes Weinerce00a962014-09-05 08:43:57 -04002855
Tejun Heo791badb2013-12-05 12:28:02 -05002856static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002857 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002858{
Tejun Heo182446d2013-08-08 20:11:24 -04002859 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002860 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002861
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002862 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002863 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002864 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002865 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002866 case _MEMSWAP:
2867 counter = &memcg->memsw;
2868 break;
2869 case _KMEM:
2870 counter = &memcg->kmem;
2871 break;
2872 default:
2873 BUG();
2874 }
2875
2876 switch (MEMFILE_ATTR(cft->private)) {
2877 case RES_USAGE:
2878 if (counter == &memcg->memory)
2879 return mem_cgroup_usage(memcg, false);
2880 if (counter == &memcg->memsw)
2881 return mem_cgroup_usage(memcg, true);
2882 return (u64)page_counter_read(counter) * PAGE_SIZE;
2883 case RES_LIMIT:
2884 return (u64)counter->limit * PAGE_SIZE;
2885 case RES_MAX_USAGE:
2886 return (u64)counter->watermark * PAGE_SIZE;
2887 case RES_FAILCNT:
2888 return counter->failcnt;
2889 case RES_SOFT_LIMIT:
2890 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002891 default:
2892 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002893 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002894}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002895
Glauber Costa510fc4e2012-12-18 14:21:47 -08002896#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08002897static int memcg_activate_kmem(struct mem_cgroup *memcg,
2898 unsigned long nr_pages)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002899{
2900 int err = 0;
2901 int memcg_id;
2902
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002903 BUG_ON(memcg->kmemcg_id >= 0);
Vladimir Davydov2788cf0c2015-02-12 14:59:38 -08002904 BUG_ON(memcg->kmem_acct_activated);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002905 BUG_ON(memcg->kmem_acct_active);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002906
2907 /*
Glauber Costa510fc4e2012-12-18 14:21:47 -08002908 * For simplicity, we won't allow this to be disabled. It also can't
2909 * be changed if the cgroup has children already, or if tasks had
2910 * already joined.
2911 *
2912 * If tasks join before we set the limit, a person looking at
2913 * kmem.usage_in_bytes will have no way to determine when it took
2914 * place, which makes the value quite meaningless.
2915 *
2916 * After it first became limited, changes in the value of the limit are
2917 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08002918 */
Glauber Costa09998212013-02-22 16:34:55 -08002919 mutex_lock(&memcg_create_mutex);
Tejun Heoea280e72014-05-16 13:22:48 -04002920 if (cgroup_has_tasks(memcg->css.cgroup) ||
2921 (memcg->use_hierarchy && memcg_has_children(memcg)))
Vladimir Davydovd6441632014-01-23 15:53:09 -08002922 err = -EBUSY;
Glauber Costa09998212013-02-22 16:34:55 -08002923 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002924 if (err)
2925 goto out;
2926
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002927 memcg_id = memcg_alloc_cache_id();
Vladimir Davydovd6441632014-01-23 15:53:09 -08002928 if (memcg_id < 0) {
2929 err = memcg_id;
2930 goto out;
2931 }
2932
Vladimir Davydovd6441632014-01-23 15:53:09 -08002933 /*
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002934 * We couldn't have accounted to this cgroup, because it hasn't got
2935 * activated yet, so this should succeed.
Vladimir Davydovd6441632014-01-23 15:53:09 -08002936 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002937 err = page_counter_limit(&memcg->kmem, nr_pages);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002938 VM_BUG_ON(err);
2939
2940 static_key_slow_inc(&memcg_kmem_enabled_key);
2941 /*
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002942 * A memory cgroup is considered kmem-active as soon as it gets
2943 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002944 * guarantee no one starts accounting before all call sites are
2945 * patched.
2946 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002947 memcg->kmemcg_id = memcg_id;
Vladimir Davydov2788cf0c2015-02-12 14:59:38 -08002948 memcg->kmem_acct_activated = true;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002949 memcg->kmem_acct_active = true;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002950out:
Vladimir Davydovd6441632014-01-23 15:53:09 -08002951 return err;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002952}
2953
Vladimir Davydovd6441632014-01-23 15:53:09 -08002954static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002955 unsigned long limit)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002956{
2957 int ret;
2958
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002959 mutex_lock(&memcg_limit_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002960 if (!memcg_kmem_is_active(memcg))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002961 ret = memcg_activate_kmem(memcg, limit);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002962 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002963 ret = page_counter_limit(&memcg->kmem, limit);
2964 mutex_unlock(&memcg_limit_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002965 return ret;
2966}
2967
Glauber Costa55007d82012-12-18 14:22:38 -08002968static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08002969{
Glauber Costa55007d82012-12-18 14:22:38 -08002970 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002971 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002972
Glauber Costa510fc4e2012-12-18 14:21:47 -08002973 if (!parent)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002974 return 0;
Glauber Costa55007d82012-12-18 14:22:38 -08002975
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08002976 mutex_lock(&memcg_limit_mutex);
Glauber Costaa8964b92012-12-18 14:22:09 -08002977 /*
Vladimir Davydovd6441632014-01-23 15:53:09 -08002978 * If the parent cgroup is not kmem-active now, it cannot be activated
2979 * after this point, because it has at least one child already.
Glauber Costaa8964b92012-12-18 14:22:09 -08002980 */
Vladimir Davydovd6441632014-01-23 15:53:09 -08002981 if (memcg_kmem_is_active(parent))
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08002982 ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX);
2983 mutex_unlock(&memcg_limit_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08002984 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002985}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002986#else
2987static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002988 unsigned long limit)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002989{
2990 return -EINVAL;
2991}
Hugh Dickins6d0439902013-02-22 16:35:50 -08002992#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08002993
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002994/*
2995 * The user of this function is...
2996 * RES_LIMIT.
2997 */
Tejun Heo451af502014-05-13 12:16:21 -04002998static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2999 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003000{
Tejun Heo451af502014-05-13 12:16:21 -04003001 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003002 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003003 int ret;
3004
Tejun Heo451af502014-05-13 12:16:21 -04003005 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08003006 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003007 if (ret)
3008 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07003009
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003010 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003011 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003012 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3013 ret = -EINVAL;
3014 break;
3015 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003016 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3017 case _MEM:
3018 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003019 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003020 case _MEMSWAP:
3021 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3022 break;
3023 case _KMEM:
3024 ret = memcg_update_kmem_limit(memcg, nr_pages);
3025 break;
3026 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003027 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003028 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003029 memcg->soft_limit = nr_pages;
3030 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003031 break;
3032 }
Tejun Heo451af502014-05-13 12:16:21 -04003033 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003034}
3035
Tejun Heo6770c642014-05-13 12:16:21 -04003036static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3037 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003038{
Tejun Heo6770c642014-05-13 12:16:21 -04003039 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003040 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003041
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003042 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3043 case _MEM:
3044 counter = &memcg->memory;
3045 break;
3046 case _MEMSWAP:
3047 counter = &memcg->memsw;
3048 break;
3049 case _KMEM:
3050 counter = &memcg->kmem;
3051 break;
3052 default:
3053 BUG();
3054 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003055
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003056 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003057 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003058 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003059 break;
3060 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003061 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003062 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003063 default:
3064 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003065 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003066
Tejun Heo6770c642014-05-13 12:16:21 -04003067 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003068}
3069
Tejun Heo182446d2013-08-08 20:11:24 -04003070static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003071 struct cftype *cft)
3072{
Tejun Heo182446d2013-08-08 20:11:24 -04003073 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003074}
3075
Daisuke Nishimura02491442010-03-10 15:22:17 -08003076#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003077static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003078 struct cftype *cft, u64 val)
3079{
Tejun Heo182446d2013-08-08 20:11:24 -04003080 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003081
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003082 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003083 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003084
Glauber Costaee5e8472013-02-22 16:34:50 -08003085 /*
3086 * No kind of locking is needed in here, because ->can_attach() will
3087 * check this value once in the beginning of the process, and then carry
3088 * on with stale data. This means that changes to this value will only
3089 * affect task migrations starting after the change.
3090 */
3091 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003092 return 0;
3093}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003094#else
Tejun Heo182446d2013-08-08 20:11:24 -04003095static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003096 struct cftype *cft, u64 val)
3097{
3098 return -ENOSYS;
3099}
3100#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003101
Ying Han406eb0c2011-05-26 16:25:37 -07003102#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003103static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003104{
Greg Thelen25485de2013-11-12 15:07:40 -08003105 struct numa_stat {
3106 const char *name;
3107 unsigned int lru_mask;
3108 };
3109
3110 static const struct numa_stat stats[] = {
3111 { "total", LRU_ALL },
3112 { "file", LRU_ALL_FILE },
3113 { "anon", LRU_ALL_ANON },
3114 { "unevictable", BIT(LRU_UNEVICTABLE) },
3115 };
3116 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003117 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003118 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003119 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003120
Greg Thelen25485de2013-11-12 15:07:40 -08003121 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3122 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3123 seq_printf(m, "%s=%lu", stat->name, nr);
3124 for_each_node_state(nid, N_MEMORY) {
3125 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3126 stat->lru_mask);
3127 seq_printf(m, " N%d=%lu", nid, nr);
3128 }
3129 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003130 }
Ying Han406eb0c2011-05-26 16:25:37 -07003131
Ying Han071aee12013-11-12 15:07:41 -08003132 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3133 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003134
Ying Han071aee12013-11-12 15:07:41 -08003135 nr = 0;
3136 for_each_mem_cgroup_tree(iter, memcg)
3137 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3138 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3139 for_each_node_state(nid, N_MEMORY) {
3140 nr = 0;
3141 for_each_mem_cgroup_tree(iter, memcg)
3142 nr += mem_cgroup_node_nr_lru_pages(
3143 iter, nid, stat->lru_mask);
3144 seq_printf(m, " N%d=%lu", nid, nr);
3145 }
3146 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003147 }
Ying Han406eb0c2011-05-26 16:25:37 -07003148
Ying Han406eb0c2011-05-26 16:25:37 -07003149 return 0;
3150}
3151#endif /* CONFIG_NUMA */
3152
Tejun Heo2da8ca82013-12-05 12:28:04 -05003153static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003154{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003155 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003156 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003157 struct mem_cgroup *mi;
3158 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003159
Greg Thelen0ca44b12015-02-11 15:25:58 -08003160 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3161 MEM_CGROUP_STAT_NSTATS);
3162 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3163 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003164 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3165
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003166 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07003167 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003168 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003169 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
3170 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003171 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003172
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003173 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3174 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3175 mem_cgroup_read_events(memcg, i));
3176
3177 for (i = 0; i < NR_LRU_LISTS; i++)
3178 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3179 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3180
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003181 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003182 memory = memsw = PAGE_COUNTER_MAX;
3183 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3184 memory = min(memory, mi->memory.limit);
3185 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003186 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003187 seq_printf(m, "hierarchical_memory_limit %llu\n",
3188 (u64)memory * PAGE_SIZE);
3189 if (do_swap_account)
3190 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3191 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003192
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003193 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
3194 long long val = 0;
3195
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07003196 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003197 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003198 for_each_mem_cgroup_tree(mi, memcg)
3199 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
3200 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
3201 }
3202
3203 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3204 unsigned long long val = 0;
3205
3206 for_each_mem_cgroup_tree(mi, memcg)
3207 val += mem_cgroup_read_events(mi, i);
3208 seq_printf(m, "total_%s %llu\n",
3209 mem_cgroup_events_names[i], val);
3210 }
3211
3212 for (i = 0; i < NR_LRU_LISTS; i++) {
3213 unsigned long long val = 0;
3214
3215 for_each_mem_cgroup_tree(mi, memcg)
3216 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3217 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003218 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003219
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003220#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003221 {
3222 int nid, zid;
3223 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003224 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003225 unsigned long recent_rotated[2] = {0, 0};
3226 unsigned long recent_scanned[2] = {0, 0};
3227
3228 for_each_online_node(nid)
3229 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Jianyu Zhane2318752014-06-06 14:38:20 -07003230 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
Hugh Dickins89abfab2012-05-29 15:06:53 -07003231 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003232
Hugh Dickins89abfab2012-05-29 15:06:53 -07003233 recent_rotated[0] += rstat->recent_rotated[0];
3234 recent_rotated[1] += rstat->recent_rotated[1];
3235 recent_scanned[0] += rstat->recent_scanned[0];
3236 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003237 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003238 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3239 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3240 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3241 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003242 }
3243#endif
3244
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003245 return 0;
3246}
3247
Tejun Heo182446d2013-08-08 20:11:24 -04003248static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3249 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003250{
Tejun Heo182446d2013-08-08 20:11:24 -04003251 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003252
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003253 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003254}
3255
Tejun Heo182446d2013-08-08 20:11:24 -04003256static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3257 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003258{
Tejun Heo182446d2013-08-08 20:11:24 -04003259 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003260
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003261 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003262 return -EINVAL;
3263
Linus Torvalds14208b02014-06-09 15:03:33 -07003264 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003265 memcg->swappiness = val;
3266 else
3267 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003268
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003269 return 0;
3270}
3271
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003272static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3273{
3274 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003275 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003276 int i;
3277
3278 rcu_read_lock();
3279 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003280 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003281 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003282 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003283
3284 if (!t)
3285 goto unlock;
3286
Johannes Weinerce00a962014-09-05 08:43:57 -04003287 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003288
3289 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003290 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003291 * If it's not true, a threshold was crossed after last
3292 * call of __mem_cgroup_threshold().
3293 */
Phil Carmody5407a562010-05-26 14:42:42 -07003294 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003295
3296 /*
3297 * Iterate backward over array of thresholds starting from
3298 * current_threshold and check if a threshold is crossed.
3299 * If none of thresholds below usage is crossed, we read
3300 * only one element of the array here.
3301 */
3302 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3303 eventfd_signal(t->entries[i].eventfd, 1);
3304
3305 /* i = current_threshold + 1 */
3306 i++;
3307
3308 /*
3309 * Iterate forward over array of thresholds starting from
3310 * current_threshold+1 and check if a threshold is crossed.
3311 * If none of thresholds above usage is crossed, we read
3312 * only one element of the array here.
3313 */
3314 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3315 eventfd_signal(t->entries[i].eventfd, 1);
3316
3317 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003318 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003319unlock:
3320 rcu_read_unlock();
3321}
3322
3323static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3324{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003325 while (memcg) {
3326 __mem_cgroup_threshold(memcg, false);
3327 if (do_swap_account)
3328 __mem_cgroup_threshold(memcg, true);
3329
3330 memcg = parent_mem_cgroup(memcg);
3331 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003332}
3333
3334static int compare_thresholds(const void *a, const void *b)
3335{
3336 const struct mem_cgroup_threshold *_a = a;
3337 const struct mem_cgroup_threshold *_b = b;
3338
Greg Thelen2bff24a2013-09-11 14:23:08 -07003339 if (_a->threshold > _b->threshold)
3340 return 1;
3341
3342 if (_a->threshold < _b->threshold)
3343 return -1;
3344
3345 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003346}
3347
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003348static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003349{
3350 struct mem_cgroup_eventfd_list *ev;
3351
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003352 spin_lock(&memcg_oom_lock);
3353
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003354 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003355 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003356
3357 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003358 return 0;
3359}
3360
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003361static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003362{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003363 struct mem_cgroup *iter;
3364
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003365 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003366 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003367}
3368
Tejun Heo59b6f872013-11-22 18:20:43 -05003369static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003370 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003371{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003372 struct mem_cgroup_thresholds *thresholds;
3373 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003374 unsigned long threshold;
3375 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003376 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003377
Johannes Weiner650c5e52015-02-11 15:26:03 -08003378 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003379 if (ret)
3380 return ret;
Shaohua Li2105f9a2015-10-15 15:28:29 -07003381 threshold <<= PAGE_SHIFT;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003382
3383 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003384
Johannes Weiner05b84302014-08-06 16:05:59 -07003385 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003386 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003387 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003388 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003389 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003390 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003391 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003392 BUG();
3393
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003394 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003395 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003396 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3397
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003398 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003399
3400 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003401 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003402 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003403 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003404 ret = -ENOMEM;
3405 goto unlock;
3406 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003407 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003408
3409 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003410 if (thresholds->primary) {
3411 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003412 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003413 }
3414
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003415 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003416 new->entries[size - 1].eventfd = eventfd;
3417 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003418
3419 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003420 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003421 compare_thresholds, NULL);
3422
3423 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003424 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003425 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003426 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003427 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003428 * new->current_threshold will not be used until
3429 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003430 * it here.
3431 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003432 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003433 } else
3434 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003435 }
3436
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003437 /* Free old spare buffer and save old primary buffer as spare */
3438 kfree(thresholds->spare);
3439 thresholds->spare = thresholds->primary;
3440
3441 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003442
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003443 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003444 synchronize_rcu();
3445
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003446unlock:
3447 mutex_unlock(&memcg->thresholds_lock);
3448
3449 return ret;
3450}
3451
Tejun Heo59b6f872013-11-22 18:20:43 -05003452static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003453 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003454{
Tejun Heo59b6f872013-11-22 18:20:43 -05003455 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003456}
3457
Tejun Heo59b6f872013-11-22 18:20:43 -05003458static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003459 struct eventfd_ctx *eventfd, const char *args)
3460{
Tejun Heo59b6f872013-11-22 18:20:43 -05003461 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003462}
3463
Tejun Heo59b6f872013-11-22 18:20:43 -05003464static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003465 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003466{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003467 struct mem_cgroup_thresholds *thresholds;
3468 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003469 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003470 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003471
3472 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003473
3474 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003475 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003476 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003477 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003478 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003479 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003480 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003481 BUG();
3482
Anton Vorontsov371528c2012-02-24 05:14:46 +04003483 if (!thresholds->primary)
3484 goto unlock;
3485
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003486 /* Check if a threshold crossed before removing */
3487 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3488
3489 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003490 size = 0;
3491 for (i = 0; i < thresholds->primary->size; i++) {
3492 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003493 size++;
3494 }
3495
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003496 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003497
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003498 /* Set thresholds array to NULL if we don't have thresholds */
3499 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003500 kfree(new);
3501 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003502 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003503 }
3504
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003505 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003506
3507 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003508 new->current_threshold = -1;
3509 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3510 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003511 continue;
3512
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003513 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003514 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003515 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003516 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003517 * until rcu_assign_pointer(), so it's safe to increment
3518 * it here.
3519 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003520 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003521 }
3522 j++;
3523 }
3524
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003525swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003526 /* Swap primary and spare array */
3527 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003528
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003529 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003530
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003531 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003532 synchronize_rcu();
Martijn Coenen2c641f52016-01-15 16:57:49 -08003533
3534 /* If all events are unregistered, free the spare array */
3535 if (!new) {
3536 kfree(thresholds->spare);
3537 thresholds->spare = NULL;
3538 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04003539unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003540 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003541}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003542
Tejun Heo59b6f872013-11-22 18:20:43 -05003543static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003544 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003545{
Tejun Heo59b6f872013-11-22 18:20:43 -05003546 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003547}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003548
Tejun Heo59b6f872013-11-22 18:20:43 -05003549static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003550 struct eventfd_ctx *eventfd)
3551{
Tejun Heo59b6f872013-11-22 18:20:43 -05003552 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003553}
3554
Tejun Heo59b6f872013-11-22 18:20:43 -05003555static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003556 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003557{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003558 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003559
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003560 event = kmalloc(sizeof(*event), GFP_KERNEL);
3561 if (!event)
3562 return -ENOMEM;
3563
Michal Hocko1af8efe2011-07-26 16:08:24 -07003564 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003565
3566 event->eventfd = eventfd;
3567 list_add(&event->list, &memcg->oom_notify);
3568
3569 /* already in OOM ? */
Tejun Heo47088812015-06-24 16:58:23 -07003570 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003571 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003572 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003573
3574 return 0;
3575}
3576
Tejun Heo59b6f872013-11-22 18:20:43 -05003577static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003578 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003579{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003580 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003581
Michal Hocko1af8efe2011-07-26 16:08:24 -07003582 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003583
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003584 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003585 if (ev->eventfd == eventfd) {
3586 list_del(&ev->list);
3587 kfree(ev);
3588 }
3589 }
3590
Michal Hocko1af8efe2011-07-26 16:08:24 -07003591 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003592}
3593
Tejun Heo2da8ca82013-12-05 12:28:04 -05003594static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003595{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003596 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003597
Tejun Heo791badb2013-12-05 12:28:02 -05003598 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heo47088812015-06-24 16:58:23 -07003599 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003600 return 0;
3601}
3602
Tejun Heo182446d2013-08-08 20:11:24 -04003603static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003604 struct cftype *cft, u64 val)
3605{
Tejun Heo182446d2013-08-08 20:11:24 -04003606 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003607
3608 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003609 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003610 return -EINVAL;
3611
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003612 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003613 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003614 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003615
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003616 return 0;
3617}
3618
Andrew Mortonc255a452012-07-31 16:43:02 -07003619#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03003620static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00003621{
Glauber Costa55007d82012-12-18 14:22:38 -08003622 int ret;
3623
Glauber Costa55007d82012-12-18 14:22:38 -08003624 ret = memcg_propagate_kmem(memcg);
3625 if (ret)
3626 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08003627
Glauber Costa1d62e432012-04-09 19:36:33 -03003628 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07003629}
Glauber Costae5671df2011-12-11 21:47:01 +00003630
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003631static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3632{
Vladimir Davydov2788cf0c2015-02-12 14:59:38 -08003633 struct cgroup_subsys_state *css;
3634 struct mem_cgroup *parent, *child;
3635 int kmemcg_id;
3636
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003637 if (!memcg->kmem_acct_active)
3638 return;
3639
3640 /*
3641 * Clear the 'active' flag before clearing memcg_caches arrays entries.
3642 * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it
3643 * guarantees no cache will be created for this cgroup after we are
3644 * done (see memcg_create_kmem_cache()).
3645 */
3646 memcg->kmem_acct_active = false;
3647
3648 memcg_deactivate_kmem_caches(memcg);
Vladimir Davydov2788cf0c2015-02-12 14:59:38 -08003649
3650 kmemcg_id = memcg->kmemcg_id;
3651 BUG_ON(kmemcg_id < 0);
3652
3653 parent = parent_mem_cgroup(memcg);
3654 if (!parent)
3655 parent = root_mem_cgroup;
3656
3657 /*
3658 * Change kmemcg_id of this cgroup and all its descendants to the
3659 * parent's id, and then move all entries from this cgroup's list_lrus
3660 * to ones of the parent. After we have finished, all list_lrus
3661 * corresponding to this cgroup are guaranteed to remain empty. The
3662 * ordering is imposed by list_lru_node->lock taken by
3663 * memcg_drain_all_list_lrus().
3664 */
3665 css_for_each_descendant_pre(css, &memcg->css) {
3666 child = mem_cgroup_from_css(css);
3667 BUG_ON(child->kmemcg_id != kmemcg_id);
3668 child->kmemcg_id = parent->kmemcg_id;
3669 if (!memcg->use_hierarchy)
3670 break;
3671 }
3672 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
3673
3674 memcg_free_cache_id(kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003675}
3676
Li Zefan10d5ebf2013-07-08 16:00:33 -07003677static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00003678{
Vladimir Davydovf48b80a2015-02-12 14:59:56 -08003679 if (memcg->kmem_acct_activated) {
3680 memcg_destroy_kmem_caches(memcg);
3681 static_key_slow_dec(&memcg_kmem_enabled_key);
3682 WARN_ON(page_counter_read(&memcg->kmem));
3683 }
Glauber Costa1d62e432012-04-09 19:36:33 -03003684 mem_cgroup_sockets_destroy(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07003685}
Glauber Costae5671df2011-12-11 21:47:01 +00003686#else
Glauber Costacbe128e32012-04-09 19:36:34 -03003687static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00003688{
3689 return 0;
3690}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00003691
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003692static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3693{
3694}
3695
Li Zefan10d5ebf2013-07-08 16:00:33 -07003696static void memcg_destroy_kmem(struct mem_cgroup *memcg)
3697{
3698}
Glauber Costae5671df2011-12-11 21:47:01 +00003699#endif
3700
Tejun Heo6b385782015-05-22 17:13:37 -04003701#ifdef CONFIG_CGROUP_WRITEBACK
3702
3703struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3704{
3705 return &memcg->cgwb_list;
3706}
3707
Tejun Heo8c93c2f2015-05-22 18:23:33 -04003708static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3709{
3710 return wb_domain_init(&memcg->cgwb_domain, gfp);
3711}
3712
3713static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3714{
3715 wb_domain_exit(&memcg->cgwb_domain);
3716}
3717
Tejun Heoe65e8b62015-05-22 18:23:34 -04003718static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3719{
3720 wb_domain_size_changed(&memcg->cgwb_domain);
3721}
3722
Tejun Heo8c93c2f2015-05-22 18:23:33 -04003723struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3724{
3725 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3726
3727 if (!memcg->css.parent)
3728 return NULL;
3729
3730 return &memcg->cgwb_domain;
3731}
3732
Tejun Heo6a139c82015-05-22 18:23:35 -04003733/**
3734 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3735 * @wb: bdi_writeback in question
Tejun Heo8debf2d2015-09-29 13:04:26 -04003736 * @pfilepages: out parameter for number of file pages
3737 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heo6a139c82015-05-22 18:23:35 -04003738 * @pdirty: out parameter for number of dirty pages
3739 * @pwriteback: out parameter for number of pages under writeback
3740 *
Tejun Heo8debf2d2015-09-29 13:04:26 -04003741 * Determine the numbers of file, headroom, dirty, and writeback pages in
3742 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3743 * is a bit more involved.
Tejun Heo6a139c82015-05-22 18:23:35 -04003744 *
Tejun Heo8debf2d2015-09-29 13:04:26 -04003745 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3746 * headroom is calculated as the lowest headroom of itself and the
3747 * ancestors. Note that this doesn't consider the actual amount of
3748 * available memory in the system. The caller should further cap
3749 * *@pheadroom accordingly.
Tejun Heo6a139c82015-05-22 18:23:35 -04003750 */
Tejun Heo8debf2d2015-09-29 13:04:26 -04003751void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3752 unsigned long *pheadroom, unsigned long *pdirty,
3753 unsigned long *pwriteback)
Tejun Heo6a139c82015-05-22 18:23:35 -04003754{
3755 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3756 struct mem_cgroup *parent;
Tejun Heo6a139c82015-05-22 18:23:35 -04003757
3758 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3759
3760 /* this should eventually include NR_UNSTABLE_NFS */
3761 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heo8debf2d2015-09-29 13:04:26 -04003762 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3763 (1 << LRU_ACTIVE_FILE));
3764 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heo6a139c82015-05-22 18:23:35 -04003765
Tejun Heo6a139c82015-05-22 18:23:35 -04003766 while ((parent = parent_mem_cgroup(memcg))) {
3767 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3768 unsigned long used = page_counter_read(&memcg->memory);
3769
Tejun Heo8debf2d2015-09-29 13:04:26 -04003770 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heo6a139c82015-05-22 18:23:35 -04003771 memcg = parent;
3772 }
Tejun Heo6a139c82015-05-22 18:23:35 -04003773}
3774
Tejun Heo8c93c2f2015-05-22 18:23:33 -04003775#else /* CONFIG_CGROUP_WRITEBACK */
3776
3777static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3778{
3779 return 0;
3780}
3781
3782static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3783{
3784}
3785
Tejun Heoe65e8b62015-05-22 18:23:34 -04003786static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3787{
3788}
3789
Tejun Heo6b385782015-05-22 17:13:37 -04003790#endif /* CONFIG_CGROUP_WRITEBACK */
3791
Tejun Heo79bd9812013-11-22 18:20:42 -05003792/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003793 * DO NOT USE IN NEW FILES.
3794 *
3795 * "cgroup.event_control" implementation.
3796 *
3797 * This is way over-engineered. It tries to support fully configurable
3798 * events for each user. Such level of flexibility is completely
3799 * unnecessary especially in the light of the planned unified hierarchy.
3800 *
3801 * Please deprecate this and replace with something simpler if at all
3802 * possible.
3803 */
3804
3805/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003806 * Unregister event and free resources.
3807 *
3808 * Gets called from workqueue.
3809 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003810static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003811{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003812 struct mem_cgroup_event *event =
3813 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003814 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003815
3816 remove_wait_queue(event->wqh, &event->wait);
3817
Tejun Heo59b6f872013-11-22 18:20:43 -05003818 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003819
3820 /* Notify userspace the event is going away. */
3821 eventfd_signal(event->eventfd, 1);
3822
3823 eventfd_ctx_put(event->eventfd);
3824 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003825 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003826}
3827
3828/*
3829 * Gets called on POLLHUP on eventfd when user closes it.
3830 *
3831 * Called with wqh->lock held and interrupts disabled.
3832 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003833static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3834 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003835{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003836 struct mem_cgroup_event *event =
3837 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003838 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003839 unsigned long flags = (unsigned long)key;
3840
3841 if (flags & POLLHUP) {
3842 /*
3843 * If the event has been detached at cgroup removal, we
3844 * can simply return knowing the other side will cleanup
3845 * for us.
3846 *
3847 * We can't race against event freeing since the other
3848 * side will require wqh->lock via remove_wait_queue(),
3849 * which we hold.
3850 */
Tejun Heofba94802013-11-22 18:20:43 -05003851 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003852 if (!list_empty(&event->list)) {
3853 list_del_init(&event->list);
3854 /*
3855 * We are in atomic context, but cgroup_event_remove()
3856 * may sleep, so we have to call it in workqueue.
3857 */
3858 schedule_work(&event->remove);
3859 }
Tejun Heofba94802013-11-22 18:20:43 -05003860 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003861 }
3862
3863 return 0;
3864}
3865
Tejun Heo3bc942f2013-11-22 18:20:44 -05003866static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003867 wait_queue_head_t *wqh, poll_table *pt)
3868{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003869 struct mem_cgroup_event *event =
3870 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003871
3872 event->wqh = wqh;
3873 add_wait_queue(wqh, &event->wait);
3874}
3875
3876/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003877 * DO NOT USE IN NEW FILES.
3878 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003879 * Parse input and register new cgroup event handler.
3880 *
3881 * Input must be in format '<event_fd> <control_fd> <args>'.
3882 * Interpretation of args is defined by control file implementation.
3883 */
Tejun Heo451af502014-05-13 12:16:21 -04003884static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3885 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003886{
Tejun Heo451af502014-05-13 12:16:21 -04003887 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003888 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003889 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003890 struct cgroup_subsys_state *cfile_css;
3891 unsigned int efd, cfd;
3892 struct fd efile;
3893 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003894 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003895 char *endp;
3896 int ret;
3897
Tejun Heo451af502014-05-13 12:16:21 -04003898 buf = strstrip(buf);
3899
3900 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003901 if (*endp != ' ')
3902 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003903 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003904
Tejun Heo451af502014-05-13 12:16:21 -04003905 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003906 if ((*endp != ' ') && (*endp != '\0'))
3907 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003908 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003909
3910 event = kzalloc(sizeof(*event), GFP_KERNEL);
3911 if (!event)
3912 return -ENOMEM;
3913
Tejun Heo59b6f872013-11-22 18:20:43 -05003914 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003915 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003916 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3917 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3918 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003919
3920 efile = fdget(efd);
3921 if (!efile.file) {
3922 ret = -EBADF;
3923 goto out_kfree;
3924 }
3925
3926 event->eventfd = eventfd_ctx_fileget(efile.file);
3927 if (IS_ERR(event->eventfd)) {
3928 ret = PTR_ERR(event->eventfd);
3929 goto out_put_efile;
3930 }
3931
3932 cfile = fdget(cfd);
3933 if (!cfile.file) {
3934 ret = -EBADF;
3935 goto out_put_eventfd;
3936 }
3937
3938 /* the process need read permission on control file */
3939 /* AV: shouldn't we check that it's been opened for read instead? */
3940 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3941 if (ret < 0)
3942 goto out_put_cfile;
3943
Tejun Heo79bd9812013-11-22 18:20:42 -05003944 /*
Tejun Heofba94802013-11-22 18:20:43 -05003945 * Determine the event callbacks and set them in @event. This used
3946 * to be done via struct cftype but cgroup core no longer knows
3947 * about these events. The following is crude but the whole thing
3948 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003949 *
3950 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003951 */
Al Virob5830432014-10-31 01:22:04 -04003952 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003953
3954 if (!strcmp(name, "memory.usage_in_bytes")) {
3955 event->register_event = mem_cgroup_usage_register_event;
3956 event->unregister_event = mem_cgroup_usage_unregister_event;
3957 } else if (!strcmp(name, "memory.oom_control")) {
3958 event->register_event = mem_cgroup_oom_register_event;
3959 event->unregister_event = mem_cgroup_oom_unregister_event;
3960 } else if (!strcmp(name, "memory.pressure_level")) {
3961 event->register_event = vmpressure_register_event;
3962 event->unregister_event = vmpressure_unregister_event;
3963 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003964 event->register_event = memsw_cgroup_usage_register_event;
3965 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003966 } else {
3967 ret = -EINVAL;
3968 goto out_put_cfile;
3969 }
3970
3971 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003972 * Verify @cfile should belong to @css. Also, remaining events are
3973 * automatically removed on cgroup destruction but the removal is
3974 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003975 */
Al Virob5830432014-10-31 01:22:04 -04003976 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003977 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003978 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003979 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003980 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003981 if (cfile_css != css) {
3982 css_put(cfile_css);
3983 goto out_put_cfile;
3984 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003985
Tejun Heo451af502014-05-13 12:16:21 -04003986 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003987 if (ret)
3988 goto out_put_css;
3989
3990 efile.file->f_op->poll(efile.file, &event->pt);
3991
Tejun Heofba94802013-11-22 18:20:43 -05003992 spin_lock(&memcg->event_list_lock);
3993 list_add(&event->list, &memcg->event_list);
3994 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003995
3996 fdput(cfile);
3997 fdput(efile);
3998
Tejun Heo451af502014-05-13 12:16:21 -04003999 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05004000
4001out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05004002 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05004003out_put_cfile:
4004 fdput(cfile);
4005out_put_eventfd:
4006 eventfd_ctx_put(event->eventfd);
4007out_put_efile:
4008 fdput(efile);
4009out_kfree:
4010 kfree(event);
4011
4012 return ret;
4013}
4014
Johannes Weiner241994ed2015-02-11 15:26:06 -08004015static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004016 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004017 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004018 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004019 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004020 },
4021 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004022 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004023 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004024 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004025 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004026 },
4027 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004028 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004029 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004030 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004031 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004032 },
4033 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004034 .name = "soft_limit_in_bytes",
4035 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004036 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004037 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07004038 },
4039 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004040 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004041 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004042 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004043 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004044 },
Balbir Singh8697d332008-02-07 00:13:59 -08004045 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004046 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004047 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004048 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004049 {
4050 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04004051 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004052 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004053 {
4054 .name = "use_hierarchy",
4055 .write_u64 = mem_cgroup_hierarchy_write,
4056 .read_u64 = mem_cgroup_hierarchy_read,
4057 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004058 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05004059 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04004060 .write = memcg_write_event_control,
Tejun Heo6c106f52015-09-18 17:54:23 -04004061 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05004062 },
4063 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004064 .name = "swappiness",
4065 .read_u64 = mem_cgroup_swappiness_read,
4066 .write_u64 = mem_cgroup_swappiness_write,
4067 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004068 {
4069 .name = "move_charge_at_immigrate",
4070 .read_u64 = mem_cgroup_move_charge_read,
4071 .write_u64 = mem_cgroup_move_charge_write,
4072 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004073 {
4074 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004075 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004076 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004077 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4078 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004079 {
4080 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004081 },
Ying Han406eb0c2011-05-26 16:25:37 -07004082#ifdef CONFIG_NUMA
4083 {
4084 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004085 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07004086 },
4087#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004088#ifdef CONFIG_MEMCG_KMEM
4089 {
4090 .name = "kmem.limit_in_bytes",
4091 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004092 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004093 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004094 },
4095 {
4096 .name = "kmem.usage_in_bytes",
4097 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004098 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004099 },
4100 {
4101 .name = "kmem.failcnt",
4102 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004103 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004104 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004105 },
4106 {
4107 .name = "kmem.max_usage_in_bytes",
4108 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004109 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004110 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004111 },
Glauber Costa749c5412012-12-18 14:23:01 -08004112#ifdef CONFIG_SLABINFO
4113 {
4114 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004115 .seq_start = slab_start,
4116 .seq_next = slab_next,
4117 .seq_stop = slab_stop,
4118 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004119 },
4120#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004121#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07004122 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004123};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004124
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004125static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004126{
4127 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004128 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004129 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004130 /*
4131 * This routine is called against possible nodes.
4132 * But it's BUG to call kmalloc() against offline node.
4133 *
4134 * TODO: this routine can waste much memory for nodes which will
4135 * never be onlined. It's better to use memory hotplug callback
4136 * function.
4137 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004138 if (!node_state(node, N_NORMAL_MEMORY))
4139 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004140 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004141 if (!pn)
4142 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004143
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004144 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4145 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004146 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07004147 mz->usage_in_excess = 0;
4148 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004149 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004150 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004151 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004152 return 0;
4153}
4154
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004155static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004156{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004157 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004158}
4159
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004160static struct mem_cgroup *mem_cgroup_alloc(void)
4161{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004162 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004163 size_t size;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004164
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004165 size = sizeof(struct mem_cgroup);
4166 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004167
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004168 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004169 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004170 return NULL;
4171
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004172 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4173 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004174 goto out_free;
Tejun Heo8c93c2f2015-05-22 18:23:33 -04004175
4176 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4177 goto out_free_stat;
4178
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004179 spin_lock_init(&memcg->pcp_counter_lock);
4180 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004181
Tejun Heo8c93c2f2015-05-22 18:23:33 -04004182out_free_stat:
4183 free_percpu(memcg->stat);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004184out_free:
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004185 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004186 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004187}
4188
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004189/*
Glauber Costac8b2a362012-12-18 14:22:13 -08004190 * At destroying mem_cgroup, references from swap_cgroup can remain.
4191 * (scanning all at force_empty is too costly...)
4192 *
4193 * Instead of clearing all references at force_empty, we remember
4194 * the number of reference from swap_cgroup and free mem_cgroup when
4195 * it goes down to 0.
4196 *
4197 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07004198 */
Glauber Costac8b2a362012-12-18 14:22:13 -08004199
4200static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07004201{
Glauber Costac8b2a362012-12-18 14:22:13 -08004202 int node;
Hugh Dickins59927fb2012-03-15 15:17:07 -07004203
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07004204 mem_cgroup_remove_from_trees(memcg);
Glauber Costac8b2a362012-12-18 14:22:13 -08004205
4206 for_each_node(node)
4207 free_mem_cgroup_per_zone_info(memcg, node);
4208
4209 free_percpu(memcg->stat);
Tejun Heo8c93c2f2015-05-22 18:23:33 -04004210 memcg_wb_domain_exit(memcg);
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004211 kfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07004212}
Glauber Costa3afe36b2012-05-29 15:07:10 -07004213
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004214/*
4215 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4216 */
Glauber Costae1aab162011-12-11 21:47:03 +00004217struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004218{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004219 if (!memcg->memory.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004220 return NULL;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004221 return mem_cgroup_from_counter(memcg->memory.parent, memory);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004222}
Glauber Costae1aab162011-12-11 21:47:03 +00004223EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004224
Li Zefan0eb253e2009-01-15 13:51:25 -08004225static struct cgroup_subsys_state * __ref
Tejun Heoeb954192013-08-08 20:11:23 -04004226mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004227{
Glauber Costad142e3e2013-02-22 16:34:52 -08004228 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004229 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004230 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004231
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004232 memcg = mem_cgroup_alloc();
4233 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004234 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004235
Bob Liu3ed28fa2012-01-12 17:19:04 -08004236 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004237 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004238 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004239
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004240 /* root ? */
Tejun Heoeb954192013-08-08 20:11:23 -04004241 if (parent_css == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08004242 root_mem_cgroup = memcg;
Tejun Heodbee2272015-05-22 17:13:20 -04004243 mem_cgroup_root_css = &memcg->css;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004244 page_counter_init(&memcg->memory, NULL);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004245 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004246 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004247 page_counter_init(&memcg->memsw, NULL);
4248 page_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004249 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004250
Glauber Costad142e3e2013-02-22 16:34:52 -08004251 memcg->last_scanned_node = MAX_NUMNODES;
4252 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004253 memcg->move_charge_at_immigrate = 0;
4254 mutex_init(&memcg->thresholds_lock);
4255 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004256 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004257 INIT_LIST_HEAD(&memcg->event_list);
4258 spin_lock_init(&memcg->event_list_lock);
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004259#ifdef CONFIG_MEMCG_KMEM
4260 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004261#endif
Tejun Heo6b385782015-05-22 17:13:37 -04004262#ifdef CONFIG_CGROUP_WRITEBACK
4263 INIT_LIST_HEAD(&memcg->cgwb_list);
4264#endif
Glauber Costad142e3e2013-02-22 16:34:52 -08004265 return &memcg->css;
4266
4267free_out:
4268 __mem_cgroup_free(memcg);
4269 return ERR_PTR(error);
4270}
4271
4272static int
Tejun Heoeb954192013-08-08 20:11:23 -04004273mem_cgroup_css_online(struct cgroup_subsys_state *css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004274{
Tejun Heoeb954192013-08-08 20:11:23 -04004275 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04004276 struct mem_cgroup *parent = mem_cgroup_from_css(css->parent);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004277 int ret;
Glauber Costad142e3e2013-02-22 16:34:52 -08004278
Tejun Heo15a4c832014-05-04 15:09:14 -04004279 if (css->id > MEM_CGROUP_ID_MAX)
Li Zefan4219b2d2013-09-23 16:56:29 +08004280 return -ENOSPC;
4281
Tejun Heo63876982013-08-08 20:11:23 -04004282 if (!parent)
Glauber Costad142e3e2013-02-22 16:34:52 -08004283 return 0;
4284
Glauber Costa09998212013-02-22 16:34:55 -08004285 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08004286
4287 memcg->use_hierarchy = parent->use_hierarchy;
4288 memcg->oom_kill_disable = parent->oom_kill_disable;
4289 memcg->swappiness = mem_cgroup_swappiness(parent);
4290
4291 if (parent->use_hierarchy) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004292 page_counter_init(&memcg->memory, &parent->memory);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004293 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004294 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004295 page_counter_init(&memcg->memsw, &parent->memsw);
4296 page_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08004297
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004298 /*
Li Zefan8d76a972013-07-08 16:00:36 -07004299 * No need to take a reference to the parent because cgroup
4300 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004301 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08004302 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004303 page_counter_init(&memcg->memory, NULL);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004304 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004305 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004306 page_counter_init(&memcg->memsw, NULL);
4307 page_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004308 /*
4309 * Deeper hierachy with use_hierarchy == false doesn't make
4310 * much sense so let cgroup subsystem know about this
4311 * unfortunate state in our controller.
4312 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004313 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004314 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004315 }
Glauber Costa09998212013-02-22 16:34:55 -08004316 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08004317
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004318 ret = memcg_init_kmem(memcg, &memory_cgrp_subsys);
4319 if (ret)
4320 return ret;
4321
4322 /*
4323 * Make sure the memcg is initialized: mem_cgroup_iter()
4324 * orders reading memcg->initialized against its callers
4325 * reading the memcg members.
4326 */
4327 smp_store_release(&memcg->initialized, 1);
4328
4329 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004330}
4331
Tejun Heoeb954192013-08-08 20:11:23 -04004332static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004333{
Tejun Heoeb954192013-08-08 20:11:23 -04004334 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004335 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004336
4337 /*
4338 * Unregister events and notify userspace.
4339 * Notify userspace about cgroup removing only after rmdir of cgroup
4340 * directory to avoid race between userspace and kernelspace.
4341 */
Tejun Heofba94802013-11-22 18:20:43 -05004342 spin_lock(&memcg->event_list_lock);
4343 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004344 list_del_init(&event->list);
4345 schedule_work(&event->remove);
4346 }
Tejun Heofba94802013-11-22 18:20:43 -05004347 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004348
Michal Hocko33cb8762013-07-31 13:53:51 -07004349 vmpressure_cleanup(&memcg->vmpressure);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08004350
4351 memcg_deactivate_kmem(memcg);
Tejun Heo6b385782015-05-22 17:13:37 -04004352
4353 wb_memcg_offline(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004354}
4355
Tejun Heoeb954192013-08-08 20:11:23 -04004356static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004357{
Tejun Heoeb954192013-08-08 20:11:23 -04004358 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004359
Li Zefan10d5ebf2013-07-08 16:00:33 -07004360 memcg_destroy_kmem(memcg);
Li Zefan465939a2013-07-08 16:00:38 -07004361 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004362}
4363
Tejun Heo1ced9532014-07-08 18:02:57 -04004364/**
4365 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4366 * @css: the target css
4367 *
4368 * Reset the states of the mem_cgroup associated with @css. This is
4369 * invoked when the userland requests disabling on the default hierarchy
4370 * but the memcg is pinned through dependency. The memcg should stop
4371 * applying policies and should revert to the vanilla state as it may be
4372 * made visible again.
4373 *
4374 * The current implementation only resets the essential configurations.
4375 * This needs to be expanded to cover all the visible parts.
4376 */
4377static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4378{
4379 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4380
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004381 mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX);
4382 mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX);
4383 memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004384 memcg->low = 0;
4385 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004386 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heoe65e8b62015-05-22 18:23:34 -04004387 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004388}
4389
Daisuke Nishimura02491442010-03-10 15:22:17 -08004390#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004391/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004392static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004393{
Johannes Weiner05b84302014-08-06 16:05:59 -07004394 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004395
4396 /* Try a single bulk charge without reclaim first */
Johannes Weiner00501b52014-08-08 14:19:20 -07004397 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004398 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004399 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004400 return ret;
4401 }
Johannes Weiner692e7c42014-08-06 16:05:57 -07004402 if (ret == -EINTR) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004403 cancel_charge(root_mem_cgroup, count);
Johannes Weiner692e7c42014-08-06 16:05:57 -07004404 return ret;
4405 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004406
4407 /* Try charges one by one with reclaim */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004408 while (count--) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004409 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
Johannes Weiner9476db92014-08-06 16:05:55 -07004410 /*
4411 * In case of failure, any residual charges against
4412 * mc.to will be dropped by mem_cgroup_clear_mc()
Johannes Weiner692e7c42014-08-06 16:05:57 -07004413 * later on. However, cancel any charges that are
4414 * bypassed to root right away or they'll be lost.
Johannes Weiner9476db92014-08-06 16:05:55 -07004415 */
Johannes Weiner692e7c42014-08-06 16:05:57 -07004416 if (ret == -EINTR)
Johannes Weiner00501b52014-08-08 14:19:20 -07004417 cancel_charge(root_mem_cgroup, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004418 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004419 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004420 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004421 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004422 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004423 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004424}
4425
4426/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004427 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004428 * @vma: the vma the pte to be checked belongs
4429 * @addr: the address corresponding to the pte to be checked
4430 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08004431 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004432 *
4433 * Returns
4434 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4435 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4436 * move charge. if @target is not NULL, the page is stored in target->page
4437 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08004438 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4439 * target for charge migration. if @target is not NULL, the entry is stored
4440 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004441 *
4442 * Called with pte lock held.
4443 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004444union mc_target {
4445 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004446 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004447};
4448
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004449enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004450 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004451 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004452 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004453};
4454
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004455static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4456 unsigned long addr, pte_t ptent)
4457{
4458 struct page *page = vm_normal_page(vma, addr, ptent);
4459
4460 if (!page || !page_mapped(page))
4461 return NULL;
4462 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004463 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004464 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004465 } else {
4466 if (!(mc.flags & MOVE_FILE))
4467 return NULL;
4468 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004469 if (!get_page_unless_zero(page))
4470 return NULL;
4471
4472 return page;
4473}
4474
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004475#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004476static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4477 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4478{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004479 struct page *page = NULL;
4480 swp_entry_t ent = pte_to_swp_entry(ptent);
4481
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004482 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004483 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004484 /*
4485 * Because lookup_swap_cache() updates some statistics counter,
4486 * we call find_get_page() with swapper_space directly.
4487 */
Shaohua Li33806f02013-02-22 16:34:37 -08004488 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004489 if (do_swap_account)
4490 entry->val = ent.val;
4491
4492 return page;
4493}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004494#else
4495static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4496 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4497{
4498 return NULL;
4499}
4500#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004501
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004502static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4503 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4504{
4505 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004506 struct address_space *mapping;
4507 pgoff_t pgoff;
4508
4509 if (!vma->vm_file) /* anonymous vma */
4510 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004511 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004512 return NULL;
4513
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004514 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004515 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004516
4517 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004518#ifdef CONFIG_SWAP
4519 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004520 if (shmem_mapping(mapping)) {
4521 page = find_get_entry(mapping, pgoff);
4522 if (radix_tree_exceptional_entry(page)) {
4523 swp_entry_t swp = radix_to_swp_entry(page);
4524 if (do_swap_account)
4525 *entry = swp;
4526 page = find_get_page(swap_address_space(swp), swp.val);
4527 }
4528 } else
4529 page = find_get_page(mapping, pgoff);
4530#else
4531 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004532#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004533 return page;
4534}
4535
Chen Gangb1b0dea2015-04-14 15:47:35 -07004536/**
4537 * mem_cgroup_move_account - move account of the page
4538 * @page: the page
4539 * @nr_pages: number of regular pages (>1 for huge pages)
4540 * @from: mem_cgroup which the page is moved from.
4541 * @to: mem_cgroup which the page is moved to. @from != @to.
4542 *
4543 * The caller must confirm following.
4544 * - page is not on LRU (isolate_page() is useful.)
4545 * - compound_lock is held when nr_pages > 1
4546 *
4547 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4548 * from old cgroup.
4549 */
4550static int mem_cgroup_move_account(struct page *page,
4551 unsigned int nr_pages,
4552 struct mem_cgroup *from,
4553 struct mem_cgroup *to)
4554{
4555 unsigned long flags;
4556 int ret;
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04004557 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004558
4559 VM_BUG_ON(from == to);
4560 VM_BUG_ON_PAGE(PageLRU(page), page);
4561 /*
4562 * The page is isolated from LRU. So, collapse function
4563 * will not handle this page. But page splitting can happen.
4564 * Do this check under compound_page_lock(). The caller should
4565 * hold it.
4566 */
4567 ret = -EBUSY;
4568 if (nr_pages > 1 && !PageTransHuge(page))
4569 goto out;
4570
4571 /*
4572 * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup
4573 * of its source page while we change it: page migration takes
4574 * both pages off the LRU, but page cache replacement doesn't.
4575 */
4576 if (!trylock_page(page))
4577 goto out;
4578
4579 ret = -EINVAL;
4580 if (page->mem_cgroup != from)
4581 goto out_unlock;
4582
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04004583 anon = PageAnon(page);
4584
Chen Gangb1b0dea2015-04-14 15:47:35 -07004585 spin_lock_irqsave(&from->move_lock, flags);
4586
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04004587 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004588 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4589 nr_pages);
4590 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4591 nr_pages);
4592 }
4593
Greg Thelen7c9d3ff2015-05-22 17:13:16 -04004594 /*
4595 * move_lock grabbed above and caller set from->moving_account, so
4596 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4597 * So mapping should be stable for dirty pages.
4598 */
4599 if (!anon && PageDirty(page)) {
4600 struct address_space *mapping = page_mapping(page);
4601
4602 if (mapping_cap_account_dirty(mapping)) {
4603 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4604 nr_pages);
4605 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4606 nr_pages);
4607 }
4608 }
4609
Chen Gangb1b0dea2015-04-14 15:47:35 -07004610 if (PageWriteback(page)) {
4611 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4612 nr_pages);
4613 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4614 nr_pages);
4615 }
4616
4617 /*
4618 * It is safe to change page->mem_cgroup here because the page
4619 * is referenced, charged, and isolated - we can't race with
4620 * uncharging, charging, migration, or LRU putback.
4621 */
4622
4623 /* caller should have done css_get */
4624 page->mem_cgroup = to;
4625 spin_unlock_irqrestore(&from->move_lock, flags);
4626
4627 ret = 0;
4628
4629 local_irq_disable();
4630 mem_cgroup_charge_statistics(to, page, nr_pages);
4631 memcg_check_events(to, page);
4632 mem_cgroup_charge_statistics(from, page, -nr_pages);
4633 memcg_check_events(from, page);
4634 local_irq_enable();
4635out_unlock:
4636 unlock_page(page);
4637out:
4638 return ret;
4639}
4640
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004641static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004642 unsigned long addr, pte_t ptent, union mc_target *target)
4643{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004644 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004645 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004646 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004647
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004648 if (pte_present(ptent))
4649 page = mc_handle_present_pte(vma, addr, ptent);
4650 else if (is_swap_pte(ptent))
4651 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004652 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004653 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004654
4655 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004656 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004657 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004658 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004659 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004660 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004661 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004662 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004663 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004664 ret = MC_TARGET_PAGE;
4665 if (target)
4666 target->page = page;
4667 }
4668 if (!ret || !target)
4669 put_page(page);
4670 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004671 /* There is a swap entry and a page doesn't exist or isn't charged */
4672 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004673 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004674 ret = MC_TARGET_SWAP;
4675 if (target)
4676 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004677 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004678 return ret;
4679}
4680
Naoya Horiguchi12724852012-03-21 16:34:28 -07004681#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4682/*
4683 * We don't consider swapping or file mapped pages because THP does not
4684 * support them for now.
4685 * Caller should make sure that pmd_trans_huge(pmd) is true.
4686 */
4687static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4688 unsigned long addr, pmd_t pmd, union mc_target *target)
4689{
4690 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004691 enum mc_target_type ret = MC_TARGET_NONE;
4692
4693 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004694 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004695 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004696 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004697 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004698 ret = MC_TARGET_PAGE;
4699 if (target) {
4700 get_page(page);
4701 target->page = page;
4702 }
4703 }
4704 return ret;
4705}
4706#else
4707static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4708 unsigned long addr, pmd_t pmd, union mc_target *target)
4709{
4710 return MC_TARGET_NONE;
4711}
4712#endif
4713
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004714static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4715 unsigned long addr, unsigned long end,
4716 struct mm_walk *walk)
4717{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004718 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004719 pte_t *pte;
4720 spinlock_t *ptl;
4721
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004722 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004723 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4724 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004725 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004726 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004727 }
Dave Hansen03319322011-03-22 16:32:56 -07004728
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004729 if (pmd_trans_unstable(pmd))
4730 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004731 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4732 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004733 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004734 mc.precharge++; /* increment precharge temporarily */
4735 pte_unmap_unlock(pte - 1, ptl);
4736 cond_resched();
4737
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004738 return 0;
4739}
4740
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004741static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4742{
4743 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004744
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004745 struct mm_walk mem_cgroup_count_precharge_walk = {
4746 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4747 .mm = mm,
4748 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004749 down_read(&mm->mmap_sem);
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004750 walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004751 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004752
4753 precharge = mc.precharge;
4754 mc.precharge = 0;
4755
4756 return precharge;
4757}
4758
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004759static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4760{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004761 unsigned long precharge = mem_cgroup_count_precharge(mm);
4762
4763 VM_BUG_ON(mc.moving_task);
4764 mc.moving_task = current;
4765 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004766}
4767
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004768/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4769static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004770{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004771 struct mem_cgroup *from = mc.from;
4772 struct mem_cgroup *to = mc.to;
4773
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004774 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004775 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004776 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004777 mc.precharge = 0;
4778 }
4779 /*
4780 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4781 * we must uncharge here.
4782 */
4783 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004784 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004785 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004786 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004787 /* we must fixup refcnts and charges */
4788 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004789 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004790 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004791 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004792
Johannes Weiner05b84302014-08-06 16:05:59 -07004793 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004794 * we charged both to->memory and to->memsw, so we
4795 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004796 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004797 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004798 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004799
Johannes Weinere8ea14c2014-12-10 15:42:42 -08004800 css_put_many(&mc.from->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004801
Li Zefan40503772013-07-08 16:00:34 -07004802 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004803 mc.moved_swap = 0;
4804 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004805 memcg_oom_recover(from);
4806 memcg_oom_recover(to);
4807 wake_up_all(&mc.waitq);
4808}
4809
4810static void mem_cgroup_clear_mc(void)
4811{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004812 /*
4813 * we must clear moving_task before waking up waiters at the end of
4814 * task migration.
4815 */
4816 mc.moving_task = NULL;
4817 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004818 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004819 mc.from = NULL;
4820 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004821 spin_unlock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004822}
4823
Tejun Heoeb954192013-08-08 20:11:23 -04004824static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08004825 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004826{
Tejun Heoeb954192013-08-08 20:11:23 -04004827 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo67d6b192015-09-08 15:01:10 -07004828 struct mem_cgroup *from;
4829 struct task_struct *p;
4830 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004831 unsigned long move_flags;
Tejun Heo67d6b192015-09-08 15:01:10 -07004832 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004833
Glauber Costaee5e8472013-02-22 16:34:50 -08004834 /*
4835 * We are now commited to this value whatever it is. Changes in this
4836 * tunable will only affect upcoming migrations, not the current one.
4837 * So we need to save it, and keep it going.
4838 */
Jason Low4db0c3c2015-04-15 16:14:08 -07004839 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
Tejun Heo67d6b192015-09-08 15:01:10 -07004840 if (!move_flags)
4841 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004842
Tejun Heo67d6b192015-09-08 15:01:10 -07004843 p = cgroup_taskset_first(tset);
4844 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004845
Tejun Heo67d6b192015-09-08 15:01:10 -07004846 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004847
Tejun Heo67d6b192015-09-08 15:01:10 -07004848 mm = get_task_mm(p);
4849 if (!mm)
4850 return 0;
4851 /* We move charges only when we move a owner of the mm */
4852 if (mm->owner == p) {
4853 VM_BUG_ON(mc.from);
4854 VM_BUG_ON(mc.to);
4855 VM_BUG_ON(mc.precharge);
4856 VM_BUG_ON(mc.moved_charge);
4857 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004858
Tejun Heo67d6b192015-09-08 15:01:10 -07004859 spin_lock(&mc.lock);
4860 mc.from = from;
4861 mc.to = memcg;
4862 mc.flags = move_flags;
4863 spin_unlock(&mc.lock);
4864 /* We set mc.moving_task later */
4865
4866 ret = mem_cgroup_precharge_mc(mm);
4867 if (ret)
4868 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004869 }
Tejun Heo67d6b192015-09-08 15:01:10 -07004870 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004871 return ret;
4872}
4873
Tejun Heoeb954192013-08-08 20:11:23 -04004874static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08004875 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004876{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004877 if (mc.to)
4878 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004879}
4880
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004881static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4882 unsigned long addr, unsigned long end,
4883 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004884{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004885 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004886 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004887 pte_t *pte;
4888 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004889 enum mc_target_type target_type;
4890 union mc_target target;
4891 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004892
Naoya Horiguchi12724852012-03-21 16:34:28 -07004893 /*
4894 * We don't take compound_lock() here but no race with splitting thp
4895 * happens because:
4896 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
4897 * under splitting, which means there's no concurrent thp split,
4898 * - if another thread runs into split_huge_page() just after we
4899 * entered this if-block, the thread must wait for page table lock
4900 * to be unlocked in __split_huge_page_splitting(), where the main
4901 * part of thp split is not executed yet.
4902 */
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004903 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004904 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004905 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004906 return 0;
4907 }
4908 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4909 if (target_type == MC_TARGET_PAGE) {
4910 page = target.page;
4911 if (!isolate_lru_page(page)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004912 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
Johannes Weiner1306a852014-12-10 15:44:52 -08004913 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004914 mc.precharge -= HPAGE_PMD_NR;
4915 mc.moved_charge += HPAGE_PMD_NR;
4916 }
4917 putback_lru_page(page);
4918 }
4919 put_page(page);
4920 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004921 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004922 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004923 }
4924
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004925 if (pmd_trans_unstable(pmd))
4926 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004927retry:
4928 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4929 for (; addr != end; addr += PAGE_SIZE) {
4930 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004931 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004932
4933 if (!mc.precharge)
4934 break;
4935
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004936 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004937 case MC_TARGET_PAGE:
4938 page = target.page;
4939 if (isolate_lru_page(page))
4940 goto put;
Johannes Weiner1306a852014-12-10 15:44:52 -08004941 if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004942 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004943 /* we uncharge from mc.from later. */
4944 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004945 }
4946 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004947put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004948 put_page(page);
4949 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004950 case MC_TARGET_SWAP:
4951 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004952 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004953 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004954 /* we fixup refcnts and charges later. */
4955 mc.moved_swap++;
4956 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004957 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004958 default:
4959 break;
4960 }
4961 }
4962 pte_unmap_unlock(pte - 1, ptl);
4963 cond_resched();
4964
4965 if (addr != end) {
4966 /*
4967 * We have consumed all precharges we got in can_attach().
4968 * We try charge one by one, but don't do any additional
4969 * charges to mc.to if we have failed in charge once in attach()
4970 * phase.
4971 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004972 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004973 if (!ret)
4974 goto retry;
4975 }
4976
4977 return ret;
4978}
4979
4980static void mem_cgroup_move_charge(struct mm_struct *mm)
4981{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004982 struct mm_walk mem_cgroup_move_charge_walk = {
4983 .pmd_entry = mem_cgroup_move_charge_pte_range,
4984 .mm = mm,
4985 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004986
4987 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004988 /*
4989 * Signal mem_cgroup_begin_page_stat() to take the memcg's
4990 * move_lock while we're moving its pages to another memcg.
4991 * Then wait for already started RCU-only updates to finish.
4992 */
4993 atomic_inc(&mc.from->moving_account);
4994 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004995retry:
4996 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
4997 /*
4998 * Someone who are holding the mmap_sem might be waiting in
4999 * waitq. So we cancel all extra charges, wake up all waiters,
5000 * and retry. Because we cancel precharges, we might not be able
5001 * to move enough charges, but moving charge is a best-effort
5002 * feature anyway, so it wouldn't be a big problem.
5003 */
5004 __mem_cgroup_clear_mc();
5005 cond_resched();
5006 goto retry;
5007 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005008 /*
5009 * When we have consumed all precharges and failed in doing
5010 * additional charge, the page walk just aborts.
5011 */
5012 walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005013 up_read(&mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08005014 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005015}
5016
Tejun Heoeb954192013-08-08 20:11:23 -04005017static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005018 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08005019{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005020 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005021 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005022
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005023 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005024 if (mc.to)
5025 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005026 mmput(mm);
5027 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005028 if (mc.to)
5029 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005030}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005031#else /* !CONFIG_MMU */
Tejun Heoeb954192013-08-08 20:11:23 -04005032static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005033 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005034{
5035 return 0;
5036}
Tejun Heoeb954192013-08-08 20:11:23 -04005037static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005038 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005039{
5040}
Tejun Heoeb954192013-08-08 20:11:23 -04005041static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005042 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005043{
5044}
5045#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005046
Tejun Heof00baae2013-04-15 13:41:15 -07005047/*
5048 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005049 * to verify whether we're attached to the default hierarchy on each mount
5050 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005051 */
Tejun Heoeb954192013-08-08 20:11:23 -04005052static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005053{
5054 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005055 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005056 * guarantees that @root doesn't have any children, so turning it
5057 * on for the root memcg is enough.
5058 */
Tejun Heoded8e212015-09-18 11:56:28 -04005059 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee592015-03-12 16:26:19 -07005060 root_mem_cgroup->use_hierarchy = true;
5061 else
5062 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005063}
5064
Johannes Weiner241994ed2015-02-11 15:26:06 -08005065static u64 memory_current_read(struct cgroup_subsys_state *css,
5066 struct cftype *cft)
5067{
5068 return mem_cgroup_usage(mem_cgroup_from_css(css), false);
5069}
5070
5071static int memory_low_show(struct seq_file *m, void *v)
5072{
5073 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005074 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005075
5076 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005077 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005078 else
5079 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5080
5081 return 0;
5082}
5083
5084static ssize_t memory_low_write(struct kernfs_open_file *of,
5085 char *buf, size_t nbytes, loff_t off)
5086{
5087 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5088 unsigned long low;
5089 int err;
5090
5091 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005092 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005093 if (err)
5094 return err;
5095
5096 memcg->low = low;
5097
5098 return nbytes;
5099}
5100
5101static int memory_high_show(struct seq_file *m, void *v)
5102{
5103 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005104 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005105
5106 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005107 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005108 else
5109 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5110
5111 return 0;
5112}
5113
5114static ssize_t memory_high_write(struct kernfs_open_file *of,
5115 char *buf, size_t nbytes, loff_t off)
5116{
5117 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5118 unsigned long high;
5119 int err;
5120
5121 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005122 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005123 if (err)
5124 return err;
5125
5126 memcg->high = high;
5127
Tejun Heoe65e8b62015-05-22 18:23:34 -04005128 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005129 return nbytes;
5130}
5131
5132static int memory_max_show(struct seq_file *m, void *v)
5133{
5134 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005135 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005136
5137 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005138 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005139 else
5140 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5141
5142 return 0;
5143}
5144
5145static ssize_t memory_max_write(struct kernfs_open_file *of,
5146 char *buf, size_t nbytes, loff_t off)
5147{
5148 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5149 unsigned long max;
5150 int err;
5151
5152 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005153 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005154 if (err)
5155 return err;
5156
5157 err = mem_cgroup_resize_limit(memcg, max);
5158 if (err)
5159 return err;
5160
Tejun Heoe65e8b62015-05-22 18:23:34 -04005161 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005162 return nbytes;
5163}
5164
5165static int memory_events_show(struct seq_file *m, void *v)
5166{
5167 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5168
5169 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5170 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5171 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5172 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5173
5174 return 0;
5175}
5176
5177static struct cftype memory_files[] = {
5178 {
5179 .name = "current",
5180 .read_u64 = memory_current_read,
5181 },
5182 {
5183 .name = "low",
5184 .flags = CFTYPE_NOT_ON_ROOT,
5185 .seq_show = memory_low_show,
5186 .write = memory_low_write,
5187 },
5188 {
5189 .name = "high",
5190 .flags = CFTYPE_NOT_ON_ROOT,
5191 .seq_show = memory_high_show,
5192 .write = memory_high_write,
5193 },
5194 {
5195 .name = "max",
5196 .flags = CFTYPE_NOT_ON_ROOT,
5197 .seq_show = memory_max_show,
5198 .write = memory_max_write,
5199 },
5200 {
5201 .name = "events",
5202 .flags = CFTYPE_NOT_ON_ROOT,
5203 .seq_show = memory_events_show,
5204 },
5205 { } /* terminate */
5206};
5207
Tejun Heo073219e2014-02-08 10:36:58 -05005208struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005209 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005210 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005211 .css_offline = mem_cgroup_css_offline,
5212 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005213 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005214 .can_attach = mem_cgroup_can_attach,
5215 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005216 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005217 .bind = mem_cgroup_bind,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005218 .dfl_cftypes = memory_files,
5219 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005220 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005221};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005222
Johannes Weiner241994ed2015-02-11 15:26:06 -08005223/**
Johannes Weiner241994ed2015-02-11 15:26:06 -08005224 * mem_cgroup_low - check if memory consumption is below the normal range
5225 * @root: the highest ancestor to consider
5226 * @memcg: the memory cgroup to check
5227 *
5228 * Returns %true if memory consumption of @memcg, and that of all
5229 * configurable ancestors up to @root, is below the normal range.
5230 */
5231bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5232{
5233 if (mem_cgroup_disabled())
5234 return false;
5235
5236 /*
5237 * The toplevel group doesn't have a configurable range, so
5238 * it's never low when looked at directly, and it is not
5239 * considered an ancestor when assessing the hierarchy.
5240 */
5241
5242 if (memcg == root_mem_cgroup)
5243 return false;
5244
Michal Hocko4e54ded2015-02-27 15:51:46 -08005245 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005246 return false;
5247
5248 while (memcg != root) {
5249 memcg = parent_mem_cgroup(memcg);
5250
5251 if (memcg == root_mem_cgroup)
5252 break;
5253
Michal Hocko4e54ded2015-02-27 15:51:46 -08005254 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005255 return false;
5256 }
5257 return true;
5258}
5259
Johannes Weiner00501b52014-08-08 14:19:20 -07005260/**
5261 * mem_cgroup_try_charge - try charging a page
5262 * @page: page to charge
5263 * @mm: mm context of the victim
5264 * @gfp_mask: reclaim mode
5265 * @memcgp: charged memcg return
5266 *
5267 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5268 * pages according to @gfp_mask if necessary.
5269 *
5270 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5271 * Otherwise, an error code is returned.
5272 *
5273 * After page->mapping has been set up, the caller must finalize the
5274 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5275 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5276 */
5277int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
5278 gfp_t gfp_mask, struct mem_cgroup **memcgp)
5279{
5280 struct mem_cgroup *memcg = NULL;
5281 unsigned int nr_pages = 1;
5282 int ret = 0;
5283
5284 if (mem_cgroup_disabled())
5285 goto out;
5286
5287 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005288 /*
5289 * Every swap fault against a single page tries to charge the
5290 * page, bail as early as possible. shmem_unuse() encounters
5291 * already charged pages, too. The USED bit is protected by
5292 * the page lock, which serializes swap cache removal, which
5293 * in turn serializes uncharging.
5294 */
Johannes Weiner1306a852014-12-10 15:44:52 -08005295 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005296 goto out;
5297 }
5298
5299 if (PageTransHuge(page)) {
5300 nr_pages <<= compound_order(page);
5301 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5302 }
5303
5304 if (do_swap_account && PageSwapCache(page))
5305 memcg = try_get_mem_cgroup_from_page(page);
5306 if (!memcg)
5307 memcg = get_mem_cgroup_from_mm(mm);
5308
5309 ret = try_charge(memcg, gfp_mask, nr_pages);
5310
5311 css_put(&memcg->css);
5312
5313 if (ret == -EINTR) {
5314 memcg = root_mem_cgroup;
5315 ret = 0;
5316 }
5317out:
5318 *memcgp = memcg;
5319 return ret;
5320}
5321
5322/**
5323 * mem_cgroup_commit_charge - commit a page charge
5324 * @page: page to charge
5325 * @memcg: memcg to charge the page to
5326 * @lrucare: page might be on LRU already
5327 *
5328 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5329 * after page->mapping has been set up. This must happen atomically
5330 * as part of the page instantiation, i.e. under the page table lock
5331 * for anonymous pages, under the page lock for page and swap cache.
5332 *
5333 * In addition, the page must not be on the LRU during the commit, to
5334 * prevent racing with task migration. If it might be, use @lrucare.
5335 *
5336 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5337 */
5338void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
5339 bool lrucare)
5340{
5341 unsigned int nr_pages = 1;
5342
5343 VM_BUG_ON_PAGE(!page->mapping, page);
5344 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5345
5346 if (mem_cgroup_disabled())
5347 return;
5348 /*
5349 * Swap faults will attempt to charge the same page multiple
5350 * times. But reuse_swap_page() might have removed the page
5351 * from swapcache already, so we can't check PageSwapCache().
5352 */
5353 if (!memcg)
5354 return;
5355
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005356 commit_charge(page, memcg, lrucare);
5357
Johannes Weiner00501b52014-08-08 14:19:20 -07005358 if (PageTransHuge(page)) {
5359 nr_pages <<= compound_order(page);
5360 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5361 }
5362
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005363 local_irq_disable();
5364 mem_cgroup_charge_statistics(memcg, page, nr_pages);
5365 memcg_check_events(memcg, page);
5366 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005367
5368 if (do_swap_account && PageSwapCache(page)) {
5369 swp_entry_t entry = { .val = page_private(page) };
5370 /*
5371 * The swap entry might not get freed for a long time,
5372 * let's not wait for it. The page already received a
5373 * memory+swap charge, drop the swap entry duplicate.
5374 */
5375 mem_cgroup_uncharge_swap(entry);
5376 }
5377}
5378
5379/**
5380 * mem_cgroup_cancel_charge - cancel a page charge
5381 * @page: page to charge
5382 * @memcg: memcg to charge the page to
5383 *
5384 * Cancel a charge transaction started by mem_cgroup_try_charge().
5385 */
5386void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg)
5387{
5388 unsigned int nr_pages = 1;
5389
5390 if (mem_cgroup_disabled())
5391 return;
5392 /*
5393 * Swap faults will attempt to charge the same page multiple
5394 * times. But reuse_swap_page() might have removed the page
5395 * from swapcache already, so we can't check PageSwapCache().
5396 */
5397 if (!memcg)
5398 return;
5399
5400 if (PageTransHuge(page)) {
5401 nr_pages <<= compound_order(page);
5402 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5403 }
5404
5405 cancel_charge(memcg, nr_pages);
5406}
5407
Johannes Weiner747db952014-08-08 14:19:24 -07005408static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005409 unsigned long nr_anon, unsigned long nr_file,
5410 unsigned long nr_huge, struct page *dummy_page)
5411{
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005412 unsigned long nr_pages = nr_anon + nr_file;
Johannes Weiner747db952014-08-08 14:19:24 -07005413 unsigned long flags;
5414
Johannes Weinerce00a962014-09-05 08:43:57 -04005415 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005416 page_counter_uncharge(&memcg->memory, nr_pages);
5417 if (do_swap_account)
5418 page_counter_uncharge(&memcg->memsw, nr_pages);
Johannes Weinerce00a962014-09-05 08:43:57 -04005419 memcg_oom_recover(memcg);
5420 }
Johannes Weiner747db952014-08-08 14:19:24 -07005421
5422 local_irq_save(flags);
5423 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5424 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5425 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5426 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005427 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005428 memcg_check_events(memcg, dummy_page);
5429 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005430
5431 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005432 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005433}
5434
5435static void uncharge_list(struct list_head *page_list)
5436{
5437 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005438 unsigned long nr_anon = 0;
5439 unsigned long nr_file = 0;
5440 unsigned long nr_huge = 0;
5441 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005442 struct list_head *next;
5443 struct page *page;
5444
5445 next = page_list->next;
5446 do {
5447 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005448
5449 page = list_entry(next, struct page, lru);
5450 next = page->lru.next;
5451
5452 VM_BUG_ON_PAGE(PageLRU(page), page);
5453 VM_BUG_ON_PAGE(page_count(page), page);
5454
Johannes Weiner1306a852014-12-10 15:44:52 -08005455 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005456 continue;
5457
5458 /*
5459 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005460 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005461 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005462 */
5463
Johannes Weiner1306a852014-12-10 15:44:52 -08005464 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005465 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005466 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5467 nr_huge, page);
5468 pgpgout = nr_anon = nr_file = nr_huge = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005469 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005470 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005471 }
5472
5473 if (PageTransHuge(page)) {
5474 nr_pages <<= compound_order(page);
5475 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5476 nr_huge += nr_pages;
5477 }
5478
5479 if (PageAnon(page))
5480 nr_anon += nr_pages;
5481 else
5482 nr_file += nr_pages;
5483
Johannes Weiner1306a852014-12-10 15:44:52 -08005484 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005485
5486 pgpgout++;
5487 } while (next != page_list);
5488
5489 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005490 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5491 nr_huge, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005492}
5493
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005494/**
5495 * mem_cgroup_uncharge - uncharge a page
5496 * @page: page to uncharge
5497 *
5498 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5499 * mem_cgroup_commit_charge().
5500 */
5501void mem_cgroup_uncharge(struct page *page)
5502{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005503 if (mem_cgroup_disabled())
5504 return;
5505
Johannes Weiner747db952014-08-08 14:19:24 -07005506 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005507 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005508 return;
5509
Johannes Weiner747db952014-08-08 14:19:24 -07005510 INIT_LIST_HEAD(&page->lru);
5511 uncharge_list(&page->lru);
5512}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005513
Johannes Weiner747db952014-08-08 14:19:24 -07005514/**
5515 * mem_cgroup_uncharge_list - uncharge a list of page
5516 * @page_list: list of pages to uncharge
5517 *
5518 * Uncharge a list of pages previously charged with
5519 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5520 */
5521void mem_cgroup_uncharge_list(struct list_head *page_list)
5522{
5523 if (mem_cgroup_disabled())
5524 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005525
Johannes Weiner747db952014-08-08 14:19:24 -07005526 if (!list_empty(page_list))
5527 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005528}
5529
5530/**
5531 * mem_cgroup_migrate - migrate a charge to another page
5532 * @oldpage: currently charged page
5533 * @newpage: page to transfer the charge to
Michal Hockof5e03a42015-02-05 12:25:14 -08005534 * @lrucare: either or both pages might be on the LRU already
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005535 *
5536 * Migrate the charge from @oldpage to @newpage.
5537 *
5538 * Both pages must be locked, @newpage->mapping must be set up.
5539 */
5540void mem_cgroup_migrate(struct page *oldpage, struct page *newpage,
5541 bool lrucare)
5542{
Johannes Weiner29833312014-12-10 15:44:02 -08005543 struct mem_cgroup *memcg;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005544 int isolated;
5545
5546 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5547 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
5548 VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage);
5549 VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage);
5550 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005551 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5552 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005553
5554 if (mem_cgroup_disabled())
5555 return;
5556
5557 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005558 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005559 return;
5560
Johannes Weiner7d5e3242014-12-10 15:43:46 -08005561 /*
5562 * Swapcache readahead pages can get migrated before being
5563 * charged, and migration from compaction can happen to an
5564 * uncharged page when the PFN walker finds a page that
5565 * reclaim just put back on the LRU but has not released yet.
5566 */
Johannes Weiner1306a852014-12-10 15:44:52 -08005567 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005568 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005569 return;
5570
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005571 if (lrucare)
5572 lock_page_lru(oldpage, &isolated);
5573
Johannes Weiner1306a852014-12-10 15:44:52 -08005574 oldpage->mem_cgroup = NULL;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005575
5576 if (lrucare)
5577 unlock_page_lru(oldpage, isolated);
5578
Johannes Weiner29833312014-12-10 15:44:02 -08005579 commit_charge(newpage, memcg, lrucare);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005580}
5581
Michal Hocko2d110852013-02-22 16:34:43 -08005582/*
Michal Hocko10813122013-02-22 16:35:41 -08005583 * subsys_initcall() for memory controller.
5584 *
5585 * Some parts like hotcpu_notifier() have to be initialized from this context
5586 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5587 * everything that doesn't depend on a specific mem_cgroup structure should
5588 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005589 */
5590static int __init mem_cgroup_init(void)
5591{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005592 int cpu, node;
5593
Michal Hocko2d110852013-02-22 16:34:43 -08005594 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005595
5596 for_each_possible_cpu(cpu)
5597 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5598 drain_local_stock);
5599
5600 for_each_node(node) {
5601 struct mem_cgroup_tree_per_node *rtpn;
5602 int zone;
5603
5604 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5605 node_online(node) ? node : NUMA_NO_NODE);
5606
5607 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5608 struct mem_cgroup_tree_per_zone *rtpz;
5609
5610 rtpz = &rtpn->rb_tree_per_zone[zone];
5611 rtpz->rb_root = RB_ROOT;
5612 spin_lock_init(&rtpz->lock);
5613 }
5614 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5615 }
5616
Michal Hocko2d110852013-02-22 16:34:43 -08005617 return 0;
5618}
5619subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005620
5621#ifdef CONFIG_MEMCG_SWAP
5622/**
5623 * mem_cgroup_swapout - transfer a memsw charge to swap
5624 * @page: page whose memsw charge to transfer
5625 * @entry: swap entry to move the charge to
5626 *
5627 * Transfer the memsw charge of @page to @entry.
5628 */
5629void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5630{
5631 struct mem_cgroup *memcg;
5632 unsigned short oldid;
5633
5634 VM_BUG_ON_PAGE(PageLRU(page), page);
5635 VM_BUG_ON_PAGE(page_count(page), page);
5636
5637 if (!do_swap_account)
5638 return;
5639
5640 memcg = page->mem_cgroup;
5641
5642 /* Readahead page, never charged */
5643 if (!memcg)
5644 return;
5645
5646 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5647 VM_BUG_ON_PAGE(oldid, page);
5648 mem_cgroup_swap_statistics(memcg, true);
5649
5650 page->mem_cgroup = NULL;
5651
5652 if (!mem_cgroup_is_root(memcg))
5653 page_counter_uncharge(&memcg->memory, 1);
5654
Johannes Weinerf3717632015-06-10 11:14:54 -07005655 /* Caller disabled preemption with mapping->tree_lock */
Johannes Weiner21afa382015-02-11 15:26:36 -08005656 mem_cgroup_charge_statistics(memcg, page, -1);
5657 memcg_check_events(memcg, page);
5658}
5659
5660/**
5661 * mem_cgroup_uncharge_swap - uncharge a swap entry
5662 * @entry: swap entry to uncharge
5663 *
5664 * Drop the memsw charge associated with @entry.
5665 */
5666void mem_cgroup_uncharge_swap(swp_entry_t entry)
5667{
5668 struct mem_cgroup *memcg;
5669 unsigned short id;
5670
5671 if (!do_swap_account)
5672 return;
5673
5674 id = swap_cgroup_record(entry, 0);
5675 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005676 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005677 if (memcg) {
5678 if (!mem_cgroup_is_root(memcg))
5679 page_counter_uncharge(&memcg->memsw, 1);
5680 mem_cgroup_swap_statistics(memcg, false);
5681 css_put(&memcg->css);
5682 }
5683 rcu_read_unlock();
5684}
5685
5686/* for remember boot option*/
5687#ifdef CONFIG_MEMCG_SWAP_ENABLED
5688static int really_do_swap_account __initdata = 1;
5689#else
5690static int really_do_swap_account __initdata;
5691#endif
5692
5693static int __init enable_swap_account(char *s)
5694{
5695 if (!strcmp(s, "1"))
5696 really_do_swap_account = 1;
5697 else if (!strcmp(s, "0"))
5698 really_do_swap_account = 0;
5699 return 1;
5700}
5701__setup("swapaccount=", enable_swap_account);
5702
5703static struct cftype memsw_cgroup_files[] = {
5704 {
5705 .name = "memsw.usage_in_bytes",
5706 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
5707 .read_u64 = mem_cgroup_read_u64,
5708 },
5709 {
5710 .name = "memsw.max_usage_in_bytes",
5711 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
5712 .write = mem_cgroup_reset,
5713 .read_u64 = mem_cgroup_read_u64,
5714 },
5715 {
5716 .name = "memsw.limit_in_bytes",
5717 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
5718 .write = mem_cgroup_write,
5719 .read_u64 = mem_cgroup_read_u64,
5720 },
5721 {
5722 .name = "memsw.failcnt",
5723 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
5724 .write = mem_cgroup_reset,
5725 .read_u64 = mem_cgroup_read_u64,
5726 },
5727 { }, /* terminate */
5728};
5729
5730static int __init mem_cgroup_swap_init(void)
5731{
5732 if (!mem_cgroup_disabled() && really_do_swap_account) {
5733 do_swap_account = 1;
5734 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
5735 memsw_cgroup_files));
5736 }
5737 return 0;
5738}
5739subsys_initcall(mem_cgroup_swap_init);
5740
5741#endif /* CONFIG_MEMCG_SWAP */