blob: fe787f5c41bd1332eeca88e85a3f0bf483bb552b [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Johannes Weiner1575e682015-04-14 15:44:51 -070017 * Native page reclaim
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
22 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080023 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
27 *
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
32 */
33
Johannes Weiner3e32cb22014-12-10 15:42:31 -080034#include <linux/page_counter.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080035#include <linux/memcontrol.h>
36#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080037#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080038#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080039#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080040#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080041#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080043#include <linux/bit_spinlock.h>
44#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070045#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040046#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080047#include <linux/mutex.h>
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -070048#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070049#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080051#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080052#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080053#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050054#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080055#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080056#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080057#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070058#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070059#include <linux/mm_inline.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080060#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080061#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070062#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070063#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050064#include <linux/file.h>
Tejun Heob23afb92015-11-05 18:46:11 -080065#include <linux/tracehook.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080066#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000067#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070068#include <net/ip.h>
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
Johannes Weiner7d828602016-01-14 15:20:56 -080078struct mem_cgroup *root_mem_cgroup __read_mostly;
79
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070080#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh8cdea7c2008-02-07 00:13:50 -080081
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080082/* Socket memory accounting disabled? */
83static bool cgroup_memory_nosocket;
84
Vladimir Davydov04823c82016-01-20 15:02:38 -080085/* Kernel memory accounting disabled? */
86static bool cgroup_memory_nokmem;
87
Johannes Weiner21afa382015-02-11 15:26:36 -080088/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070089#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080090int do_swap_account __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080091#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070092#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080093#endif
94
Johannes Weiner7941d212016-01-14 15:21:23 -080095/* Whether legacy memory+swap accounting is active */
96static bool do_memsw_account(void)
97{
98 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account;
99}
100
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700101static const char * const mem_cgroup_stat_names[] = {
102 "cache",
103 "rss",
David Rientjesb070e652013-05-07 16:18:09 -0700104 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700105 "mapped_file",
Greg Thelenc4843a72015-05-22 17:13:16 -0400106 "dirty",
Sha Zhengju3ea67d02013-09-12 15:13:53 -0700107 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700108 "swap",
109};
110
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700111static const char * const mem_cgroup_events_names[] = {
112 "pgpgin",
113 "pgpgout",
114 "pgfault",
115 "pgmajfault",
116};
117
Sha Zhengju58cf1882013-02-22 16:32:05 -0800118static const char * const mem_cgroup_lru_names[] = {
119 "inactive_anon",
120 "active_anon",
121 "inactive_file",
122 "active_file",
123 "unevictable",
124};
125
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700126#define THRESHOLDS_EVENTS_TARGET 128
127#define SOFTLIMIT_EVENTS_TARGET 1024
128#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700129
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700130/*
131 * Cgroups above their limits are maintained in a RB-Tree, independent of
132 * their hierarchy representation
133 */
134
135struct mem_cgroup_tree_per_zone {
136 struct rb_root rb_root;
137 spinlock_t lock;
138};
139
140struct mem_cgroup_tree_per_node {
141 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
142};
143
144struct mem_cgroup_tree {
145 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
146};
147
148static struct mem_cgroup_tree soft_limit_tree __read_mostly;
149
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700150/* for OOM */
151struct mem_cgroup_eventfd_list {
152 struct list_head list;
153 struct eventfd_ctx *eventfd;
154};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800155
Tejun Heo79bd9812013-11-22 18:20:42 -0500156/*
157 * cgroup_event represents events which userspace want to receive.
158 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500159struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500160 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500161 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500162 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500163 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500164 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500165 * eventfd to signal userspace about the event.
166 */
167 struct eventfd_ctx *eventfd;
168 /*
169 * Each of these stored in a list by the cgroup.
170 */
171 struct list_head list;
172 /*
Tejun Heofba94802013-11-22 18:20:43 -0500173 * register_event() callback will be used to add new userspace
174 * waiter for changes related to this event. Use eventfd_signal()
175 * on eventfd to send notification to userspace.
176 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500177 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500178 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500179 /*
180 * unregister_event() callback will be called when userspace closes
181 * the eventfd or on cgroup removing. This callback must be set,
182 * if you want provide notification functionality.
183 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500184 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500185 struct eventfd_ctx *eventfd);
186 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500187 * All fields below needed to unregister event when
188 * userspace closes eventfd.
189 */
190 poll_table pt;
191 wait_queue_head_t *wqh;
192 wait_queue_t wait;
193 struct work_struct remove;
194};
195
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700196static void mem_cgroup_threshold(struct mem_cgroup *memcg);
197static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800198
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800199/* Stuffs for move charges at task migration. */
200/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800201 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800202 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800203#define MOVE_ANON 0x1U
204#define MOVE_FILE 0x2U
205#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800206
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800207/* "mc" and its members are protected by cgroup_mutex */
208static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800209 spinlock_t lock; /* for from, to */
Tejun Heo264a0ae2016-04-21 19:09:02 -0400210 struct mm_struct *mm;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800211 struct mem_cgroup *from;
212 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800213 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800214 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800215 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800216 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800217 struct task_struct *moving_task; /* a task moving charges */
218 wait_queue_head_t waitq; /* a waitq for other context */
219} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700220 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800221 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
222};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800223
Balbir Singh4e416952009-09-23 15:56:39 -0700224/*
225 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
226 * limit reclaim to prevent infinite loops, if they ever occur.
227 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700228#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700229#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700230
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800231enum charge_type {
232 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700233 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800234 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700235 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700236 NR_CHARGE_TYPE,
237};
238
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800239/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800240enum res_type {
241 _MEM,
242 _MEMSWAP,
243 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800244 _KMEM,
Vladimir Davydovd55f90b2016-01-20 15:02:44 -0800245 _TCP,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800246};
247
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700248#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
249#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800250#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700251/* Used for OOM nofiier */
252#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800253
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700254/* Some nice accessors for the vmpressure. */
255struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
256{
257 if (!memcg)
258 memcg = root_mem_cgroup;
259 return &memcg->vmpressure;
260}
261
262struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
263{
264 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
265}
266
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700267static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
268{
269 return (memcg == root_mem_cgroup);
270}
271
Johannes Weiner127424c2016-01-20 15:02:32 -0800272#ifndef CONFIG_SLOB
Glauber Costa55007d82012-12-18 14:22:38 -0800273/*
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800274 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800275 * The main reason for not using cgroup id for this:
276 * this works better in sparse environments, where we have a lot of memcgs,
277 * but only a few kmem-limited. Or also, if we have, for instance, 200
278 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
279 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800280 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800281 * The current size of the caches array is stored in memcg_nr_cache_ids. It
282 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800283 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800284static DEFINE_IDA(memcg_cache_ida);
285int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800286
Vladimir Davydov05257a12015-02-12 14:59:01 -0800287/* Protects memcg_nr_cache_ids */
288static DECLARE_RWSEM(memcg_cache_ids_sem);
289
290void memcg_get_cache_ids(void)
291{
292 down_read(&memcg_cache_ids_sem);
293}
294
295void memcg_put_cache_ids(void)
296{
297 up_read(&memcg_cache_ids_sem);
298}
299
Glauber Costa55007d82012-12-18 14:22:38 -0800300/*
301 * MIN_SIZE is different than 1, because we would like to avoid going through
302 * the alloc/free process all the time. In a small machine, 4 kmem-limited
303 * cgroups is a reasonable guess. In the future, it could be a parameter or
304 * tunable, but that is strictly not necessary.
305 *
Li Zefanb8627832013-09-23 16:56:47 +0800306 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800307 * this constant directly from cgroup, but it is understandable that this is
308 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800309 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800310 * increase ours as well if it increases.
311 */
312#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800313#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800314
Glauber Costad7f25f82012-12-18 14:22:40 -0800315/*
316 * A lot of the calls to the cache allocation functions are expected to be
317 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
318 * conditional to this static branch, we'll have to allow modules that does
319 * kmem_cache_alloc and the such to see this symbol as well
320 */
Johannes Weineref129472016-01-14 15:21:34 -0800321DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
Glauber Costad7f25f82012-12-18 14:22:40 -0800322EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800323
Johannes Weiner127424c2016-01-20 15:02:32 -0800324#endif /* !CONFIG_SLOB */
Glauber Costaa8964b92012-12-18 14:22:09 -0800325
Balbir Singhf64c3f52009-09-23 15:56:37 -0700326static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700327mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700328{
Jianyu Zhane2318752014-06-06 14:38:20 -0700329 int nid = zone_to_nid(zone);
330 int zid = zone_idx(zone);
331
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700332 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700333}
334
Tejun Heoad7fa852015-05-27 20:00:02 -0400335/**
336 * mem_cgroup_css_from_page - css of the memcg associated with a page
337 * @page: page of interest
338 *
339 * If memcg is bound to the default hierarchy, css of the memcg associated
340 * with @page is returned. The returned css remains associated with @page
341 * until it is released.
342 *
343 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
344 * is returned.
Tejun Heoad7fa852015-05-27 20:00:02 -0400345 */
346struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
347{
348 struct mem_cgroup *memcg;
349
Tejun Heoad7fa852015-05-27 20:00:02 -0400350 memcg = page->mem_cgroup;
351
Tejun Heo9e10a132015-09-18 11:56:28 -0400352 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400353 memcg = root_mem_cgroup;
354
Tejun Heoad7fa852015-05-27 20:00:02 -0400355 return &memcg->css;
356}
357
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700358/**
359 * page_cgroup_ino - return inode number of the memcg a page is charged to
360 * @page: the page
361 *
362 * Look up the closest online ancestor of the memory cgroup @page is charged to
363 * and return its inode number or 0 if @page is not charged to any cgroup. It
364 * is safe to call this function without holding a reference to @page.
365 *
366 * Note, this function is inherently racy, because there is nothing to prevent
367 * the cgroup inode from getting torn down and potentially reallocated a moment
368 * after page_cgroup_ino() returns, so it only should be used by callers that
369 * do not care (such as procfs interfaces).
370 */
371ino_t page_cgroup_ino(struct page *page)
372{
373 struct mem_cgroup *memcg;
374 unsigned long ino = 0;
375
376 rcu_read_lock();
377 memcg = READ_ONCE(page->mem_cgroup);
378 while (memcg && !(memcg->css.flags & CSS_ONLINE))
379 memcg = parent_mem_cgroup(memcg);
380 if (memcg)
381 ino = cgroup_ino(memcg->css.cgroup);
382 rcu_read_unlock();
383 return ino;
384}
385
Balbir Singhf64c3f52009-09-23 15:56:37 -0700386static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700387mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700388{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700389 int nid = page_to_nid(page);
390 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700391
Jianyu Zhane2318752014-06-06 14:38:20 -0700392 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700393}
394
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700395static struct mem_cgroup_tree_per_zone *
396soft_limit_tree_node_zone(int nid, int zid)
397{
398 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
399}
400
401static struct mem_cgroup_tree_per_zone *
402soft_limit_tree_from_page(struct page *page)
403{
404 int nid = page_to_nid(page);
405 int zid = page_zonenum(page);
406
407 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
408}
409
Johannes Weinercf2c8122014-06-06 14:38:21 -0700410static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz,
411 struct mem_cgroup_tree_per_zone *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800412 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700413{
414 struct rb_node **p = &mctz->rb_root.rb_node;
415 struct rb_node *parent = NULL;
416 struct mem_cgroup_per_zone *mz_node;
417
418 if (mz->on_tree)
419 return;
420
421 mz->usage_in_excess = new_usage_in_excess;
422 if (!mz->usage_in_excess)
423 return;
424 while (*p) {
425 parent = *p;
426 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
427 tree_node);
428 if (mz->usage_in_excess < mz_node->usage_in_excess)
429 p = &(*p)->rb_left;
430 /*
431 * We can't avoid mem cgroups that are over their soft
432 * limit by the same amount
433 */
434 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
435 p = &(*p)->rb_right;
436 }
437 rb_link_node(&mz->tree_node, parent, p);
438 rb_insert_color(&mz->tree_node, &mctz->rb_root);
439 mz->on_tree = true;
440}
441
Johannes Weinercf2c8122014-06-06 14:38:21 -0700442static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
443 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700444{
445 if (!mz->on_tree)
446 return;
447 rb_erase(&mz->tree_node, &mctz->rb_root);
448 mz->on_tree = false;
449}
450
Johannes Weinercf2c8122014-06-06 14:38:21 -0700451static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
452 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700453{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700454 unsigned long flags;
455
456 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700457 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700458 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700459}
460
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800461static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
462{
463 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700464 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800465 unsigned long excess = 0;
466
467 if (nr_pages > soft_limit)
468 excess = nr_pages - soft_limit;
469
470 return excess;
471}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700472
473static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
474{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800475 unsigned long excess;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700476 struct mem_cgroup_per_zone *mz;
477 struct mem_cgroup_tree_per_zone *mctz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700478
Jianyu Zhane2318752014-06-06 14:38:20 -0700479 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700480 /*
481 * Necessary to update all ancestors when hierarchy is used.
482 * because their event counter is not touched.
483 */
484 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Jianyu Zhane2318752014-06-06 14:38:20 -0700485 mz = mem_cgroup_page_zoneinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800486 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700487 /*
488 * We have to update the tree if mz is on RB-tree or
489 * mem is over its softlimit.
490 */
491 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700492 unsigned long flags;
493
494 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700495 /* if on-tree, remove it */
496 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700497 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700498 /*
499 * Insert again. mz->usage_in_excess will be updated.
500 * If excess is 0, no tree ops.
501 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700502 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700503 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700504 }
505 }
506}
507
508static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
509{
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700510 struct mem_cgroup_tree_per_zone *mctz;
Jianyu Zhane2318752014-06-06 14:38:20 -0700511 struct mem_cgroup_per_zone *mz;
512 int nid, zid;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700513
Jianyu Zhane2318752014-06-06 14:38:20 -0700514 for_each_node(nid) {
515 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
516 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
517 mctz = soft_limit_tree_node_zone(nid, zid);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700518 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700519 }
520 }
521}
522
523static struct mem_cgroup_per_zone *
524__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
525{
526 struct rb_node *rightmost = NULL;
527 struct mem_cgroup_per_zone *mz;
528
529retry:
530 mz = NULL;
531 rightmost = rb_last(&mctz->rb_root);
532 if (!rightmost)
533 goto done; /* Nothing to reclaim from */
534
535 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
536 /*
537 * Remove the node now but someone else can add it back,
538 * we will to add it back at the end of reclaim to its correct
539 * position in the tree.
540 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700541 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800542 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400543 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700544 goto retry;
545done:
546 return mz;
547}
548
549static struct mem_cgroup_per_zone *
550mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
551{
552 struct mem_cgroup_per_zone *mz;
553
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700554 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700555 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700556 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700557 return mz;
558}
559
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700560/*
Greg Thelen484ebb32015-10-01 15:37:05 -0700561 * Return page count for single (non recursive) @memcg.
562 *
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700563 * Implementation Note: reading percpu statistics for memcg.
564 *
565 * Both of vmstat[] and percpu_counter has threshold and do periodic
566 * synchronization to implement "quick" read. There are trade-off between
567 * reading cost and precision of value. Then, we may have a chance to implement
Greg Thelen484ebb32015-10-01 15:37:05 -0700568 * a periodic synchronization of counter in memcg's counter.
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700569 *
570 * But this _read() function is used for user interface now. The user accounts
571 * memory usage by memory cgroup and he _always_ requires exact value because
572 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
573 * have to visit all online cpus and make sum. So, for now, unnecessary
574 * synchronization is not implemented. (just implemented for cpu hotplug)
575 *
576 * If there are kernel internal actions which can make use of some not-exact
577 * value, and reading all cpu value can be performance bottleneck in some
Greg Thelen484ebb32015-10-01 15:37:05 -0700578 * common workload, threshold and synchronization as vmstat[] should be
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700579 * implemented.
580 */
Greg Thelen484ebb32015-10-01 15:37:05 -0700581static unsigned long
582mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800583{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700584 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800585 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800586
Greg Thelen484ebb32015-10-01 15:37:05 -0700587 /* Per-cpu values can be negative, use a signed accumulator */
Tejun Heo733a5722015-05-22 18:23:18 -0400588 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700589 val += per_cpu(memcg->stat->count[idx], cpu);
Greg Thelen484ebb32015-10-01 15:37:05 -0700590 /*
591 * Summing races with updates, so val may be negative. Avoid exposing
592 * transient negative values.
593 */
594 if (val < 0)
595 val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800596 return val;
597}
598
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700599static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700600 enum mem_cgroup_events_index idx)
601{
602 unsigned long val = 0;
603 int cpu;
604
Tejun Heo733a5722015-05-22 18:23:18 -0400605 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700606 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700607 return val;
608}
609
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700610static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700611 struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800612 bool compound, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800613{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700614 /*
615 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
616 * counted as CACHE even if it's on ANON LRU.
617 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700618 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700619 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700620 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800621 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700622 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700623 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700624
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800625 if (compound) {
626 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
David Rientjesb070e652013-05-07 16:18:09 -0700627 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
628 nr_pages);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800629 }
David Rientjesb070e652013-05-07 16:18:09 -0700630
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800631 /* pagein of a big page is an event. So, ignore page size */
632 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700633 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800634 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700635 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800636 nr_pages = -nr_pages; /* for event */
637 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800638
Johannes Weiner13114712012-05-29 15:07:07 -0700639 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800640}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800641
Vladimir Davydov0a6b76d2016-03-17 14:18:42 -0700642unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
643 int nid, unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700644{
Jianyu Zhane2318752014-06-06 14:38:20 -0700645 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700646 int zid;
647
Jianyu Zhane2318752014-06-06 14:38:20 -0700648 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700649
Jianyu Zhane2318752014-06-06 14:38:20 -0700650 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
651 struct mem_cgroup_per_zone *mz;
652 enum lru_list lru;
653
654 for_each_lru(lru) {
655 if (!(BIT(lru) & lru_mask))
656 continue;
657 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
658 nr += mz->lru_size[lru];
659 }
660 }
661 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700662}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700663
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700664static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700665 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800666{
Jianyu Zhane2318752014-06-06 14:38:20 -0700667 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700668 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800669
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800670 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700671 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
672 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800673}
674
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800675static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
676 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800677{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700678 unsigned long val, next;
679
Johannes Weiner13114712012-05-29 15:07:07 -0700680 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700681 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700682 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800683 if ((long)next - (long)val < 0) {
684 switch (target) {
685 case MEM_CGROUP_TARGET_THRESH:
686 next = val + THRESHOLDS_EVENTS_TARGET;
687 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700688 case MEM_CGROUP_TARGET_SOFTLIMIT:
689 next = val + SOFTLIMIT_EVENTS_TARGET;
690 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800691 case MEM_CGROUP_TARGET_NUMAINFO:
692 next = val + NUMAINFO_EVENTS_TARGET;
693 break;
694 default:
695 break;
696 }
697 __this_cpu_write(memcg->stat->targets[target], next);
698 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700699 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800700 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800701}
702
703/*
704 * Check events in order.
705 *
706 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700707static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800708{
709 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800710 if (unlikely(mem_cgroup_event_ratelimit(memcg,
711 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700712 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800713 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800714
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700715 do_softlimit = mem_cgroup_event_ratelimit(memcg,
716 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700717#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800718 do_numainfo = mem_cgroup_event_ratelimit(memcg,
719 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700720#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800721 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700722 if (unlikely(do_softlimit))
723 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800724#if MAX_NUMNODES > 1
725 if (unlikely(do_numainfo))
726 atomic_inc(&memcg->numainfo_events);
727#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700728 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800729}
730
Balbir Singhcf475ad2008-04-29 01:00:16 -0700731struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800732{
Balbir Singh31a78f22008-09-28 23:09:31 +0100733 /*
734 * mm_update_next_owner() may clear mm->owner to NULL
735 * if it races with swapoff, page migration, etc.
736 * So this can be called with p == NULL.
737 */
738 if (unlikely(!p))
739 return NULL;
740
Tejun Heo073219e2014-02-08 10:36:58 -0500741 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800742}
Michal Hocko33398cf2015-09-08 15:01:02 -0700743EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800744
Johannes Weinerdf381972014-04-07 15:37:43 -0700745static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800746{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700747 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700748
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800749 rcu_read_lock();
750 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700751 /*
752 * Page cache insertions can happen withou an
753 * actual mm context, e.g. during disk probing
754 * on boot, loopback IO, acct() writes etc.
755 */
756 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700757 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700758 else {
759 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
760 if (unlikely(!memcg))
761 memcg = root_mem_cgroup;
762 }
Tejun Heoec903c02014-05-13 12:11:01 -0400763 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800764 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700765 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800766}
767
Johannes Weiner56600482012-01-12 17:17:59 -0800768/**
769 * mem_cgroup_iter - iterate over memory cgroup hierarchy
770 * @root: hierarchy root
771 * @prev: previously returned memcg, NULL on first invocation
772 * @reclaim: cookie for shared reclaim walks, NULL for full walks
773 *
774 * Returns references to children of the hierarchy below @root, or
775 * @root itself, or %NULL after a full round-trip.
776 *
777 * Caller must pass the return value in @prev on subsequent
778 * invocations for reference counting, or use mem_cgroup_iter_break()
779 * to cancel a hierarchy walk before the round-trip is complete.
780 *
781 * Reclaimers can specify a zone and a priority level in @reclaim to
782 * divide up the memcgs in the hierarchy among all concurrent
783 * reclaimers operating on the same zone and priority.
784 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700785struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800786 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700787 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700788{
Michal Hocko33398cf2015-09-08 15:01:02 -0700789 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800790 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800791 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800792 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700793
Andrew Morton694fbc02013-09-24 15:27:37 -0700794 if (mem_cgroup_disabled())
795 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800796
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700797 if (!root)
798 root = root_mem_cgroup;
799
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800800 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800801 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800802
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800803 if (!root->use_hierarchy && root != root_mem_cgroup) {
804 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800805 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700806 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800807 }
808
Michal Hocko542f85f2013-04-29 15:07:15 -0700809 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800810
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800811 if (reclaim) {
812 struct mem_cgroup_per_zone *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800813
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800814 mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone);
815 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700816
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800817 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700818 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800819
Vladimir Davydov6df38682015-12-29 14:54:10 -0800820 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -0700821 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800822 if (!pos || css_tryget(&pos->css))
823 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800824 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800825 * css reference reached zero, so iter->position will
826 * be cleared by ->css_released. However, we should not
827 * rely on this happening soon, because ->css_released
828 * is called from a work queue, and by busy-waiting we
829 * might block it. So we clear iter->position right
830 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800831 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800832 (void)cmpxchg(&iter->position, pos, NULL);
833 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800834 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800835
836 if (pos)
837 css = &pos->css;
838
839 for (;;) {
840 css = css_next_descendant_pre(css, &root->css);
841 if (!css) {
842 /*
843 * Reclaimers share the hierarchy walk, and a
844 * new one might jump in right at the end of
845 * the hierarchy - make sure they see at least
846 * one group and restart from the beginning.
847 */
848 if (!prev)
849 continue;
850 break;
851 }
852
853 /*
854 * Verify the css and acquire a reference. The root
855 * is provided by the caller, so we know it's alive
856 * and kicking, and don't take an extra reference.
857 */
858 memcg = mem_cgroup_from_css(css);
859
860 if (css == &root->css)
861 break;
862
Johannes Weiner0b8f73e2016-01-20 15:02:53 -0800863 if (css_tryget(css))
864 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800865
866 memcg = NULL;
867 }
868
869 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800870 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800871 * The position could have already been updated by a competing
872 * thread, so check that the value hasn't changed since we read
873 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800874 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800875 (void)cmpxchg(&iter->position, pos, memcg);
876
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800877 if (pos)
878 css_put(&pos->css);
879
880 if (!memcg)
881 iter->generation++;
882 else if (!prev)
883 reclaim->generation = iter->generation;
884 }
885
Michal Hocko542f85f2013-04-29 15:07:15 -0700886out_unlock:
887 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800888out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700889 if (prev && prev != root)
890 css_put(&prev->css);
891
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800892 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700893}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800894
Johannes Weiner56600482012-01-12 17:17:59 -0800895/**
896 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
897 * @root: hierarchy root
898 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
899 */
900void mem_cgroup_iter_break(struct mem_cgroup *root,
901 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800902{
903 if (!root)
904 root = root_mem_cgroup;
905 if (prev && prev != root)
906 css_put(&prev->css);
907}
908
Vladimir Davydov6df38682015-12-29 14:54:10 -0800909static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
910{
911 struct mem_cgroup *memcg = dead_memcg;
912 struct mem_cgroup_reclaim_iter *iter;
913 struct mem_cgroup_per_zone *mz;
914 int nid, zid;
915 int i;
916
917 while ((memcg = parent_mem_cgroup(memcg))) {
918 for_each_node(nid) {
919 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
920 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
921 for (i = 0; i <= DEF_PRIORITY; i++) {
922 iter = &mz->iter[i];
923 cmpxchg(&iter->position,
924 dead_memcg, NULL);
925 }
926 }
927 }
928 }
929}
930
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700931/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800932 * Iteration constructs for visiting all cgroups (under a tree). If
933 * loops are exited prematurely (break), mem_cgroup_iter_break() must
934 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700935 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800936#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800937 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800938 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800939 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700940
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800941#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800942 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800943 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800944 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700945
Johannes Weiner925b7672012-01-12 17:18:15 -0800946/**
947 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
948 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700949 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -0800950 *
951 * Returns the lru list vector holding pages for the given @zone and
952 * @mem. This can be the global zone lruvec, if the memory controller
953 * is disabled.
954 */
955struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
956 struct mem_cgroup *memcg)
957{
958 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800959 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -0800960
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800961 if (mem_cgroup_disabled()) {
962 lruvec = &zone->lruvec;
963 goto out;
964 }
Johannes Weiner925b7672012-01-12 17:18:15 -0800965
Jianyu Zhane2318752014-06-06 14:38:20 -0700966 mz = mem_cgroup_zone_zoneinfo(memcg, zone);
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800967 lruvec = &mz->lruvec;
968out:
969 /*
970 * Since a node can be onlined after the mem_cgroup was created,
971 * we have to be prepared to initialize lruvec->zone here;
972 * and if offlined then reonlined, we need to reinitialize it.
973 */
974 if (unlikely(lruvec->zone != zone))
975 lruvec->zone = zone;
976 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -0800977}
978
Johannes Weiner925b7672012-01-12 17:18:15 -0800979/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800980 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -0800981 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700982 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800983 *
984 * This function is only safe when following the LRU page isolation
985 * and putback protocol: the LRU lock must be held, and the page must
986 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -0700987 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700988struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -0700989{
990 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -0800991 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800992 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800993
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800994 if (mem_cgroup_disabled()) {
995 lruvec = &zone->lruvec;
996 goto out;
997 }
Christoph Lameterb69408e2008-10-18 20:26:14 -0700998
Johannes Weiner1306a852014-12-10 15:44:52 -0800999 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001000 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001001 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -08001002 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -08001003 */
Johannes Weiner29833312014-12-10 15:44:02 -08001004 if (!memcg)
1005 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001006
Jianyu Zhane2318752014-06-06 14:38:20 -07001007 mz = mem_cgroup_page_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001008 lruvec = &mz->lruvec;
1009out:
1010 /*
1011 * Since a node can be onlined after the mem_cgroup was created,
1012 * we have to be prepared to initialize lruvec->zone here;
1013 * and if offlined then reonlined, we need to reinitialize it.
1014 */
1015 if (unlikely(lruvec->zone != zone))
1016 lruvec->zone = zone;
1017 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001018}
1019
1020/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001021 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1022 * @lruvec: mem_cgroup per zone lru vector
1023 * @lru: index of lru list the page is sitting on
1024 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001025 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001026 * This function must be called when a page is added to or removed from an
1027 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001028 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001029void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1030 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001031{
1032 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001033 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001034
1035 if (mem_cgroup_disabled())
1036 return;
1037
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001038 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1039 lru_size = mz->lru_size + lru;
1040 *lru_size += nr_pages;
1041 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001042}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001043
Johannes Weiner2314b422014-12-10 15:44:33 -08001044bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001045{
Johannes Weiner2314b422014-12-10 15:44:33 -08001046 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001047 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001048 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001049
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001050 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001051 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001052 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001053 task_unlock(p);
1054 } else {
1055 /*
1056 * All threads may have already detached their mm's, but the oom
1057 * killer still needs to detect if they have already been oom
1058 * killed to prevent needlessly killing additional tasks.
1059 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001060 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001061 task_memcg = mem_cgroup_from_task(task);
1062 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001063 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001064 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001065 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1066 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001067 return ret;
1068}
1069
Johannes Weiner19942822011-02-01 15:52:43 -08001070/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001071 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001072 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001073 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001074 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001075 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001076 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001077static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001078{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001079 unsigned long margin = 0;
1080 unsigned long count;
1081 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001082
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001083 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001084 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001085 if (count < limit)
1086 margin = limit - count;
1087
Johannes Weiner7941d212016-01-14 15:21:23 -08001088 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001089 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001090 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001091 if (count <= limit)
1092 margin = min(margin, limit - count);
1093 }
1094
1095 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001096}
1097
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001098/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001099 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001100 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001101 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1102 * moving cgroups. This is for waiting at high-memory pressure
1103 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001104 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001105static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001106{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001107 struct mem_cgroup *from;
1108 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001109 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001110 /*
1111 * Unlike task_move routines, we access mc.to, mc.from not under
1112 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1113 */
1114 spin_lock(&mc.lock);
1115 from = mc.from;
1116 to = mc.to;
1117 if (!from)
1118 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001119
Johannes Weiner2314b422014-12-10 15:44:33 -08001120 ret = mem_cgroup_is_descendant(from, memcg) ||
1121 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001122unlock:
1123 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001124 return ret;
1125}
1126
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001127static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001128{
1129 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001130 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001131 DEFINE_WAIT(wait);
1132 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1133 /* moving charge context might have finished. */
1134 if (mc.moving_task)
1135 schedule();
1136 finish_wait(&mc.waitq, &wait);
1137 return true;
1138 }
1139 }
1140 return false;
1141}
1142
Sha Zhengju58cf1882013-02-22 16:32:05 -08001143#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001144/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001145 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001146 * @memcg: The memory cgroup that went over limit
1147 * @p: Task that is going to be killed
1148 *
1149 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1150 * enabled
1151 */
1152void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1153{
Sha Zhengju58cf1882013-02-22 16:32:05 -08001154 struct mem_cgroup *iter;
1155 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001156
Balbir Singhe2224322009-04-02 16:57:39 -07001157 rcu_read_lock();
1158
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001159 if (p) {
1160 pr_info("Task in ");
1161 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1162 pr_cont(" killed as a result of limit of ");
1163 } else {
1164 pr_info("Memory limit reached of cgroup ");
1165 }
1166
Tejun Heoe61734c2014-02-12 09:29:50 -05001167 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001168 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001169
Balbir Singhe2224322009-04-02 16:57:39 -07001170 rcu_read_unlock();
1171
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001172 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1173 K((u64)page_counter_read(&memcg->memory)),
1174 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1175 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1176 K((u64)page_counter_read(&memcg->memsw)),
1177 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1178 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1179 K((u64)page_counter_read(&memcg->kmem)),
1180 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001181
1182 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001183 pr_info("Memory cgroup stats for ");
1184 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001185 pr_cont(":");
1186
1187 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08001188 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Sha Zhengju58cf1882013-02-22 16:32:05 -08001189 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07001190 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
Sha Zhengju58cf1882013-02-22 16:32:05 -08001191 K(mem_cgroup_read_stat(iter, i)));
1192 }
1193
1194 for (i = 0; i < NR_LRU_LISTS; i++)
1195 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1196 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1197
1198 pr_cont("\n");
1199 }
Balbir Singhe2224322009-04-02 16:57:39 -07001200}
1201
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001202/*
1203 * This function returns the number of memcg under hierarchy tree. Returns
1204 * 1(self count) if no children.
1205 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001206static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001207{
1208 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001209 struct mem_cgroup *iter;
1210
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001211 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001212 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001213 return num;
1214}
1215
Balbir Singh6d61ef42009-01-07 18:08:06 -08001216/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001217 * Return the memory (and swap, if configured) limit for a memcg.
1218 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001219static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001220{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001221 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001222
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001223 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001224 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001225 unsigned long memsw_limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001226 unsigned long swap_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001227
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001228 memsw_limit = memcg->memsw.limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001229 swap_limit = memcg->swap.limit;
1230 swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1231 limit = min(limit + swap_limit, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001232 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001233 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001234}
1235
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001236static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001237 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001238{
David Rientjes6e0fc462015-09-08 15:00:36 -07001239 struct oom_control oc = {
1240 .zonelist = NULL,
1241 .nodemask = NULL,
1242 .gfp_mask = gfp_mask,
1243 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001244 };
David Rientjes9cbb78b2012-07-31 16:43:44 -07001245 struct mem_cgroup *iter;
1246 unsigned long chosen_points = 0;
1247 unsigned long totalpages;
1248 unsigned int points = 0;
1249 struct task_struct *chosen = NULL;
1250
Johannes Weinerdc564012015-06-24 16:57:19 -07001251 mutex_lock(&oom_lock);
1252
David Rientjes876aafb2012-07-31 16:43:48 -07001253 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001254 * If current has a pending SIGKILL or is exiting, then automatically
1255 * select it. The goal is to allow it to allocate so that it may
1256 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001257 */
Oleg Nesterovd003f372014-12-12 16:56:24 -08001258 if (fatal_signal_pending(current) || task_will_free_mem(current)) {
Johannes Weiner16e95192015-06-24 16:57:07 -07001259 mark_oom_victim(current);
Johannes Weinerdc564012015-06-24 16:57:19 -07001260 goto unlock;
David Rientjes876aafb2012-07-31 16:43:48 -07001261 }
1262
David Rientjes6e0fc462015-09-08 15:00:36 -07001263 check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001264 totalpages = mem_cgroup_get_limit(memcg) ? : 1;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001265 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001266 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001267 struct task_struct *task;
1268
Tejun Heo72ec7022013-08-08 20:11:26 -04001269 css_task_iter_start(&iter->css, &it);
1270 while ((task = css_task_iter_next(&it))) {
David Rientjes6e0fc462015-09-08 15:00:36 -07001271 switch (oom_scan_process_thread(&oc, task, totalpages)) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001272 case OOM_SCAN_SELECT:
1273 if (chosen)
1274 put_task_struct(chosen);
1275 chosen = task;
1276 chosen_points = ULONG_MAX;
1277 get_task_struct(chosen);
1278 /* fall through */
1279 case OOM_SCAN_CONTINUE:
1280 continue;
1281 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001282 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001283 mem_cgroup_iter_break(memcg, iter);
1284 if (chosen)
1285 put_task_struct(chosen);
Johannes Weinerdc564012015-06-24 16:57:19 -07001286 goto unlock;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001287 case OOM_SCAN_OK:
1288 break;
1289 };
1290 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001291 if (!points || points < chosen_points)
1292 continue;
1293 /* Prefer thread group leaders for display purposes */
1294 if (points == chosen_points &&
1295 thread_group_leader(chosen))
1296 continue;
1297
1298 if (chosen)
1299 put_task_struct(chosen);
1300 chosen = task;
1301 chosen_points = points;
1302 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001303 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001304 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001305 }
1306
Johannes Weinerdc564012015-06-24 16:57:19 -07001307 if (chosen) {
1308 points = chosen_points * 1000 / totalpages;
David Rientjes6e0fc462015-09-08 15:00:36 -07001309 oom_kill_process(&oc, chosen, points, totalpages, memcg,
1310 "Memory cgroup out of memory");
Johannes Weinerdc564012015-06-24 16:57:19 -07001311 }
1312unlock:
1313 mutex_unlock(&oom_lock);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001314 return chosen;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001315}
1316
Michele Curtiae6e71d2014-12-12 16:56:35 -08001317#if MAX_NUMNODES > 1
1318
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001319/**
1320 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001321 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001322 * @nid: the node ID to be checked.
1323 * @noswap : specify true here if the user wants flle only information.
1324 *
1325 * This function returns whether the specified memcg contains any
1326 * reclaimable pages on a node. Returns true if there are any reclaimable
1327 * pages in the node.
1328 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001329static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001330 int nid, bool noswap)
1331{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001332 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001333 return true;
1334 if (noswap || !total_swap_pages)
1335 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001336 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001337 return true;
1338 return false;
1339
1340}
Ying Han889976d2011-05-26 16:25:33 -07001341
1342/*
1343 * Always updating the nodemask is not very good - even if we have an empty
1344 * list or the wrong list here, we can start from some node and traverse all
1345 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1346 *
1347 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001348static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001349{
1350 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001351 /*
1352 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1353 * pagein/pageout changes since the last update.
1354 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001355 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001356 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001357 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001358 return;
1359
Ying Han889976d2011-05-26 16:25:33 -07001360 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001361 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001362
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001363 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001364
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001365 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1366 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001367 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001368
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001369 atomic_set(&memcg->numainfo_events, 0);
1370 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001371}
1372
1373/*
1374 * Selecting a node where we start reclaim from. Because what we need is just
1375 * reducing usage counter, start from anywhere is O,K. Considering
1376 * memory reclaim from current node, there are pros. and cons.
1377 *
1378 * Freeing memory from current node means freeing memory from a node which
1379 * we'll use or we've used. So, it may make LRU bad. And if several threads
1380 * hit limits, it will see a contention on a node. But freeing from remote
1381 * node means more costs for memory reclaim because of memory latency.
1382 *
1383 * Now, we use round-robin. Better algorithm is welcomed.
1384 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001385int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001386{
1387 int node;
1388
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001389 mem_cgroup_may_update_nodemask(memcg);
1390 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001391
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001392 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001393 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001394 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001395 /*
1396 * We call this when we hit limit, not when pages are added to LRU.
1397 * No LRU may hold pages because all pages are UNEVICTABLE or
1398 * memcg is too small and all pages are not on LRU. In that case,
1399 * we use curret node.
1400 */
1401 if (unlikely(node == MAX_NUMNODES))
1402 node = numa_node_id();
1403
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001404 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001405 return node;
1406}
Ying Han889976d2011-05-26 16:25:33 -07001407#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001408int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001409{
1410 return 0;
1411}
1412#endif
1413
Andrew Morton0608f432013-09-24 15:27:41 -07001414static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1415 struct zone *zone,
1416 gfp_t gfp_mask,
1417 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001418{
Andrew Morton0608f432013-09-24 15:27:41 -07001419 struct mem_cgroup *victim = NULL;
1420 int total = 0;
1421 int loop = 0;
1422 unsigned long excess;
1423 unsigned long nr_scanned;
1424 struct mem_cgroup_reclaim_cookie reclaim = {
1425 .zone = zone,
1426 .priority = 0,
1427 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001428
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001429 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001430
Andrew Morton0608f432013-09-24 15:27:41 -07001431 while (1) {
1432 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1433 if (!victim) {
1434 loop++;
1435 if (loop >= 2) {
1436 /*
1437 * If we have not been able to reclaim
1438 * anything, it might because there are
1439 * no reclaimable pages under this hierarchy
1440 */
1441 if (!total)
1442 break;
1443 /*
1444 * We want to do more targeted reclaim.
1445 * excess >> 2 is not to excessive so as to
1446 * reclaim too much, nor too less that we keep
1447 * coming back to reclaim from this cgroup
1448 */
1449 if (total >= (excess >> 2) ||
1450 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1451 break;
1452 }
1453 continue;
1454 }
Andrew Morton0608f432013-09-24 15:27:41 -07001455 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1456 zone, &nr_scanned);
1457 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001458 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001459 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001460 }
Andrew Morton0608f432013-09-24 15:27:41 -07001461 mem_cgroup_iter_break(root_memcg, victim);
1462 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001463}
1464
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001465#ifdef CONFIG_LOCKDEP
1466static struct lockdep_map memcg_oom_lock_dep_map = {
1467 .name = "memcg_oom_lock",
1468};
1469#endif
1470
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001471static DEFINE_SPINLOCK(memcg_oom_lock);
1472
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001473/*
1474 * Check OOM-Killer is already running under our hierarchy.
1475 * If someone is running, return false.
1476 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001477static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001478{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001479 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001480
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001481 spin_lock(&memcg_oom_lock);
1482
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001483 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001484 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001485 /*
1486 * this subtree of our hierarchy is already locked
1487 * so we cannot give a lock.
1488 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001489 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001490 mem_cgroup_iter_break(memcg, iter);
1491 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001492 } else
1493 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001494 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001495
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001496 if (failed) {
1497 /*
1498 * OK, we failed to lock the whole subtree so we have
1499 * to clean up what we set up to the failing subtree
1500 */
1501 for_each_mem_cgroup_tree(iter, memcg) {
1502 if (iter == failed) {
1503 mem_cgroup_iter_break(memcg, iter);
1504 break;
1505 }
1506 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001507 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001508 } else
1509 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001510
1511 spin_unlock(&memcg_oom_lock);
1512
1513 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001514}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001515
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001516static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001517{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001518 struct mem_cgroup *iter;
1519
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001520 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001521 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001522 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001523 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001524 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001525}
1526
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001527static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001528{
1529 struct mem_cgroup *iter;
1530
Tejun Heoc2b42d32015-06-24 16:58:23 -07001531 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001532 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001533 iter->under_oom++;
1534 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001535}
1536
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001537static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001538{
1539 struct mem_cgroup *iter;
1540
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001541 /*
1542 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001543 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001544 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001545 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001546 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001547 if (iter->under_oom > 0)
1548 iter->under_oom--;
1549 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001550}
1551
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001552static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1553
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001554struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001555 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001556 wait_queue_t wait;
1557};
1558
1559static int memcg_oom_wake_function(wait_queue_t *wait,
1560 unsigned mode, int sync, void *arg)
1561{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001562 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1563 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001564 struct oom_wait_info *oom_wait_info;
1565
1566 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001567 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001568
Johannes Weiner2314b422014-12-10 15:44:33 -08001569 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1570 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001571 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001572 return autoremove_wake_function(wait, mode, sync, arg);
1573}
1574
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001575static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001576{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001577 /*
1578 * For the following lockless ->under_oom test, the only required
1579 * guarantee is that it must see the state asserted by an OOM when
1580 * this function is called as a result of userland actions
1581 * triggered by the notification of the OOM. This is trivially
1582 * achieved by invoking mem_cgroup_mark_under_oom() before
1583 * triggering notification.
1584 */
1585 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001586 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001587}
1588
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001589static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001590{
Tejun Heo626ebc42015-11-05 18:46:09 -08001591 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001592 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001593 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001594 * We are in the middle of the charge context here, so we
1595 * don't want to block when potentially sitting on a callstack
1596 * that holds all kinds of filesystem and mm locks.
1597 *
1598 * Also, the caller may handle a failed allocation gracefully
1599 * (like optional page cache readahead) and so an OOM killer
1600 * invocation might not even be necessary.
1601 *
1602 * That's why we don't do anything here except remember the
1603 * OOM context and then deal with it at the end of the page
1604 * fault when the stack is unwound, the locks are released,
1605 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001606 */
Johannes Weiner49426422013-10-16 13:46:59 -07001607 css_get(&memcg->css);
Tejun Heo626ebc42015-11-05 18:46:09 -08001608 current->memcg_in_oom = memcg;
1609 current->memcg_oom_gfp_mask = mask;
1610 current->memcg_oom_order = order;
Johannes Weiner49426422013-10-16 13:46:59 -07001611}
1612
1613/**
1614 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1615 * @handle: actually kill/wait or just clean up the OOM state
1616 *
1617 * This has to be called at the end of a page fault if the memcg OOM
1618 * handler was enabled.
1619 *
1620 * Memcg supports userspace OOM handling where failed allocations must
1621 * sleep on a waitqueue until the userspace task resolves the
1622 * situation. Sleeping directly in the charge context with all kinds
1623 * of locks held is not a good idea, instead we remember an OOM state
1624 * in the task and mem_cgroup_oom_synchronize() has to be called at
1625 * the end of the page fault to complete the OOM handling.
1626 *
1627 * Returns %true if an ongoing memcg OOM situation was detected and
1628 * completed, %false otherwise.
1629 */
1630bool mem_cgroup_oom_synchronize(bool handle)
1631{
Tejun Heo626ebc42015-11-05 18:46:09 -08001632 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001633 struct oom_wait_info owait;
1634 bool locked;
1635
1636 /* OOM is global, do not handle */
1637 if (!memcg)
1638 return false;
1639
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001640 if (!handle || oom_killer_disabled)
Johannes Weiner49426422013-10-16 13:46:59 -07001641 goto cleanup;
1642
1643 owait.memcg = memcg;
1644 owait.wait.flags = 0;
1645 owait.wait.func = memcg_oom_wake_function;
1646 owait.wait.private = current;
1647 INIT_LIST_HEAD(&owait.wait.task_list);
1648
1649 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001650 mem_cgroup_mark_under_oom(memcg);
1651
1652 locked = mem_cgroup_oom_trylock(memcg);
1653
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001654 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001655 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001656
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001657 if (locked && !memcg->oom_kill_disable) {
1658 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001659 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001660 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1661 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001662 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001663 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001664 mem_cgroup_unmark_under_oom(memcg);
1665 finish_wait(&memcg_oom_waitq, &owait.wait);
1666 }
1667
1668 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001669 mem_cgroup_oom_unlock(memcg);
1670 /*
1671 * There is no guarantee that an OOM-lock contender
1672 * sees the wakeups triggered by the OOM kill
1673 * uncharges. Wake any sleepers explicitely.
1674 */
1675 memcg_oom_recover(memcg);
1676 }
Johannes Weiner49426422013-10-16 13:46:59 -07001677cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001678 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001679 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001680 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001681}
1682
Johannes Weinerd7365e72014-10-29 14:50:48 -07001683/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001684 * lock_page_memcg - lock a page->mem_cgroup binding
1685 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001686 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001687 * This function protects unlocked LRU pages from being moved to
1688 * another cgroup and stabilizes their page->mem_cgroup binding.
Balbir Singhd69b0422009-06-17 16:26:34 -07001689 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001690void lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001691{
1692 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001693 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001694
Johannes Weiner6de22612015-02-11 15:25:01 -08001695 /*
1696 * The RCU lock is held throughout the transaction. The fast
1697 * path can get away without acquiring the memcg->move_lock
1698 * because page moving starts with an RCU grace period.
Johannes Weiner6de22612015-02-11 15:25:01 -08001699 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001700 rcu_read_lock();
1701
1702 if (mem_cgroup_disabled())
Johannes Weiner62cccb82016-03-15 14:57:22 -07001703 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001704again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001705 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001706 if (unlikely(!memcg))
Johannes Weiner62cccb82016-03-15 14:57:22 -07001707 return;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001708
Qiang Huangbdcbb652014-06-04 16:08:21 -07001709 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001710 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001711
Johannes Weiner6de22612015-02-11 15:25:01 -08001712 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001713 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001714 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001715 goto again;
1716 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001717
1718 /*
1719 * When charge migration first begins, we can have locked and
1720 * unlocked page stat updates happening concurrently. Track
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001721 * the task who has the lock for unlock_page_memcg().
Johannes Weiner6de22612015-02-11 15:25:01 -08001722 */
1723 memcg->move_lock_task = current;
1724 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001725
Johannes Weiner62cccb82016-03-15 14:57:22 -07001726 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001727}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001728EXPORT_SYMBOL(lock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001729
Johannes Weinerd7365e72014-10-29 14:50:48 -07001730/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001731 * unlock_page_memcg - unlock a page->mem_cgroup binding
Johannes Weiner62cccb82016-03-15 14:57:22 -07001732 * @page: the page
Johannes Weinerd7365e72014-10-29 14:50:48 -07001733 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001734void unlock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001735{
Johannes Weiner62cccb82016-03-15 14:57:22 -07001736 struct mem_cgroup *memcg = page->mem_cgroup;
1737
Johannes Weiner6de22612015-02-11 15:25:01 -08001738 if (memcg && memcg->move_lock_task == current) {
1739 unsigned long flags = memcg->move_lock_flags;
1740
1741 memcg->move_lock_task = NULL;
1742 memcg->move_lock_flags = 0;
1743
1744 spin_unlock_irqrestore(&memcg->move_lock, flags);
1745 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001746
Johannes Weinerd7365e72014-10-29 14:50:48 -07001747 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001748}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001749EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001750
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001751/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001752 * size of first charge trial. "32" comes from vmscan.c's magic value.
1753 * TODO: maybe necessary to use big numbers in big irons.
1754 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001755#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001756struct memcg_stock_pcp {
1757 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001758 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001759 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001760 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001761#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001762};
1763static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001764static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001765
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001766/**
1767 * consume_stock: Try to consume stocked charge on this cpu.
1768 * @memcg: memcg to consume from.
1769 * @nr_pages: how many pages to charge.
1770 *
1771 * The charges will only happen if @memcg matches the current cpu's memcg
1772 * stock, and at least @nr_pages are available in that stock. Failure to
1773 * service an allocation will refill the stock.
1774 *
1775 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001776 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001777static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001778{
1779 struct memcg_stock_pcp *stock;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001780 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001781
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001782 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001783 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001784
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001785 stock = &get_cpu_var(memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001786 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001787 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001788 ret = true;
1789 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001790 put_cpu_var(memcg_stock);
1791 return ret;
1792}
1793
1794/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001795 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001796 */
1797static void drain_stock(struct memcg_stock_pcp *stock)
1798{
1799 struct mem_cgroup *old = stock->cached;
1800
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001801 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001802 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001803 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001804 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001805 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001806 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001807 }
1808 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001809}
1810
1811/*
1812 * This must be called under preempt disabled or must be called by
1813 * a thread which is pinned to local cpu.
1814 */
1815static void drain_local_stock(struct work_struct *dummy)
1816{
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001817 struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001818 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001819 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001820}
1821
1822/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001823 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001824 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001825 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001826static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001827{
1828 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1829
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001830 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001831 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001832 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001833 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001834 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001835 put_cpu_var(memcg_stock);
1836}
1837
1838/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001839 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001840 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001841 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001842static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001843{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001844 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001845
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001846 /* If someone's already draining, avoid adding running more workers. */
1847 if (!mutex_trylock(&percpu_charge_mutex))
1848 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001849 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001850 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001851 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001852 for_each_online_cpu(cpu) {
1853 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001854 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001855
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001856 memcg = stock->cached;
1857 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001858 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001859 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001860 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001861 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1862 if (cpu == curcpu)
1863 drain_local_stock(&stock->work);
1864 else
1865 schedule_work_on(cpu, &stock->work);
1866 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001867 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001868 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001869 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001870 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001871}
1872
Paul Gortmaker0db06282013-06-19 14:53:51 -04001873static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001874 unsigned long action,
1875 void *hcpu)
1876{
1877 int cpu = (unsigned long)hcpu;
1878 struct memcg_stock_pcp *stock;
1879
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001880 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001881 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001882
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07001883 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001884 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001885
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001886 stock = &per_cpu(memcg_stock, cpu);
1887 drain_stock(stock);
1888 return NOTIFY_OK;
1889}
1890
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001891static void reclaim_high(struct mem_cgroup *memcg,
1892 unsigned int nr_pages,
1893 gfp_t gfp_mask)
1894{
1895 do {
1896 if (page_counter_read(&memcg->memory) <= memcg->high)
1897 continue;
1898 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
1899 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1900 } while ((memcg = parent_mem_cgroup(memcg)));
1901}
1902
1903static void high_work_func(struct work_struct *work)
1904{
1905 struct mem_cgroup *memcg;
1906
1907 memcg = container_of(work, struct mem_cgroup, high_work);
1908 reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1909}
1910
Tejun Heob23afb92015-11-05 18:46:11 -08001911/*
1912 * Scheduled by try_charge() to be executed from the userland return path
1913 * and reclaims memory over the high limit.
1914 */
1915void mem_cgroup_handle_over_high(void)
1916{
1917 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001918 struct mem_cgroup *memcg;
Tejun Heob23afb92015-11-05 18:46:11 -08001919
1920 if (likely(!nr_pages))
1921 return;
1922
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001923 memcg = get_mem_cgroup_from_mm(current->mm);
1924 reclaim_high(memcg, nr_pages, GFP_KERNEL);
Tejun Heob23afb92015-11-05 18:46:11 -08001925 css_put(&memcg->css);
1926 current->memcg_nr_pages_over_high = 0;
1927}
1928
Johannes Weiner00501b52014-08-08 14:19:20 -07001929static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1930 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001931{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001932 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001933 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001934 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001935 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001936 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001937 bool may_swap = true;
1938 bool drained = false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001939
Johannes Weinerce00a962014-09-05 08:43:57 -04001940 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08001941 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001942retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001943 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08001944 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001945
Johannes Weiner7941d212016-01-14 15:21:23 -08001946 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08001947 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1948 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001949 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08001950 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001951 page_counter_uncharge(&memcg->memsw, batch);
1952 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001953 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001954 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001955 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001956 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001957
Johannes Weiner6539cc02014-08-06 16:05:42 -07001958 if (batch > nr_pages) {
1959 batch = nr_pages;
1960 goto retry;
1961 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001962
Johannes Weiner06b078f2014-08-06 16:05:44 -07001963 /*
1964 * Unlike in global OOM situations, memcg is not in a physical
1965 * memory shortage. Allow dying and OOM-killed tasks to
1966 * bypass the last charges so that they can exit quickly and
1967 * free their memory.
1968 */
1969 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1970 fatal_signal_pending(current) ||
1971 current->flags & PF_EXITING))
Tejun Heo10d53c72015-11-05 18:46:17 -08001972 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001973
1974 if (unlikely(task_in_memcg_oom(current)))
1975 goto nomem;
1976
Mel Gormand0164ad2015-11-06 16:28:21 -08001977 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001978 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001979
Johannes Weiner241994e2015-02-11 15:26:06 -08001980 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
1981
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001982 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1983 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07001984
Johannes Weiner61e02c72014-08-06 16:08:16 -07001985 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07001986 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07001987
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001988 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001989 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001990 drained = true;
1991 goto retry;
1992 }
1993
Johannes Weiner28c34c22014-08-06 16:05:47 -07001994 if (gfp_mask & __GFP_NORETRY)
1995 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001996 /*
1997 * Even though the limit is exceeded at this point, reclaim
1998 * may have been able to free some pages. Retry the charge
1999 * before killing the task.
2000 *
2001 * Only for regular pages, though: huge pages are rather
2002 * unlikely to succeed so close to the limit, and we fall back
2003 * to regular pages anyway in case of failure.
2004 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002005 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002006 goto retry;
2007 /*
2008 * At task move, charge accounts can be doubly counted. So, it's
2009 * better to wait until the end of task_move if something is going on.
2010 */
2011 if (mem_cgroup_wait_acct_move(mem_over_limit))
2012 goto retry;
2013
Johannes Weiner9b130612014-08-06 16:05:51 -07002014 if (nr_retries--)
2015 goto retry;
2016
Johannes Weiner06b078f2014-08-06 16:05:44 -07002017 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08002018 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07002019
Johannes Weiner6539cc02014-08-06 16:05:42 -07002020 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08002021 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002022
Johannes Weiner241994e2015-02-11 15:26:06 -08002023 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
2024
Jerome Marchand3608de02015-11-05 18:47:29 -08002025 mem_cgroup_oom(mem_over_limit, gfp_mask,
2026 get_order(nr_pages * PAGE_SIZE));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002027nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002028 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002029 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08002030force:
2031 /*
2032 * The allocation either can't fail or will lead to more memory
2033 * being freed very soon. Allow memory usage go over the limit
2034 * temporarily by force charging it.
2035 */
2036 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002037 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08002038 page_counter_charge(&memcg->memsw, nr_pages);
2039 css_get_many(&memcg->css, nr_pages);
2040
2041 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002042
2043done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002044 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002045 if (batch > nr_pages)
2046 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002047
Johannes Weiner241994e2015-02-11 15:26:06 -08002048 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002049 * If the hierarchy is above the normal consumption range, schedule
2050 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002051 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002052 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2053 * not recorded as it most likely matches current's and won't
2054 * change in the meantime. As high limit is checked again before
2055 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994e2015-02-11 15:26:06 -08002056 */
2057 do {
Tejun Heob23afb92015-11-05 18:46:11 -08002058 if (page_counter_read(&memcg->memory) > memcg->high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002059 /* Don't bother a random interrupted task */
2060 if (in_interrupt()) {
2061 schedule_work(&memcg->high_work);
2062 break;
2063 }
Vladimir Davydov9516a182015-12-11 13:40:24 -08002064 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002065 set_notify_resume(current);
2066 break;
2067 }
Johannes Weiner241994e2015-02-11 15:26:06 -08002068 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002069
2070 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002071}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002072
Johannes Weiner00501b52014-08-08 14:19:20 -07002073static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002074{
Johannes Weinerce00a962014-09-05 08:43:57 -04002075 if (mem_cgroup_is_root(memcg))
2076 return;
2077
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002078 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002079 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002080 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002081
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002082 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002083}
2084
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002085static void lock_page_lru(struct page *page, int *isolated)
2086{
2087 struct zone *zone = page_zone(page);
2088
2089 spin_lock_irq(&zone->lru_lock);
2090 if (PageLRU(page)) {
2091 struct lruvec *lruvec;
2092
2093 lruvec = mem_cgroup_page_lruvec(page, zone);
2094 ClearPageLRU(page);
2095 del_page_from_lru_list(page, lruvec, page_lru(page));
2096 *isolated = 1;
2097 } else
2098 *isolated = 0;
2099}
2100
2101static void unlock_page_lru(struct page *page, int isolated)
2102{
2103 struct zone *zone = page_zone(page);
2104
2105 if (isolated) {
2106 struct lruvec *lruvec;
2107
2108 lruvec = mem_cgroup_page_lruvec(page, zone);
2109 VM_BUG_ON_PAGE(PageLRU(page), page);
2110 SetPageLRU(page);
2111 add_page_to_lru_list(page, lruvec, page_lru(page));
2112 }
2113 spin_unlock_irq(&zone->lru_lock);
2114}
2115
Johannes Weiner00501b52014-08-08 14:19:20 -07002116static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002117 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002118{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002119 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002120
Johannes Weiner1306a852014-12-10 15:44:52 -08002121 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002122
2123 /*
2124 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2125 * may already be on some other mem_cgroup's LRU. Take care of it.
2126 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002127 if (lrucare)
2128 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002129
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002130 /*
2131 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002132 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002133 *
2134 * - the page is uncharged
2135 *
2136 * - the page is off-LRU
2137 *
2138 * - an anonymous fault has exclusive page access, except for
2139 * a locked page table
2140 *
2141 * - a page cache insertion, a swapin fault, or a migration
2142 * have the page locked
2143 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002144 page->mem_cgroup = memcg;
Hugh Dickins3be91272008-02-07 00:14:19 -08002145
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002146 if (lrucare)
2147 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002148}
2149
Johannes Weiner127424c2016-01-20 15:02:32 -08002150#ifndef CONFIG_SLOB
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002151static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002152{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002153 int id, size;
2154 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002155
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002156 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002157 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2158 if (id < 0)
2159 return id;
2160
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002161 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002162 return id;
2163
2164 /*
2165 * There's no space for the new id in memcg_caches arrays,
2166 * so we have to grow them.
2167 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002168 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002169
2170 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002171 if (size < MEMCG_CACHES_MIN_SIZE)
2172 size = MEMCG_CACHES_MIN_SIZE;
2173 else if (size > MEMCG_CACHES_MAX_SIZE)
2174 size = MEMCG_CACHES_MAX_SIZE;
2175
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002176 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002177 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002178 err = memcg_update_all_list_lrus(size);
2179 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002180 memcg_nr_cache_ids = size;
2181
2182 up_write(&memcg_cache_ids_sem);
2183
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002184 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002185 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002186 return err;
2187 }
2188 return id;
2189}
2190
2191static void memcg_free_cache_id(int id)
2192{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002193 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002194}
2195
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002196struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002197 struct mem_cgroup *memcg;
2198 struct kmem_cache *cachep;
2199 struct work_struct work;
2200};
2201
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002202static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002203{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002204 struct memcg_kmem_cache_create_work *cw =
2205 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002206 struct mem_cgroup *memcg = cw->memcg;
2207 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002208
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002209 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002210
Vladimir Davydov5722d092014-04-07 15:39:24 -07002211 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002212 kfree(cw);
2213}
2214
2215/*
2216 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002217 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002218static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2219 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002220{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002221 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002222
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002223 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002224 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002225 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002226
2227 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002228
2229 cw->memcg = memcg;
2230 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002231 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002232
Glauber Costad7f25f82012-12-18 14:22:40 -08002233 schedule_work(&cw->work);
2234}
2235
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002236static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2237 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002238{
2239 /*
2240 * We need to stop accounting when we kmalloc, because if the
2241 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002242 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002243 *
2244 * However, it is better to enclose the whole function. Depending on
2245 * the debugging options enabled, INIT_WORK(), for instance, can
2246 * trigger an allocation. This too, will make us recurse. Because at
2247 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2248 * the safest choice is to do it like this, wrapping the whole function.
2249 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002250 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002251 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002252 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002253}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002254
Glauber Costad7f25f82012-12-18 14:22:40 -08002255/*
2256 * Return the kmem_cache we're supposed to use for a slab allocation.
2257 * We try to use the current memcg's version of the cache.
2258 *
2259 * If the cache does not exist yet, if we are the first user of it,
2260 * we either create it immediately, if possible, or create it asynchronously
2261 * in a workqueue.
2262 * In the latter case, we will let the current allocation go through with
2263 * the original cache.
2264 *
2265 * Can't be called in interrupt context or from kernel threads.
2266 * This function needs to be called with rcu_read_lock() held.
2267 */
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002268struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
Glauber Costad7f25f82012-12-18 14:22:40 -08002269{
2270 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002271 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002272 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002273
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002274 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002275
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002276 if (cachep->flags & SLAB_ACCOUNT)
2277 gfp |= __GFP_ACCOUNT;
2278
2279 if (!(gfp & __GFP_ACCOUNT))
2280 return cachep;
2281
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002282 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002283 return cachep;
2284
Vladimir Davydov8135be52014-12-12 16:56:38 -08002285 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002286 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002287 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002288 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002289
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002290 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002291 if (likely(memcg_cachep))
2292 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002293
2294 /*
2295 * If we are in a safe context (can wait, and not in interrupt
2296 * context), we could be be predictable and return right away.
2297 * This would guarantee that the allocation being performed
2298 * already belongs in the new cache.
2299 *
2300 * However, there are some clashes that can arrive from locking.
2301 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002302 * memcg_create_kmem_cache, this means no further allocation
2303 * could happen with the slab_mutex held. So it's better to
2304 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002305 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002306 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002307out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002308 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002309 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002310}
Glauber Costad7f25f82012-12-18 14:22:40 -08002311
Vladimir Davydov8135be52014-12-12 16:56:38 -08002312void __memcg_kmem_put_cache(struct kmem_cache *cachep)
2313{
2314 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002315 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002316}
2317
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002318int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2319 struct mem_cgroup *memcg)
2320{
2321 unsigned int nr_pages = 1 << order;
2322 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002323 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002324
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002325 ret = try_charge(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08002326 if (ret)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002327 return ret;
Johannes Weiner52c29b02016-01-20 15:02:35 -08002328
2329 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2330 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2331 cancel_charge(memcg, nr_pages);
2332 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002333 }
2334
2335 page->mem_cgroup = memcg;
2336
2337 return 0;
2338}
2339
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002340int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002341{
2342 struct mem_cgroup *memcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002343 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002344
Johannes Weinerdf381972014-04-07 15:37:43 -07002345 memcg = get_mem_cgroup_from_mm(current->mm);
Vladimir Davydovb6ecd2d2016-03-17 14:18:33 -07002346 if (!mem_cgroup_is_root(memcg))
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002347 ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002348 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002349 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002350}
2351
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002352void __memcg_kmem_uncharge(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002353{
Johannes Weiner1306a852014-12-10 15:44:52 -08002354 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002355 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002356
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002357 if (!memcg)
2358 return;
2359
Sasha Levin309381fea2014-01-23 15:52:54 -08002360 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002361
Johannes Weiner52c29b02016-01-20 15:02:35 -08002362 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2363 page_counter_uncharge(&memcg->kmem, nr_pages);
2364
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002365 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002366 if (do_memsw_account())
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002367 page_counter_uncharge(&memcg->memsw, nr_pages);
2368
Johannes Weiner1306a852014-12-10 15:44:52 -08002369 page->mem_cgroup = NULL;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002370 css_put_many(&memcg->css, nr_pages);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002371}
Johannes Weiner127424c2016-01-20 15:02:32 -08002372#endif /* !CONFIG_SLOB */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002373
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002374#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2375
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002376/*
2377 * Because tail pages are not marked as "used", set it. We're under
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08002378 * zone->lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002379 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002380void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002381{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002382 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002383
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002384 if (mem_cgroup_disabled())
2385 return;
David Rientjesb070e652013-05-07 16:18:09 -07002386
Johannes Weiner29833312014-12-10 15:44:02 -08002387 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002388 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002389
Johannes Weiner1306a852014-12-10 15:44:52 -08002390 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002391 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002392}
Hugh Dickins12d27102012-01-12 17:19:52 -08002393#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002394
Andrew Mortonc255a452012-07-31 16:43:02 -07002395#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002396static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2397 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002398{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002399 int val = (charge) ? 1 : -1;
2400 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002401}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002402
2403/**
2404 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2405 * @entry: swap entry to be moved
2406 * @from: mem_cgroup which the entry is moved from
2407 * @to: mem_cgroup which the entry is moved to
2408 *
2409 * It succeeds only when the swap_cgroup's record for this entry is the same
2410 * as the mem_cgroup's id of @from.
2411 *
2412 * Returns 0 on success, -EINVAL on failure.
2413 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002414 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002415 * both res and memsw, and called css_get().
2416 */
2417static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002418 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002419{
2420 unsigned short old_id, new_id;
2421
Li Zefan34c00c32013-09-23 16:56:01 +08002422 old_id = mem_cgroup_id(from);
2423 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002424
2425 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002426 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002427 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002428 return 0;
2429 }
2430 return -EINVAL;
2431}
2432#else
2433static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002434 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002435{
2436 return -EINVAL;
2437}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002438#endif
2439
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002440static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002441
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002442static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002443 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002444{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002445 unsigned long curusage;
2446 unsigned long oldusage;
2447 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002448 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002449 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002450
2451 /*
2452 * For keeping hierarchical_reclaim simple, how long we should retry
2453 * is depends on callers. We set our retry-count to be function
2454 * of # of children which we should visit in this loop.
2455 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002456 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2457 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002458
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002459 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002460
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002461 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002462 if (signal_pending(current)) {
2463 ret = -EINTR;
2464 break;
2465 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002466
2467 mutex_lock(&memcg_limit_mutex);
2468 if (limit > memcg->memsw.limit) {
2469 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002470 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002471 break;
2472 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002473 if (limit > memcg->memory.limit)
2474 enlarge = true;
2475 ret = page_counter_limit(&memcg->memory, limit);
2476 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002477
2478 if (!ret)
2479 break;
2480
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002481 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2482
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002483 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002484 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002485 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002486 retry_count--;
2487 else
2488 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002489 } while (retry_count);
2490
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002491 if (!ret && enlarge)
2492 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002493
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002494 return ret;
2495}
2496
Li Zefan338c8432009-06-17 16:27:15 -07002497static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002498 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002499{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002500 unsigned long curusage;
2501 unsigned long oldusage;
2502 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002503 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002504 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002505
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002506 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002507 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2508 mem_cgroup_count_children(memcg);
2509
2510 oldusage = page_counter_read(&memcg->memsw);
2511
2512 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002513 if (signal_pending(current)) {
2514 ret = -EINTR;
2515 break;
2516 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002517
2518 mutex_lock(&memcg_limit_mutex);
2519 if (limit < memcg->memory.limit) {
2520 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002521 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002522 break;
2523 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002524 if (limit > memcg->memsw.limit)
2525 enlarge = true;
2526 ret = page_counter_limit(&memcg->memsw, limit);
2527 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002528
2529 if (!ret)
2530 break;
2531
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002532 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2533
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002534 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002535 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002536 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002537 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002538 else
2539 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002540 } while (retry_count);
2541
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002542 if (!ret && enlarge)
2543 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002544
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002545 return ret;
2546}
2547
Andrew Morton0608f432013-09-24 15:27:41 -07002548unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2549 gfp_t gfp_mask,
2550 unsigned long *total_scanned)
2551{
2552 unsigned long nr_reclaimed = 0;
2553 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2554 unsigned long reclaimed;
2555 int loop = 0;
2556 struct mem_cgroup_tree_per_zone *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002557 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002558 unsigned long nr_scanned;
2559
2560 if (order > 0)
2561 return 0;
2562
2563 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
2564 /*
2565 * This loop can run a while, specially if mem_cgroup's continuously
2566 * keep exceeding their soft limit and putting the system under
2567 * pressure
2568 */
2569 do {
2570 if (next_mz)
2571 mz = next_mz;
2572 else
2573 mz = mem_cgroup_largest_soft_limit_node(mctz);
2574 if (!mz)
2575 break;
2576
2577 nr_scanned = 0;
2578 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
2579 gfp_mask, &nr_scanned);
2580 nr_reclaimed += reclaimed;
2581 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002582 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002583 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002584
2585 /*
2586 * If we failed to reclaim anything from this memory cgroup
2587 * it is time to move on to the next cgroup
2588 */
2589 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002590 if (!reclaimed)
2591 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2592
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002593 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002594 /*
2595 * One school of thought says that we should not add
2596 * back the node to the tree if reclaim returns 0.
2597 * But our reclaim could return 0, simply because due
2598 * to priority we are exposing a smaller subset of
2599 * memory to reclaim from. Consider this as a longer
2600 * term TODO.
2601 */
2602 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002603 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002604 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002605 css_put(&mz->memcg->css);
2606 loop++;
2607 /*
2608 * Could not reclaim anything and there are no more
2609 * mem cgroups to try or we seem to be looping without
2610 * reclaiming anything.
2611 */
2612 if (!nr_reclaimed &&
2613 (next_mz == NULL ||
2614 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2615 break;
2616 } while (!nr_reclaimed);
2617 if (next_mz)
2618 css_put(&next_mz->memcg->css);
2619 return nr_reclaimed;
2620}
2621
Tejun Heoea280e72014-05-16 13:22:48 -04002622/*
2623 * Test whether @memcg has children, dead or alive. Note that this
2624 * function doesn't care whether @memcg has use_hierarchy enabled and
2625 * returns %true if there are child csses according to the cgroup
2626 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2627 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002628static inline bool memcg_has_children(struct mem_cgroup *memcg)
2629{
Tejun Heoea280e72014-05-16 13:22:48 -04002630 bool ret;
2631
Tejun Heoea280e72014-05-16 13:22:48 -04002632 rcu_read_lock();
2633 ret = css_next_child(NULL, &memcg->css);
2634 rcu_read_unlock();
2635 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002636}
2637
2638/*
Michal Hockoc26251f2012-10-26 13:37:28 +02002639 * Reclaims as many pages from the given memcg as possible and moves
2640 * the rest to the parent.
2641 *
2642 * Caller is responsible for holding css reference for memcg.
2643 */
2644static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2645{
2646 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002647
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002648 /* we call try-to-free pages for make this cgroup empty */
2649 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002650 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002651 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002652 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002653
Michal Hockoc26251f2012-10-26 13:37:28 +02002654 if (signal_pending(current))
2655 return -EINTR;
2656
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002657 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2658 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002659 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002660 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002661 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002662 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002663 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002664
2665 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002666
2667 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002668}
2669
Tejun Heo6770c642014-05-13 12:16:21 -04002670static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2671 char *buf, size_t nbytes,
2672 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002673{
Tejun Heo6770c642014-05-13 12:16:21 -04002674 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002675
Michal Hockod8423012012-10-26 13:37:29 +02002676 if (mem_cgroup_is_root(memcg))
2677 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002678 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002679}
2680
Tejun Heo182446d2013-08-08 20:11:24 -04002681static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2682 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002683{
Tejun Heo182446d2013-08-08 20:11:24 -04002684 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002685}
2686
Tejun Heo182446d2013-08-08 20:11:24 -04002687static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2688 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002689{
2690 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002691 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002692 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002693
Glauber Costa567fb432012-07-31 16:43:07 -07002694 if (memcg->use_hierarchy == val)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002695 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07002696
Balbir Singh18f59ea2009-01-07 18:08:07 -08002697 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002698 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002699 * in the child subtrees. If it is unset, then the change can
2700 * occur, provided the current cgroup has no children.
2701 *
2702 * For the root cgroup, parent_mem is NULL, we allow value to be
2703 * set if there are no children.
2704 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002705 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002706 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002707 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002708 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002709 else
2710 retval = -EBUSY;
2711 } else
2712 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002713
Balbir Singh18f59ea2009-01-07 18:08:07 -08002714 return retval;
2715}
2716
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002717static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
Johannes Weinerce00a962014-09-05 08:43:57 -04002718{
2719 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002720 int i;
Johannes Weinerce00a962014-09-05 08:43:57 -04002721
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002722 memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
Johannes Weinerce00a962014-09-05 08:43:57 -04002723
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002724 for_each_mem_cgroup_tree(iter, memcg) {
2725 for (i = 0; i < MEMCG_NR_STAT; i++)
2726 stat[i] += mem_cgroup_read_stat(iter, i);
2727 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002728}
2729
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002730static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
Johannes Weiner587d9f72016-01-20 15:03:19 -08002731{
2732 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002733 int i;
Johannes Weiner587d9f72016-01-20 15:03:19 -08002734
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002735 memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
Johannes Weiner587d9f72016-01-20 15:03:19 -08002736
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002737 for_each_mem_cgroup_tree(iter, memcg) {
2738 for (i = 0; i < MEMCG_NR_EVENTS; i++)
2739 events[i] += mem_cgroup_read_events(iter, i);
2740 }
Johannes Weiner587d9f72016-01-20 15:03:19 -08002741}
2742
Andrew Morton6f646152015-11-06 16:28:58 -08002743static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04002744{
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002745 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002746
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002747 if (mem_cgroup_is_root(memcg)) {
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002748 struct mem_cgroup *iter;
2749
2750 for_each_mem_cgroup_tree(iter, memcg) {
2751 val += mem_cgroup_read_stat(iter,
2752 MEM_CGROUP_STAT_CACHE);
2753 val += mem_cgroup_read_stat(iter,
2754 MEM_CGROUP_STAT_RSS);
2755 if (swap)
2756 val += mem_cgroup_read_stat(iter,
2757 MEM_CGROUP_STAT_SWAP);
2758 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002759 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002760 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002761 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002762 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002763 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002764 }
Michal Hockoc12176d2015-11-05 18:50:29 -08002765 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002766}
2767
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002768enum {
2769 RES_USAGE,
2770 RES_LIMIT,
2771 RES_MAX_USAGE,
2772 RES_FAILCNT,
2773 RES_SOFT_LIMIT,
2774};
Johannes Weinerce00a962014-09-05 08:43:57 -04002775
Tejun Heo791badb2013-12-05 12:28:02 -05002776static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002777 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002778{
Tejun Heo182446d2013-08-08 20:11:24 -04002779 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002780 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002781
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002782 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002783 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002784 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002785 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002786 case _MEMSWAP:
2787 counter = &memcg->memsw;
2788 break;
2789 case _KMEM:
2790 counter = &memcg->kmem;
2791 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002792 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08002793 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002794 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002795 default:
2796 BUG();
2797 }
2798
2799 switch (MEMFILE_ATTR(cft->private)) {
2800 case RES_USAGE:
2801 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08002802 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002803 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08002804 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002805 return (u64)page_counter_read(counter) * PAGE_SIZE;
2806 case RES_LIMIT:
2807 return (u64)counter->limit * PAGE_SIZE;
2808 case RES_MAX_USAGE:
2809 return (u64)counter->watermark * PAGE_SIZE;
2810 case RES_FAILCNT:
2811 return counter->failcnt;
2812 case RES_SOFT_LIMIT:
2813 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002814 default:
2815 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002816 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002817}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002818
Johannes Weiner127424c2016-01-20 15:02:32 -08002819#ifndef CONFIG_SLOB
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002820static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002821{
Vladimir Davydovd6441632014-01-23 15:53:09 -08002822 int memcg_id;
2823
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002824 if (cgroup_memory_nokmem)
2825 return 0;
2826
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002827 BUG_ON(memcg->kmemcg_id >= 0);
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002828 BUG_ON(memcg->kmem_state);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002829
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002830 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002831 if (memcg_id < 0)
2832 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002833
Johannes Weineref129472016-01-14 15:21:34 -08002834 static_branch_inc(&memcg_kmem_enabled_key);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002835 /*
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002836 * A memory cgroup is considered kmem-online as soon as it gets
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002837 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002838 * guarantee no one starts accounting before all call sites are
2839 * patched.
2840 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002841 memcg->kmemcg_id = memcg_id;
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002842 memcg->kmem_state = KMEM_ONLINE;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002843
2844 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002845}
2846
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002847static void memcg_offline_kmem(struct mem_cgroup *memcg)
2848{
2849 struct cgroup_subsys_state *css;
2850 struct mem_cgroup *parent, *child;
2851 int kmemcg_id;
2852
2853 if (memcg->kmem_state != KMEM_ONLINE)
2854 return;
2855 /*
2856 * Clear the online state before clearing memcg_caches array
2857 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2858 * guarantees that no cache will be created for this cgroup
2859 * after we are done (see memcg_create_kmem_cache()).
2860 */
2861 memcg->kmem_state = KMEM_ALLOCATED;
2862
2863 memcg_deactivate_kmem_caches(memcg);
2864
2865 kmemcg_id = memcg->kmemcg_id;
2866 BUG_ON(kmemcg_id < 0);
2867
2868 parent = parent_mem_cgroup(memcg);
2869 if (!parent)
2870 parent = root_mem_cgroup;
2871
2872 /*
2873 * Change kmemcg_id of this cgroup and all its descendants to the
2874 * parent's id, and then move all entries from this cgroup's list_lrus
2875 * to ones of the parent. After we have finished, all list_lrus
2876 * corresponding to this cgroup are guaranteed to remain empty. The
2877 * ordering is imposed by list_lru_node->lock taken by
2878 * memcg_drain_all_list_lrus().
2879 */
2880 css_for_each_descendant_pre(css, &memcg->css) {
2881 child = mem_cgroup_from_css(css);
2882 BUG_ON(child->kmemcg_id != kmemcg_id);
2883 child->kmemcg_id = parent->kmemcg_id;
2884 if (!memcg->use_hierarchy)
2885 break;
2886 }
2887 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2888
2889 memcg_free_cache_id(kmemcg_id);
2890}
2891
2892static void memcg_free_kmem(struct mem_cgroup *memcg)
2893{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002894 /* css_alloc() failed, offlining didn't happen */
2895 if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2896 memcg_offline_kmem(memcg);
2897
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002898 if (memcg->kmem_state == KMEM_ALLOCATED) {
2899 memcg_destroy_kmem_caches(memcg);
2900 static_branch_dec(&memcg_kmem_enabled_key);
2901 WARN_ON(page_counter_read(&memcg->kmem));
2902 }
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002903}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002904#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002905static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08002906{
2907 return 0;
2908}
2909static void memcg_offline_kmem(struct mem_cgroup *memcg)
2910{
2911}
2912static void memcg_free_kmem(struct mem_cgroup *memcg)
2913{
2914}
2915#endif /* !CONFIG_SLOB */
2916
Johannes Weiner127424c2016-01-20 15:02:32 -08002917static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2918 unsigned long limit)
2919{
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002920 int ret;
Johannes Weiner127424c2016-01-20 15:02:32 -08002921
2922 mutex_lock(&memcg_limit_mutex);
Johannes Weiner127424c2016-01-20 15:02:32 -08002923 ret = page_counter_limit(&memcg->kmem, limit);
Johannes Weiner127424c2016-01-20 15:02:32 -08002924 mutex_unlock(&memcg_limit_mutex);
2925 return ret;
2926}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002927
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002928static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2929{
2930 int ret;
2931
2932 mutex_lock(&memcg_limit_mutex);
2933
Johannes Weiner0db15292016-01-20 15:02:50 -08002934 ret = page_counter_limit(&memcg->tcpmem, limit);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002935 if (ret)
2936 goto out;
2937
Johannes Weiner0db15292016-01-20 15:02:50 -08002938 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002939 /*
2940 * The active flag needs to be written after the static_key
2941 * update. This is what guarantees that the socket activation
2942 * function is the last one to run. See sock_update_memcg() for
2943 * details, and note that we don't mark any socket as belonging
2944 * to this memcg until that flag is up.
2945 *
2946 * We need to do this, because static_keys will span multiple
2947 * sites, but we can't control their order. If we mark a socket
2948 * as accounted, but the accounting functions are not patched in
2949 * yet, we'll lose accounting.
2950 *
2951 * We never race with the readers in sock_update_memcg(),
2952 * because when this value change, the code to process it is not
2953 * patched in yet.
2954 */
2955 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08002956 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002957 }
2958out:
2959 mutex_unlock(&memcg_limit_mutex);
2960 return ret;
2961}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002962
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002963/*
2964 * The user of this function is...
2965 * RES_LIMIT.
2966 */
Tejun Heo451af502014-05-13 12:16:21 -04002967static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2968 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002969{
Tejun Heo451af502014-05-13 12:16:21 -04002970 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002971 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002972 int ret;
2973
Tejun Heo451af502014-05-13 12:16:21 -04002974 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08002975 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002976 if (ret)
2977 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07002978
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002979 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002980 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07002981 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2982 ret = -EINVAL;
2983 break;
2984 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002985 switch (MEMFILE_TYPE(of_cft(of)->private)) {
2986 case _MEM:
2987 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002988 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002989 case _MEMSWAP:
2990 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
2991 break;
2992 case _KMEM:
2993 ret = memcg_update_kmem_limit(memcg, nr_pages);
2994 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002995 case _TCP:
2996 ret = memcg_update_tcp_limit(memcg, nr_pages);
2997 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002998 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002999 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003000 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003001 memcg->soft_limit = nr_pages;
3002 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003003 break;
3004 }
Tejun Heo451af502014-05-13 12:16:21 -04003005 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003006}
3007
Tejun Heo6770c642014-05-13 12:16:21 -04003008static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3009 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003010{
Tejun Heo6770c642014-05-13 12:16:21 -04003011 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003012 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003013
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003014 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3015 case _MEM:
3016 counter = &memcg->memory;
3017 break;
3018 case _MEMSWAP:
3019 counter = &memcg->memsw;
3020 break;
3021 case _KMEM:
3022 counter = &memcg->kmem;
3023 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003024 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003025 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003026 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003027 default:
3028 BUG();
3029 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003030
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003031 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003032 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003033 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003034 break;
3035 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003036 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003037 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003038 default:
3039 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003040 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003041
Tejun Heo6770c642014-05-13 12:16:21 -04003042 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003043}
3044
Tejun Heo182446d2013-08-08 20:11:24 -04003045static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003046 struct cftype *cft)
3047{
Tejun Heo182446d2013-08-08 20:11:24 -04003048 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003049}
3050
Daisuke Nishimura02491442010-03-10 15:22:17 -08003051#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003052static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003053 struct cftype *cft, u64 val)
3054{
Tejun Heo182446d2013-08-08 20:11:24 -04003055 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003056
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003057 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003058 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003059
Glauber Costaee5e8472013-02-22 16:34:50 -08003060 /*
3061 * No kind of locking is needed in here, because ->can_attach() will
3062 * check this value once in the beginning of the process, and then carry
3063 * on with stale data. This means that changes to this value will only
3064 * affect task migrations starting after the change.
3065 */
3066 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003067 return 0;
3068}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003069#else
Tejun Heo182446d2013-08-08 20:11:24 -04003070static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003071 struct cftype *cft, u64 val)
3072{
3073 return -ENOSYS;
3074}
3075#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003076
Ying Han406eb0c2011-05-26 16:25:37 -07003077#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003078static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003079{
Greg Thelen25485de2013-11-12 15:07:40 -08003080 struct numa_stat {
3081 const char *name;
3082 unsigned int lru_mask;
3083 };
3084
3085 static const struct numa_stat stats[] = {
3086 { "total", LRU_ALL },
3087 { "file", LRU_ALL_FILE },
3088 { "anon", LRU_ALL_ANON },
3089 { "unevictable", BIT(LRU_UNEVICTABLE) },
3090 };
3091 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003092 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003093 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003094 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003095
Greg Thelen25485de2013-11-12 15:07:40 -08003096 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3097 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3098 seq_printf(m, "%s=%lu", stat->name, nr);
3099 for_each_node_state(nid, N_MEMORY) {
3100 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3101 stat->lru_mask);
3102 seq_printf(m, " N%d=%lu", nid, nr);
3103 }
3104 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003105 }
Ying Han406eb0c2011-05-26 16:25:37 -07003106
Ying Han071aee12013-11-12 15:07:41 -08003107 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3108 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003109
Ying Han071aee12013-11-12 15:07:41 -08003110 nr = 0;
3111 for_each_mem_cgroup_tree(iter, memcg)
3112 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3113 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3114 for_each_node_state(nid, N_MEMORY) {
3115 nr = 0;
3116 for_each_mem_cgroup_tree(iter, memcg)
3117 nr += mem_cgroup_node_nr_lru_pages(
3118 iter, nid, stat->lru_mask);
3119 seq_printf(m, " N%d=%lu", nid, nr);
3120 }
3121 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003122 }
Ying Han406eb0c2011-05-26 16:25:37 -07003123
Ying Han406eb0c2011-05-26 16:25:37 -07003124 return 0;
3125}
3126#endif /* CONFIG_NUMA */
3127
Tejun Heo2da8ca82013-12-05 12:28:04 -05003128static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003129{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003130 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003131 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003132 struct mem_cgroup *mi;
3133 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003134
Greg Thelen0ca44b12015-02-11 15:25:58 -08003135 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3136 MEM_CGROUP_STAT_NSTATS);
3137 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3138 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003139 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3140
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003141 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Johannes Weiner7941d212016-01-14 15:21:23 -08003142 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003143 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07003144 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003145 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003146 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003147
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003148 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3149 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3150 mem_cgroup_read_events(memcg, i));
3151
3152 for (i = 0; i < NR_LRU_LISTS; i++)
3153 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3154 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3155
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003156 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003157 memory = memsw = PAGE_COUNTER_MAX;
3158 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3159 memory = min(memory, mi->memory.limit);
3160 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003161 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003162 seq_printf(m, "hierarchical_memory_limit %llu\n",
3163 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08003164 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003165 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3166 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003167
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003168 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003169 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003170
Johannes Weiner7941d212016-01-14 15:21:23 -08003171 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003172 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003173 for_each_mem_cgroup_tree(mi, memcg)
3174 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
Greg Thelen484ebb32015-10-01 15:37:05 -07003175 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003176 }
3177
3178 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3179 unsigned long long val = 0;
3180
3181 for_each_mem_cgroup_tree(mi, memcg)
3182 val += mem_cgroup_read_events(mi, i);
3183 seq_printf(m, "total_%s %llu\n",
3184 mem_cgroup_events_names[i], val);
3185 }
3186
3187 for (i = 0; i < NR_LRU_LISTS; i++) {
3188 unsigned long long val = 0;
3189
3190 for_each_mem_cgroup_tree(mi, memcg)
3191 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3192 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003193 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003194
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003195#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003196 {
3197 int nid, zid;
3198 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003199 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003200 unsigned long recent_rotated[2] = {0, 0};
3201 unsigned long recent_scanned[2] = {0, 0};
3202
3203 for_each_online_node(nid)
3204 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Jianyu Zhane2318752014-06-06 14:38:20 -07003205 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
Hugh Dickins89abfab2012-05-29 15:06:53 -07003206 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003207
Hugh Dickins89abfab2012-05-29 15:06:53 -07003208 recent_rotated[0] += rstat->recent_rotated[0];
3209 recent_rotated[1] += rstat->recent_rotated[1];
3210 recent_scanned[0] += rstat->recent_scanned[0];
3211 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003212 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003213 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3214 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3215 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3216 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003217 }
3218#endif
3219
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003220 return 0;
3221}
3222
Tejun Heo182446d2013-08-08 20:11:24 -04003223static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3224 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003225{
Tejun Heo182446d2013-08-08 20:11:24 -04003226 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003227
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003228 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003229}
3230
Tejun Heo182446d2013-08-08 20:11:24 -04003231static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3232 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003233{
Tejun Heo182446d2013-08-08 20:11:24 -04003234 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003235
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003236 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003237 return -EINVAL;
3238
Linus Torvalds14208b02014-06-09 15:03:33 -07003239 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003240 memcg->swappiness = val;
3241 else
3242 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003243
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003244 return 0;
3245}
3246
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003247static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3248{
3249 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003250 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003251 int i;
3252
3253 rcu_read_lock();
3254 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003255 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003256 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003257 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003258
3259 if (!t)
3260 goto unlock;
3261
Johannes Weinerce00a962014-09-05 08:43:57 -04003262 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003263
3264 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003265 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003266 * If it's not true, a threshold was crossed after last
3267 * call of __mem_cgroup_threshold().
3268 */
Phil Carmody5407a562010-05-26 14:42:42 -07003269 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003270
3271 /*
3272 * Iterate backward over array of thresholds starting from
3273 * current_threshold and check if a threshold is crossed.
3274 * If none of thresholds below usage is crossed, we read
3275 * only one element of the array here.
3276 */
3277 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3278 eventfd_signal(t->entries[i].eventfd, 1);
3279
3280 /* i = current_threshold + 1 */
3281 i++;
3282
3283 /*
3284 * Iterate forward over array of thresholds starting from
3285 * current_threshold+1 and check if a threshold is crossed.
3286 * If none of thresholds above usage is crossed, we read
3287 * only one element of the array here.
3288 */
3289 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3290 eventfd_signal(t->entries[i].eventfd, 1);
3291
3292 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003293 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003294unlock:
3295 rcu_read_unlock();
3296}
3297
3298static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3299{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003300 while (memcg) {
3301 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08003302 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003303 __mem_cgroup_threshold(memcg, true);
3304
3305 memcg = parent_mem_cgroup(memcg);
3306 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003307}
3308
3309static int compare_thresholds(const void *a, const void *b)
3310{
3311 const struct mem_cgroup_threshold *_a = a;
3312 const struct mem_cgroup_threshold *_b = b;
3313
Greg Thelen2bff24a2013-09-11 14:23:08 -07003314 if (_a->threshold > _b->threshold)
3315 return 1;
3316
3317 if (_a->threshold < _b->threshold)
3318 return -1;
3319
3320 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003321}
3322
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003323static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003324{
3325 struct mem_cgroup_eventfd_list *ev;
3326
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003327 spin_lock(&memcg_oom_lock);
3328
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003329 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003330 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003331
3332 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003333 return 0;
3334}
3335
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003336static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003337{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003338 struct mem_cgroup *iter;
3339
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003340 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003341 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003342}
3343
Tejun Heo59b6f872013-11-22 18:20:43 -05003344static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003345 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003346{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003347 struct mem_cgroup_thresholds *thresholds;
3348 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003349 unsigned long threshold;
3350 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003351 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003352
Johannes Weiner650c5e52015-02-11 15:26:03 -08003353 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003354 if (ret)
3355 return ret;
3356
3357 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003358
Johannes Weiner05b84302014-08-06 16:05:59 -07003359 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003360 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003361 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003362 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003363 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003364 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003365 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003366 BUG();
3367
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003368 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003369 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003370 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3371
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003372 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003373
3374 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003375 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003376 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003377 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003378 ret = -ENOMEM;
3379 goto unlock;
3380 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003381 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003382
3383 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003384 if (thresholds->primary) {
3385 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003386 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003387 }
3388
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003389 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003390 new->entries[size - 1].eventfd = eventfd;
3391 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003392
3393 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003394 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003395 compare_thresholds, NULL);
3396
3397 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003398 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003399 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003400 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003401 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003402 * new->current_threshold will not be used until
3403 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003404 * it here.
3405 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003406 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003407 } else
3408 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003409 }
3410
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003411 /* Free old spare buffer and save old primary buffer as spare */
3412 kfree(thresholds->spare);
3413 thresholds->spare = thresholds->primary;
3414
3415 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003416
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003417 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003418 synchronize_rcu();
3419
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003420unlock:
3421 mutex_unlock(&memcg->thresholds_lock);
3422
3423 return ret;
3424}
3425
Tejun Heo59b6f872013-11-22 18:20:43 -05003426static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003427 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003428{
Tejun Heo59b6f872013-11-22 18:20:43 -05003429 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003430}
3431
Tejun Heo59b6f872013-11-22 18:20:43 -05003432static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003433 struct eventfd_ctx *eventfd, const char *args)
3434{
Tejun Heo59b6f872013-11-22 18:20:43 -05003435 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003436}
3437
Tejun Heo59b6f872013-11-22 18:20:43 -05003438static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003439 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003440{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003441 struct mem_cgroup_thresholds *thresholds;
3442 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003443 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003444 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003445
3446 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003447
3448 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003449 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003450 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003451 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003452 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003453 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003454 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003455 BUG();
3456
Anton Vorontsov371528c2012-02-24 05:14:46 +04003457 if (!thresholds->primary)
3458 goto unlock;
3459
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003460 /* Check if a threshold crossed before removing */
3461 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3462
3463 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003464 size = 0;
3465 for (i = 0; i < thresholds->primary->size; i++) {
3466 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003467 size++;
3468 }
3469
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003470 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003471
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003472 /* Set thresholds array to NULL if we don't have thresholds */
3473 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003474 kfree(new);
3475 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003476 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003477 }
3478
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003479 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003480
3481 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003482 new->current_threshold = -1;
3483 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3484 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003485 continue;
3486
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003487 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003488 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003489 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003490 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003491 * until rcu_assign_pointer(), so it's safe to increment
3492 * it here.
3493 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003494 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003495 }
3496 j++;
3497 }
3498
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003499swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003500 /* Swap primary and spare array */
3501 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003502
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003503 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003504
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003505 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003506 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08003507
3508 /* If all events are unregistered, free the spare array */
3509 if (!new) {
3510 kfree(thresholds->spare);
3511 thresholds->spare = NULL;
3512 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04003513unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003514 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003515}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003516
Tejun Heo59b6f872013-11-22 18:20:43 -05003517static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003518 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003519{
Tejun Heo59b6f872013-11-22 18:20:43 -05003520 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003521}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003522
Tejun Heo59b6f872013-11-22 18:20:43 -05003523static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003524 struct eventfd_ctx *eventfd)
3525{
Tejun Heo59b6f872013-11-22 18:20:43 -05003526 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003527}
3528
Tejun Heo59b6f872013-11-22 18:20:43 -05003529static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003530 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003531{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003532 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003533
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003534 event = kmalloc(sizeof(*event), GFP_KERNEL);
3535 if (!event)
3536 return -ENOMEM;
3537
Michal Hocko1af8efe2011-07-26 16:08:24 -07003538 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003539
3540 event->eventfd = eventfd;
3541 list_add(&event->list, &memcg->oom_notify);
3542
3543 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07003544 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003545 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003546 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003547
3548 return 0;
3549}
3550
Tejun Heo59b6f872013-11-22 18:20:43 -05003551static void mem_cgroup_oom_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{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003554 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003555
Michal Hocko1af8efe2011-07-26 16:08:24 -07003556 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003557
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003558 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003559 if (ev->eventfd == eventfd) {
3560 list_del(&ev->list);
3561 kfree(ev);
3562 }
3563 }
3564
Michal Hocko1af8efe2011-07-26 16:08:24 -07003565 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003566}
3567
Tejun Heo2da8ca82013-12-05 12:28:04 -05003568static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003569{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003570 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003571
Tejun Heo791badb2013-12-05 12:28:02 -05003572 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07003573 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003574 return 0;
3575}
3576
Tejun Heo182446d2013-08-08 20:11:24 -04003577static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003578 struct cftype *cft, u64 val)
3579{
Tejun Heo182446d2013-08-08 20:11:24 -04003580 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003581
3582 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003583 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003584 return -EINVAL;
3585
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003586 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003587 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003588 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003589
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003590 return 0;
3591}
3592
Tejun Heo52ebea72015-05-22 17:13:37 -04003593#ifdef CONFIG_CGROUP_WRITEBACK
3594
3595struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3596{
3597 return &memcg->cgwb_list;
3598}
3599
Tejun Heo841710a2015-05-22 18:23:33 -04003600static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3601{
3602 return wb_domain_init(&memcg->cgwb_domain, gfp);
3603}
3604
3605static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3606{
3607 wb_domain_exit(&memcg->cgwb_domain);
3608}
3609
Tejun Heo2529bb32015-05-22 18:23:34 -04003610static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3611{
3612 wb_domain_size_changed(&memcg->cgwb_domain);
3613}
3614
Tejun Heo841710a2015-05-22 18:23:33 -04003615struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3616{
3617 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3618
3619 if (!memcg->css.parent)
3620 return NULL;
3621
3622 return &memcg->cgwb_domain;
3623}
3624
Tejun Heoc2aa7232015-05-22 18:23:35 -04003625/**
3626 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3627 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003628 * @pfilepages: out parameter for number of file pages
3629 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04003630 * @pdirty: out parameter for number of dirty pages
3631 * @pwriteback: out parameter for number of pages under writeback
3632 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003633 * Determine the numbers of file, headroom, dirty, and writeback pages in
3634 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3635 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003636 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003637 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3638 * headroom is calculated as the lowest headroom of itself and the
3639 * ancestors. Note that this doesn't consider the actual amount of
3640 * available memory in the system. The caller should further cap
3641 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003642 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003643void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3644 unsigned long *pheadroom, unsigned long *pdirty,
3645 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04003646{
3647 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3648 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003649
3650 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3651
3652 /* this should eventually include NR_UNSTABLE_NFS */
3653 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003654 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3655 (1 << LRU_ACTIVE_FILE));
3656 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003657
Tejun Heoc2aa7232015-05-22 18:23:35 -04003658 while ((parent = parent_mem_cgroup(memcg))) {
3659 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3660 unsigned long used = page_counter_read(&memcg->memory);
3661
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003662 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04003663 memcg = parent;
3664 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04003665}
3666
Tejun Heo841710a2015-05-22 18:23:33 -04003667#else /* CONFIG_CGROUP_WRITEBACK */
3668
3669static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3670{
3671 return 0;
3672}
3673
3674static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3675{
3676}
3677
Tejun Heo2529bb32015-05-22 18:23:34 -04003678static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3679{
3680}
3681
Tejun Heo52ebea72015-05-22 17:13:37 -04003682#endif /* CONFIG_CGROUP_WRITEBACK */
3683
Tejun Heo79bd9812013-11-22 18:20:42 -05003684/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003685 * DO NOT USE IN NEW FILES.
3686 *
3687 * "cgroup.event_control" implementation.
3688 *
3689 * This is way over-engineered. It tries to support fully configurable
3690 * events for each user. Such level of flexibility is completely
3691 * unnecessary especially in the light of the planned unified hierarchy.
3692 *
3693 * Please deprecate this and replace with something simpler if at all
3694 * possible.
3695 */
3696
3697/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003698 * Unregister event and free resources.
3699 *
3700 * Gets called from workqueue.
3701 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003702static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003703{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003704 struct mem_cgroup_event *event =
3705 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003706 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003707
3708 remove_wait_queue(event->wqh, &event->wait);
3709
Tejun Heo59b6f872013-11-22 18:20:43 -05003710 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003711
3712 /* Notify userspace the event is going away. */
3713 eventfd_signal(event->eventfd, 1);
3714
3715 eventfd_ctx_put(event->eventfd);
3716 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003717 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003718}
3719
3720/*
3721 * Gets called on POLLHUP on eventfd when user closes it.
3722 *
3723 * Called with wqh->lock held and interrupts disabled.
3724 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003725static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3726 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003727{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003728 struct mem_cgroup_event *event =
3729 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003730 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003731 unsigned long flags = (unsigned long)key;
3732
3733 if (flags & POLLHUP) {
3734 /*
3735 * If the event has been detached at cgroup removal, we
3736 * can simply return knowing the other side will cleanup
3737 * for us.
3738 *
3739 * We can't race against event freeing since the other
3740 * side will require wqh->lock via remove_wait_queue(),
3741 * which we hold.
3742 */
Tejun Heofba94802013-11-22 18:20:43 -05003743 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003744 if (!list_empty(&event->list)) {
3745 list_del_init(&event->list);
3746 /*
3747 * We are in atomic context, but cgroup_event_remove()
3748 * may sleep, so we have to call it in workqueue.
3749 */
3750 schedule_work(&event->remove);
3751 }
Tejun Heofba94802013-11-22 18:20:43 -05003752 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003753 }
3754
3755 return 0;
3756}
3757
Tejun Heo3bc942f2013-11-22 18:20:44 -05003758static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003759 wait_queue_head_t *wqh, poll_table *pt)
3760{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003761 struct mem_cgroup_event *event =
3762 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003763
3764 event->wqh = wqh;
3765 add_wait_queue(wqh, &event->wait);
3766}
3767
3768/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003769 * DO NOT USE IN NEW FILES.
3770 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003771 * Parse input and register new cgroup event handler.
3772 *
3773 * Input must be in format '<event_fd> <control_fd> <args>'.
3774 * Interpretation of args is defined by control file implementation.
3775 */
Tejun Heo451af502014-05-13 12:16:21 -04003776static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3777 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003778{
Tejun Heo451af502014-05-13 12:16:21 -04003779 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003780 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003781 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003782 struct cgroup_subsys_state *cfile_css;
3783 unsigned int efd, cfd;
3784 struct fd efile;
3785 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003786 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003787 char *endp;
3788 int ret;
3789
Tejun Heo451af502014-05-13 12:16:21 -04003790 buf = strstrip(buf);
3791
3792 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003793 if (*endp != ' ')
3794 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003795 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003796
Tejun Heo451af502014-05-13 12:16:21 -04003797 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003798 if ((*endp != ' ') && (*endp != '\0'))
3799 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003800 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003801
3802 event = kzalloc(sizeof(*event), GFP_KERNEL);
3803 if (!event)
3804 return -ENOMEM;
3805
Tejun Heo59b6f872013-11-22 18:20:43 -05003806 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003807 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003808 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3809 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3810 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003811
3812 efile = fdget(efd);
3813 if (!efile.file) {
3814 ret = -EBADF;
3815 goto out_kfree;
3816 }
3817
3818 event->eventfd = eventfd_ctx_fileget(efile.file);
3819 if (IS_ERR(event->eventfd)) {
3820 ret = PTR_ERR(event->eventfd);
3821 goto out_put_efile;
3822 }
3823
3824 cfile = fdget(cfd);
3825 if (!cfile.file) {
3826 ret = -EBADF;
3827 goto out_put_eventfd;
3828 }
3829
3830 /* the process need read permission on control file */
3831 /* AV: shouldn't we check that it's been opened for read instead? */
3832 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3833 if (ret < 0)
3834 goto out_put_cfile;
3835
Tejun Heo79bd9812013-11-22 18:20:42 -05003836 /*
Tejun Heofba94802013-11-22 18:20:43 -05003837 * Determine the event callbacks and set them in @event. This used
3838 * to be done via struct cftype but cgroup core no longer knows
3839 * about these events. The following is crude but the whole thing
3840 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003841 *
3842 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003843 */
Al Virob5830432014-10-31 01:22:04 -04003844 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003845
3846 if (!strcmp(name, "memory.usage_in_bytes")) {
3847 event->register_event = mem_cgroup_usage_register_event;
3848 event->unregister_event = mem_cgroup_usage_unregister_event;
3849 } else if (!strcmp(name, "memory.oom_control")) {
3850 event->register_event = mem_cgroup_oom_register_event;
3851 event->unregister_event = mem_cgroup_oom_unregister_event;
3852 } else if (!strcmp(name, "memory.pressure_level")) {
3853 event->register_event = vmpressure_register_event;
3854 event->unregister_event = vmpressure_unregister_event;
3855 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003856 event->register_event = memsw_cgroup_usage_register_event;
3857 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003858 } else {
3859 ret = -EINVAL;
3860 goto out_put_cfile;
3861 }
3862
3863 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003864 * Verify @cfile should belong to @css. Also, remaining events are
3865 * automatically removed on cgroup destruction but the removal is
3866 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003867 */
Al Virob5830432014-10-31 01:22:04 -04003868 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003869 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003870 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003871 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003872 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003873 if (cfile_css != css) {
3874 css_put(cfile_css);
3875 goto out_put_cfile;
3876 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003877
Tejun Heo451af502014-05-13 12:16:21 -04003878 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003879 if (ret)
3880 goto out_put_css;
3881
3882 efile.file->f_op->poll(efile.file, &event->pt);
3883
Tejun Heofba94802013-11-22 18:20:43 -05003884 spin_lock(&memcg->event_list_lock);
3885 list_add(&event->list, &memcg->event_list);
3886 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003887
3888 fdput(cfile);
3889 fdput(efile);
3890
Tejun Heo451af502014-05-13 12:16:21 -04003891 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05003892
3893out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05003894 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003895out_put_cfile:
3896 fdput(cfile);
3897out_put_eventfd:
3898 eventfd_ctx_put(event->eventfd);
3899out_put_efile:
3900 fdput(efile);
3901out_kfree:
3902 kfree(event);
3903
3904 return ret;
3905}
3906
Johannes Weiner241994e2015-02-11 15:26:06 -08003907static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003908 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003909 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003910 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003911 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003912 },
3913 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003914 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003915 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04003916 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003917 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003918 },
3919 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003920 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003921 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003922 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003923 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003924 },
3925 {
Balbir Singh296c81d2009-09-23 15:56:36 -07003926 .name = "soft_limit_in_bytes",
3927 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003928 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003929 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07003930 },
3931 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003932 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003933 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04003934 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003935 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003936 },
Balbir Singh8697d332008-02-07 00:13:59 -08003937 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003938 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003939 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003940 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003941 {
3942 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04003943 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003944 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08003945 {
3946 .name = "use_hierarchy",
3947 .write_u64 = mem_cgroup_hierarchy_write,
3948 .read_u64 = mem_cgroup_hierarchy_read,
3949 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003950 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05003951 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04003952 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04003953 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05003954 },
3955 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003956 .name = "swappiness",
3957 .read_u64 = mem_cgroup_swappiness_read,
3958 .write_u64 = mem_cgroup_swappiness_write,
3959 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003960 {
3961 .name = "move_charge_at_immigrate",
3962 .read_u64 = mem_cgroup_move_charge_read,
3963 .write_u64 = mem_cgroup_move_charge_write,
3964 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003965 {
3966 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003967 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003968 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003969 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3970 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003971 {
3972 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003973 },
Ying Han406eb0c2011-05-26 16:25:37 -07003974#ifdef CONFIG_NUMA
3975 {
3976 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003977 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07003978 },
3979#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08003980 {
3981 .name = "kmem.limit_in_bytes",
3982 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003983 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003984 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08003985 },
3986 {
3987 .name = "kmem.usage_in_bytes",
3988 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003989 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08003990 },
3991 {
3992 .name = "kmem.failcnt",
3993 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04003994 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003995 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08003996 },
3997 {
3998 .name = "kmem.max_usage_in_bytes",
3999 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004000 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004001 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004002 },
Glauber Costa749c5412012-12-18 14:23:01 -08004003#ifdef CONFIG_SLABINFO
4004 {
4005 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004006 .seq_start = slab_start,
4007 .seq_next = slab_next,
4008 .seq_stop = slab_stop,
4009 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004010 },
4011#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004012 {
4013 .name = "kmem.tcp.limit_in_bytes",
4014 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4015 .write = mem_cgroup_write,
4016 .read_u64 = mem_cgroup_read_u64,
4017 },
4018 {
4019 .name = "kmem.tcp.usage_in_bytes",
4020 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4021 .read_u64 = mem_cgroup_read_u64,
4022 },
4023 {
4024 .name = "kmem.tcp.failcnt",
4025 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4026 .write = mem_cgroup_reset,
4027 .read_u64 = mem_cgroup_read_u64,
4028 },
4029 {
4030 .name = "kmem.tcp.max_usage_in_bytes",
4031 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4032 .write = mem_cgroup_reset,
4033 .read_u64 = mem_cgroup_read_u64,
4034 },
Tejun Heo6bc10342012-04-01 12:09:55 -07004035 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004036};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004037
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004038static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004039{
4040 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004041 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004042 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004043 /*
4044 * This routine is called against possible nodes.
4045 * But it's BUG to call kmalloc() against offline node.
4046 *
4047 * TODO: this routine can waste much memory for nodes which will
4048 * never be onlined. It's better to use memory hotplug callback
4049 * function.
4050 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004051 if (!node_state(node, N_NORMAL_MEMORY))
4052 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004053 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004054 if (!pn)
4055 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004056
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004057 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4058 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004059 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07004060 mz->usage_in_excess = 0;
4061 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004062 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004063 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004064 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004065 return 0;
4066}
4067
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004068static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004069{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004070 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004071}
4072
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004073static void mem_cgroup_free(struct mem_cgroup *memcg)
4074{
4075 int node;
4076
4077 memcg_wb_domain_exit(memcg);
4078 for_each_node(node)
4079 free_mem_cgroup_per_zone_info(memcg, node);
4080 free_percpu(memcg->stat);
4081 kfree(memcg);
4082}
4083
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004084static struct mem_cgroup *mem_cgroup_alloc(void)
4085{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004086 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004087 size_t size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004088 int node;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004089
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004090 size = sizeof(struct mem_cgroup);
4091 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004092
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004093 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004094 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004095 return NULL;
4096
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004097 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4098 if (!memcg->stat)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004099 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004100
Bob Liu3ed28fa2012-01-12 17:19:04 -08004101 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004102 if (alloc_mem_cgroup_per_zone_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004103 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004104
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004105 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4106 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004107
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004108 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08004109 memcg->last_scanned_node = MAX_NUMNODES;
4110 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004111 mutex_init(&memcg->thresholds_lock);
4112 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004113 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004114 INIT_LIST_HEAD(&memcg->event_list);
4115 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004116 memcg->socket_pressure = jiffies;
Johannes Weiner127424c2016-01-20 15:02:32 -08004117#ifndef CONFIG_SLOB
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004118 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004119#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004120#ifdef CONFIG_CGROUP_WRITEBACK
4121 INIT_LIST_HEAD(&memcg->cgwb_list);
4122#endif
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004123 return memcg;
4124fail:
4125 mem_cgroup_free(memcg);
4126 return NULL;
Glauber Costad142e3e2013-02-22 16:34:52 -08004127}
4128
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004129static struct cgroup_subsys_state * __ref
4130mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004131{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004132 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4133 struct mem_cgroup *memcg;
4134 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08004135
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004136 memcg = mem_cgroup_alloc();
4137 if (!memcg)
4138 return ERR_PTR(error);
Li Zefan4219b2d2013-09-23 16:56:29 +08004139
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004140 memcg->high = PAGE_COUNTER_MAX;
4141 memcg->soft_limit = PAGE_COUNTER_MAX;
4142 if (parent) {
4143 memcg->swappiness = mem_cgroup_swappiness(parent);
4144 memcg->oom_kill_disable = parent->oom_kill_disable;
4145 }
4146 if (parent && parent->use_hierarchy) {
4147 memcg->use_hierarchy = true;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004148 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004149 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004150 page_counter_init(&memcg->memsw, &parent->memsw);
4151 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08004152 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004153 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004154 page_counter_init(&memcg->memory, NULL);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004155 page_counter_init(&memcg->swap, NULL);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004156 page_counter_init(&memcg->memsw, NULL);
4157 page_counter_init(&memcg->kmem, NULL);
Johannes Weiner0db15292016-01-20 15:02:50 -08004158 page_counter_init(&memcg->tcpmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004159 /*
4160 * Deeper hierachy with use_hierarchy == false doesn't make
4161 * much sense so let cgroup subsystem know about this
4162 * unfortunate state in our controller.
4163 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004164 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004165 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004166 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08004167
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004168 /* The following stuff does not apply to the root */
4169 if (!parent) {
4170 root_mem_cgroup = memcg;
4171 return &memcg->css;
4172 }
4173
Vladimir Davydovb313aee2016-03-17 14:18:27 -07004174 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004175 if (error)
4176 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08004177
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004178 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004179 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004180
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004181 return &memcg->css;
4182fail:
4183 mem_cgroup_free(memcg);
4184 return NULL;
4185}
4186
4187static int
4188mem_cgroup_css_online(struct cgroup_subsys_state *css)
4189{
4190 if (css->id > MEM_CGROUP_ID_MAX)
4191 return -ENOSPC;
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004192
4193 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004194}
4195
Tejun Heoeb954192013-08-08 20:11:23 -04004196static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004197{
Tejun Heoeb954192013-08-08 20:11:23 -04004198 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004199 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004200
4201 /*
4202 * Unregister events and notify userspace.
4203 * Notify userspace about cgroup removing only after rmdir of cgroup
4204 * directory to avoid race between userspace and kernelspace.
4205 */
Tejun Heofba94802013-11-22 18:20:43 -05004206 spin_lock(&memcg->event_list_lock);
4207 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004208 list_del_init(&event->list);
4209 schedule_work(&event->remove);
4210 }
Tejun Heofba94802013-11-22 18:20:43 -05004211 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004212
Johannes Weiner567e9ab2016-01-20 15:02:24 -08004213 memcg_offline_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004214 wb_memcg_offline(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004215}
4216
Vladimir Davydov6df38682015-12-29 14:54:10 -08004217static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4218{
4219 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4220
4221 invalidate_reclaim_iterators(memcg);
4222}
4223
Tejun Heoeb954192013-08-08 20:11:23 -04004224static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004225{
Tejun Heoeb954192013-08-08 20:11:23 -04004226 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004227
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004228 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004229 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004230
Johannes Weiner0db15292016-01-20 15:02:50 -08004231 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004232 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004233
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004234 vmpressure_cleanup(&memcg->vmpressure);
4235 cancel_work_sync(&memcg->high_work);
4236 mem_cgroup_remove_from_trees(memcg);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004237 memcg_free_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004238 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004239}
4240
Tejun Heo1ced9532014-07-08 18:02:57 -04004241/**
4242 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4243 * @css: the target css
4244 *
4245 * Reset the states of the mem_cgroup associated with @css. This is
4246 * invoked when the userland requests disabling on the default hierarchy
4247 * but the memcg is pinned through dependency. The memcg should stop
4248 * applying policies and should revert to the vanilla state as it may be
4249 * made visible again.
4250 *
4251 * The current implementation only resets the essential configurations.
4252 * This needs to be expanded to cover all the visible parts.
4253 */
4254static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4255{
4256 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4257
Vladimir Davydovd334c9b2016-03-17 14:19:38 -07004258 page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4259 page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4260 page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4261 page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4262 page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
Johannes Weiner241994e2015-02-11 15:26:06 -08004263 memcg->low = 0;
4264 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004265 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004266 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004267}
4268
Daisuke Nishimura02491442010-03-10 15:22:17 -08004269#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004270/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004271static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004272{
Johannes Weiner05b84302014-08-06 16:05:59 -07004273 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004274
Mel Gormand0164ad2015-11-06 16:28:21 -08004275 /* Try a single bulk charge without reclaim first, kswapd may wake */
4276 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004277 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004278 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004279 return ret;
4280 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004281
4282 /* Try charges one by one with reclaim */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004283 while (count--) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004284 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004285 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004286 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004287 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004288 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004289 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004290 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004291}
4292
4293/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004294 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004295 * @vma: the vma the pte to be checked belongs
4296 * @addr: the address corresponding to the pte to be checked
4297 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08004298 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004299 *
4300 * Returns
4301 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4302 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4303 * move charge. if @target is not NULL, the page is stored in target->page
4304 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08004305 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4306 * target for charge migration. if @target is not NULL, the entry is stored
4307 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004308 *
4309 * Called with pte lock held.
4310 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004311union mc_target {
4312 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004313 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004314};
4315
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004316enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004317 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004318 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004319 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004320};
4321
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004322static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4323 unsigned long addr, pte_t ptent)
4324{
4325 struct page *page = vm_normal_page(vma, addr, ptent);
4326
4327 if (!page || !page_mapped(page))
4328 return NULL;
4329 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004330 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004331 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004332 } else {
4333 if (!(mc.flags & MOVE_FILE))
4334 return NULL;
4335 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004336 if (!get_page_unless_zero(page))
4337 return NULL;
4338
4339 return page;
4340}
4341
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004342#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004343static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4344 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4345{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004346 struct page *page = NULL;
4347 swp_entry_t ent = pte_to_swp_entry(ptent);
4348
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004349 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004350 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004351 /*
4352 * Because lookup_swap_cache() updates some statistics counter,
4353 * we call find_get_page() with swapper_space directly.
4354 */
Shaohua Li33806f02013-02-22 16:34:37 -08004355 page = find_get_page(swap_address_space(ent), ent.val);
Johannes Weiner7941d212016-01-14 15:21:23 -08004356 if (do_memsw_account())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004357 entry->val = ent.val;
4358
4359 return page;
4360}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004361#else
4362static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4363 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4364{
4365 return NULL;
4366}
4367#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004368
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004369static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4370 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4371{
4372 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004373 struct address_space *mapping;
4374 pgoff_t pgoff;
4375
4376 if (!vma->vm_file) /* anonymous vma */
4377 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004378 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004379 return NULL;
4380
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004381 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004382 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004383
4384 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004385#ifdef CONFIG_SWAP
4386 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004387 if (shmem_mapping(mapping)) {
4388 page = find_get_entry(mapping, pgoff);
4389 if (radix_tree_exceptional_entry(page)) {
4390 swp_entry_t swp = radix_to_swp_entry(page);
Johannes Weiner7941d212016-01-14 15:21:23 -08004391 if (do_memsw_account())
Johannes Weiner139b6a62014-05-06 12:50:05 -07004392 *entry = swp;
4393 page = find_get_page(swap_address_space(swp), swp.val);
4394 }
4395 } else
4396 page = find_get_page(mapping, pgoff);
4397#else
4398 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004399#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004400 return page;
4401}
4402
Chen Gangb1b0dea2015-04-14 15:47:35 -07004403/**
4404 * mem_cgroup_move_account - move account of the page
4405 * @page: the page
4406 * @nr_pages: number of regular pages (>1 for huge pages)
4407 * @from: mem_cgroup which the page is moved from.
4408 * @to: mem_cgroup which the page is moved to. @from != @to.
4409 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08004410 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07004411 *
4412 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4413 * from old cgroup.
4414 */
4415static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004416 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07004417 struct mem_cgroup *from,
4418 struct mem_cgroup *to)
4419{
4420 unsigned long flags;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004421 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004422 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004423 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004424
4425 VM_BUG_ON(from == to);
4426 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004427 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07004428
4429 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07004430 * Prevent mem_cgroup_migrate() from looking at
Hugh Dickins45637ba2015-11-05 18:49:40 -08004431 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07004432 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004433 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004434 if (!trylock_page(page))
4435 goto out;
4436
4437 ret = -EINVAL;
4438 if (page->mem_cgroup != from)
4439 goto out_unlock;
4440
Greg Thelenc4843a72015-05-22 17:13:16 -04004441 anon = PageAnon(page);
4442
Chen Gangb1b0dea2015-04-14 15:47:35 -07004443 spin_lock_irqsave(&from->move_lock, flags);
4444
Greg Thelenc4843a72015-05-22 17:13:16 -04004445 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004446 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4447 nr_pages);
4448 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4449 nr_pages);
4450 }
4451
Greg Thelenc4843a72015-05-22 17:13:16 -04004452 /*
4453 * move_lock grabbed above and caller set from->moving_account, so
4454 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4455 * So mapping should be stable for dirty pages.
4456 */
4457 if (!anon && PageDirty(page)) {
4458 struct address_space *mapping = page_mapping(page);
4459
4460 if (mapping_cap_account_dirty(mapping)) {
4461 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4462 nr_pages);
4463 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4464 nr_pages);
4465 }
4466 }
4467
Chen Gangb1b0dea2015-04-14 15:47:35 -07004468 if (PageWriteback(page)) {
4469 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4470 nr_pages);
4471 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4472 nr_pages);
4473 }
4474
4475 /*
4476 * It is safe to change page->mem_cgroup here because the page
4477 * is referenced, charged, and isolated - we can't race with
4478 * uncharging, charging, migration, or LRU putback.
4479 */
4480
4481 /* caller should have done css_get */
4482 page->mem_cgroup = to;
4483 spin_unlock_irqrestore(&from->move_lock, flags);
4484
4485 ret = 0;
4486
4487 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004488 mem_cgroup_charge_statistics(to, page, compound, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004489 memcg_check_events(to, page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004490 mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004491 memcg_check_events(from, page);
4492 local_irq_enable();
4493out_unlock:
4494 unlock_page(page);
4495out:
4496 return ret;
4497}
4498
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004499static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004500 unsigned long addr, pte_t ptent, union mc_target *target)
4501{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004502 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004503 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004504 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004505
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004506 if (pte_present(ptent))
4507 page = mc_handle_present_pte(vma, addr, ptent);
4508 else if (is_swap_pte(ptent))
4509 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004510 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004511 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004512
4513 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004514 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004515 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004516 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004517 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004518 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004519 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004520 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004521 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004522 ret = MC_TARGET_PAGE;
4523 if (target)
4524 target->page = page;
4525 }
4526 if (!ret || !target)
4527 put_page(page);
4528 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004529 /* There is a swap entry and a page doesn't exist or isn't charged */
4530 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004531 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004532 ret = MC_TARGET_SWAP;
4533 if (target)
4534 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004535 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004536 return ret;
4537}
4538
Naoya Horiguchi12724852012-03-21 16:34:28 -07004539#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4540/*
4541 * We don't consider swapping or file mapped pages because THP does not
4542 * support them for now.
4543 * Caller should make sure that pmd_trans_huge(pmd) is true.
4544 */
4545static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4546 unsigned long addr, pmd_t pmd, union mc_target *target)
4547{
4548 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004549 enum mc_target_type ret = MC_TARGET_NONE;
4550
4551 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004552 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004553 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004554 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004555 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004556 ret = MC_TARGET_PAGE;
4557 if (target) {
4558 get_page(page);
4559 target->page = page;
4560 }
4561 }
4562 return ret;
4563}
4564#else
4565static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4566 unsigned long addr, pmd_t pmd, union mc_target *target)
4567{
4568 return MC_TARGET_NONE;
4569}
4570#endif
4571
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004572static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4573 unsigned long addr, unsigned long end,
4574 struct mm_walk *walk)
4575{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004576 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004577 pte_t *pte;
4578 spinlock_t *ptl;
4579
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004580 ptl = pmd_trans_huge_lock(pmd, vma);
4581 if (ptl) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004582 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4583 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004584 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004585 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004586 }
Dave Hansen03319322011-03-22 16:32:56 -07004587
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004588 if (pmd_trans_unstable(pmd))
4589 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004590 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4591 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004592 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004593 mc.precharge++; /* increment precharge temporarily */
4594 pte_unmap_unlock(pte - 1, ptl);
4595 cond_resched();
4596
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004597 return 0;
4598}
4599
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004600static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4601{
4602 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004603
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004604 struct mm_walk mem_cgroup_count_precharge_walk = {
4605 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4606 .mm = mm,
4607 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004608 down_read(&mm->mmap_sem);
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004609 walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004610 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004611
4612 precharge = mc.precharge;
4613 mc.precharge = 0;
4614
4615 return precharge;
4616}
4617
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004618static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4619{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004620 unsigned long precharge = mem_cgroup_count_precharge(mm);
4621
4622 VM_BUG_ON(mc.moving_task);
4623 mc.moving_task = current;
4624 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004625}
4626
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004627/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4628static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004629{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004630 struct mem_cgroup *from = mc.from;
4631 struct mem_cgroup *to = mc.to;
4632
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004633 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004634 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004635 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004636 mc.precharge = 0;
4637 }
4638 /*
4639 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4640 * we must uncharge here.
4641 */
4642 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004643 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004644 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004645 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004646 /* we must fixup refcnts and charges */
4647 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004648 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004649 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004650 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004651
Johannes Weiner05b84302014-08-06 16:05:59 -07004652 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004653 * we charged both to->memory and to->memsw, so we
4654 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004655 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004656 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004657 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004658
Johannes Weinere8ea14c2014-12-10 15:42:42 -08004659 css_put_many(&mc.from->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004660
Li Zefan40503772013-07-08 16:00:34 -07004661 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004662 mc.moved_swap = 0;
4663 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004664 memcg_oom_recover(from);
4665 memcg_oom_recover(to);
4666 wake_up_all(&mc.waitq);
4667}
4668
4669static void mem_cgroup_clear_mc(void)
4670{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004671 struct mm_struct *mm = mc.mm;
4672
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004673 /*
4674 * we must clear moving_task before waking up waiters at the end of
4675 * task migration.
4676 */
4677 mc.moving_task = NULL;
4678 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004679 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004680 mc.from = NULL;
4681 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04004682 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004683 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004684
4685 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004686}
4687
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004688static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004689{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004690 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07004691 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07004692 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04004693 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004694 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004695 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004696 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004697
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004698 /* charge immigration isn't supported on the default hierarchy */
4699 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07004700 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004701
Tejun Heo4530edd2015-09-11 15:00:19 -04004702 /*
4703 * Multi-process migrations only happen on the default hierarchy
4704 * where charge immigration is not used. Perform charge
4705 * immigration if @tset contains a leader and whine if there are
4706 * multiple.
4707 */
4708 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004709 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04004710 WARN_ON_ONCE(p);
4711 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004712 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04004713 }
4714 if (!p)
4715 return 0;
4716
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004717 /*
4718 * We are now commited to this value whatever it is. Changes in this
4719 * tunable will only affect upcoming migrations, not the current one.
4720 * So we need to save it, and keep it going.
4721 */
4722 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4723 if (!move_flags)
4724 return 0;
4725
Tejun Heo9f2115f2015-09-08 15:01:10 -07004726 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004727
Tejun Heo9f2115f2015-09-08 15:01:10 -07004728 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004729
Tejun Heo9f2115f2015-09-08 15:01:10 -07004730 mm = get_task_mm(p);
4731 if (!mm)
4732 return 0;
4733 /* We move charges only when we move a owner of the mm */
4734 if (mm->owner == p) {
4735 VM_BUG_ON(mc.from);
4736 VM_BUG_ON(mc.to);
4737 VM_BUG_ON(mc.precharge);
4738 VM_BUG_ON(mc.moved_charge);
4739 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004740
Tejun Heo9f2115f2015-09-08 15:01:10 -07004741 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004742 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004743 mc.from = from;
4744 mc.to = memcg;
4745 mc.flags = move_flags;
4746 spin_unlock(&mc.lock);
4747 /* We set mc.moving_task later */
4748
4749 ret = mem_cgroup_precharge_mc(mm);
4750 if (ret)
4751 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004752 } else {
4753 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004754 }
4755 return ret;
4756}
4757
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004758static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004759{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004760 if (mc.to)
4761 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004762}
4763
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004764static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4765 unsigned long addr, unsigned long end,
4766 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004767{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004768 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004769 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004770 pte_t *pte;
4771 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004772 enum mc_target_type target_type;
4773 union mc_target target;
4774 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004775
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004776 ptl = pmd_trans_huge_lock(pmd, vma);
4777 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004778 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004779 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004780 return 0;
4781 }
4782 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4783 if (target_type == MC_TARGET_PAGE) {
4784 page = target.page;
4785 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004786 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08004787 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004788 mc.precharge -= HPAGE_PMD_NR;
4789 mc.moved_charge += HPAGE_PMD_NR;
4790 }
4791 putback_lru_page(page);
4792 }
4793 put_page(page);
4794 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004795 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004796 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004797 }
4798
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004799 if (pmd_trans_unstable(pmd))
4800 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004801retry:
4802 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4803 for (; addr != end; addr += PAGE_SIZE) {
4804 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004805 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004806
4807 if (!mc.precharge)
4808 break;
4809
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004810 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004811 case MC_TARGET_PAGE:
4812 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08004813 /*
4814 * We can have a part of the split pmd here. Moving it
4815 * can be done but it would be too convoluted so simply
4816 * ignore such a partial THP and keep it in original
4817 * memcg. There should be somebody mapping the head.
4818 */
4819 if (PageTransCompound(page))
4820 goto put;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004821 if (isolate_lru_page(page))
4822 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004823 if (!mem_cgroup_move_account(page, false,
4824 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004825 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004826 /* we uncharge from mc.from later. */
4827 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004828 }
4829 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004830put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004831 put_page(page);
4832 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004833 case MC_TARGET_SWAP:
4834 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004835 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004836 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004837 /* we fixup refcnts and charges later. */
4838 mc.moved_swap++;
4839 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004840 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004841 default:
4842 break;
4843 }
4844 }
4845 pte_unmap_unlock(pte - 1, ptl);
4846 cond_resched();
4847
4848 if (addr != end) {
4849 /*
4850 * We have consumed all precharges we got in can_attach().
4851 * We try charge one by one, but don't do any additional
4852 * charges to mc.to if we have failed in charge once in attach()
4853 * phase.
4854 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004855 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004856 if (!ret)
4857 goto retry;
4858 }
4859
4860 return ret;
4861}
4862
Tejun Heo264a0ae2016-04-21 19:09:02 -04004863static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004864{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004865 struct mm_walk mem_cgroup_move_charge_walk = {
4866 .pmd_entry = mem_cgroup_move_charge_pte_range,
Tejun Heo264a0ae2016-04-21 19:09:02 -04004867 .mm = mc.mm,
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004868 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004869
4870 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004871 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07004872 * Signal lock_page_memcg() to take the memcg's move_lock
4873 * while we're moving its pages to another memcg. Then wait
4874 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08004875 */
4876 atomic_inc(&mc.from->moving_account);
4877 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004878retry:
Tejun Heo264a0ae2016-04-21 19:09:02 -04004879 if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004880 /*
4881 * Someone who are holding the mmap_sem might be waiting in
4882 * waitq. So we cancel all extra charges, wake up all waiters,
4883 * and retry. Because we cancel precharges, we might not be able
4884 * to move enough charges, but moving charge is a best-effort
4885 * feature anyway, so it wouldn't be a big problem.
4886 */
4887 __mem_cgroup_clear_mc();
4888 cond_resched();
4889 goto retry;
4890 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004891 /*
4892 * When we have consumed all precharges and failed in doing
4893 * additional charge, the page walk just aborts.
4894 */
4895 walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004896 up_read(&mc.mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08004897 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004898}
4899
Tejun Heo264a0ae2016-04-21 19:09:02 -04004900static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08004901{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004902 if (mc.to) {
4903 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004904 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004905 }
Balbir Singh67e465a2008-02-07 00:13:54 -08004906}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004907#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004908static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004909{
4910 return 0;
4911}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004912static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004913{
4914}
Tejun Heo264a0ae2016-04-21 19:09:02 -04004915static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004916{
4917}
4918#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08004919
Tejun Heof00baae2013-04-15 13:41:15 -07004920/*
4921 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04004922 * to verify whether we're attached to the default hierarchy on each mount
4923 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07004924 */
Tejun Heoeb954192013-08-08 20:11:23 -04004925static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07004926{
4927 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04004928 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07004929 * guarantees that @root doesn't have any children, so turning it
4930 * on for the root memcg is enough.
4931 */
Tejun Heo9e10a132015-09-18 11:56:28 -04004932 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee592015-03-12 16:26:19 -07004933 root_mem_cgroup->use_hierarchy = true;
4934 else
4935 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07004936}
4937
Johannes Weiner241994e2015-02-11 15:26:06 -08004938static u64 memory_current_read(struct cgroup_subsys_state *css,
4939 struct cftype *cft)
4940{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08004941 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4942
4943 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994e2015-02-11 15:26:06 -08004944}
4945
4946static int memory_low_show(struct seq_file *m, void *v)
4947{
4948 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07004949 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994e2015-02-11 15:26:06 -08004950
4951 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08004952 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08004953 else
4954 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
4955
4956 return 0;
4957}
4958
4959static ssize_t memory_low_write(struct kernfs_open_file *of,
4960 char *buf, size_t nbytes, loff_t off)
4961{
4962 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
4963 unsigned long low;
4964 int err;
4965
4966 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08004967 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994e2015-02-11 15:26:06 -08004968 if (err)
4969 return err;
4970
4971 memcg->low = low;
4972
4973 return nbytes;
4974}
4975
4976static int memory_high_show(struct seq_file *m, void *v)
4977{
4978 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07004979 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994e2015-02-11 15:26:06 -08004980
4981 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08004982 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08004983 else
4984 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
4985
4986 return 0;
4987}
4988
4989static ssize_t memory_high_write(struct kernfs_open_file *of,
4990 char *buf, size_t nbytes, loff_t off)
4991{
4992 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner588083b2016-03-17 14:20:25 -07004993 unsigned long nr_pages;
Johannes Weiner241994e2015-02-11 15:26:06 -08004994 unsigned long high;
4995 int err;
4996
4997 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08004998 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994e2015-02-11 15:26:06 -08004999 if (err)
5000 return err;
5001
5002 memcg->high = high;
5003
Johannes Weiner588083b2016-03-17 14:20:25 -07005004 nr_pages = page_counter_read(&memcg->memory);
5005 if (nr_pages > high)
5006 try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5007 GFP_KERNEL, true);
5008
Tejun Heo2529bb32015-05-22 18:23:34 -04005009 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005010 return nbytes;
5011}
5012
5013static int memory_max_show(struct seq_file *m, void *v)
5014{
5015 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005016 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994e2015-02-11 15:26:06 -08005017
5018 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005019 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005020 else
5021 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5022
5023 return 0;
5024}
5025
5026static ssize_t memory_max_write(struct kernfs_open_file *of,
5027 char *buf, size_t nbytes, loff_t off)
5028{
5029 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005030 unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5031 bool drained = false;
Johannes Weiner241994e2015-02-11 15:26:06 -08005032 unsigned long max;
5033 int err;
5034
5035 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005036 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994e2015-02-11 15:26:06 -08005037 if (err)
5038 return err;
5039
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005040 xchg(&memcg->memory.limit, max);
5041
5042 for (;;) {
5043 unsigned long nr_pages = page_counter_read(&memcg->memory);
5044
5045 if (nr_pages <= max)
5046 break;
5047
5048 if (signal_pending(current)) {
5049 err = -EINTR;
5050 break;
5051 }
5052
5053 if (!drained) {
5054 drain_all_stock(memcg);
5055 drained = true;
5056 continue;
5057 }
5058
5059 if (nr_reclaims) {
5060 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5061 GFP_KERNEL, true))
5062 nr_reclaims--;
5063 continue;
5064 }
5065
5066 mem_cgroup_events(memcg, MEMCG_OOM, 1);
5067 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5068 break;
5069 }
Johannes Weiner241994e2015-02-11 15:26:06 -08005070
Tejun Heo2529bb32015-05-22 18:23:34 -04005071 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005072 return nbytes;
5073}
5074
5075static int memory_events_show(struct seq_file *m, void *v)
5076{
5077 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5078
5079 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5080 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5081 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5082 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5083
5084 return 0;
5085}
5086
Johannes Weiner587d9f72016-01-20 15:03:19 -08005087static int memory_stat_show(struct seq_file *m, void *v)
5088{
5089 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005090 unsigned long stat[MEMCG_NR_STAT];
5091 unsigned long events[MEMCG_NR_EVENTS];
Johannes Weiner587d9f72016-01-20 15:03:19 -08005092 int i;
5093
5094 /*
5095 * Provide statistics on the state of the memory subsystem as
5096 * well as cumulative event counters that show past behavior.
5097 *
5098 * This list is ordered following a combination of these gradients:
5099 * 1) generic big picture -> specifics and details
5100 * 2) reflecting userspace activity -> reflecting kernel heuristics
5101 *
5102 * Current memory state:
5103 */
5104
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005105 tree_stat(memcg, stat);
5106 tree_events(memcg, events);
5107
Johannes Weiner587d9f72016-01-20 15:03:19 -08005108 seq_printf(m, "anon %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005109 (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005110 seq_printf(m, "file %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005111 (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE);
Vladimir Davydov12580e42016-03-17 14:17:38 -07005112 seq_printf(m, "kernel_stack %llu\n",
5113 (u64)stat[MEMCG_KERNEL_STACK] * PAGE_SIZE);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005114 seq_printf(m, "slab %llu\n",
5115 (u64)(stat[MEMCG_SLAB_RECLAIMABLE] +
5116 stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
Johannes Weinerb2807f02016-01-20 15:03:22 -08005117 seq_printf(m, "sock %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005118 (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005119
5120 seq_printf(m, "file_mapped %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005121 (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005122 seq_printf(m, "file_dirty %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005123 (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005124 seq_printf(m, "file_writeback %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005125 (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005126
5127 for (i = 0; i < NR_LRU_LISTS; i++) {
5128 struct mem_cgroup *mi;
5129 unsigned long val = 0;
5130
5131 for_each_mem_cgroup_tree(mi, memcg)
5132 val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5133 seq_printf(m, "%s %llu\n",
5134 mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5135 }
5136
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005137 seq_printf(m, "slab_reclaimable %llu\n",
5138 (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE);
5139 seq_printf(m, "slab_unreclaimable %llu\n",
5140 (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5141
Johannes Weiner587d9f72016-01-20 15:03:19 -08005142 /* Accumulated memory events */
5143
5144 seq_printf(m, "pgfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005145 events[MEM_CGROUP_EVENTS_PGFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005146 seq_printf(m, "pgmajfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005147 events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005148
5149 return 0;
5150}
5151
Johannes Weiner241994e2015-02-11 15:26:06 -08005152static struct cftype memory_files[] = {
5153 {
5154 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005155 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994e2015-02-11 15:26:06 -08005156 .read_u64 = memory_current_read,
5157 },
5158 {
5159 .name = "low",
5160 .flags = CFTYPE_NOT_ON_ROOT,
5161 .seq_show = memory_low_show,
5162 .write = memory_low_write,
5163 },
5164 {
5165 .name = "high",
5166 .flags = CFTYPE_NOT_ON_ROOT,
5167 .seq_show = memory_high_show,
5168 .write = memory_high_write,
5169 },
5170 {
5171 .name = "max",
5172 .flags = CFTYPE_NOT_ON_ROOT,
5173 .seq_show = memory_max_show,
5174 .write = memory_max_write,
5175 },
5176 {
5177 .name = "events",
5178 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04005179 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994e2015-02-11 15:26:06 -08005180 .seq_show = memory_events_show,
5181 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08005182 {
5183 .name = "stat",
5184 .flags = CFTYPE_NOT_ON_ROOT,
5185 .seq_show = memory_stat_show,
5186 },
Johannes Weiner241994e2015-02-11 15:26:06 -08005187 { } /* terminate */
5188};
5189
Tejun Heo073219e2014-02-08 10:36:58 -05005190struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005191 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005192 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005193 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08005194 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08005195 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005196 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005197 .can_attach = mem_cgroup_can_attach,
5198 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04005199 .post_attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005200 .bind = mem_cgroup_bind,
Johannes Weiner241994e2015-02-11 15:26:06 -08005201 .dfl_cftypes = memory_files,
5202 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005203 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005204};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005205
Johannes Weiner241994e2015-02-11 15:26:06 -08005206/**
Johannes Weiner241994e2015-02-11 15:26:06 -08005207 * mem_cgroup_low - check if memory consumption is below the normal range
5208 * @root: the highest ancestor to consider
5209 * @memcg: the memory cgroup to check
5210 *
5211 * Returns %true if memory consumption of @memcg, and that of all
5212 * configurable ancestors up to @root, is below the normal range.
5213 */
5214bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5215{
5216 if (mem_cgroup_disabled())
5217 return false;
5218
5219 /*
5220 * The toplevel group doesn't have a configurable range, so
5221 * it's never low when looked at directly, and it is not
5222 * considered an ancestor when assessing the hierarchy.
5223 */
5224
5225 if (memcg == root_mem_cgroup)
5226 return false;
5227
Michal Hocko4e54ded2015-02-27 15:51:46 -08005228 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005229 return false;
5230
5231 while (memcg != root) {
5232 memcg = parent_mem_cgroup(memcg);
5233
5234 if (memcg == root_mem_cgroup)
5235 break;
5236
Michal Hocko4e54ded2015-02-27 15:51:46 -08005237 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005238 return false;
5239 }
5240 return true;
5241}
5242
Johannes Weiner00501b52014-08-08 14:19:20 -07005243/**
5244 * mem_cgroup_try_charge - try charging a page
5245 * @page: page to charge
5246 * @mm: mm context of the victim
5247 * @gfp_mask: reclaim mode
5248 * @memcgp: charged memcg return
5249 *
5250 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5251 * pages according to @gfp_mask if necessary.
5252 *
5253 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5254 * Otherwise, an error code is returned.
5255 *
5256 * After page->mapping has been set up, the caller must finalize the
5257 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5258 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5259 */
5260int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005261 gfp_t gfp_mask, struct mem_cgroup **memcgp,
5262 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005263{
5264 struct mem_cgroup *memcg = NULL;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005265 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005266 int ret = 0;
5267
5268 if (mem_cgroup_disabled())
5269 goto out;
5270
5271 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005272 /*
5273 * Every swap fault against a single page tries to charge the
5274 * page, bail as early as possible. shmem_unuse() encounters
5275 * already charged pages, too. The USED bit is protected by
5276 * the page lock, which serializes swap cache removal, which
5277 * in turn serializes uncharging.
5278 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005279 VM_BUG_ON_PAGE(!PageLocked(page), page);
Johannes Weiner1306a852014-12-10 15:44:52 -08005280 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005281 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005282
Vladimir Davydov37e84352016-01-20 15:02:56 -08005283 if (do_swap_account) {
Vladimir Davydove993d902015-09-09 15:35:35 -07005284 swp_entry_t ent = { .val = page_private(page), };
5285 unsigned short id = lookup_swap_cgroup_id(ent);
5286
5287 rcu_read_lock();
5288 memcg = mem_cgroup_from_id(id);
5289 if (memcg && !css_tryget_online(&memcg->css))
5290 memcg = NULL;
5291 rcu_read_unlock();
5292 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005293 }
5294
Johannes Weiner00501b52014-08-08 14:19:20 -07005295 if (!memcg)
5296 memcg = get_mem_cgroup_from_mm(mm);
5297
5298 ret = try_charge(memcg, gfp_mask, nr_pages);
5299
5300 css_put(&memcg->css);
Johannes Weiner00501b52014-08-08 14:19:20 -07005301out:
5302 *memcgp = memcg;
5303 return ret;
5304}
5305
5306/**
5307 * mem_cgroup_commit_charge - commit a page charge
5308 * @page: page to charge
5309 * @memcg: memcg to charge the page to
5310 * @lrucare: page might be on LRU already
5311 *
5312 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5313 * after page->mapping has been set up. This must happen atomically
5314 * as part of the page instantiation, i.e. under the page table lock
5315 * for anonymous pages, under the page lock for page and swap cache.
5316 *
5317 * In addition, the page must not be on the LRU during the commit, to
5318 * prevent racing with task migration. If it might be, use @lrucare.
5319 *
5320 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5321 */
5322void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005323 bool lrucare, bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005324{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005325 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005326
5327 VM_BUG_ON_PAGE(!page->mapping, page);
5328 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5329
5330 if (mem_cgroup_disabled())
5331 return;
5332 /*
5333 * Swap faults will attempt to charge the same page multiple
5334 * times. But reuse_swap_page() might have removed the page
5335 * from swapcache already, so we can't check PageSwapCache().
5336 */
5337 if (!memcg)
5338 return;
5339
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005340 commit_charge(page, memcg, lrucare);
5341
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005342 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005343 mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005344 memcg_check_events(memcg, page);
5345 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005346
Johannes Weiner7941d212016-01-14 15:21:23 -08005347 if (do_memsw_account() && PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005348 swp_entry_t entry = { .val = page_private(page) };
5349 /*
5350 * The swap entry might not get freed for a long time,
5351 * let's not wait for it. The page already received a
5352 * memory+swap charge, drop the swap entry duplicate.
5353 */
5354 mem_cgroup_uncharge_swap(entry);
5355 }
5356}
5357
5358/**
5359 * mem_cgroup_cancel_charge - cancel a page charge
5360 * @page: page to charge
5361 * @memcg: memcg to charge the page to
5362 *
5363 * Cancel a charge transaction started by mem_cgroup_try_charge().
5364 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005365void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5366 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005367{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005368 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005369
5370 if (mem_cgroup_disabled())
5371 return;
5372 /*
5373 * Swap faults will attempt to charge the same page multiple
5374 * times. But reuse_swap_page() might have removed the page
5375 * from swapcache already, so we can't check PageSwapCache().
5376 */
5377 if (!memcg)
5378 return;
5379
Johannes Weiner00501b52014-08-08 14:19:20 -07005380 cancel_charge(memcg, nr_pages);
5381}
5382
Johannes Weiner747db952014-08-08 14:19:24 -07005383static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005384 unsigned long nr_anon, unsigned long nr_file,
5385 unsigned long nr_huge, struct page *dummy_page)
5386{
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005387 unsigned long nr_pages = nr_anon + nr_file;
Johannes Weiner747db952014-08-08 14:19:24 -07005388 unsigned long flags;
5389
Johannes Weinerce00a962014-09-05 08:43:57 -04005390 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005391 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08005392 if (do_memsw_account())
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005393 page_counter_uncharge(&memcg->memsw, nr_pages);
Johannes Weinerce00a962014-09-05 08:43:57 -04005394 memcg_oom_recover(memcg);
5395 }
Johannes Weiner747db952014-08-08 14:19:24 -07005396
5397 local_irq_save(flags);
5398 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5399 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5400 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5401 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005402 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005403 memcg_check_events(memcg, dummy_page);
5404 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005405
5406 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005407 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005408}
5409
5410static void uncharge_list(struct list_head *page_list)
5411{
5412 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005413 unsigned long nr_anon = 0;
5414 unsigned long nr_file = 0;
5415 unsigned long nr_huge = 0;
5416 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005417 struct list_head *next;
5418 struct page *page;
5419
Johannes Weiner8b592652016-03-17 14:20:31 -07005420 /*
5421 * Note that the list can be a single page->lru; hence the
5422 * do-while loop instead of a simple list_for_each_entry().
5423 */
Johannes Weiner747db952014-08-08 14:19:24 -07005424 next = page_list->next;
5425 do {
5426 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005427
5428 page = list_entry(next, struct page, lru);
5429 next = page->lru.next;
5430
5431 VM_BUG_ON_PAGE(PageLRU(page), page);
5432 VM_BUG_ON_PAGE(page_count(page), page);
5433
Johannes Weiner1306a852014-12-10 15:44:52 -08005434 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005435 continue;
5436
5437 /*
5438 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005439 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005440 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005441 */
5442
Johannes Weiner1306a852014-12-10 15:44:52 -08005443 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005444 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005445 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5446 nr_huge, page);
5447 pgpgout = nr_anon = nr_file = nr_huge = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005448 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005449 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005450 }
5451
5452 if (PageTransHuge(page)) {
5453 nr_pages <<= compound_order(page);
5454 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5455 nr_huge += nr_pages;
5456 }
5457
5458 if (PageAnon(page))
5459 nr_anon += nr_pages;
5460 else
5461 nr_file += nr_pages;
5462
Johannes Weiner1306a852014-12-10 15:44:52 -08005463 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005464
5465 pgpgout++;
5466 } while (next != page_list);
5467
5468 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005469 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5470 nr_huge, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005471}
5472
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005473/**
5474 * mem_cgroup_uncharge - uncharge a page
5475 * @page: page to uncharge
5476 *
5477 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5478 * mem_cgroup_commit_charge().
5479 */
5480void mem_cgroup_uncharge(struct page *page)
5481{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005482 if (mem_cgroup_disabled())
5483 return;
5484
Johannes Weiner747db952014-08-08 14:19:24 -07005485 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005486 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005487 return;
5488
Johannes Weiner747db952014-08-08 14:19:24 -07005489 INIT_LIST_HEAD(&page->lru);
5490 uncharge_list(&page->lru);
5491}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005492
Johannes Weiner747db952014-08-08 14:19:24 -07005493/**
5494 * mem_cgroup_uncharge_list - uncharge a list of page
5495 * @page_list: list of pages to uncharge
5496 *
5497 * Uncharge a list of pages previously charged with
5498 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5499 */
5500void mem_cgroup_uncharge_list(struct list_head *page_list)
5501{
5502 if (mem_cgroup_disabled())
5503 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005504
Johannes Weiner747db952014-08-08 14:19:24 -07005505 if (!list_empty(page_list))
5506 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005507}
5508
5509/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005510 * mem_cgroup_migrate - charge a page's replacement
5511 * @oldpage: currently circulating page
5512 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005513 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005514 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5515 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005516 *
5517 * Both pages must be locked, @newpage->mapping must be set up.
5518 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005519void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005520{
Johannes Weiner29833312014-12-10 15:44:02 -08005521 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005522 unsigned int nr_pages;
5523 bool compound;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005524
5525 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5526 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005527 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005528 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5529 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005530
5531 if (mem_cgroup_disabled())
5532 return;
5533
5534 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005535 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005536 return;
5537
Hugh Dickins45637ba2015-11-05 18:49:40 -08005538 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08005539 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005540 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005541 return;
5542
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005543 /* Force-charge the new page. The old one will be freed soon */
5544 compound = PageTransHuge(newpage);
5545 nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5546
5547 page_counter_charge(&memcg->memory, nr_pages);
5548 if (do_memsw_account())
5549 page_counter_charge(&memcg->memsw, nr_pages);
5550 css_get_many(&memcg->css, nr_pages);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005551
Johannes Weiner9cf76662016-03-15 14:57:58 -07005552 commit_charge(newpage, memcg, false);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005553
5554 local_irq_disable();
5555 mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5556 memcg_check_events(memcg, newpage);
5557 local_irq_enable();
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005558}
5559
Johannes Weineref129472016-01-14 15:21:34 -08005560DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08005561EXPORT_SYMBOL(memcg_sockets_enabled_key);
5562
5563void sock_update_memcg(struct sock *sk)
5564{
5565 struct mem_cgroup *memcg;
5566
5567 /* Socket cloning can throw us here with sk_cgrp already
5568 * filled. It won't however, necessarily happen from
5569 * process context. So the test for root memcg given
5570 * the current task's memcg won't help us in this case.
5571 *
5572 * Respecting the original socket's memcg is a better
5573 * decision in this case.
5574 */
5575 if (sk->sk_memcg) {
5576 BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5577 css_get(&sk->sk_memcg->css);
5578 return;
5579 }
5580
5581 rcu_read_lock();
5582 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005583 if (memcg == root_mem_cgroup)
5584 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08005585 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005586 goto out;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005587 if (css_tryget_online(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08005588 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005589out:
Johannes Weiner11092082016-01-14 15:21:26 -08005590 rcu_read_unlock();
5591}
5592EXPORT_SYMBOL(sock_update_memcg);
5593
5594void sock_release_memcg(struct sock *sk)
5595{
5596 WARN_ON(!sk->sk_memcg);
5597 css_put(&sk->sk_memcg->css);
5598}
5599
5600/**
5601 * mem_cgroup_charge_skmem - charge socket memory
5602 * @memcg: memcg to charge
5603 * @nr_pages: number of pages to charge
5604 *
5605 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5606 * @memcg's configured limit, %false if the charge had to be forced.
5607 */
5608bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5609{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005610 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08005611
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005612 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005613 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005614
Johannes Weiner0db15292016-01-20 15:02:50 -08005615 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5616 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005617 return true;
5618 }
Johannes Weiner0db15292016-01-20 15:02:50 -08005619 page_counter_charge(&memcg->tcpmem, nr_pages);
5620 memcg->tcpmem_pressure = 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005621 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08005622 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005623
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005624 /* Don't block in the packet receive path */
5625 if (in_softirq())
5626 gfp_mask = GFP_NOWAIT;
5627
Johannes Weinerb2807f02016-01-20 15:03:22 -08005628 this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5629
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005630 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5631 return true;
5632
5633 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005634 return false;
5635}
5636
5637/**
5638 * mem_cgroup_uncharge_skmem - uncharge socket memory
5639 * @memcg - memcg to uncharge
5640 * @nr_pages - number of pages to uncharge
5641 */
5642void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5643{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005644 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005645 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005646 return;
5647 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005648
Johannes Weinerb2807f02016-01-20 15:03:22 -08005649 this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5650
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005651 page_counter_uncharge(&memcg->memory, nr_pages);
5652 css_put_many(&memcg->css, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005653}
5654
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005655static int __init cgroup_memory(char *s)
5656{
5657 char *token;
5658
5659 while ((token = strsep(&s, ",")) != NULL) {
5660 if (!*token)
5661 continue;
5662 if (!strcmp(token, "nosocket"))
5663 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08005664 if (!strcmp(token, "nokmem"))
5665 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005666 }
5667 return 0;
5668}
5669__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08005670
Michal Hocko2d110852013-02-22 16:34:43 -08005671/*
Michal Hocko10813122013-02-22 16:35:41 -08005672 * subsys_initcall() for memory controller.
5673 *
5674 * Some parts like hotcpu_notifier() have to be initialized from this context
5675 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5676 * everything that doesn't depend on a specific mem_cgroup structure should
5677 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005678 */
5679static int __init mem_cgroup_init(void)
5680{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005681 int cpu, node;
5682
Michal Hocko2d110852013-02-22 16:34:43 -08005683 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005684
5685 for_each_possible_cpu(cpu)
5686 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5687 drain_local_stock);
5688
5689 for_each_node(node) {
5690 struct mem_cgroup_tree_per_node *rtpn;
5691 int zone;
5692
5693 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5694 node_online(node) ? node : NUMA_NO_NODE);
5695
5696 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5697 struct mem_cgroup_tree_per_zone *rtpz;
5698
5699 rtpz = &rtpn->rb_tree_per_zone[zone];
5700 rtpz->rb_root = RB_ROOT;
5701 spin_lock_init(&rtpz->lock);
5702 }
5703 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5704 }
5705
Michal Hocko2d110852013-02-22 16:34:43 -08005706 return 0;
5707}
5708subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005709
5710#ifdef CONFIG_MEMCG_SWAP
5711/**
5712 * mem_cgroup_swapout - transfer a memsw charge to swap
5713 * @page: page whose memsw charge to transfer
5714 * @entry: swap entry to move the charge to
5715 *
5716 * Transfer the memsw charge of @page to @entry.
5717 */
5718void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5719{
5720 struct mem_cgroup *memcg;
5721 unsigned short oldid;
5722
5723 VM_BUG_ON_PAGE(PageLRU(page), page);
5724 VM_BUG_ON_PAGE(page_count(page), page);
5725
Johannes Weiner7941d212016-01-14 15:21:23 -08005726 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08005727 return;
5728
5729 memcg = page->mem_cgroup;
5730
5731 /* Readahead page, never charged */
5732 if (!memcg)
5733 return;
5734
5735 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5736 VM_BUG_ON_PAGE(oldid, page);
5737 mem_cgroup_swap_statistics(memcg, true);
5738
5739 page->mem_cgroup = NULL;
5740
5741 if (!mem_cgroup_is_root(memcg))
5742 page_counter_uncharge(&memcg->memory, 1);
5743
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07005744 /*
5745 * Interrupts should be disabled here because the caller holds the
5746 * mapping->tree_lock lock which is taken with interrupts-off. It is
5747 * important here to have the interrupts disabled because it is the
5748 * only synchronisation we have for udpating the per-CPU variables.
5749 */
5750 VM_BUG_ON(!irqs_disabled());
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005751 mem_cgroup_charge_statistics(memcg, page, false, -1);
Johannes Weiner21afa382015-02-11 15:26:36 -08005752 memcg_check_events(memcg, page);
5753}
5754
Vladimir Davydov37e84352016-01-20 15:02:56 -08005755/*
5756 * mem_cgroup_try_charge_swap - try charging a swap entry
5757 * @page: page being added to swap
5758 * @entry: swap entry to charge
5759 *
5760 * Try to charge @entry to the memcg that @page belongs to.
5761 *
5762 * Returns 0 on success, -ENOMEM on failure.
5763 */
5764int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5765{
5766 struct mem_cgroup *memcg;
5767 struct page_counter *counter;
5768 unsigned short oldid;
5769
5770 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5771 return 0;
5772
5773 memcg = page->mem_cgroup;
5774
5775 /* Readahead page, never charged */
5776 if (!memcg)
5777 return 0;
5778
5779 if (!mem_cgroup_is_root(memcg) &&
5780 !page_counter_try_charge(&memcg->swap, 1, &counter))
5781 return -ENOMEM;
5782
5783 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5784 VM_BUG_ON_PAGE(oldid, page);
5785 mem_cgroup_swap_statistics(memcg, true);
5786
5787 css_get(&memcg->css);
5788 return 0;
5789}
5790
Johannes Weiner21afa382015-02-11 15:26:36 -08005791/**
5792 * mem_cgroup_uncharge_swap - uncharge a swap entry
5793 * @entry: swap entry to uncharge
5794 *
Vladimir Davydov37e84352016-01-20 15:02:56 -08005795 * Drop the swap charge associated with @entry.
Johannes Weiner21afa382015-02-11 15:26:36 -08005796 */
5797void mem_cgroup_uncharge_swap(swp_entry_t entry)
5798{
5799 struct mem_cgroup *memcg;
5800 unsigned short id;
5801
Vladimir Davydov37e84352016-01-20 15:02:56 -08005802 if (!do_swap_account)
Johannes Weiner21afa382015-02-11 15:26:36 -08005803 return;
5804
5805 id = swap_cgroup_record(entry, 0);
5806 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005807 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005808 if (memcg) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08005809 if (!mem_cgroup_is_root(memcg)) {
5810 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5811 page_counter_uncharge(&memcg->swap, 1);
5812 else
5813 page_counter_uncharge(&memcg->memsw, 1);
5814 }
Johannes Weiner21afa382015-02-11 15:26:36 -08005815 mem_cgroup_swap_statistics(memcg, false);
5816 css_put(&memcg->css);
5817 }
5818 rcu_read_unlock();
5819}
5820
Vladimir Davydovd8b38432016-01-20 15:03:07 -08005821long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
5822{
5823 long nr_swap_pages = get_nr_swap_pages();
5824
5825 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5826 return nr_swap_pages;
5827 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5828 nr_swap_pages = min_t(long, nr_swap_pages,
5829 READ_ONCE(memcg->swap.limit) -
5830 page_counter_read(&memcg->swap));
5831 return nr_swap_pages;
5832}
5833
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08005834bool mem_cgroup_swap_full(struct page *page)
5835{
5836 struct mem_cgroup *memcg;
5837
5838 VM_BUG_ON_PAGE(!PageLocked(page), page);
5839
5840 if (vm_swap_full())
5841 return true;
5842 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5843 return false;
5844
5845 memcg = page->mem_cgroup;
5846 if (!memcg)
5847 return false;
5848
5849 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5850 if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
5851 return true;
5852
5853 return false;
5854}
5855
Johannes Weiner21afa382015-02-11 15:26:36 -08005856/* for remember boot option*/
5857#ifdef CONFIG_MEMCG_SWAP_ENABLED
5858static int really_do_swap_account __initdata = 1;
5859#else
5860static int really_do_swap_account __initdata;
5861#endif
5862
5863static int __init enable_swap_account(char *s)
5864{
5865 if (!strcmp(s, "1"))
5866 really_do_swap_account = 1;
5867 else if (!strcmp(s, "0"))
5868 really_do_swap_account = 0;
5869 return 1;
5870}
5871__setup("swapaccount=", enable_swap_account);
5872
Vladimir Davydov37e84352016-01-20 15:02:56 -08005873static u64 swap_current_read(struct cgroup_subsys_state *css,
5874 struct cftype *cft)
5875{
5876 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5877
5878 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
5879}
5880
5881static int swap_max_show(struct seq_file *m, void *v)
5882{
5883 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5884 unsigned long max = READ_ONCE(memcg->swap.limit);
5885
5886 if (max == PAGE_COUNTER_MAX)
5887 seq_puts(m, "max\n");
5888 else
5889 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5890
5891 return 0;
5892}
5893
5894static ssize_t swap_max_write(struct kernfs_open_file *of,
5895 char *buf, size_t nbytes, loff_t off)
5896{
5897 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5898 unsigned long max;
5899 int err;
5900
5901 buf = strstrip(buf);
5902 err = page_counter_memparse(buf, "max", &max);
5903 if (err)
5904 return err;
5905
5906 mutex_lock(&memcg_limit_mutex);
5907 err = page_counter_limit(&memcg->swap, max);
5908 mutex_unlock(&memcg_limit_mutex);
5909 if (err)
5910 return err;
5911
5912 return nbytes;
5913}
5914
5915static struct cftype swap_files[] = {
5916 {
5917 .name = "swap.current",
5918 .flags = CFTYPE_NOT_ON_ROOT,
5919 .read_u64 = swap_current_read,
5920 },
5921 {
5922 .name = "swap.max",
5923 .flags = CFTYPE_NOT_ON_ROOT,
5924 .seq_show = swap_max_show,
5925 .write = swap_max_write,
5926 },
5927 { } /* terminate */
5928};
5929
Johannes Weiner21afa382015-02-11 15:26:36 -08005930static struct cftype memsw_cgroup_files[] = {
5931 {
5932 .name = "memsw.usage_in_bytes",
5933 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
5934 .read_u64 = mem_cgroup_read_u64,
5935 },
5936 {
5937 .name = "memsw.max_usage_in_bytes",
5938 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
5939 .write = mem_cgroup_reset,
5940 .read_u64 = mem_cgroup_read_u64,
5941 },
5942 {
5943 .name = "memsw.limit_in_bytes",
5944 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
5945 .write = mem_cgroup_write,
5946 .read_u64 = mem_cgroup_read_u64,
5947 },
5948 {
5949 .name = "memsw.failcnt",
5950 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
5951 .write = mem_cgroup_reset,
5952 .read_u64 = mem_cgroup_read_u64,
5953 },
5954 { }, /* terminate */
5955};
5956
5957static int __init mem_cgroup_swap_init(void)
5958{
5959 if (!mem_cgroup_disabled() && really_do_swap_account) {
5960 do_swap_account = 1;
Vladimir Davydov37e84352016-01-20 15:02:56 -08005961 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
5962 swap_files));
Johannes Weiner21afa382015-02-11 15:26:36 -08005963 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
5964 memsw_cgroup_files));
5965 }
5966 return 0;
5967}
5968subsys_initcall(mem_cgroup_swap_init);
5969
5970#endif /* CONFIG_MEMCG_SWAP */