blob: b3f16ab4b4311d00efd34e9e53022a42b0d43d28 [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 Mortonbb4cc1a82013-09-24 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 Weiner5d1ea48b2014-12-10 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 15:27:40 -0700476 struct mem_cgroup_per_zone *mz;
477 struct mem_cgroup_tree_per_zone *mctz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700478
Jianyu Zhane2318752014-06-06 14:38:20 -0700479 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 15:27:40 -0700504 }
505 }
506}
507
508static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
509{
Andrew Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Mortonbb4cc1a82013-09-24 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 Weinerf53d7ce32012-01-12 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 Weinerf53d7ce32012-01-12 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 Mortonbb4cc1a82013-09-24 15:27:40 -0700688 case MEM_CGROUP_TARGET_SOFTLIMIT:
689 next = val + SOFTLIMIT_EVENTS_TARGET;
690 break;
Johannes Weinerf53d7ce32012-01-12 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 Weinerf53d7ce32012-01-12 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 Weinerf53d7ce32012-01-12 17:18:23 -0800710 if (unlikely(mem_cgroup_event_ratelimit(memcg,
711 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700712 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800713 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800714
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700715 do_softlimit = mem_cgroup_event_ratelimit(memcg,
716 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -0700717#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800718 do_numainfo = mem_cgroup_event_ratelimit(memcg,
719 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -0700720#endif
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800721 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700722 if (unlikely(do_softlimit))
723 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce32012-01-12 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 Dickinsca707232016-05-19 17:12:35 -07001026 * This function must be called under lru_lock, just before a page is added
1027 * to or just after a page is removed from an lru list (that ordering being
1028 * so as to allow it to check that lru_size 0 is consistent with list_empty).
Johannes Weiner925b7672012-01-12 17:18:15 -08001029 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001030void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1031 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001032{
1033 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001034 unsigned long *lru_size;
Hugh Dickinsca707232016-05-19 17:12:35 -07001035 long size;
1036 bool empty;
Johannes Weiner925b7672012-01-12 17:18:15 -08001037
Hugh Dickins9d5e6a92016-05-19 17:12:38 -07001038 __update_lru_size(lruvec, lru, nr_pages);
1039
Johannes Weiner925b7672012-01-12 17:18:15 -08001040 if (mem_cgroup_disabled())
1041 return;
1042
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001043 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1044 lru_size = mz->lru_size + lru;
Hugh Dickinsca707232016-05-19 17:12:35 -07001045 empty = list_empty(lruvec->lists + lru);
1046
1047 if (nr_pages < 0)
1048 *lru_size += nr_pages;
1049
1050 size = *lru_size;
1051 if (WARN_ONCE(size < 0 || empty != !size,
1052 "%s(%p, %d, %d): lru_size %ld but %sempty\n",
1053 __func__, lruvec, lru, nr_pages, size, empty ? "" : "not ")) {
1054 VM_BUG_ON(1);
1055 *lru_size = 0;
1056 }
1057
1058 if (nr_pages > 0)
1059 *lru_size += nr_pages;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001060}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001061
Johannes Weiner2314b422014-12-10 15:44:33 -08001062bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001063{
Johannes Weiner2314b422014-12-10 15:44:33 -08001064 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001065 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001066 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001067
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001068 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001069 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001070 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001071 task_unlock(p);
1072 } else {
1073 /*
1074 * All threads may have already detached their mm's, but the oom
1075 * killer still needs to detect if they have already been oom
1076 * killed to prevent needlessly killing additional tasks.
1077 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001078 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001079 task_memcg = mem_cgroup_from_task(task);
1080 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001081 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001082 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001083 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1084 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001085 return ret;
1086}
1087
Johannes Weiner19942822011-02-01 15:52:43 -08001088/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001089 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001090 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001091 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001092 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001093 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001094 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001095static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001096{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001097 unsigned long margin = 0;
1098 unsigned long count;
1099 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001100
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001101 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001102 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001103 if (count < limit)
1104 margin = limit - count;
1105
Johannes Weiner7941d212016-01-14 15:21:23 -08001106 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001107 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001108 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001109 if (count <= limit)
1110 margin = min(margin, limit - count);
1111 }
1112
1113 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001114}
1115
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001116/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001117 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001118 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001119 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1120 * moving cgroups. This is for waiting at high-memory pressure
1121 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001122 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001123static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001124{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001125 struct mem_cgroup *from;
1126 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001127 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001128 /*
1129 * Unlike task_move routines, we access mc.to, mc.from not under
1130 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1131 */
1132 spin_lock(&mc.lock);
1133 from = mc.from;
1134 to = mc.to;
1135 if (!from)
1136 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001137
Johannes Weiner2314b422014-12-10 15:44:33 -08001138 ret = mem_cgroup_is_descendant(from, memcg) ||
1139 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001140unlock:
1141 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001142 return ret;
1143}
1144
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001145static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001146{
1147 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001148 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001149 DEFINE_WAIT(wait);
1150 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1151 /* moving charge context might have finished. */
1152 if (mc.moving_task)
1153 schedule();
1154 finish_wait(&mc.waitq, &wait);
1155 return true;
1156 }
1157 }
1158 return false;
1159}
1160
Sha Zhengju58cf1882013-02-22 16:32:05 -08001161#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001162/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001163 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001164 * @memcg: The memory cgroup that went over limit
1165 * @p: Task that is going to be killed
1166 *
1167 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1168 * enabled
1169 */
1170void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1171{
Sha Zhengju58cf1882013-02-22 16:32:05 -08001172 struct mem_cgroup *iter;
1173 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001174
Balbir Singhe2224322009-04-02 16:57:39 -07001175 rcu_read_lock();
1176
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001177 if (p) {
1178 pr_info("Task in ");
1179 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1180 pr_cont(" killed as a result of limit of ");
1181 } else {
1182 pr_info("Memory limit reached of cgroup ");
1183 }
1184
Tejun Heoe61734c2014-02-12 09:29:50 -05001185 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001186 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001187
Balbir Singhe2224322009-04-02 16:57:39 -07001188 rcu_read_unlock();
1189
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001190 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1191 K((u64)page_counter_read(&memcg->memory)),
1192 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1193 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1194 K((u64)page_counter_read(&memcg->memsw)),
1195 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1196 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1197 K((u64)page_counter_read(&memcg->kmem)),
1198 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001199
1200 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001201 pr_info("Memory cgroup stats for ");
1202 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001203 pr_cont(":");
1204
1205 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08001206 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Sha Zhengju58cf1882013-02-22 16:32:05 -08001207 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07001208 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
Sha Zhengju58cf1882013-02-22 16:32:05 -08001209 K(mem_cgroup_read_stat(iter, i)));
1210 }
1211
1212 for (i = 0; i < NR_LRU_LISTS; i++)
1213 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1214 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1215
1216 pr_cont("\n");
1217 }
Balbir Singhe2224322009-04-02 16:57:39 -07001218}
1219
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001220/*
1221 * This function returns the number of memcg under hierarchy tree. Returns
1222 * 1(self count) if no children.
1223 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001224static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001225{
1226 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001227 struct mem_cgroup *iter;
1228
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001229 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001230 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001231 return num;
1232}
1233
Balbir Singh6d61ef42009-01-07 18:08:06 -08001234/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001235 * Return the memory (and swap, if configured) limit for a memcg.
1236 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001237static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001238{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001239 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001240
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001241 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001242 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001243 unsigned long memsw_limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001244 unsigned long swap_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001245
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001246 memsw_limit = memcg->memsw.limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001247 swap_limit = memcg->swap.limit;
1248 swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1249 limit = min(limit + swap_limit, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001250 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001251 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001252}
1253
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001254static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001255 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001256{
David Rientjes6e0fc462015-09-08 15:00:36 -07001257 struct oom_control oc = {
1258 .zonelist = NULL,
1259 .nodemask = NULL,
1260 .gfp_mask = gfp_mask,
1261 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001262 };
David Rientjes9cbb78b2012-07-31 16:43:44 -07001263 struct mem_cgroup *iter;
1264 unsigned long chosen_points = 0;
1265 unsigned long totalpages;
1266 unsigned int points = 0;
1267 struct task_struct *chosen = NULL;
1268
Johannes Weinerdc564012015-06-24 16:57:19 -07001269 mutex_lock(&oom_lock);
1270
David Rientjes876aafb2012-07-31 16:43:48 -07001271 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001272 * If current has a pending SIGKILL or is exiting, then automatically
1273 * select it. The goal is to allow it to allocate so that it may
1274 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001275 */
Oleg Nesterovd003f372014-12-12 16:56:24 -08001276 if (fatal_signal_pending(current) || task_will_free_mem(current)) {
Johannes Weiner16e95192015-06-24 16:57:07 -07001277 mark_oom_victim(current);
Michal Hocko3ef22df2016-05-19 17:13:12 -07001278 try_oom_reaper(current);
Johannes Weinerdc564012015-06-24 16:57:19 -07001279 goto unlock;
David Rientjes876aafb2012-07-31 16:43:48 -07001280 }
1281
David Rientjes6e0fc462015-09-08 15:00:36 -07001282 check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001283 totalpages = mem_cgroup_get_limit(memcg) ? : 1;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001284 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001285 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001286 struct task_struct *task;
1287
Tejun Heo72ec7022013-08-08 20:11:26 -04001288 css_task_iter_start(&iter->css, &it);
1289 while ((task = css_task_iter_next(&it))) {
David Rientjes6e0fc462015-09-08 15:00:36 -07001290 switch (oom_scan_process_thread(&oc, task, totalpages)) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001291 case OOM_SCAN_SELECT:
1292 if (chosen)
1293 put_task_struct(chosen);
1294 chosen = task;
1295 chosen_points = ULONG_MAX;
1296 get_task_struct(chosen);
1297 /* fall through */
1298 case OOM_SCAN_CONTINUE:
1299 continue;
1300 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001301 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001302 mem_cgroup_iter_break(memcg, iter);
1303 if (chosen)
1304 put_task_struct(chosen);
Johannes Weinerdc564012015-06-24 16:57:19 -07001305 goto unlock;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001306 case OOM_SCAN_OK:
1307 break;
1308 };
1309 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001310 if (!points || points < chosen_points)
1311 continue;
1312 /* Prefer thread group leaders for display purposes */
1313 if (points == chosen_points &&
1314 thread_group_leader(chosen))
1315 continue;
1316
1317 if (chosen)
1318 put_task_struct(chosen);
1319 chosen = task;
1320 chosen_points = points;
1321 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001322 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001323 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001324 }
1325
Johannes Weinerdc564012015-06-24 16:57:19 -07001326 if (chosen) {
1327 points = chosen_points * 1000 / totalpages;
David Rientjes6e0fc462015-09-08 15:00:36 -07001328 oom_kill_process(&oc, chosen, points, totalpages, memcg,
1329 "Memory cgroup out of memory");
Johannes Weinerdc564012015-06-24 16:57:19 -07001330 }
1331unlock:
1332 mutex_unlock(&oom_lock);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001333 return chosen;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001334}
1335
Michele Curtiae6e71d2014-12-12 16:56:35 -08001336#if MAX_NUMNODES > 1
1337
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001338/**
1339 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001340 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001341 * @nid: the node ID to be checked.
1342 * @noswap : specify true here if the user wants flle only information.
1343 *
1344 * This function returns whether the specified memcg contains any
1345 * reclaimable pages on a node. Returns true if there are any reclaimable
1346 * pages in the node.
1347 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001348static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001349 int nid, bool noswap)
1350{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001351 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001352 return true;
1353 if (noswap || !total_swap_pages)
1354 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001355 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001356 return true;
1357 return false;
1358
1359}
Ying Han889976d2011-05-26 16:25:33 -07001360
1361/*
1362 * Always updating the nodemask is not very good - even if we have an empty
1363 * list or the wrong list here, we can start from some node and traverse all
1364 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1365 *
1366 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001367static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001368{
1369 int nid;
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001370 /*
1371 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1372 * pagein/pageout changes since the last update.
1373 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001374 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001375 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001376 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001377 return;
1378
Ying Han889976d2011-05-26 16:25:33 -07001379 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001380 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001381
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001382 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001383
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001384 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1385 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001386 }
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001387
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001388 atomic_set(&memcg->numainfo_events, 0);
1389 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001390}
1391
1392/*
1393 * Selecting a node where we start reclaim from. Because what we need is just
1394 * reducing usage counter, start from anywhere is O,K. Considering
1395 * memory reclaim from current node, there are pros. and cons.
1396 *
1397 * Freeing memory from current node means freeing memory from a node which
1398 * we'll use or we've used. So, it may make LRU bad. And if several threads
1399 * hit limits, it will see a contention on a node. But freeing from remote
1400 * node means more costs for memory reclaim because of memory latency.
1401 *
1402 * Now, we use round-robin. Better algorithm is welcomed.
1403 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001404int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001405{
1406 int node;
1407
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001408 mem_cgroup_may_update_nodemask(memcg);
1409 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001410
Andrew Morton0edaf862016-05-19 17:10:58 -07001411 node = next_node_in(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001412 /*
Michal Hockofda3d692016-05-19 17:11:34 -07001413 * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1414 * last time it really checked all the LRUs due to rate limiting.
1415 * Fallback to the current node in that case for simplicity.
Ying Han889976d2011-05-26 16:25:33 -07001416 */
1417 if (unlikely(node == MAX_NUMNODES))
1418 node = numa_node_id();
1419
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001420 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001421 return node;
1422}
Ying Han889976d2011-05-26 16:25:33 -07001423#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001424int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001425{
1426 return 0;
1427}
1428#endif
1429
Andrew Morton0608f432013-09-24 15:27:41 -07001430static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1431 struct zone *zone,
1432 gfp_t gfp_mask,
1433 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001434{
Andrew Morton0608f432013-09-24 15:27:41 -07001435 struct mem_cgroup *victim = NULL;
1436 int total = 0;
1437 int loop = 0;
1438 unsigned long excess;
1439 unsigned long nr_scanned;
1440 struct mem_cgroup_reclaim_cookie reclaim = {
1441 .zone = zone,
1442 .priority = 0,
1443 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001444
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001445 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001446
Andrew Morton0608f432013-09-24 15:27:41 -07001447 while (1) {
1448 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1449 if (!victim) {
1450 loop++;
1451 if (loop >= 2) {
1452 /*
1453 * If we have not been able to reclaim
1454 * anything, it might because there are
1455 * no reclaimable pages under this hierarchy
1456 */
1457 if (!total)
1458 break;
1459 /*
1460 * We want to do more targeted reclaim.
1461 * excess >> 2 is not to excessive so as to
1462 * reclaim too much, nor too less that we keep
1463 * coming back to reclaim from this cgroup
1464 */
1465 if (total >= (excess >> 2) ||
1466 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1467 break;
1468 }
1469 continue;
1470 }
Andrew Morton0608f432013-09-24 15:27:41 -07001471 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1472 zone, &nr_scanned);
1473 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001474 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001475 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001476 }
Andrew Morton0608f432013-09-24 15:27:41 -07001477 mem_cgroup_iter_break(root_memcg, victim);
1478 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001479}
1480
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001481#ifdef CONFIG_LOCKDEP
1482static struct lockdep_map memcg_oom_lock_dep_map = {
1483 .name = "memcg_oom_lock",
1484};
1485#endif
1486
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001487static DEFINE_SPINLOCK(memcg_oom_lock);
1488
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001489/*
1490 * Check OOM-Killer is already running under our hierarchy.
1491 * If someone is running, return false.
1492 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001493static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001494{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001495 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001496
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001497 spin_lock(&memcg_oom_lock);
1498
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001499 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001500 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001501 /*
1502 * this subtree of our hierarchy is already locked
1503 * so we cannot give a lock.
1504 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001505 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001506 mem_cgroup_iter_break(memcg, iter);
1507 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001508 } else
1509 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001510 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001511
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001512 if (failed) {
1513 /*
1514 * OK, we failed to lock the whole subtree so we have
1515 * to clean up what we set up to the failing subtree
1516 */
1517 for_each_mem_cgroup_tree(iter, memcg) {
1518 if (iter == failed) {
1519 mem_cgroup_iter_break(memcg, iter);
1520 break;
1521 }
1522 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001523 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001524 } else
1525 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001526
1527 spin_unlock(&memcg_oom_lock);
1528
1529 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001530}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001531
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001532static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001533{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001534 struct mem_cgroup *iter;
1535
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001536 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001537 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001538 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001539 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001540 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001541}
1542
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001543static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001544{
1545 struct mem_cgroup *iter;
1546
Tejun Heoc2b42d32015-06-24 16:58:23 -07001547 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001548 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001549 iter->under_oom++;
1550 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001551}
1552
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001553static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001554{
1555 struct mem_cgroup *iter;
1556
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001557 /*
1558 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001559 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001560 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001561 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001562 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001563 if (iter->under_oom > 0)
1564 iter->under_oom--;
1565 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001566}
1567
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001568static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1569
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001570struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001571 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001572 wait_queue_t wait;
1573};
1574
1575static int memcg_oom_wake_function(wait_queue_t *wait,
1576 unsigned mode, int sync, void *arg)
1577{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001578 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1579 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001580 struct oom_wait_info *oom_wait_info;
1581
1582 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001583 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001584
Johannes Weiner2314b422014-12-10 15:44:33 -08001585 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1586 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001587 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001588 return autoremove_wake_function(wait, mode, sync, arg);
1589}
1590
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001591static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001592{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001593 /*
1594 * For the following lockless ->under_oom test, the only required
1595 * guarantee is that it must see the state asserted by an OOM when
1596 * this function is called as a result of userland actions
1597 * triggered by the notification of the OOM. This is trivially
1598 * achieved by invoking mem_cgroup_mark_under_oom() before
1599 * triggering notification.
1600 */
1601 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001602 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001603}
1604
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001605static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001606{
Tejun Heo626ebc42015-11-05 18:46:09 -08001607 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001608 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001609 /*
Johannes Weiner494264202013-10-16 13:46:59 -07001610 * We are in the middle of the charge context here, so we
1611 * don't want to block when potentially sitting on a callstack
1612 * that holds all kinds of filesystem and mm locks.
1613 *
1614 * Also, the caller may handle a failed allocation gracefully
1615 * (like optional page cache readahead) and so an OOM killer
1616 * invocation might not even be necessary.
1617 *
1618 * That's why we don't do anything here except remember the
1619 * OOM context and then deal with it at the end of the page
1620 * fault when the stack is unwound, the locks are released,
1621 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001622 */
Johannes Weiner494264202013-10-16 13:46:59 -07001623 css_get(&memcg->css);
Tejun Heo626ebc42015-11-05 18:46:09 -08001624 current->memcg_in_oom = memcg;
1625 current->memcg_oom_gfp_mask = mask;
1626 current->memcg_oom_order = order;
Johannes Weiner494264202013-10-16 13:46:59 -07001627}
1628
1629/**
1630 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1631 * @handle: actually kill/wait or just clean up the OOM state
1632 *
1633 * This has to be called at the end of a page fault if the memcg OOM
1634 * handler was enabled.
1635 *
1636 * Memcg supports userspace OOM handling where failed allocations must
1637 * sleep on a waitqueue until the userspace task resolves the
1638 * situation. Sleeping directly in the charge context with all kinds
1639 * of locks held is not a good idea, instead we remember an OOM state
1640 * in the task and mem_cgroup_oom_synchronize() has to be called at
1641 * the end of the page fault to complete the OOM handling.
1642 *
1643 * Returns %true if an ongoing memcg OOM situation was detected and
1644 * completed, %false otherwise.
1645 */
1646bool mem_cgroup_oom_synchronize(bool handle)
1647{
Tejun Heo626ebc42015-11-05 18:46:09 -08001648 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner494264202013-10-16 13:46:59 -07001649 struct oom_wait_info owait;
1650 bool locked;
1651
1652 /* OOM is global, do not handle */
1653 if (!memcg)
1654 return false;
1655
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001656 if (!handle || oom_killer_disabled)
Johannes Weiner494264202013-10-16 13:46:59 -07001657 goto cleanup;
1658
1659 owait.memcg = memcg;
1660 owait.wait.flags = 0;
1661 owait.wait.func = memcg_oom_wake_function;
1662 owait.wait.private = current;
1663 INIT_LIST_HEAD(&owait.wait.task_list);
1664
1665 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001666 mem_cgroup_mark_under_oom(memcg);
1667
1668 locked = mem_cgroup_oom_trylock(memcg);
1669
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001670 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001671 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001672
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001673 if (locked && !memcg->oom_kill_disable) {
1674 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner494264202013-10-16 13:46:59 -07001675 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001676 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1677 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001678 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001679 schedule();
Johannes Weiner494264202013-10-16 13:46:59 -07001680 mem_cgroup_unmark_under_oom(memcg);
1681 finish_wait(&memcg_oom_waitq, &owait.wait);
1682 }
1683
1684 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001685 mem_cgroup_oom_unlock(memcg);
1686 /*
1687 * There is no guarantee that an OOM-lock contender
1688 * sees the wakeups triggered by the OOM kill
1689 * uncharges. Wake any sleepers explicitely.
1690 */
1691 memcg_oom_recover(memcg);
1692 }
Johannes Weiner494264202013-10-16 13:46:59 -07001693cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001694 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001695 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001696 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001697}
1698
Johannes Weinerd7365e72014-10-29 14:50:48 -07001699/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001700 * lock_page_memcg - lock a page->mem_cgroup binding
1701 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001702 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001703 * This function protects unlocked LRU pages from being moved to
1704 * another cgroup and stabilizes their page->mem_cgroup binding.
Balbir Singhd69b0422009-06-17 16:26:34 -07001705 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001706void lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001707{
1708 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001709 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001710
Johannes Weiner6de22612015-02-11 15:25:01 -08001711 /*
1712 * The RCU lock is held throughout the transaction. The fast
1713 * path can get away without acquiring the memcg->move_lock
1714 * because page moving starts with an RCU grace period.
Johannes Weiner6de22612015-02-11 15:25:01 -08001715 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001716 rcu_read_lock();
1717
1718 if (mem_cgroup_disabled())
Johannes Weiner62cccb82016-03-15 14:57:22 -07001719 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001720again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001721 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001722 if (unlikely(!memcg))
Johannes Weiner62cccb82016-03-15 14:57:22 -07001723 return;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001724
Qiang Huangbdcbb652014-06-04 16:08:21 -07001725 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001726 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001727
Johannes Weiner6de22612015-02-11 15:25:01 -08001728 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001729 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001730 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001731 goto again;
1732 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001733
1734 /*
1735 * When charge migration first begins, we can have locked and
1736 * unlocked page stat updates happening concurrently. Track
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001737 * the task who has the lock for unlock_page_memcg().
Johannes Weiner6de22612015-02-11 15:25:01 -08001738 */
1739 memcg->move_lock_task = current;
1740 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001741
Johannes Weiner62cccb82016-03-15 14:57:22 -07001742 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001743}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001744EXPORT_SYMBOL(lock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001745
Johannes Weinerd7365e72014-10-29 14:50:48 -07001746/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001747 * unlock_page_memcg - unlock a page->mem_cgroup binding
Johannes Weiner62cccb82016-03-15 14:57:22 -07001748 * @page: the page
Johannes Weinerd7365e72014-10-29 14:50:48 -07001749 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001750void unlock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001751{
Johannes Weiner62cccb82016-03-15 14:57:22 -07001752 struct mem_cgroup *memcg = page->mem_cgroup;
1753
Johannes Weiner6de22612015-02-11 15:25:01 -08001754 if (memcg && memcg->move_lock_task == current) {
1755 unsigned long flags = memcg->move_lock_flags;
1756
1757 memcg->move_lock_task = NULL;
1758 memcg->move_lock_flags = 0;
1759
1760 spin_unlock_irqrestore(&memcg->move_lock, flags);
1761 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001762
Johannes Weinerd7365e72014-10-29 14:50:48 -07001763 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001764}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001765EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001766
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001767/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001768 * size of first charge trial. "32" comes from vmscan.c's magic value.
1769 * TODO: maybe necessary to use big numbers in big irons.
1770 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001771#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001772struct memcg_stock_pcp {
1773 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001774 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001775 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001776 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001777#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001778};
1779static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001780static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001781
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001782/**
1783 * consume_stock: Try to consume stocked charge on this cpu.
1784 * @memcg: memcg to consume from.
1785 * @nr_pages: how many pages to charge.
1786 *
1787 * The charges will only happen if @memcg matches the current cpu's memcg
1788 * stock, and at least @nr_pages are available in that stock. Failure to
1789 * service an allocation will refill the stock.
1790 *
1791 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001792 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001793static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001794{
1795 struct memcg_stock_pcp *stock;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001796 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001797
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001798 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001799 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001800
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001801 stock = &get_cpu_var(memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001802 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001803 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001804 ret = true;
1805 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001806 put_cpu_var(memcg_stock);
1807 return ret;
1808}
1809
1810/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001811 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001812 */
1813static void drain_stock(struct memcg_stock_pcp *stock)
1814{
1815 struct mem_cgroup *old = stock->cached;
1816
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001817 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001818 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001819 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001820 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001821 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001822 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001823 }
1824 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001825}
1826
1827/*
1828 * This must be called under preempt disabled or must be called by
1829 * a thread which is pinned to local cpu.
1830 */
1831static void drain_local_stock(struct work_struct *dummy)
1832{
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001833 struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001834 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001835 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001836}
1837
1838/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001839 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001840 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001841 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001842static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001843{
1844 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1845
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001846 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001847 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001848 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001849 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001850 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001851 put_cpu_var(memcg_stock);
1852}
1853
1854/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001855 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001856 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001857 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001858static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001859{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001860 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001861
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001862 /* If someone's already draining, avoid adding running more workers. */
1863 if (!mutex_trylock(&percpu_charge_mutex))
1864 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001865 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001866 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001867 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001868 for_each_online_cpu(cpu) {
1869 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001870 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001871
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001872 memcg = stock->cached;
1873 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001874 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001875 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001876 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001877 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1878 if (cpu == curcpu)
1879 drain_local_stock(&stock->work);
1880 else
1881 schedule_work_on(cpu, &stock->work);
1882 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001883 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001884 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001885 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001886 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001887}
1888
Paul Gortmaker0db06282013-06-19 14:53:51 -04001889static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001890 unsigned long action,
1891 void *hcpu)
1892{
1893 int cpu = (unsigned long)hcpu;
1894 struct memcg_stock_pcp *stock;
1895
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001896 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001897 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001898
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07001899 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001900 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001901
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001902 stock = &per_cpu(memcg_stock, cpu);
1903 drain_stock(stock);
1904 return NOTIFY_OK;
1905}
1906
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001907static void reclaim_high(struct mem_cgroup *memcg,
1908 unsigned int nr_pages,
1909 gfp_t gfp_mask)
1910{
1911 do {
1912 if (page_counter_read(&memcg->memory) <= memcg->high)
1913 continue;
1914 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
1915 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1916 } while ((memcg = parent_mem_cgroup(memcg)));
1917}
1918
1919static void high_work_func(struct work_struct *work)
1920{
1921 struct mem_cgroup *memcg;
1922
1923 memcg = container_of(work, struct mem_cgroup, high_work);
1924 reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1925}
1926
Tejun Heob23afb92015-11-05 18:46:11 -08001927/*
1928 * Scheduled by try_charge() to be executed from the userland return path
1929 * and reclaims memory over the high limit.
1930 */
1931void mem_cgroup_handle_over_high(void)
1932{
1933 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001934 struct mem_cgroup *memcg;
Tejun Heob23afb92015-11-05 18:46:11 -08001935
1936 if (likely(!nr_pages))
1937 return;
1938
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001939 memcg = get_mem_cgroup_from_mm(current->mm);
1940 reclaim_high(memcg, nr_pages, GFP_KERNEL);
Tejun Heob23afb92015-11-05 18:46:11 -08001941 css_put(&memcg->css);
1942 current->memcg_nr_pages_over_high = 0;
1943}
1944
Johannes Weiner00501b52014-08-08 14:19:20 -07001945static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1946 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001947{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001948 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001949 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001950 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001951 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001952 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001953 bool may_swap = true;
1954 bool drained = false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001955
Johannes Weinerce00a962014-09-05 08:43:57 -04001956 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08001957 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001958retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001959 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08001960 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001961
Johannes Weiner7941d212016-01-14 15:21:23 -08001962 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08001963 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1964 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001965 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08001966 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001967 page_counter_uncharge(&memcg->memsw, batch);
1968 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001969 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001970 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001971 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001972 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001973
Johannes Weiner6539cc02014-08-06 16:05:42 -07001974 if (batch > nr_pages) {
1975 batch = nr_pages;
1976 goto retry;
1977 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001978
Johannes Weiner06b078fc02014-08-06 16:05:44 -07001979 /*
1980 * Unlike in global OOM situations, memcg is not in a physical
1981 * memory shortage. Allow dying and OOM-killed tasks to
1982 * bypass the last charges so that they can exit quickly and
1983 * free their memory.
1984 */
1985 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1986 fatal_signal_pending(current) ||
1987 current->flags & PF_EXITING))
Tejun Heo10d53c72015-11-05 18:46:17 -08001988 goto force;
Johannes Weiner06b078fc02014-08-06 16:05:44 -07001989
1990 if (unlikely(task_in_memcg_oom(current)))
1991 goto nomem;
1992
Mel Gormand0164ad2015-11-06 16:28:21 -08001993 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001994 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001995
Johannes Weiner241994ed2015-02-11 15:26:06 -08001996 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
1997
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001998 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1999 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002000
Johannes Weiner61e02c72014-08-06 16:08:16 -07002001 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07002002 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07002003
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002004 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002005 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002006 drained = true;
2007 goto retry;
2008 }
2009
Johannes Weiner28c34c22014-08-06 16:05:47 -07002010 if (gfp_mask & __GFP_NORETRY)
2011 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002012 /*
2013 * Even though the limit is exceeded at this point, reclaim
2014 * may have been able to free some pages. Retry the charge
2015 * before killing the task.
2016 *
2017 * Only for regular pages, though: huge pages are rather
2018 * unlikely to succeed so close to the limit, and we fall back
2019 * to regular pages anyway in case of failure.
2020 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002021 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002022 goto retry;
2023 /*
2024 * At task move, charge accounts can be doubly counted. So, it's
2025 * better to wait until the end of task_move if something is going on.
2026 */
2027 if (mem_cgroup_wait_acct_move(mem_over_limit))
2028 goto retry;
2029
Johannes Weiner9b130612014-08-06 16:05:51 -07002030 if (nr_retries--)
2031 goto retry;
2032
Johannes Weiner06b078fc02014-08-06 16:05:44 -07002033 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08002034 goto force;
Johannes Weiner06b078fc02014-08-06 16:05:44 -07002035
Johannes Weiner6539cc02014-08-06 16:05:42 -07002036 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08002037 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002038
Johannes Weiner241994ed2015-02-11 15:26:06 -08002039 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
2040
Jerome Marchand3608de02015-11-05 18:47:29 -08002041 mem_cgroup_oom(mem_over_limit, gfp_mask,
2042 get_order(nr_pages * PAGE_SIZE));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002043nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002044 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002045 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08002046force:
2047 /*
2048 * The allocation either can't fail or will lead to more memory
2049 * being freed very soon. Allow memory usage go over the limit
2050 * temporarily by force charging it.
2051 */
2052 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002053 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08002054 page_counter_charge(&memcg->memsw, nr_pages);
2055 css_get_many(&memcg->css, nr_pages);
2056
2057 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002058
2059done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002060 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002061 if (batch > nr_pages)
2062 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002063
Johannes Weiner241994ed2015-02-11 15:26:06 -08002064 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002065 * If the hierarchy is above the normal consumption range, schedule
2066 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002067 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002068 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2069 * not recorded as it most likely matches current's and won't
2070 * change in the meantime. As high limit is checked again before
2071 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994ed2015-02-11 15:26:06 -08002072 */
2073 do {
Tejun Heob23afb92015-11-05 18:46:11 -08002074 if (page_counter_read(&memcg->memory) > memcg->high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002075 /* Don't bother a random interrupted task */
2076 if (in_interrupt()) {
2077 schedule_work(&memcg->high_work);
2078 break;
2079 }
Vladimir Davydov9516a182015-12-11 13:40:24 -08002080 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002081 set_notify_resume(current);
2082 break;
2083 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08002084 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002085
2086 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002087}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002088
Johannes Weiner00501b52014-08-08 14:19:20 -07002089static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002090{
Johannes Weinerce00a962014-09-05 08:43:57 -04002091 if (mem_cgroup_is_root(memcg))
2092 return;
2093
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002094 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002095 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002096 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002097
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002098 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002099}
2100
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002101static void lock_page_lru(struct page *page, int *isolated)
2102{
2103 struct zone *zone = page_zone(page);
2104
2105 spin_lock_irq(&zone->lru_lock);
2106 if (PageLRU(page)) {
2107 struct lruvec *lruvec;
2108
2109 lruvec = mem_cgroup_page_lruvec(page, zone);
2110 ClearPageLRU(page);
2111 del_page_from_lru_list(page, lruvec, page_lru(page));
2112 *isolated = 1;
2113 } else
2114 *isolated = 0;
2115}
2116
2117static void unlock_page_lru(struct page *page, int isolated)
2118{
2119 struct zone *zone = page_zone(page);
2120
2121 if (isolated) {
2122 struct lruvec *lruvec;
2123
2124 lruvec = mem_cgroup_page_lruvec(page, zone);
2125 VM_BUG_ON_PAGE(PageLRU(page), page);
2126 SetPageLRU(page);
2127 add_page_to_lru_list(page, lruvec, page_lru(page));
2128 }
2129 spin_unlock_irq(&zone->lru_lock);
2130}
2131
Johannes Weiner00501b52014-08-08 14:19:20 -07002132static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002133 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002134{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002135 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002136
Johannes Weiner1306a852014-12-10 15:44:52 -08002137 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002138
2139 /*
2140 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2141 * may already be on some other mem_cgroup's LRU. Take care of it.
2142 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002143 if (lrucare)
2144 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002145
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002146 /*
2147 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002148 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002149 *
2150 * - the page is uncharged
2151 *
2152 * - the page is off-LRU
2153 *
2154 * - an anonymous fault has exclusive page access, except for
2155 * a locked page table
2156 *
2157 * - a page cache insertion, a swapin fault, or a migration
2158 * have the page locked
2159 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002160 page->mem_cgroup = memcg;
Hugh Dickins3be912772008-02-07 00:14:19 -08002161
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002162 if (lrucare)
2163 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002164}
2165
Johannes Weiner127424c2016-01-20 15:02:32 -08002166#ifndef CONFIG_SLOB
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002167static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002168{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002169 int id, size;
2170 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002171
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002172 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002173 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2174 if (id < 0)
2175 return id;
2176
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002177 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002178 return id;
2179
2180 /*
2181 * There's no space for the new id in memcg_caches arrays,
2182 * so we have to grow them.
2183 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002184 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002185
2186 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002187 if (size < MEMCG_CACHES_MIN_SIZE)
2188 size = MEMCG_CACHES_MIN_SIZE;
2189 else if (size > MEMCG_CACHES_MAX_SIZE)
2190 size = MEMCG_CACHES_MAX_SIZE;
2191
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002192 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002193 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002194 err = memcg_update_all_list_lrus(size);
2195 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002196 memcg_nr_cache_ids = size;
2197
2198 up_write(&memcg_cache_ids_sem);
2199
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002200 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002201 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002202 return err;
2203 }
2204 return id;
2205}
2206
2207static void memcg_free_cache_id(int id)
2208{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002209 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002210}
2211
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002212struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002213 struct mem_cgroup *memcg;
2214 struct kmem_cache *cachep;
2215 struct work_struct work;
2216};
2217
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002218static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002219{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002220 struct memcg_kmem_cache_create_work *cw =
2221 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002222 struct mem_cgroup *memcg = cw->memcg;
2223 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002224
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002225 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002226
Vladimir Davydov5722d092014-04-07 15:39:24 -07002227 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002228 kfree(cw);
2229}
2230
2231/*
2232 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002233 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002234static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2235 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002236{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002237 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002238
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002239 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002240 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002241 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002242
2243 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002244
2245 cw->memcg = memcg;
2246 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002247 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002248
Glauber Costad7f25f82012-12-18 14:22:40 -08002249 schedule_work(&cw->work);
2250}
2251
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002252static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2253 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002254{
2255 /*
2256 * We need to stop accounting when we kmalloc, because if the
2257 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002258 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002259 *
2260 * However, it is better to enclose the whole function. Depending on
2261 * the debugging options enabled, INIT_WORK(), for instance, can
2262 * trigger an allocation. This too, will make us recurse. Because at
2263 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2264 * the safest choice is to do it like this, wrapping the whole function.
2265 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002266 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002267 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002268 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002269}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002270
Glauber Costad7f25f82012-12-18 14:22:40 -08002271/*
2272 * Return the kmem_cache we're supposed to use for a slab allocation.
2273 * We try to use the current memcg's version of the cache.
2274 *
2275 * If the cache does not exist yet, if we are the first user of it,
2276 * we either create it immediately, if possible, or create it asynchronously
2277 * in a workqueue.
2278 * In the latter case, we will let the current allocation go through with
2279 * the original cache.
2280 *
2281 * Can't be called in interrupt context or from kernel threads.
2282 * This function needs to be called with rcu_read_lock() held.
2283 */
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002284struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
Glauber Costad7f25f82012-12-18 14:22:40 -08002285{
2286 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002287 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002288 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002289
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002290 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002291
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002292 if (cachep->flags & SLAB_ACCOUNT)
2293 gfp |= __GFP_ACCOUNT;
2294
2295 if (!(gfp & __GFP_ACCOUNT))
2296 return cachep;
2297
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002298 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002299 return cachep;
2300
Vladimir Davydov8135be52014-12-12 16:56:38 -08002301 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002302 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002303 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002304 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002305
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002306 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002307 if (likely(memcg_cachep))
2308 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002309
2310 /*
2311 * If we are in a safe context (can wait, and not in interrupt
2312 * context), we could be be predictable and return right away.
2313 * This would guarantee that the allocation being performed
2314 * already belongs in the new cache.
2315 *
2316 * However, there are some clashes that can arrive from locking.
2317 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002318 * memcg_create_kmem_cache, this means no further allocation
2319 * could happen with the slab_mutex held. So it's better to
2320 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002321 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002322 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002323out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002324 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002325 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002326}
Glauber Costad7f25f82012-12-18 14:22:40 -08002327
Vladimir Davydov8135be52014-12-12 16:56:38 -08002328void __memcg_kmem_put_cache(struct kmem_cache *cachep)
2329{
2330 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002331 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002332}
2333
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002334int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2335 struct mem_cgroup *memcg)
2336{
2337 unsigned int nr_pages = 1 << order;
2338 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002339 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002340
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002341 ret = try_charge(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08002342 if (ret)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002343 return ret;
Johannes Weiner52c29b02016-01-20 15:02:35 -08002344
2345 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2346 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2347 cancel_charge(memcg, nr_pages);
2348 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002349 }
2350
2351 page->mem_cgroup = memcg;
2352
2353 return 0;
2354}
2355
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002356int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002357{
2358 struct mem_cgroup *memcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002359 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002360
Johannes Weinerdf381972014-04-07 15:37:43 -07002361 memcg = get_mem_cgroup_from_mm(current->mm);
Vladimir Davydovb6ecd2d2016-03-17 14:18:33 -07002362 if (!mem_cgroup_is_root(memcg))
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002363 ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002364 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002365 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002366}
2367
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002368void __memcg_kmem_uncharge(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002369{
Johannes Weiner1306a852014-12-10 15:44:52 -08002370 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002371 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002372
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002373 if (!memcg)
2374 return;
2375
Sasha Levin309381fea2014-01-23 15:52:54 -08002376 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002377
Johannes Weiner52c29b02016-01-20 15:02:35 -08002378 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2379 page_counter_uncharge(&memcg->kmem, nr_pages);
2380
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002381 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002382 if (do_memsw_account())
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002383 page_counter_uncharge(&memcg->memsw, nr_pages);
2384
Johannes Weiner1306a852014-12-10 15:44:52 -08002385 page->mem_cgroup = NULL;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002386 css_put_many(&memcg->css, nr_pages);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002387}
Johannes Weiner127424c2016-01-20 15:02:32 -08002388#endif /* !CONFIG_SLOB */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002389
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002390#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2391
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002392/*
2393 * Because tail pages are not marked as "used", set it. We're under
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08002394 * zone->lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002395 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002396void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002397{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002398 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002399
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002400 if (mem_cgroup_disabled())
2401 return;
David Rientjesb070e652013-05-07 16:18:09 -07002402
Johannes Weiner29833312014-12-10 15:44:02 -08002403 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002404 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002405
Johannes Weiner1306a852014-12-10 15:44:52 -08002406 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002407 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002408}
Hugh Dickins12d27102012-01-12 17:19:52 -08002409#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002410
Andrew Mortonc255a452012-07-31 16:43:02 -07002411#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002412static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2413 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002414{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002415 int val = (charge) ? 1 : -1;
2416 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002417}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002418
2419/**
2420 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2421 * @entry: swap entry to be moved
2422 * @from: mem_cgroup which the entry is moved from
2423 * @to: mem_cgroup which the entry is moved to
2424 *
2425 * It succeeds only when the swap_cgroup's record for this entry is the same
2426 * as the mem_cgroup's id of @from.
2427 *
2428 * Returns 0 on success, -EINVAL on failure.
2429 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002430 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002431 * both res and memsw, and called css_get().
2432 */
2433static 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 unsigned short old_id, new_id;
2437
Li Zefan34c00c32013-09-23 16:56:01 +08002438 old_id = mem_cgroup_id(from);
2439 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002440
2441 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002442 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002443 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002444 return 0;
2445 }
2446 return -EINVAL;
2447}
2448#else
2449static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002450 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002451{
2452 return -EINVAL;
2453}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002454#endif
2455
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002456static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002457
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002458static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002459 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002460{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002461 unsigned long curusage;
2462 unsigned long oldusage;
2463 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002464 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002465 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002466
2467 /*
2468 * For keeping hierarchical_reclaim simple, how long we should retry
2469 * is depends on callers. We set our retry-count to be function
2470 * of # of children which we should visit in this loop.
2471 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002472 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2473 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002474
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002475 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002476
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002477 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002478 if (signal_pending(current)) {
2479 ret = -EINTR;
2480 break;
2481 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002482
2483 mutex_lock(&memcg_limit_mutex);
2484 if (limit > memcg->memsw.limit) {
2485 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002486 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002487 break;
2488 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002489 if (limit > memcg->memory.limit)
2490 enlarge = true;
2491 ret = page_counter_limit(&memcg->memory, limit);
2492 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002493
2494 if (!ret)
2495 break;
2496
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002497 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2498
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002499 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002500 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002501 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002502 retry_count--;
2503 else
2504 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002505 } while (retry_count);
2506
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002507 if (!ret && enlarge)
2508 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002509
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002510 return ret;
2511}
2512
Li Zefan338c8432009-06-17 16:27:15 -07002513static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002514 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002515{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002516 unsigned long curusage;
2517 unsigned long oldusage;
2518 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002519 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002520 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002521
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002522 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002523 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2524 mem_cgroup_count_children(memcg);
2525
2526 oldusage = page_counter_read(&memcg->memsw);
2527
2528 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002529 if (signal_pending(current)) {
2530 ret = -EINTR;
2531 break;
2532 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002533
2534 mutex_lock(&memcg_limit_mutex);
2535 if (limit < memcg->memory.limit) {
2536 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002537 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002538 break;
2539 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002540 if (limit > memcg->memsw.limit)
2541 enlarge = true;
2542 ret = page_counter_limit(&memcg->memsw, limit);
2543 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002544
2545 if (!ret)
2546 break;
2547
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002548 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2549
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002550 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002551 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002552 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002553 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002554 else
2555 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002556 } while (retry_count);
2557
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002558 if (!ret && enlarge)
2559 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002560
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002561 return ret;
2562}
2563
Andrew Morton0608f432013-09-24 15:27:41 -07002564unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2565 gfp_t gfp_mask,
2566 unsigned long *total_scanned)
2567{
2568 unsigned long nr_reclaimed = 0;
2569 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2570 unsigned long reclaimed;
2571 int loop = 0;
2572 struct mem_cgroup_tree_per_zone *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002573 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002574 unsigned long nr_scanned;
2575
2576 if (order > 0)
2577 return 0;
2578
2579 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
2580 /*
2581 * This loop can run a while, specially if mem_cgroup's continuously
2582 * keep exceeding their soft limit and putting the system under
2583 * pressure
2584 */
2585 do {
2586 if (next_mz)
2587 mz = next_mz;
2588 else
2589 mz = mem_cgroup_largest_soft_limit_node(mctz);
2590 if (!mz)
2591 break;
2592
2593 nr_scanned = 0;
2594 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
2595 gfp_mask, &nr_scanned);
2596 nr_reclaimed += reclaimed;
2597 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002598 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002599 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002600
2601 /*
2602 * If we failed to reclaim anything from this memory cgroup
2603 * it is time to move on to the next cgroup
2604 */
2605 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002606 if (!reclaimed)
2607 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2608
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002609 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002610 /*
2611 * One school of thought says that we should not add
2612 * back the node to the tree if reclaim returns 0.
2613 * But our reclaim could return 0, simply because due
2614 * to priority we are exposing a smaller subset of
2615 * memory to reclaim from. Consider this as a longer
2616 * term TODO.
2617 */
2618 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002619 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002620 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002621 css_put(&mz->memcg->css);
2622 loop++;
2623 /*
2624 * Could not reclaim anything and there are no more
2625 * mem cgroups to try or we seem to be looping without
2626 * reclaiming anything.
2627 */
2628 if (!nr_reclaimed &&
2629 (next_mz == NULL ||
2630 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2631 break;
2632 } while (!nr_reclaimed);
2633 if (next_mz)
2634 css_put(&next_mz->memcg->css);
2635 return nr_reclaimed;
2636}
2637
Tejun Heoea280e72014-05-16 13:22:48 -04002638/*
2639 * Test whether @memcg has children, dead or alive. Note that this
2640 * function doesn't care whether @memcg has use_hierarchy enabled and
2641 * returns %true if there are child csses according to the cgroup
2642 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2643 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002644static inline bool memcg_has_children(struct mem_cgroup *memcg)
2645{
Tejun Heoea280e72014-05-16 13:22:48 -04002646 bool ret;
2647
Tejun Heoea280e72014-05-16 13:22:48 -04002648 rcu_read_lock();
2649 ret = css_next_child(NULL, &memcg->css);
2650 rcu_read_unlock();
2651 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002652}
2653
2654/*
Greg Thelen51038172016-05-20 16:58:18 -07002655 * Reclaims as many pages from the given memcg as possible.
Michal Hockoc26251f2012-10-26 13:37:28 +02002656 *
2657 * Caller is responsible for holding css reference for memcg.
2658 */
2659static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2660{
2661 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002662
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002663 /* we call try-to-free pages for make this cgroup empty */
2664 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002665 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002666 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002667 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002668
Michal Hockoc26251f2012-10-26 13:37:28 +02002669 if (signal_pending(current))
2670 return -EINTR;
2671
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002672 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2673 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002674 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002675 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002676 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002677 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002678 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002679
2680 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002681
2682 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002683}
2684
Tejun Heo6770c642014-05-13 12:16:21 -04002685static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2686 char *buf, size_t nbytes,
2687 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002688{
Tejun Heo6770c642014-05-13 12:16:21 -04002689 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002690
Michal Hockod8423012012-10-26 13:37:29 +02002691 if (mem_cgroup_is_root(memcg))
2692 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002693 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002694}
2695
Tejun Heo182446d2013-08-08 20:11:24 -04002696static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2697 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002698{
Tejun Heo182446d2013-08-08 20:11:24 -04002699 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002700}
2701
Tejun Heo182446d2013-08-08 20:11:24 -04002702static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2703 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002704{
2705 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002706 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002707 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002708
Glauber Costa567fb432012-07-31 16:43:07 -07002709 if (memcg->use_hierarchy == val)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002710 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07002711
Balbir Singh18f59ea2009-01-07 18:08:07 -08002712 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002713 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002714 * in the child subtrees. If it is unset, then the change can
2715 * occur, provided the current cgroup has no children.
2716 *
2717 * For the root cgroup, parent_mem is NULL, we allow value to be
2718 * set if there are no children.
2719 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002720 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002721 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002722 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002723 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002724 else
2725 retval = -EBUSY;
2726 } else
2727 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002728
Balbir Singh18f59ea2009-01-07 18:08:07 -08002729 return retval;
2730}
2731
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002732static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
Johannes Weinerce00a962014-09-05 08:43:57 -04002733{
2734 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002735 int i;
Johannes Weinerce00a962014-09-05 08:43:57 -04002736
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002737 memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
Johannes Weinerce00a962014-09-05 08:43:57 -04002738
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002739 for_each_mem_cgroup_tree(iter, memcg) {
2740 for (i = 0; i < MEMCG_NR_STAT; i++)
2741 stat[i] += mem_cgroup_read_stat(iter, i);
2742 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002743}
2744
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002745static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
Johannes Weiner587d9f72016-01-20 15:03:19 -08002746{
2747 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002748 int i;
Johannes Weiner587d9f72016-01-20 15:03:19 -08002749
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002750 memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
Johannes Weiner587d9f72016-01-20 15:03:19 -08002751
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002752 for_each_mem_cgroup_tree(iter, memcg) {
2753 for (i = 0; i < MEMCG_NR_EVENTS; i++)
2754 events[i] += mem_cgroup_read_events(iter, i);
2755 }
Johannes Weiner587d9f72016-01-20 15:03:19 -08002756}
2757
Andrew Morton6f646152015-11-06 16:28:58 -08002758static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04002759{
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002760 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002761
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002762 if (mem_cgroup_is_root(memcg)) {
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002763 struct mem_cgroup *iter;
2764
2765 for_each_mem_cgroup_tree(iter, memcg) {
2766 val += mem_cgroup_read_stat(iter,
2767 MEM_CGROUP_STAT_CACHE);
2768 val += mem_cgroup_read_stat(iter,
2769 MEM_CGROUP_STAT_RSS);
2770 if (swap)
2771 val += mem_cgroup_read_stat(iter,
2772 MEM_CGROUP_STAT_SWAP);
2773 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002774 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002775 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002776 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002777 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002778 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002779 }
Michal Hockoc12176d2015-11-05 18:50:29 -08002780 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002781}
2782
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002783enum {
2784 RES_USAGE,
2785 RES_LIMIT,
2786 RES_MAX_USAGE,
2787 RES_FAILCNT,
2788 RES_SOFT_LIMIT,
2789};
Johannes Weinerce00a962014-09-05 08:43:57 -04002790
Tejun Heo791badb2013-12-05 12:28:02 -05002791static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002792 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002793{
Tejun Heo182446d2013-08-08 20:11:24 -04002794 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002795 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002796
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002797 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002798 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002799 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002800 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002801 case _MEMSWAP:
2802 counter = &memcg->memsw;
2803 break;
2804 case _KMEM:
2805 counter = &memcg->kmem;
2806 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002807 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08002808 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002809 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002810 default:
2811 BUG();
2812 }
2813
2814 switch (MEMFILE_ATTR(cft->private)) {
2815 case RES_USAGE:
2816 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08002817 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002818 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08002819 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002820 return (u64)page_counter_read(counter) * PAGE_SIZE;
2821 case RES_LIMIT:
2822 return (u64)counter->limit * PAGE_SIZE;
2823 case RES_MAX_USAGE:
2824 return (u64)counter->watermark * PAGE_SIZE;
2825 case RES_FAILCNT:
2826 return counter->failcnt;
2827 case RES_SOFT_LIMIT:
2828 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002829 default:
2830 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002831 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002832}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002833
Johannes Weiner127424c2016-01-20 15:02:32 -08002834#ifndef CONFIG_SLOB
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002835static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002836{
Vladimir Davydovd6441632014-01-23 15:53:09 -08002837 int memcg_id;
2838
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002839 if (cgroup_memory_nokmem)
2840 return 0;
2841
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002842 BUG_ON(memcg->kmemcg_id >= 0);
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002843 BUG_ON(memcg->kmem_state);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002844
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002845 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002846 if (memcg_id < 0)
2847 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002848
Johannes Weineref129472016-01-14 15:21:34 -08002849 static_branch_inc(&memcg_kmem_enabled_key);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002850 /*
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002851 * A memory cgroup is considered kmem-online as soon as it gets
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002852 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002853 * guarantee no one starts accounting before all call sites are
2854 * patched.
2855 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002856 memcg->kmemcg_id = memcg_id;
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002857 memcg->kmem_state = KMEM_ONLINE;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002858
2859 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002860}
2861
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002862static void memcg_offline_kmem(struct mem_cgroup *memcg)
2863{
2864 struct cgroup_subsys_state *css;
2865 struct mem_cgroup *parent, *child;
2866 int kmemcg_id;
2867
2868 if (memcg->kmem_state != KMEM_ONLINE)
2869 return;
2870 /*
2871 * Clear the online state before clearing memcg_caches array
2872 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2873 * guarantees that no cache will be created for this cgroup
2874 * after we are done (see memcg_create_kmem_cache()).
2875 */
2876 memcg->kmem_state = KMEM_ALLOCATED;
2877
2878 memcg_deactivate_kmem_caches(memcg);
2879
2880 kmemcg_id = memcg->kmemcg_id;
2881 BUG_ON(kmemcg_id < 0);
2882
2883 parent = parent_mem_cgroup(memcg);
2884 if (!parent)
2885 parent = root_mem_cgroup;
2886
2887 /*
2888 * Change kmemcg_id of this cgroup and all its descendants to the
2889 * parent's id, and then move all entries from this cgroup's list_lrus
2890 * to ones of the parent. After we have finished, all list_lrus
2891 * corresponding to this cgroup are guaranteed to remain empty. The
2892 * ordering is imposed by list_lru_node->lock taken by
2893 * memcg_drain_all_list_lrus().
2894 */
2895 css_for_each_descendant_pre(css, &memcg->css) {
2896 child = mem_cgroup_from_css(css);
2897 BUG_ON(child->kmemcg_id != kmemcg_id);
2898 child->kmemcg_id = parent->kmemcg_id;
2899 if (!memcg->use_hierarchy)
2900 break;
2901 }
2902 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2903
2904 memcg_free_cache_id(kmemcg_id);
2905}
2906
2907static void memcg_free_kmem(struct mem_cgroup *memcg)
2908{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002909 /* css_alloc() failed, offlining didn't happen */
2910 if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2911 memcg_offline_kmem(memcg);
2912
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002913 if (memcg->kmem_state == KMEM_ALLOCATED) {
2914 memcg_destroy_kmem_caches(memcg);
2915 static_branch_dec(&memcg_kmem_enabled_key);
2916 WARN_ON(page_counter_read(&memcg->kmem));
2917 }
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002918}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002919#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002920static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08002921{
2922 return 0;
2923}
2924static void memcg_offline_kmem(struct mem_cgroup *memcg)
2925{
2926}
2927static void memcg_free_kmem(struct mem_cgroup *memcg)
2928{
2929}
2930#endif /* !CONFIG_SLOB */
2931
Johannes Weiner127424c2016-01-20 15:02:32 -08002932static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2933 unsigned long limit)
2934{
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002935 int ret;
Johannes Weiner127424c2016-01-20 15:02:32 -08002936
2937 mutex_lock(&memcg_limit_mutex);
Johannes Weiner127424c2016-01-20 15:02:32 -08002938 ret = page_counter_limit(&memcg->kmem, limit);
Johannes Weiner127424c2016-01-20 15:02:32 -08002939 mutex_unlock(&memcg_limit_mutex);
2940 return ret;
2941}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002942
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002943static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2944{
2945 int ret;
2946
2947 mutex_lock(&memcg_limit_mutex);
2948
Johannes Weiner0db15292016-01-20 15:02:50 -08002949 ret = page_counter_limit(&memcg->tcpmem, limit);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002950 if (ret)
2951 goto out;
2952
Johannes Weiner0db15292016-01-20 15:02:50 -08002953 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002954 /*
2955 * The active flag needs to be written after the static_key
2956 * update. This is what guarantees that the socket activation
2957 * function is the last one to run. See sock_update_memcg() for
2958 * details, and note that we don't mark any socket as belonging
2959 * to this memcg until that flag is up.
2960 *
2961 * We need to do this, because static_keys will span multiple
2962 * sites, but we can't control their order. If we mark a socket
2963 * as accounted, but the accounting functions are not patched in
2964 * yet, we'll lose accounting.
2965 *
2966 * We never race with the readers in sock_update_memcg(),
2967 * because when this value change, the code to process it is not
2968 * patched in yet.
2969 */
2970 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08002971 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002972 }
2973out:
2974 mutex_unlock(&memcg_limit_mutex);
2975 return ret;
2976}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002977
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002978/*
2979 * The user of this function is...
2980 * RES_LIMIT.
2981 */
Tejun Heo451af5042014-05-13 12:16:21 -04002982static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2983 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002984{
Tejun Heo451af5042014-05-13 12:16:21 -04002985 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002986 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002987 int ret;
2988
Tejun Heo451af5042014-05-13 12:16:21 -04002989 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08002990 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002991 if (ret)
2992 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07002993
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002994 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002995 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07002996 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2997 ret = -EINVAL;
2998 break;
2999 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003000 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3001 case _MEM:
3002 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003003 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003004 case _MEMSWAP:
3005 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3006 break;
3007 case _KMEM:
3008 ret = memcg_update_kmem_limit(memcg, nr_pages);
3009 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003010 case _TCP:
3011 ret = memcg_update_tcp_limit(memcg, nr_pages);
3012 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003013 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003014 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003015 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003016 memcg->soft_limit = nr_pages;
3017 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003018 break;
3019 }
Tejun Heo451af5042014-05-13 12:16:21 -04003020 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003021}
3022
Tejun Heo6770c642014-05-13 12:16:21 -04003023static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3024 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003025{
Tejun Heo6770c642014-05-13 12:16:21 -04003026 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003027 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003028
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003029 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3030 case _MEM:
3031 counter = &memcg->memory;
3032 break;
3033 case _MEMSWAP:
3034 counter = &memcg->memsw;
3035 break;
3036 case _KMEM:
3037 counter = &memcg->kmem;
3038 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003039 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003040 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003041 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003042 default:
3043 BUG();
3044 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003045
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003046 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003047 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003048 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003049 break;
3050 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003051 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003052 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003053 default:
3054 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003055 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003056
Tejun Heo6770c642014-05-13 12:16:21 -04003057 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003058}
3059
Tejun Heo182446d2013-08-08 20:11:24 -04003060static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003061 struct cftype *cft)
3062{
Tejun Heo182446d2013-08-08 20:11:24 -04003063 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003064}
3065
Daisuke Nishimura02491442010-03-10 15:22:17 -08003066#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003067static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003068 struct cftype *cft, u64 val)
3069{
Tejun Heo182446d2013-08-08 20:11:24 -04003070 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003071
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003072 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003073 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003074
Glauber Costaee5e8472013-02-22 16:34:50 -08003075 /*
3076 * No kind of locking is needed in here, because ->can_attach() will
3077 * check this value once in the beginning of the process, and then carry
3078 * on with stale data. This means that changes to this value will only
3079 * affect task migrations starting after the change.
3080 */
3081 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003082 return 0;
3083}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003084#else
Tejun Heo182446d2013-08-08 20:11:24 -04003085static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003086 struct cftype *cft, u64 val)
3087{
3088 return -ENOSYS;
3089}
3090#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003091
Ying Han406eb0c2011-05-26 16:25:37 -07003092#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003093static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003094{
Greg Thelen25485de2013-11-12 15:07:40 -08003095 struct numa_stat {
3096 const char *name;
3097 unsigned int lru_mask;
3098 };
3099
3100 static const struct numa_stat stats[] = {
3101 { "total", LRU_ALL },
3102 { "file", LRU_ALL_FILE },
3103 { "anon", LRU_ALL_ANON },
3104 { "unevictable", BIT(LRU_UNEVICTABLE) },
3105 };
3106 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003107 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003108 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003109 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003110
Greg Thelen25485de2013-11-12 15:07:40 -08003111 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3112 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3113 seq_printf(m, "%s=%lu", stat->name, nr);
3114 for_each_node_state(nid, N_MEMORY) {
3115 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3116 stat->lru_mask);
3117 seq_printf(m, " N%d=%lu", nid, nr);
3118 }
3119 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003120 }
Ying Han406eb0c2011-05-26 16:25:37 -07003121
Ying Han071aee12013-11-12 15:07:41 -08003122 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3123 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003124
Ying Han071aee12013-11-12 15:07:41 -08003125 nr = 0;
3126 for_each_mem_cgroup_tree(iter, memcg)
3127 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3128 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3129 for_each_node_state(nid, N_MEMORY) {
3130 nr = 0;
3131 for_each_mem_cgroup_tree(iter, memcg)
3132 nr += mem_cgroup_node_nr_lru_pages(
3133 iter, nid, stat->lru_mask);
3134 seq_printf(m, " N%d=%lu", nid, nr);
3135 }
3136 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003137 }
Ying Han406eb0c2011-05-26 16:25:37 -07003138
Ying Han406eb0c2011-05-26 16:25:37 -07003139 return 0;
3140}
3141#endif /* CONFIG_NUMA */
3142
Tejun Heo2da8ca82013-12-05 12:28:04 -05003143static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003144{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003145 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003146 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003147 struct mem_cgroup *mi;
3148 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003149
Greg Thelen0ca44b12015-02-11 15:25:58 -08003150 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3151 MEM_CGROUP_STAT_NSTATS);
3152 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3153 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003154 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3155
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003156 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Johannes Weiner7941d212016-01-14 15:21:23 -08003157 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003158 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07003159 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003160 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003161 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003162
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003163 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3164 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3165 mem_cgroup_read_events(memcg, i));
3166
3167 for (i = 0; i < NR_LRU_LISTS; i++)
3168 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3169 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3170
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003171 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003172 memory = memsw = PAGE_COUNTER_MAX;
3173 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3174 memory = min(memory, mi->memory.limit);
3175 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003176 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003177 seq_printf(m, "hierarchical_memory_limit %llu\n",
3178 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08003179 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003180 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3181 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003182
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003183 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003184 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003185
Johannes Weiner7941d212016-01-14 15:21:23 -08003186 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003187 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003188 for_each_mem_cgroup_tree(mi, memcg)
3189 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
Greg Thelen484ebb32015-10-01 15:37:05 -07003190 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003191 }
3192
3193 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3194 unsigned long long val = 0;
3195
3196 for_each_mem_cgroup_tree(mi, memcg)
3197 val += mem_cgroup_read_events(mi, i);
3198 seq_printf(m, "total_%s %llu\n",
3199 mem_cgroup_events_names[i], val);
3200 }
3201
3202 for (i = 0; i < NR_LRU_LISTS; i++) {
3203 unsigned long long val = 0;
3204
3205 for_each_mem_cgroup_tree(mi, memcg)
3206 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3207 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003208 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003209
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003210#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003211 {
3212 int nid, zid;
3213 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003214 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003215 unsigned long recent_rotated[2] = {0, 0};
3216 unsigned long recent_scanned[2] = {0, 0};
3217
3218 for_each_online_node(nid)
3219 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Jianyu Zhane2318752014-06-06 14:38:20 -07003220 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
Hugh Dickins89abfab2012-05-29 15:06:53 -07003221 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003222
Hugh Dickins89abfab2012-05-29 15:06:53 -07003223 recent_rotated[0] += rstat->recent_rotated[0];
3224 recent_rotated[1] += rstat->recent_rotated[1];
3225 recent_scanned[0] += rstat->recent_scanned[0];
3226 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003227 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003228 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3229 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3230 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3231 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003232 }
3233#endif
3234
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003235 return 0;
3236}
3237
Tejun Heo182446d2013-08-08 20:11:24 -04003238static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3239 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003240{
Tejun Heo182446d2013-08-08 20:11:24 -04003241 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003242
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003243 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003244}
3245
Tejun Heo182446d2013-08-08 20:11:24 -04003246static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3247 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003248{
Tejun Heo182446d2013-08-08 20:11:24 -04003249 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003250
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003251 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003252 return -EINVAL;
3253
Linus Torvalds14208b02014-06-09 15:03:33 -07003254 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003255 memcg->swappiness = val;
3256 else
3257 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003258
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003259 return 0;
3260}
3261
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003262static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3263{
3264 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003265 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003266 int i;
3267
3268 rcu_read_lock();
3269 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003270 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003271 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003272 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003273
3274 if (!t)
3275 goto unlock;
3276
Johannes Weinerce00a962014-09-05 08:43:57 -04003277 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003278
3279 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003280 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003281 * If it's not true, a threshold was crossed after last
3282 * call of __mem_cgroup_threshold().
3283 */
Phil Carmody5407a562010-05-26 14:42:42 -07003284 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003285
3286 /*
3287 * Iterate backward over array of thresholds starting from
3288 * current_threshold and check if a threshold is crossed.
3289 * If none of thresholds below usage is crossed, we read
3290 * only one element of the array here.
3291 */
3292 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3293 eventfd_signal(t->entries[i].eventfd, 1);
3294
3295 /* i = current_threshold + 1 */
3296 i++;
3297
3298 /*
3299 * Iterate forward over array of thresholds starting from
3300 * current_threshold+1 and check if a threshold is crossed.
3301 * If none of thresholds above usage is crossed, we read
3302 * only one element of the array here.
3303 */
3304 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3305 eventfd_signal(t->entries[i].eventfd, 1);
3306
3307 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003308 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003309unlock:
3310 rcu_read_unlock();
3311}
3312
3313static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3314{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003315 while (memcg) {
3316 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08003317 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003318 __mem_cgroup_threshold(memcg, true);
3319
3320 memcg = parent_mem_cgroup(memcg);
3321 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003322}
3323
3324static int compare_thresholds(const void *a, const void *b)
3325{
3326 const struct mem_cgroup_threshold *_a = a;
3327 const struct mem_cgroup_threshold *_b = b;
3328
Greg Thelen2bff24a2013-09-11 14:23:08 -07003329 if (_a->threshold > _b->threshold)
3330 return 1;
3331
3332 if (_a->threshold < _b->threshold)
3333 return -1;
3334
3335 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003336}
3337
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003338static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003339{
3340 struct mem_cgroup_eventfd_list *ev;
3341
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003342 spin_lock(&memcg_oom_lock);
3343
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003344 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003345 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003346
3347 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003348 return 0;
3349}
3350
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003351static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003352{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003353 struct mem_cgroup *iter;
3354
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003355 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003356 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003357}
3358
Tejun Heo59b6f872013-11-22 18:20:43 -05003359static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003360 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003361{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003362 struct mem_cgroup_thresholds *thresholds;
3363 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003364 unsigned long threshold;
3365 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003366 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003367
Johannes Weiner650c5e52015-02-11 15:26:03 -08003368 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003369 if (ret)
3370 return ret;
3371
3372 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003373
Johannes Weiner05b84302014-08-06 16:05:59 -07003374 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003375 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003376 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003377 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003378 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003379 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003380 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003381 BUG();
3382
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003383 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003384 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003385 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3386
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003387 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003388
3389 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003390 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003391 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003392 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003393 ret = -ENOMEM;
3394 goto unlock;
3395 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003396 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003397
3398 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003399 if (thresholds->primary) {
3400 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003401 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003402 }
3403
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003404 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003405 new->entries[size - 1].eventfd = eventfd;
3406 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003407
3408 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003409 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003410 compare_thresholds, NULL);
3411
3412 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003413 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003414 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003415 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003416 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003417 * new->current_threshold will not be used until
3418 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003419 * it here.
3420 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003421 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003422 } else
3423 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003424 }
3425
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003426 /* Free old spare buffer and save old primary buffer as spare */
3427 kfree(thresholds->spare);
3428 thresholds->spare = thresholds->primary;
3429
3430 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003431
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003432 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003433 synchronize_rcu();
3434
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003435unlock:
3436 mutex_unlock(&memcg->thresholds_lock);
3437
3438 return ret;
3439}
3440
Tejun Heo59b6f872013-11-22 18:20:43 -05003441static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003442 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003443{
Tejun Heo59b6f872013-11-22 18:20:43 -05003444 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003445}
3446
Tejun Heo59b6f872013-11-22 18:20:43 -05003447static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003448 struct eventfd_ctx *eventfd, const char *args)
3449{
Tejun Heo59b6f872013-11-22 18:20:43 -05003450 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003451}
3452
Tejun Heo59b6f872013-11-22 18:20:43 -05003453static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003454 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003455{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003456 struct mem_cgroup_thresholds *thresholds;
3457 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003458 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003459 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003460
3461 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003462
3463 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003464 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003465 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003466 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003467 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003468 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003469 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003470 BUG();
3471
Anton Vorontsov371528c2012-02-24 05:14:46 +04003472 if (!thresholds->primary)
3473 goto unlock;
3474
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003475 /* Check if a threshold crossed before removing */
3476 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3477
3478 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003479 size = 0;
3480 for (i = 0; i < thresholds->primary->size; i++) {
3481 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003482 size++;
3483 }
3484
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003485 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003486
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003487 /* Set thresholds array to NULL if we don't have thresholds */
3488 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003489 kfree(new);
3490 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003491 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003492 }
3493
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003494 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003495
3496 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003497 new->current_threshold = -1;
3498 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3499 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003500 continue;
3501
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003502 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003503 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003504 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003505 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003506 * until rcu_assign_pointer(), so it's safe to increment
3507 * it here.
3508 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003509 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003510 }
3511 j++;
3512 }
3513
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003514swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003515 /* Swap primary and spare array */
3516 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003517
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003518 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003519
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003520 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003521 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08003522
3523 /* If all events are unregistered, free the spare array */
3524 if (!new) {
3525 kfree(thresholds->spare);
3526 thresholds->spare = NULL;
3527 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04003528unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003529 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003530}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003531
Tejun Heo59b6f872013-11-22 18:20:43 -05003532static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003533 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003534{
Tejun Heo59b6f872013-11-22 18:20:43 -05003535 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003536}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003537
Tejun Heo59b6f872013-11-22 18:20:43 -05003538static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003539 struct eventfd_ctx *eventfd)
3540{
Tejun Heo59b6f872013-11-22 18:20:43 -05003541 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003542}
3543
Tejun Heo59b6f872013-11-22 18:20:43 -05003544static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003545 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003546{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003547 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003548
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003549 event = kmalloc(sizeof(*event), GFP_KERNEL);
3550 if (!event)
3551 return -ENOMEM;
3552
Michal Hocko1af8efe2011-07-26 16:08:24 -07003553 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003554
3555 event->eventfd = eventfd;
3556 list_add(&event->list, &memcg->oom_notify);
3557
3558 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07003559 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003560 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003561 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003562
3563 return 0;
3564}
3565
Tejun Heo59b6f872013-11-22 18:20:43 -05003566static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003567 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003568{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003569 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003570
Michal Hocko1af8efe2011-07-26 16:08:24 -07003571 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003572
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003573 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003574 if (ev->eventfd == eventfd) {
3575 list_del(&ev->list);
3576 kfree(ev);
3577 }
3578 }
3579
Michal Hocko1af8efe2011-07-26 16:08:24 -07003580 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003581}
3582
Tejun Heo2da8ca82013-12-05 12:28:04 -05003583static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003584{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003585 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003586
Tejun Heo791badb2013-12-05 12:28:02 -05003587 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07003588 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003589 return 0;
3590}
3591
Tejun Heo182446d2013-08-08 20:11:24 -04003592static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003593 struct cftype *cft, u64 val)
3594{
Tejun Heo182446d2013-08-08 20:11:24 -04003595 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003596
3597 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003598 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003599 return -EINVAL;
3600
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003601 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003602 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003603 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003604
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003605 return 0;
3606}
3607
Tejun Heo52ebea72015-05-22 17:13:37 -04003608#ifdef CONFIG_CGROUP_WRITEBACK
3609
3610struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3611{
3612 return &memcg->cgwb_list;
3613}
3614
Tejun Heo841710a2015-05-22 18:23:33 -04003615static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3616{
3617 return wb_domain_init(&memcg->cgwb_domain, gfp);
3618}
3619
3620static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3621{
3622 wb_domain_exit(&memcg->cgwb_domain);
3623}
3624
Tejun Heo2529bb32015-05-22 18:23:34 -04003625static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3626{
3627 wb_domain_size_changed(&memcg->cgwb_domain);
3628}
3629
Tejun Heo841710a2015-05-22 18:23:33 -04003630struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3631{
3632 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3633
3634 if (!memcg->css.parent)
3635 return NULL;
3636
3637 return &memcg->cgwb_domain;
3638}
3639
Tejun Heoc2aa7232015-05-22 18:23:35 -04003640/**
3641 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3642 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003643 * @pfilepages: out parameter for number of file pages
3644 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04003645 * @pdirty: out parameter for number of dirty pages
3646 * @pwriteback: out parameter for number of pages under writeback
3647 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003648 * Determine the numbers of file, headroom, dirty, and writeback pages in
3649 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3650 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003651 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003652 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3653 * headroom is calculated as the lowest headroom of itself and the
3654 * ancestors. Note that this doesn't consider the actual amount of
3655 * available memory in the system. The caller should further cap
3656 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003657 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003658void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3659 unsigned long *pheadroom, unsigned long *pdirty,
3660 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04003661{
3662 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3663 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003664
3665 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3666
3667 /* this should eventually include NR_UNSTABLE_NFS */
3668 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003669 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3670 (1 << LRU_ACTIVE_FILE));
3671 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003672
Tejun Heoc2aa7232015-05-22 18:23:35 -04003673 while ((parent = parent_mem_cgroup(memcg))) {
3674 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3675 unsigned long used = page_counter_read(&memcg->memory);
3676
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003677 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04003678 memcg = parent;
3679 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04003680}
3681
Tejun Heo841710a2015-05-22 18:23:33 -04003682#else /* CONFIG_CGROUP_WRITEBACK */
3683
3684static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3685{
3686 return 0;
3687}
3688
3689static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3690{
3691}
3692
Tejun Heo2529bb32015-05-22 18:23:34 -04003693static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3694{
3695}
3696
Tejun Heo52ebea72015-05-22 17:13:37 -04003697#endif /* CONFIG_CGROUP_WRITEBACK */
3698
Tejun Heo79bd9812013-11-22 18:20:42 -05003699/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003700 * DO NOT USE IN NEW FILES.
3701 *
3702 * "cgroup.event_control" implementation.
3703 *
3704 * This is way over-engineered. It tries to support fully configurable
3705 * events for each user. Such level of flexibility is completely
3706 * unnecessary especially in the light of the planned unified hierarchy.
3707 *
3708 * Please deprecate this and replace with something simpler if at all
3709 * possible.
3710 */
3711
3712/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003713 * Unregister event and free resources.
3714 *
3715 * Gets called from workqueue.
3716 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003717static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003718{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003719 struct mem_cgroup_event *event =
3720 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003721 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003722
3723 remove_wait_queue(event->wqh, &event->wait);
3724
Tejun Heo59b6f872013-11-22 18:20:43 -05003725 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003726
3727 /* Notify userspace the event is going away. */
3728 eventfd_signal(event->eventfd, 1);
3729
3730 eventfd_ctx_put(event->eventfd);
3731 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003732 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003733}
3734
3735/*
3736 * Gets called on POLLHUP on eventfd when user closes it.
3737 *
3738 * Called with wqh->lock held and interrupts disabled.
3739 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003740static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3741 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003742{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003743 struct mem_cgroup_event *event =
3744 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003745 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003746 unsigned long flags = (unsigned long)key;
3747
3748 if (flags & POLLHUP) {
3749 /*
3750 * If the event has been detached at cgroup removal, we
3751 * can simply return knowing the other side will cleanup
3752 * for us.
3753 *
3754 * We can't race against event freeing since the other
3755 * side will require wqh->lock via remove_wait_queue(),
3756 * which we hold.
3757 */
Tejun Heofba94802013-11-22 18:20:43 -05003758 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003759 if (!list_empty(&event->list)) {
3760 list_del_init(&event->list);
3761 /*
3762 * We are in atomic context, but cgroup_event_remove()
3763 * may sleep, so we have to call it in workqueue.
3764 */
3765 schedule_work(&event->remove);
3766 }
Tejun Heofba94802013-11-22 18:20:43 -05003767 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003768 }
3769
3770 return 0;
3771}
3772
Tejun Heo3bc942f2013-11-22 18:20:44 -05003773static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003774 wait_queue_head_t *wqh, poll_table *pt)
3775{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003776 struct mem_cgroup_event *event =
3777 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003778
3779 event->wqh = wqh;
3780 add_wait_queue(wqh, &event->wait);
3781}
3782
3783/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003784 * DO NOT USE IN NEW FILES.
3785 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003786 * Parse input and register new cgroup event handler.
3787 *
3788 * Input must be in format '<event_fd> <control_fd> <args>'.
3789 * Interpretation of args is defined by control file implementation.
3790 */
Tejun Heo451af5042014-05-13 12:16:21 -04003791static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3792 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003793{
Tejun Heo451af5042014-05-13 12:16:21 -04003794 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003795 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003796 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003797 struct cgroup_subsys_state *cfile_css;
3798 unsigned int efd, cfd;
3799 struct fd efile;
3800 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003801 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003802 char *endp;
3803 int ret;
3804
Tejun Heo451af5042014-05-13 12:16:21 -04003805 buf = strstrip(buf);
3806
3807 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003808 if (*endp != ' ')
3809 return -EINVAL;
Tejun Heo451af5042014-05-13 12:16:21 -04003810 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003811
Tejun Heo451af5042014-05-13 12:16:21 -04003812 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003813 if ((*endp != ' ') && (*endp != '\0'))
3814 return -EINVAL;
Tejun Heo451af5042014-05-13 12:16:21 -04003815 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003816
3817 event = kzalloc(sizeof(*event), GFP_KERNEL);
3818 if (!event)
3819 return -ENOMEM;
3820
Tejun Heo59b6f872013-11-22 18:20:43 -05003821 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003822 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003823 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3824 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3825 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003826
3827 efile = fdget(efd);
3828 if (!efile.file) {
3829 ret = -EBADF;
3830 goto out_kfree;
3831 }
3832
3833 event->eventfd = eventfd_ctx_fileget(efile.file);
3834 if (IS_ERR(event->eventfd)) {
3835 ret = PTR_ERR(event->eventfd);
3836 goto out_put_efile;
3837 }
3838
3839 cfile = fdget(cfd);
3840 if (!cfile.file) {
3841 ret = -EBADF;
3842 goto out_put_eventfd;
3843 }
3844
3845 /* the process need read permission on control file */
3846 /* AV: shouldn't we check that it's been opened for read instead? */
3847 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3848 if (ret < 0)
3849 goto out_put_cfile;
3850
Tejun Heo79bd9812013-11-22 18:20:42 -05003851 /*
Tejun Heofba94802013-11-22 18:20:43 -05003852 * Determine the event callbacks and set them in @event. This used
3853 * to be done via struct cftype but cgroup core no longer knows
3854 * about these events. The following is crude but the whole thing
3855 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003856 *
3857 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003858 */
Al Virob5830432014-10-31 01:22:04 -04003859 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003860
3861 if (!strcmp(name, "memory.usage_in_bytes")) {
3862 event->register_event = mem_cgroup_usage_register_event;
3863 event->unregister_event = mem_cgroup_usage_unregister_event;
3864 } else if (!strcmp(name, "memory.oom_control")) {
3865 event->register_event = mem_cgroup_oom_register_event;
3866 event->unregister_event = mem_cgroup_oom_unregister_event;
3867 } else if (!strcmp(name, "memory.pressure_level")) {
3868 event->register_event = vmpressure_register_event;
3869 event->unregister_event = vmpressure_unregister_event;
3870 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003871 event->register_event = memsw_cgroup_usage_register_event;
3872 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003873 } else {
3874 ret = -EINVAL;
3875 goto out_put_cfile;
3876 }
3877
3878 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003879 * Verify @cfile should belong to @css. Also, remaining events are
3880 * automatically removed on cgroup destruction but the removal is
3881 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003882 */
Al Virob5830432014-10-31 01:22:04 -04003883 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003884 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003885 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003886 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003887 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003888 if (cfile_css != css) {
3889 css_put(cfile_css);
3890 goto out_put_cfile;
3891 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003892
Tejun Heo451af5042014-05-13 12:16:21 -04003893 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003894 if (ret)
3895 goto out_put_css;
3896
3897 efile.file->f_op->poll(efile.file, &event->pt);
3898
Tejun Heofba94802013-11-22 18:20:43 -05003899 spin_lock(&memcg->event_list_lock);
3900 list_add(&event->list, &memcg->event_list);
3901 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003902
3903 fdput(cfile);
3904 fdput(efile);
3905
Tejun Heo451af5042014-05-13 12:16:21 -04003906 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05003907
3908out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05003909 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003910out_put_cfile:
3911 fdput(cfile);
3912out_put_eventfd:
3913 eventfd_ctx_put(event->eventfd);
3914out_put_efile:
3915 fdput(efile);
3916out_kfree:
3917 kfree(event);
3918
3919 return ret;
3920}
3921
Johannes Weiner241994ed2015-02-11 15:26:06 -08003922static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003923 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003924 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003925 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003926 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003927 },
3928 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003929 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003930 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04003931 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003932 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003933 },
3934 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003935 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003936 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af5042014-05-13 12:16:21 -04003937 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003938 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003939 },
3940 {
Balbir Singh296c81d2009-09-23 15:56:36 -07003941 .name = "soft_limit_in_bytes",
3942 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af5042014-05-13 12:16:21 -04003943 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003944 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07003945 },
3946 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003947 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003948 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04003949 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003950 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003951 },
Balbir Singh8697d332008-02-07 00:13:59 -08003952 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003953 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003954 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003955 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003956 {
3957 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04003958 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003959 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08003960 {
3961 .name = "use_hierarchy",
3962 .write_u64 = mem_cgroup_hierarchy_write,
3963 .read_u64 = mem_cgroup_hierarchy_read,
3964 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003965 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05003966 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af5042014-05-13 12:16:21 -04003967 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04003968 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05003969 },
3970 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003971 .name = "swappiness",
3972 .read_u64 = mem_cgroup_swappiness_read,
3973 .write_u64 = mem_cgroup_swappiness_write,
3974 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003975 {
3976 .name = "move_charge_at_immigrate",
3977 .read_u64 = mem_cgroup_move_charge_read,
3978 .write_u64 = mem_cgroup_move_charge_write,
3979 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003980 {
3981 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003982 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003983 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003984 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3985 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003986 {
3987 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003988 },
Ying Han406eb0c2011-05-26 16:25:37 -07003989#ifdef CONFIG_NUMA
3990 {
3991 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003992 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07003993 },
3994#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08003995 {
3996 .name = "kmem.limit_in_bytes",
3997 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af5042014-05-13 12:16:21 -04003998 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003999 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004000 },
4001 {
4002 .name = "kmem.usage_in_bytes",
4003 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004004 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004005 },
4006 {
4007 .name = "kmem.failcnt",
4008 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004009 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004010 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004011 },
4012 {
4013 .name = "kmem.max_usage_in_bytes",
4014 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004015 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004016 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004017 },
Glauber Costa749c5412012-12-18 14:23:01 -08004018#ifdef CONFIG_SLABINFO
4019 {
4020 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004021 .seq_start = slab_start,
4022 .seq_next = slab_next,
4023 .seq_stop = slab_stop,
4024 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004025 },
4026#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004027 {
4028 .name = "kmem.tcp.limit_in_bytes",
4029 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4030 .write = mem_cgroup_write,
4031 .read_u64 = mem_cgroup_read_u64,
4032 },
4033 {
4034 .name = "kmem.tcp.usage_in_bytes",
4035 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4036 .read_u64 = mem_cgroup_read_u64,
4037 },
4038 {
4039 .name = "kmem.tcp.failcnt",
4040 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4041 .write = mem_cgroup_reset,
4042 .read_u64 = mem_cgroup_read_u64,
4043 },
4044 {
4045 .name = "kmem.tcp.max_usage_in_bytes",
4046 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4047 .write = mem_cgroup_reset,
4048 .read_u64 = mem_cgroup_read_u64,
4049 },
Tejun Heo6bc10342012-04-01 12:09:55 -07004050 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004051};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004052
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004053static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004054{
4055 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004056 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004057 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004058 /*
4059 * This routine is called against possible nodes.
4060 * But it's BUG to call kmalloc() against offline node.
4061 *
4062 * TODO: this routine can waste much memory for nodes which will
4063 * never be onlined. It's better to use memory hotplug callback
4064 * function.
4065 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004066 if (!node_state(node, N_NORMAL_MEMORY))
4067 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004068 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004069 if (!pn)
4070 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004071
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004072 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4073 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004074 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -07004075 mz->usage_in_excess = 0;
4076 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004077 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004078 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004079 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004080 return 0;
4081}
4082
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004083static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004084{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004085 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004086}
4087
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004088static void mem_cgroup_free(struct mem_cgroup *memcg)
4089{
4090 int node;
4091
4092 memcg_wb_domain_exit(memcg);
4093 for_each_node(node)
4094 free_mem_cgroup_per_zone_info(memcg, node);
4095 free_percpu(memcg->stat);
4096 kfree(memcg);
4097}
4098
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004099static struct mem_cgroup *mem_cgroup_alloc(void)
4100{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004101 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004102 size_t size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004103 int node;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004104
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004105 size = sizeof(struct mem_cgroup);
4106 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004107
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004108 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004109 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004110 return NULL;
4111
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004112 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4113 if (!memcg->stat)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004114 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004115
Bob Liu3ed28fa2012-01-12 17:19:04 -08004116 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004117 if (alloc_mem_cgroup_per_zone_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004118 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004119
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004120 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4121 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004122
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004123 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08004124 memcg->last_scanned_node = MAX_NUMNODES;
4125 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004126 mutex_init(&memcg->thresholds_lock);
4127 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004128 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004129 INIT_LIST_HEAD(&memcg->event_list);
4130 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004131 memcg->socket_pressure = jiffies;
Johannes Weiner127424c2016-01-20 15:02:32 -08004132#ifndef CONFIG_SLOB
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004133 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004134#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004135#ifdef CONFIG_CGROUP_WRITEBACK
4136 INIT_LIST_HEAD(&memcg->cgwb_list);
4137#endif
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004138 return memcg;
4139fail:
4140 mem_cgroup_free(memcg);
4141 return NULL;
Glauber Costad142e3e2013-02-22 16:34:52 -08004142}
4143
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004144static struct cgroup_subsys_state * __ref
4145mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004146{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004147 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4148 struct mem_cgroup *memcg;
4149 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08004150
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004151 memcg = mem_cgroup_alloc();
4152 if (!memcg)
4153 return ERR_PTR(error);
Li Zefan4219b2d2013-09-23 16:56:29 +08004154
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004155 memcg->high = PAGE_COUNTER_MAX;
4156 memcg->soft_limit = PAGE_COUNTER_MAX;
4157 if (parent) {
4158 memcg->swappiness = mem_cgroup_swappiness(parent);
4159 memcg->oom_kill_disable = parent->oom_kill_disable;
4160 }
4161 if (parent && parent->use_hierarchy) {
4162 memcg->use_hierarchy = true;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004163 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004164 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004165 page_counter_init(&memcg->memsw, &parent->memsw);
4166 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08004167 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004168 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004169 page_counter_init(&memcg->memory, NULL);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004170 page_counter_init(&memcg->swap, NULL);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004171 page_counter_init(&memcg->memsw, NULL);
4172 page_counter_init(&memcg->kmem, NULL);
Johannes Weiner0db15292016-01-20 15:02:50 -08004173 page_counter_init(&memcg->tcpmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004174 /*
4175 * Deeper hierachy with use_hierarchy == false doesn't make
4176 * much sense so let cgroup subsystem know about this
4177 * unfortunate state in our controller.
4178 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004179 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004180 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004181 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08004182
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004183 /* The following stuff does not apply to the root */
4184 if (!parent) {
4185 root_mem_cgroup = memcg;
4186 return &memcg->css;
4187 }
4188
Vladimir Davydovb313aee2016-03-17 14:18:27 -07004189 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004190 if (error)
4191 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08004192
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004193 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004194 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004195
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004196 return &memcg->css;
4197fail:
4198 mem_cgroup_free(memcg);
4199 return NULL;
4200}
4201
4202static int
4203mem_cgroup_css_online(struct cgroup_subsys_state *css)
4204{
4205 if (css->id > MEM_CGROUP_ID_MAX)
4206 return -ENOSPC;
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004207
4208 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004209}
4210
Tejun Heoeb954192013-08-08 20:11:23 -04004211static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004212{
Tejun Heoeb954192013-08-08 20:11:23 -04004213 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004214 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004215
4216 /*
4217 * Unregister events and notify userspace.
4218 * Notify userspace about cgroup removing only after rmdir of cgroup
4219 * directory to avoid race between userspace and kernelspace.
4220 */
Tejun Heofba94802013-11-22 18:20:43 -05004221 spin_lock(&memcg->event_list_lock);
4222 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004223 list_del_init(&event->list);
4224 schedule_work(&event->remove);
4225 }
Tejun Heofba94802013-11-22 18:20:43 -05004226 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004227
Johannes Weiner567e9ab2016-01-20 15:02:24 -08004228 memcg_offline_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004229 wb_memcg_offline(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004230}
4231
Vladimir Davydov6df38682015-12-29 14:54:10 -08004232static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4233{
4234 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4235
4236 invalidate_reclaim_iterators(memcg);
4237}
4238
Tejun Heoeb954192013-08-08 20:11:23 -04004239static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004240{
Tejun Heoeb954192013-08-08 20:11:23 -04004241 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004242
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004243 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004244 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004245
Johannes Weiner0db15292016-01-20 15:02:50 -08004246 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004247 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004248
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004249 vmpressure_cleanup(&memcg->vmpressure);
4250 cancel_work_sync(&memcg->high_work);
4251 mem_cgroup_remove_from_trees(memcg);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004252 memcg_free_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004253 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004254}
4255
Tejun Heo1ced953b2014-07-08 18:02:57 -04004256/**
4257 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4258 * @css: the target css
4259 *
4260 * Reset the states of the mem_cgroup associated with @css. This is
4261 * invoked when the userland requests disabling on the default hierarchy
4262 * but the memcg is pinned through dependency. The memcg should stop
4263 * applying policies and should revert to the vanilla state as it may be
4264 * made visible again.
4265 *
4266 * The current implementation only resets the essential configurations.
4267 * This needs to be expanded to cover all the visible parts.
4268 */
4269static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4270{
4271 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4272
Vladimir Davydovd334c9bc2016-03-17 14:19:38 -07004273 page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4274 page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4275 page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4276 page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4277 page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004278 memcg->low = 0;
4279 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004280 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004281 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced953b2014-07-08 18:02:57 -04004282}
4283
Daisuke Nishimura02491442010-03-10 15:22:17 -08004284#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004285/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004286static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004287{
Johannes Weiner05b84302014-08-06 16:05:59 -07004288 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004289
Mel Gormand0164ad2015-11-06 16:28:21 -08004290 /* Try a single bulk charge without reclaim first, kswapd may wake */
4291 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004292 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004293 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004294 return ret;
4295 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004296
4297 /* Try charges one by one with reclaim */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004298 while (count--) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004299 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004300 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004301 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004302 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004303 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004304 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004305 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004306}
4307
4308/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004309 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004310 * @vma: the vma the pte to be checked belongs
4311 * @addr: the address corresponding to the pte to be checked
4312 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08004313 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004314 *
4315 * Returns
4316 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4317 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4318 * move charge. if @target is not NULL, the page is stored in target->page
4319 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08004320 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4321 * target for charge migration. if @target is not NULL, the entry is stored
4322 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004323 *
4324 * Called with pte lock held.
4325 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004326union mc_target {
4327 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004328 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004329};
4330
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004331enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004332 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004333 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004334 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004335};
4336
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004337static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4338 unsigned long addr, pte_t ptent)
4339{
4340 struct page *page = vm_normal_page(vma, addr, ptent);
4341
4342 if (!page || !page_mapped(page))
4343 return NULL;
4344 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004345 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004346 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004347 } else {
4348 if (!(mc.flags & MOVE_FILE))
4349 return NULL;
4350 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004351 if (!get_page_unless_zero(page))
4352 return NULL;
4353
4354 return page;
4355}
4356
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004357#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004358static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4359 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4360{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004361 struct page *page = NULL;
4362 swp_entry_t ent = pte_to_swp_entry(ptent);
4363
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004364 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004365 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004366 /*
4367 * Because lookup_swap_cache() updates some statistics counter,
4368 * we call find_get_page() with swapper_space directly.
4369 */
Shaohua Li33806f02013-02-22 16:34:37 -08004370 page = find_get_page(swap_address_space(ent), ent.val);
Johannes Weiner7941d212016-01-14 15:21:23 -08004371 if (do_memsw_account())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004372 entry->val = ent.val;
4373
4374 return page;
4375}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004376#else
4377static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4378 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4379{
4380 return NULL;
4381}
4382#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004383
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004384static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4385 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4386{
4387 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004388 struct address_space *mapping;
4389 pgoff_t pgoff;
4390
4391 if (!vma->vm_file) /* anonymous vma */
4392 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004393 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004394 return NULL;
4395
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004396 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004397 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004398
4399 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004400#ifdef CONFIG_SWAP
4401 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004402 if (shmem_mapping(mapping)) {
4403 page = find_get_entry(mapping, pgoff);
4404 if (radix_tree_exceptional_entry(page)) {
4405 swp_entry_t swp = radix_to_swp_entry(page);
Johannes Weiner7941d212016-01-14 15:21:23 -08004406 if (do_memsw_account())
Johannes Weiner139b6a62014-05-06 12:50:05 -07004407 *entry = swp;
4408 page = find_get_page(swap_address_space(swp), swp.val);
4409 }
4410 } else
4411 page = find_get_page(mapping, pgoff);
4412#else
4413 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004414#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004415 return page;
4416}
4417
Chen Gangb1b0dea2015-04-14 15:47:35 -07004418/**
4419 * mem_cgroup_move_account - move account of the page
4420 * @page: the page
4421 * @nr_pages: number of regular pages (>1 for huge pages)
4422 * @from: mem_cgroup which the page is moved from.
4423 * @to: mem_cgroup which the page is moved to. @from != @to.
4424 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08004425 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07004426 *
4427 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4428 * from old cgroup.
4429 */
4430static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004431 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07004432 struct mem_cgroup *from,
4433 struct mem_cgroup *to)
4434{
4435 unsigned long flags;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004436 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004437 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004438 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004439
4440 VM_BUG_ON(from == to);
4441 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004442 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07004443
4444 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07004445 * Prevent mem_cgroup_migrate() from looking at
Hugh Dickins45637ba2015-11-05 18:49:40 -08004446 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07004447 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004448 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004449 if (!trylock_page(page))
4450 goto out;
4451
4452 ret = -EINVAL;
4453 if (page->mem_cgroup != from)
4454 goto out_unlock;
4455
Greg Thelenc4843a72015-05-22 17:13:16 -04004456 anon = PageAnon(page);
4457
Chen Gangb1b0dea2015-04-14 15:47:35 -07004458 spin_lock_irqsave(&from->move_lock, flags);
4459
Greg Thelenc4843a72015-05-22 17:13:16 -04004460 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004461 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4462 nr_pages);
4463 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4464 nr_pages);
4465 }
4466
Greg Thelenc4843a72015-05-22 17:13:16 -04004467 /*
4468 * move_lock grabbed above and caller set from->moving_account, so
4469 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4470 * So mapping should be stable for dirty pages.
4471 */
4472 if (!anon && PageDirty(page)) {
4473 struct address_space *mapping = page_mapping(page);
4474
4475 if (mapping_cap_account_dirty(mapping)) {
4476 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4477 nr_pages);
4478 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4479 nr_pages);
4480 }
4481 }
4482
Chen Gangb1b0dea2015-04-14 15:47:35 -07004483 if (PageWriteback(page)) {
4484 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4485 nr_pages);
4486 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4487 nr_pages);
4488 }
4489
4490 /*
4491 * It is safe to change page->mem_cgroup here because the page
4492 * is referenced, charged, and isolated - we can't race with
4493 * uncharging, charging, migration, or LRU putback.
4494 */
4495
4496 /* caller should have done css_get */
4497 page->mem_cgroup = to;
4498 spin_unlock_irqrestore(&from->move_lock, flags);
4499
4500 ret = 0;
4501
4502 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004503 mem_cgroup_charge_statistics(to, page, compound, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004504 memcg_check_events(to, page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004505 mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004506 memcg_check_events(from, page);
4507 local_irq_enable();
4508out_unlock:
4509 unlock_page(page);
4510out:
4511 return ret;
4512}
4513
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004514static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004515 unsigned long addr, pte_t ptent, union mc_target *target)
4516{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004517 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004518 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004519 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004520
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004521 if (pte_present(ptent))
4522 page = mc_handle_present_pte(vma, addr, ptent);
4523 else if (is_swap_pte(ptent))
4524 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004525 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004526 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004527
4528 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004529 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004530 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004531 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004532 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004533 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004534 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004535 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004536 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004537 ret = MC_TARGET_PAGE;
4538 if (target)
4539 target->page = page;
4540 }
4541 if (!ret || !target)
4542 put_page(page);
4543 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004544 /* There is a swap entry and a page doesn't exist or isn't charged */
4545 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004546 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004547 ret = MC_TARGET_SWAP;
4548 if (target)
4549 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004550 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004551 return ret;
4552}
4553
Naoya Horiguchi12724852012-03-21 16:34:28 -07004554#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4555/*
4556 * We don't consider swapping or file mapped pages because THP does not
4557 * support them for now.
4558 * Caller should make sure that pmd_trans_huge(pmd) is true.
4559 */
4560static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4561 unsigned long addr, pmd_t pmd, union mc_target *target)
4562{
4563 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004564 enum mc_target_type ret = MC_TARGET_NONE;
4565
4566 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004567 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004568 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004569 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004570 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004571 ret = MC_TARGET_PAGE;
4572 if (target) {
4573 get_page(page);
4574 target->page = page;
4575 }
4576 }
4577 return ret;
4578}
4579#else
4580static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4581 unsigned long addr, pmd_t pmd, union mc_target *target)
4582{
4583 return MC_TARGET_NONE;
4584}
4585#endif
4586
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004587static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4588 unsigned long addr, unsigned long end,
4589 struct mm_walk *walk)
4590{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004591 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004592 pte_t *pte;
4593 spinlock_t *ptl;
4594
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004595 ptl = pmd_trans_huge_lock(pmd, vma);
4596 if (ptl) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004597 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4598 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004599 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004600 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004601 }
Dave Hansen03319322011-03-22 16:32:56 -07004602
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004603 if (pmd_trans_unstable(pmd))
4604 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004605 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4606 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004607 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004608 mc.precharge++; /* increment precharge temporarily */
4609 pte_unmap_unlock(pte - 1, ptl);
4610 cond_resched();
4611
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004612 return 0;
4613}
4614
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004615static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4616{
4617 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004618
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004619 struct mm_walk mem_cgroup_count_precharge_walk = {
4620 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4621 .mm = mm,
4622 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004623 down_read(&mm->mmap_sem);
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004624 walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004625 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004626
4627 precharge = mc.precharge;
4628 mc.precharge = 0;
4629
4630 return precharge;
4631}
4632
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004633static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4634{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004635 unsigned long precharge = mem_cgroup_count_precharge(mm);
4636
4637 VM_BUG_ON(mc.moving_task);
4638 mc.moving_task = current;
4639 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004640}
4641
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004642/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4643static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004644{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004645 struct mem_cgroup *from = mc.from;
4646 struct mem_cgroup *to = mc.to;
4647
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004648 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004649 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004650 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004651 mc.precharge = 0;
4652 }
4653 /*
4654 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4655 * we must uncharge here.
4656 */
4657 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004658 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004659 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004660 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004661 /* we must fixup refcnts and charges */
4662 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004663 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004664 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004665 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004666
Johannes Weiner05b84302014-08-06 16:05:59 -07004667 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004668 * we charged both to->memory and to->memsw, so we
4669 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004670 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004671 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004672 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004673
Johannes Weinere8ea14c2014-12-10 15:42:42 -08004674 css_put_many(&mc.from->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004675
Li Zefan40503772013-07-08 16:00:34 -07004676 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004677 mc.moved_swap = 0;
4678 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004679 memcg_oom_recover(from);
4680 memcg_oom_recover(to);
4681 wake_up_all(&mc.waitq);
4682}
4683
4684static void mem_cgroup_clear_mc(void)
4685{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004686 struct mm_struct *mm = mc.mm;
4687
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004688 /*
4689 * we must clear moving_task before waking up waiters at the end of
4690 * task migration.
4691 */
4692 mc.moving_task = NULL;
4693 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004694 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004695 mc.from = NULL;
4696 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04004697 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004698 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004699
4700 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004701}
4702
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004703static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004704{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004705 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07004706 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07004707 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04004708 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004709 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004710 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004711 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004712
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004713 /* charge immigration isn't supported on the default hierarchy */
4714 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07004715 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004716
Tejun Heo4530edd2015-09-11 15:00:19 -04004717 /*
4718 * Multi-process migrations only happen on the default hierarchy
4719 * where charge immigration is not used. Perform charge
4720 * immigration if @tset contains a leader and whine if there are
4721 * multiple.
4722 */
4723 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004724 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04004725 WARN_ON_ONCE(p);
4726 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004727 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04004728 }
4729 if (!p)
4730 return 0;
4731
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004732 /*
4733 * We are now commited to this value whatever it is. Changes in this
4734 * tunable will only affect upcoming migrations, not the current one.
4735 * So we need to save it, and keep it going.
4736 */
4737 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4738 if (!move_flags)
4739 return 0;
4740
Tejun Heo9f2115f2015-09-08 15:01:10 -07004741 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004742
Tejun Heo9f2115f2015-09-08 15:01:10 -07004743 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004744
Tejun Heo9f2115f2015-09-08 15:01:10 -07004745 mm = get_task_mm(p);
4746 if (!mm)
4747 return 0;
4748 /* We move charges only when we move a owner of the mm */
4749 if (mm->owner == p) {
4750 VM_BUG_ON(mc.from);
4751 VM_BUG_ON(mc.to);
4752 VM_BUG_ON(mc.precharge);
4753 VM_BUG_ON(mc.moved_charge);
4754 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004755
Tejun Heo9f2115f2015-09-08 15:01:10 -07004756 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004757 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004758 mc.from = from;
4759 mc.to = memcg;
4760 mc.flags = move_flags;
4761 spin_unlock(&mc.lock);
4762 /* We set mc.moving_task later */
4763
4764 ret = mem_cgroup_precharge_mc(mm);
4765 if (ret)
4766 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004767 } else {
4768 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004769 }
4770 return ret;
4771}
4772
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004773static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004774{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004775 if (mc.to)
4776 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004777}
4778
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004779static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4780 unsigned long addr, unsigned long end,
4781 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004782{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004783 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004784 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004785 pte_t *pte;
4786 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004787 enum mc_target_type target_type;
4788 union mc_target target;
4789 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004790
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004791 ptl = pmd_trans_huge_lock(pmd, vma);
4792 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004793 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004794 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004795 return 0;
4796 }
4797 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4798 if (target_type == MC_TARGET_PAGE) {
4799 page = target.page;
4800 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004801 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08004802 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004803 mc.precharge -= HPAGE_PMD_NR;
4804 mc.moved_charge += HPAGE_PMD_NR;
4805 }
4806 putback_lru_page(page);
4807 }
4808 put_page(page);
4809 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004810 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004811 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004812 }
4813
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004814 if (pmd_trans_unstable(pmd))
4815 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004816retry:
4817 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4818 for (; addr != end; addr += PAGE_SIZE) {
4819 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004820 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004821
4822 if (!mc.precharge)
4823 break;
4824
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004825 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004826 case MC_TARGET_PAGE:
4827 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08004828 /*
4829 * We can have a part of the split pmd here. Moving it
4830 * can be done but it would be too convoluted so simply
4831 * ignore such a partial THP and keep it in original
4832 * memcg. There should be somebody mapping the head.
4833 */
4834 if (PageTransCompound(page))
4835 goto put;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004836 if (isolate_lru_page(page))
4837 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004838 if (!mem_cgroup_move_account(page, false,
4839 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004840 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004841 /* we uncharge from mc.from later. */
4842 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004843 }
4844 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004845put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004846 put_page(page);
4847 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004848 case MC_TARGET_SWAP:
4849 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004850 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004851 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004852 /* we fixup refcnts and charges later. */
4853 mc.moved_swap++;
4854 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004855 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004856 default:
4857 break;
4858 }
4859 }
4860 pte_unmap_unlock(pte - 1, ptl);
4861 cond_resched();
4862
4863 if (addr != end) {
4864 /*
4865 * We have consumed all precharges we got in can_attach().
4866 * We try charge one by one, but don't do any additional
4867 * charges to mc.to if we have failed in charge once in attach()
4868 * phase.
4869 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004870 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004871 if (!ret)
4872 goto retry;
4873 }
4874
4875 return ret;
4876}
4877
Tejun Heo264a0ae2016-04-21 19:09:02 -04004878static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004879{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004880 struct mm_walk mem_cgroup_move_charge_walk = {
4881 .pmd_entry = mem_cgroup_move_charge_pte_range,
Tejun Heo264a0ae2016-04-21 19:09:02 -04004882 .mm = mc.mm,
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004883 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004884
4885 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004886 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07004887 * Signal lock_page_memcg() to take the memcg's move_lock
4888 * while we're moving its pages to another memcg. Then wait
4889 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08004890 */
4891 atomic_inc(&mc.from->moving_account);
4892 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004893retry:
Tejun Heo264a0ae2016-04-21 19:09:02 -04004894 if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004895 /*
4896 * Someone who are holding the mmap_sem might be waiting in
4897 * waitq. So we cancel all extra charges, wake up all waiters,
4898 * and retry. Because we cancel precharges, we might not be able
4899 * to move enough charges, but moving charge is a best-effort
4900 * feature anyway, so it wouldn't be a big problem.
4901 */
4902 __mem_cgroup_clear_mc();
4903 cond_resched();
4904 goto retry;
4905 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004906 /*
4907 * When we have consumed all precharges and failed in doing
4908 * additional charge, the page walk just aborts.
4909 */
4910 walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004911 up_read(&mc.mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08004912 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004913}
4914
Tejun Heo264a0ae2016-04-21 19:09:02 -04004915static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08004916{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004917 if (mc.to) {
4918 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004919 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004920 }
Balbir Singh67e465a2008-02-07 00:13:54 -08004921}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004922#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004923static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004924{
4925 return 0;
4926}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004927static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004928{
4929}
Tejun Heo264a0ae2016-04-21 19:09:02 -04004930static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004931{
4932}
4933#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08004934
Tejun Heof00baae2013-04-15 13:41:15 -07004935/*
4936 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04004937 * to verify whether we're attached to the default hierarchy on each mount
4938 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07004939 */
Tejun Heoeb954192013-08-08 20:11:23 -04004940static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07004941{
4942 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04004943 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07004944 * guarantees that @root doesn't have any children, so turning it
4945 * on for the root memcg is enough.
4946 */
Tejun Heo9e10a132015-09-18 11:56:28 -04004947 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee5902015-03-12 16:26:19 -07004948 root_mem_cgroup->use_hierarchy = true;
4949 else
4950 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07004951}
4952
Johannes Weiner241994ed2015-02-11 15:26:06 -08004953static u64 memory_current_read(struct cgroup_subsys_state *css,
4954 struct cftype *cft)
4955{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08004956 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4957
4958 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994ed2015-02-11 15:26:06 -08004959}
4960
4961static int memory_low_show(struct seq_file *m, void *v)
4962{
4963 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07004964 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004965
4966 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08004967 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08004968 else
4969 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
4970
4971 return 0;
4972}
4973
4974static ssize_t memory_low_write(struct kernfs_open_file *of,
4975 char *buf, size_t nbytes, loff_t off)
4976{
4977 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
4978 unsigned long low;
4979 int err;
4980
4981 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08004982 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004983 if (err)
4984 return err;
4985
4986 memcg->low = low;
4987
4988 return nbytes;
4989}
4990
4991static int memory_high_show(struct seq_file *m, void *v)
4992{
4993 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07004994 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004995
4996 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08004997 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08004998 else
4999 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5000
5001 return 0;
5002}
5003
5004static ssize_t memory_high_write(struct kernfs_open_file *of,
5005 char *buf, size_t nbytes, loff_t off)
5006{
5007 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner588083b2016-03-17 14:20:25 -07005008 unsigned long nr_pages;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005009 unsigned long high;
5010 int err;
5011
5012 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005013 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005014 if (err)
5015 return err;
5016
5017 memcg->high = high;
5018
Johannes Weiner588083b2016-03-17 14:20:25 -07005019 nr_pages = page_counter_read(&memcg->memory);
5020 if (nr_pages > high)
5021 try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5022 GFP_KERNEL, true);
5023
Tejun Heo2529bb32015-05-22 18:23:34 -04005024 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005025 return nbytes;
5026}
5027
5028static int memory_max_show(struct seq_file *m, void *v)
5029{
5030 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005031 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005032
5033 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005034 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005035 else
5036 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5037
5038 return 0;
5039}
5040
5041static ssize_t memory_max_write(struct kernfs_open_file *of,
5042 char *buf, size_t nbytes, loff_t off)
5043{
5044 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005045 unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5046 bool drained = false;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005047 unsigned long max;
5048 int err;
5049
5050 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005051 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005052 if (err)
5053 return err;
5054
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005055 xchg(&memcg->memory.limit, max);
5056
5057 for (;;) {
5058 unsigned long nr_pages = page_counter_read(&memcg->memory);
5059
5060 if (nr_pages <= max)
5061 break;
5062
5063 if (signal_pending(current)) {
5064 err = -EINTR;
5065 break;
5066 }
5067
5068 if (!drained) {
5069 drain_all_stock(memcg);
5070 drained = true;
5071 continue;
5072 }
5073
5074 if (nr_reclaims) {
5075 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5076 GFP_KERNEL, true))
5077 nr_reclaims--;
5078 continue;
5079 }
5080
5081 mem_cgroup_events(memcg, MEMCG_OOM, 1);
5082 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5083 break;
5084 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08005085
Tejun Heo2529bb32015-05-22 18:23:34 -04005086 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005087 return nbytes;
5088}
5089
5090static int memory_events_show(struct seq_file *m, void *v)
5091{
5092 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5093
5094 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5095 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5096 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5097 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5098
5099 return 0;
5100}
5101
Johannes Weiner587d9f72016-01-20 15:03:19 -08005102static int memory_stat_show(struct seq_file *m, void *v)
5103{
5104 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005105 unsigned long stat[MEMCG_NR_STAT];
5106 unsigned long events[MEMCG_NR_EVENTS];
Johannes Weiner587d9f72016-01-20 15:03:19 -08005107 int i;
5108
5109 /*
5110 * Provide statistics on the state of the memory subsystem as
5111 * well as cumulative event counters that show past behavior.
5112 *
5113 * This list is ordered following a combination of these gradients:
5114 * 1) generic big picture -> specifics and details
5115 * 2) reflecting userspace activity -> reflecting kernel heuristics
5116 *
5117 * Current memory state:
5118 */
5119
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005120 tree_stat(memcg, stat);
5121 tree_events(memcg, events);
5122
Johannes Weiner587d9f72016-01-20 15:03:19 -08005123 seq_printf(m, "anon %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005124 (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005125 seq_printf(m, "file %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005126 (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE);
Vladimir Davydov12580e42016-03-17 14:17:38 -07005127 seq_printf(m, "kernel_stack %llu\n",
5128 (u64)stat[MEMCG_KERNEL_STACK] * PAGE_SIZE);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005129 seq_printf(m, "slab %llu\n",
5130 (u64)(stat[MEMCG_SLAB_RECLAIMABLE] +
5131 stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
Johannes Weinerb2807f02016-01-20 15:03:22 -08005132 seq_printf(m, "sock %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005133 (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005134
5135 seq_printf(m, "file_mapped %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005136 (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005137 seq_printf(m, "file_dirty %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005138 (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005139 seq_printf(m, "file_writeback %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005140 (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005141
5142 for (i = 0; i < NR_LRU_LISTS; i++) {
5143 struct mem_cgroup *mi;
5144 unsigned long val = 0;
5145
5146 for_each_mem_cgroup_tree(mi, memcg)
5147 val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5148 seq_printf(m, "%s %llu\n",
5149 mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5150 }
5151
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005152 seq_printf(m, "slab_reclaimable %llu\n",
5153 (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE);
5154 seq_printf(m, "slab_unreclaimable %llu\n",
5155 (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5156
Johannes Weiner587d9f72016-01-20 15:03:19 -08005157 /* Accumulated memory events */
5158
5159 seq_printf(m, "pgfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005160 events[MEM_CGROUP_EVENTS_PGFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005161 seq_printf(m, "pgmajfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005162 events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005163
5164 return 0;
5165}
5166
Johannes Weiner241994ed2015-02-11 15:26:06 -08005167static struct cftype memory_files[] = {
5168 {
5169 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005170 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005171 .read_u64 = memory_current_read,
5172 },
5173 {
5174 .name = "low",
5175 .flags = CFTYPE_NOT_ON_ROOT,
5176 .seq_show = memory_low_show,
5177 .write = memory_low_write,
5178 },
5179 {
5180 .name = "high",
5181 .flags = CFTYPE_NOT_ON_ROOT,
5182 .seq_show = memory_high_show,
5183 .write = memory_high_write,
5184 },
5185 {
5186 .name = "max",
5187 .flags = CFTYPE_NOT_ON_ROOT,
5188 .seq_show = memory_max_show,
5189 .write = memory_max_write,
5190 },
5191 {
5192 .name = "events",
5193 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04005194 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994ed2015-02-11 15:26:06 -08005195 .seq_show = memory_events_show,
5196 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08005197 {
5198 .name = "stat",
5199 .flags = CFTYPE_NOT_ON_ROOT,
5200 .seq_show = memory_stat_show,
5201 },
Johannes Weiner241994ed2015-02-11 15:26:06 -08005202 { } /* terminate */
5203};
5204
Tejun Heo073219e2014-02-08 10:36:58 -05005205struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005206 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005207 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005208 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08005209 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08005210 .css_free = mem_cgroup_css_free,
Tejun Heo1ced953b2014-07-08 18:02:57 -04005211 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005212 .can_attach = mem_cgroup_can_attach,
5213 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04005214 .post_attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005215 .bind = mem_cgroup_bind,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005216 .dfl_cftypes = memory_files,
5217 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005218 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005219};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005220
Johannes Weiner241994ed2015-02-11 15:26:06 -08005221/**
Johannes Weiner241994ed2015-02-11 15:26:06 -08005222 * mem_cgroup_low - check if memory consumption is below the normal range
5223 * @root: the highest ancestor to consider
5224 * @memcg: the memory cgroup to check
5225 *
5226 * Returns %true if memory consumption of @memcg, and that of all
5227 * configurable ancestors up to @root, is below the normal range.
5228 */
5229bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5230{
5231 if (mem_cgroup_disabled())
5232 return false;
5233
5234 /*
5235 * The toplevel group doesn't have a configurable range, so
5236 * it's never low when looked at directly, and it is not
5237 * considered an ancestor when assessing the hierarchy.
5238 */
5239
5240 if (memcg == root_mem_cgroup)
5241 return false;
5242
Michal Hocko4e54ded2015-02-27 15:51:46 -08005243 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005244 return false;
5245
5246 while (memcg != root) {
5247 memcg = parent_mem_cgroup(memcg);
5248
5249 if (memcg == root_mem_cgroup)
5250 break;
5251
Michal Hocko4e54ded2015-02-27 15:51:46 -08005252 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005253 return false;
5254 }
5255 return true;
5256}
5257
Johannes Weiner00501b52014-08-08 14:19:20 -07005258/**
5259 * mem_cgroup_try_charge - try charging a page
5260 * @page: page to charge
5261 * @mm: mm context of the victim
5262 * @gfp_mask: reclaim mode
5263 * @memcgp: charged memcg return
5264 *
5265 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5266 * pages according to @gfp_mask if necessary.
5267 *
5268 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5269 * Otherwise, an error code is returned.
5270 *
5271 * After page->mapping has been set up, the caller must finalize the
5272 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5273 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5274 */
5275int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005276 gfp_t gfp_mask, struct mem_cgroup **memcgp,
5277 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005278{
5279 struct mem_cgroup *memcg = NULL;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005280 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005281 int ret = 0;
5282
5283 if (mem_cgroup_disabled())
5284 goto out;
5285
5286 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005287 /*
5288 * Every swap fault against a single page tries to charge the
5289 * page, bail as early as possible. shmem_unuse() encounters
5290 * already charged pages, too. The USED bit is protected by
5291 * the page lock, which serializes swap cache removal, which
5292 * in turn serializes uncharging.
5293 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005294 VM_BUG_ON_PAGE(!PageLocked(page), page);
Johannes Weiner1306a852014-12-10 15:44:52 -08005295 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005296 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005297
Vladimir Davydov37e84352016-01-20 15:02:56 -08005298 if (do_swap_account) {
Vladimir Davydove993d902015-09-09 15:35:35 -07005299 swp_entry_t ent = { .val = page_private(page), };
5300 unsigned short id = lookup_swap_cgroup_id(ent);
5301
5302 rcu_read_lock();
5303 memcg = mem_cgroup_from_id(id);
5304 if (memcg && !css_tryget_online(&memcg->css))
5305 memcg = NULL;
5306 rcu_read_unlock();
5307 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005308 }
5309
Johannes Weiner00501b52014-08-08 14:19:20 -07005310 if (!memcg)
5311 memcg = get_mem_cgroup_from_mm(mm);
5312
5313 ret = try_charge(memcg, gfp_mask, nr_pages);
5314
5315 css_put(&memcg->css);
Johannes Weiner00501b52014-08-08 14:19:20 -07005316out:
5317 *memcgp = memcg;
5318 return ret;
5319}
5320
5321/**
5322 * mem_cgroup_commit_charge - commit a page charge
5323 * @page: page to charge
5324 * @memcg: memcg to charge the page to
5325 * @lrucare: page might be on LRU already
5326 *
5327 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5328 * after page->mapping has been set up. This must happen atomically
5329 * as part of the page instantiation, i.e. under the page table lock
5330 * for anonymous pages, under the page lock for page and swap cache.
5331 *
5332 * In addition, the page must not be on the LRU during the commit, to
5333 * prevent racing with task migration. If it might be, use @lrucare.
5334 *
5335 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5336 */
5337void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005338 bool lrucare, bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005339{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005340 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005341
5342 VM_BUG_ON_PAGE(!page->mapping, page);
5343 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5344
5345 if (mem_cgroup_disabled())
5346 return;
5347 /*
5348 * Swap faults will attempt to charge the same page multiple
5349 * times. But reuse_swap_page() might have removed the page
5350 * from swapcache already, so we can't check PageSwapCache().
5351 */
5352 if (!memcg)
5353 return;
5354
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005355 commit_charge(page, memcg, lrucare);
5356
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005357 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005358 mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005359 memcg_check_events(memcg, page);
5360 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005361
Johannes Weiner7941d212016-01-14 15:21:23 -08005362 if (do_memsw_account() && PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005363 swp_entry_t entry = { .val = page_private(page) };
5364 /*
5365 * The swap entry might not get freed for a long time,
5366 * let's not wait for it. The page already received a
5367 * memory+swap charge, drop the swap entry duplicate.
5368 */
5369 mem_cgroup_uncharge_swap(entry);
5370 }
5371}
5372
5373/**
5374 * mem_cgroup_cancel_charge - cancel a page charge
5375 * @page: page to charge
5376 * @memcg: memcg to charge the page to
5377 *
5378 * Cancel a charge transaction started by mem_cgroup_try_charge().
5379 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005380void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5381 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005382{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005383 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005384
5385 if (mem_cgroup_disabled())
5386 return;
5387 /*
5388 * Swap faults will attempt to charge the same page multiple
5389 * times. But reuse_swap_page() might have removed the page
5390 * from swapcache already, so we can't check PageSwapCache().
5391 */
5392 if (!memcg)
5393 return;
5394
Johannes Weiner00501b52014-08-08 14:19:20 -07005395 cancel_charge(memcg, nr_pages);
5396}
5397
Johannes Weiner747db952014-08-08 14:19:24 -07005398static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005399 unsigned long nr_anon, unsigned long nr_file,
5400 unsigned long nr_huge, struct page *dummy_page)
5401{
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005402 unsigned long nr_pages = nr_anon + nr_file;
Johannes Weiner747db952014-08-08 14:19:24 -07005403 unsigned long flags;
5404
Johannes Weinerce00a962014-09-05 08:43:57 -04005405 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005406 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08005407 if (do_memsw_account())
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005408 page_counter_uncharge(&memcg->memsw, nr_pages);
Johannes Weinerce00a962014-09-05 08:43:57 -04005409 memcg_oom_recover(memcg);
5410 }
Johannes Weiner747db952014-08-08 14:19:24 -07005411
5412 local_irq_save(flags);
5413 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5414 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5415 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5416 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005417 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005418 memcg_check_events(memcg, dummy_page);
5419 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005420
5421 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005422 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005423}
5424
5425static void uncharge_list(struct list_head *page_list)
5426{
5427 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005428 unsigned long nr_anon = 0;
5429 unsigned long nr_file = 0;
5430 unsigned long nr_huge = 0;
5431 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005432 struct list_head *next;
5433 struct page *page;
5434
Johannes Weiner8b592652016-03-17 14:20:31 -07005435 /*
5436 * Note that the list can be a single page->lru; hence the
5437 * do-while loop instead of a simple list_for_each_entry().
5438 */
Johannes Weiner747db952014-08-08 14:19:24 -07005439 next = page_list->next;
5440 do {
5441 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005442
5443 page = list_entry(next, struct page, lru);
5444 next = page->lru.next;
5445
5446 VM_BUG_ON_PAGE(PageLRU(page), page);
5447 VM_BUG_ON_PAGE(page_count(page), page);
5448
Johannes Weiner1306a852014-12-10 15:44:52 -08005449 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005450 continue;
5451
5452 /*
5453 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005454 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005455 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005456 */
5457
Johannes Weiner1306a852014-12-10 15:44:52 -08005458 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005459 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005460 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5461 nr_huge, page);
5462 pgpgout = nr_anon = nr_file = nr_huge = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005463 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005464 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005465 }
5466
5467 if (PageTransHuge(page)) {
5468 nr_pages <<= compound_order(page);
5469 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5470 nr_huge += nr_pages;
5471 }
5472
5473 if (PageAnon(page))
5474 nr_anon += nr_pages;
5475 else
5476 nr_file += nr_pages;
5477
Johannes Weiner1306a852014-12-10 15:44:52 -08005478 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005479
5480 pgpgout++;
5481 } while (next != page_list);
5482
5483 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005484 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5485 nr_huge, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005486}
5487
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005488/**
5489 * mem_cgroup_uncharge - uncharge a page
5490 * @page: page to uncharge
5491 *
5492 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5493 * mem_cgroup_commit_charge().
5494 */
5495void mem_cgroup_uncharge(struct page *page)
5496{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005497 if (mem_cgroup_disabled())
5498 return;
5499
Johannes Weiner747db952014-08-08 14:19:24 -07005500 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005501 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005502 return;
5503
Johannes Weiner747db952014-08-08 14:19:24 -07005504 INIT_LIST_HEAD(&page->lru);
5505 uncharge_list(&page->lru);
5506}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005507
Johannes Weiner747db952014-08-08 14:19:24 -07005508/**
5509 * mem_cgroup_uncharge_list - uncharge a list of page
5510 * @page_list: list of pages to uncharge
5511 *
5512 * Uncharge a list of pages previously charged with
5513 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5514 */
5515void mem_cgroup_uncharge_list(struct list_head *page_list)
5516{
5517 if (mem_cgroup_disabled())
5518 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005519
Johannes Weiner747db952014-08-08 14:19:24 -07005520 if (!list_empty(page_list))
5521 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005522}
5523
5524/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005525 * mem_cgroup_migrate - charge a page's replacement
5526 * @oldpage: currently circulating page
5527 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005528 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005529 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5530 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005531 *
5532 * Both pages must be locked, @newpage->mapping must be set up.
5533 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005534void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005535{
Johannes Weiner29833312014-12-10 15:44:02 -08005536 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005537 unsigned int nr_pages;
5538 bool compound;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005539
5540 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5541 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005542 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005543 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5544 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005545
5546 if (mem_cgroup_disabled())
5547 return;
5548
5549 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005550 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005551 return;
5552
Hugh Dickins45637ba2015-11-05 18:49:40 -08005553 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08005554 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005555 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005556 return;
5557
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005558 /* Force-charge the new page. The old one will be freed soon */
5559 compound = PageTransHuge(newpage);
5560 nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5561
5562 page_counter_charge(&memcg->memory, nr_pages);
5563 if (do_memsw_account())
5564 page_counter_charge(&memcg->memsw, nr_pages);
5565 css_get_many(&memcg->css, nr_pages);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005566
Johannes Weiner9cf76662016-03-15 14:57:58 -07005567 commit_charge(newpage, memcg, false);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005568
5569 local_irq_disable();
5570 mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5571 memcg_check_events(memcg, newpage);
5572 local_irq_enable();
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005573}
5574
Johannes Weineref129472016-01-14 15:21:34 -08005575DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08005576EXPORT_SYMBOL(memcg_sockets_enabled_key);
5577
5578void sock_update_memcg(struct sock *sk)
5579{
5580 struct mem_cgroup *memcg;
5581
5582 /* Socket cloning can throw us here with sk_cgrp already
5583 * filled. It won't however, necessarily happen from
5584 * process context. So the test for root memcg given
5585 * the current task's memcg won't help us in this case.
5586 *
5587 * Respecting the original socket's memcg is a better
5588 * decision in this case.
5589 */
5590 if (sk->sk_memcg) {
5591 BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5592 css_get(&sk->sk_memcg->css);
5593 return;
5594 }
5595
5596 rcu_read_lock();
5597 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005598 if (memcg == root_mem_cgroup)
5599 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08005600 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005601 goto out;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005602 if (css_tryget_online(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08005603 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005604out:
Johannes Weiner11092082016-01-14 15:21:26 -08005605 rcu_read_unlock();
5606}
5607EXPORT_SYMBOL(sock_update_memcg);
5608
5609void sock_release_memcg(struct sock *sk)
5610{
5611 WARN_ON(!sk->sk_memcg);
5612 css_put(&sk->sk_memcg->css);
5613}
5614
5615/**
5616 * mem_cgroup_charge_skmem - charge socket memory
5617 * @memcg: memcg to charge
5618 * @nr_pages: number of pages to charge
5619 *
5620 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5621 * @memcg's configured limit, %false if the charge had to be forced.
5622 */
5623bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5624{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005625 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08005626
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005627 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005628 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005629
Johannes Weiner0db15292016-01-20 15:02:50 -08005630 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5631 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005632 return true;
5633 }
Johannes Weiner0db15292016-01-20 15:02:50 -08005634 page_counter_charge(&memcg->tcpmem, nr_pages);
5635 memcg->tcpmem_pressure = 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005636 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08005637 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005638
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005639 /* Don't block in the packet receive path */
5640 if (in_softirq())
5641 gfp_mask = GFP_NOWAIT;
5642
Johannes Weinerb2807f02016-01-20 15:03:22 -08005643 this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5644
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005645 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5646 return true;
5647
5648 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005649 return false;
5650}
5651
5652/**
5653 * mem_cgroup_uncharge_skmem - uncharge socket memory
5654 * @memcg - memcg to uncharge
5655 * @nr_pages - number of pages to uncharge
5656 */
5657void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5658{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005659 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005660 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005661 return;
5662 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005663
Johannes Weinerb2807f02016-01-20 15:03:22 -08005664 this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5665
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005666 page_counter_uncharge(&memcg->memory, nr_pages);
5667 css_put_many(&memcg->css, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005668}
5669
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005670static int __init cgroup_memory(char *s)
5671{
5672 char *token;
5673
5674 while ((token = strsep(&s, ",")) != NULL) {
5675 if (!*token)
5676 continue;
5677 if (!strcmp(token, "nosocket"))
5678 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08005679 if (!strcmp(token, "nokmem"))
5680 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005681 }
5682 return 0;
5683}
5684__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08005685
Michal Hocko2d110852013-02-22 16:34:43 -08005686/*
Michal Hocko10813122013-02-22 16:35:41 -08005687 * subsys_initcall() for memory controller.
5688 *
5689 * Some parts like hotcpu_notifier() have to be initialized from this context
5690 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5691 * everything that doesn't depend on a specific mem_cgroup structure should
5692 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005693 */
5694static int __init mem_cgroup_init(void)
5695{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005696 int cpu, node;
5697
Michal Hocko2d110852013-02-22 16:34:43 -08005698 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005699
5700 for_each_possible_cpu(cpu)
5701 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5702 drain_local_stock);
5703
5704 for_each_node(node) {
5705 struct mem_cgroup_tree_per_node *rtpn;
5706 int zone;
5707
5708 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5709 node_online(node) ? node : NUMA_NO_NODE);
5710
5711 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5712 struct mem_cgroup_tree_per_zone *rtpz;
5713
5714 rtpz = &rtpn->rb_tree_per_zone[zone];
5715 rtpz->rb_root = RB_ROOT;
5716 spin_lock_init(&rtpz->lock);
5717 }
5718 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5719 }
5720
Michal Hocko2d110852013-02-22 16:34:43 -08005721 return 0;
5722}
5723subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005724
5725#ifdef CONFIG_MEMCG_SWAP
5726/**
5727 * mem_cgroup_swapout - transfer a memsw charge to swap
5728 * @page: page whose memsw charge to transfer
5729 * @entry: swap entry to move the charge to
5730 *
5731 * Transfer the memsw charge of @page to @entry.
5732 */
5733void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5734{
5735 struct mem_cgroup *memcg;
5736 unsigned short oldid;
5737
5738 VM_BUG_ON_PAGE(PageLRU(page), page);
5739 VM_BUG_ON_PAGE(page_count(page), page);
5740
Johannes Weiner7941d212016-01-14 15:21:23 -08005741 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08005742 return;
5743
5744 memcg = page->mem_cgroup;
5745
5746 /* Readahead page, never charged */
5747 if (!memcg)
5748 return;
5749
5750 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5751 VM_BUG_ON_PAGE(oldid, page);
5752 mem_cgroup_swap_statistics(memcg, true);
5753
5754 page->mem_cgroup = NULL;
5755
5756 if (!mem_cgroup_is_root(memcg))
5757 page_counter_uncharge(&memcg->memory, 1);
5758
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07005759 /*
5760 * Interrupts should be disabled here because the caller holds the
5761 * mapping->tree_lock lock which is taken with interrupts-off. It is
5762 * important here to have the interrupts disabled because it is the
5763 * only synchronisation we have for udpating the per-CPU variables.
5764 */
5765 VM_BUG_ON(!irqs_disabled());
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005766 mem_cgroup_charge_statistics(memcg, page, false, -1);
Johannes Weiner21afa382015-02-11 15:26:36 -08005767 memcg_check_events(memcg, page);
5768}
5769
Vladimir Davydov37e84352016-01-20 15:02:56 -08005770/*
5771 * mem_cgroup_try_charge_swap - try charging a swap entry
5772 * @page: page being added to swap
5773 * @entry: swap entry to charge
5774 *
5775 * Try to charge @entry to the memcg that @page belongs to.
5776 *
5777 * Returns 0 on success, -ENOMEM on failure.
5778 */
5779int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5780{
5781 struct mem_cgroup *memcg;
5782 struct page_counter *counter;
5783 unsigned short oldid;
5784
5785 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5786 return 0;
5787
5788 memcg = page->mem_cgroup;
5789
5790 /* Readahead page, never charged */
5791 if (!memcg)
5792 return 0;
5793
5794 if (!mem_cgroup_is_root(memcg) &&
5795 !page_counter_try_charge(&memcg->swap, 1, &counter))
5796 return -ENOMEM;
5797
5798 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5799 VM_BUG_ON_PAGE(oldid, page);
5800 mem_cgroup_swap_statistics(memcg, true);
5801
5802 css_get(&memcg->css);
5803 return 0;
5804}
5805
Johannes Weiner21afa382015-02-11 15:26:36 -08005806/**
5807 * mem_cgroup_uncharge_swap - uncharge a swap entry
5808 * @entry: swap entry to uncharge
5809 *
Vladimir Davydov37e84352016-01-20 15:02:56 -08005810 * Drop the swap charge associated with @entry.
Johannes Weiner21afa382015-02-11 15:26:36 -08005811 */
5812void mem_cgroup_uncharge_swap(swp_entry_t entry)
5813{
5814 struct mem_cgroup *memcg;
5815 unsigned short id;
5816
Vladimir Davydov37e84352016-01-20 15:02:56 -08005817 if (!do_swap_account)
Johannes Weiner21afa382015-02-11 15:26:36 -08005818 return;
5819
5820 id = swap_cgroup_record(entry, 0);
5821 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005822 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005823 if (memcg) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08005824 if (!mem_cgroup_is_root(memcg)) {
5825 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5826 page_counter_uncharge(&memcg->swap, 1);
5827 else
5828 page_counter_uncharge(&memcg->memsw, 1);
5829 }
Johannes Weiner21afa382015-02-11 15:26:36 -08005830 mem_cgroup_swap_statistics(memcg, false);
5831 css_put(&memcg->css);
5832 }
5833 rcu_read_unlock();
5834}
5835
Vladimir Davydovd8b38432016-01-20 15:03:07 -08005836long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
5837{
5838 long nr_swap_pages = get_nr_swap_pages();
5839
5840 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5841 return nr_swap_pages;
5842 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5843 nr_swap_pages = min_t(long, nr_swap_pages,
5844 READ_ONCE(memcg->swap.limit) -
5845 page_counter_read(&memcg->swap));
5846 return nr_swap_pages;
5847}
5848
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08005849bool mem_cgroup_swap_full(struct page *page)
5850{
5851 struct mem_cgroup *memcg;
5852
5853 VM_BUG_ON_PAGE(!PageLocked(page), page);
5854
5855 if (vm_swap_full())
5856 return true;
5857 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5858 return false;
5859
5860 memcg = page->mem_cgroup;
5861 if (!memcg)
5862 return false;
5863
5864 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5865 if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
5866 return true;
5867
5868 return false;
5869}
5870
Johannes Weiner21afa382015-02-11 15:26:36 -08005871/* for remember boot option*/
5872#ifdef CONFIG_MEMCG_SWAP_ENABLED
5873static int really_do_swap_account __initdata = 1;
5874#else
5875static int really_do_swap_account __initdata;
5876#endif
5877
5878static int __init enable_swap_account(char *s)
5879{
5880 if (!strcmp(s, "1"))
5881 really_do_swap_account = 1;
5882 else if (!strcmp(s, "0"))
5883 really_do_swap_account = 0;
5884 return 1;
5885}
5886__setup("swapaccount=", enable_swap_account);
5887
Vladimir Davydov37e84352016-01-20 15:02:56 -08005888static u64 swap_current_read(struct cgroup_subsys_state *css,
5889 struct cftype *cft)
5890{
5891 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5892
5893 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
5894}
5895
5896static int swap_max_show(struct seq_file *m, void *v)
5897{
5898 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5899 unsigned long max = READ_ONCE(memcg->swap.limit);
5900
5901 if (max == PAGE_COUNTER_MAX)
5902 seq_puts(m, "max\n");
5903 else
5904 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5905
5906 return 0;
5907}
5908
5909static ssize_t swap_max_write(struct kernfs_open_file *of,
5910 char *buf, size_t nbytes, loff_t off)
5911{
5912 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5913 unsigned long max;
5914 int err;
5915
5916 buf = strstrip(buf);
5917 err = page_counter_memparse(buf, "max", &max);
5918 if (err)
5919 return err;
5920
5921 mutex_lock(&memcg_limit_mutex);
5922 err = page_counter_limit(&memcg->swap, max);
5923 mutex_unlock(&memcg_limit_mutex);
5924 if (err)
5925 return err;
5926
5927 return nbytes;
5928}
5929
5930static struct cftype swap_files[] = {
5931 {
5932 .name = "swap.current",
5933 .flags = CFTYPE_NOT_ON_ROOT,
5934 .read_u64 = swap_current_read,
5935 },
5936 {
5937 .name = "swap.max",
5938 .flags = CFTYPE_NOT_ON_ROOT,
5939 .seq_show = swap_max_show,
5940 .write = swap_max_write,
5941 },
5942 { } /* terminate */
5943};
5944
Johannes Weiner21afa382015-02-11 15:26:36 -08005945static struct cftype memsw_cgroup_files[] = {
5946 {
5947 .name = "memsw.usage_in_bytes",
5948 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
5949 .read_u64 = mem_cgroup_read_u64,
5950 },
5951 {
5952 .name = "memsw.max_usage_in_bytes",
5953 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
5954 .write = mem_cgroup_reset,
5955 .read_u64 = mem_cgroup_read_u64,
5956 },
5957 {
5958 .name = "memsw.limit_in_bytes",
5959 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
5960 .write = mem_cgroup_write,
5961 .read_u64 = mem_cgroup_read_u64,
5962 },
5963 {
5964 .name = "memsw.failcnt",
5965 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
5966 .write = mem_cgroup_reset,
5967 .read_u64 = mem_cgroup_read_u64,
5968 },
5969 { }, /* terminate */
5970};
5971
5972static int __init mem_cgroup_swap_init(void)
5973{
5974 if (!mem_cgroup_disabled() && really_do_swap_account) {
5975 do_swap_account = 1;
Vladimir Davydov37e84352016-01-20 15:02:56 -08005976 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
5977 swap_files));
Johannes Weiner21afa382015-02-11 15:26:36 -08005978 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
5979 memsw_cgroup_files));
5980 }
5981 return 0;
5982}
5983subsys_initcall(mem_cgroup_swap_init);
5984
5985#endif /* CONFIG_MEMCG_SWAP */