blob: c57c4423c68837d14816c5ff230435e1567e7c20 [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 Heo56161632015-05-22 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 Thelenc4843a72015-05-22 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 Hockoe752eb62015-09-08 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. Biederman2e685cad2013-10-19 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 Heoad7fa852015-05-27 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
437 if (!memcg || !cgroup_on_dfl(memcg->css.cgroup))
438 memcg = root_mem_cgroup;
439
440 rcu_read_unlock();
441 return &memcg->css;
442}
443
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700444/**
445 * page_cgroup_ino - return inode number of the memcg a page is charged to
446 * @page: the page
447 *
448 * Look up the closest online ancestor of the memory cgroup @page is charged to
449 * and return its inode number or 0 if @page is not charged to any cgroup. It
450 * is safe to call this function without holding a reference to @page.
451 *
452 * Note, this function is inherently racy, because there is nothing to prevent
453 * the cgroup inode from getting torn down and potentially reallocated a moment
454 * after page_cgroup_ino() returns, so it only should be used by callers that
455 * do not care (such as procfs interfaces).
456 */
457ino_t page_cgroup_ino(struct page *page)
458{
459 struct mem_cgroup *memcg;
460 unsigned long ino = 0;
461
462 rcu_read_lock();
463 memcg = READ_ONCE(page->mem_cgroup);
464 while (memcg && !(memcg->css.flags & CSS_ONLINE))
465 memcg = parent_mem_cgroup(memcg);
466 if (memcg)
467 ino = cgroup_ino(memcg->css.cgroup);
468 rcu_read_unlock();
469 return ino;
470}
471
Balbir Singhf64c3f52009-09-23 15:56:37 -0700472static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700473mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700474{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700475 int nid = page_to_nid(page);
476 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700477
Jianyu Zhane2318752014-06-06 14:38:20 -0700478 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700479}
480
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700481static struct mem_cgroup_tree_per_zone *
482soft_limit_tree_node_zone(int nid, int zid)
483{
484 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
485}
486
487static struct mem_cgroup_tree_per_zone *
488soft_limit_tree_from_page(struct page *page)
489{
490 int nid = page_to_nid(page);
491 int zid = page_zonenum(page);
492
493 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
494}
495
Johannes Weinercf2c8122014-06-06 14:38:21 -0700496static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz,
497 struct mem_cgroup_tree_per_zone *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800498 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700499{
500 struct rb_node **p = &mctz->rb_root.rb_node;
501 struct rb_node *parent = NULL;
502 struct mem_cgroup_per_zone *mz_node;
503
504 if (mz->on_tree)
505 return;
506
507 mz->usage_in_excess = new_usage_in_excess;
508 if (!mz->usage_in_excess)
509 return;
510 while (*p) {
511 parent = *p;
512 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
513 tree_node);
514 if (mz->usage_in_excess < mz_node->usage_in_excess)
515 p = &(*p)->rb_left;
516 /*
517 * We can't avoid mem cgroups that are over their soft
518 * limit by the same amount
519 */
520 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
521 p = &(*p)->rb_right;
522 }
523 rb_link_node(&mz->tree_node, parent, p);
524 rb_insert_color(&mz->tree_node, &mctz->rb_root);
525 mz->on_tree = true;
526}
527
Johannes Weinercf2c8122014-06-06 14:38:21 -0700528static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
529 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700530{
531 if (!mz->on_tree)
532 return;
533 rb_erase(&mz->tree_node, &mctz->rb_root);
534 mz->on_tree = false;
535}
536
Johannes Weinercf2c8122014-06-06 14:38:21 -0700537static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
538 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700539{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700540 unsigned long flags;
541
542 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700543 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700544 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700545}
546
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800547static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
548{
549 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700550 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800551 unsigned long excess = 0;
552
553 if (nr_pages > soft_limit)
554 excess = nr_pages - soft_limit;
555
556 return excess;
557}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700558
559static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
560{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800561 unsigned long excess;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700562 struct mem_cgroup_per_zone *mz;
563 struct mem_cgroup_tree_per_zone *mctz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700564
Jianyu Zhane2318752014-06-06 14:38:20 -0700565 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700566 /*
567 * Necessary to update all ancestors when hierarchy is used.
568 * because their event counter is not touched.
569 */
570 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Jianyu Zhane2318752014-06-06 14:38:20 -0700571 mz = mem_cgroup_page_zoneinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800572 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700573 /*
574 * We have to update the tree if mz is on RB-tree or
575 * mem is over its softlimit.
576 */
577 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700578 unsigned long flags;
579
580 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700581 /* if on-tree, remove it */
582 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700583 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700584 /*
585 * Insert again. mz->usage_in_excess will be updated.
586 * If excess is 0, no tree ops.
587 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700588 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700589 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700590 }
591 }
592}
593
594static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
595{
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700596 struct mem_cgroup_tree_per_zone *mctz;
Jianyu Zhane2318752014-06-06 14:38:20 -0700597 struct mem_cgroup_per_zone *mz;
598 int nid, zid;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700599
Jianyu Zhane2318752014-06-06 14:38:20 -0700600 for_each_node(nid) {
601 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
602 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
603 mctz = soft_limit_tree_node_zone(nid, zid);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700604 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700605 }
606 }
607}
608
609static struct mem_cgroup_per_zone *
610__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
611{
612 struct rb_node *rightmost = NULL;
613 struct mem_cgroup_per_zone *mz;
614
615retry:
616 mz = NULL;
617 rightmost = rb_last(&mctz->rb_root);
618 if (!rightmost)
619 goto done; /* Nothing to reclaim from */
620
621 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
622 /*
623 * Remove the node now but someone else can add it back,
624 * we will to add it back at the end of reclaim to its correct
625 * position in the tree.
626 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700627 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800628 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400629 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700630 goto retry;
631done:
632 return mz;
633}
634
635static struct mem_cgroup_per_zone *
636mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
637{
638 struct mem_cgroup_per_zone *mz;
639
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700640 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700641 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700642 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700643 return mz;
644}
645
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700646/*
Greg Thelen484ebb32015-10-01 15:37:05 -0700647 * Return page count for single (non recursive) @memcg.
648 *
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700649 * Implementation Note: reading percpu statistics for memcg.
650 *
651 * Both of vmstat[] and percpu_counter has threshold and do periodic
652 * synchronization to implement "quick" read. There are trade-off between
653 * reading cost and precision of value. Then, we may have a chance to implement
Greg Thelen484ebb32015-10-01 15:37:05 -0700654 * a periodic synchronization of counter in memcg's counter.
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700655 *
656 * But this _read() function is used for user interface now. The user accounts
657 * memory usage by memory cgroup and he _always_ requires exact value because
658 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
659 * have to visit all online cpus and make sum. So, for now, unnecessary
660 * synchronization is not implemented. (just implemented for cpu hotplug)
661 *
662 * If there are kernel internal actions which can make use of some not-exact
663 * value, and reading all cpu value can be performance bottleneck in some
Greg Thelen484ebb32015-10-01 15:37:05 -0700664 * common workload, threshold and synchronization as vmstat[] should be
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700665 * implemented.
666 */
Greg Thelen484ebb32015-10-01 15:37:05 -0700667static unsigned long
668mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800669{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700670 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800671 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800672
Greg Thelen484ebb32015-10-01 15:37:05 -0700673 /* Per-cpu values can be negative, use a signed accumulator */
Tejun Heo733a5722015-05-22 18:23:18 -0400674 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700675 val += per_cpu(memcg->stat->count[idx], cpu);
Greg Thelen484ebb32015-10-01 15:37:05 -0700676 /*
677 * Summing races with updates, so val may be negative. Avoid exposing
678 * transient negative values.
679 */
680 if (val < 0)
681 val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800682 return val;
683}
684
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700685static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700686 enum mem_cgroup_events_index idx)
687{
688 unsigned long val = 0;
689 int cpu;
690
Tejun Heo733a5722015-05-22 18:23:18 -0400691 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700692 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700693 return val;
694}
695
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700696static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700697 struct page *page,
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700698 int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800699{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700700 /*
701 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
702 * counted as CACHE even if it's on ANON LRU.
703 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700704 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700705 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700706 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800707 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700708 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700709 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700710
David Rientjesb070e652013-05-07 16:18:09 -0700711 if (PageTransHuge(page))
712 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
713 nr_pages);
714
KAMEZAWA Hiroyukie401f1762011-01-20 14:44:23 -0800715 /* pagein of a big page is an event. So, ignore page size */
716 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700717 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800718 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700719 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800720 nr_pages = -nr_pages; /* for event */
721 }
KAMEZAWA Hiroyukie401f1762011-01-20 14:44:23 -0800722
Johannes Weiner13114712012-05-29 15:07:07 -0700723 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800724}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800725
Jianyu Zhane2318752014-06-06 14:38:20 -0700726static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
727 int nid,
728 unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700729{
Jianyu Zhane2318752014-06-06 14:38:20 -0700730 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700731 int zid;
732
Jianyu Zhane2318752014-06-06 14:38:20 -0700733 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700734
Jianyu Zhane2318752014-06-06 14:38:20 -0700735 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
736 struct mem_cgroup_per_zone *mz;
737 enum lru_list lru;
738
739 for_each_lru(lru) {
740 if (!(BIT(lru) & lru_mask))
741 continue;
742 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
743 nr += mz->lru_size[lru];
744 }
745 }
746 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700747}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700748
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700749static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700750 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800751{
Jianyu Zhane2318752014-06-06 14:38:20 -0700752 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700753 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800754
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800755 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700756 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
757 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800758}
759
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800760static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
761 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800762{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700763 unsigned long val, next;
764
Johannes Weiner13114712012-05-29 15:07:07 -0700765 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700766 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700767 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800768 if ((long)next - (long)val < 0) {
769 switch (target) {
770 case MEM_CGROUP_TARGET_THRESH:
771 next = val + THRESHOLDS_EVENTS_TARGET;
772 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700773 case MEM_CGROUP_TARGET_SOFTLIMIT:
774 next = val + SOFTLIMIT_EVENTS_TARGET;
775 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800776 case MEM_CGROUP_TARGET_NUMAINFO:
777 next = val + NUMAINFO_EVENTS_TARGET;
778 break;
779 default:
780 break;
781 }
782 __this_cpu_write(memcg->stat->targets[target], next);
783 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700784 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800785 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800786}
787
788/*
789 * Check events in order.
790 *
791 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700792static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800793{
794 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800795 if (unlikely(mem_cgroup_event_ratelimit(memcg,
796 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700797 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800798 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800799
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700800 do_softlimit = mem_cgroup_event_ratelimit(memcg,
801 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700802#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800803 do_numainfo = mem_cgroup_event_ratelimit(memcg,
804 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700805#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800806 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700807 if (unlikely(do_softlimit))
808 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800809#if MAX_NUMNODES > 1
810 if (unlikely(do_numainfo))
811 atomic_inc(&memcg->numainfo_events);
812#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700813 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800814}
815
Balbir Singhcf475ad2008-04-29 01:00:16 -0700816struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800817{
Balbir Singh31a78f22008-09-28 23:09:31 +0100818 /*
819 * mm_update_next_owner() may clear mm->owner to NULL
820 * if it races with swapoff, page migration, etc.
821 * So this can be called with p == NULL.
822 */
823 if (unlikely(!p))
824 return NULL;
825
Tejun Heo073219e2014-02-08 10:36:58 -0500826 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800827}
Michal Hocko33398cf2015-09-08 15:01:02 -0700828EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800829
Johannes Weinerdf381972014-04-07 15:37:43 -0700830static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800831{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700832 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700833
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800834 rcu_read_lock();
835 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700836 /*
837 * Page cache insertions can happen withou an
838 * actual mm context, e.g. during disk probing
839 * on boot, loopback IO, acct() writes etc.
840 */
841 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700842 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700843 else {
844 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
845 if (unlikely(!memcg))
846 memcg = root_mem_cgroup;
847 }
Tejun Heoec903c02014-05-13 12:11:01 -0400848 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800849 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700850 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800851}
852
Johannes Weiner56600482012-01-12 17:17:59 -0800853/**
854 * mem_cgroup_iter - iterate over memory cgroup hierarchy
855 * @root: hierarchy root
856 * @prev: previously returned memcg, NULL on first invocation
857 * @reclaim: cookie for shared reclaim walks, NULL for full walks
858 *
859 * Returns references to children of the hierarchy below @root, or
860 * @root itself, or %NULL after a full round-trip.
861 *
862 * Caller must pass the return value in @prev on subsequent
863 * invocations for reference counting, or use mem_cgroup_iter_break()
864 * to cancel a hierarchy walk before the round-trip is complete.
865 *
866 * Reclaimers can specify a zone and a priority level in @reclaim to
867 * divide up the memcgs in the hierarchy among all concurrent
868 * reclaimers operating on the same zone and priority.
869 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700870struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800871 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700872 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700873{
Michal Hocko33398cf2015-09-08 15:01:02 -0700874 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800875 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800876 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800877 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700878
Andrew Morton694fbc02013-09-24 15:27:37 -0700879 if (mem_cgroup_disabled())
880 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800881
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700882 if (!root)
883 root = root_mem_cgroup;
884
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800885 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800886 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800887
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800888 if (!root->use_hierarchy && root != root_mem_cgroup) {
889 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800890 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700891 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800892 }
893
Michal Hocko542f85f2013-04-29 15:07:15 -0700894 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800895
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800896 if (reclaim) {
897 struct mem_cgroup_per_zone *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800898
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800899 mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone);
900 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700901
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800902 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700903 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800904
905 do {
Jason Low4db0c3c2015-04-15 16:14:08 -0700906 pos = READ_ONCE(iter->position);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800907 /*
908 * A racing update may change the position and
909 * put the last reference, hence css_tryget(),
910 * or retry to see the updated position.
911 */
912 } while (pos && !css_tryget(&pos->css));
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800913 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800914
915 if (pos)
916 css = &pos->css;
917
918 for (;;) {
919 css = css_next_descendant_pre(css, &root->css);
920 if (!css) {
921 /*
922 * Reclaimers share the hierarchy walk, and a
923 * new one might jump in right at the end of
924 * the hierarchy - make sure they see at least
925 * one group and restart from the beginning.
926 */
927 if (!prev)
928 continue;
929 break;
930 }
931
932 /*
933 * Verify the css and acquire a reference. The root
934 * is provided by the caller, so we know it's alive
935 * and kicking, and don't take an extra reference.
936 */
937 memcg = mem_cgroup_from_css(css);
938
939 if (css == &root->css)
940 break;
941
Johannes Weinerb2052562014-12-10 15:42:48 -0800942 if (css_tryget(css)) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800943 /*
944 * Make sure the memcg is initialized:
945 * mem_cgroup_css_online() orders the the
946 * initialization against setting the flag.
947 */
948 if (smp_load_acquire(&memcg->initialized))
949 break;
950
951 css_put(css);
952 }
953
954 memcg = NULL;
955 }
956
957 if (reclaim) {
958 if (cmpxchg(&iter->position, pos, memcg) == pos) {
959 if (memcg)
960 css_get(&memcg->css);
961 if (pos)
962 css_put(&pos->css);
963 }
964
965 /*
966 * pairs with css_tryget when dereferencing iter->position
967 * above.
968 */
969 if (pos)
970 css_put(&pos->css);
971
972 if (!memcg)
973 iter->generation++;
974 else if (!prev)
975 reclaim->generation = iter->generation;
976 }
977
Michal Hocko542f85f2013-04-29 15:07:15 -0700978out_unlock:
979 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800980out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700981 if (prev && prev != root)
982 css_put(&prev->css);
983
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800984 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700985}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800986
Johannes Weiner56600482012-01-12 17:17:59 -0800987/**
988 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
989 * @root: hierarchy root
990 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
991 */
992void mem_cgroup_iter_break(struct mem_cgroup *root,
993 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800994{
995 if (!root)
996 root = root_mem_cgroup;
997 if (prev && prev != root)
998 css_put(&prev->css);
999}
1000
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001001/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001002 * Iteration constructs for visiting all cgroups (under a tree). If
1003 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1004 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001005 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001006#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001007 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001008 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001009 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001010
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001011#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001012 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001013 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001014 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001015
Johannes Weiner925b7672012-01-12 17:18:15 -08001016/**
1017 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1018 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001019 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -08001020 *
1021 * Returns the lru list vector holding pages for the given @zone and
1022 * @mem. This can be the global zone lruvec, if the memory controller
1023 * is disabled.
1024 */
1025struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1026 struct mem_cgroup *memcg)
1027{
1028 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001029 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001030
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001031 if (mem_cgroup_disabled()) {
1032 lruvec = &zone->lruvec;
1033 goto out;
1034 }
Johannes Weiner925b7672012-01-12 17:18:15 -08001035
Jianyu Zhane2318752014-06-06 14:38:20 -07001036 mz = mem_cgroup_zone_zoneinfo(memcg, zone);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001037 lruvec = &mz->lruvec;
1038out:
1039 /*
1040 * Since a node can be onlined after the mem_cgroup was created,
1041 * we have to be prepared to initialize lruvec->zone here;
1042 * and if offlined then reonlined, we need to reinitialize it.
1043 */
1044 if (unlikely(lruvec->zone != zone))
1045 lruvec->zone = zone;
1046 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001047}
1048
Johannes Weiner925b7672012-01-12 17:18:15 -08001049/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001050 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -08001051 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001052 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001053 *
1054 * This function is only safe when following the LRU page isolation
1055 * and putback protocol: the LRU lock must be held, and the page must
1056 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -07001057 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001058struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -07001059{
1060 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001061 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001062 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001063
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001064 if (mem_cgroup_disabled()) {
1065 lruvec = &zone->lruvec;
1066 goto out;
1067 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001068
Johannes Weiner1306a852014-12-10 15:44:52 -08001069 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001070 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001071 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -08001072 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -08001073 */
Johannes Weiner29833312014-12-10 15:44:02 -08001074 if (!memcg)
1075 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001076
Jianyu Zhane2318752014-06-06 14:38:20 -07001077 mz = mem_cgroup_page_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001078 lruvec = &mz->lruvec;
1079out:
1080 /*
1081 * Since a node can be onlined after the mem_cgroup was created,
1082 * we have to be prepared to initialize lruvec->zone here;
1083 * and if offlined then reonlined, we need to reinitialize it.
1084 */
1085 if (unlikely(lruvec->zone != zone))
1086 lruvec->zone = zone;
1087 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001088}
1089
1090/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001091 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1092 * @lruvec: mem_cgroup per zone lru vector
1093 * @lru: index of lru list the page is sitting on
1094 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001095 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001096 * This function must be called when a page is added to or removed from an
1097 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001098 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001099void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1100 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001101{
1102 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001103 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001104
1105 if (mem_cgroup_disabled())
1106 return;
1107
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001108 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1109 lru_size = mz->lru_size + lru;
1110 *lru_size += nr_pages;
1111 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001112}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001113
Johannes Weiner2314b422014-12-10 15:44:33 -08001114bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001115{
Johannes Weiner2314b422014-12-10 15:44:33 -08001116 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001117 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001118 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001119
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001120 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001121 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001122 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001123 task_unlock(p);
1124 } else {
1125 /*
1126 * All threads may have already detached their mm's, but the oom
1127 * killer still needs to detect if they have already been oom
1128 * killed to prevent needlessly killing additional tasks.
1129 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001130 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001131 task_memcg = mem_cgroup_from_task(task);
1132 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001133 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001134 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001135 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1136 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001137 return ret;
1138}
1139
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001140#define mem_cgroup_from_counter(counter, member) \
Balbir Singh6d61ef42009-01-07 18:08:06 -08001141 container_of(counter, struct mem_cgroup, member)
1142
Johannes Weiner19942822011-02-01 15:52:43 -08001143/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001144 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001145 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001146 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001147 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001148 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001149 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001150static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001151{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001152 unsigned long margin = 0;
1153 unsigned long count;
1154 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001155
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001156 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001157 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001158 if (count < limit)
1159 margin = limit - count;
1160
1161 if (do_swap_account) {
1162 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001163 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001164 if (count <= limit)
1165 margin = min(margin, limit - count);
1166 }
1167
1168 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001169}
1170
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001171/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001172 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001173 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001174 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1175 * moving cgroups. This is for waiting at high-memory pressure
1176 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001177 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001178static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001179{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001180 struct mem_cgroup *from;
1181 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001182 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001183 /*
1184 * Unlike task_move routines, we access mc.to, mc.from not under
1185 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1186 */
1187 spin_lock(&mc.lock);
1188 from = mc.from;
1189 to = mc.to;
1190 if (!from)
1191 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001192
Johannes Weiner2314b422014-12-10 15:44:33 -08001193 ret = mem_cgroup_is_descendant(from, memcg) ||
1194 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001195unlock:
1196 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001197 return ret;
1198}
1199
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001200static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001201{
1202 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001203 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001204 DEFINE_WAIT(wait);
1205 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1206 /* moving charge context might have finished. */
1207 if (mc.moving_task)
1208 schedule();
1209 finish_wait(&mc.waitq, &wait);
1210 return true;
1211 }
1212 }
1213 return false;
1214}
1215
Sha Zhengju58cf1882013-02-22 16:32:05 -08001216#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001217/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001218 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001219 * @memcg: The memory cgroup that went over limit
1220 * @p: Task that is going to be killed
1221 *
1222 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1223 * enabled
1224 */
1225void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1226{
Tejun Heoe61734c2014-02-12 09:29:50 -05001227 /* oom_info_lock ensures that parallel ooms do not interleave */
Michal Hocko08088cb2014-02-25 15:01:44 -08001228 static DEFINE_MUTEX(oom_info_lock);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001229 struct mem_cgroup *iter;
1230 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001231
Michal Hocko08088cb2014-02-25 15:01:44 -08001232 mutex_lock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001233 rcu_read_lock();
1234
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001235 if (p) {
1236 pr_info("Task in ");
1237 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1238 pr_cont(" killed as a result of limit of ");
1239 } else {
1240 pr_info("Memory limit reached of cgroup ");
1241 }
1242
Tejun Heoe61734c2014-02-12 09:29:50 -05001243 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001244 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001245
Balbir Singhe2224322009-04-02 16:57:39 -07001246 rcu_read_unlock();
1247
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001248 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1249 K((u64)page_counter_read(&memcg->memory)),
1250 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1251 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1252 K((u64)page_counter_read(&memcg->memsw)),
1253 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1254 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1255 K((u64)page_counter_read(&memcg->kmem)),
1256 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001257
1258 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001259 pr_info("Memory cgroup stats for ");
1260 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001261 pr_cont(":");
1262
1263 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1264 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1265 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07001266 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
Sha Zhengju58cf1882013-02-22 16:32:05 -08001267 K(mem_cgroup_read_stat(iter, i)));
1268 }
1269
1270 for (i = 0; i < NR_LRU_LISTS; i++)
1271 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1272 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1273
1274 pr_cont("\n");
1275 }
Michal Hocko08088cb2014-02-25 15:01:44 -08001276 mutex_unlock(&oom_info_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001277}
1278
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001279/*
1280 * This function returns the number of memcg under hierarchy tree. Returns
1281 * 1(self count) if no children.
1282 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001283static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001284{
1285 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001286 struct mem_cgroup *iter;
1287
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001288 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001289 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001290 return num;
1291}
1292
Balbir Singh6d61ef42009-01-07 18:08:06 -08001293/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001294 * Return the memory (and swap, if configured) limit for a memcg.
1295 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001296static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001297{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001298 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001299
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001300 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001301 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001302 unsigned long memsw_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001303
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001304 memsw_limit = memcg->memsw.limit;
1305 limit = min(limit + total_swap_pages, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001306 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001307 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001308}
1309
David Rientjes19965462012-12-11 16:00:26 -08001310static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1311 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001312{
David Rientjes6e0fc462015-09-08 15:00:36 -07001313 struct oom_control oc = {
1314 .zonelist = NULL,
1315 .nodemask = NULL,
1316 .gfp_mask = gfp_mask,
1317 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001318 };
David Rientjes9cbb78b2012-07-31 16:43:44 -07001319 struct mem_cgroup *iter;
1320 unsigned long chosen_points = 0;
1321 unsigned long totalpages;
1322 unsigned int points = 0;
1323 struct task_struct *chosen = NULL;
1324
Johannes Weinerdc564012015-06-24 16:57:19 -07001325 mutex_lock(&oom_lock);
1326
David Rientjes876aafb2012-07-31 16:43:48 -07001327 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001328 * If current has a pending SIGKILL or is exiting, then automatically
1329 * select it. The goal is to allow it to allocate so that it may
1330 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001331 */
Oleg Nesterovd003f372014-12-12 16:56:24 -08001332 if (fatal_signal_pending(current) || task_will_free_mem(current)) {
Johannes Weiner16e95192015-06-24 16:57:07 -07001333 mark_oom_victim(current);
Johannes Weinerdc564012015-06-24 16:57:19 -07001334 goto unlock;
David Rientjes876aafb2012-07-31 16:43:48 -07001335 }
1336
David Rientjes6e0fc462015-09-08 15:00:36 -07001337 check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001338 totalpages = mem_cgroup_get_limit(memcg) ? : 1;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001339 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001340 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001341 struct task_struct *task;
1342
Tejun Heo72ec7022013-08-08 20:11:26 -04001343 css_task_iter_start(&iter->css, &it);
1344 while ((task = css_task_iter_next(&it))) {
David Rientjes6e0fc462015-09-08 15:00:36 -07001345 switch (oom_scan_process_thread(&oc, task, totalpages)) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001346 case OOM_SCAN_SELECT:
1347 if (chosen)
1348 put_task_struct(chosen);
1349 chosen = task;
1350 chosen_points = ULONG_MAX;
1351 get_task_struct(chosen);
1352 /* fall through */
1353 case OOM_SCAN_CONTINUE:
1354 continue;
1355 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001356 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001357 mem_cgroup_iter_break(memcg, iter);
1358 if (chosen)
1359 put_task_struct(chosen);
Johannes Weinerdc564012015-06-24 16:57:19 -07001360 goto unlock;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001361 case OOM_SCAN_OK:
1362 break;
1363 };
1364 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001365 if (!points || points < chosen_points)
1366 continue;
1367 /* Prefer thread group leaders for display purposes */
1368 if (points == chosen_points &&
1369 thread_group_leader(chosen))
1370 continue;
1371
1372 if (chosen)
1373 put_task_struct(chosen);
1374 chosen = task;
1375 chosen_points = points;
1376 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001377 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001378 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001379 }
1380
Johannes Weinerdc564012015-06-24 16:57:19 -07001381 if (chosen) {
1382 points = chosen_points * 1000 / totalpages;
David Rientjes6e0fc462015-09-08 15:00:36 -07001383 oom_kill_process(&oc, chosen, points, totalpages, memcg,
1384 "Memory cgroup out of memory");
Johannes Weinerdc564012015-06-24 16:57:19 -07001385 }
1386unlock:
1387 mutex_unlock(&oom_lock);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001388}
1389
Michele Curtiae6e71d2014-12-12 16:56:35 -08001390#if MAX_NUMNODES > 1
1391
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001392/**
1393 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001394 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001395 * @nid: the node ID to be checked.
1396 * @noswap : specify true here if the user wants flle only information.
1397 *
1398 * This function returns whether the specified memcg contains any
1399 * reclaimable pages on a node. Returns true if there are any reclaimable
1400 * pages in the node.
1401 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001402static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001403 int nid, bool noswap)
1404{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001405 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001406 return true;
1407 if (noswap || !total_swap_pages)
1408 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001409 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001410 return true;
1411 return false;
1412
1413}
Ying Han889976d2011-05-26 16:25:33 -07001414
1415/*
1416 * Always updating the nodemask is not very good - even if we have an empty
1417 * list or the wrong list here, we can start from some node and traverse all
1418 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1419 *
1420 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001421static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001422{
1423 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001424 /*
1425 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1426 * pagein/pageout changes since the last update.
1427 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001428 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001429 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001430 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001431 return;
1432
Ying Han889976d2011-05-26 16:25:33 -07001433 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001434 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001435
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001436 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001437
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001438 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1439 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001440 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001441
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001442 atomic_set(&memcg->numainfo_events, 0);
1443 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001444}
1445
1446/*
1447 * Selecting a node where we start reclaim from. Because what we need is just
1448 * reducing usage counter, start from anywhere is O,K. Considering
1449 * memory reclaim from current node, there are pros. and cons.
1450 *
1451 * Freeing memory from current node means freeing memory from a node which
1452 * we'll use or we've used. So, it may make LRU bad. And if several threads
1453 * hit limits, it will see a contention on a node. But freeing from remote
1454 * node means more costs for memory reclaim because of memory latency.
1455 *
1456 * Now, we use round-robin. Better algorithm is welcomed.
1457 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001458int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001459{
1460 int node;
1461
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001462 mem_cgroup_may_update_nodemask(memcg);
1463 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001464
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001465 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001466 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001467 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001468 /*
1469 * We call this when we hit limit, not when pages are added to LRU.
1470 * No LRU may hold pages because all pages are UNEVICTABLE or
1471 * memcg is too small and all pages are not on LRU. In that case,
1472 * we use curret node.
1473 */
1474 if (unlikely(node == MAX_NUMNODES))
1475 node = numa_node_id();
1476
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001477 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001478 return node;
1479}
Ying Han889976d2011-05-26 16:25:33 -07001480#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001481int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001482{
1483 return 0;
1484}
1485#endif
1486
Andrew Morton0608f432013-09-24 15:27:41 -07001487static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1488 struct zone *zone,
1489 gfp_t gfp_mask,
1490 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001491{
Andrew Morton0608f432013-09-24 15:27:41 -07001492 struct mem_cgroup *victim = NULL;
1493 int total = 0;
1494 int loop = 0;
1495 unsigned long excess;
1496 unsigned long nr_scanned;
1497 struct mem_cgroup_reclaim_cookie reclaim = {
1498 .zone = zone,
1499 .priority = 0,
1500 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001501
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001502 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001503
Andrew Morton0608f432013-09-24 15:27:41 -07001504 while (1) {
1505 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1506 if (!victim) {
1507 loop++;
1508 if (loop >= 2) {
1509 /*
1510 * If we have not been able to reclaim
1511 * anything, it might because there are
1512 * no reclaimable pages under this hierarchy
1513 */
1514 if (!total)
1515 break;
1516 /*
1517 * We want to do more targeted reclaim.
1518 * excess >> 2 is not to excessive so as to
1519 * reclaim too much, nor too less that we keep
1520 * coming back to reclaim from this cgroup
1521 */
1522 if (total >= (excess >> 2) ||
1523 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1524 break;
1525 }
1526 continue;
1527 }
Andrew Morton0608f432013-09-24 15:27:41 -07001528 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1529 zone, &nr_scanned);
1530 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001531 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001532 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001533 }
Andrew Morton0608f432013-09-24 15:27:41 -07001534 mem_cgroup_iter_break(root_memcg, victim);
1535 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001536}
1537
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001538#ifdef CONFIG_LOCKDEP
1539static struct lockdep_map memcg_oom_lock_dep_map = {
1540 .name = "memcg_oom_lock",
1541};
1542#endif
1543
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001544static DEFINE_SPINLOCK(memcg_oom_lock);
1545
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001546/*
1547 * Check OOM-Killer is already running under our hierarchy.
1548 * If someone is running, return false.
1549 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001550static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001551{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001552 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001553
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001554 spin_lock(&memcg_oom_lock);
1555
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001556 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001557 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001558 /*
1559 * this subtree of our hierarchy is already locked
1560 * so we cannot give a lock.
1561 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001562 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001563 mem_cgroup_iter_break(memcg, iter);
1564 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001565 } else
1566 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001567 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001568
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001569 if (failed) {
1570 /*
1571 * OK, we failed to lock the whole subtree so we have
1572 * to clean up what we set up to the failing subtree
1573 */
1574 for_each_mem_cgroup_tree(iter, memcg) {
1575 if (iter == failed) {
1576 mem_cgroup_iter_break(memcg, iter);
1577 break;
1578 }
1579 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001580 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001581 } else
1582 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001583
1584 spin_unlock(&memcg_oom_lock);
1585
1586 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001587}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001588
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001589static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001590{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001591 struct mem_cgroup *iter;
1592
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001593 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001594 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001595 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001596 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001597 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001598}
1599
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001600static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001601{
1602 struct mem_cgroup *iter;
1603
Tejun Heoc2b42d32015-06-24 16:58:23 -07001604 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001605 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001606 iter->under_oom++;
1607 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001608}
1609
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001610static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001611{
1612 struct mem_cgroup *iter;
1613
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001614 /*
1615 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001616 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001617 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001618 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001619 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001620 if (iter->under_oom > 0)
1621 iter->under_oom--;
1622 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001623}
1624
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001625static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1626
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001627struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001628 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001629 wait_queue_t wait;
1630};
1631
1632static int memcg_oom_wake_function(wait_queue_t *wait,
1633 unsigned mode, int sync, void *arg)
1634{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001635 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1636 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001637 struct oom_wait_info *oom_wait_info;
1638
1639 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001640 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001641
Johannes Weiner2314b422014-12-10 15:44:33 -08001642 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1643 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001644 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001645 return autoremove_wake_function(wait, mode, sync, arg);
1646}
1647
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001648static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001649{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001650 /*
1651 * For the following lockless ->under_oom test, the only required
1652 * guarantee is that it must see the state asserted by an OOM when
1653 * this function is called as a result of userland actions
1654 * triggered by the notification of the OOM. This is trivially
1655 * achieved by invoking mem_cgroup_mark_under_oom() before
1656 * triggering notification.
1657 */
1658 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001659 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001660}
1661
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001662static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001663{
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001664 if (!current->memcg_oom.may_oom)
1665 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001666 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001667 * We are in the middle of the charge context here, so we
1668 * don't want to block when potentially sitting on a callstack
1669 * that holds all kinds of filesystem and mm locks.
1670 *
1671 * Also, the caller may handle a failed allocation gracefully
1672 * (like optional page cache readahead) and so an OOM killer
1673 * invocation might not even be necessary.
1674 *
1675 * That's why we don't do anything here except remember the
1676 * OOM context and then deal with it at the end of the page
1677 * fault when the stack is unwound, the locks are released,
1678 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001679 */
Johannes Weiner49426422013-10-16 13:46:59 -07001680 css_get(&memcg->css);
1681 current->memcg_oom.memcg = memcg;
1682 current->memcg_oom.gfp_mask = mask;
1683 current->memcg_oom.order = order;
1684}
1685
1686/**
1687 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1688 * @handle: actually kill/wait or just clean up the OOM state
1689 *
1690 * This has to be called at the end of a page fault if the memcg OOM
1691 * handler was enabled.
1692 *
1693 * Memcg supports userspace OOM handling where failed allocations must
1694 * sleep on a waitqueue until the userspace task resolves the
1695 * situation. Sleeping directly in the charge context with all kinds
1696 * of locks held is not a good idea, instead we remember an OOM state
1697 * in the task and mem_cgroup_oom_synchronize() has to be called at
1698 * the end of the page fault to complete the OOM handling.
1699 *
1700 * Returns %true if an ongoing memcg OOM situation was detected and
1701 * completed, %false otherwise.
1702 */
1703bool mem_cgroup_oom_synchronize(bool handle)
1704{
1705 struct mem_cgroup *memcg = current->memcg_oom.memcg;
1706 struct oom_wait_info owait;
1707 bool locked;
1708
1709 /* OOM is global, do not handle */
1710 if (!memcg)
1711 return false;
1712
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001713 if (!handle || oom_killer_disabled)
Johannes Weiner49426422013-10-16 13:46:59 -07001714 goto cleanup;
1715
1716 owait.memcg = memcg;
1717 owait.wait.flags = 0;
1718 owait.wait.func = memcg_oom_wake_function;
1719 owait.wait.private = current;
1720 INIT_LIST_HEAD(&owait.wait.task_list);
1721
1722 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001723 mem_cgroup_mark_under_oom(memcg);
1724
1725 locked = mem_cgroup_oom_trylock(memcg);
1726
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001727 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001728 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001729
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001730 if (locked && !memcg->oom_kill_disable) {
1731 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001732 finish_wait(&memcg_oom_waitq, &owait.wait);
1733 mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask,
1734 current->memcg_oom.order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001735 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001736 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001737 mem_cgroup_unmark_under_oom(memcg);
1738 finish_wait(&memcg_oom_waitq, &owait.wait);
1739 }
1740
1741 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001742 mem_cgroup_oom_unlock(memcg);
1743 /*
1744 * There is no guarantee that an OOM-lock contender
1745 * sees the wakeups triggered by the OOM kill
1746 * uncharges. Wake any sleepers explicitely.
1747 */
1748 memcg_oom_recover(memcg);
1749 }
Johannes Weiner49426422013-10-16 13:46:59 -07001750cleanup:
1751 current->memcg_oom.memcg = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001752 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001753 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001754}
1755
Johannes Weinerd7365e72014-10-29 14:50:48 -07001756/**
1757 * mem_cgroup_begin_page_stat - begin a page state statistics transaction
1758 * @page: page that is going to change accounted state
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001759 *
Johannes Weinerd7365e72014-10-29 14:50:48 -07001760 * This function must mark the beginning of an accounted page state
1761 * change to prevent double accounting when the page is concurrently
1762 * being moved to another memcg:
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001763 *
Johannes Weiner6de22612015-02-11 15:25:01 -08001764 * memcg = mem_cgroup_begin_page_stat(page);
Johannes Weinerd7365e72014-10-29 14:50:48 -07001765 * if (TestClearPageState(page))
1766 * mem_cgroup_update_page_stat(memcg, state, -1);
Johannes Weiner6de22612015-02-11 15:25:01 -08001767 * mem_cgroup_end_page_stat(memcg);
Balbir Singhd69b0422009-06-17 16:26:34 -07001768 */
Johannes Weiner6de22612015-02-11 15:25:01 -08001769struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001770{
1771 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001772 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001773
Johannes Weiner6de22612015-02-11 15:25:01 -08001774 /*
1775 * The RCU lock is held throughout the transaction. The fast
1776 * path can get away without acquiring the memcg->move_lock
1777 * because page moving starts with an RCU grace period.
1778 *
1779 * The RCU lock also protects the memcg from being freed when
1780 * the page state that is going to change is the only thing
1781 * preventing the page from being uncharged.
1782 * E.g. end-writeback clearing PageWriteback(), which allows
1783 * migration to go ahead and uncharge the page before the
1784 * account transaction might be complete.
1785 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001786 rcu_read_lock();
1787
1788 if (mem_cgroup_disabled())
1789 return NULL;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001790again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001791 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001792 if (unlikely(!memcg))
Johannes Weinerd7365e72014-10-29 14:50:48 -07001793 return NULL;
1794
Qiang Huangbdcbb652014-06-04 16:08:21 -07001795 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weinerd7365e72014-10-29 14:50:48 -07001796 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001797
Johannes Weiner6de22612015-02-11 15:25:01 -08001798 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001799 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001800 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001801 goto again;
1802 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001803
1804 /*
1805 * When charge migration first begins, we can have locked and
1806 * unlocked page stat updates happening concurrently. Track
1807 * the task who has the lock for mem_cgroup_end_page_stat().
1808 */
1809 memcg->move_lock_task = current;
1810 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001811
1812 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001813}
Greg Thelenc4843a72015-05-22 17:13:16 -04001814EXPORT_SYMBOL(mem_cgroup_begin_page_stat);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001815
Johannes Weinerd7365e72014-10-29 14:50:48 -07001816/**
1817 * mem_cgroup_end_page_stat - finish a page state statistics transaction
1818 * @memcg: the memcg that was accounted against
Johannes Weinerd7365e72014-10-29 14:50:48 -07001819 */
Johannes Weiner6de22612015-02-11 15:25:01 -08001820void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001821{
Johannes Weiner6de22612015-02-11 15:25:01 -08001822 if (memcg && memcg->move_lock_task == current) {
1823 unsigned long flags = memcg->move_lock_flags;
1824
1825 memcg->move_lock_task = NULL;
1826 memcg->move_lock_flags = 0;
1827
1828 spin_unlock_irqrestore(&memcg->move_lock, flags);
1829 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001830
Johannes Weinerd7365e72014-10-29 14:50:48 -07001831 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001832}
Greg Thelenc4843a72015-05-22 17:13:16 -04001833EXPORT_SYMBOL(mem_cgroup_end_page_stat);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001834
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001835/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001836 * size of first charge trial. "32" comes from vmscan.c's magic value.
1837 * TODO: maybe necessary to use big numbers in big irons.
1838 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001839#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001840struct memcg_stock_pcp {
1841 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001842 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001843 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001844 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001845#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001846};
1847static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001848static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001849
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001850/**
1851 * consume_stock: Try to consume stocked charge on this cpu.
1852 * @memcg: memcg to consume from.
1853 * @nr_pages: how many pages to charge.
1854 *
1855 * The charges will only happen if @memcg matches the current cpu's memcg
1856 * stock, and at least @nr_pages are available in that stock. Failure to
1857 * service an allocation will refill the stock.
1858 *
1859 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001860 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001861static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001862{
1863 struct memcg_stock_pcp *stock;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001864 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001865
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001866 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001867 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001868
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001869 stock = &get_cpu_var(memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001870 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001871 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001872 ret = true;
1873 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001874 put_cpu_var(memcg_stock);
1875 return ret;
1876}
1877
1878/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001879 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001880 */
1881static void drain_stock(struct memcg_stock_pcp *stock)
1882{
1883 struct mem_cgroup *old = stock->cached;
1884
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001885 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001886 page_counter_uncharge(&old->memory, stock->nr_pages);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001887 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001888 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001889 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001890 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001891 }
1892 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001893}
1894
1895/*
1896 * This must be called under preempt disabled or must be called by
1897 * a thread which is pinned to local cpu.
1898 */
1899static void drain_local_stock(struct work_struct *dummy)
1900{
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001901 struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001902 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001903 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001904}
1905
1906/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001907 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001908 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001909 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001910static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001911{
1912 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1913
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001914 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001915 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001916 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001917 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001918 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001919 put_cpu_var(memcg_stock);
1920}
1921
1922/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001923 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001924 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001925 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001926static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001927{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001928 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001929
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001930 /* If someone's already draining, avoid adding running more workers. */
1931 if (!mutex_trylock(&percpu_charge_mutex))
1932 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001933 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001934 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001935 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001936 for_each_online_cpu(cpu) {
1937 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001938 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001939
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001940 memcg = stock->cached;
1941 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001942 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001943 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001944 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001945 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1946 if (cpu == curcpu)
1947 drain_local_stock(&stock->work);
1948 else
1949 schedule_work_on(cpu, &stock->work);
1950 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001951 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001952 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001953 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001954 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001955}
1956
Paul Gortmaker0db06282013-06-19 14:53:51 -04001957static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001958 unsigned long action,
1959 void *hcpu)
1960{
1961 int cpu = (unsigned long)hcpu;
1962 struct memcg_stock_pcp *stock;
1963
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001964 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001965 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001966
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07001967 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001968 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001969
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001970 stock = &per_cpu(memcg_stock, cpu);
1971 drain_stock(stock);
1972 return NOTIFY_OK;
1973}
1974
Johannes Weiner00501b52014-08-08 14:19:20 -07001975static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1976 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001977{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001978 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001979 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001980 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001981 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001982 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001983 bool may_swap = true;
1984 bool drained = false;
Johannes Weiner05b84302014-08-06 16:05:59 -07001985 int ret = 0;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001986
Johannes Weinerce00a962014-09-05 08:43:57 -04001987 if (mem_cgroup_is_root(memcg))
1988 goto done;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001989retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001990 if (consume_stock(memcg, nr_pages))
1991 goto done;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001992
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001993 if (!do_swap_account ||
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001994 !page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1995 if (!page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001996 goto done_restock;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001997 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001998 page_counter_uncharge(&memcg->memsw, batch);
1999 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002000 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002001 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002002 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002003 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002004
Johannes Weiner6539cc02014-08-06 16:05:42 -07002005 if (batch > nr_pages) {
2006 batch = nr_pages;
2007 goto retry;
2008 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002009
Johannes Weiner06b078f2014-08-06 16:05:44 -07002010 /*
2011 * Unlike in global OOM situations, memcg is not in a physical
2012 * memory shortage. Allow dying and OOM-killed tasks to
2013 * bypass the last charges so that they can exit quickly and
2014 * free their memory.
2015 */
2016 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
2017 fatal_signal_pending(current) ||
2018 current->flags & PF_EXITING))
2019 goto bypass;
2020
2021 if (unlikely(task_in_memcg_oom(current)))
2022 goto nomem;
2023
Johannes Weiner6539cc02014-08-06 16:05:42 -07002024 if (!(gfp_mask & __GFP_WAIT))
2025 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002026
Johannes Weiner241994ed2015-02-11 15:26:06 -08002027 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
2028
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002029 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
2030 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002031
Johannes Weiner61e02c72014-08-06 16:08:16 -07002032 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07002033 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07002034
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002035 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002036 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002037 drained = true;
2038 goto retry;
2039 }
2040
Johannes Weiner28c34c22014-08-06 16:05:47 -07002041 if (gfp_mask & __GFP_NORETRY)
2042 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002043 /*
2044 * Even though the limit is exceeded at this point, reclaim
2045 * may have been able to free some pages. Retry the charge
2046 * before killing the task.
2047 *
2048 * Only for regular pages, though: huge pages are rather
2049 * unlikely to succeed so close to the limit, and we fall back
2050 * to regular pages anyway in case of failure.
2051 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002052 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002053 goto retry;
2054 /*
2055 * At task move, charge accounts can be doubly counted. So, it's
2056 * better to wait until the end of task_move if something is going on.
2057 */
2058 if (mem_cgroup_wait_acct_move(mem_over_limit))
2059 goto retry;
2060
Johannes Weiner9b130612014-08-06 16:05:51 -07002061 if (nr_retries--)
2062 goto retry;
2063
Johannes Weiner06b078f2014-08-06 16:05:44 -07002064 if (gfp_mask & __GFP_NOFAIL)
2065 goto bypass;
2066
Johannes Weiner6539cc02014-08-06 16:05:42 -07002067 if (fatal_signal_pending(current))
2068 goto bypass;
2069
Johannes Weiner241994ed2015-02-11 15:26:06 -08002070 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
2071
Johannes Weiner61e02c72014-08-06 16:08:16 -07002072 mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002073nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002074 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002075 return -ENOMEM;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002076bypass:
Johannes Weinerce00a962014-09-05 08:43:57 -04002077 return -EINTR;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002078
2079done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002080 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002081 if (batch > nr_pages)
2082 refill_stock(memcg, batch - nr_pages);
Vladimir Davydov7d638092015-06-10 11:14:46 -07002083 if (!(gfp_mask & __GFP_WAIT))
2084 goto done;
Johannes Weiner241994ed2015-02-11 15:26:06 -08002085 /*
2086 * If the hierarchy is above the normal consumption range,
2087 * make the charging task trim their excess contribution.
2088 */
2089 do {
2090 if (page_counter_read(&memcg->memory) <= memcg->high)
2091 continue;
2092 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
2093 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
2094 } while ((memcg = parent_mem_cgroup(memcg)));
Johannes Weiner6539cc02014-08-06 16:05:42 -07002095done:
Johannes Weiner05b84302014-08-06 16:05:59 -07002096 return ret;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002097}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002098
Johannes Weiner00501b52014-08-08 14:19:20 -07002099static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002100{
Johannes Weinerce00a962014-09-05 08:43:57 -04002101 if (mem_cgroup_is_root(memcg))
2102 return;
2103
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002104 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner05b84302014-08-06 16:05:59 -07002105 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002106 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002107
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002108 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002109}
2110
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002111static void lock_page_lru(struct page *page, int *isolated)
2112{
2113 struct zone *zone = page_zone(page);
2114
2115 spin_lock_irq(&zone->lru_lock);
2116 if (PageLRU(page)) {
2117 struct lruvec *lruvec;
2118
2119 lruvec = mem_cgroup_page_lruvec(page, zone);
2120 ClearPageLRU(page);
2121 del_page_from_lru_list(page, lruvec, page_lru(page));
2122 *isolated = 1;
2123 } else
2124 *isolated = 0;
2125}
2126
2127static void unlock_page_lru(struct page *page, int isolated)
2128{
2129 struct zone *zone = page_zone(page);
2130
2131 if (isolated) {
2132 struct lruvec *lruvec;
2133
2134 lruvec = mem_cgroup_page_lruvec(page, zone);
2135 VM_BUG_ON_PAGE(PageLRU(page), page);
2136 SetPageLRU(page);
2137 add_page_to_lru_list(page, lruvec, page_lru(page));
2138 }
2139 spin_unlock_irq(&zone->lru_lock);
2140}
2141
Johannes Weiner00501b52014-08-08 14:19:20 -07002142static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002143 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002144{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002145 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002146
Johannes Weiner1306a852014-12-10 15:44:52 -08002147 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002148
2149 /*
2150 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2151 * may already be on some other mem_cgroup's LRU. Take care of it.
2152 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002153 if (lrucare)
2154 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002155
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002156 /*
2157 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002158 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002159 *
2160 * - the page is uncharged
2161 *
2162 * - the page is off-LRU
2163 *
2164 * - an anonymous fault has exclusive page access, except for
2165 * a locked page table
2166 *
2167 * - a page cache insertion, a swapin fault, or a migration
2168 * have the page locked
2169 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002170 page->mem_cgroup = memcg;
Hugh Dickins3be91272008-02-07 00:14:19 -08002171
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002172 if (lrucare)
2173 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002174}
2175
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002176#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydovdbf22eb2015-02-10 14:11:41 -08002177int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp,
2178 unsigned long nr_pages)
Glauber Costa749c5412012-12-18 14:23:01 -08002179{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002180 struct page_counter *counter;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002181 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002182
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002183 ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter);
2184 if (ret < 0)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002185 return ret;
2186
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002187 ret = try_charge(memcg, gfp, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002188 if (ret == -EINTR) {
2189 /*
Johannes Weiner00501b52014-08-08 14:19:20 -07002190 * try_charge() chose to bypass to root due to OOM kill or
2191 * fatal signal. Since our only options are to either fail
2192 * the allocation or charge it to this cgroup, do it as a
2193 * temporary condition. But we can't fail. From a kmem/slab
2194 * perspective, the cache has already been selected, by
2195 * mem_cgroup_kmem_get_cache(), so it is too late to change
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002196 * our minds.
2197 *
2198 * This condition will only trigger if the task entered
Johannes Weiner00501b52014-08-08 14:19:20 -07002199 * memcg_charge_kmem in a sane state, but was OOM-killed
2200 * during try_charge() above. Tasks that were already dying
2201 * when the allocation triggers should have been already
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002202 * directed to the root cgroup in memcontrol.h
2203 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002204 page_counter_charge(&memcg->memory, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002205 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002206 page_counter_charge(&memcg->memsw, nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002207 css_get_many(&memcg->css, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002208 ret = 0;
2209 } else if (ret)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002210 page_counter_uncharge(&memcg->kmem, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002211
2212 return ret;
2213}
2214
Vladimir Davydovdbf22eb2015-02-10 14:11:41 -08002215void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002216{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002217 page_counter_uncharge(&memcg->memory, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002218 if (do_swap_account)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002219 page_counter_uncharge(&memcg->memsw, nr_pages);
Glauber Costa7de37682012-12-18 14:22:07 -08002220
Johannes Weiner64f21992014-12-10 15:42:45 -08002221 page_counter_uncharge(&memcg->kmem, nr_pages);
Glauber Costa7de37682012-12-18 14:22:07 -08002222
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002223 css_put_many(&memcg->css, nr_pages);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002224}
2225
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002226static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002227{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002228 int id, size;
2229 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002230
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002231 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002232 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2233 if (id < 0)
2234 return id;
2235
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002236 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002237 return id;
2238
2239 /*
2240 * There's no space for the new id in memcg_caches arrays,
2241 * so we have to grow them.
2242 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002243 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002244
2245 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002246 if (size < MEMCG_CACHES_MIN_SIZE)
2247 size = MEMCG_CACHES_MIN_SIZE;
2248 else if (size > MEMCG_CACHES_MAX_SIZE)
2249 size = MEMCG_CACHES_MAX_SIZE;
2250
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002251 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002252 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002253 err = memcg_update_all_list_lrus(size);
2254 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002255 memcg_nr_cache_ids = size;
2256
2257 up_write(&memcg_cache_ids_sem);
2258
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002259 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002260 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002261 return err;
2262 }
2263 return id;
2264}
2265
2266static void memcg_free_cache_id(int id)
2267{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002268 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002269}
2270
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002271struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002272 struct mem_cgroup *memcg;
2273 struct kmem_cache *cachep;
2274 struct work_struct work;
2275};
2276
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002277static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002278{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002279 struct memcg_kmem_cache_create_work *cw =
2280 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002281 struct mem_cgroup *memcg = cw->memcg;
2282 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002283
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002284 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002285
Vladimir Davydov5722d092014-04-07 15:39:24 -07002286 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002287 kfree(cw);
2288}
2289
2290/*
2291 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002292 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002293static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2294 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002295{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002296 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002297
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002298 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002299 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002300 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002301
2302 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002303
2304 cw->memcg = memcg;
2305 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002306 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002307
Glauber Costad7f25f82012-12-18 14:22:40 -08002308 schedule_work(&cw->work);
2309}
2310
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002311static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2312 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002313{
2314 /*
2315 * We need to stop accounting when we kmalloc, because if the
2316 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002317 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002318 *
2319 * However, it is better to enclose the whole function. Depending on
2320 * the debugging options enabled, INIT_WORK(), for instance, can
2321 * trigger an allocation. This too, will make us recurse. Because at
2322 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2323 * the safest choice is to do it like this, wrapping the whole function.
2324 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002325 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002326 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002327 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002328}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002329
Glauber Costad7f25f82012-12-18 14:22:40 -08002330/*
2331 * Return the kmem_cache we're supposed to use for a slab allocation.
2332 * We try to use the current memcg's version of the cache.
2333 *
2334 * If the cache does not exist yet, if we are the first user of it,
2335 * we either create it immediately, if possible, or create it asynchronously
2336 * in a workqueue.
2337 * In the latter case, we will let the current allocation go through with
2338 * the original cache.
2339 *
2340 * Can't be called in interrupt context or from kernel threads.
2341 * This function needs to be called with rcu_read_lock() held.
2342 */
Zhang Zhen056b7cc2014-12-12 16:55:38 -08002343struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002344{
2345 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002346 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002347 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002348
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002349 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002350
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002351 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002352 return cachep;
2353
Vladimir Davydov8135be52014-12-12 16:56:38 -08002354 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002355 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002356 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002357 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002358
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002359 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002360 if (likely(memcg_cachep))
2361 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002362
2363 /*
2364 * If we are in a safe context (can wait, and not in interrupt
2365 * context), we could be be predictable and return right away.
2366 * This would guarantee that the allocation being performed
2367 * already belongs in the new cache.
2368 *
2369 * However, there are some clashes that can arrive from locking.
2370 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002371 * memcg_create_kmem_cache, this means no further allocation
2372 * could happen with the slab_mutex held. So it's better to
2373 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002374 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002375 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002376out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002377 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002378 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002379}
Glauber Costad7f25f82012-12-18 14:22:40 -08002380
Vladimir Davydov8135be52014-12-12 16:56:38 -08002381void __memcg_kmem_put_cache(struct kmem_cache *cachep)
2382{
2383 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002384 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002385}
2386
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002387/*
2388 * We need to verify if the allocation against current->mm->owner's memcg is
2389 * possible for the given order. But the page is not allocated yet, so we'll
2390 * need a further commit step to do the final arrangements.
2391 *
2392 * It is possible for the task to switch cgroups in this mean time, so at
2393 * commit time, we can't rely on task conversion any longer. We'll then use
2394 * the handle argument to return to the caller which cgroup we should commit
2395 * against. We could also return the memcg directly and avoid the pointer
2396 * passing, but a boolean return value gives better semantics considering
2397 * the compiled-out case as well.
2398 *
2399 * Returning true means the allocation is possible.
2400 */
2401bool
2402__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
2403{
2404 struct mem_cgroup *memcg;
2405 int ret;
2406
2407 *_memcg = NULL;
Glauber Costa6d42c232013-07-08 16:00:00 -07002408
Johannes Weinerdf381972014-04-07 15:37:43 -07002409 memcg = get_mem_cgroup_from_mm(current->mm);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002410
Vladimir Davydovcf2b8fb2014-10-09 15:28:59 -07002411 if (!memcg_kmem_is_active(memcg)) {
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002412 css_put(&memcg->css);
2413 return true;
2414 }
2415
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002416 ret = memcg_charge_kmem(memcg, gfp, 1 << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002417 if (!ret)
2418 *_memcg = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002419
2420 css_put(&memcg->css);
2421 return (ret == 0);
2422}
2423
2424void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
2425 int order)
2426{
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002427 VM_BUG_ON(mem_cgroup_is_root(memcg));
2428
2429 /* The page allocation failed. Revert */
2430 if (!page) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002431 memcg_uncharge_kmem(memcg, 1 << order);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002432 return;
2433 }
Johannes Weiner1306a852014-12-10 15:44:52 -08002434 page->mem_cgroup = memcg;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002435}
2436
2437void __memcg_kmem_uncharge_pages(struct page *page, int order)
2438{
Johannes Weiner1306a852014-12-10 15:44:52 -08002439 struct mem_cgroup *memcg = page->mem_cgroup;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002440
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002441 if (!memcg)
2442 return;
2443
Sasha Levin309381fea2014-01-23 15:52:54 -08002444 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002445
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002446 memcg_uncharge_kmem(memcg, 1 << order);
Johannes Weiner1306a852014-12-10 15:44:52 -08002447 page->mem_cgroup = NULL;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002448}
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002449
2450struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr)
2451{
2452 struct mem_cgroup *memcg = NULL;
2453 struct kmem_cache *cachep;
2454 struct page *page;
2455
2456 page = virt_to_head_page(ptr);
2457 if (PageSlab(page)) {
2458 cachep = page->slab_cache;
2459 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002460 memcg = cachep->memcg_params.memcg;
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002461 } else
2462 /* page allocated by alloc_kmem_pages */
2463 memcg = page->mem_cgroup;
2464
2465 return memcg;
2466}
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002467#endif /* CONFIG_MEMCG_KMEM */
2468
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002469#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2470
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002471/*
2472 * Because tail pages are not marked as "used", set it. We're under
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002473 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2474 * charge/uncharge will be never happen and move_account() is done under
2475 * compound_lock(), so we don't have to take care of races.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002476 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002477void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002478{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002479 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002480
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002481 if (mem_cgroup_disabled())
2482 return;
David Rientjesb070e652013-05-07 16:18:09 -07002483
Johannes Weiner29833312014-12-10 15:44:02 -08002484 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002485 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002486
Johannes Weiner1306a852014-12-10 15:44:52 -08002487 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002488 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002489}
Hugh Dickins12d27102012-01-12 17:19:52 -08002490#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002491
Andrew Mortonc255a452012-07-31 16:43:02 -07002492#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002493static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2494 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002495{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002496 int val = (charge) ? 1 : -1;
2497 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002498}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002499
2500/**
2501 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2502 * @entry: swap entry to be moved
2503 * @from: mem_cgroup which the entry is moved from
2504 * @to: mem_cgroup which the entry is moved to
2505 *
2506 * It succeeds only when the swap_cgroup's record for this entry is the same
2507 * as the mem_cgroup's id of @from.
2508 *
2509 * Returns 0 on success, -EINVAL on failure.
2510 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002511 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002512 * both res and memsw, and called css_get().
2513 */
2514static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002515 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002516{
2517 unsigned short old_id, new_id;
2518
Li Zefan34c00c32013-09-23 16:56:01 +08002519 old_id = mem_cgroup_id(from);
2520 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002521
2522 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002523 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002524 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002525 return 0;
2526 }
2527 return -EINVAL;
2528}
2529#else
2530static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002531 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002532{
2533 return -EINVAL;
2534}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002535#endif
2536
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002537static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002538
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002539static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002540 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002541{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002542 unsigned long curusage;
2543 unsigned long oldusage;
2544 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002545 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002546 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002547
2548 /*
2549 * For keeping hierarchical_reclaim simple, how long we should retry
2550 * is depends on callers. We set our retry-count to be function
2551 * of # of children which we should visit in this loop.
2552 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002553 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2554 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002555
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002556 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002557
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002558 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002559 if (signal_pending(current)) {
2560 ret = -EINTR;
2561 break;
2562 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002563
2564 mutex_lock(&memcg_limit_mutex);
2565 if (limit > memcg->memsw.limit) {
2566 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002567 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002568 break;
2569 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002570 if (limit > memcg->memory.limit)
2571 enlarge = true;
2572 ret = page_counter_limit(&memcg->memory, limit);
2573 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002574
2575 if (!ret)
2576 break;
2577
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002578 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2579
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002580 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002581 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002582 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002583 retry_count--;
2584 else
2585 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002586 } while (retry_count);
2587
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002588 if (!ret && enlarge)
2589 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002590
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002591 return ret;
2592}
2593
Li Zefan338c8432009-06-17 16:27:15 -07002594static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002595 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002596{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002597 unsigned long curusage;
2598 unsigned long oldusage;
2599 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002600 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002601 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002602
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002603 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002604 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2605 mem_cgroup_count_children(memcg);
2606
2607 oldusage = page_counter_read(&memcg->memsw);
2608
2609 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002610 if (signal_pending(current)) {
2611 ret = -EINTR;
2612 break;
2613 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002614
2615 mutex_lock(&memcg_limit_mutex);
2616 if (limit < memcg->memory.limit) {
2617 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002618 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002619 break;
2620 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002621 if (limit > memcg->memsw.limit)
2622 enlarge = true;
2623 ret = page_counter_limit(&memcg->memsw, limit);
2624 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002625
2626 if (!ret)
2627 break;
2628
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002629 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2630
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002631 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002632 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002633 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002634 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002635 else
2636 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002637 } while (retry_count);
2638
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002639 if (!ret && enlarge)
2640 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002641
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002642 return ret;
2643}
2644
Andrew Morton0608f432013-09-24 15:27:41 -07002645unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2646 gfp_t gfp_mask,
2647 unsigned long *total_scanned)
2648{
2649 unsigned long nr_reclaimed = 0;
2650 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2651 unsigned long reclaimed;
2652 int loop = 0;
2653 struct mem_cgroup_tree_per_zone *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002654 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002655 unsigned long nr_scanned;
2656
2657 if (order > 0)
2658 return 0;
2659
2660 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
2661 /*
2662 * This loop can run a while, specially if mem_cgroup's continuously
2663 * keep exceeding their soft limit and putting the system under
2664 * pressure
2665 */
2666 do {
2667 if (next_mz)
2668 mz = next_mz;
2669 else
2670 mz = mem_cgroup_largest_soft_limit_node(mctz);
2671 if (!mz)
2672 break;
2673
2674 nr_scanned = 0;
2675 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
2676 gfp_mask, &nr_scanned);
2677 nr_reclaimed += reclaimed;
2678 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002679 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002680 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002681
2682 /*
2683 * If we failed to reclaim anything from this memory cgroup
2684 * it is time to move on to the next cgroup
2685 */
2686 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002687 if (!reclaimed)
2688 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2689
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002690 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002691 /*
2692 * One school of thought says that we should not add
2693 * back the node to the tree if reclaim returns 0.
2694 * But our reclaim could return 0, simply because due
2695 * to priority we are exposing a smaller subset of
2696 * memory to reclaim from. Consider this as a longer
2697 * term TODO.
2698 */
2699 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002700 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002701 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002702 css_put(&mz->memcg->css);
2703 loop++;
2704 /*
2705 * Could not reclaim anything and there are no more
2706 * mem cgroups to try or we seem to be looping without
2707 * reclaiming anything.
2708 */
2709 if (!nr_reclaimed &&
2710 (next_mz == NULL ||
2711 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2712 break;
2713 } while (!nr_reclaimed);
2714 if (next_mz)
2715 css_put(&next_mz->memcg->css);
2716 return nr_reclaimed;
2717}
2718
Tejun Heoea280e72014-05-16 13:22:48 -04002719/*
2720 * Test whether @memcg has children, dead or alive. Note that this
2721 * function doesn't care whether @memcg has use_hierarchy enabled and
2722 * returns %true if there are child csses according to the cgroup
2723 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2724 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002725static inline bool memcg_has_children(struct mem_cgroup *memcg)
2726{
Tejun Heoea280e72014-05-16 13:22:48 -04002727 bool ret;
2728
Johannes Weiner696ac172013-10-31 16:34:15 -07002729 /*
Tejun Heoea280e72014-05-16 13:22:48 -04002730 * The lock does not prevent addition or deletion of children, but
2731 * it prevents a new child from being initialized based on this
2732 * parent in css_online(), so it's enough to decide whether
2733 * hierarchically inherited attributes can still be changed or not.
Johannes Weiner696ac172013-10-31 16:34:15 -07002734 */
Tejun Heoea280e72014-05-16 13:22:48 -04002735 lockdep_assert_held(&memcg_create_mutex);
2736
2737 rcu_read_lock();
2738 ret = css_next_child(NULL, &memcg->css);
2739 rcu_read_unlock();
2740 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002741}
2742
2743/*
Michal Hockoc26251f2012-10-26 13:37:28 +02002744 * Reclaims as many pages from the given memcg as possible and moves
2745 * the rest to the parent.
2746 *
2747 * Caller is responsible for holding css reference for memcg.
2748 */
2749static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2750{
2751 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002752
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002753 /* we call try-to-free pages for make this cgroup empty */
2754 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002755 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002756 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002757 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002758
Michal Hockoc26251f2012-10-26 13:37:28 +02002759 if (signal_pending(current))
2760 return -EINTR;
2761
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002762 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2763 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002764 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002765 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002766 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002767 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002768 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002769
2770 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002771
2772 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002773}
2774
Tejun Heo6770c642014-05-13 12:16:21 -04002775static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2776 char *buf, size_t nbytes,
2777 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002778{
Tejun Heo6770c642014-05-13 12:16:21 -04002779 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002780
Michal Hockod8423012012-10-26 13:37:29 +02002781 if (mem_cgroup_is_root(memcg))
2782 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002783 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002784}
2785
Tejun Heo182446d2013-08-08 20:11:24 -04002786static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2787 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002788{
Tejun Heo182446d2013-08-08 20:11:24 -04002789 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002790}
2791
Tejun Heo182446d2013-08-08 20:11:24 -04002792static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2793 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002794{
2795 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002796 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002797 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002798
Glauber Costa09998212013-02-22 16:34:55 -08002799 mutex_lock(&memcg_create_mutex);
Glauber Costa567fb432012-07-31 16:43:07 -07002800
2801 if (memcg->use_hierarchy == val)
2802 goto out;
2803
Balbir Singh18f59ea2009-01-07 18:08:07 -08002804 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002805 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002806 * in the child subtrees. If it is unset, then the change can
2807 * occur, provided the current cgroup has no children.
2808 *
2809 * For the root cgroup, parent_mem is NULL, we allow value to be
2810 * set if there are no children.
2811 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002812 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002813 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002814 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002815 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002816 else
2817 retval = -EBUSY;
2818 } else
2819 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002820
2821out:
Glauber Costa09998212013-02-22 16:34:55 -08002822 mutex_unlock(&memcg_create_mutex);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002823
2824 return retval;
2825}
2826
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002827static unsigned long tree_stat(struct mem_cgroup *memcg,
2828 enum mem_cgroup_stat_index idx)
Johannes Weinerce00a962014-09-05 08:43:57 -04002829{
2830 struct mem_cgroup *iter;
Greg Thelen484ebb32015-10-01 15:37:05 -07002831 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002832
Johannes Weinerce00a962014-09-05 08:43:57 -04002833 for_each_mem_cgroup_tree(iter, memcg)
2834 val += mem_cgroup_read_stat(iter, idx);
2835
Johannes Weinerce00a962014-09-05 08:43:57 -04002836 return val;
2837}
2838
2839static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
2840{
2841 u64 val;
2842
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002843 if (mem_cgroup_is_root(memcg)) {
2844 val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE);
2845 val += tree_stat(memcg, MEM_CGROUP_STAT_RSS);
2846 if (swap)
2847 val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP);
2848 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002849 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002850 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002851 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002852 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002853 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002854 return val << PAGE_SHIFT;
2855}
2856
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002857enum {
2858 RES_USAGE,
2859 RES_LIMIT,
2860 RES_MAX_USAGE,
2861 RES_FAILCNT,
2862 RES_SOFT_LIMIT,
2863};
Johannes Weinerce00a962014-09-05 08:43:57 -04002864
Tejun Heo791badb2013-12-05 12:28:02 -05002865static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002866 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002867{
Tejun Heo182446d2013-08-08 20:11:24 -04002868 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002869 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002870
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002871 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002872 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002873 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002874 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002875 case _MEMSWAP:
2876 counter = &memcg->memsw;
2877 break;
2878 case _KMEM:
2879 counter = &memcg->kmem;
2880 break;
2881 default:
2882 BUG();
2883 }
2884
2885 switch (MEMFILE_ATTR(cft->private)) {
2886 case RES_USAGE:
2887 if (counter == &memcg->memory)
2888 return mem_cgroup_usage(memcg, false);
2889 if (counter == &memcg->memsw)
2890 return mem_cgroup_usage(memcg, true);
2891 return (u64)page_counter_read(counter) * PAGE_SIZE;
2892 case RES_LIMIT:
2893 return (u64)counter->limit * PAGE_SIZE;
2894 case RES_MAX_USAGE:
2895 return (u64)counter->watermark * PAGE_SIZE;
2896 case RES_FAILCNT:
2897 return counter->failcnt;
2898 case RES_SOFT_LIMIT:
2899 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002900 default:
2901 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002902 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002903}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002904
Glauber Costa510fc4e2012-12-18 14:21:47 -08002905#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08002906static int memcg_activate_kmem(struct mem_cgroup *memcg,
2907 unsigned long nr_pages)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002908{
2909 int err = 0;
2910 int memcg_id;
2911
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002912 BUG_ON(memcg->kmemcg_id >= 0);
Vladimir Davydov2788cf02015-02-12 14:59:38 -08002913 BUG_ON(memcg->kmem_acct_activated);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002914 BUG_ON(memcg->kmem_acct_active);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002915
2916 /*
Glauber Costa510fc4e2012-12-18 14:21:47 -08002917 * For simplicity, we won't allow this to be disabled. It also can't
2918 * be changed if the cgroup has children already, or if tasks had
2919 * already joined.
2920 *
2921 * If tasks join before we set the limit, a person looking at
2922 * kmem.usage_in_bytes will have no way to determine when it took
2923 * place, which makes the value quite meaningless.
2924 *
2925 * After it first became limited, changes in the value of the limit are
2926 * of course permitted.
Glauber Costa510fc4e2012-12-18 14:21:47 -08002927 */
Glauber Costa09998212013-02-22 16:34:55 -08002928 mutex_lock(&memcg_create_mutex);
Tejun Heoea280e72014-05-16 13:22:48 -04002929 if (cgroup_has_tasks(memcg->css.cgroup) ||
2930 (memcg->use_hierarchy && memcg_has_children(memcg)))
Vladimir Davydovd6441632014-01-23 15:53:09 -08002931 err = -EBUSY;
Glauber Costa09998212013-02-22 16:34:55 -08002932 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002933 if (err)
2934 goto out;
2935
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002936 memcg_id = memcg_alloc_cache_id();
Vladimir Davydovd6441632014-01-23 15:53:09 -08002937 if (memcg_id < 0) {
2938 err = memcg_id;
2939 goto out;
2940 }
2941
Vladimir Davydovd6441632014-01-23 15:53:09 -08002942 /*
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002943 * We couldn't have accounted to this cgroup, because it hasn't got
2944 * activated yet, so this should succeed.
Vladimir Davydovd6441632014-01-23 15:53:09 -08002945 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002946 err = page_counter_limit(&memcg->kmem, nr_pages);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002947 VM_BUG_ON(err);
2948
2949 static_key_slow_inc(&memcg_kmem_enabled_key);
2950 /*
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002951 * A memory cgroup is considered kmem-active as soon as it gets
2952 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002953 * guarantee no one starts accounting before all call sites are
2954 * patched.
2955 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002956 memcg->kmemcg_id = memcg_id;
Vladimir Davydov2788cf02015-02-12 14:59:38 -08002957 memcg->kmem_acct_activated = true;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002958 memcg->kmem_acct_active = true;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002959out:
Vladimir Davydovd6441632014-01-23 15:53:09 -08002960 return err;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002961}
2962
Vladimir Davydovd6441632014-01-23 15:53:09 -08002963static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002964 unsigned long limit)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002965{
2966 int ret;
2967
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002968 mutex_lock(&memcg_limit_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002969 if (!memcg_kmem_is_active(memcg))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002970 ret = memcg_activate_kmem(memcg, limit);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002971 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002972 ret = page_counter_limit(&memcg->kmem, limit);
2973 mutex_unlock(&memcg_limit_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002974 return ret;
2975}
2976
Glauber Costa55007d82012-12-18 14:22:38 -08002977static int memcg_propagate_kmem(struct mem_cgroup *memcg)
Glauber Costa510fc4e2012-12-18 14:21:47 -08002978{
Glauber Costa55007d82012-12-18 14:22:38 -08002979 int ret = 0;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002980 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002981
Glauber Costa510fc4e2012-12-18 14:21:47 -08002982 if (!parent)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002983 return 0;
Glauber Costa55007d82012-12-18 14:22:38 -08002984
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08002985 mutex_lock(&memcg_limit_mutex);
Glauber Costaa8964b92012-12-18 14:22:09 -08002986 /*
Vladimir Davydovd6441632014-01-23 15:53:09 -08002987 * If the parent cgroup is not kmem-active now, it cannot be activated
2988 * after this point, because it has at least one child already.
Glauber Costaa8964b92012-12-18 14:22:09 -08002989 */
Vladimir Davydovd6441632014-01-23 15:53:09 -08002990 if (memcg_kmem_is_active(parent))
Vladimir Davydov8c0145b2014-12-10 15:43:48 -08002991 ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX);
2992 mutex_unlock(&memcg_limit_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -08002993 return ret;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002994}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002995#else
2996static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002997 unsigned long limit)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002998{
2999 return -EINVAL;
3000}
Hugh Dickins6d0439902013-02-22 16:35:50 -08003001#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa510fc4e2012-12-18 14:21:47 -08003002
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003003/*
3004 * The user of this function is...
3005 * RES_LIMIT.
3006 */
Tejun Heo451af502014-05-13 12:16:21 -04003007static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
3008 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003009{
Tejun Heo451af502014-05-13 12:16:21 -04003010 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003011 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003012 int ret;
3013
Tejun Heo451af502014-05-13 12:16:21 -04003014 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08003015 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003016 if (ret)
3017 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07003018
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003019 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003020 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003021 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3022 ret = -EINVAL;
3023 break;
3024 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003025 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3026 case _MEM:
3027 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003028 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003029 case _MEMSWAP:
3030 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3031 break;
3032 case _KMEM:
3033 ret = memcg_update_kmem_limit(memcg, nr_pages);
3034 break;
3035 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003036 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003037 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003038 memcg->soft_limit = nr_pages;
3039 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003040 break;
3041 }
Tejun Heo451af502014-05-13 12:16:21 -04003042 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003043}
3044
Tejun Heo6770c642014-05-13 12:16:21 -04003045static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3046 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003047{
Tejun Heo6770c642014-05-13 12:16:21 -04003048 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003049 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003050
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003051 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3052 case _MEM:
3053 counter = &memcg->memory;
3054 break;
3055 case _MEMSWAP:
3056 counter = &memcg->memsw;
3057 break;
3058 case _KMEM:
3059 counter = &memcg->kmem;
3060 break;
3061 default:
3062 BUG();
3063 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003064
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003065 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003066 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003067 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003068 break;
3069 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003070 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003071 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003072 default:
3073 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003074 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003075
Tejun Heo6770c642014-05-13 12:16:21 -04003076 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003077}
3078
Tejun Heo182446d2013-08-08 20:11:24 -04003079static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003080 struct cftype *cft)
3081{
Tejun Heo182446d2013-08-08 20:11:24 -04003082 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003083}
3084
Daisuke Nishimura02491442010-03-10 15:22:17 -08003085#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003086static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003087 struct cftype *cft, u64 val)
3088{
Tejun Heo182446d2013-08-08 20:11:24 -04003089 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003090
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003091 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003092 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003093
Glauber Costaee5e8472013-02-22 16:34:50 -08003094 /*
3095 * No kind of locking is needed in here, because ->can_attach() will
3096 * check this value once in the beginning of the process, and then carry
3097 * on with stale data. This means that changes to this value will only
3098 * affect task migrations starting after the change.
3099 */
3100 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003101 return 0;
3102}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003103#else
Tejun Heo182446d2013-08-08 20:11:24 -04003104static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003105 struct cftype *cft, u64 val)
3106{
3107 return -ENOSYS;
3108}
3109#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003110
Ying Han406eb0c2011-05-26 16:25:37 -07003111#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003112static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003113{
Greg Thelen25485de2013-11-12 15:07:40 -08003114 struct numa_stat {
3115 const char *name;
3116 unsigned int lru_mask;
3117 };
3118
3119 static const struct numa_stat stats[] = {
3120 { "total", LRU_ALL },
3121 { "file", LRU_ALL_FILE },
3122 { "anon", LRU_ALL_ANON },
3123 { "unevictable", BIT(LRU_UNEVICTABLE) },
3124 };
3125 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003126 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003127 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003128 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003129
Greg Thelen25485de2013-11-12 15:07:40 -08003130 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3131 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3132 seq_printf(m, "%s=%lu", stat->name, nr);
3133 for_each_node_state(nid, N_MEMORY) {
3134 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3135 stat->lru_mask);
3136 seq_printf(m, " N%d=%lu", nid, nr);
3137 }
3138 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003139 }
Ying Han406eb0c2011-05-26 16:25:37 -07003140
Ying Han071aee12013-11-12 15:07:41 -08003141 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3142 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003143
Ying Han071aee12013-11-12 15:07:41 -08003144 nr = 0;
3145 for_each_mem_cgroup_tree(iter, memcg)
3146 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3147 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3148 for_each_node_state(nid, N_MEMORY) {
3149 nr = 0;
3150 for_each_mem_cgroup_tree(iter, memcg)
3151 nr += mem_cgroup_node_nr_lru_pages(
3152 iter, nid, stat->lru_mask);
3153 seq_printf(m, " N%d=%lu", nid, nr);
3154 }
3155 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003156 }
Ying Han406eb0c2011-05-26 16:25:37 -07003157
Ying Han406eb0c2011-05-26 16:25:37 -07003158 return 0;
3159}
3160#endif /* CONFIG_NUMA */
3161
Tejun Heo2da8ca82013-12-05 12:28:04 -05003162static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003163{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003164 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003165 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003166 struct mem_cgroup *mi;
3167 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003168
Greg Thelen0ca44b12015-02-11 15:25:58 -08003169 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3170 MEM_CGROUP_STAT_NSTATS);
3171 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3172 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003173 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3174
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003175 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07003176 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003177 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07003178 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003179 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003180 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003181
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003182 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3183 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3184 mem_cgroup_read_events(memcg, i));
3185
3186 for (i = 0; i < NR_LRU_LISTS; i++)
3187 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3188 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3189
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003190 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003191 memory = memsw = PAGE_COUNTER_MAX;
3192 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3193 memory = min(memory, mi->memory.limit);
3194 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003195 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003196 seq_printf(m, "hierarchical_memory_limit %llu\n",
3197 (u64)memory * PAGE_SIZE);
3198 if (do_swap_account)
3199 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3200 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003201
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003202 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003203 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003204
Kamezawa Hiroyukibff6bb82012-07-31 16:41:38 -07003205 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003206 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003207 for_each_mem_cgroup_tree(mi, memcg)
3208 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
Greg Thelen484ebb32015-10-01 15:37:05 -07003209 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003210 }
3211
3212 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3213 unsigned long long val = 0;
3214
3215 for_each_mem_cgroup_tree(mi, memcg)
3216 val += mem_cgroup_read_events(mi, i);
3217 seq_printf(m, "total_%s %llu\n",
3218 mem_cgroup_events_names[i], val);
3219 }
3220
3221 for (i = 0; i < NR_LRU_LISTS; i++) {
3222 unsigned long long val = 0;
3223
3224 for_each_mem_cgroup_tree(mi, memcg)
3225 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3226 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003227 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003228
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003229#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003230 {
3231 int nid, zid;
3232 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003233 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003234 unsigned long recent_rotated[2] = {0, 0};
3235 unsigned long recent_scanned[2] = {0, 0};
3236
3237 for_each_online_node(nid)
3238 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Jianyu Zhane2318752014-06-06 14:38:20 -07003239 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
Hugh Dickins89abfab2012-05-29 15:06:53 -07003240 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003241
Hugh Dickins89abfab2012-05-29 15:06:53 -07003242 recent_rotated[0] += rstat->recent_rotated[0];
3243 recent_rotated[1] += rstat->recent_rotated[1];
3244 recent_scanned[0] += rstat->recent_scanned[0];
3245 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003246 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003247 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3248 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3249 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3250 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003251 }
3252#endif
3253
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003254 return 0;
3255}
3256
Tejun Heo182446d2013-08-08 20:11:24 -04003257static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3258 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003259{
Tejun Heo182446d2013-08-08 20:11:24 -04003260 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003261
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003262 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003263}
3264
Tejun Heo182446d2013-08-08 20:11:24 -04003265static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3266 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003267{
Tejun Heo182446d2013-08-08 20:11:24 -04003268 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003269
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003270 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003271 return -EINVAL;
3272
Linus Torvalds14208b02014-06-09 15:03:33 -07003273 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003274 memcg->swappiness = val;
3275 else
3276 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003277
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003278 return 0;
3279}
3280
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003281static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3282{
3283 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003284 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003285 int i;
3286
3287 rcu_read_lock();
3288 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003289 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003290 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003291 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003292
3293 if (!t)
3294 goto unlock;
3295
Johannes Weinerce00a962014-09-05 08:43:57 -04003296 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003297
3298 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003299 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003300 * If it's not true, a threshold was crossed after last
3301 * call of __mem_cgroup_threshold().
3302 */
Phil Carmody5407a562010-05-26 14:42:42 -07003303 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003304
3305 /*
3306 * Iterate backward over array of thresholds starting from
3307 * current_threshold and check if a threshold is crossed.
3308 * If none of thresholds below usage is crossed, we read
3309 * only one element of the array here.
3310 */
3311 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3312 eventfd_signal(t->entries[i].eventfd, 1);
3313
3314 /* i = current_threshold + 1 */
3315 i++;
3316
3317 /*
3318 * Iterate forward over array of thresholds starting from
3319 * current_threshold+1 and check if a threshold is crossed.
3320 * If none of thresholds above usage is crossed, we read
3321 * only one element of the array here.
3322 */
3323 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3324 eventfd_signal(t->entries[i].eventfd, 1);
3325
3326 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003327 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003328unlock:
3329 rcu_read_unlock();
3330}
3331
3332static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3333{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003334 while (memcg) {
3335 __mem_cgroup_threshold(memcg, false);
3336 if (do_swap_account)
3337 __mem_cgroup_threshold(memcg, true);
3338
3339 memcg = parent_mem_cgroup(memcg);
3340 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003341}
3342
3343static int compare_thresholds(const void *a, const void *b)
3344{
3345 const struct mem_cgroup_threshold *_a = a;
3346 const struct mem_cgroup_threshold *_b = b;
3347
Greg Thelen2bff24a2013-09-11 14:23:08 -07003348 if (_a->threshold > _b->threshold)
3349 return 1;
3350
3351 if (_a->threshold < _b->threshold)
3352 return -1;
3353
3354 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003355}
3356
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003357static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003358{
3359 struct mem_cgroup_eventfd_list *ev;
3360
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003361 spin_lock(&memcg_oom_lock);
3362
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003363 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003364 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003365
3366 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003367 return 0;
3368}
3369
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003370static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003371{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003372 struct mem_cgroup *iter;
3373
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003374 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003375 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003376}
3377
Tejun Heo59b6f872013-11-22 18:20:43 -05003378static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003379 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003380{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003381 struct mem_cgroup_thresholds *thresholds;
3382 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003383 unsigned long threshold;
3384 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003385 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003386
Johannes Weiner650c5e52015-02-11 15:26:03 -08003387 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003388 if (ret)
3389 return ret;
Shaohua Li424cdc12015-10-15 15:28:29 -07003390 threshold <<= PAGE_SHIFT;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003391
3392 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003393
Johannes Weiner05b84302014-08-06 16:05:59 -07003394 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003395 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003396 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003397 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003398 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003399 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003400 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003401 BUG();
3402
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003403 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003404 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003405 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3406
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003407 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003408
3409 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003410 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003411 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003412 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003413 ret = -ENOMEM;
3414 goto unlock;
3415 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003416 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003417
3418 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003419 if (thresholds->primary) {
3420 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003421 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003422 }
3423
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003424 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003425 new->entries[size - 1].eventfd = eventfd;
3426 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003427
3428 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003429 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003430 compare_thresholds, NULL);
3431
3432 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003433 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003434 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003435 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003436 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003437 * new->current_threshold will not be used until
3438 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003439 * it here.
3440 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003441 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003442 } else
3443 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003444 }
3445
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003446 /* Free old spare buffer and save old primary buffer as spare */
3447 kfree(thresholds->spare);
3448 thresholds->spare = thresholds->primary;
3449
3450 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003451
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003452 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003453 synchronize_rcu();
3454
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003455unlock:
3456 mutex_unlock(&memcg->thresholds_lock);
3457
3458 return ret;
3459}
3460
Tejun Heo59b6f872013-11-22 18:20:43 -05003461static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003462 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003463{
Tejun Heo59b6f872013-11-22 18:20:43 -05003464 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003465}
3466
Tejun Heo59b6f872013-11-22 18:20:43 -05003467static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003468 struct eventfd_ctx *eventfd, const char *args)
3469{
Tejun Heo59b6f872013-11-22 18:20:43 -05003470 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003471}
3472
Tejun Heo59b6f872013-11-22 18:20:43 -05003473static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003474 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003475{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003476 struct mem_cgroup_thresholds *thresholds;
3477 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003478 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003479 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003480
3481 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003482
3483 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003484 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003485 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003486 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003487 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003488 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003489 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003490 BUG();
3491
Anton Vorontsov371528c2012-02-24 05:14:46 +04003492 if (!thresholds->primary)
3493 goto unlock;
3494
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003495 /* Check if a threshold crossed before removing */
3496 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3497
3498 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003499 size = 0;
3500 for (i = 0; i < thresholds->primary->size; i++) {
3501 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003502 size++;
3503 }
3504
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003505 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003506
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003507 /* Set thresholds array to NULL if we don't have thresholds */
3508 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003509 kfree(new);
3510 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003511 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003512 }
3513
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003514 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003515
3516 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003517 new->current_threshold = -1;
3518 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3519 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003520 continue;
3521
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003522 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003523 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003524 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003525 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003526 * until rcu_assign_pointer(), so it's safe to increment
3527 * it here.
3528 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003529 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003530 }
3531 j++;
3532 }
3533
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003534swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003535 /* Swap primary and spare array */
3536 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003537 /* If all events are unregistered, free the spare array */
3538 if (!new) {
3539 kfree(thresholds->spare);
3540 thresholds->spare = NULL;
3541 }
3542
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003543 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003544
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003545 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003546 synchronize_rcu();
Anton Vorontsov371528c2012-02-24 05:14:46 +04003547unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003548 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003549}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003550
Tejun Heo59b6f872013-11-22 18:20:43 -05003551static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003552 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003553{
Tejun Heo59b6f872013-11-22 18:20:43 -05003554 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003555}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003556
Tejun Heo59b6f872013-11-22 18:20:43 -05003557static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003558 struct eventfd_ctx *eventfd)
3559{
Tejun Heo59b6f872013-11-22 18:20:43 -05003560 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003561}
3562
Tejun Heo59b6f872013-11-22 18:20:43 -05003563static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003564 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003565{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003566 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003567
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003568 event = kmalloc(sizeof(*event), GFP_KERNEL);
3569 if (!event)
3570 return -ENOMEM;
3571
Michal Hocko1af8efe2011-07-26 16:08:24 -07003572 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003573
3574 event->eventfd = eventfd;
3575 list_add(&event->list, &memcg->oom_notify);
3576
3577 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07003578 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003579 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003580 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003581
3582 return 0;
3583}
3584
Tejun Heo59b6f872013-11-22 18:20:43 -05003585static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003586 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003587{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003588 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003589
Michal Hocko1af8efe2011-07-26 16:08:24 -07003590 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003591
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003592 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003593 if (ev->eventfd == eventfd) {
3594 list_del(&ev->list);
3595 kfree(ev);
3596 }
3597 }
3598
Michal Hocko1af8efe2011-07-26 16:08:24 -07003599 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003600}
3601
Tejun Heo2da8ca82013-12-05 12:28:04 -05003602static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003603{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003604 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003605
Tejun Heo791badb2013-12-05 12:28:02 -05003606 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07003607 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003608 return 0;
3609}
3610
Tejun Heo182446d2013-08-08 20:11:24 -04003611static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003612 struct cftype *cft, u64 val)
3613{
Tejun Heo182446d2013-08-08 20:11:24 -04003614 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003615
3616 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003617 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003618 return -EINVAL;
3619
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003620 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003621 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003622 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003623
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003624 return 0;
3625}
3626
Andrew Mortonc255a452012-07-31 16:43:02 -07003627#ifdef CONFIG_MEMCG_KMEM
Glauber Costacbe128e32012-04-09 19:36:34 -03003628static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00003629{
Glauber Costa55007d82012-12-18 14:22:38 -08003630 int ret;
3631
Glauber Costa55007d82012-12-18 14:22:38 -08003632 ret = memcg_propagate_kmem(memcg);
3633 if (ret)
3634 return ret;
Glauber Costa2633d7a2012-12-18 14:22:34 -08003635
Glauber Costa1d62e432012-04-09 19:36:33 -03003636 return mem_cgroup_sockets_init(memcg, ss);
Michel Lespinasse573b4002013-04-29 15:08:13 -07003637}
Glauber Costae5671df2011-12-11 21:47:01 +00003638
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003639static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3640{
Vladimir Davydov2788cf02015-02-12 14:59:38 -08003641 struct cgroup_subsys_state *css;
3642 struct mem_cgroup *parent, *child;
3643 int kmemcg_id;
3644
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003645 if (!memcg->kmem_acct_active)
3646 return;
3647
3648 /*
3649 * Clear the 'active' flag before clearing memcg_caches arrays entries.
3650 * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it
3651 * guarantees no cache will be created for this cgroup after we are
3652 * done (see memcg_create_kmem_cache()).
3653 */
3654 memcg->kmem_acct_active = false;
3655
3656 memcg_deactivate_kmem_caches(memcg);
Vladimir Davydov2788cf02015-02-12 14:59:38 -08003657
3658 kmemcg_id = memcg->kmemcg_id;
3659 BUG_ON(kmemcg_id < 0);
3660
3661 parent = parent_mem_cgroup(memcg);
3662 if (!parent)
3663 parent = root_mem_cgroup;
3664
3665 /*
3666 * Change kmemcg_id of this cgroup and all its descendants to the
3667 * parent's id, and then move all entries from this cgroup's list_lrus
3668 * to ones of the parent. After we have finished, all list_lrus
3669 * corresponding to this cgroup are guaranteed to remain empty. The
3670 * ordering is imposed by list_lru_node->lock taken by
3671 * memcg_drain_all_list_lrus().
3672 */
3673 css_for_each_descendant_pre(css, &memcg->css) {
3674 child = mem_cgroup_from_css(css);
3675 BUG_ON(child->kmemcg_id != kmemcg_id);
3676 child->kmemcg_id = parent->kmemcg_id;
3677 if (!memcg->use_hierarchy)
3678 break;
3679 }
3680 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
3681
3682 memcg_free_cache_id(kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003683}
3684
Li Zefan10d5ebf2013-07-08 16:00:33 -07003685static void memcg_destroy_kmem(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +00003686{
Vladimir Davydovf48b80a2015-02-12 14:59:56 -08003687 if (memcg->kmem_acct_activated) {
3688 memcg_destroy_kmem_caches(memcg);
3689 static_key_slow_dec(&memcg_kmem_enabled_key);
3690 WARN_ON(page_counter_read(&memcg->kmem));
3691 }
Glauber Costa1d62e432012-04-09 19:36:33 -03003692 mem_cgroup_sockets_destroy(memcg);
Li Zefan10d5ebf2013-07-08 16:00:33 -07003693}
Glauber Costae5671df2011-12-11 21:47:01 +00003694#else
Glauber Costacbe128e32012-04-09 19:36:34 -03003695static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costae5671df2011-12-11 21:47:01 +00003696{
3697 return 0;
3698}
Glauber Costad1a4c0b2011-12-11 21:47:04 +00003699
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003700static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3701{
3702}
3703
Li Zefan10d5ebf2013-07-08 16:00:33 -07003704static void memcg_destroy_kmem(struct mem_cgroup *memcg)
3705{
3706}
Glauber Costae5671df2011-12-11 21:47:01 +00003707#endif
3708
Tejun Heo52ebea72015-05-22 17:13:37 -04003709#ifdef CONFIG_CGROUP_WRITEBACK
3710
3711struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3712{
3713 return &memcg->cgwb_list;
3714}
3715
Tejun Heo841710a2015-05-22 18:23:33 -04003716static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3717{
3718 return wb_domain_init(&memcg->cgwb_domain, gfp);
3719}
3720
3721static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3722{
3723 wb_domain_exit(&memcg->cgwb_domain);
3724}
3725
Tejun Heo2529bb32015-05-22 18:23:34 -04003726static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3727{
3728 wb_domain_size_changed(&memcg->cgwb_domain);
3729}
3730
Tejun Heo841710a2015-05-22 18:23:33 -04003731struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3732{
3733 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3734
3735 if (!memcg->css.parent)
3736 return NULL;
3737
3738 return &memcg->cgwb_domain;
3739}
3740
Tejun Heoc2aa7232015-05-22 18:23:35 -04003741/**
3742 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3743 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003744 * @pfilepages: out parameter for number of file pages
3745 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04003746 * @pdirty: out parameter for number of dirty pages
3747 * @pwriteback: out parameter for number of pages under writeback
3748 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003749 * Determine the numbers of file, headroom, dirty, and writeback pages in
3750 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3751 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003752 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003753 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3754 * headroom is calculated as the lowest headroom of itself and the
3755 * ancestors. Note that this doesn't consider the actual amount of
3756 * available memory in the system. The caller should further cap
3757 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003758 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003759void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3760 unsigned long *pheadroom, unsigned long *pdirty,
3761 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04003762{
3763 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3764 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003765
3766 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3767
3768 /* this should eventually include NR_UNSTABLE_NFS */
3769 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003770 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3771 (1 << LRU_ACTIVE_FILE));
3772 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003773
Tejun Heoc2aa7232015-05-22 18:23:35 -04003774 while ((parent = parent_mem_cgroup(memcg))) {
3775 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3776 unsigned long used = page_counter_read(&memcg->memory);
3777
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003778 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04003779 memcg = parent;
3780 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04003781}
3782
Tejun Heo841710a2015-05-22 18:23:33 -04003783#else /* CONFIG_CGROUP_WRITEBACK */
3784
3785static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3786{
3787 return 0;
3788}
3789
3790static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3791{
3792}
3793
Tejun Heo2529bb32015-05-22 18:23:34 -04003794static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3795{
3796}
3797
Tejun Heo52ebea72015-05-22 17:13:37 -04003798#endif /* CONFIG_CGROUP_WRITEBACK */
3799
Tejun Heo79bd9812013-11-22 18:20:42 -05003800/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003801 * DO NOT USE IN NEW FILES.
3802 *
3803 * "cgroup.event_control" implementation.
3804 *
3805 * This is way over-engineered. It tries to support fully configurable
3806 * events for each user. Such level of flexibility is completely
3807 * unnecessary especially in the light of the planned unified hierarchy.
3808 *
3809 * Please deprecate this and replace with something simpler if at all
3810 * possible.
3811 */
3812
3813/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003814 * Unregister event and free resources.
3815 *
3816 * Gets called from workqueue.
3817 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003818static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003819{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003820 struct mem_cgroup_event *event =
3821 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003822 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003823
3824 remove_wait_queue(event->wqh, &event->wait);
3825
Tejun Heo59b6f872013-11-22 18:20:43 -05003826 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003827
3828 /* Notify userspace the event is going away. */
3829 eventfd_signal(event->eventfd, 1);
3830
3831 eventfd_ctx_put(event->eventfd);
3832 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003833 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003834}
3835
3836/*
3837 * Gets called on POLLHUP on eventfd when user closes it.
3838 *
3839 * Called with wqh->lock held and interrupts disabled.
3840 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003841static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3842 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003843{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003844 struct mem_cgroup_event *event =
3845 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003846 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003847 unsigned long flags = (unsigned long)key;
3848
3849 if (flags & POLLHUP) {
3850 /*
3851 * If the event has been detached at cgroup removal, we
3852 * can simply return knowing the other side will cleanup
3853 * for us.
3854 *
3855 * We can't race against event freeing since the other
3856 * side will require wqh->lock via remove_wait_queue(),
3857 * which we hold.
3858 */
Tejun Heofba94802013-11-22 18:20:43 -05003859 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003860 if (!list_empty(&event->list)) {
3861 list_del_init(&event->list);
3862 /*
3863 * We are in atomic context, but cgroup_event_remove()
3864 * may sleep, so we have to call it in workqueue.
3865 */
3866 schedule_work(&event->remove);
3867 }
Tejun Heofba94802013-11-22 18:20:43 -05003868 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003869 }
3870
3871 return 0;
3872}
3873
Tejun Heo3bc942f2013-11-22 18:20:44 -05003874static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003875 wait_queue_head_t *wqh, poll_table *pt)
3876{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003877 struct mem_cgroup_event *event =
3878 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003879
3880 event->wqh = wqh;
3881 add_wait_queue(wqh, &event->wait);
3882}
3883
3884/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003885 * DO NOT USE IN NEW FILES.
3886 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003887 * Parse input and register new cgroup event handler.
3888 *
3889 * Input must be in format '<event_fd> <control_fd> <args>'.
3890 * Interpretation of args is defined by control file implementation.
3891 */
Tejun Heo451af502014-05-13 12:16:21 -04003892static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3893 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003894{
Tejun Heo451af502014-05-13 12:16:21 -04003895 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003896 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003897 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003898 struct cgroup_subsys_state *cfile_css;
3899 unsigned int efd, cfd;
3900 struct fd efile;
3901 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003902 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003903 char *endp;
3904 int ret;
3905
Tejun Heo451af502014-05-13 12:16:21 -04003906 buf = strstrip(buf);
3907
3908 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003909 if (*endp != ' ')
3910 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003911 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003912
Tejun Heo451af502014-05-13 12:16:21 -04003913 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003914 if ((*endp != ' ') && (*endp != '\0'))
3915 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003916 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003917
3918 event = kzalloc(sizeof(*event), GFP_KERNEL);
3919 if (!event)
3920 return -ENOMEM;
3921
Tejun Heo59b6f872013-11-22 18:20:43 -05003922 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003923 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003924 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3925 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3926 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003927
3928 efile = fdget(efd);
3929 if (!efile.file) {
3930 ret = -EBADF;
3931 goto out_kfree;
3932 }
3933
3934 event->eventfd = eventfd_ctx_fileget(efile.file);
3935 if (IS_ERR(event->eventfd)) {
3936 ret = PTR_ERR(event->eventfd);
3937 goto out_put_efile;
3938 }
3939
3940 cfile = fdget(cfd);
3941 if (!cfile.file) {
3942 ret = -EBADF;
3943 goto out_put_eventfd;
3944 }
3945
3946 /* the process need read permission on control file */
3947 /* AV: shouldn't we check that it's been opened for read instead? */
3948 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3949 if (ret < 0)
3950 goto out_put_cfile;
3951
Tejun Heo79bd9812013-11-22 18:20:42 -05003952 /*
Tejun Heofba94802013-11-22 18:20:43 -05003953 * Determine the event callbacks and set them in @event. This used
3954 * to be done via struct cftype but cgroup core no longer knows
3955 * about these events. The following is crude but the whole thing
3956 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003957 *
3958 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003959 */
Al Virob5830432014-10-31 01:22:04 -04003960 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003961
3962 if (!strcmp(name, "memory.usage_in_bytes")) {
3963 event->register_event = mem_cgroup_usage_register_event;
3964 event->unregister_event = mem_cgroup_usage_unregister_event;
3965 } else if (!strcmp(name, "memory.oom_control")) {
3966 event->register_event = mem_cgroup_oom_register_event;
3967 event->unregister_event = mem_cgroup_oom_unregister_event;
3968 } else if (!strcmp(name, "memory.pressure_level")) {
3969 event->register_event = vmpressure_register_event;
3970 event->unregister_event = vmpressure_unregister_event;
3971 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003972 event->register_event = memsw_cgroup_usage_register_event;
3973 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003974 } else {
3975 ret = -EINVAL;
3976 goto out_put_cfile;
3977 }
3978
3979 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003980 * Verify @cfile should belong to @css. Also, remaining events are
3981 * automatically removed on cgroup destruction but the removal is
3982 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003983 */
Al Virob5830432014-10-31 01:22:04 -04003984 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003985 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003986 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003987 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003988 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003989 if (cfile_css != css) {
3990 css_put(cfile_css);
3991 goto out_put_cfile;
3992 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003993
Tejun Heo451af502014-05-13 12:16:21 -04003994 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003995 if (ret)
3996 goto out_put_css;
3997
3998 efile.file->f_op->poll(efile.file, &event->pt);
3999
Tejun Heofba94802013-11-22 18:20:43 -05004000 spin_lock(&memcg->event_list_lock);
4001 list_add(&event->list, &memcg->event_list);
4002 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004003
4004 fdput(cfile);
4005 fdput(efile);
4006
Tejun Heo451af502014-05-13 12:16:21 -04004007 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05004008
4009out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05004010 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05004011out_put_cfile:
4012 fdput(cfile);
4013out_put_eventfd:
4014 eventfd_ctx_put(event->eventfd);
4015out_put_efile:
4016 fdput(efile);
4017out_kfree:
4018 kfree(event);
4019
4020 return ret;
4021}
4022
Johannes Weiner241994ed2015-02-11 15:26:06 -08004023static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004024 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004025 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004026 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004027 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004028 },
4029 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004030 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004031 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004032 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004033 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004034 },
4035 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004036 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004037 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004038 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004039 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004040 },
4041 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004042 .name = "soft_limit_in_bytes",
4043 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004044 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004045 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07004046 },
4047 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004048 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004049 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004050 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004051 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004052 },
Balbir Singh8697d332008-02-07 00:13:59 -08004053 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004054 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004055 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004056 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004057 {
4058 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04004059 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004060 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004061 {
4062 .name = "use_hierarchy",
4063 .write_u64 = mem_cgroup_hierarchy_write,
4064 .read_u64 = mem_cgroup_hierarchy_read,
4065 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004066 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05004067 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04004068 .write = memcg_write_event_control,
Tejun Heo79bd9812013-11-22 18:20:42 -05004069 .flags = CFTYPE_NO_PREFIX,
4070 .mode = S_IWUGO,
4071 },
4072 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004073 .name = "swappiness",
4074 .read_u64 = mem_cgroup_swappiness_read,
4075 .write_u64 = mem_cgroup_swappiness_write,
4076 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004077 {
4078 .name = "move_charge_at_immigrate",
4079 .read_u64 = mem_cgroup_move_charge_read,
4080 .write_u64 = mem_cgroup_move_charge_write,
4081 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004082 {
4083 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004084 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004085 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004086 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4087 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004088 {
4089 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004090 },
Ying Han406eb0c2011-05-26 16:25:37 -07004091#ifdef CONFIG_NUMA
4092 {
4093 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004094 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07004095 },
4096#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004097#ifdef CONFIG_MEMCG_KMEM
4098 {
4099 .name = "kmem.limit_in_bytes",
4100 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004101 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004102 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004103 },
4104 {
4105 .name = "kmem.usage_in_bytes",
4106 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004107 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004108 },
4109 {
4110 .name = "kmem.failcnt",
4111 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004112 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004113 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004114 },
4115 {
4116 .name = "kmem.max_usage_in_bytes",
4117 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004118 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004119 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004120 },
Glauber Costa749c5412012-12-18 14:23:01 -08004121#ifdef CONFIG_SLABINFO
4122 {
4123 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004124 .seq_start = slab_start,
4125 .seq_next = slab_next,
4126 .seq_stop = slab_stop,
4127 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004128 },
4129#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004130#endif
Tejun Heo6bc10342012-04-01 12:09:55 -07004131 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004132};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004133
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004134static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004135{
4136 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004137 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004138 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004139 /*
4140 * This routine is called against possible nodes.
4141 * But it's BUG to call kmalloc() against offline node.
4142 *
4143 * TODO: this routine can waste much memory for nodes which will
4144 * never be onlined. It's better to use memory hotplug callback
4145 * function.
4146 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004147 if (!node_state(node, N_NORMAL_MEMORY))
4148 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004149 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004150 if (!pn)
4151 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004152
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004153 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4154 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004155 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07004156 mz->usage_in_excess = 0;
4157 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004158 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004159 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004160 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004161 return 0;
4162}
4163
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004164static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004165{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004166 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004167}
4168
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004169static struct mem_cgroup *mem_cgroup_alloc(void)
4170{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004171 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004172 size_t size;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004173
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004174 size = sizeof(struct mem_cgroup);
4175 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004176
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004177 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004178 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004179 return NULL;
4180
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004181 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4182 if (!memcg->stat)
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004183 goto out_free;
Tejun Heo841710a2015-05-22 18:23:33 -04004184
4185 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4186 goto out_free_stat;
4187
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004188 return memcg;
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004189
Tejun Heo841710a2015-05-22 18:23:33 -04004190out_free_stat:
4191 free_percpu(memcg->stat);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004192out_free:
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004193 kfree(memcg);
Dan Carpenterd2e61b82010-11-11 14:05:12 -08004194 return NULL;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004195}
4196
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004197/*
Glauber Costac8b2a362012-12-18 14:22:13 -08004198 * At destroying mem_cgroup, references from swap_cgroup can remain.
4199 * (scanning all at force_empty is too costly...)
4200 *
4201 * Instead of clearing all references at force_empty, we remember
4202 * the number of reference from swap_cgroup and free mem_cgroup when
4203 * it goes down to 0.
4204 *
4205 * Removal of cgroup itself succeeds regardless of refs from swap.
Hugh Dickins59927fb2012-03-15 15:17:07 -07004206 */
Glauber Costac8b2a362012-12-18 14:22:13 -08004207
4208static void __mem_cgroup_free(struct mem_cgroup *memcg)
Hugh Dickins59927fb2012-03-15 15:17:07 -07004209{
Glauber Costac8b2a362012-12-18 14:22:13 -08004210 int node;
Hugh Dickins59927fb2012-03-15 15:17:07 -07004211
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07004212 mem_cgroup_remove_from_trees(memcg);
Glauber Costac8b2a362012-12-18 14:22:13 -08004213
4214 for_each_node(node)
4215 free_mem_cgroup_per_zone_info(memcg, node);
4216
4217 free_percpu(memcg->stat);
Tejun Heo841710a2015-05-22 18:23:33 -04004218 memcg_wb_domain_exit(memcg);
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004219 kfree(memcg);
Hugh Dickins59927fb2012-03-15 15:17:07 -07004220}
Glauber Costa3afe36b2012-05-29 15:07:10 -07004221
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004222/*
4223 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4224 */
Glauber Costae1aab162011-12-11 21:47:03 +00004225struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004226{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004227 if (!memcg->memory.parent)
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004228 return NULL;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004229 return mem_cgroup_from_counter(memcg->memory.parent, memory);
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004230}
Glauber Costae1aab162011-12-11 21:47:03 +00004231EXPORT_SYMBOL(parent_mem_cgroup);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004232
Li Zefan0eb253e2009-01-15 13:51:25 -08004233static struct cgroup_subsys_state * __ref
Tejun Heoeb954192013-08-08 20:11:23 -04004234mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004235{
Glauber Costad142e3e2013-02-22 16:34:52 -08004236 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004237 long error = -ENOMEM;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004238 int node;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004239
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004240 memcg = mem_cgroup_alloc();
4241 if (!memcg)
KAMEZAWA Hiroyuki04046e12009-04-02 16:57:33 -07004242 return ERR_PTR(error);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004243
Bob Liu3ed28fa2012-01-12 17:19:04 -08004244 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004245 if (alloc_mem_cgroup_per_zone_info(memcg, node))
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004246 goto free_out;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004247
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08004248 /* root ? */
Tejun Heoeb954192013-08-08 20:11:23 -04004249 if (parent_css == NULL) {
Hillf Dantona41c58a2011-12-19 17:11:57 -08004250 root_mem_cgroup = memcg;
Tejun Heo56161632015-05-22 17:13:20 -04004251 mem_cgroup_root_css = &memcg->css;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004252 page_counter_init(&memcg->memory, NULL);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004253 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004254 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004255 page_counter_init(&memcg->memsw, NULL);
4256 page_counter_init(&memcg->kmem, NULL);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004257 }
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004258
Glauber Costad142e3e2013-02-22 16:34:52 -08004259 memcg->last_scanned_node = MAX_NUMNODES;
4260 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004261 memcg->move_charge_at_immigrate = 0;
4262 mutex_init(&memcg->thresholds_lock);
4263 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004264 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004265 INIT_LIST_HEAD(&memcg->event_list);
4266 spin_lock_init(&memcg->event_list_lock);
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004267#ifdef CONFIG_MEMCG_KMEM
4268 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004269#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004270#ifdef CONFIG_CGROUP_WRITEBACK
4271 INIT_LIST_HEAD(&memcg->cgwb_list);
4272#endif
Glauber Costad142e3e2013-02-22 16:34:52 -08004273 return &memcg->css;
4274
4275free_out:
4276 __mem_cgroup_free(memcg);
4277 return ERR_PTR(error);
4278}
4279
4280static int
Tejun Heoeb954192013-08-08 20:11:23 -04004281mem_cgroup_css_online(struct cgroup_subsys_state *css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004282{
Tejun Heoeb954192013-08-08 20:11:23 -04004283 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04004284 struct mem_cgroup *parent = mem_cgroup_from_css(css->parent);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004285 int ret;
Glauber Costad142e3e2013-02-22 16:34:52 -08004286
Tejun Heo15a4c832014-05-04 15:09:14 -04004287 if (css->id > MEM_CGROUP_ID_MAX)
Li Zefan4219b2d2013-09-23 16:56:29 +08004288 return -ENOSPC;
4289
Tejun Heo63876982013-08-08 20:11:23 -04004290 if (!parent)
Glauber Costad142e3e2013-02-22 16:34:52 -08004291 return 0;
4292
Glauber Costa09998212013-02-22 16:34:55 -08004293 mutex_lock(&memcg_create_mutex);
Glauber Costad142e3e2013-02-22 16:34:52 -08004294
4295 memcg->use_hierarchy = parent->use_hierarchy;
4296 memcg->oom_kill_disable = parent->oom_kill_disable;
4297 memcg->swappiness = mem_cgroup_swappiness(parent);
4298
4299 if (parent->use_hierarchy) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004300 page_counter_init(&memcg->memory, &parent->memory);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004301 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004302 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004303 page_counter_init(&memcg->memsw, &parent->memsw);
4304 page_counter_init(&memcg->kmem, &parent->kmem);
Glauber Costa55007d82012-12-18 14:22:38 -08004305
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004306 /*
Li Zefan8d76a972013-07-08 16:00:36 -07004307 * No need to take a reference to the parent because cgroup
4308 * core guarantees its existence.
Daisuke Nishimura7bcc1bb2009-01-29 14:25:11 -08004309 */
Balbir Singh18f59ea2009-01-07 18:08:07 -08004310 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004311 page_counter_init(&memcg->memory, NULL);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004312 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004313 memcg->soft_limit = PAGE_COUNTER_MAX;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004314 page_counter_init(&memcg->memsw, NULL);
4315 page_counter_init(&memcg->kmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004316 /*
4317 * Deeper hierachy with use_hierarchy == false doesn't make
4318 * much sense so let cgroup subsystem know about this
4319 * unfortunate state in our controller.
4320 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004321 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004322 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004323 }
Glauber Costa09998212013-02-22 16:34:55 -08004324 mutex_unlock(&memcg_create_mutex);
Vladimir Davydovd6441632014-01-23 15:53:09 -08004325
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004326 ret = memcg_init_kmem(memcg, &memory_cgrp_subsys);
4327 if (ret)
4328 return ret;
4329
4330 /*
4331 * Make sure the memcg is initialized: mem_cgroup_iter()
4332 * orders reading memcg->initialized against its callers
4333 * reading the memcg members.
4334 */
4335 smp_store_release(&memcg->initialized, 1);
4336
4337 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004338}
4339
Tejun Heoeb954192013-08-08 20:11:23 -04004340static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004341{
Tejun Heoeb954192013-08-08 20:11:23 -04004342 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004343 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004344
4345 /*
4346 * Unregister events and notify userspace.
4347 * Notify userspace about cgroup removing only after rmdir of cgroup
4348 * directory to avoid race between userspace and kernelspace.
4349 */
Tejun Heofba94802013-11-22 18:20:43 -05004350 spin_lock(&memcg->event_list_lock);
4351 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004352 list_del_init(&event->list);
4353 schedule_work(&event->remove);
4354 }
Tejun Heofba94802013-11-22 18:20:43 -05004355 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004356
Michal Hocko33cb8762013-07-31 13:53:51 -07004357 vmpressure_cleanup(&memcg->vmpressure);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08004358
4359 memcg_deactivate_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004360
4361 wb_memcg_offline(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004362}
4363
Tejun Heoeb954192013-08-08 20:11:23 -04004364static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004365{
Tejun Heoeb954192013-08-08 20:11:23 -04004366 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004367
Li Zefan10d5ebf2013-07-08 16:00:33 -07004368 memcg_destroy_kmem(memcg);
Li Zefan465939a2013-07-08 16:00:38 -07004369 __mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004370}
4371
Tejun Heo1ced9532014-07-08 18:02:57 -04004372/**
4373 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4374 * @css: the target css
4375 *
4376 * Reset the states of the mem_cgroup associated with @css. This is
4377 * invoked when the userland requests disabling on the default hierarchy
4378 * but the memcg is pinned through dependency. The memcg should stop
4379 * applying policies and should revert to the vanilla state as it may be
4380 * made visible again.
4381 *
4382 * The current implementation only resets the essential configurations.
4383 * This needs to be expanded to cover all the visible parts.
4384 */
4385static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4386{
4387 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4388
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004389 mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX);
4390 mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX);
4391 memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004392 memcg->low = 0;
4393 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004394 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004395 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004396}
4397
Daisuke Nishimura02491442010-03-10 15:22:17 -08004398#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004399/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004400static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004401{
Johannes Weiner05b84302014-08-06 16:05:59 -07004402 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004403
4404 /* Try a single bulk charge without reclaim first */
Johannes Weiner00501b52014-08-08 14:19:20 -07004405 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004406 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004407 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004408 return ret;
4409 }
Johannes Weiner692e7c42014-08-06 16:05:57 -07004410 if (ret == -EINTR) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004411 cancel_charge(root_mem_cgroup, count);
Johannes Weiner692e7c42014-08-06 16:05:57 -07004412 return ret;
4413 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004414
4415 /* Try charges one by one with reclaim */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004416 while (count--) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004417 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
Johannes Weiner9476db92014-08-06 16:05:55 -07004418 /*
4419 * In case of failure, any residual charges against
4420 * mc.to will be dropped by mem_cgroup_clear_mc()
Johannes Weiner692e7c42014-08-06 16:05:57 -07004421 * later on. However, cancel any charges that are
4422 * bypassed to root right away or they'll be lost.
Johannes Weiner9476db92014-08-06 16:05:55 -07004423 */
Johannes Weiner692e7c42014-08-06 16:05:57 -07004424 if (ret == -EINTR)
Johannes Weiner00501b52014-08-08 14:19:20 -07004425 cancel_charge(root_mem_cgroup, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004426 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004427 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004428 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004429 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004430 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004431 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004432}
4433
4434/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004435 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004436 * @vma: the vma the pte to be checked belongs
4437 * @addr: the address corresponding to the pte to be checked
4438 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08004439 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004440 *
4441 * Returns
4442 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4443 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4444 * move charge. if @target is not NULL, the page is stored in target->page
4445 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08004446 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4447 * target for charge migration. if @target is not NULL, the entry is stored
4448 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004449 *
4450 * Called with pte lock held.
4451 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004452union mc_target {
4453 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004454 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004455};
4456
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004457enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004458 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004459 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004460 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004461};
4462
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004463static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4464 unsigned long addr, pte_t ptent)
4465{
4466 struct page *page = vm_normal_page(vma, addr, ptent);
4467
4468 if (!page || !page_mapped(page))
4469 return NULL;
4470 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004471 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004472 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004473 } else {
4474 if (!(mc.flags & MOVE_FILE))
4475 return NULL;
4476 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004477 if (!get_page_unless_zero(page))
4478 return NULL;
4479
4480 return page;
4481}
4482
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004483#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004484static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4485 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4486{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004487 struct page *page = NULL;
4488 swp_entry_t ent = pte_to_swp_entry(ptent);
4489
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004490 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004491 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004492 /*
4493 * Because lookup_swap_cache() updates some statistics counter,
4494 * we call find_get_page() with swapper_space directly.
4495 */
Shaohua Li33806f02013-02-22 16:34:37 -08004496 page = find_get_page(swap_address_space(ent), ent.val);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004497 if (do_swap_account)
4498 entry->val = ent.val;
4499
4500 return page;
4501}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004502#else
4503static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4504 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4505{
4506 return NULL;
4507}
4508#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004509
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004510static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4511 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4512{
4513 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004514 struct address_space *mapping;
4515 pgoff_t pgoff;
4516
4517 if (!vma->vm_file) /* anonymous vma */
4518 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004519 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004520 return NULL;
4521
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004522 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004523 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004524
4525 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004526#ifdef CONFIG_SWAP
4527 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004528 if (shmem_mapping(mapping)) {
4529 page = find_get_entry(mapping, pgoff);
4530 if (radix_tree_exceptional_entry(page)) {
4531 swp_entry_t swp = radix_to_swp_entry(page);
4532 if (do_swap_account)
4533 *entry = swp;
4534 page = find_get_page(swap_address_space(swp), swp.val);
4535 }
4536 } else
4537 page = find_get_page(mapping, pgoff);
4538#else
4539 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004540#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004541 return page;
4542}
4543
Chen Gangb1b0dea2015-04-14 15:47:35 -07004544/**
4545 * mem_cgroup_move_account - move account of the page
4546 * @page: the page
4547 * @nr_pages: number of regular pages (>1 for huge pages)
4548 * @from: mem_cgroup which the page is moved from.
4549 * @to: mem_cgroup which the page is moved to. @from != @to.
4550 *
4551 * The caller must confirm following.
4552 * - page is not on LRU (isolate_page() is useful.)
4553 * - compound_lock is held when nr_pages > 1
4554 *
4555 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4556 * from old cgroup.
4557 */
4558static int mem_cgroup_move_account(struct page *page,
4559 unsigned int nr_pages,
4560 struct mem_cgroup *from,
4561 struct mem_cgroup *to)
4562{
4563 unsigned long flags;
4564 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004565 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004566
4567 VM_BUG_ON(from == to);
4568 VM_BUG_ON_PAGE(PageLRU(page), page);
4569 /*
4570 * The page is isolated from LRU. So, collapse function
4571 * will not handle this page. But page splitting can happen.
4572 * Do this check under compound_page_lock(). The caller should
4573 * hold it.
4574 */
4575 ret = -EBUSY;
4576 if (nr_pages > 1 && !PageTransHuge(page))
4577 goto out;
4578
4579 /*
4580 * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup
4581 * of its source page while we change it: page migration takes
4582 * both pages off the LRU, but page cache replacement doesn't.
4583 */
4584 if (!trylock_page(page))
4585 goto out;
4586
4587 ret = -EINVAL;
4588 if (page->mem_cgroup != from)
4589 goto out_unlock;
4590
Greg Thelenc4843a72015-05-22 17:13:16 -04004591 anon = PageAnon(page);
4592
Chen Gangb1b0dea2015-04-14 15:47:35 -07004593 spin_lock_irqsave(&from->move_lock, flags);
4594
Greg Thelenc4843a72015-05-22 17:13:16 -04004595 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004596 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4597 nr_pages);
4598 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4599 nr_pages);
4600 }
4601
Greg Thelenc4843a72015-05-22 17:13:16 -04004602 /*
4603 * move_lock grabbed above and caller set from->moving_account, so
4604 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4605 * So mapping should be stable for dirty pages.
4606 */
4607 if (!anon && PageDirty(page)) {
4608 struct address_space *mapping = page_mapping(page);
4609
4610 if (mapping_cap_account_dirty(mapping)) {
4611 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4612 nr_pages);
4613 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4614 nr_pages);
4615 }
4616 }
4617
Chen Gangb1b0dea2015-04-14 15:47:35 -07004618 if (PageWriteback(page)) {
4619 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4620 nr_pages);
4621 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4622 nr_pages);
4623 }
4624
4625 /*
4626 * It is safe to change page->mem_cgroup here because the page
4627 * is referenced, charged, and isolated - we can't race with
4628 * uncharging, charging, migration, or LRU putback.
4629 */
4630
4631 /* caller should have done css_get */
4632 page->mem_cgroup = to;
4633 spin_unlock_irqrestore(&from->move_lock, flags);
4634
4635 ret = 0;
4636
4637 local_irq_disable();
4638 mem_cgroup_charge_statistics(to, page, nr_pages);
4639 memcg_check_events(to, page);
4640 mem_cgroup_charge_statistics(from, page, -nr_pages);
4641 memcg_check_events(from, page);
4642 local_irq_enable();
4643out_unlock:
4644 unlock_page(page);
4645out:
4646 return ret;
4647}
4648
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004649static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004650 unsigned long addr, pte_t ptent, union mc_target *target)
4651{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004652 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004653 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004654 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004655
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004656 if (pte_present(ptent))
4657 page = mc_handle_present_pte(vma, addr, ptent);
4658 else if (is_swap_pte(ptent))
4659 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004660 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004661 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004662
4663 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004664 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004665 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004666 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004667 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004668 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004669 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004670 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004671 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004672 ret = MC_TARGET_PAGE;
4673 if (target)
4674 target->page = page;
4675 }
4676 if (!ret || !target)
4677 put_page(page);
4678 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004679 /* There is a swap entry and a page doesn't exist or isn't charged */
4680 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004681 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004682 ret = MC_TARGET_SWAP;
4683 if (target)
4684 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004685 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004686 return ret;
4687}
4688
Naoya Horiguchi12724852012-03-21 16:34:28 -07004689#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4690/*
4691 * We don't consider swapping or file mapped pages because THP does not
4692 * support them for now.
4693 * Caller should make sure that pmd_trans_huge(pmd) is true.
4694 */
4695static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4696 unsigned long addr, pmd_t pmd, union mc_target *target)
4697{
4698 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004699 enum mc_target_type ret = MC_TARGET_NONE;
4700
4701 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004702 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004703 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004704 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004705 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004706 ret = MC_TARGET_PAGE;
4707 if (target) {
4708 get_page(page);
4709 target->page = page;
4710 }
4711 }
4712 return ret;
4713}
4714#else
4715static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4716 unsigned long addr, pmd_t pmd, union mc_target *target)
4717{
4718 return MC_TARGET_NONE;
4719}
4720#endif
4721
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004722static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4723 unsigned long addr, unsigned long end,
4724 struct mm_walk *walk)
4725{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004726 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004727 pte_t *pte;
4728 spinlock_t *ptl;
4729
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004730 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004731 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4732 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004733 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004734 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004735 }
Dave Hansen03319322011-03-22 16:32:56 -07004736
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004737 if (pmd_trans_unstable(pmd))
4738 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004739 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4740 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004741 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004742 mc.precharge++; /* increment precharge temporarily */
4743 pte_unmap_unlock(pte - 1, ptl);
4744 cond_resched();
4745
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004746 return 0;
4747}
4748
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004749static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4750{
4751 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004752
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004753 struct mm_walk mem_cgroup_count_precharge_walk = {
4754 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4755 .mm = mm,
4756 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004757 down_read(&mm->mmap_sem);
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004758 walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004759 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004760
4761 precharge = mc.precharge;
4762 mc.precharge = 0;
4763
4764 return precharge;
4765}
4766
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004767static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4768{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004769 unsigned long precharge = mem_cgroup_count_precharge(mm);
4770
4771 VM_BUG_ON(mc.moving_task);
4772 mc.moving_task = current;
4773 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004774}
4775
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004776/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4777static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004778{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004779 struct mem_cgroup *from = mc.from;
4780 struct mem_cgroup *to = mc.to;
4781
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004782 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004783 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004784 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004785 mc.precharge = 0;
4786 }
4787 /*
4788 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4789 * we must uncharge here.
4790 */
4791 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004792 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004793 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004794 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004795 /* we must fixup refcnts and charges */
4796 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004797 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004798 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004799 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004800
Johannes Weiner05b84302014-08-06 16:05:59 -07004801 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004802 * we charged both to->memory and to->memsw, so we
4803 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004804 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004805 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004806 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004807
Johannes Weinere8ea14c2014-12-10 15:42:42 -08004808 css_put_many(&mc.from->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004809
Li Zefan40503772013-07-08 16:00:34 -07004810 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004811 mc.moved_swap = 0;
4812 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004813 memcg_oom_recover(from);
4814 memcg_oom_recover(to);
4815 wake_up_all(&mc.waitq);
4816}
4817
4818static void mem_cgroup_clear_mc(void)
4819{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004820 /*
4821 * we must clear moving_task before waking up waiters at the end of
4822 * task migration.
4823 */
4824 mc.moving_task = NULL;
4825 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004826 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004827 mc.from = NULL;
4828 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004829 spin_unlock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004830}
4831
Tejun Heoeb954192013-08-08 20:11:23 -04004832static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08004833 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004834{
Tejun Heoeb954192013-08-08 20:11:23 -04004835 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo9f2115f2015-09-08 15:01:10 -07004836 struct mem_cgroup *from;
4837 struct task_struct *p;
4838 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004839 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004840 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004841
Glauber Costaee5e8472013-02-22 16:34:50 -08004842 /*
4843 * We are now commited to this value whatever it is. Changes in this
4844 * tunable will only affect upcoming migrations, not the current one.
4845 * So we need to save it, and keep it going.
4846 */
Jason Low4db0c3c2015-04-15 16:14:08 -07004847 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
Tejun Heo9f2115f2015-09-08 15:01:10 -07004848 if (!move_flags)
4849 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004850
Tejun Heo9f2115f2015-09-08 15:01:10 -07004851 p = cgroup_taskset_first(tset);
4852 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004853
Tejun Heo9f2115f2015-09-08 15:01:10 -07004854 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004855
Tejun Heo9f2115f2015-09-08 15:01:10 -07004856 mm = get_task_mm(p);
4857 if (!mm)
4858 return 0;
4859 /* We move charges only when we move a owner of the mm */
4860 if (mm->owner == p) {
4861 VM_BUG_ON(mc.from);
4862 VM_BUG_ON(mc.to);
4863 VM_BUG_ON(mc.precharge);
4864 VM_BUG_ON(mc.moved_charge);
4865 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004866
Tejun Heo9f2115f2015-09-08 15:01:10 -07004867 spin_lock(&mc.lock);
4868 mc.from = from;
4869 mc.to = memcg;
4870 mc.flags = move_flags;
4871 spin_unlock(&mc.lock);
4872 /* We set mc.moving_task later */
4873
4874 ret = mem_cgroup_precharge_mc(mm);
4875 if (ret)
4876 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004877 }
Tejun Heo9f2115f2015-09-08 15:01:10 -07004878 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004879 return ret;
4880}
4881
Tejun Heoeb954192013-08-08 20:11:23 -04004882static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08004883 struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004884{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004885 if (mc.to)
4886 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004887}
4888
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004889static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4890 unsigned long addr, unsigned long end,
4891 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004892{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004893 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004894 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004895 pte_t *pte;
4896 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004897 enum mc_target_type target_type;
4898 union mc_target target;
4899 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004900
Naoya Horiguchi12724852012-03-21 16:34:28 -07004901 /*
4902 * We don't take compound_lock() here but no race with splitting thp
4903 * happens because:
4904 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
4905 * under splitting, which means there's no concurrent thp split,
4906 * - if another thread runs into split_huge_page() just after we
4907 * entered this if-block, the thread must wait for page table lock
4908 * to be unlocked in __split_huge_page_splitting(), where the main
4909 * part of thp split is not executed yet.
4910 */
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004911 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004912 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004913 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004914 return 0;
4915 }
4916 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4917 if (target_type == MC_TARGET_PAGE) {
4918 page = target.page;
4919 if (!isolate_lru_page(page)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004920 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
Johannes Weiner1306a852014-12-10 15:44:52 -08004921 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004922 mc.precharge -= HPAGE_PMD_NR;
4923 mc.moved_charge += HPAGE_PMD_NR;
4924 }
4925 putback_lru_page(page);
4926 }
4927 put_page(page);
4928 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004929 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004930 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004931 }
4932
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004933 if (pmd_trans_unstable(pmd))
4934 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004935retry:
4936 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4937 for (; addr != end; addr += PAGE_SIZE) {
4938 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004939 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004940
4941 if (!mc.precharge)
4942 break;
4943
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004944 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004945 case MC_TARGET_PAGE:
4946 page = target.page;
4947 if (isolate_lru_page(page))
4948 goto put;
Johannes Weiner1306a852014-12-10 15:44:52 -08004949 if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004950 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004951 /* we uncharge from mc.from later. */
4952 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004953 }
4954 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004955put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004956 put_page(page);
4957 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004958 case MC_TARGET_SWAP:
4959 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004960 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004961 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004962 /* we fixup refcnts and charges later. */
4963 mc.moved_swap++;
4964 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004965 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004966 default:
4967 break;
4968 }
4969 }
4970 pte_unmap_unlock(pte - 1, ptl);
4971 cond_resched();
4972
4973 if (addr != end) {
4974 /*
4975 * We have consumed all precharges we got in can_attach().
4976 * We try charge one by one, but don't do any additional
4977 * charges to mc.to if we have failed in charge once in attach()
4978 * phase.
4979 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004980 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004981 if (!ret)
4982 goto retry;
4983 }
4984
4985 return ret;
4986}
4987
4988static void mem_cgroup_move_charge(struct mm_struct *mm)
4989{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004990 struct mm_walk mem_cgroup_move_charge_walk = {
4991 .pmd_entry = mem_cgroup_move_charge_pte_range,
4992 .mm = mm,
4993 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004994
4995 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004996 /*
4997 * Signal mem_cgroup_begin_page_stat() to take the memcg's
4998 * move_lock while we're moving its pages to another memcg.
4999 * Then wait for already started RCU-only updates to finish.
5000 */
5001 atomic_inc(&mc.from->moving_account);
5002 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005003retry:
5004 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5005 /*
5006 * Someone who are holding the mmap_sem might be waiting in
5007 * waitq. So we cancel all extra charges, wake up all waiters,
5008 * and retry. Because we cancel precharges, we might not be able
5009 * to move enough charges, but moving charge is a best-effort
5010 * feature anyway, so it wouldn't be a big problem.
5011 */
5012 __mem_cgroup_clear_mc();
5013 cond_resched();
5014 goto retry;
5015 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005016 /*
5017 * When we have consumed all precharges and failed in doing
5018 * additional charge, the page walk just aborts.
5019 */
5020 walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005021 up_read(&mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08005022 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005023}
5024
Tejun Heoeb954192013-08-08 20:11:23 -04005025static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005026 struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08005027{
Tejun Heo2f7ee562011-12-12 18:12:21 -08005028 struct task_struct *p = cgroup_taskset_first(tset);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005029 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005030
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005031 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005032 if (mc.to)
5033 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005034 mmput(mm);
5035 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005036 if (mc.to)
5037 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08005038}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005039#else /* !CONFIG_MMU */
Tejun Heoeb954192013-08-08 20:11:23 -04005040static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005041 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005042{
5043 return 0;
5044}
Tejun Heoeb954192013-08-08 20:11:23 -04005045static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005046 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005047{
5048}
Tejun Heoeb954192013-08-08 20:11:23 -04005049static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
Li Zefan761b3ef2012-01-31 13:47:36 +08005050 struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005051{
5052}
5053#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005054
Tejun Heof00baae2013-04-15 13:41:15 -07005055/*
5056 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005057 * to verify whether we're attached to the default hierarchy on each mount
5058 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005059 */
Tejun Heoeb954192013-08-08 20:11:23 -04005060static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005061{
5062 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005063 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005064 * guarantees that @root doesn't have any children, so turning it
5065 * on for the root memcg is enough.
5066 */
Tejun Heoaa6ec292014-07-09 10:08:08 -04005067 if (cgroup_on_dfl(root_css->cgroup))
Vladimir Davydov7feee592015-03-12 16:26:19 -07005068 root_mem_cgroup->use_hierarchy = true;
5069 else
5070 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005071}
5072
Johannes Weiner241994ed2015-02-11 15:26:06 -08005073static u64 memory_current_read(struct cgroup_subsys_state *css,
5074 struct cftype *cft)
5075{
5076 return mem_cgroup_usage(mem_cgroup_from_css(css), false);
5077}
5078
5079static int memory_low_show(struct seq_file *m, void *v)
5080{
5081 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005082 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005083
5084 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005085 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005086 else
5087 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5088
5089 return 0;
5090}
5091
5092static ssize_t memory_low_write(struct kernfs_open_file *of,
5093 char *buf, size_t nbytes, loff_t off)
5094{
5095 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5096 unsigned long low;
5097 int err;
5098
5099 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005100 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005101 if (err)
5102 return err;
5103
5104 memcg->low = low;
5105
5106 return nbytes;
5107}
5108
5109static int memory_high_show(struct seq_file *m, void *v)
5110{
5111 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005112 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005113
5114 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005115 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005116 else
5117 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5118
5119 return 0;
5120}
5121
5122static ssize_t memory_high_write(struct kernfs_open_file *of,
5123 char *buf, size_t nbytes, loff_t off)
5124{
5125 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5126 unsigned long high;
5127 int err;
5128
5129 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005130 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005131 if (err)
5132 return err;
5133
5134 memcg->high = high;
5135
Tejun Heo2529bb32015-05-22 18:23:34 -04005136 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005137 return nbytes;
5138}
5139
5140static int memory_max_show(struct seq_file *m, void *v)
5141{
5142 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005143 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005144
5145 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005146 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005147 else
5148 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5149
5150 return 0;
5151}
5152
5153static ssize_t memory_max_write(struct kernfs_open_file *of,
5154 char *buf, size_t nbytes, loff_t off)
5155{
5156 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5157 unsigned long max;
5158 int err;
5159
5160 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005161 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005162 if (err)
5163 return err;
5164
5165 err = mem_cgroup_resize_limit(memcg, max);
5166 if (err)
5167 return err;
5168
Tejun Heo2529bb32015-05-22 18:23:34 -04005169 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005170 return nbytes;
5171}
5172
5173static int memory_events_show(struct seq_file *m, void *v)
5174{
5175 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5176
5177 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5178 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5179 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5180 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5181
5182 return 0;
5183}
5184
5185static struct cftype memory_files[] = {
5186 {
5187 .name = "current",
5188 .read_u64 = memory_current_read,
5189 },
5190 {
5191 .name = "low",
5192 .flags = CFTYPE_NOT_ON_ROOT,
5193 .seq_show = memory_low_show,
5194 .write = memory_low_write,
5195 },
5196 {
5197 .name = "high",
5198 .flags = CFTYPE_NOT_ON_ROOT,
5199 .seq_show = memory_high_show,
5200 .write = memory_high_write,
5201 },
5202 {
5203 .name = "max",
5204 .flags = CFTYPE_NOT_ON_ROOT,
5205 .seq_show = memory_max_show,
5206 .write = memory_max_write,
5207 },
5208 {
5209 .name = "events",
5210 .flags = CFTYPE_NOT_ON_ROOT,
5211 .seq_show = memory_events_show,
5212 },
5213 { } /* terminate */
5214};
5215
Tejun Heo073219e2014-02-08 10:36:58 -05005216struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005217 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005218 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005219 .css_offline = mem_cgroup_css_offline,
5220 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005221 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005222 .can_attach = mem_cgroup_can_attach,
5223 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005224 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005225 .bind = mem_cgroup_bind,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005226 .dfl_cftypes = memory_files,
5227 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005228 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005229};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005230
Johannes Weiner241994ed2015-02-11 15:26:06 -08005231/**
Johannes Weiner241994ed2015-02-11 15:26:06 -08005232 * mem_cgroup_low - check if memory consumption is below the normal range
5233 * @root: the highest ancestor to consider
5234 * @memcg: the memory cgroup to check
5235 *
5236 * Returns %true if memory consumption of @memcg, and that of all
5237 * configurable ancestors up to @root, is below the normal range.
5238 */
5239bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5240{
5241 if (mem_cgroup_disabled())
5242 return false;
5243
5244 /*
5245 * The toplevel group doesn't have a configurable range, so
5246 * it's never low when looked at directly, and it is not
5247 * considered an ancestor when assessing the hierarchy.
5248 */
5249
5250 if (memcg == root_mem_cgroup)
5251 return false;
5252
Michal Hocko4e54ded2015-02-27 15:51:46 -08005253 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005254 return false;
5255
5256 while (memcg != root) {
5257 memcg = parent_mem_cgroup(memcg);
5258
5259 if (memcg == root_mem_cgroup)
5260 break;
5261
Michal Hocko4e54ded2015-02-27 15:51:46 -08005262 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005263 return false;
5264 }
5265 return true;
5266}
5267
Johannes Weiner00501b52014-08-08 14:19:20 -07005268/**
5269 * mem_cgroup_try_charge - try charging a page
5270 * @page: page to charge
5271 * @mm: mm context of the victim
5272 * @gfp_mask: reclaim mode
5273 * @memcgp: charged memcg return
5274 *
5275 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5276 * pages according to @gfp_mask if necessary.
5277 *
5278 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5279 * Otherwise, an error code is returned.
5280 *
5281 * After page->mapping has been set up, the caller must finalize the
5282 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5283 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5284 */
5285int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
5286 gfp_t gfp_mask, struct mem_cgroup **memcgp)
5287{
5288 struct mem_cgroup *memcg = NULL;
5289 unsigned int nr_pages = 1;
5290 int ret = 0;
5291
5292 if (mem_cgroup_disabled())
5293 goto out;
5294
5295 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005296 /*
5297 * Every swap fault against a single page tries to charge the
5298 * page, bail as early as possible. shmem_unuse() encounters
5299 * already charged pages, too. The USED bit is protected by
5300 * the page lock, which serializes swap cache removal, which
5301 * in turn serializes uncharging.
5302 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005303 VM_BUG_ON_PAGE(!PageLocked(page), page);
Johannes Weiner1306a852014-12-10 15:44:52 -08005304 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005305 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005306
5307 if (do_swap_account) {
5308 swp_entry_t ent = { .val = page_private(page), };
5309 unsigned short id = lookup_swap_cgroup_id(ent);
5310
5311 rcu_read_lock();
5312 memcg = mem_cgroup_from_id(id);
5313 if (memcg && !css_tryget_online(&memcg->css))
5314 memcg = NULL;
5315 rcu_read_unlock();
5316 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005317 }
5318
5319 if (PageTransHuge(page)) {
5320 nr_pages <<= compound_order(page);
5321 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5322 }
5323
Johannes Weiner00501b52014-08-08 14:19:20 -07005324 if (!memcg)
5325 memcg = get_mem_cgroup_from_mm(mm);
5326
5327 ret = try_charge(memcg, gfp_mask, nr_pages);
5328
5329 css_put(&memcg->css);
5330
5331 if (ret == -EINTR) {
5332 memcg = root_mem_cgroup;
5333 ret = 0;
5334 }
5335out:
5336 *memcgp = memcg;
5337 return ret;
5338}
5339
5340/**
5341 * mem_cgroup_commit_charge - commit a page charge
5342 * @page: page to charge
5343 * @memcg: memcg to charge the page to
5344 * @lrucare: page might be on LRU already
5345 *
5346 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5347 * after page->mapping has been set up. This must happen atomically
5348 * as part of the page instantiation, i.e. under the page table lock
5349 * for anonymous pages, under the page lock for page and swap cache.
5350 *
5351 * In addition, the page must not be on the LRU during the commit, to
5352 * prevent racing with task migration. If it might be, use @lrucare.
5353 *
5354 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5355 */
5356void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
5357 bool lrucare)
5358{
5359 unsigned int nr_pages = 1;
5360
5361 VM_BUG_ON_PAGE(!page->mapping, page);
5362 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5363
5364 if (mem_cgroup_disabled())
5365 return;
5366 /*
5367 * Swap faults will attempt to charge the same page multiple
5368 * times. But reuse_swap_page() might have removed the page
5369 * from swapcache already, so we can't check PageSwapCache().
5370 */
5371 if (!memcg)
5372 return;
5373
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005374 commit_charge(page, memcg, lrucare);
5375
Johannes Weiner00501b52014-08-08 14:19:20 -07005376 if (PageTransHuge(page)) {
5377 nr_pages <<= compound_order(page);
5378 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5379 }
5380
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005381 local_irq_disable();
5382 mem_cgroup_charge_statistics(memcg, page, nr_pages);
5383 memcg_check_events(memcg, page);
5384 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005385
5386 if (do_swap_account && PageSwapCache(page)) {
5387 swp_entry_t entry = { .val = page_private(page) };
5388 /*
5389 * The swap entry might not get freed for a long time,
5390 * let's not wait for it. The page already received a
5391 * memory+swap charge, drop the swap entry duplicate.
5392 */
5393 mem_cgroup_uncharge_swap(entry);
5394 }
5395}
5396
5397/**
5398 * mem_cgroup_cancel_charge - cancel a page charge
5399 * @page: page to charge
5400 * @memcg: memcg to charge the page to
5401 *
5402 * Cancel a charge transaction started by mem_cgroup_try_charge().
5403 */
5404void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg)
5405{
5406 unsigned int nr_pages = 1;
5407
5408 if (mem_cgroup_disabled())
5409 return;
5410 /*
5411 * Swap faults will attempt to charge the same page multiple
5412 * times. But reuse_swap_page() might have removed the page
5413 * from swapcache already, so we can't check PageSwapCache().
5414 */
5415 if (!memcg)
5416 return;
5417
5418 if (PageTransHuge(page)) {
5419 nr_pages <<= compound_order(page);
5420 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5421 }
5422
5423 cancel_charge(memcg, nr_pages);
5424}
5425
Johannes Weiner747db952014-08-08 14:19:24 -07005426static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005427 unsigned long nr_anon, unsigned long nr_file,
5428 unsigned long nr_huge, struct page *dummy_page)
5429{
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005430 unsigned long nr_pages = nr_anon + nr_file;
Johannes Weiner747db952014-08-08 14:19:24 -07005431 unsigned long flags;
5432
Johannes Weinerce00a962014-09-05 08:43:57 -04005433 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005434 page_counter_uncharge(&memcg->memory, nr_pages);
5435 if (do_swap_account)
5436 page_counter_uncharge(&memcg->memsw, nr_pages);
Johannes Weinerce00a962014-09-05 08:43:57 -04005437 memcg_oom_recover(memcg);
5438 }
Johannes Weiner747db952014-08-08 14:19:24 -07005439
5440 local_irq_save(flags);
5441 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5442 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5443 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5444 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005445 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005446 memcg_check_events(memcg, dummy_page);
5447 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005448
5449 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005450 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005451}
5452
5453static void uncharge_list(struct list_head *page_list)
5454{
5455 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005456 unsigned long nr_anon = 0;
5457 unsigned long nr_file = 0;
5458 unsigned long nr_huge = 0;
5459 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005460 struct list_head *next;
5461 struct page *page;
5462
5463 next = page_list->next;
5464 do {
5465 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005466
5467 page = list_entry(next, struct page, lru);
5468 next = page->lru.next;
5469
5470 VM_BUG_ON_PAGE(PageLRU(page), page);
5471 VM_BUG_ON_PAGE(page_count(page), page);
5472
Johannes Weiner1306a852014-12-10 15:44:52 -08005473 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005474 continue;
5475
5476 /*
5477 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005478 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005479 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005480 */
5481
Johannes Weiner1306a852014-12-10 15:44:52 -08005482 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005483 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005484 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5485 nr_huge, page);
5486 pgpgout = nr_anon = nr_file = nr_huge = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005487 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005488 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005489 }
5490
5491 if (PageTransHuge(page)) {
5492 nr_pages <<= compound_order(page);
5493 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5494 nr_huge += nr_pages;
5495 }
5496
5497 if (PageAnon(page))
5498 nr_anon += nr_pages;
5499 else
5500 nr_file += nr_pages;
5501
Johannes Weiner1306a852014-12-10 15:44:52 -08005502 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005503
5504 pgpgout++;
5505 } while (next != page_list);
5506
5507 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005508 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5509 nr_huge, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005510}
5511
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005512/**
5513 * mem_cgroup_uncharge - uncharge a page
5514 * @page: page to uncharge
5515 *
5516 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5517 * mem_cgroup_commit_charge().
5518 */
5519void mem_cgroup_uncharge(struct page *page)
5520{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005521 if (mem_cgroup_disabled())
5522 return;
5523
Johannes Weiner747db952014-08-08 14:19:24 -07005524 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005525 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005526 return;
5527
Johannes Weiner747db952014-08-08 14:19:24 -07005528 INIT_LIST_HEAD(&page->lru);
5529 uncharge_list(&page->lru);
5530}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005531
Johannes Weiner747db952014-08-08 14:19:24 -07005532/**
5533 * mem_cgroup_uncharge_list - uncharge a list of page
5534 * @page_list: list of pages to uncharge
5535 *
5536 * Uncharge a list of pages previously charged with
5537 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5538 */
5539void mem_cgroup_uncharge_list(struct list_head *page_list)
5540{
5541 if (mem_cgroup_disabled())
5542 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005543
Johannes Weiner747db952014-08-08 14:19:24 -07005544 if (!list_empty(page_list))
5545 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005546}
5547
5548/**
5549 * mem_cgroup_migrate - migrate a charge to another page
5550 * @oldpage: currently charged page
5551 * @newpage: page to transfer the charge to
Michal Hockof5e03a42015-02-05 12:25:14 -08005552 * @lrucare: either or both pages might be on the LRU already
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005553 *
5554 * Migrate the charge from @oldpage to @newpage.
5555 *
5556 * Both pages must be locked, @newpage->mapping must be set up.
5557 */
5558void mem_cgroup_migrate(struct page *oldpage, struct page *newpage,
5559 bool lrucare)
5560{
Johannes Weiner29833312014-12-10 15:44:02 -08005561 struct mem_cgroup *memcg;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005562 int isolated;
5563
5564 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5565 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
5566 VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage);
5567 VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage);
5568 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005569 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5570 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005571
5572 if (mem_cgroup_disabled())
5573 return;
5574
5575 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005576 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005577 return;
5578
Johannes Weiner7d5e3242014-12-10 15:43:46 -08005579 /*
5580 * Swapcache readahead pages can get migrated before being
5581 * charged, and migration from compaction can happen to an
5582 * uncharged page when the PFN walker finds a page that
5583 * reclaim just put back on the LRU but has not released yet.
5584 */
Johannes Weiner1306a852014-12-10 15:44:52 -08005585 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005586 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005587 return;
5588
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005589 if (lrucare)
5590 lock_page_lru(oldpage, &isolated);
5591
Johannes Weiner1306a852014-12-10 15:44:52 -08005592 oldpage->mem_cgroup = NULL;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005593
5594 if (lrucare)
5595 unlock_page_lru(oldpage, isolated);
5596
Johannes Weiner29833312014-12-10 15:44:02 -08005597 commit_charge(newpage, memcg, lrucare);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005598}
5599
Michal Hocko2d110852013-02-22 16:34:43 -08005600/*
Michal Hocko10813122013-02-22 16:35:41 -08005601 * subsys_initcall() for memory controller.
5602 *
5603 * Some parts like hotcpu_notifier() have to be initialized from this context
5604 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5605 * everything that doesn't depend on a specific mem_cgroup structure should
5606 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005607 */
5608static int __init mem_cgroup_init(void)
5609{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005610 int cpu, node;
5611
Michal Hocko2d110852013-02-22 16:34:43 -08005612 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005613
5614 for_each_possible_cpu(cpu)
5615 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5616 drain_local_stock);
5617
5618 for_each_node(node) {
5619 struct mem_cgroup_tree_per_node *rtpn;
5620 int zone;
5621
5622 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5623 node_online(node) ? node : NUMA_NO_NODE);
5624
5625 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5626 struct mem_cgroup_tree_per_zone *rtpz;
5627
5628 rtpz = &rtpn->rb_tree_per_zone[zone];
5629 rtpz->rb_root = RB_ROOT;
5630 spin_lock_init(&rtpz->lock);
5631 }
5632 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5633 }
5634
Michal Hocko2d110852013-02-22 16:34:43 -08005635 return 0;
5636}
5637subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005638
5639#ifdef CONFIG_MEMCG_SWAP
5640/**
5641 * mem_cgroup_swapout - transfer a memsw charge to swap
5642 * @page: page whose memsw charge to transfer
5643 * @entry: swap entry to move the charge to
5644 *
5645 * Transfer the memsw charge of @page to @entry.
5646 */
5647void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5648{
5649 struct mem_cgroup *memcg;
5650 unsigned short oldid;
5651
5652 VM_BUG_ON_PAGE(PageLRU(page), page);
5653 VM_BUG_ON_PAGE(page_count(page), page);
5654
5655 if (!do_swap_account)
5656 return;
5657
5658 memcg = page->mem_cgroup;
5659
5660 /* Readahead page, never charged */
5661 if (!memcg)
5662 return;
5663
5664 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5665 VM_BUG_ON_PAGE(oldid, page);
5666 mem_cgroup_swap_statistics(memcg, true);
5667
5668 page->mem_cgroup = NULL;
5669
5670 if (!mem_cgroup_is_root(memcg))
5671 page_counter_uncharge(&memcg->memory, 1);
5672
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07005673 /*
5674 * Interrupts should be disabled here because the caller holds the
5675 * mapping->tree_lock lock which is taken with interrupts-off. It is
5676 * important here to have the interrupts disabled because it is the
5677 * only synchronisation we have for udpating the per-CPU variables.
5678 */
5679 VM_BUG_ON(!irqs_disabled());
Johannes Weiner21afa382015-02-11 15:26:36 -08005680 mem_cgroup_charge_statistics(memcg, page, -1);
5681 memcg_check_events(memcg, page);
5682}
5683
5684/**
5685 * mem_cgroup_uncharge_swap - uncharge a swap entry
5686 * @entry: swap entry to uncharge
5687 *
5688 * Drop the memsw charge associated with @entry.
5689 */
5690void mem_cgroup_uncharge_swap(swp_entry_t entry)
5691{
5692 struct mem_cgroup *memcg;
5693 unsigned short id;
5694
5695 if (!do_swap_account)
5696 return;
5697
5698 id = swap_cgroup_record(entry, 0);
5699 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005700 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005701 if (memcg) {
5702 if (!mem_cgroup_is_root(memcg))
5703 page_counter_uncharge(&memcg->memsw, 1);
5704 mem_cgroup_swap_statistics(memcg, false);
5705 css_put(&memcg->css);
5706 }
5707 rcu_read_unlock();
5708}
5709
5710/* for remember boot option*/
5711#ifdef CONFIG_MEMCG_SWAP_ENABLED
5712static int really_do_swap_account __initdata = 1;
5713#else
5714static int really_do_swap_account __initdata;
5715#endif
5716
5717static int __init enable_swap_account(char *s)
5718{
5719 if (!strcmp(s, "1"))
5720 really_do_swap_account = 1;
5721 else if (!strcmp(s, "0"))
5722 really_do_swap_account = 0;
5723 return 1;
5724}
5725__setup("swapaccount=", enable_swap_account);
5726
5727static struct cftype memsw_cgroup_files[] = {
5728 {
5729 .name = "memsw.usage_in_bytes",
5730 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
5731 .read_u64 = mem_cgroup_read_u64,
5732 },
5733 {
5734 .name = "memsw.max_usage_in_bytes",
5735 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
5736 .write = mem_cgroup_reset,
5737 .read_u64 = mem_cgroup_read_u64,
5738 },
5739 {
5740 .name = "memsw.limit_in_bytes",
5741 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
5742 .write = mem_cgroup_write,
5743 .read_u64 = mem_cgroup_read_u64,
5744 },
5745 {
5746 .name = "memsw.failcnt",
5747 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
5748 .write = mem_cgroup_reset,
5749 .read_u64 = mem_cgroup_read_u64,
5750 },
5751 { }, /* terminate */
5752};
5753
5754static int __init mem_cgroup_swap_init(void)
5755{
5756 if (!mem_cgroup_disabled() && really_do_swap_account) {
5757 do_swap_account = 1;
5758 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
5759 memsw_cgroup_files));
5760 }
5761 return 0;
5762}
5763subsys_initcall(mem_cgroup_swap_init);
5764
5765#endif /* CONFIG_MEMCG_SWAP */