blob: 5339c89dff6317510b2710e9ab2770c23ac71e1e [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Johannes Weiner1575e682015-04-14 15:44:51 -070017 * Native page reclaim
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
22 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080023 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
27 *
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
32 */
33
Johannes Weiner3e32cb22014-12-10 15:42:31 -080034#include <linux/page_counter.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080035#include <linux/memcontrol.h>
36#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080037#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080038#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080039#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080040#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080041#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080043#include <linux/bit_spinlock.h>
44#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070045#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040046#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080047#include <linux/mutex.h>
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -070048#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070049#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080051#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080052#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080053#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050054#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080055#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080056#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080057#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070058#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070059#include <linux/mm_inline.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080060#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080061#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070062#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070063#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050064#include <linux/file.h>
Tejun Heob23afb92015-11-05 18:46:11 -080065#include <linux/tracehook.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080066#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000067#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070068#include <net/ip.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080069#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080070
Balbir Singh8697d332008-02-07 00:13:59 -080071#include <asm/uaccess.h>
72
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070073#include <trace/events/vmscan.h>
74
Tejun Heo073219e2014-02-08 10:36:58 -050075struct cgroup_subsys memory_cgrp_subsys __read_mostly;
76EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080077
Johannes Weiner7d828602016-01-14 15:20:56 -080078struct mem_cgroup *root_mem_cgroup __read_mostly;
79
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070080#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh8cdea7c2008-02-07 00:13:50 -080081
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080082/* Socket memory accounting disabled? */
83static bool cgroup_memory_nosocket;
84
Vladimir Davydov04823c82016-01-20 15:02:38 -080085/* Kernel memory accounting disabled? */
86static bool cgroup_memory_nokmem;
87
Johannes Weiner21afa382015-02-11 15:26:36 -080088/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070089#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080090int do_swap_account __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080091#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070092#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080093#endif
94
Johannes Weiner7941d212016-01-14 15:21:23 -080095/* Whether legacy memory+swap accounting is active */
96static bool do_memsw_account(void)
97{
98 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account;
99}
100
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700101static const char * const mem_cgroup_stat_names[] = {
102 "cache",
103 "rss",
David Rientjesb070e652013-05-07 16:18:09 -0700104 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700105 "mapped_file",
Greg Thelenc4843a72015-05-22 17:13:16 -0400106 "dirty",
Sha Zhengju3ea67d02013-09-12 15:13:53 -0700107 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700108 "swap",
109};
110
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700111static const char * const mem_cgroup_events_names[] = {
112 "pgpgin",
113 "pgpgout",
114 "pgfault",
115 "pgmajfault",
116};
117
Sha Zhengju58cf1882013-02-22 16:32:05 -0800118static const char * const mem_cgroup_lru_names[] = {
119 "inactive_anon",
120 "active_anon",
121 "inactive_file",
122 "active_file",
123 "unevictable",
124};
125
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700126#define THRESHOLDS_EVENTS_TARGET 128
127#define SOFTLIMIT_EVENTS_TARGET 1024
128#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700129
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700130/*
131 * Cgroups above their limits are maintained in a RB-Tree, independent of
132 * their hierarchy representation
133 */
134
135struct mem_cgroup_tree_per_zone {
136 struct rb_root rb_root;
137 spinlock_t lock;
138};
139
140struct mem_cgroup_tree_per_node {
141 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
142};
143
144struct mem_cgroup_tree {
145 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
146};
147
148static struct mem_cgroup_tree soft_limit_tree __read_mostly;
149
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700150/* for OOM */
151struct mem_cgroup_eventfd_list {
152 struct list_head list;
153 struct eventfd_ctx *eventfd;
154};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800155
Tejun Heo79bd9812013-11-22 18:20:42 -0500156/*
157 * cgroup_event represents events which userspace want to receive.
158 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500159struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500160 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500161 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500162 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500163 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500164 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500165 * eventfd to signal userspace about the event.
166 */
167 struct eventfd_ctx *eventfd;
168 /*
169 * Each of these stored in a list by the cgroup.
170 */
171 struct list_head list;
172 /*
Tejun Heofba94802013-11-22 18:20:43 -0500173 * register_event() callback will be used to add new userspace
174 * waiter for changes related to this event. Use eventfd_signal()
175 * on eventfd to send notification to userspace.
176 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500177 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500178 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500179 /*
180 * unregister_event() callback will be called when userspace closes
181 * the eventfd or on cgroup removing. This callback must be set,
182 * if you want provide notification functionality.
183 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500184 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500185 struct eventfd_ctx *eventfd);
186 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500187 * All fields below needed to unregister event when
188 * userspace closes eventfd.
189 */
190 poll_table pt;
191 wait_queue_head_t *wqh;
192 wait_queue_t wait;
193 struct work_struct remove;
194};
195
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700196static void mem_cgroup_threshold(struct mem_cgroup *memcg);
197static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800198
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800199/* Stuffs for move charges at task migration. */
200/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800201 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800202 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800203#define MOVE_ANON 0x1U
204#define MOVE_FILE 0x2U
205#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800206
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800207/* "mc" and its members are protected by cgroup_mutex */
208static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800209 spinlock_t lock; /* for from, to */
Tejun Heo264a0ae2016-04-21 19:09:02 -0400210 struct mm_struct *mm;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800211 struct mem_cgroup *from;
212 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800213 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800214 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800215 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800216 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800217 struct task_struct *moving_task; /* a task moving charges */
218 wait_queue_head_t waitq; /* a waitq for other context */
219} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700220 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800221 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
222};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800223
Balbir Singh4e416952009-09-23 15:56:39 -0700224/*
225 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
226 * limit reclaim to prevent infinite loops, if they ever occur.
227 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700228#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700229#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700230
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800231enum charge_type {
232 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700233 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800234 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700235 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700236 NR_CHARGE_TYPE,
237};
238
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800239/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800240enum res_type {
241 _MEM,
242 _MEMSWAP,
243 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800244 _KMEM,
Vladimir Davydovd55f90b2016-01-20 15:02:44 -0800245 _TCP,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800246};
247
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700248#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
249#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800250#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700251/* Used for OOM nofiier */
252#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800253
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700254/* Some nice accessors for the vmpressure. */
255struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
256{
257 if (!memcg)
258 memcg = root_mem_cgroup;
259 return &memcg->vmpressure;
260}
261
262struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
263{
264 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
265}
266
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700267static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
268{
269 return (memcg == root_mem_cgroup);
270}
271
Johannes Weiner127424c2016-01-20 15:02:32 -0800272#ifndef CONFIG_SLOB
Glauber Costa55007d82012-12-18 14:22:38 -0800273/*
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800274 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800275 * The main reason for not using cgroup id for this:
276 * this works better in sparse environments, where we have a lot of memcgs,
277 * but only a few kmem-limited. Or also, if we have, for instance, 200
278 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
279 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800280 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800281 * The current size of the caches array is stored in memcg_nr_cache_ids. It
282 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800283 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800284static DEFINE_IDA(memcg_cache_ida);
285int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800286
Vladimir Davydov05257a12015-02-12 14:59:01 -0800287/* Protects memcg_nr_cache_ids */
288static DECLARE_RWSEM(memcg_cache_ids_sem);
289
290void memcg_get_cache_ids(void)
291{
292 down_read(&memcg_cache_ids_sem);
293}
294
295void memcg_put_cache_ids(void)
296{
297 up_read(&memcg_cache_ids_sem);
298}
299
Glauber Costa55007d82012-12-18 14:22:38 -0800300/*
301 * MIN_SIZE is different than 1, because we would like to avoid going through
302 * the alloc/free process all the time. In a small machine, 4 kmem-limited
303 * cgroups is a reasonable guess. In the future, it could be a parameter or
304 * tunable, but that is strictly not necessary.
305 *
Li Zefanb8627832013-09-23 16:56:47 +0800306 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800307 * this constant directly from cgroup, but it is understandable that this is
308 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800309 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800310 * increase ours as well if it increases.
311 */
312#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800313#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800314
Glauber Costad7f25f82012-12-18 14:22:40 -0800315/*
316 * A lot of the calls to the cache allocation functions are expected to be
317 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
318 * conditional to this static branch, we'll have to allow modules that does
319 * kmem_cache_alloc and the such to see this symbol as well
320 */
Johannes Weineref129472016-01-14 15:21:34 -0800321DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
Glauber Costad7f25f82012-12-18 14:22:40 -0800322EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800323
Johannes Weiner127424c2016-01-20 15:02:32 -0800324#endif /* !CONFIG_SLOB */
Glauber Costaa8964b92012-12-18 14:22:09 -0800325
Balbir Singhf64c3f52009-09-23 15:56:37 -0700326static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700327mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700328{
Jianyu Zhane2318752014-06-06 14:38:20 -0700329 int nid = zone_to_nid(zone);
330 int zid = zone_idx(zone);
331
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700332 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700333}
334
Tejun Heoad7fa852015-05-27 20:00:02 -0400335/**
336 * mem_cgroup_css_from_page - css of the memcg associated with a page
337 * @page: page of interest
338 *
339 * If memcg is bound to the default hierarchy, css of the memcg associated
340 * with @page is returned. The returned css remains associated with @page
341 * until it is released.
342 *
343 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
344 * is returned.
Tejun Heoad7fa852015-05-27 20:00:02 -0400345 */
346struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
347{
348 struct mem_cgroup *memcg;
349
Tejun Heoad7fa852015-05-27 20:00:02 -0400350 memcg = page->mem_cgroup;
351
Tejun Heo9e10a132015-09-18 11:56:28 -0400352 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400353 memcg = root_mem_cgroup;
354
Tejun Heoad7fa852015-05-27 20:00:02 -0400355 return &memcg->css;
356}
357
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700358/**
359 * page_cgroup_ino - return inode number of the memcg a page is charged to
360 * @page: the page
361 *
362 * Look up the closest online ancestor of the memory cgroup @page is charged to
363 * and return its inode number or 0 if @page is not charged to any cgroup. It
364 * is safe to call this function without holding a reference to @page.
365 *
366 * Note, this function is inherently racy, because there is nothing to prevent
367 * the cgroup inode from getting torn down and potentially reallocated a moment
368 * after page_cgroup_ino() returns, so it only should be used by callers that
369 * do not care (such as procfs interfaces).
370 */
371ino_t page_cgroup_ino(struct page *page)
372{
373 struct mem_cgroup *memcg;
374 unsigned long ino = 0;
375
376 rcu_read_lock();
377 memcg = READ_ONCE(page->mem_cgroup);
378 while (memcg && !(memcg->css.flags & CSS_ONLINE))
379 memcg = parent_mem_cgroup(memcg);
380 if (memcg)
381 ino = cgroup_ino(memcg->css.cgroup);
382 rcu_read_unlock();
383 return ino;
384}
385
Balbir Singhf64c3f52009-09-23 15:56:37 -0700386static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700387mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700388{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700389 int nid = page_to_nid(page);
390 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700391
Jianyu Zhane2318752014-06-06 14:38:20 -0700392 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700393}
394
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700395static struct mem_cgroup_tree_per_zone *
396soft_limit_tree_node_zone(int nid, int zid)
397{
398 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
399}
400
401static struct mem_cgroup_tree_per_zone *
402soft_limit_tree_from_page(struct page *page)
403{
404 int nid = page_to_nid(page);
405 int zid = page_zonenum(page);
406
407 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
408}
409
Johannes Weinercf2c8122014-06-06 14:38:21 -0700410static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz,
411 struct mem_cgroup_tree_per_zone *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800412 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700413{
414 struct rb_node **p = &mctz->rb_root.rb_node;
415 struct rb_node *parent = NULL;
416 struct mem_cgroup_per_zone *mz_node;
417
418 if (mz->on_tree)
419 return;
420
421 mz->usage_in_excess = new_usage_in_excess;
422 if (!mz->usage_in_excess)
423 return;
424 while (*p) {
425 parent = *p;
426 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
427 tree_node);
428 if (mz->usage_in_excess < mz_node->usage_in_excess)
429 p = &(*p)->rb_left;
430 /*
431 * We can't avoid mem cgroups that are over their soft
432 * limit by the same amount
433 */
434 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
435 p = &(*p)->rb_right;
436 }
437 rb_link_node(&mz->tree_node, parent, p);
438 rb_insert_color(&mz->tree_node, &mctz->rb_root);
439 mz->on_tree = true;
440}
441
Johannes Weinercf2c8122014-06-06 14:38:21 -0700442static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
443 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700444{
445 if (!mz->on_tree)
446 return;
447 rb_erase(&mz->tree_node, &mctz->rb_root);
448 mz->on_tree = false;
449}
450
Johannes Weinercf2c8122014-06-06 14:38:21 -0700451static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
452 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700453{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700454 unsigned long flags;
455
456 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700457 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700458 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700459}
460
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800461static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
462{
463 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700464 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800465 unsigned long excess = 0;
466
467 if (nr_pages > soft_limit)
468 excess = nr_pages - soft_limit;
469
470 return excess;
471}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700472
473static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
474{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800475 unsigned long excess;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700476 struct mem_cgroup_per_zone *mz;
477 struct mem_cgroup_tree_per_zone *mctz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700478
Jianyu Zhane2318752014-06-06 14:38:20 -0700479 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700480 /*
481 * Necessary to update all ancestors when hierarchy is used.
482 * because their event counter is not touched.
483 */
484 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Jianyu Zhane2318752014-06-06 14:38:20 -0700485 mz = mem_cgroup_page_zoneinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800486 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700487 /*
488 * We have to update the tree if mz is on RB-tree or
489 * mem is over its softlimit.
490 */
491 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700492 unsigned long flags;
493
494 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700495 /* if on-tree, remove it */
496 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700497 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700498 /*
499 * Insert again. mz->usage_in_excess will be updated.
500 * If excess is 0, no tree ops.
501 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700502 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700503 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700504 }
505 }
506}
507
508static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
509{
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700510 struct mem_cgroup_tree_per_zone *mctz;
Jianyu Zhane2318752014-06-06 14:38:20 -0700511 struct mem_cgroup_per_zone *mz;
512 int nid, zid;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700513
Jianyu Zhane2318752014-06-06 14:38:20 -0700514 for_each_node(nid) {
515 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
516 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
517 mctz = soft_limit_tree_node_zone(nid, zid);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700518 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700519 }
520 }
521}
522
523static struct mem_cgroup_per_zone *
524__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
525{
526 struct rb_node *rightmost = NULL;
527 struct mem_cgroup_per_zone *mz;
528
529retry:
530 mz = NULL;
531 rightmost = rb_last(&mctz->rb_root);
532 if (!rightmost)
533 goto done; /* Nothing to reclaim from */
534
535 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
536 /*
537 * Remove the node now but someone else can add it back,
538 * we will to add it back at the end of reclaim to its correct
539 * position in the tree.
540 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700541 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800542 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400543 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700544 goto retry;
545done:
546 return mz;
547}
548
549static struct mem_cgroup_per_zone *
550mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
551{
552 struct mem_cgroup_per_zone *mz;
553
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700554 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700555 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700556 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700557 return mz;
558}
559
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700560/*
Greg Thelen484ebb32015-10-01 15:37:05 -0700561 * Return page count for single (non recursive) @memcg.
562 *
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700563 * Implementation Note: reading percpu statistics for memcg.
564 *
565 * Both of vmstat[] and percpu_counter has threshold and do periodic
566 * synchronization to implement "quick" read. There are trade-off between
567 * reading cost and precision of value. Then, we may have a chance to implement
Greg Thelen484ebb32015-10-01 15:37:05 -0700568 * a periodic synchronization of counter in memcg's counter.
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700569 *
570 * But this _read() function is used for user interface now. The user accounts
571 * memory usage by memory cgroup and he _always_ requires exact value because
572 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
573 * have to visit all online cpus and make sum. So, for now, unnecessary
574 * synchronization is not implemented. (just implemented for cpu hotplug)
575 *
576 * If there are kernel internal actions which can make use of some not-exact
577 * value, and reading all cpu value can be performance bottleneck in some
Greg Thelen484ebb32015-10-01 15:37:05 -0700578 * common workload, threshold and synchronization as vmstat[] should be
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700579 * implemented.
580 */
Greg Thelen484ebb32015-10-01 15:37:05 -0700581static unsigned long
582mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800583{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700584 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800585 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800586
Greg Thelen484ebb32015-10-01 15:37:05 -0700587 /* Per-cpu values can be negative, use a signed accumulator */
Tejun Heo733a5722015-05-22 18:23:18 -0400588 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700589 val += per_cpu(memcg->stat->count[idx], cpu);
Greg Thelen484ebb32015-10-01 15:37:05 -0700590 /*
591 * Summing races with updates, so val may be negative. Avoid exposing
592 * transient negative values.
593 */
594 if (val < 0)
595 val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800596 return val;
597}
598
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700599static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700600 enum mem_cgroup_events_index idx)
601{
602 unsigned long val = 0;
603 int cpu;
604
Tejun Heo733a5722015-05-22 18:23:18 -0400605 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700606 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700607 return val;
608}
609
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700610static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700611 struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800612 bool compound, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800613{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700614 /*
615 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
616 * counted as CACHE even if it's on ANON LRU.
617 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700618 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700619 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700620 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800621 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700622 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700623 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700624
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800625 if (compound) {
626 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
David Rientjesb070e652013-05-07 16:18:09 -0700627 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
628 nr_pages);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800629 }
David Rientjesb070e652013-05-07 16:18:09 -0700630
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800631 /* pagein of a big page is an event. So, ignore page size */
632 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700633 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800634 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700635 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800636 nr_pages = -nr_pages; /* for event */
637 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800638
Johannes Weiner13114712012-05-29 15:07:07 -0700639 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800640}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800641
Vladimir Davydov0a6b76d2016-03-17 14:18:42 -0700642unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
643 int nid, unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700644{
Jianyu Zhane2318752014-06-06 14:38:20 -0700645 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700646 int zid;
647
Jianyu Zhane2318752014-06-06 14:38:20 -0700648 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700649
Jianyu Zhane2318752014-06-06 14:38:20 -0700650 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
651 struct mem_cgroup_per_zone *mz;
652 enum lru_list lru;
653
654 for_each_lru(lru) {
655 if (!(BIT(lru) & lru_mask))
656 continue;
657 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
658 nr += mz->lru_size[lru];
659 }
660 }
661 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700662}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700663
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700664static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700665 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800666{
Jianyu Zhane2318752014-06-06 14:38:20 -0700667 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700668 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800669
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800670 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700671 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
672 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800673}
674
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800675static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
676 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800677{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700678 unsigned long val, next;
679
Johannes Weiner13114712012-05-29 15:07:07 -0700680 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700681 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700682 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800683 if ((long)next - (long)val < 0) {
684 switch (target) {
685 case MEM_CGROUP_TARGET_THRESH:
686 next = val + THRESHOLDS_EVENTS_TARGET;
687 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700688 case MEM_CGROUP_TARGET_SOFTLIMIT:
689 next = val + SOFTLIMIT_EVENTS_TARGET;
690 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800691 case MEM_CGROUP_TARGET_NUMAINFO:
692 next = val + NUMAINFO_EVENTS_TARGET;
693 break;
694 default:
695 break;
696 }
697 __this_cpu_write(memcg->stat->targets[target], next);
698 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700699 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800700 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800701}
702
703/*
704 * Check events in order.
705 *
706 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700707static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800708{
709 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800710 if (unlikely(mem_cgroup_event_ratelimit(memcg,
711 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700712 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800713 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800714
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700715 do_softlimit = mem_cgroup_event_ratelimit(memcg,
716 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700717#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800718 do_numainfo = mem_cgroup_event_ratelimit(memcg,
719 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700720#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800721 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700722 if (unlikely(do_softlimit))
723 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800724#if MAX_NUMNODES > 1
725 if (unlikely(do_numainfo))
726 atomic_inc(&memcg->numainfo_events);
727#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700728 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800729}
730
Balbir Singhcf475ad2008-04-29 01:00:16 -0700731struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800732{
Balbir Singh31a78f22008-09-28 23:09:31 +0100733 /*
734 * mm_update_next_owner() may clear mm->owner to NULL
735 * if it races with swapoff, page migration, etc.
736 * So this can be called with p == NULL.
737 */
738 if (unlikely(!p))
739 return NULL;
740
Tejun Heo073219e2014-02-08 10:36:58 -0500741 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800742}
Michal Hocko33398cf2015-09-08 15:01:02 -0700743EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800744
Johannes Weinerdf381972014-04-07 15:37:43 -0700745static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800746{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700747 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700748
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800749 rcu_read_lock();
750 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700751 /*
752 * Page cache insertions can happen withou an
753 * actual mm context, e.g. during disk probing
754 * on boot, loopback IO, acct() writes etc.
755 */
756 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700757 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700758 else {
759 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
760 if (unlikely(!memcg))
761 memcg = root_mem_cgroup;
762 }
Tejun Heoec903c02014-05-13 12:11:01 -0400763 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800764 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700765 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800766}
767
Johannes Weiner56600482012-01-12 17:17:59 -0800768/**
769 * mem_cgroup_iter - iterate over memory cgroup hierarchy
770 * @root: hierarchy root
771 * @prev: previously returned memcg, NULL on first invocation
772 * @reclaim: cookie for shared reclaim walks, NULL for full walks
773 *
774 * Returns references to children of the hierarchy below @root, or
775 * @root itself, or %NULL after a full round-trip.
776 *
777 * Caller must pass the return value in @prev on subsequent
778 * invocations for reference counting, or use mem_cgroup_iter_break()
779 * to cancel a hierarchy walk before the round-trip is complete.
780 *
781 * Reclaimers can specify a zone and a priority level in @reclaim to
782 * divide up the memcgs in the hierarchy among all concurrent
783 * reclaimers operating on the same zone and priority.
784 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700785struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800786 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700787 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700788{
Michal Hocko33398cf2015-09-08 15:01:02 -0700789 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800790 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800791 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800792 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700793
Andrew Morton694fbc02013-09-24 15:27:37 -0700794 if (mem_cgroup_disabled())
795 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800796
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700797 if (!root)
798 root = root_mem_cgroup;
799
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800800 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800801 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800802
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800803 if (!root->use_hierarchy && root != root_mem_cgroup) {
804 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800805 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700806 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800807 }
808
Michal Hocko542f85f2013-04-29 15:07:15 -0700809 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800810
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800811 if (reclaim) {
812 struct mem_cgroup_per_zone *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800813
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800814 mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone);
815 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700816
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800817 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700818 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800819
Vladimir Davydov6df38682015-12-29 14:54:10 -0800820 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -0700821 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800822 if (!pos || css_tryget(&pos->css))
823 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800824 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800825 * css reference reached zero, so iter->position will
826 * be cleared by ->css_released. However, we should not
827 * rely on this happening soon, because ->css_released
828 * is called from a work queue, and by busy-waiting we
829 * might block it. So we clear iter->position right
830 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800831 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800832 (void)cmpxchg(&iter->position, pos, NULL);
833 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800834 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800835
836 if (pos)
837 css = &pos->css;
838
839 for (;;) {
840 css = css_next_descendant_pre(css, &root->css);
841 if (!css) {
842 /*
843 * Reclaimers share the hierarchy walk, and a
844 * new one might jump in right at the end of
845 * the hierarchy - make sure they see at least
846 * one group and restart from the beginning.
847 */
848 if (!prev)
849 continue;
850 break;
851 }
852
853 /*
854 * Verify the css and acquire a reference. The root
855 * is provided by the caller, so we know it's alive
856 * and kicking, and don't take an extra reference.
857 */
858 memcg = mem_cgroup_from_css(css);
859
860 if (css == &root->css)
861 break;
862
Johannes Weiner0b8f73e2016-01-20 15:02:53 -0800863 if (css_tryget(css))
864 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800865
866 memcg = NULL;
867 }
868
869 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800870 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800871 * The position could have already been updated by a competing
872 * thread, so check that the value hasn't changed since we read
873 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800874 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800875 (void)cmpxchg(&iter->position, pos, memcg);
876
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800877 if (pos)
878 css_put(&pos->css);
879
880 if (!memcg)
881 iter->generation++;
882 else if (!prev)
883 reclaim->generation = iter->generation;
884 }
885
Michal Hocko542f85f2013-04-29 15:07:15 -0700886out_unlock:
887 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800888out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700889 if (prev && prev != root)
890 css_put(&prev->css);
891
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800892 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700893}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800894
Johannes Weiner56600482012-01-12 17:17:59 -0800895/**
896 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
897 * @root: hierarchy root
898 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
899 */
900void mem_cgroup_iter_break(struct mem_cgroup *root,
901 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800902{
903 if (!root)
904 root = root_mem_cgroup;
905 if (prev && prev != root)
906 css_put(&prev->css);
907}
908
Vladimir Davydov6df38682015-12-29 14:54:10 -0800909static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
910{
911 struct mem_cgroup *memcg = dead_memcg;
912 struct mem_cgroup_reclaim_iter *iter;
913 struct mem_cgroup_per_zone *mz;
914 int nid, zid;
915 int i;
916
917 while ((memcg = parent_mem_cgroup(memcg))) {
918 for_each_node(nid) {
919 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
920 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
921 for (i = 0; i <= DEF_PRIORITY; i++) {
922 iter = &mz->iter[i];
923 cmpxchg(&iter->position,
924 dead_memcg, NULL);
925 }
926 }
927 }
928 }
929}
930
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700931/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800932 * Iteration constructs for visiting all cgroups (under a tree). If
933 * loops are exited prematurely (break), mem_cgroup_iter_break() must
934 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700935 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800936#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800937 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800938 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800939 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700940
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800941#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800942 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800943 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800944 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700945
Johannes Weiner925b7672012-01-12 17:18:15 -0800946/**
947 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
948 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700949 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -0800950 *
951 * Returns the lru list vector holding pages for the given @zone and
952 * @mem. This can be the global zone lruvec, if the memory controller
953 * is disabled.
954 */
955struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
956 struct mem_cgroup *memcg)
957{
958 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800959 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -0800960
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800961 if (mem_cgroup_disabled()) {
962 lruvec = &zone->lruvec;
963 goto out;
964 }
Johannes Weiner925b7672012-01-12 17:18:15 -0800965
Jianyu Zhane2318752014-06-06 14:38:20 -0700966 mz = mem_cgroup_zone_zoneinfo(memcg, zone);
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800967 lruvec = &mz->lruvec;
968out:
969 /*
970 * Since a node can be onlined after the mem_cgroup was created,
971 * we have to be prepared to initialize lruvec->zone here;
972 * and if offlined then reonlined, we need to reinitialize it.
973 */
974 if (unlikely(lruvec->zone != zone))
975 lruvec->zone = zone;
976 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -0800977}
978
Johannes Weiner925b7672012-01-12 17:18:15 -0800979/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800980 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -0800981 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700982 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800983 *
984 * This function is only safe when following the LRU page isolation
985 * and putback protocol: the LRU lock must be held, and the page must
986 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -0700987 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700988struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -0700989{
990 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -0800991 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800992 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800993
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800994 if (mem_cgroup_disabled()) {
995 lruvec = &zone->lruvec;
996 goto out;
997 }
Christoph Lameterb69408e2008-10-18 20:26:14 -0700998
Johannes Weiner1306a852014-12-10 15:44:52 -0800999 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001000 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001001 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -08001002 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -08001003 */
Johannes Weiner29833312014-12-10 15:44:02 -08001004 if (!memcg)
1005 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001006
Jianyu Zhane2318752014-06-06 14:38:20 -07001007 mz = mem_cgroup_page_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001008 lruvec = &mz->lruvec;
1009out:
1010 /*
1011 * Since a node can be onlined after the mem_cgroup was created,
1012 * we have to be prepared to initialize lruvec->zone here;
1013 * and if offlined then reonlined, we need to reinitialize it.
1014 */
1015 if (unlikely(lruvec->zone != zone))
1016 lruvec->zone = zone;
1017 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001018}
1019
1020/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001021 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1022 * @lruvec: mem_cgroup per zone lru vector
1023 * @lru: index of lru list the page is sitting on
1024 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001025 *
Hugh 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);
Li RongQingcbedbac2016-05-27 14:27:43 -07001111 else
1112 margin = 0;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001113 }
1114
1115 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001116}
1117
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001118/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001119 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001120 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001121 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1122 * moving cgroups. This is for waiting at high-memory pressure
1123 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001124 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001125static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001126{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001127 struct mem_cgroup *from;
1128 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001129 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001130 /*
1131 * Unlike task_move routines, we access mc.to, mc.from not under
1132 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1133 */
1134 spin_lock(&mc.lock);
1135 from = mc.from;
1136 to = mc.to;
1137 if (!from)
1138 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001139
Johannes Weiner2314b422014-12-10 15:44:33 -08001140 ret = mem_cgroup_is_descendant(from, memcg) ||
1141 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001142unlock:
1143 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001144 return ret;
1145}
1146
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001147static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001148{
1149 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001150 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001151 DEFINE_WAIT(wait);
1152 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1153 /* moving charge context might have finished. */
1154 if (mc.moving_task)
1155 schedule();
1156 finish_wait(&mc.waitq, &wait);
1157 return true;
1158 }
1159 }
1160 return false;
1161}
1162
Sha Zhengju58cf1882013-02-22 16:32:05 -08001163#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001164/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001165 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001166 * @memcg: The memory cgroup that went over limit
1167 * @p: Task that is going to be killed
1168 *
1169 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1170 * enabled
1171 */
1172void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1173{
Sha Zhengju58cf1882013-02-22 16:32:05 -08001174 struct mem_cgroup *iter;
1175 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001176
Balbir Singhe2224322009-04-02 16:57:39 -07001177 rcu_read_lock();
1178
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001179 if (p) {
1180 pr_info("Task in ");
1181 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1182 pr_cont(" killed as a result of limit of ");
1183 } else {
1184 pr_info("Memory limit reached of cgroup ");
1185 }
1186
Tejun Heoe61734c2014-02-12 09:29:50 -05001187 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001188 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001189
Balbir Singhe2224322009-04-02 16:57:39 -07001190 rcu_read_unlock();
1191
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001192 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1193 K((u64)page_counter_read(&memcg->memory)),
1194 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1195 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1196 K((u64)page_counter_read(&memcg->memsw)),
1197 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1198 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1199 K((u64)page_counter_read(&memcg->kmem)),
1200 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001201
1202 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001203 pr_info("Memory cgroup stats for ");
1204 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001205 pr_cont(":");
1206
1207 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08001208 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Sha Zhengju58cf1882013-02-22 16:32:05 -08001209 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07001210 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
Sha Zhengju58cf1882013-02-22 16:32:05 -08001211 K(mem_cgroup_read_stat(iter, i)));
1212 }
1213
1214 for (i = 0; i < NR_LRU_LISTS; i++)
1215 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1216 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1217
1218 pr_cont("\n");
1219 }
Balbir Singhe2224322009-04-02 16:57:39 -07001220}
1221
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001222/*
1223 * This function returns the number of memcg under hierarchy tree. Returns
1224 * 1(self count) if no children.
1225 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001226static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001227{
1228 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001229 struct mem_cgroup *iter;
1230
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001231 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001232 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001233 return num;
1234}
1235
Balbir Singh6d61ef42009-01-07 18:08:06 -08001236/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001237 * Return the memory (and swap, if configured) limit for a memcg.
1238 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001239static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001240{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001241 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001242
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001243 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001244 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001245 unsigned long memsw_limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001246 unsigned long swap_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001247
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001248 memsw_limit = memcg->memsw.limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001249 swap_limit = memcg->swap.limit;
1250 swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1251 limit = min(limit + swap_limit, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001252 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001253 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001254}
1255
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001256static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001257 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001258{
David Rientjes6e0fc462015-09-08 15:00:36 -07001259 struct oom_control oc = {
1260 .zonelist = NULL,
1261 .nodemask = NULL,
1262 .gfp_mask = gfp_mask,
1263 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001264 };
David Rientjes9cbb78b2012-07-31 16:43:44 -07001265 struct mem_cgroup *iter;
1266 unsigned long chosen_points = 0;
1267 unsigned long totalpages;
1268 unsigned int points = 0;
1269 struct task_struct *chosen = NULL;
1270
Johannes Weinerdc564012015-06-24 16:57:19 -07001271 mutex_lock(&oom_lock);
1272
David Rientjes876aafb2012-07-31 16:43:48 -07001273 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001274 * If current has a pending SIGKILL or is exiting, then automatically
1275 * select it. The goal is to allow it to allocate so that it may
1276 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001277 */
Oleg Nesterovd003f372014-12-12 16:56:24 -08001278 if (fatal_signal_pending(current) || task_will_free_mem(current)) {
Johannes Weiner16e95192015-06-24 16:57:07 -07001279 mark_oom_victim(current);
Michal Hocko3ef22df2016-05-19 17:13:12 -07001280 try_oom_reaper(current);
Johannes Weinerdc564012015-06-24 16:57:19 -07001281 goto unlock;
David Rientjes876aafb2012-07-31 16:43:48 -07001282 }
1283
David Rientjes6e0fc462015-09-08 15:00:36 -07001284 check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001285 totalpages = mem_cgroup_get_limit(memcg) ? : 1;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001286 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001287 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001288 struct task_struct *task;
1289
Tejun Heo72ec7022013-08-08 20:11:26 -04001290 css_task_iter_start(&iter->css, &it);
1291 while ((task = css_task_iter_next(&it))) {
David Rientjes6e0fc462015-09-08 15:00:36 -07001292 switch (oom_scan_process_thread(&oc, task, totalpages)) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001293 case OOM_SCAN_SELECT:
1294 if (chosen)
1295 put_task_struct(chosen);
1296 chosen = task;
1297 chosen_points = ULONG_MAX;
1298 get_task_struct(chosen);
1299 /* fall through */
1300 case OOM_SCAN_CONTINUE:
1301 continue;
1302 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001303 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001304 mem_cgroup_iter_break(memcg, iter);
1305 if (chosen)
1306 put_task_struct(chosen);
Tetsuo Handa1ebab2d2016-05-26 15:16:19 -07001307 /* Set a dummy value to return "true". */
1308 chosen = (void *) 1;
Johannes Weinerdc564012015-06-24 16:57:19 -07001309 goto unlock;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001310 case OOM_SCAN_OK:
1311 break;
1312 };
1313 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001314 if (!points || points < chosen_points)
1315 continue;
1316 /* Prefer thread group leaders for display purposes */
1317 if (points == chosen_points &&
1318 thread_group_leader(chosen))
1319 continue;
1320
1321 if (chosen)
1322 put_task_struct(chosen);
1323 chosen = task;
1324 chosen_points = points;
1325 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001326 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001327 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001328 }
1329
Johannes Weinerdc564012015-06-24 16:57:19 -07001330 if (chosen) {
1331 points = chosen_points * 1000 / totalpages;
David Rientjes6e0fc462015-09-08 15:00:36 -07001332 oom_kill_process(&oc, chosen, points, totalpages, memcg,
1333 "Memory cgroup out of memory");
Johannes Weinerdc564012015-06-24 16:57:19 -07001334 }
1335unlock:
1336 mutex_unlock(&oom_lock);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001337 return chosen;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001338}
1339
Michele Curtiae6e71d2014-12-12 16:56:35 -08001340#if MAX_NUMNODES > 1
1341
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001342/**
1343 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001344 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001345 * @nid: the node ID to be checked.
1346 * @noswap : specify true here if the user wants flle only information.
1347 *
1348 * This function returns whether the specified memcg contains any
1349 * reclaimable pages on a node. Returns true if there are any reclaimable
1350 * pages in the node.
1351 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001352static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001353 int nid, bool noswap)
1354{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001355 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001356 return true;
1357 if (noswap || !total_swap_pages)
1358 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001359 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001360 return true;
1361 return false;
1362
1363}
Ying Han889976d2011-05-26 16:25:33 -07001364
1365/*
1366 * Always updating the nodemask is not very good - even if we have an empty
1367 * list or the wrong list here, we can start from some node and traverse all
1368 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1369 *
1370 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001371static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001372{
1373 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001374 /*
1375 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1376 * pagein/pageout changes since the last update.
1377 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001378 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001379 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001380 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001381 return;
1382
Ying Han889976d2011-05-26 16:25:33 -07001383 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001384 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001385
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001386 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001387
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001388 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1389 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001390 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001391
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001392 atomic_set(&memcg->numainfo_events, 0);
1393 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001394}
1395
1396/*
1397 * Selecting a node where we start reclaim from. Because what we need is just
1398 * reducing usage counter, start from anywhere is O,K. Considering
1399 * memory reclaim from current node, there are pros. and cons.
1400 *
1401 * Freeing memory from current node means freeing memory from a node which
1402 * we'll use or we've used. So, it may make LRU bad. And if several threads
1403 * hit limits, it will see a contention on a node. But freeing from remote
1404 * node means more costs for memory reclaim because of memory latency.
1405 *
1406 * Now, we use round-robin. Better algorithm is welcomed.
1407 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001408int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001409{
1410 int node;
1411
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001412 mem_cgroup_may_update_nodemask(memcg);
1413 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001414
Andrew Morton0edaf862016-05-19 17:10:58 -07001415 node = next_node_in(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001416 /*
Michal Hockofda3d692016-05-19 17:11:34 -07001417 * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1418 * last time it really checked all the LRUs due to rate limiting.
1419 * Fallback to the current node in that case for simplicity.
Ying Han889976d2011-05-26 16:25:33 -07001420 */
1421 if (unlikely(node == MAX_NUMNODES))
1422 node = numa_node_id();
1423
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001424 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001425 return node;
1426}
Ying Han889976d2011-05-26 16:25:33 -07001427#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001428int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001429{
1430 return 0;
1431}
1432#endif
1433
Andrew Morton0608f432013-09-24 15:27:41 -07001434static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1435 struct zone *zone,
1436 gfp_t gfp_mask,
1437 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001438{
Andrew Morton0608f432013-09-24 15:27:41 -07001439 struct mem_cgroup *victim = NULL;
1440 int total = 0;
1441 int loop = 0;
1442 unsigned long excess;
1443 unsigned long nr_scanned;
1444 struct mem_cgroup_reclaim_cookie reclaim = {
1445 .zone = zone,
1446 .priority = 0,
1447 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001448
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001449 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001450
Andrew Morton0608f432013-09-24 15:27:41 -07001451 while (1) {
1452 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1453 if (!victim) {
1454 loop++;
1455 if (loop >= 2) {
1456 /*
1457 * If we have not been able to reclaim
1458 * anything, it might because there are
1459 * no reclaimable pages under this hierarchy
1460 */
1461 if (!total)
1462 break;
1463 /*
1464 * We want to do more targeted reclaim.
1465 * excess >> 2 is not to excessive so as to
1466 * reclaim too much, nor too less that we keep
1467 * coming back to reclaim from this cgroup
1468 */
1469 if (total >= (excess >> 2) ||
1470 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1471 break;
1472 }
1473 continue;
1474 }
Andrew Morton0608f432013-09-24 15:27:41 -07001475 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1476 zone, &nr_scanned);
1477 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001478 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001479 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001480 }
Andrew Morton0608f432013-09-24 15:27:41 -07001481 mem_cgroup_iter_break(root_memcg, victim);
1482 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001483}
1484
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001485#ifdef CONFIG_LOCKDEP
1486static struct lockdep_map memcg_oom_lock_dep_map = {
1487 .name = "memcg_oom_lock",
1488};
1489#endif
1490
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001491static DEFINE_SPINLOCK(memcg_oom_lock);
1492
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001493/*
1494 * Check OOM-Killer is already running under our hierarchy.
1495 * If someone is running, return false.
1496 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001497static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001498{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001499 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001500
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001501 spin_lock(&memcg_oom_lock);
1502
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001503 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001504 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001505 /*
1506 * this subtree of our hierarchy is already locked
1507 * so we cannot give a lock.
1508 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001509 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001510 mem_cgroup_iter_break(memcg, iter);
1511 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001512 } else
1513 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001514 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001515
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001516 if (failed) {
1517 /*
1518 * OK, we failed to lock the whole subtree so we have
1519 * to clean up what we set up to the failing subtree
1520 */
1521 for_each_mem_cgroup_tree(iter, memcg) {
1522 if (iter == failed) {
1523 mem_cgroup_iter_break(memcg, iter);
1524 break;
1525 }
1526 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001527 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001528 } else
1529 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001530
1531 spin_unlock(&memcg_oom_lock);
1532
1533 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001534}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001535
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001536static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001537{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001538 struct mem_cgroup *iter;
1539
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001540 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001541 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001542 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001543 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001544 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001545}
1546
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001547static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001548{
1549 struct mem_cgroup *iter;
1550
Tejun Heoc2b42d32015-06-24 16:58:23 -07001551 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001552 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001553 iter->under_oom++;
1554 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001555}
1556
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001557static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001558{
1559 struct mem_cgroup *iter;
1560
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001561 /*
1562 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001563 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001564 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001565 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001566 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001567 if (iter->under_oom > 0)
1568 iter->under_oom--;
1569 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001570}
1571
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001572static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1573
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001574struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001575 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001576 wait_queue_t wait;
1577};
1578
1579static int memcg_oom_wake_function(wait_queue_t *wait,
1580 unsigned mode, int sync, void *arg)
1581{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001582 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1583 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001584 struct oom_wait_info *oom_wait_info;
1585
1586 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001587 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001588
Johannes Weiner2314b422014-12-10 15:44:33 -08001589 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1590 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001591 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001592 return autoremove_wake_function(wait, mode, sync, arg);
1593}
1594
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001595static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001596{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001597 /*
1598 * For the following lockless ->under_oom test, the only required
1599 * guarantee is that it must see the state asserted by an OOM when
1600 * this function is called as a result of userland actions
1601 * triggered by the notification of the OOM. This is trivially
1602 * achieved by invoking mem_cgroup_mark_under_oom() before
1603 * triggering notification.
1604 */
1605 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001606 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001607}
1608
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001609static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001610{
Andrew Mortond0db7af2016-06-08 15:33:47 -07001611 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001612 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001613 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001614 * We are in the middle of the charge context here, so we
1615 * don't want to block when potentially sitting on a callstack
1616 * that holds all kinds of filesystem and mm locks.
1617 *
1618 * Also, the caller may handle a failed allocation gracefully
1619 * (like optional page cache readahead) and so an OOM killer
1620 * invocation might not even be necessary.
1621 *
1622 * That's why we don't do anything here except remember the
1623 * OOM context and then deal with it at the end of the page
1624 * fault when the stack is unwound, the locks are released,
1625 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001626 */
Johannes Weiner49426422013-10-16 13:46:59 -07001627 css_get(&memcg->css);
Tejun Heo626ebc42015-11-05 18:46:09 -08001628 current->memcg_in_oom = memcg;
1629 current->memcg_oom_gfp_mask = mask;
1630 current->memcg_oom_order = order;
Johannes Weiner49426422013-10-16 13:46:59 -07001631}
1632
1633/**
1634 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1635 * @handle: actually kill/wait or just clean up the OOM state
1636 *
1637 * This has to be called at the end of a page fault if the memcg OOM
1638 * handler was enabled.
1639 *
1640 * Memcg supports userspace OOM handling where failed allocations must
1641 * sleep on a waitqueue until the userspace task resolves the
1642 * situation. Sleeping directly in the charge context with all kinds
1643 * of locks held is not a good idea, instead we remember an OOM state
1644 * in the task and mem_cgroup_oom_synchronize() has to be called at
1645 * the end of the page fault to complete the OOM handling.
1646 *
1647 * Returns %true if an ongoing memcg OOM situation was detected and
1648 * completed, %false otherwise.
1649 */
1650bool mem_cgroup_oom_synchronize(bool handle)
1651{
Tejun Heo626ebc42015-11-05 18:46:09 -08001652 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001653 struct oom_wait_info owait;
1654 bool locked;
1655
1656 /* OOM is global, do not handle */
1657 if (!memcg)
1658 return false;
1659
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001660 if (!handle || oom_killer_disabled)
Johannes Weiner49426422013-10-16 13:46:59 -07001661 goto cleanup;
1662
1663 owait.memcg = memcg;
1664 owait.wait.flags = 0;
1665 owait.wait.func = memcg_oom_wake_function;
1666 owait.wait.private = current;
1667 INIT_LIST_HEAD(&owait.wait.task_list);
1668
1669 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001670 mem_cgroup_mark_under_oom(memcg);
1671
1672 locked = mem_cgroup_oom_trylock(memcg);
1673
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001674 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001675 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001676
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001677 if (locked && !memcg->oom_kill_disable) {
1678 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001679 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001680 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1681 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001682 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001683 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001684 mem_cgroup_unmark_under_oom(memcg);
1685 finish_wait(&memcg_oom_waitq, &owait.wait);
1686 }
1687
1688 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001689 mem_cgroup_oom_unlock(memcg);
1690 /*
1691 * There is no guarantee that an OOM-lock contender
1692 * sees the wakeups triggered by the OOM kill
1693 * uncharges. Wake any sleepers explicitely.
1694 */
1695 memcg_oom_recover(memcg);
1696 }
Johannes Weiner49426422013-10-16 13:46:59 -07001697cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001698 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001699 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001700 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001701}
1702
Johannes Weinerd7365e72014-10-29 14:50:48 -07001703/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001704 * lock_page_memcg - lock a page->mem_cgroup binding
1705 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001706 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001707 * This function protects unlocked LRU pages from being moved to
1708 * another cgroup and stabilizes their page->mem_cgroup binding.
Balbir Singhd69b0422009-06-17 16:26:34 -07001709 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001710void lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001711{
1712 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001713 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001714
Johannes Weiner6de22612015-02-11 15:25:01 -08001715 /*
1716 * The RCU lock is held throughout the transaction. The fast
1717 * path can get away without acquiring the memcg->move_lock
1718 * because page moving starts with an RCU grace period.
Johannes Weiner6de22612015-02-11 15:25:01 -08001719 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001720 rcu_read_lock();
1721
1722 if (mem_cgroup_disabled())
Johannes Weiner62cccb82016-03-15 14:57:22 -07001723 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001724again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001725 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001726 if (unlikely(!memcg))
Johannes Weiner62cccb82016-03-15 14:57:22 -07001727 return;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001728
Qiang Huangbdcbb652014-06-04 16:08:21 -07001729 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001730 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001731
Johannes Weiner6de22612015-02-11 15:25:01 -08001732 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001733 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001734 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001735 goto again;
1736 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001737
1738 /*
1739 * When charge migration first begins, we can have locked and
1740 * unlocked page stat updates happening concurrently. Track
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001741 * the task who has the lock for unlock_page_memcg().
Johannes Weiner6de22612015-02-11 15:25:01 -08001742 */
1743 memcg->move_lock_task = current;
1744 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001745
Johannes Weiner62cccb82016-03-15 14:57:22 -07001746 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001747}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001748EXPORT_SYMBOL(lock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001749
Johannes Weinerd7365e72014-10-29 14:50:48 -07001750/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001751 * unlock_page_memcg - unlock a page->mem_cgroup binding
Johannes Weiner62cccb82016-03-15 14:57:22 -07001752 * @page: the page
Johannes Weinerd7365e72014-10-29 14:50:48 -07001753 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001754void unlock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001755{
Johannes Weiner62cccb82016-03-15 14:57:22 -07001756 struct mem_cgroup *memcg = page->mem_cgroup;
1757
Johannes Weiner6de22612015-02-11 15:25:01 -08001758 if (memcg && memcg->move_lock_task == current) {
1759 unsigned long flags = memcg->move_lock_flags;
1760
1761 memcg->move_lock_task = NULL;
1762 memcg->move_lock_flags = 0;
1763
1764 spin_unlock_irqrestore(&memcg->move_lock, flags);
1765 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001766
Johannes Weinerd7365e72014-10-29 14:50:48 -07001767 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001768}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001769EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001770
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001771/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001772 * size of first charge trial. "32" comes from vmscan.c's magic value.
1773 * TODO: maybe necessary to use big numbers in big irons.
1774 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001775#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001776struct memcg_stock_pcp {
1777 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001778 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001779 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001780 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001781#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001782};
1783static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001784static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001785
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001786/**
1787 * consume_stock: Try to consume stocked charge on this cpu.
1788 * @memcg: memcg to consume from.
1789 * @nr_pages: how many pages to charge.
1790 *
1791 * The charges will only happen if @memcg matches the current cpu's memcg
1792 * stock, and at least @nr_pages are available in that stock. Failure to
1793 * service an allocation will refill the stock.
1794 *
1795 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001796 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001797static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001798{
1799 struct memcg_stock_pcp *stock;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001800 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001801
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001802 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001803 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001804
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001805 stock = &get_cpu_var(memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001806 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001807 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001808 ret = true;
1809 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001810 put_cpu_var(memcg_stock);
1811 return ret;
1812}
1813
1814/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001815 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001816 */
1817static void drain_stock(struct memcg_stock_pcp *stock)
1818{
1819 struct mem_cgroup *old = stock->cached;
1820
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001821 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001822 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001823 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001824 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001825 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001826 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001827 }
1828 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001829}
1830
1831/*
1832 * This must be called under preempt disabled or must be called by
1833 * a thread which is pinned to local cpu.
1834 */
1835static void drain_local_stock(struct work_struct *dummy)
1836{
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001837 struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001838 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001839 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001840}
1841
1842/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001843 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001844 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001845 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001846static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001847{
1848 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1849
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001850 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001851 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001852 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001853 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001854 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001855 put_cpu_var(memcg_stock);
1856}
1857
1858/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001859 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001860 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001861 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001862static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001863{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001864 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001865
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001866 /* If someone's already draining, avoid adding running more workers. */
1867 if (!mutex_trylock(&percpu_charge_mutex))
1868 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001869 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001870 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001871 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001872 for_each_online_cpu(cpu) {
1873 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001874 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001875
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001876 memcg = stock->cached;
1877 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001878 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001879 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001880 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001881 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1882 if (cpu == curcpu)
1883 drain_local_stock(&stock->work);
1884 else
1885 schedule_work_on(cpu, &stock->work);
1886 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001887 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001888 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001889 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001890 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001891}
1892
Paul Gortmaker0db06282013-06-19 14:53:51 -04001893static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001894 unsigned long action,
1895 void *hcpu)
1896{
1897 int cpu = (unsigned long)hcpu;
1898 struct memcg_stock_pcp *stock;
1899
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001900 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001901 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001902
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07001903 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001904 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001905
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001906 stock = &per_cpu(memcg_stock, cpu);
1907 drain_stock(stock);
1908 return NOTIFY_OK;
1909}
1910
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001911static void reclaim_high(struct mem_cgroup *memcg,
1912 unsigned int nr_pages,
1913 gfp_t gfp_mask)
1914{
1915 do {
1916 if (page_counter_read(&memcg->memory) <= memcg->high)
1917 continue;
1918 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
1919 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1920 } while ((memcg = parent_mem_cgroup(memcg)));
1921}
1922
1923static void high_work_func(struct work_struct *work)
1924{
1925 struct mem_cgroup *memcg;
1926
1927 memcg = container_of(work, struct mem_cgroup, high_work);
1928 reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1929}
1930
Tejun Heob23afb92015-11-05 18:46:11 -08001931/*
1932 * Scheduled by try_charge() to be executed from the userland return path
1933 * and reclaims memory over the high limit.
1934 */
1935void mem_cgroup_handle_over_high(void)
1936{
1937 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001938 struct mem_cgroup *memcg;
Tejun Heob23afb92015-11-05 18:46:11 -08001939
1940 if (likely(!nr_pages))
1941 return;
1942
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001943 memcg = get_mem_cgroup_from_mm(current->mm);
1944 reclaim_high(memcg, nr_pages, GFP_KERNEL);
Tejun Heob23afb92015-11-05 18:46:11 -08001945 css_put(&memcg->css);
1946 current->memcg_nr_pages_over_high = 0;
1947}
1948
Johannes Weiner00501b52014-08-08 14:19:20 -07001949static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1950 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001951{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001952 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001953 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001954 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001955 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001956 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001957 bool may_swap = true;
1958 bool drained = false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001959
Johannes Weinerce00a962014-09-05 08:43:57 -04001960 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08001961 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001962retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001963 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08001964 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001965
Johannes Weiner7941d212016-01-14 15:21:23 -08001966 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08001967 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1968 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001969 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08001970 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001971 page_counter_uncharge(&memcg->memsw, batch);
1972 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001973 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001974 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001975 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001976 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001977
Johannes Weiner6539cc02014-08-06 16:05:42 -07001978 if (batch > nr_pages) {
1979 batch = nr_pages;
1980 goto retry;
1981 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001982
Johannes Weiner06b078f2014-08-06 16:05:44 -07001983 /*
1984 * Unlike in global OOM situations, memcg is not in a physical
1985 * memory shortage. Allow dying and OOM-killed tasks to
1986 * bypass the last charges so that they can exit quickly and
1987 * free their memory.
1988 */
1989 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1990 fatal_signal_pending(current) ||
1991 current->flags & PF_EXITING))
Tejun Heo10d53c72015-11-05 18:46:17 -08001992 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001993
1994 if (unlikely(task_in_memcg_oom(current)))
1995 goto nomem;
1996
Mel Gormand0164ad2015-11-06 16:28:21 -08001997 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001998 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001999
Johannes Weiner241994e2015-02-11 15:26:06 -08002000 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
2001
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002002 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
2003 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002004
Johannes Weiner61e02c72014-08-06 16:08:16 -07002005 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07002006 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07002007
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002008 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002009 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002010 drained = true;
2011 goto retry;
2012 }
2013
Johannes Weiner28c34c22014-08-06 16:05:47 -07002014 if (gfp_mask & __GFP_NORETRY)
2015 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002016 /*
2017 * Even though the limit is exceeded at this point, reclaim
2018 * may have been able to free some pages. Retry the charge
2019 * before killing the task.
2020 *
2021 * Only for regular pages, though: huge pages are rather
2022 * unlikely to succeed so close to the limit, and we fall back
2023 * to regular pages anyway in case of failure.
2024 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002025 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002026 goto retry;
2027 /*
2028 * At task move, charge accounts can be doubly counted. So, it's
2029 * better to wait until the end of task_move if something is going on.
2030 */
2031 if (mem_cgroup_wait_acct_move(mem_over_limit))
2032 goto retry;
2033
Johannes Weiner9b130612014-08-06 16:05:51 -07002034 if (nr_retries--)
2035 goto retry;
2036
Johannes Weiner06b078f2014-08-06 16:05:44 -07002037 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08002038 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07002039
Johannes Weiner6539cc02014-08-06 16:05:42 -07002040 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08002041 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002042
Johannes Weiner241994e2015-02-11 15:26:06 -08002043 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
2044
Jerome Marchand3608de02015-11-05 18:47:29 -08002045 mem_cgroup_oom(mem_over_limit, gfp_mask,
2046 get_order(nr_pages * PAGE_SIZE));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002047nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002048 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002049 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08002050force:
2051 /*
2052 * The allocation either can't fail or will lead to more memory
2053 * being freed very soon. Allow memory usage go over the limit
2054 * temporarily by force charging it.
2055 */
2056 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002057 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08002058 page_counter_charge(&memcg->memsw, nr_pages);
2059 css_get_many(&memcg->css, nr_pages);
2060
2061 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002062
2063done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002064 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002065 if (batch > nr_pages)
2066 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002067
Johannes Weiner241994e2015-02-11 15:26:06 -08002068 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002069 * If the hierarchy is above the normal consumption range, schedule
2070 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002071 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002072 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2073 * not recorded as it most likely matches current's and won't
2074 * change in the meantime. As high limit is checked again before
2075 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994e2015-02-11 15:26:06 -08002076 */
2077 do {
Tejun Heob23afb92015-11-05 18:46:11 -08002078 if (page_counter_read(&memcg->memory) > memcg->high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002079 /* Don't bother a random interrupted task */
2080 if (in_interrupt()) {
2081 schedule_work(&memcg->high_work);
2082 break;
2083 }
Vladimir Davydov9516a182015-12-11 13:40:24 -08002084 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002085 set_notify_resume(current);
2086 break;
2087 }
Johannes Weiner241994e2015-02-11 15:26:06 -08002088 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002089
2090 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002091}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002092
Johannes Weiner00501b52014-08-08 14:19:20 -07002093static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002094{
Johannes Weinerce00a962014-09-05 08:43:57 -04002095 if (mem_cgroup_is_root(memcg))
2096 return;
2097
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002098 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002099 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002100 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002101
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002102 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002103}
2104
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002105static void lock_page_lru(struct page *page, int *isolated)
2106{
2107 struct zone *zone = page_zone(page);
2108
2109 spin_lock_irq(&zone->lru_lock);
2110 if (PageLRU(page)) {
2111 struct lruvec *lruvec;
2112
2113 lruvec = mem_cgroup_page_lruvec(page, zone);
2114 ClearPageLRU(page);
2115 del_page_from_lru_list(page, lruvec, page_lru(page));
2116 *isolated = 1;
2117 } else
2118 *isolated = 0;
2119}
2120
2121static void unlock_page_lru(struct page *page, int isolated)
2122{
2123 struct zone *zone = page_zone(page);
2124
2125 if (isolated) {
2126 struct lruvec *lruvec;
2127
2128 lruvec = mem_cgroup_page_lruvec(page, zone);
2129 VM_BUG_ON_PAGE(PageLRU(page), page);
2130 SetPageLRU(page);
2131 add_page_to_lru_list(page, lruvec, page_lru(page));
2132 }
2133 spin_unlock_irq(&zone->lru_lock);
2134}
2135
Johannes Weiner00501b52014-08-08 14:19:20 -07002136static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002137 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002138{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002139 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002140
Johannes Weiner1306a852014-12-10 15:44:52 -08002141 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002142
2143 /*
2144 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2145 * may already be on some other mem_cgroup's LRU. Take care of it.
2146 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002147 if (lrucare)
2148 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002149
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002150 /*
2151 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002152 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002153 *
2154 * - the page is uncharged
2155 *
2156 * - the page is off-LRU
2157 *
2158 * - an anonymous fault has exclusive page access, except for
2159 * a locked page table
2160 *
2161 * - a page cache insertion, a swapin fault, or a migration
2162 * have the page locked
2163 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002164 page->mem_cgroup = memcg;
Hugh Dickins3be91272008-02-07 00:14:19 -08002165
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002166 if (lrucare)
2167 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002168}
2169
Johannes Weiner127424c2016-01-20 15:02:32 -08002170#ifndef CONFIG_SLOB
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002171static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002172{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002173 int id, size;
2174 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002175
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002176 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002177 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2178 if (id < 0)
2179 return id;
2180
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002181 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002182 return id;
2183
2184 /*
2185 * There's no space for the new id in memcg_caches arrays,
2186 * so we have to grow them.
2187 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002188 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002189
2190 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002191 if (size < MEMCG_CACHES_MIN_SIZE)
2192 size = MEMCG_CACHES_MIN_SIZE;
2193 else if (size > MEMCG_CACHES_MAX_SIZE)
2194 size = MEMCG_CACHES_MAX_SIZE;
2195
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002196 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002197 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002198 err = memcg_update_all_list_lrus(size);
2199 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002200 memcg_nr_cache_ids = size;
2201
2202 up_write(&memcg_cache_ids_sem);
2203
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002204 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002205 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002206 return err;
2207 }
2208 return id;
2209}
2210
2211static void memcg_free_cache_id(int id)
2212{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002213 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002214}
2215
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002216struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002217 struct mem_cgroup *memcg;
2218 struct kmem_cache *cachep;
2219 struct work_struct work;
2220};
2221
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002222static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002223{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002224 struct memcg_kmem_cache_create_work *cw =
2225 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002226 struct mem_cgroup *memcg = cw->memcg;
2227 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002228
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002229 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002230
Vladimir Davydov5722d092014-04-07 15:39:24 -07002231 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002232 kfree(cw);
2233}
2234
2235/*
2236 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002237 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002238static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2239 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002240{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002241 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002242
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002243 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002244 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002245 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002246
2247 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002248
2249 cw->memcg = memcg;
2250 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002251 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002252
Glauber Costad7f25f82012-12-18 14:22:40 -08002253 schedule_work(&cw->work);
2254}
2255
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002256static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2257 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002258{
2259 /*
2260 * We need to stop accounting when we kmalloc, because if the
2261 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002262 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002263 *
2264 * However, it is better to enclose the whole function. Depending on
2265 * the debugging options enabled, INIT_WORK(), for instance, can
2266 * trigger an allocation. This too, will make us recurse. Because at
2267 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2268 * the safest choice is to do it like this, wrapping the whole function.
2269 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002270 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002271 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002272 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002273}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002274
Glauber Costad7f25f82012-12-18 14:22:40 -08002275/*
2276 * Return the kmem_cache we're supposed to use for a slab allocation.
2277 * We try to use the current memcg's version of the cache.
2278 *
2279 * If the cache does not exist yet, if we are the first user of it,
2280 * we either create it immediately, if possible, or create it asynchronously
2281 * in a workqueue.
2282 * In the latter case, we will let the current allocation go through with
2283 * the original cache.
2284 *
2285 * Can't be called in interrupt context or from kernel threads.
2286 * This function needs to be called with rcu_read_lock() held.
2287 */
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002288struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
Glauber Costad7f25f82012-12-18 14:22:40 -08002289{
2290 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002291 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002292 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002293
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002294 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002295
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002296 if (cachep->flags & SLAB_ACCOUNT)
2297 gfp |= __GFP_ACCOUNT;
2298
2299 if (!(gfp & __GFP_ACCOUNT))
2300 return cachep;
2301
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002302 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002303 return cachep;
2304
Vladimir Davydov8135be52014-12-12 16:56:38 -08002305 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002306 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002307 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002308 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002309
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002310 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002311 if (likely(memcg_cachep))
2312 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002313
2314 /*
2315 * If we are in a safe context (can wait, and not in interrupt
2316 * context), we could be be predictable and return right away.
2317 * This would guarantee that the allocation being performed
2318 * already belongs in the new cache.
2319 *
2320 * However, there are some clashes that can arrive from locking.
2321 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002322 * memcg_create_kmem_cache, this means no further allocation
2323 * could happen with the slab_mutex held. So it's better to
2324 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002325 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002326 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002327out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002328 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002329 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002330}
Glauber Costad7f25f82012-12-18 14:22:40 -08002331
Vladimir Davydov8135be52014-12-12 16:56:38 -08002332void __memcg_kmem_put_cache(struct kmem_cache *cachep)
2333{
2334 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002335 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002336}
2337
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002338int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2339 struct mem_cgroup *memcg)
2340{
2341 unsigned int nr_pages = 1 << order;
2342 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002343 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002344
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002345 ret = try_charge(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08002346 if (ret)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002347 return ret;
Johannes Weiner52c29b02016-01-20 15:02:35 -08002348
2349 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2350 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2351 cancel_charge(memcg, nr_pages);
2352 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002353 }
2354
2355 page->mem_cgroup = memcg;
2356
2357 return 0;
2358}
2359
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002360int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002361{
2362 struct mem_cgroup *memcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002363 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002364
Johannes Weinerdf381972014-04-07 15:37:43 -07002365 memcg = get_mem_cgroup_from_mm(current->mm);
Vladimir Davydovb6ecd2d2016-03-17 14:18:33 -07002366 if (!mem_cgroup_is_root(memcg))
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002367 ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002368 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002369 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002370}
2371
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002372void __memcg_kmem_uncharge(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002373{
Johannes Weiner1306a852014-12-10 15:44:52 -08002374 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002375 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002376
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002377 if (!memcg)
2378 return;
2379
Sasha Levin309381fea2014-01-23 15:52:54 -08002380 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002381
Johannes Weiner52c29b02016-01-20 15:02:35 -08002382 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2383 page_counter_uncharge(&memcg->kmem, nr_pages);
2384
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002385 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002386 if (do_memsw_account())
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002387 page_counter_uncharge(&memcg->memsw, nr_pages);
2388
Johannes Weiner1306a852014-12-10 15:44:52 -08002389 page->mem_cgroup = NULL;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002390 css_put_many(&memcg->css, nr_pages);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002391}
Johannes Weiner127424c2016-01-20 15:02:32 -08002392#endif /* !CONFIG_SLOB */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002393
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002394#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2395
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002396/*
2397 * Because tail pages are not marked as "used", set it. We're under
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08002398 * zone->lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002399 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002400void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002401{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002402 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002403
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002404 if (mem_cgroup_disabled())
2405 return;
David Rientjesb070e652013-05-07 16:18:09 -07002406
Johannes Weiner29833312014-12-10 15:44:02 -08002407 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002408 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002409
Johannes Weiner1306a852014-12-10 15:44:52 -08002410 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002411 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002412}
Hugh Dickins12d27102012-01-12 17:19:52 -08002413#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002414
Andrew Mortonc255a452012-07-31 16:43:02 -07002415#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002416static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2417 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002418{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002419 int val = (charge) ? 1 : -1;
2420 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002421}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002422
2423/**
2424 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2425 * @entry: swap entry to be moved
2426 * @from: mem_cgroup which the entry is moved from
2427 * @to: mem_cgroup which the entry is moved to
2428 *
2429 * It succeeds only when the swap_cgroup's record for this entry is the same
2430 * as the mem_cgroup's id of @from.
2431 *
2432 * Returns 0 on success, -EINVAL on failure.
2433 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002434 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002435 * both res and memsw, and called css_get().
2436 */
2437static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002438 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002439{
2440 unsigned short old_id, new_id;
2441
Li Zefan34c00c32013-09-23 16:56:01 +08002442 old_id = mem_cgroup_id(from);
2443 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002444
2445 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002446 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002447 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002448 return 0;
2449 }
2450 return -EINVAL;
2451}
2452#else
2453static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002454 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002455{
2456 return -EINVAL;
2457}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002458#endif
2459
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002460static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002461
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002462static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002463 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002464{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002465 unsigned long curusage;
2466 unsigned long oldusage;
2467 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002468 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002469 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002470
2471 /*
2472 * For keeping hierarchical_reclaim simple, how long we should retry
2473 * is depends on callers. We set our retry-count to be function
2474 * of # of children which we should visit in this loop.
2475 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002476 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2477 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002478
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002479 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002480
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002481 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002482 if (signal_pending(current)) {
2483 ret = -EINTR;
2484 break;
2485 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002486
2487 mutex_lock(&memcg_limit_mutex);
2488 if (limit > memcg->memsw.limit) {
2489 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002490 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002491 break;
2492 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002493 if (limit > memcg->memory.limit)
2494 enlarge = true;
2495 ret = page_counter_limit(&memcg->memory, limit);
2496 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002497
2498 if (!ret)
2499 break;
2500
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002501 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2502
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002503 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002504 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002505 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002506 retry_count--;
2507 else
2508 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002509 } while (retry_count);
2510
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002511 if (!ret && enlarge)
2512 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002513
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002514 return ret;
2515}
2516
Li Zefan338c8432009-06-17 16:27:15 -07002517static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002518 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002519{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002520 unsigned long curusage;
2521 unsigned long oldusage;
2522 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002523 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002524 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002525
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002526 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002527 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2528 mem_cgroup_count_children(memcg);
2529
2530 oldusage = page_counter_read(&memcg->memsw);
2531
2532 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002533 if (signal_pending(current)) {
2534 ret = -EINTR;
2535 break;
2536 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002537
2538 mutex_lock(&memcg_limit_mutex);
2539 if (limit < memcg->memory.limit) {
2540 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002541 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002542 break;
2543 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002544 if (limit > memcg->memsw.limit)
2545 enlarge = true;
2546 ret = page_counter_limit(&memcg->memsw, limit);
2547 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002548
2549 if (!ret)
2550 break;
2551
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002552 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2553
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002554 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002555 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002556 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002557 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002558 else
2559 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002560 } while (retry_count);
2561
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002562 if (!ret && enlarge)
2563 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002564
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002565 return ret;
2566}
2567
Andrew Morton0608f432013-09-24 15:27:41 -07002568unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2569 gfp_t gfp_mask,
2570 unsigned long *total_scanned)
2571{
2572 unsigned long nr_reclaimed = 0;
2573 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2574 unsigned long reclaimed;
2575 int loop = 0;
2576 struct mem_cgroup_tree_per_zone *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002577 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002578 unsigned long nr_scanned;
2579
2580 if (order > 0)
2581 return 0;
2582
2583 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
2584 /*
2585 * This loop can run a while, specially if mem_cgroup's continuously
2586 * keep exceeding their soft limit and putting the system under
2587 * pressure
2588 */
2589 do {
2590 if (next_mz)
2591 mz = next_mz;
2592 else
2593 mz = mem_cgroup_largest_soft_limit_node(mctz);
2594 if (!mz)
2595 break;
2596
2597 nr_scanned = 0;
2598 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
2599 gfp_mask, &nr_scanned);
2600 nr_reclaimed += reclaimed;
2601 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002602 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002603 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002604
2605 /*
2606 * If we failed to reclaim anything from this memory cgroup
2607 * it is time to move on to the next cgroup
2608 */
2609 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002610 if (!reclaimed)
2611 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2612
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002613 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002614 /*
2615 * One school of thought says that we should not add
2616 * back the node to the tree if reclaim returns 0.
2617 * But our reclaim could return 0, simply because due
2618 * to priority we are exposing a smaller subset of
2619 * memory to reclaim from. Consider this as a longer
2620 * term TODO.
2621 */
2622 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002623 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002624 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002625 css_put(&mz->memcg->css);
2626 loop++;
2627 /*
2628 * Could not reclaim anything and there are no more
2629 * mem cgroups to try or we seem to be looping without
2630 * reclaiming anything.
2631 */
2632 if (!nr_reclaimed &&
2633 (next_mz == NULL ||
2634 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2635 break;
2636 } while (!nr_reclaimed);
2637 if (next_mz)
2638 css_put(&next_mz->memcg->css);
2639 return nr_reclaimed;
2640}
2641
Tejun Heoea280e72014-05-16 13:22:48 -04002642/*
2643 * Test whether @memcg has children, dead or alive. Note that this
2644 * function doesn't care whether @memcg has use_hierarchy enabled and
2645 * returns %true if there are child csses according to the cgroup
2646 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2647 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002648static inline bool memcg_has_children(struct mem_cgroup *memcg)
2649{
Tejun Heoea280e72014-05-16 13:22:48 -04002650 bool ret;
2651
Tejun Heoea280e72014-05-16 13:22:48 -04002652 rcu_read_lock();
2653 ret = css_next_child(NULL, &memcg->css);
2654 rcu_read_unlock();
2655 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002656}
2657
2658/*
Greg Thelen51038172016-05-20 16:58:18 -07002659 * Reclaims as many pages from the given memcg as possible.
Michal Hockoc26251f2012-10-26 13:37:28 +02002660 *
2661 * Caller is responsible for holding css reference for memcg.
2662 */
2663static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2664{
2665 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002666
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002667 /* we call try-to-free pages for make this cgroup empty */
2668 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002669 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002670 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002671 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002672
Michal Hockoc26251f2012-10-26 13:37:28 +02002673 if (signal_pending(current))
2674 return -EINTR;
2675
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002676 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2677 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002678 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002679 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002680 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002681 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002682 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002683
2684 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002685
2686 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002687}
2688
Tejun Heo6770c642014-05-13 12:16:21 -04002689static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2690 char *buf, size_t nbytes,
2691 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002692{
Tejun Heo6770c642014-05-13 12:16:21 -04002693 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002694
Michal Hockod8423012012-10-26 13:37:29 +02002695 if (mem_cgroup_is_root(memcg))
2696 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002697 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002698}
2699
Tejun Heo182446d2013-08-08 20:11:24 -04002700static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2701 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002702{
Tejun Heo182446d2013-08-08 20:11:24 -04002703 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002704}
2705
Tejun Heo182446d2013-08-08 20:11:24 -04002706static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2707 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002708{
2709 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002710 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002711 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002712
Glauber Costa567fb432012-07-31 16:43:07 -07002713 if (memcg->use_hierarchy == val)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002714 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07002715
Balbir Singh18f59ea2009-01-07 18:08:07 -08002716 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002717 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002718 * in the child subtrees. If it is unset, then the change can
2719 * occur, provided the current cgroup has no children.
2720 *
2721 * For the root cgroup, parent_mem is NULL, we allow value to be
2722 * set if there are no children.
2723 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002724 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002725 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002726 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002727 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002728 else
2729 retval = -EBUSY;
2730 } else
2731 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002732
Balbir Singh18f59ea2009-01-07 18:08:07 -08002733 return retval;
2734}
2735
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002736static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
Johannes Weinerce00a962014-09-05 08:43:57 -04002737{
2738 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002739 int i;
Johannes Weinerce00a962014-09-05 08:43:57 -04002740
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002741 memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
Johannes Weinerce00a962014-09-05 08:43:57 -04002742
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002743 for_each_mem_cgroup_tree(iter, memcg) {
2744 for (i = 0; i < MEMCG_NR_STAT; i++)
2745 stat[i] += mem_cgroup_read_stat(iter, i);
2746 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002747}
2748
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002749static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
Johannes Weiner587d9f72016-01-20 15:03:19 -08002750{
2751 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002752 int i;
Johannes Weiner587d9f72016-01-20 15:03:19 -08002753
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002754 memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
Johannes Weiner587d9f72016-01-20 15:03:19 -08002755
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002756 for_each_mem_cgroup_tree(iter, memcg) {
2757 for (i = 0; i < MEMCG_NR_EVENTS; i++)
2758 events[i] += mem_cgroup_read_events(iter, i);
2759 }
Johannes Weiner587d9f72016-01-20 15:03:19 -08002760}
2761
Andrew Morton6f646152015-11-06 16:28:58 -08002762static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04002763{
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002764 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002765
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002766 if (mem_cgroup_is_root(memcg)) {
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002767 struct mem_cgroup *iter;
2768
2769 for_each_mem_cgroup_tree(iter, memcg) {
2770 val += mem_cgroup_read_stat(iter,
2771 MEM_CGROUP_STAT_CACHE);
2772 val += mem_cgroup_read_stat(iter,
2773 MEM_CGROUP_STAT_RSS);
2774 if (swap)
2775 val += mem_cgroup_read_stat(iter,
2776 MEM_CGROUP_STAT_SWAP);
2777 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002778 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002779 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002780 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002781 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002782 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002783 }
Michal Hockoc12176d2015-11-05 18:50:29 -08002784 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002785}
2786
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002787enum {
2788 RES_USAGE,
2789 RES_LIMIT,
2790 RES_MAX_USAGE,
2791 RES_FAILCNT,
2792 RES_SOFT_LIMIT,
2793};
Johannes Weinerce00a962014-09-05 08:43:57 -04002794
Tejun Heo791badb2013-12-05 12:28:02 -05002795static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002796 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002797{
Tejun Heo182446d2013-08-08 20:11:24 -04002798 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002799 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002800
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002801 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002802 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002803 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002804 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002805 case _MEMSWAP:
2806 counter = &memcg->memsw;
2807 break;
2808 case _KMEM:
2809 counter = &memcg->kmem;
2810 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002811 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08002812 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002813 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002814 default:
2815 BUG();
2816 }
2817
2818 switch (MEMFILE_ATTR(cft->private)) {
2819 case RES_USAGE:
2820 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08002821 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002822 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08002823 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002824 return (u64)page_counter_read(counter) * PAGE_SIZE;
2825 case RES_LIMIT:
2826 return (u64)counter->limit * PAGE_SIZE;
2827 case RES_MAX_USAGE:
2828 return (u64)counter->watermark * PAGE_SIZE;
2829 case RES_FAILCNT:
2830 return counter->failcnt;
2831 case RES_SOFT_LIMIT:
2832 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002833 default:
2834 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002835 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002836}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002837
Johannes Weiner127424c2016-01-20 15:02:32 -08002838#ifndef CONFIG_SLOB
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002839static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002840{
Vladimir Davydovd6441632014-01-23 15:53:09 -08002841 int memcg_id;
2842
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002843 if (cgroup_memory_nokmem)
2844 return 0;
2845
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002846 BUG_ON(memcg->kmemcg_id >= 0);
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002847 BUG_ON(memcg->kmem_state);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002848
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002849 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002850 if (memcg_id < 0)
2851 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002852
Johannes Weineref129472016-01-14 15:21:34 -08002853 static_branch_inc(&memcg_kmem_enabled_key);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002854 /*
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002855 * A memory cgroup is considered kmem-online as soon as it gets
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002856 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002857 * guarantee no one starts accounting before all call sites are
2858 * patched.
2859 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002860 memcg->kmemcg_id = memcg_id;
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002861 memcg->kmem_state = KMEM_ONLINE;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002862
2863 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002864}
2865
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002866static void memcg_offline_kmem(struct mem_cgroup *memcg)
2867{
2868 struct cgroup_subsys_state *css;
2869 struct mem_cgroup *parent, *child;
2870 int kmemcg_id;
2871
2872 if (memcg->kmem_state != KMEM_ONLINE)
2873 return;
2874 /*
2875 * Clear the online state before clearing memcg_caches array
2876 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2877 * guarantees that no cache will be created for this cgroup
2878 * after we are done (see memcg_create_kmem_cache()).
2879 */
2880 memcg->kmem_state = KMEM_ALLOCATED;
2881
2882 memcg_deactivate_kmem_caches(memcg);
2883
2884 kmemcg_id = memcg->kmemcg_id;
2885 BUG_ON(kmemcg_id < 0);
2886
2887 parent = parent_mem_cgroup(memcg);
2888 if (!parent)
2889 parent = root_mem_cgroup;
2890
2891 /*
2892 * Change kmemcg_id of this cgroup and all its descendants to the
2893 * parent's id, and then move all entries from this cgroup's list_lrus
2894 * to ones of the parent. After we have finished, all list_lrus
2895 * corresponding to this cgroup are guaranteed to remain empty. The
2896 * ordering is imposed by list_lru_node->lock taken by
2897 * memcg_drain_all_list_lrus().
2898 */
Tejun Heo3a06bb72016-06-03 14:55:44 -07002899 rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002900 css_for_each_descendant_pre(css, &memcg->css) {
2901 child = mem_cgroup_from_css(css);
2902 BUG_ON(child->kmemcg_id != kmemcg_id);
2903 child->kmemcg_id = parent->kmemcg_id;
2904 if (!memcg->use_hierarchy)
2905 break;
2906 }
Tejun Heo3a06bb72016-06-03 14:55:44 -07002907 rcu_read_unlock();
2908
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002909 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2910
2911 memcg_free_cache_id(kmemcg_id);
2912}
2913
2914static void memcg_free_kmem(struct mem_cgroup *memcg)
2915{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002916 /* css_alloc() failed, offlining didn't happen */
2917 if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2918 memcg_offline_kmem(memcg);
2919
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002920 if (memcg->kmem_state == KMEM_ALLOCATED) {
2921 memcg_destroy_kmem_caches(memcg);
2922 static_branch_dec(&memcg_kmem_enabled_key);
2923 WARN_ON(page_counter_read(&memcg->kmem));
2924 }
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002925}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002926#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002927static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08002928{
2929 return 0;
2930}
2931static void memcg_offline_kmem(struct mem_cgroup *memcg)
2932{
2933}
2934static void memcg_free_kmem(struct mem_cgroup *memcg)
2935{
2936}
2937#endif /* !CONFIG_SLOB */
2938
Johannes Weiner127424c2016-01-20 15:02:32 -08002939static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2940 unsigned long limit)
2941{
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002942 int ret;
Johannes Weiner127424c2016-01-20 15:02:32 -08002943
2944 mutex_lock(&memcg_limit_mutex);
Johannes Weiner127424c2016-01-20 15:02:32 -08002945 ret = page_counter_limit(&memcg->kmem, limit);
Johannes Weiner127424c2016-01-20 15:02:32 -08002946 mutex_unlock(&memcg_limit_mutex);
2947 return ret;
2948}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002949
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002950static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2951{
2952 int ret;
2953
2954 mutex_lock(&memcg_limit_mutex);
2955
Johannes Weiner0db15292016-01-20 15:02:50 -08002956 ret = page_counter_limit(&memcg->tcpmem, limit);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002957 if (ret)
2958 goto out;
2959
Johannes Weiner0db15292016-01-20 15:02:50 -08002960 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002961 /*
2962 * The active flag needs to be written after the static_key
2963 * update. This is what guarantees that the socket activation
2964 * function is the last one to run. See sock_update_memcg() for
2965 * details, and note that we don't mark any socket as belonging
2966 * to this memcg until that flag is up.
2967 *
2968 * We need to do this, because static_keys will span multiple
2969 * sites, but we can't control their order. If we mark a socket
2970 * as accounted, but the accounting functions are not patched in
2971 * yet, we'll lose accounting.
2972 *
2973 * We never race with the readers in sock_update_memcg(),
2974 * because when this value change, the code to process it is not
2975 * patched in yet.
2976 */
2977 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08002978 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002979 }
2980out:
2981 mutex_unlock(&memcg_limit_mutex);
2982 return ret;
2983}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002984
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002985/*
2986 * The user of this function is...
2987 * RES_LIMIT.
2988 */
Tejun Heo451af502014-05-13 12:16:21 -04002989static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2990 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002991{
Tejun Heo451af502014-05-13 12:16:21 -04002992 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002993 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002994 int ret;
2995
Tejun Heo451af502014-05-13 12:16:21 -04002996 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08002997 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002998 if (ret)
2999 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07003000
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003001 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003002 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003003 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3004 ret = -EINVAL;
3005 break;
3006 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003007 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3008 case _MEM:
3009 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003010 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003011 case _MEMSWAP:
3012 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3013 break;
3014 case _KMEM:
3015 ret = memcg_update_kmem_limit(memcg, nr_pages);
3016 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003017 case _TCP:
3018 ret = memcg_update_tcp_limit(memcg, nr_pages);
3019 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003020 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003021 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003022 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003023 memcg->soft_limit = nr_pages;
3024 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003025 break;
3026 }
Tejun Heo451af502014-05-13 12:16:21 -04003027 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003028}
3029
Tejun Heo6770c642014-05-13 12:16:21 -04003030static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3031 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003032{
Tejun Heo6770c642014-05-13 12:16:21 -04003033 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003034 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003035
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003036 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3037 case _MEM:
3038 counter = &memcg->memory;
3039 break;
3040 case _MEMSWAP:
3041 counter = &memcg->memsw;
3042 break;
3043 case _KMEM:
3044 counter = &memcg->kmem;
3045 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003046 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003047 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003048 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003049 default:
3050 BUG();
3051 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003052
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003053 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003054 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003055 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003056 break;
3057 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003058 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003059 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003060 default:
3061 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003062 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003063
Tejun Heo6770c642014-05-13 12:16:21 -04003064 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003065}
3066
Tejun Heo182446d2013-08-08 20:11:24 -04003067static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003068 struct cftype *cft)
3069{
Tejun Heo182446d2013-08-08 20:11:24 -04003070 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003071}
3072
Daisuke Nishimura02491442010-03-10 15:22:17 -08003073#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003074static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003075 struct cftype *cft, u64 val)
3076{
Tejun Heo182446d2013-08-08 20:11:24 -04003077 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003078
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003079 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003080 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003081
Glauber Costaee5e8472013-02-22 16:34:50 -08003082 /*
3083 * No kind of locking is needed in here, because ->can_attach() will
3084 * check this value once in the beginning of the process, and then carry
3085 * on with stale data. This means that changes to this value will only
3086 * affect task migrations starting after the change.
3087 */
3088 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003089 return 0;
3090}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003091#else
Tejun Heo182446d2013-08-08 20:11:24 -04003092static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003093 struct cftype *cft, u64 val)
3094{
3095 return -ENOSYS;
3096}
3097#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003098
Ying Han406eb0c2011-05-26 16:25:37 -07003099#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003100static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003101{
Greg Thelen25485de2013-11-12 15:07:40 -08003102 struct numa_stat {
3103 const char *name;
3104 unsigned int lru_mask;
3105 };
3106
3107 static const struct numa_stat stats[] = {
3108 { "total", LRU_ALL },
3109 { "file", LRU_ALL_FILE },
3110 { "anon", LRU_ALL_ANON },
3111 { "unevictable", BIT(LRU_UNEVICTABLE) },
3112 };
3113 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003114 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003115 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003116 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003117
Greg Thelen25485de2013-11-12 15:07:40 -08003118 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3119 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3120 seq_printf(m, "%s=%lu", stat->name, nr);
3121 for_each_node_state(nid, N_MEMORY) {
3122 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3123 stat->lru_mask);
3124 seq_printf(m, " N%d=%lu", nid, nr);
3125 }
3126 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003127 }
Ying Han406eb0c2011-05-26 16:25:37 -07003128
Ying Han071aee12013-11-12 15:07:41 -08003129 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3130 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003131
Ying Han071aee12013-11-12 15:07:41 -08003132 nr = 0;
3133 for_each_mem_cgroup_tree(iter, memcg)
3134 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3135 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3136 for_each_node_state(nid, N_MEMORY) {
3137 nr = 0;
3138 for_each_mem_cgroup_tree(iter, memcg)
3139 nr += mem_cgroup_node_nr_lru_pages(
3140 iter, nid, stat->lru_mask);
3141 seq_printf(m, " N%d=%lu", nid, nr);
3142 }
3143 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003144 }
Ying Han406eb0c2011-05-26 16:25:37 -07003145
Ying Han406eb0c2011-05-26 16:25:37 -07003146 return 0;
3147}
3148#endif /* CONFIG_NUMA */
3149
Tejun Heo2da8ca82013-12-05 12:28:04 -05003150static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003151{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003152 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003153 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003154 struct mem_cgroup *mi;
3155 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003156
Greg Thelen0ca44b12015-02-11 15:25:58 -08003157 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3158 MEM_CGROUP_STAT_NSTATS);
3159 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3160 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003161 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3162
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003163 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Johannes Weiner7941d212016-01-14 15:21:23 -08003164 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003165 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07003166 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003167 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003168 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003169
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003170 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3171 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3172 mem_cgroup_read_events(memcg, i));
3173
3174 for (i = 0; i < NR_LRU_LISTS; i++)
3175 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3176 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3177
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003178 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003179 memory = memsw = PAGE_COUNTER_MAX;
3180 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3181 memory = min(memory, mi->memory.limit);
3182 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003183 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003184 seq_printf(m, "hierarchical_memory_limit %llu\n",
3185 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08003186 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003187 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3188 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003189
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003190 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003191 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003192
Johannes Weiner7941d212016-01-14 15:21:23 -08003193 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003194 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003195 for_each_mem_cgroup_tree(mi, memcg)
3196 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
Greg Thelen484ebb32015-10-01 15:37:05 -07003197 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003198 }
3199
3200 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3201 unsigned long long val = 0;
3202
3203 for_each_mem_cgroup_tree(mi, memcg)
3204 val += mem_cgroup_read_events(mi, i);
3205 seq_printf(m, "total_%s %llu\n",
3206 mem_cgroup_events_names[i], val);
3207 }
3208
3209 for (i = 0; i < NR_LRU_LISTS; i++) {
3210 unsigned long long val = 0;
3211
3212 for_each_mem_cgroup_tree(mi, memcg)
3213 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3214 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003215 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003216
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003217#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003218 {
3219 int nid, zid;
3220 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003221 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003222 unsigned long recent_rotated[2] = {0, 0};
3223 unsigned long recent_scanned[2] = {0, 0};
3224
3225 for_each_online_node(nid)
3226 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Jianyu Zhane2318752014-06-06 14:38:20 -07003227 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
Hugh Dickins89abfab2012-05-29 15:06:53 -07003228 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003229
Hugh Dickins89abfab2012-05-29 15:06:53 -07003230 recent_rotated[0] += rstat->recent_rotated[0];
3231 recent_rotated[1] += rstat->recent_rotated[1];
3232 recent_scanned[0] += rstat->recent_scanned[0];
3233 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003234 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003235 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3236 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3237 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3238 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003239 }
3240#endif
3241
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003242 return 0;
3243}
3244
Tejun Heo182446d2013-08-08 20:11:24 -04003245static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3246 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003247{
Tejun Heo182446d2013-08-08 20:11:24 -04003248 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003249
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003250 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003251}
3252
Tejun Heo182446d2013-08-08 20:11:24 -04003253static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3254 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003255{
Tejun Heo182446d2013-08-08 20:11:24 -04003256 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003257
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003258 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003259 return -EINVAL;
3260
Linus Torvalds14208b02014-06-09 15:03:33 -07003261 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003262 memcg->swappiness = val;
3263 else
3264 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003265
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003266 return 0;
3267}
3268
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003269static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3270{
3271 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003272 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003273 int i;
3274
3275 rcu_read_lock();
3276 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003277 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003278 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003279 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003280
3281 if (!t)
3282 goto unlock;
3283
Johannes Weinerce00a962014-09-05 08:43:57 -04003284 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003285
3286 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003287 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003288 * If it's not true, a threshold was crossed after last
3289 * call of __mem_cgroup_threshold().
3290 */
Phil Carmody5407a562010-05-26 14:42:42 -07003291 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003292
3293 /*
3294 * Iterate backward over array of thresholds starting from
3295 * current_threshold and check if a threshold is crossed.
3296 * If none of thresholds below usage is crossed, we read
3297 * only one element of the array here.
3298 */
3299 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3300 eventfd_signal(t->entries[i].eventfd, 1);
3301
3302 /* i = current_threshold + 1 */
3303 i++;
3304
3305 /*
3306 * Iterate forward over array of thresholds starting from
3307 * current_threshold+1 and check if a threshold is crossed.
3308 * If none of thresholds above usage is crossed, we read
3309 * only one element of the array here.
3310 */
3311 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3312 eventfd_signal(t->entries[i].eventfd, 1);
3313
3314 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003315 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003316unlock:
3317 rcu_read_unlock();
3318}
3319
3320static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3321{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003322 while (memcg) {
3323 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08003324 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003325 __mem_cgroup_threshold(memcg, true);
3326
3327 memcg = parent_mem_cgroup(memcg);
3328 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003329}
3330
3331static int compare_thresholds(const void *a, const void *b)
3332{
3333 const struct mem_cgroup_threshold *_a = a;
3334 const struct mem_cgroup_threshold *_b = b;
3335
Greg Thelen2bff24a2013-09-11 14:23:08 -07003336 if (_a->threshold > _b->threshold)
3337 return 1;
3338
3339 if (_a->threshold < _b->threshold)
3340 return -1;
3341
3342 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003343}
3344
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003345static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003346{
3347 struct mem_cgroup_eventfd_list *ev;
3348
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003349 spin_lock(&memcg_oom_lock);
3350
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003351 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003352 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003353
3354 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003355 return 0;
3356}
3357
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003358static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003359{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003360 struct mem_cgroup *iter;
3361
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003362 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003363 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003364}
3365
Tejun Heo59b6f872013-11-22 18:20:43 -05003366static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003367 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003368{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003369 struct mem_cgroup_thresholds *thresholds;
3370 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003371 unsigned long threshold;
3372 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003373 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003374
Johannes Weiner650c5e52015-02-11 15:26:03 -08003375 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003376 if (ret)
3377 return ret;
3378
3379 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003380
Johannes Weiner05b84302014-08-06 16:05:59 -07003381 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003382 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003383 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003384 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003385 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003386 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003387 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003388 BUG();
3389
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003390 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003391 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003392 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3393
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003394 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003395
3396 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003397 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003398 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003399 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003400 ret = -ENOMEM;
3401 goto unlock;
3402 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003403 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003404
3405 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003406 if (thresholds->primary) {
3407 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003408 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003409 }
3410
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003411 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003412 new->entries[size - 1].eventfd = eventfd;
3413 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003414
3415 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003416 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003417 compare_thresholds, NULL);
3418
3419 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003420 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003421 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003422 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003423 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003424 * new->current_threshold will not be used until
3425 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003426 * it here.
3427 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003428 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003429 } else
3430 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003431 }
3432
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003433 /* Free old spare buffer and save old primary buffer as spare */
3434 kfree(thresholds->spare);
3435 thresholds->spare = thresholds->primary;
3436
3437 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003438
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003439 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003440 synchronize_rcu();
3441
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003442unlock:
3443 mutex_unlock(&memcg->thresholds_lock);
3444
3445 return ret;
3446}
3447
Tejun Heo59b6f872013-11-22 18:20:43 -05003448static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003449 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003450{
Tejun Heo59b6f872013-11-22 18:20:43 -05003451 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003452}
3453
Tejun Heo59b6f872013-11-22 18:20:43 -05003454static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003455 struct eventfd_ctx *eventfd, const char *args)
3456{
Tejun Heo59b6f872013-11-22 18:20:43 -05003457 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003458}
3459
Tejun Heo59b6f872013-11-22 18:20:43 -05003460static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003461 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003462{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003463 struct mem_cgroup_thresholds *thresholds;
3464 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003465 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003466 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003467
3468 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003469
3470 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003471 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003472 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003473 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003474 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003475 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003476 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003477 BUG();
3478
Anton Vorontsov371528c2012-02-24 05:14:46 +04003479 if (!thresholds->primary)
3480 goto unlock;
3481
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003482 /* Check if a threshold crossed before removing */
3483 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3484
3485 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003486 size = 0;
3487 for (i = 0; i < thresholds->primary->size; i++) {
3488 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003489 size++;
3490 }
3491
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003492 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003493
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003494 /* Set thresholds array to NULL if we don't have thresholds */
3495 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003496 kfree(new);
3497 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003498 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003499 }
3500
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003501 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003502
3503 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003504 new->current_threshold = -1;
3505 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3506 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003507 continue;
3508
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003509 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003510 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003511 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003512 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003513 * until rcu_assign_pointer(), so it's safe to increment
3514 * it here.
3515 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003516 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003517 }
3518 j++;
3519 }
3520
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003521swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003522 /* Swap primary and spare array */
3523 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003524
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003525 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003526
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003527 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003528 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08003529
3530 /* If all events are unregistered, free the spare array */
3531 if (!new) {
3532 kfree(thresholds->spare);
3533 thresholds->spare = NULL;
3534 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04003535unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003536 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003537}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003538
Tejun Heo59b6f872013-11-22 18:20:43 -05003539static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003540 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003541{
Tejun Heo59b6f872013-11-22 18:20:43 -05003542 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003543}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003544
Tejun Heo59b6f872013-11-22 18:20:43 -05003545static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003546 struct eventfd_ctx *eventfd)
3547{
Tejun Heo59b6f872013-11-22 18:20:43 -05003548 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003549}
3550
Tejun Heo59b6f872013-11-22 18:20:43 -05003551static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003552 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003553{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003554 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003555
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003556 event = kmalloc(sizeof(*event), GFP_KERNEL);
3557 if (!event)
3558 return -ENOMEM;
3559
Michal Hocko1af8efe2011-07-26 16:08:24 -07003560 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003561
3562 event->eventfd = eventfd;
3563 list_add(&event->list, &memcg->oom_notify);
3564
3565 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07003566 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003567 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003568 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003569
3570 return 0;
3571}
3572
Tejun Heo59b6f872013-11-22 18:20:43 -05003573static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003574 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003575{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003576 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003577
Michal Hocko1af8efe2011-07-26 16:08:24 -07003578 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003579
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003580 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003581 if (ev->eventfd == eventfd) {
3582 list_del(&ev->list);
3583 kfree(ev);
3584 }
3585 }
3586
Michal Hocko1af8efe2011-07-26 16:08:24 -07003587 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003588}
3589
Tejun Heo2da8ca82013-12-05 12:28:04 -05003590static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003591{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003592 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003593
Tejun Heo791badb2013-12-05 12:28:02 -05003594 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07003595 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003596 return 0;
3597}
3598
Tejun Heo182446d2013-08-08 20:11:24 -04003599static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003600 struct cftype *cft, u64 val)
3601{
Tejun Heo182446d2013-08-08 20:11:24 -04003602 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003603
3604 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003605 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003606 return -EINVAL;
3607
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003608 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003609 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003610 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003611
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003612 return 0;
3613}
3614
Tejun Heo52ebea72015-05-22 17:13:37 -04003615#ifdef CONFIG_CGROUP_WRITEBACK
3616
3617struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3618{
3619 return &memcg->cgwb_list;
3620}
3621
Tejun Heo841710a2015-05-22 18:23:33 -04003622static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3623{
3624 return wb_domain_init(&memcg->cgwb_domain, gfp);
3625}
3626
3627static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3628{
3629 wb_domain_exit(&memcg->cgwb_domain);
3630}
3631
Tejun Heo2529bb32015-05-22 18:23:34 -04003632static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3633{
3634 wb_domain_size_changed(&memcg->cgwb_domain);
3635}
3636
Tejun Heo841710a2015-05-22 18:23:33 -04003637struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3638{
3639 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3640
3641 if (!memcg->css.parent)
3642 return NULL;
3643
3644 return &memcg->cgwb_domain;
3645}
3646
Tejun Heoc2aa7232015-05-22 18:23:35 -04003647/**
3648 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3649 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003650 * @pfilepages: out parameter for number of file pages
3651 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04003652 * @pdirty: out parameter for number of dirty pages
3653 * @pwriteback: out parameter for number of pages under writeback
3654 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003655 * Determine the numbers of file, headroom, dirty, and writeback pages in
3656 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3657 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003658 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003659 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3660 * headroom is calculated as the lowest headroom of itself and the
3661 * ancestors. Note that this doesn't consider the actual amount of
3662 * available memory in the system. The caller should further cap
3663 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003664 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003665void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3666 unsigned long *pheadroom, unsigned long *pdirty,
3667 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04003668{
3669 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3670 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003671
3672 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3673
3674 /* this should eventually include NR_UNSTABLE_NFS */
3675 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003676 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3677 (1 << LRU_ACTIVE_FILE));
3678 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003679
Tejun Heoc2aa7232015-05-22 18:23:35 -04003680 while ((parent = parent_mem_cgroup(memcg))) {
3681 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3682 unsigned long used = page_counter_read(&memcg->memory);
3683
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003684 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04003685 memcg = parent;
3686 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04003687}
3688
Tejun Heo841710a2015-05-22 18:23:33 -04003689#else /* CONFIG_CGROUP_WRITEBACK */
3690
3691static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3692{
3693 return 0;
3694}
3695
3696static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3697{
3698}
3699
Tejun Heo2529bb32015-05-22 18:23:34 -04003700static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3701{
3702}
3703
Tejun Heo52ebea72015-05-22 17:13:37 -04003704#endif /* CONFIG_CGROUP_WRITEBACK */
3705
Tejun Heo79bd9812013-11-22 18:20:42 -05003706/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003707 * DO NOT USE IN NEW FILES.
3708 *
3709 * "cgroup.event_control" implementation.
3710 *
3711 * This is way over-engineered. It tries to support fully configurable
3712 * events for each user. Such level of flexibility is completely
3713 * unnecessary especially in the light of the planned unified hierarchy.
3714 *
3715 * Please deprecate this and replace with something simpler if at all
3716 * possible.
3717 */
3718
3719/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003720 * Unregister event and free resources.
3721 *
3722 * Gets called from workqueue.
3723 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003724static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003725{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003726 struct mem_cgroup_event *event =
3727 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003728 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003729
3730 remove_wait_queue(event->wqh, &event->wait);
3731
Tejun Heo59b6f872013-11-22 18:20:43 -05003732 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003733
3734 /* Notify userspace the event is going away. */
3735 eventfd_signal(event->eventfd, 1);
3736
3737 eventfd_ctx_put(event->eventfd);
3738 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003739 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003740}
3741
3742/*
3743 * Gets called on POLLHUP on eventfd when user closes it.
3744 *
3745 * Called with wqh->lock held and interrupts disabled.
3746 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003747static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3748 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003749{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003750 struct mem_cgroup_event *event =
3751 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003752 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003753 unsigned long flags = (unsigned long)key;
3754
3755 if (flags & POLLHUP) {
3756 /*
3757 * If the event has been detached at cgroup removal, we
3758 * can simply return knowing the other side will cleanup
3759 * for us.
3760 *
3761 * We can't race against event freeing since the other
3762 * side will require wqh->lock via remove_wait_queue(),
3763 * which we hold.
3764 */
Tejun Heofba94802013-11-22 18:20:43 -05003765 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003766 if (!list_empty(&event->list)) {
3767 list_del_init(&event->list);
3768 /*
3769 * We are in atomic context, but cgroup_event_remove()
3770 * may sleep, so we have to call it in workqueue.
3771 */
3772 schedule_work(&event->remove);
3773 }
Tejun Heofba94802013-11-22 18:20:43 -05003774 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003775 }
3776
3777 return 0;
3778}
3779
Tejun Heo3bc942f2013-11-22 18:20:44 -05003780static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003781 wait_queue_head_t *wqh, poll_table *pt)
3782{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003783 struct mem_cgroup_event *event =
3784 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003785
3786 event->wqh = wqh;
3787 add_wait_queue(wqh, &event->wait);
3788}
3789
3790/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003791 * DO NOT USE IN NEW FILES.
3792 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003793 * Parse input and register new cgroup event handler.
3794 *
3795 * Input must be in format '<event_fd> <control_fd> <args>'.
3796 * Interpretation of args is defined by control file implementation.
3797 */
Tejun Heo451af502014-05-13 12:16:21 -04003798static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3799 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003800{
Tejun Heo451af502014-05-13 12:16:21 -04003801 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003802 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003803 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003804 struct cgroup_subsys_state *cfile_css;
3805 unsigned int efd, cfd;
3806 struct fd efile;
3807 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003808 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003809 char *endp;
3810 int ret;
3811
Tejun Heo451af502014-05-13 12:16:21 -04003812 buf = strstrip(buf);
3813
3814 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003815 if (*endp != ' ')
3816 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003817 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003818
Tejun Heo451af502014-05-13 12:16:21 -04003819 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003820 if ((*endp != ' ') && (*endp != '\0'))
3821 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003822 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003823
3824 event = kzalloc(sizeof(*event), GFP_KERNEL);
3825 if (!event)
3826 return -ENOMEM;
3827
Tejun Heo59b6f872013-11-22 18:20:43 -05003828 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003829 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003830 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3831 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3832 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003833
3834 efile = fdget(efd);
3835 if (!efile.file) {
3836 ret = -EBADF;
3837 goto out_kfree;
3838 }
3839
3840 event->eventfd = eventfd_ctx_fileget(efile.file);
3841 if (IS_ERR(event->eventfd)) {
3842 ret = PTR_ERR(event->eventfd);
3843 goto out_put_efile;
3844 }
3845
3846 cfile = fdget(cfd);
3847 if (!cfile.file) {
3848 ret = -EBADF;
3849 goto out_put_eventfd;
3850 }
3851
3852 /* the process need read permission on control file */
3853 /* AV: shouldn't we check that it's been opened for read instead? */
3854 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3855 if (ret < 0)
3856 goto out_put_cfile;
3857
Tejun Heo79bd9812013-11-22 18:20:42 -05003858 /*
Tejun Heofba94802013-11-22 18:20:43 -05003859 * Determine the event callbacks and set them in @event. This used
3860 * to be done via struct cftype but cgroup core no longer knows
3861 * about these events. The following is crude but the whole thing
3862 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003863 *
3864 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003865 */
Al Virob5830432014-10-31 01:22:04 -04003866 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003867
3868 if (!strcmp(name, "memory.usage_in_bytes")) {
3869 event->register_event = mem_cgroup_usage_register_event;
3870 event->unregister_event = mem_cgroup_usage_unregister_event;
3871 } else if (!strcmp(name, "memory.oom_control")) {
3872 event->register_event = mem_cgroup_oom_register_event;
3873 event->unregister_event = mem_cgroup_oom_unregister_event;
3874 } else if (!strcmp(name, "memory.pressure_level")) {
3875 event->register_event = vmpressure_register_event;
3876 event->unregister_event = vmpressure_unregister_event;
3877 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003878 event->register_event = memsw_cgroup_usage_register_event;
3879 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003880 } else {
3881 ret = -EINVAL;
3882 goto out_put_cfile;
3883 }
3884
3885 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003886 * Verify @cfile should belong to @css. Also, remaining events are
3887 * automatically removed on cgroup destruction but the removal is
3888 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003889 */
Al Virob5830432014-10-31 01:22:04 -04003890 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003891 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003892 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003893 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003894 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003895 if (cfile_css != css) {
3896 css_put(cfile_css);
3897 goto out_put_cfile;
3898 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003899
Tejun Heo451af502014-05-13 12:16:21 -04003900 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003901 if (ret)
3902 goto out_put_css;
3903
3904 efile.file->f_op->poll(efile.file, &event->pt);
3905
Tejun Heofba94802013-11-22 18:20:43 -05003906 spin_lock(&memcg->event_list_lock);
3907 list_add(&event->list, &memcg->event_list);
3908 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003909
3910 fdput(cfile);
3911 fdput(efile);
3912
Tejun Heo451af502014-05-13 12:16:21 -04003913 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05003914
3915out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05003916 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003917out_put_cfile:
3918 fdput(cfile);
3919out_put_eventfd:
3920 eventfd_ctx_put(event->eventfd);
3921out_put_efile:
3922 fdput(efile);
3923out_kfree:
3924 kfree(event);
3925
3926 return ret;
3927}
3928
Johannes Weiner241994e2015-02-11 15:26:06 -08003929static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003930 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003931 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003932 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003933 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003934 },
3935 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003936 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003937 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04003938 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003939 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003940 },
3941 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003942 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003943 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003944 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003945 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003946 },
3947 {
Balbir Singh296c81d2009-09-23 15:56:36 -07003948 .name = "soft_limit_in_bytes",
3949 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003950 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003951 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07003952 },
3953 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003954 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003955 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04003956 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003957 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003958 },
Balbir Singh8697d332008-02-07 00:13:59 -08003959 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003960 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003961 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003962 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003963 {
3964 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04003965 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003966 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08003967 {
3968 .name = "use_hierarchy",
3969 .write_u64 = mem_cgroup_hierarchy_write,
3970 .read_u64 = mem_cgroup_hierarchy_read,
3971 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003972 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05003973 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04003974 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04003975 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05003976 },
3977 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003978 .name = "swappiness",
3979 .read_u64 = mem_cgroup_swappiness_read,
3980 .write_u64 = mem_cgroup_swappiness_write,
3981 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003982 {
3983 .name = "move_charge_at_immigrate",
3984 .read_u64 = mem_cgroup_move_charge_read,
3985 .write_u64 = mem_cgroup_move_charge_write,
3986 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003987 {
3988 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003989 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003990 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003991 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3992 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003993 {
3994 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003995 },
Ying Han406eb0c2011-05-26 16:25:37 -07003996#ifdef CONFIG_NUMA
3997 {
3998 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003999 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07004000 },
4001#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004002 {
4003 .name = "kmem.limit_in_bytes",
4004 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004005 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004006 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004007 },
4008 {
4009 .name = "kmem.usage_in_bytes",
4010 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004011 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004012 },
4013 {
4014 .name = "kmem.failcnt",
4015 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004016 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004017 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004018 },
4019 {
4020 .name = "kmem.max_usage_in_bytes",
4021 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004022 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004023 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004024 },
Glauber Costa749c5412012-12-18 14:23:01 -08004025#ifdef CONFIG_SLABINFO
4026 {
4027 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004028 .seq_start = slab_start,
4029 .seq_next = slab_next,
4030 .seq_stop = slab_stop,
4031 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004032 },
4033#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004034 {
4035 .name = "kmem.tcp.limit_in_bytes",
4036 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4037 .write = mem_cgroup_write,
4038 .read_u64 = mem_cgroup_read_u64,
4039 },
4040 {
4041 .name = "kmem.tcp.usage_in_bytes",
4042 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4043 .read_u64 = mem_cgroup_read_u64,
4044 },
4045 {
4046 .name = "kmem.tcp.failcnt",
4047 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4048 .write = mem_cgroup_reset,
4049 .read_u64 = mem_cgroup_read_u64,
4050 },
4051 {
4052 .name = "kmem.tcp.max_usage_in_bytes",
4053 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4054 .write = mem_cgroup_reset,
4055 .read_u64 = mem_cgroup_read_u64,
4056 },
Tejun Heo6bc10342012-04-01 12:09:55 -07004057 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004058};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004059
Johannes Weiner73f576c2016-07-20 15:44:57 -07004060/*
4061 * Private memory cgroup IDR
4062 *
4063 * Swap-out records and page cache shadow entries need to store memcg
4064 * references in constrained space, so we maintain an ID space that is
4065 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4066 * memory-controlled cgroups to 64k.
4067 *
4068 * However, there usually are many references to the oflline CSS after
4069 * the cgroup has been destroyed, such as page cache or reclaimable
4070 * slab objects, that don't need to hang on to the ID. We want to keep
4071 * those dead CSS from occupying IDs, or we might quickly exhaust the
4072 * relatively small ID space and prevent the creation of new cgroups
4073 * even when there are much fewer than 64k cgroups - possibly none.
4074 *
4075 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4076 * be freed and recycled when it's no longer needed, which is usually
4077 * when the CSS is offlined.
4078 *
4079 * The only exception to that are records of swapped out tmpfs/shmem
4080 * pages that need to be attributed to live ancestors on swapin. But
4081 * those references are manageable from userspace.
4082 */
4083
4084static DEFINE_IDR(mem_cgroup_idr);
4085
4086static void mem_cgroup_id_get(struct mem_cgroup *memcg)
4087{
4088 atomic_inc(&memcg->id.ref);
4089}
4090
4091static void mem_cgroup_id_put(struct mem_cgroup *memcg)
4092{
4093 if (atomic_dec_and_test(&memcg->id.ref)) {
4094 idr_remove(&mem_cgroup_idr, memcg->id.id);
4095 memcg->id.id = 0;
4096
4097 /* Memcg ID pins CSS */
4098 css_put(&memcg->css);
4099 }
4100}
4101
4102/**
4103 * mem_cgroup_from_id - look up a memcg from a memcg id
4104 * @id: the memcg id to look up
4105 *
4106 * Caller must hold rcu_read_lock().
4107 */
4108struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
4109{
4110 WARN_ON_ONCE(!rcu_read_lock_held());
4111 return idr_find(&mem_cgroup_idr, id);
4112}
4113
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004114static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004115{
4116 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004117 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004118 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004119 /*
4120 * This routine is called against possible nodes.
4121 * But it's BUG to call kmalloc() against offline node.
4122 *
4123 * TODO: this routine can waste much memory for nodes which will
4124 * never be onlined. It's better to use memory hotplug callback
4125 * function.
4126 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004127 if (!node_state(node, N_NORMAL_MEMORY))
4128 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004129 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004130 if (!pn)
4131 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004132
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004133 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4134 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004135 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -07004136 mz->usage_in_excess = 0;
4137 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004138 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004139 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004140 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004141 return 0;
4142}
4143
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004144static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004145{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004146 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004147}
4148
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004149static void mem_cgroup_free(struct mem_cgroup *memcg)
4150{
4151 int node;
4152
4153 memcg_wb_domain_exit(memcg);
4154 for_each_node(node)
4155 free_mem_cgroup_per_zone_info(memcg, node);
4156 free_percpu(memcg->stat);
4157 kfree(memcg);
4158}
4159
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004160static struct mem_cgroup *mem_cgroup_alloc(void)
4161{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004162 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004163 size_t size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004164 int node;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004165
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004166 size = sizeof(struct mem_cgroup);
4167 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004168
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004169 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004170 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004171 return NULL;
4172
Johannes Weiner73f576c2016-07-20 15:44:57 -07004173 memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
4174 1, MEM_CGROUP_ID_MAX,
4175 GFP_KERNEL);
4176 if (memcg->id.id < 0)
4177 goto fail;
4178
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004179 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4180 if (!memcg->stat)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004181 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004182
Bob Liu3ed28fa2012-01-12 17:19:04 -08004183 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004184 if (alloc_mem_cgroup_per_zone_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004185 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004186
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004187 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4188 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004189
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004190 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08004191 memcg->last_scanned_node = MAX_NUMNODES;
4192 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004193 mutex_init(&memcg->thresholds_lock);
4194 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004195 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004196 INIT_LIST_HEAD(&memcg->event_list);
4197 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004198 memcg->socket_pressure = jiffies;
Johannes Weiner127424c2016-01-20 15:02:32 -08004199#ifndef CONFIG_SLOB
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004200 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004201#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004202#ifdef CONFIG_CGROUP_WRITEBACK
4203 INIT_LIST_HEAD(&memcg->cgwb_list);
4204#endif
Johannes Weiner73f576c2016-07-20 15:44:57 -07004205 idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004206 return memcg;
4207fail:
Johannes Weiner73f576c2016-07-20 15:44:57 -07004208 if (memcg->id.id > 0)
4209 idr_remove(&mem_cgroup_idr, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004210 mem_cgroup_free(memcg);
4211 return NULL;
Glauber Costad142e3e2013-02-22 16:34:52 -08004212}
4213
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004214static struct cgroup_subsys_state * __ref
4215mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004216{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004217 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4218 struct mem_cgroup *memcg;
4219 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08004220
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004221 memcg = mem_cgroup_alloc();
4222 if (!memcg)
4223 return ERR_PTR(error);
Li Zefan4219b2d2013-09-23 16:56:29 +08004224
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004225 memcg->high = PAGE_COUNTER_MAX;
4226 memcg->soft_limit = PAGE_COUNTER_MAX;
4227 if (parent) {
4228 memcg->swappiness = mem_cgroup_swappiness(parent);
4229 memcg->oom_kill_disable = parent->oom_kill_disable;
4230 }
4231 if (parent && parent->use_hierarchy) {
4232 memcg->use_hierarchy = true;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004233 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004234 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004235 page_counter_init(&memcg->memsw, &parent->memsw);
4236 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08004237 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004238 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004239 page_counter_init(&memcg->memory, NULL);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004240 page_counter_init(&memcg->swap, NULL);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004241 page_counter_init(&memcg->memsw, NULL);
4242 page_counter_init(&memcg->kmem, NULL);
Johannes Weiner0db15292016-01-20 15:02:50 -08004243 page_counter_init(&memcg->tcpmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004244 /*
4245 * Deeper hierachy with use_hierarchy == false doesn't make
4246 * much sense so let cgroup subsystem know about this
4247 * unfortunate state in our controller.
4248 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004249 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004250 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004251 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08004252
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004253 /* The following stuff does not apply to the root */
4254 if (!parent) {
4255 root_mem_cgroup = memcg;
4256 return &memcg->css;
4257 }
4258
Vladimir Davydovb313aee2016-03-17 14:18:27 -07004259 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004260 if (error)
4261 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08004262
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004263 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004264 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004265
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004266 return &memcg->css;
4267fail:
4268 mem_cgroup_free(memcg);
Tejun Heoea3a9642016-06-24 14:49:58 -07004269 return ERR_PTR(-ENOMEM);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004270}
4271
Johannes Weiner73f576c2016-07-20 15:44:57 -07004272static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004273{
Johannes Weiner73f576c2016-07-20 15:44:57 -07004274 /* Online state pins memcg ID, memcg ID pins CSS */
4275 mem_cgroup_id_get(mem_cgroup_from_css(css));
4276 css_get(css);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004277 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004278}
4279
Tejun Heoeb954192013-08-08 20:11:23 -04004280static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004281{
Tejun Heoeb954192013-08-08 20:11:23 -04004282 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004283 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004284
4285 /*
4286 * Unregister events and notify userspace.
4287 * Notify userspace about cgroup removing only after rmdir of cgroup
4288 * directory to avoid race between userspace and kernelspace.
4289 */
Tejun Heofba94802013-11-22 18:20:43 -05004290 spin_lock(&memcg->event_list_lock);
4291 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004292 list_del_init(&event->list);
4293 schedule_work(&event->remove);
4294 }
Tejun Heofba94802013-11-22 18:20:43 -05004295 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004296
Johannes Weiner567e9ab2016-01-20 15:02:24 -08004297 memcg_offline_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004298 wb_memcg_offline(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004299
4300 mem_cgroup_id_put(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004301}
4302
Vladimir Davydov6df38682015-12-29 14:54:10 -08004303static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4304{
4305 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4306
4307 invalidate_reclaim_iterators(memcg);
4308}
4309
Tejun Heoeb954192013-08-08 20:11:23 -04004310static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004311{
Tejun Heoeb954192013-08-08 20:11:23 -04004312 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004313
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004314 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004315 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004316
Johannes Weiner0db15292016-01-20 15:02:50 -08004317 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004318 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004319
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004320 vmpressure_cleanup(&memcg->vmpressure);
4321 cancel_work_sync(&memcg->high_work);
4322 mem_cgroup_remove_from_trees(memcg);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004323 memcg_free_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004324 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004325}
4326
Tejun Heo1ced9532014-07-08 18:02:57 -04004327/**
4328 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4329 * @css: the target css
4330 *
4331 * Reset the states of the mem_cgroup associated with @css. This is
4332 * invoked when the userland requests disabling on the default hierarchy
4333 * but the memcg is pinned through dependency. The memcg should stop
4334 * applying policies and should revert to the vanilla state as it may be
4335 * made visible again.
4336 *
4337 * The current implementation only resets the essential configurations.
4338 * This needs to be expanded to cover all the visible parts.
4339 */
4340static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4341{
4342 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4343
Vladimir Davydovd334c9b2016-03-17 14:19:38 -07004344 page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4345 page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4346 page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4347 page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4348 page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
Johannes Weiner241994e2015-02-11 15:26:06 -08004349 memcg->low = 0;
4350 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004351 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004352 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004353}
4354
Daisuke Nishimura02491442010-03-10 15:22:17 -08004355#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004356/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004357static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004358{
Johannes Weiner05b84302014-08-06 16:05:59 -07004359 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004360
Mel Gormand0164ad2015-11-06 16:28:21 -08004361 /* Try a single bulk charge without reclaim first, kswapd may wake */
4362 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004363 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004364 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004365 return ret;
4366 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004367
4368 /* Try charges one by one with reclaim */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004369 while (count--) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004370 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004371 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004372 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004373 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004374 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004375 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004376 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004377}
4378
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004379union mc_target {
4380 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004381 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004382};
4383
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004384enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004385 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004386 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004387 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004388};
4389
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004390static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4391 unsigned long addr, pte_t ptent)
4392{
4393 struct page *page = vm_normal_page(vma, addr, ptent);
4394
4395 if (!page || !page_mapped(page))
4396 return NULL;
4397 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004398 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004399 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004400 } else {
4401 if (!(mc.flags & MOVE_FILE))
4402 return NULL;
4403 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004404 if (!get_page_unless_zero(page))
4405 return NULL;
4406
4407 return page;
4408}
4409
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004410#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004411static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4412 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4413{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004414 struct page *page = NULL;
4415 swp_entry_t ent = pte_to_swp_entry(ptent);
4416
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004417 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004418 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004419 /*
4420 * Because lookup_swap_cache() updates some statistics counter,
4421 * we call find_get_page() with swapper_space directly.
4422 */
Shaohua Li33806f02013-02-22 16:34:37 -08004423 page = find_get_page(swap_address_space(ent), ent.val);
Johannes Weiner7941d212016-01-14 15:21:23 -08004424 if (do_memsw_account())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004425 entry->val = ent.val;
4426
4427 return page;
4428}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004429#else
4430static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4431 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4432{
4433 return NULL;
4434}
4435#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004436
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004437static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4438 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4439{
4440 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004441 struct address_space *mapping;
4442 pgoff_t pgoff;
4443
4444 if (!vma->vm_file) /* anonymous vma */
4445 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004446 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004447 return NULL;
4448
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004449 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004450 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004451
4452 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004453#ifdef CONFIG_SWAP
4454 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004455 if (shmem_mapping(mapping)) {
4456 page = find_get_entry(mapping, pgoff);
4457 if (radix_tree_exceptional_entry(page)) {
4458 swp_entry_t swp = radix_to_swp_entry(page);
Johannes Weiner7941d212016-01-14 15:21:23 -08004459 if (do_memsw_account())
Johannes Weiner139b6a62014-05-06 12:50:05 -07004460 *entry = swp;
4461 page = find_get_page(swap_address_space(swp), swp.val);
4462 }
4463 } else
4464 page = find_get_page(mapping, pgoff);
4465#else
4466 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004467#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004468 return page;
4469}
4470
Chen Gangb1b0dea2015-04-14 15:47:35 -07004471/**
4472 * mem_cgroup_move_account - move account of the page
4473 * @page: the page
4474 * @nr_pages: number of regular pages (>1 for huge pages)
4475 * @from: mem_cgroup which the page is moved from.
4476 * @to: mem_cgroup which the page is moved to. @from != @to.
4477 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08004478 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07004479 *
4480 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4481 * from old cgroup.
4482 */
4483static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004484 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07004485 struct mem_cgroup *from,
4486 struct mem_cgroup *to)
4487{
4488 unsigned long flags;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004489 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004490 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004491 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004492
4493 VM_BUG_ON(from == to);
4494 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004495 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07004496
4497 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07004498 * Prevent mem_cgroup_migrate() from looking at
Hugh Dickins45637ba2015-11-05 18:49:40 -08004499 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07004500 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004501 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004502 if (!trylock_page(page))
4503 goto out;
4504
4505 ret = -EINVAL;
4506 if (page->mem_cgroup != from)
4507 goto out_unlock;
4508
Greg Thelenc4843a72015-05-22 17:13:16 -04004509 anon = PageAnon(page);
4510
Chen Gangb1b0dea2015-04-14 15:47:35 -07004511 spin_lock_irqsave(&from->move_lock, flags);
4512
Greg Thelenc4843a72015-05-22 17:13:16 -04004513 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004514 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4515 nr_pages);
4516 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4517 nr_pages);
4518 }
4519
Greg Thelenc4843a72015-05-22 17:13:16 -04004520 /*
4521 * move_lock grabbed above and caller set from->moving_account, so
4522 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4523 * So mapping should be stable for dirty pages.
4524 */
4525 if (!anon && PageDirty(page)) {
4526 struct address_space *mapping = page_mapping(page);
4527
4528 if (mapping_cap_account_dirty(mapping)) {
4529 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4530 nr_pages);
4531 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4532 nr_pages);
4533 }
4534 }
4535
Chen Gangb1b0dea2015-04-14 15:47:35 -07004536 if (PageWriteback(page)) {
4537 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4538 nr_pages);
4539 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4540 nr_pages);
4541 }
4542
4543 /*
4544 * It is safe to change page->mem_cgroup here because the page
4545 * is referenced, charged, and isolated - we can't race with
4546 * uncharging, charging, migration, or LRU putback.
4547 */
4548
4549 /* caller should have done css_get */
4550 page->mem_cgroup = to;
4551 spin_unlock_irqrestore(&from->move_lock, flags);
4552
4553 ret = 0;
4554
4555 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004556 mem_cgroup_charge_statistics(to, page, compound, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004557 memcg_check_events(to, page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004558 mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004559 memcg_check_events(from, page);
4560 local_irq_enable();
4561out_unlock:
4562 unlock_page(page);
4563out:
4564 return ret;
4565}
4566
Li RongQing7cf78062016-05-27 14:27:46 -07004567/**
4568 * get_mctgt_type - get target type of moving charge
4569 * @vma: the vma the pte to be checked belongs
4570 * @addr: the address corresponding to the pte to be checked
4571 * @ptent: the pte to be checked
4572 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4573 *
4574 * Returns
4575 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4576 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4577 * move charge. if @target is not NULL, the page is stored in target->page
4578 * with extra refcnt got(Callers should handle it).
4579 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4580 * target for charge migration. if @target is not NULL, the entry is stored
4581 * in target->ent.
4582 *
4583 * Called with pte lock held.
4584 */
4585
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004586static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004587 unsigned long addr, pte_t ptent, union mc_target *target)
4588{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004589 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004590 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004591 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004592
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004593 if (pte_present(ptent))
4594 page = mc_handle_present_pte(vma, addr, ptent);
4595 else if (is_swap_pte(ptent))
4596 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004597 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004598 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004599
4600 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004601 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004602 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004603 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004604 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004605 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004606 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004607 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004608 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004609 ret = MC_TARGET_PAGE;
4610 if (target)
4611 target->page = page;
4612 }
4613 if (!ret || !target)
4614 put_page(page);
4615 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004616 /* There is a swap entry and a page doesn't exist or isn't charged */
4617 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004618 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004619 ret = MC_TARGET_SWAP;
4620 if (target)
4621 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004622 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004623 return ret;
4624}
4625
Naoya Horiguchi12724852012-03-21 16:34:28 -07004626#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4627/*
4628 * We don't consider swapping or file mapped pages because THP does not
4629 * support them for now.
4630 * Caller should make sure that pmd_trans_huge(pmd) is true.
4631 */
4632static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4633 unsigned long addr, pmd_t pmd, union mc_target *target)
4634{
4635 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004636 enum mc_target_type ret = MC_TARGET_NONE;
4637
4638 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004639 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004640 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004641 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004642 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004643 ret = MC_TARGET_PAGE;
4644 if (target) {
4645 get_page(page);
4646 target->page = page;
4647 }
4648 }
4649 return ret;
4650}
4651#else
4652static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4653 unsigned long addr, pmd_t pmd, union mc_target *target)
4654{
4655 return MC_TARGET_NONE;
4656}
4657#endif
4658
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004659static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4660 unsigned long addr, unsigned long end,
4661 struct mm_walk *walk)
4662{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004663 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004664 pte_t *pte;
4665 spinlock_t *ptl;
4666
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004667 ptl = pmd_trans_huge_lock(pmd, vma);
4668 if (ptl) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004669 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4670 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004671 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004672 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004673 }
Dave Hansen03319322011-03-22 16:32:56 -07004674
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004675 if (pmd_trans_unstable(pmd))
4676 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004677 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4678 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004679 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004680 mc.precharge++; /* increment precharge temporarily */
4681 pte_unmap_unlock(pte - 1, ptl);
4682 cond_resched();
4683
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004684 return 0;
4685}
4686
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004687static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4688{
4689 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004690
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004691 struct mm_walk mem_cgroup_count_precharge_walk = {
4692 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4693 .mm = mm,
4694 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004695 down_read(&mm->mmap_sem);
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004696 walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004697 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004698
4699 precharge = mc.precharge;
4700 mc.precharge = 0;
4701
4702 return precharge;
4703}
4704
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004705static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4706{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004707 unsigned long precharge = mem_cgroup_count_precharge(mm);
4708
4709 VM_BUG_ON(mc.moving_task);
4710 mc.moving_task = current;
4711 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004712}
4713
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004714/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4715static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004716{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004717 struct mem_cgroup *from = mc.from;
4718 struct mem_cgroup *to = mc.to;
4719
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004720 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004721 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004722 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004723 mc.precharge = 0;
4724 }
4725 /*
4726 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4727 * we must uncharge here.
4728 */
4729 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004730 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004731 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004732 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004733 /* we must fixup refcnts and charges */
4734 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004735 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004736 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004737 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004738
Johannes Weiner05b84302014-08-06 16:05:59 -07004739 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004740 * we charged both to->memory and to->memsw, so we
4741 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004742 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004743 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004744 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004745
Johannes Weinere8ea14c2014-12-10 15:42:42 -08004746 css_put_many(&mc.from->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004747
Li Zefan40503772013-07-08 16:00:34 -07004748 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004749 mc.moved_swap = 0;
4750 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004751 memcg_oom_recover(from);
4752 memcg_oom_recover(to);
4753 wake_up_all(&mc.waitq);
4754}
4755
4756static void mem_cgroup_clear_mc(void)
4757{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004758 struct mm_struct *mm = mc.mm;
4759
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004760 /*
4761 * we must clear moving_task before waking up waiters at the end of
4762 * task migration.
4763 */
4764 mc.moving_task = NULL;
4765 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004766 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004767 mc.from = NULL;
4768 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04004769 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004770 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004771
4772 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004773}
4774
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004775static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004776{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004777 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07004778 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07004779 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04004780 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004781 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004782 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004783 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004784
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004785 /* charge immigration isn't supported on the default hierarchy */
4786 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07004787 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004788
Tejun Heo4530edd2015-09-11 15:00:19 -04004789 /*
4790 * Multi-process migrations only happen on the default hierarchy
4791 * where charge immigration is not used. Perform charge
4792 * immigration if @tset contains a leader and whine if there are
4793 * multiple.
4794 */
4795 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004796 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04004797 WARN_ON_ONCE(p);
4798 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004799 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04004800 }
4801 if (!p)
4802 return 0;
4803
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004804 /*
4805 * We are now commited to this value whatever it is. Changes in this
4806 * tunable will only affect upcoming migrations, not the current one.
4807 * So we need to save it, and keep it going.
4808 */
4809 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4810 if (!move_flags)
4811 return 0;
4812
Tejun Heo9f2115f2015-09-08 15:01:10 -07004813 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004814
Tejun Heo9f2115f2015-09-08 15:01:10 -07004815 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004816
Tejun Heo9f2115f2015-09-08 15:01:10 -07004817 mm = get_task_mm(p);
4818 if (!mm)
4819 return 0;
4820 /* We move charges only when we move a owner of the mm */
4821 if (mm->owner == p) {
4822 VM_BUG_ON(mc.from);
4823 VM_BUG_ON(mc.to);
4824 VM_BUG_ON(mc.precharge);
4825 VM_BUG_ON(mc.moved_charge);
4826 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004827
Tejun Heo9f2115f2015-09-08 15:01:10 -07004828 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004829 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004830 mc.from = from;
4831 mc.to = memcg;
4832 mc.flags = move_flags;
4833 spin_unlock(&mc.lock);
4834 /* We set mc.moving_task later */
4835
4836 ret = mem_cgroup_precharge_mc(mm);
4837 if (ret)
4838 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004839 } else {
4840 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004841 }
4842 return ret;
4843}
4844
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004845static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004846{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004847 if (mc.to)
4848 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004849}
4850
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004851static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4852 unsigned long addr, unsigned long end,
4853 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004854{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004855 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004856 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004857 pte_t *pte;
4858 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004859 enum mc_target_type target_type;
4860 union mc_target target;
4861 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004862
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004863 ptl = pmd_trans_huge_lock(pmd, vma);
4864 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004865 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004866 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004867 return 0;
4868 }
4869 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4870 if (target_type == MC_TARGET_PAGE) {
4871 page = target.page;
4872 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004873 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08004874 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004875 mc.precharge -= HPAGE_PMD_NR;
4876 mc.moved_charge += HPAGE_PMD_NR;
4877 }
4878 putback_lru_page(page);
4879 }
4880 put_page(page);
4881 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004882 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004883 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004884 }
4885
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004886 if (pmd_trans_unstable(pmd))
4887 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004888retry:
4889 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4890 for (; addr != end; addr += PAGE_SIZE) {
4891 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004892 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004893
4894 if (!mc.precharge)
4895 break;
4896
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004897 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004898 case MC_TARGET_PAGE:
4899 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08004900 /*
4901 * We can have a part of the split pmd here. Moving it
4902 * can be done but it would be too convoluted so simply
4903 * ignore such a partial THP and keep it in original
4904 * memcg. There should be somebody mapping the head.
4905 */
4906 if (PageTransCompound(page))
4907 goto put;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004908 if (isolate_lru_page(page))
4909 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004910 if (!mem_cgroup_move_account(page, false,
4911 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004912 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004913 /* we uncharge from mc.from later. */
4914 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004915 }
4916 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004917put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004918 put_page(page);
4919 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004920 case MC_TARGET_SWAP:
4921 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004922 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004923 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004924 /* we fixup refcnts and charges later. */
4925 mc.moved_swap++;
4926 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004927 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004928 default:
4929 break;
4930 }
4931 }
4932 pte_unmap_unlock(pte - 1, ptl);
4933 cond_resched();
4934
4935 if (addr != end) {
4936 /*
4937 * We have consumed all precharges we got in can_attach().
4938 * We try charge one by one, but don't do any additional
4939 * charges to mc.to if we have failed in charge once in attach()
4940 * phase.
4941 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004942 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004943 if (!ret)
4944 goto retry;
4945 }
4946
4947 return ret;
4948}
4949
Tejun Heo264a0ae2016-04-21 19:09:02 -04004950static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004951{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004952 struct mm_walk mem_cgroup_move_charge_walk = {
4953 .pmd_entry = mem_cgroup_move_charge_pte_range,
Tejun Heo264a0ae2016-04-21 19:09:02 -04004954 .mm = mc.mm,
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004955 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004956
4957 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004958 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07004959 * Signal lock_page_memcg() to take the memcg's move_lock
4960 * while we're moving its pages to another memcg. Then wait
4961 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08004962 */
4963 atomic_inc(&mc.from->moving_account);
4964 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004965retry:
Tejun Heo264a0ae2016-04-21 19:09:02 -04004966 if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004967 /*
4968 * Someone who are holding the mmap_sem might be waiting in
4969 * waitq. So we cancel all extra charges, wake up all waiters,
4970 * and retry. Because we cancel precharges, we might not be able
4971 * to move enough charges, but moving charge is a best-effort
4972 * feature anyway, so it wouldn't be a big problem.
4973 */
4974 __mem_cgroup_clear_mc();
4975 cond_resched();
4976 goto retry;
4977 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004978 /*
4979 * When we have consumed all precharges and failed in doing
4980 * additional charge, the page walk just aborts.
4981 */
4982 walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004983 up_read(&mc.mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08004984 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004985}
4986
Tejun Heo264a0ae2016-04-21 19:09:02 -04004987static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08004988{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004989 if (mc.to) {
4990 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004991 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004992 }
Balbir Singh67e465a2008-02-07 00:13:54 -08004993}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004994#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004995static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004996{
4997 return 0;
4998}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004999static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005000{
5001}
Tejun Heo264a0ae2016-04-21 19:09:02 -04005002static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005003{
5004}
5005#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005006
Tejun Heof00baae2013-04-15 13:41:15 -07005007/*
5008 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005009 * to verify whether we're attached to the default hierarchy on each mount
5010 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005011 */
Tejun Heoeb954192013-08-08 20:11:23 -04005012static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005013{
5014 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005015 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005016 * guarantees that @root doesn't have any children, so turning it
5017 * on for the root memcg is enough.
5018 */
Tejun Heo9e10a132015-09-18 11:56:28 -04005019 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee592015-03-12 16:26:19 -07005020 root_mem_cgroup->use_hierarchy = true;
5021 else
5022 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005023}
5024
Johannes Weiner241994e2015-02-11 15:26:06 -08005025static u64 memory_current_read(struct cgroup_subsys_state *css,
5026 struct cftype *cft)
5027{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005028 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5029
5030 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994e2015-02-11 15:26:06 -08005031}
5032
5033static int memory_low_show(struct seq_file *m, void *v)
5034{
5035 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005036 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994e2015-02-11 15:26:06 -08005037
5038 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005039 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005040 else
5041 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5042
5043 return 0;
5044}
5045
5046static ssize_t memory_low_write(struct kernfs_open_file *of,
5047 char *buf, size_t nbytes, loff_t off)
5048{
5049 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5050 unsigned long low;
5051 int err;
5052
5053 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005054 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994e2015-02-11 15:26:06 -08005055 if (err)
5056 return err;
5057
5058 memcg->low = low;
5059
5060 return nbytes;
5061}
5062
5063static int memory_high_show(struct seq_file *m, void *v)
5064{
5065 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005066 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994e2015-02-11 15:26:06 -08005067
5068 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005069 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005070 else
5071 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5072
5073 return 0;
5074}
5075
5076static ssize_t memory_high_write(struct kernfs_open_file *of,
5077 char *buf, size_t nbytes, loff_t off)
5078{
5079 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner588083b2016-03-17 14:20:25 -07005080 unsigned long nr_pages;
Johannes Weiner241994e2015-02-11 15:26:06 -08005081 unsigned long high;
5082 int err;
5083
5084 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005085 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994e2015-02-11 15:26:06 -08005086 if (err)
5087 return err;
5088
5089 memcg->high = high;
5090
Johannes Weiner588083b2016-03-17 14:20:25 -07005091 nr_pages = page_counter_read(&memcg->memory);
5092 if (nr_pages > high)
5093 try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5094 GFP_KERNEL, true);
5095
Tejun Heo2529bb32015-05-22 18:23:34 -04005096 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005097 return nbytes;
5098}
5099
5100static int memory_max_show(struct seq_file *m, void *v)
5101{
5102 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005103 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994e2015-02-11 15:26:06 -08005104
5105 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005106 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005107 else
5108 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5109
5110 return 0;
5111}
5112
5113static ssize_t memory_max_write(struct kernfs_open_file *of,
5114 char *buf, size_t nbytes, loff_t off)
5115{
5116 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005117 unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5118 bool drained = false;
Johannes Weiner241994e2015-02-11 15:26:06 -08005119 unsigned long max;
5120 int err;
5121
5122 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005123 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994e2015-02-11 15:26:06 -08005124 if (err)
5125 return err;
5126
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005127 xchg(&memcg->memory.limit, max);
5128
5129 for (;;) {
5130 unsigned long nr_pages = page_counter_read(&memcg->memory);
5131
5132 if (nr_pages <= max)
5133 break;
5134
5135 if (signal_pending(current)) {
5136 err = -EINTR;
5137 break;
5138 }
5139
5140 if (!drained) {
5141 drain_all_stock(memcg);
5142 drained = true;
5143 continue;
5144 }
5145
5146 if (nr_reclaims) {
5147 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5148 GFP_KERNEL, true))
5149 nr_reclaims--;
5150 continue;
5151 }
5152
5153 mem_cgroup_events(memcg, MEMCG_OOM, 1);
5154 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5155 break;
5156 }
Johannes Weiner241994e2015-02-11 15:26:06 -08005157
Tejun Heo2529bb32015-05-22 18:23:34 -04005158 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005159 return nbytes;
5160}
5161
5162static int memory_events_show(struct seq_file *m, void *v)
5163{
5164 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5165
5166 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5167 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5168 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5169 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5170
5171 return 0;
5172}
5173
Johannes Weiner587d9f72016-01-20 15:03:19 -08005174static int memory_stat_show(struct seq_file *m, void *v)
5175{
5176 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005177 unsigned long stat[MEMCG_NR_STAT];
5178 unsigned long events[MEMCG_NR_EVENTS];
Johannes Weiner587d9f72016-01-20 15:03:19 -08005179 int i;
5180
5181 /*
5182 * Provide statistics on the state of the memory subsystem as
5183 * well as cumulative event counters that show past behavior.
5184 *
5185 * This list is ordered following a combination of these gradients:
5186 * 1) generic big picture -> specifics and details
5187 * 2) reflecting userspace activity -> reflecting kernel heuristics
5188 *
5189 * Current memory state:
5190 */
5191
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005192 tree_stat(memcg, stat);
5193 tree_events(memcg, events);
5194
Johannes Weiner587d9f72016-01-20 15:03:19 -08005195 seq_printf(m, "anon %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005196 (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005197 seq_printf(m, "file %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005198 (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE);
Vladimir Davydov12580e42016-03-17 14:17:38 -07005199 seq_printf(m, "kernel_stack %llu\n",
5200 (u64)stat[MEMCG_KERNEL_STACK] * PAGE_SIZE);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005201 seq_printf(m, "slab %llu\n",
5202 (u64)(stat[MEMCG_SLAB_RECLAIMABLE] +
5203 stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
Johannes Weinerb2807f02016-01-20 15:03:22 -08005204 seq_printf(m, "sock %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005205 (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005206
5207 seq_printf(m, "file_mapped %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005208 (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005209 seq_printf(m, "file_dirty %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005210 (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005211 seq_printf(m, "file_writeback %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005212 (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005213
5214 for (i = 0; i < NR_LRU_LISTS; i++) {
5215 struct mem_cgroup *mi;
5216 unsigned long val = 0;
5217
5218 for_each_mem_cgroup_tree(mi, memcg)
5219 val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5220 seq_printf(m, "%s %llu\n",
5221 mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5222 }
5223
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005224 seq_printf(m, "slab_reclaimable %llu\n",
5225 (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE);
5226 seq_printf(m, "slab_unreclaimable %llu\n",
5227 (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5228
Johannes Weiner587d9f72016-01-20 15:03:19 -08005229 /* Accumulated memory events */
5230
5231 seq_printf(m, "pgfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005232 events[MEM_CGROUP_EVENTS_PGFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005233 seq_printf(m, "pgmajfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005234 events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005235
5236 return 0;
5237}
5238
Johannes Weiner241994e2015-02-11 15:26:06 -08005239static struct cftype memory_files[] = {
5240 {
5241 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005242 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994e2015-02-11 15:26:06 -08005243 .read_u64 = memory_current_read,
5244 },
5245 {
5246 .name = "low",
5247 .flags = CFTYPE_NOT_ON_ROOT,
5248 .seq_show = memory_low_show,
5249 .write = memory_low_write,
5250 },
5251 {
5252 .name = "high",
5253 .flags = CFTYPE_NOT_ON_ROOT,
5254 .seq_show = memory_high_show,
5255 .write = memory_high_write,
5256 },
5257 {
5258 .name = "max",
5259 .flags = CFTYPE_NOT_ON_ROOT,
5260 .seq_show = memory_max_show,
5261 .write = memory_max_write,
5262 },
5263 {
5264 .name = "events",
5265 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04005266 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994e2015-02-11 15:26:06 -08005267 .seq_show = memory_events_show,
5268 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08005269 {
5270 .name = "stat",
5271 .flags = CFTYPE_NOT_ON_ROOT,
5272 .seq_show = memory_stat_show,
5273 },
Johannes Weiner241994e2015-02-11 15:26:06 -08005274 { } /* terminate */
5275};
5276
Tejun Heo073219e2014-02-08 10:36:58 -05005277struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005278 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005279 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005280 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08005281 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08005282 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005283 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005284 .can_attach = mem_cgroup_can_attach,
5285 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04005286 .post_attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005287 .bind = mem_cgroup_bind,
Johannes Weiner241994e2015-02-11 15:26:06 -08005288 .dfl_cftypes = memory_files,
5289 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005290 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005291};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005292
Johannes Weiner241994e2015-02-11 15:26:06 -08005293/**
Johannes Weiner241994e2015-02-11 15:26:06 -08005294 * mem_cgroup_low - check if memory consumption is below the normal range
5295 * @root: the highest ancestor to consider
5296 * @memcg: the memory cgroup to check
5297 *
5298 * Returns %true if memory consumption of @memcg, and that of all
5299 * configurable ancestors up to @root, is below the normal range.
5300 */
5301bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5302{
5303 if (mem_cgroup_disabled())
5304 return false;
5305
5306 /*
5307 * The toplevel group doesn't have a configurable range, so
5308 * it's never low when looked at directly, and it is not
5309 * considered an ancestor when assessing the hierarchy.
5310 */
5311
5312 if (memcg == root_mem_cgroup)
5313 return false;
5314
Michal Hocko4e54ded2015-02-27 15:51:46 -08005315 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005316 return false;
5317
5318 while (memcg != root) {
5319 memcg = parent_mem_cgroup(memcg);
5320
5321 if (memcg == root_mem_cgroup)
5322 break;
5323
Michal Hocko4e54ded2015-02-27 15:51:46 -08005324 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005325 return false;
5326 }
5327 return true;
5328}
5329
Johannes Weiner00501b52014-08-08 14:19:20 -07005330/**
5331 * mem_cgroup_try_charge - try charging a page
5332 * @page: page to charge
5333 * @mm: mm context of the victim
5334 * @gfp_mask: reclaim mode
5335 * @memcgp: charged memcg return
5336 *
5337 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5338 * pages according to @gfp_mask if necessary.
5339 *
5340 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5341 * Otherwise, an error code is returned.
5342 *
5343 * After page->mapping has been set up, the caller must finalize the
5344 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5345 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5346 */
5347int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005348 gfp_t gfp_mask, struct mem_cgroup **memcgp,
5349 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005350{
5351 struct mem_cgroup *memcg = NULL;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005352 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005353 int ret = 0;
5354
5355 if (mem_cgroup_disabled())
5356 goto out;
5357
5358 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005359 /*
5360 * Every swap fault against a single page tries to charge the
5361 * page, bail as early as possible. shmem_unuse() encounters
5362 * already charged pages, too. The USED bit is protected by
5363 * the page lock, which serializes swap cache removal, which
5364 * in turn serializes uncharging.
5365 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005366 VM_BUG_ON_PAGE(!PageLocked(page), page);
Johannes Weiner1306a852014-12-10 15:44:52 -08005367 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005368 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005369
Vladimir Davydov37e84352016-01-20 15:02:56 -08005370 if (do_swap_account) {
Vladimir Davydove993d902015-09-09 15:35:35 -07005371 swp_entry_t ent = { .val = page_private(page), };
5372 unsigned short id = lookup_swap_cgroup_id(ent);
5373
5374 rcu_read_lock();
5375 memcg = mem_cgroup_from_id(id);
5376 if (memcg && !css_tryget_online(&memcg->css))
5377 memcg = NULL;
5378 rcu_read_unlock();
5379 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005380 }
5381
Johannes Weiner00501b52014-08-08 14:19:20 -07005382 if (!memcg)
5383 memcg = get_mem_cgroup_from_mm(mm);
5384
5385 ret = try_charge(memcg, gfp_mask, nr_pages);
5386
5387 css_put(&memcg->css);
Johannes Weiner00501b52014-08-08 14:19:20 -07005388out:
5389 *memcgp = memcg;
5390 return ret;
5391}
5392
5393/**
5394 * mem_cgroup_commit_charge - commit a page charge
5395 * @page: page to charge
5396 * @memcg: memcg to charge the page to
5397 * @lrucare: page might be on LRU already
5398 *
5399 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5400 * after page->mapping has been set up. This must happen atomically
5401 * as part of the page instantiation, i.e. under the page table lock
5402 * for anonymous pages, under the page lock for page and swap cache.
5403 *
5404 * In addition, the page must not be on the LRU during the commit, to
5405 * prevent racing with task migration. If it might be, use @lrucare.
5406 *
5407 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5408 */
5409void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005410 bool lrucare, bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005411{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005412 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005413
5414 VM_BUG_ON_PAGE(!page->mapping, page);
5415 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5416
5417 if (mem_cgroup_disabled())
5418 return;
5419 /*
5420 * Swap faults will attempt to charge the same page multiple
5421 * times. But reuse_swap_page() might have removed the page
5422 * from swapcache already, so we can't check PageSwapCache().
5423 */
5424 if (!memcg)
5425 return;
5426
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005427 commit_charge(page, memcg, lrucare);
5428
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005429 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005430 mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005431 memcg_check_events(memcg, page);
5432 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005433
Johannes Weiner7941d212016-01-14 15:21:23 -08005434 if (do_memsw_account() && PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005435 swp_entry_t entry = { .val = page_private(page) };
5436 /*
5437 * The swap entry might not get freed for a long time,
5438 * let's not wait for it. The page already received a
5439 * memory+swap charge, drop the swap entry duplicate.
5440 */
5441 mem_cgroup_uncharge_swap(entry);
5442 }
5443}
5444
5445/**
5446 * mem_cgroup_cancel_charge - cancel a page charge
5447 * @page: page to charge
5448 * @memcg: memcg to charge the page to
5449 *
5450 * Cancel a charge transaction started by mem_cgroup_try_charge().
5451 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005452void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5453 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005454{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005455 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005456
5457 if (mem_cgroup_disabled())
5458 return;
5459 /*
5460 * Swap faults will attempt to charge the same page multiple
5461 * times. But reuse_swap_page() might have removed the page
5462 * from swapcache already, so we can't check PageSwapCache().
5463 */
5464 if (!memcg)
5465 return;
5466
Johannes Weiner00501b52014-08-08 14:19:20 -07005467 cancel_charge(memcg, nr_pages);
5468}
5469
Johannes Weiner747db952014-08-08 14:19:24 -07005470static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005471 unsigned long nr_anon, unsigned long nr_file,
5472 unsigned long nr_huge, struct page *dummy_page)
5473{
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005474 unsigned long nr_pages = nr_anon + nr_file;
Johannes Weiner747db952014-08-08 14:19:24 -07005475 unsigned long flags;
5476
Johannes Weinerce00a962014-09-05 08:43:57 -04005477 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005478 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08005479 if (do_memsw_account())
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005480 page_counter_uncharge(&memcg->memsw, nr_pages);
Johannes Weinerce00a962014-09-05 08:43:57 -04005481 memcg_oom_recover(memcg);
5482 }
Johannes Weiner747db952014-08-08 14:19:24 -07005483
5484 local_irq_save(flags);
5485 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5486 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5487 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5488 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005489 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005490 memcg_check_events(memcg, dummy_page);
5491 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005492
5493 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005494 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005495}
5496
5497static void uncharge_list(struct list_head *page_list)
5498{
5499 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005500 unsigned long nr_anon = 0;
5501 unsigned long nr_file = 0;
5502 unsigned long nr_huge = 0;
5503 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005504 struct list_head *next;
5505 struct page *page;
5506
Johannes Weiner8b592652016-03-17 14:20:31 -07005507 /*
5508 * Note that the list can be a single page->lru; hence the
5509 * do-while loop instead of a simple list_for_each_entry().
5510 */
Johannes Weiner747db952014-08-08 14:19:24 -07005511 next = page_list->next;
5512 do {
5513 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005514
5515 page = list_entry(next, struct page, lru);
5516 next = page->lru.next;
5517
5518 VM_BUG_ON_PAGE(PageLRU(page), page);
5519 VM_BUG_ON_PAGE(page_count(page), page);
5520
Johannes Weiner1306a852014-12-10 15:44:52 -08005521 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005522 continue;
5523
5524 /*
5525 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005526 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005527 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005528 */
5529
Johannes Weiner1306a852014-12-10 15:44:52 -08005530 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005531 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005532 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5533 nr_huge, page);
5534 pgpgout = nr_anon = nr_file = nr_huge = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005535 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005536 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005537 }
5538
5539 if (PageTransHuge(page)) {
5540 nr_pages <<= compound_order(page);
5541 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5542 nr_huge += nr_pages;
5543 }
5544
5545 if (PageAnon(page))
5546 nr_anon += nr_pages;
5547 else
5548 nr_file += nr_pages;
5549
Johannes Weiner1306a852014-12-10 15:44:52 -08005550 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005551
5552 pgpgout++;
5553 } while (next != page_list);
5554
5555 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005556 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5557 nr_huge, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005558}
5559
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005560/**
5561 * mem_cgroup_uncharge - uncharge a page
5562 * @page: page to uncharge
5563 *
5564 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5565 * mem_cgroup_commit_charge().
5566 */
5567void mem_cgroup_uncharge(struct page *page)
5568{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005569 if (mem_cgroup_disabled())
5570 return;
5571
Johannes Weiner747db952014-08-08 14:19:24 -07005572 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005573 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005574 return;
5575
Johannes Weiner747db952014-08-08 14:19:24 -07005576 INIT_LIST_HEAD(&page->lru);
5577 uncharge_list(&page->lru);
5578}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005579
Johannes Weiner747db952014-08-08 14:19:24 -07005580/**
5581 * mem_cgroup_uncharge_list - uncharge a list of page
5582 * @page_list: list of pages to uncharge
5583 *
5584 * Uncharge a list of pages previously charged with
5585 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5586 */
5587void mem_cgroup_uncharge_list(struct list_head *page_list)
5588{
5589 if (mem_cgroup_disabled())
5590 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005591
Johannes Weiner747db952014-08-08 14:19:24 -07005592 if (!list_empty(page_list))
5593 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005594}
5595
5596/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005597 * mem_cgroup_migrate - charge a page's replacement
5598 * @oldpage: currently circulating page
5599 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005600 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005601 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5602 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005603 *
5604 * Both pages must be locked, @newpage->mapping must be set up.
5605 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005606void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005607{
Johannes Weiner29833312014-12-10 15:44:02 -08005608 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005609 unsigned int nr_pages;
5610 bool compound;
Tejun Heod93c4132016-06-24 14:49:54 -07005611 unsigned long flags;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005612
5613 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5614 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005615 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005616 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5617 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005618
5619 if (mem_cgroup_disabled())
5620 return;
5621
5622 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005623 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005624 return;
5625
Hugh Dickins45637ba2015-11-05 18:49:40 -08005626 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08005627 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005628 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005629 return;
5630
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005631 /* Force-charge the new page. The old one will be freed soon */
5632 compound = PageTransHuge(newpage);
5633 nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5634
5635 page_counter_charge(&memcg->memory, nr_pages);
5636 if (do_memsw_account())
5637 page_counter_charge(&memcg->memsw, nr_pages);
5638 css_get_many(&memcg->css, nr_pages);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005639
Johannes Weiner9cf76662016-03-15 14:57:58 -07005640 commit_charge(newpage, memcg, false);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005641
Tejun Heod93c4132016-06-24 14:49:54 -07005642 local_irq_save(flags);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005643 mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5644 memcg_check_events(memcg, newpage);
Tejun Heod93c4132016-06-24 14:49:54 -07005645 local_irq_restore(flags);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005646}
5647
Johannes Weineref129472016-01-14 15:21:34 -08005648DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08005649EXPORT_SYMBOL(memcg_sockets_enabled_key);
5650
5651void sock_update_memcg(struct sock *sk)
5652{
5653 struct mem_cgroup *memcg;
5654
5655 /* Socket cloning can throw us here with sk_cgrp already
5656 * filled. It won't however, necessarily happen from
5657 * process context. So the test for root memcg given
5658 * the current task's memcg won't help us in this case.
5659 *
5660 * Respecting the original socket's memcg is a better
5661 * decision in this case.
5662 */
5663 if (sk->sk_memcg) {
5664 BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5665 css_get(&sk->sk_memcg->css);
5666 return;
5667 }
5668
5669 rcu_read_lock();
5670 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005671 if (memcg == root_mem_cgroup)
5672 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08005673 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005674 goto out;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005675 if (css_tryget_online(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08005676 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005677out:
Johannes Weiner11092082016-01-14 15:21:26 -08005678 rcu_read_unlock();
5679}
5680EXPORT_SYMBOL(sock_update_memcg);
5681
5682void sock_release_memcg(struct sock *sk)
5683{
5684 WARN_ON(!sk->sk_memcg);
5685 css_put(&sk->sk_memcg->css);
5686}
5687
5688/**
5689 * mem_cgroup_charge_skmem - charge socket memory
5690 * @memcg: memcg to charge
5691 * @nr_pages: number of pages to charge
5692 *
5693 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5694 * @memcg's configured limit, %false if the charge had to be forced.
5695 */
5696bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5697{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005698 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08005699
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005700 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005701 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005702
Johannes Weiner0db15292016-01-20 15:02:50 -08005703 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5704 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005705 return true;
5706 }
Johannes Weiner0db15292016-01-20 15:02:50 -08005707 page_counter_charge(&memcg->tcpmem, nr_pages);
5708 memcg->tcpmem_pressure = 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005709 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08005710 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005711
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005712 /* Don't block in the packet receive path */
5713 if (in_softirq())
5714 gfp_mask = GFP_NOWAIT;
5715
Johannes Weinerb2807f02016-01-20 15:03:22 -08005716 this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5717
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005718 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5719 return true;
5720
5721 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005722 return false;
5723}
5724
5725/**
5726 * mem_cgroup_uncharge_skmem - uncharge socket memory
5727 * @memcg - memcg to uncharge
5728 * @nr_pages - number of pages to uncharge
5729 */
5730void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5731{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005732 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005733 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005734 return;
5735 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005736
Johannes Weinerb2807f02016-01-20 15:03:22 -08005737 this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5738
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005739 page_counter_uncharge(&memcg->memory, nr_pages);
5740 css_put_many(&memcg->css, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005741}
5742
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005743static int __init cgroup_memory(char *s)
5744{
5745 char *token;
5746
5747 while ((token = strsep(&s, ",")) != NULL) {
5748 if (!*token)
5749 continue;
5750 if (!strcmp(token, "nosocket"))
5751 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08005752 if (!strcmp(token, "nokmem"))
5753 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005754 }
5755 return 0;
5756}
5757__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08005758
Michal Hocko2d110852013-02-22 16:34:43 -08005759/*
Michal Hocko10813122013-02-22 16:35:41 -08005760 * subsys_initcall() for memory controller.
5761 *
5762 * Some parts like hotcpu_notifier() have to be initialized from this context
5763 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5764 * everything that doesn't depend on a specific mem_cgroup structure should
5765 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005766 */
5767static int __init mem_cgroup_init(void)
5768{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005769 int cpu, node;
5770
Michal Hocko2d110852013-02-22 16:34:43 -08005771 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005772
5773 for_each_possible_cpu(cpu)
5774 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5775 drain_local_stock);
5776
5777 for_each_node(node) {
5778 struct mem_cgroup_tree_per_node *rtpn;
5779 int zone;
5780
5781 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5782 node_online(node) ? node : NUMA_NO_NODE);
5783
5784 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5785 struct mem_cgroup_tree_per_zone *rtpz;
5786
5787 rtpz = &rtpn->rb_tree_per_zone[zone];
5788 rtpz->rb_root = RB_ROOT;
5789 spin_lock_init(&rtpz->lock);
5790 }
5791 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5792 }
5793
Michal Hocko2d110852013-02-22 16:34:43 -08005794 return 0;
5795}
5796subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005797
5798#ifdef CONFIG_MEMCG_SWAP
5799/**
5800 * mem_cgroup_swapout - transfer a memsw charge to swap
5801 * @page: page whose memsw charge to transfer
5802 * @entry: swap entry to move the charge to
5803 *
5804 * Transfer the memsw charge of @page to @entry.
5805 */
5806void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5807{
5808 struct mem_cgroup *memcg;
5809 unsigned short oldid;
5810
5811 VM_BUG_ON_PAGE(PageLRU(page), page);
5812 VM_BUG_ON_PAGE(page_count(page), page);
5813
Johannes Weiner7941d212016-01-14 15:21:23 -08005814 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08005815 return;
5816
5817 memcg = page->mem_cgroup;
5818
5819 /* Readahead page, never charged */
5820 if (!memcg)
5821 return;
5822
Johannes Weiner73f576c2016-07-20 15:44:57 -07005823 mem_cgroup_id_get(memcg);
Johannes Weiner21afa382015-02-11 15:26:36 -08005824 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5825 VM_BUG_ON_PAGE(oldid, page);
5826 mem_cgroup_swap_statistics(memcg, true);
5827
5828 page->mem_cgroup = NULL;
5829
5830 if (!mem_cgroup_is_root(memcg))
5831 page_counter_uncharge(&memcg->memory, 1);
5832
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07005833 /*
5834 * Interrupts should be disabled here because the caller holds the
5835 * mapping->tree_lock lock which is taken with interrupts-off. It is
5836 * important here to have the interrupts disabled because it is the
5837 * only synchronisation we have for udpating the per-CPU variables.
5838 */
5839 VM_BUG_ON(!irqs_disabled());
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005840 mem_cgroup_charge_statistics(memcg, page, false, -1);
Johannes Weiner21afa382015-02-11 15:26:36 -08005841 memcg_check_events(memcg, page);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005842
5843 if (!mem_cgroup_is_root(memcg))
5844 css_put(&memcg->css);
Johannes Weiner21afa382015-02-11 15:26:36 -08005845}
5846
Vladimir Davydov37e84352016-01-20 15:02:56 -08005847/*
5848 * mem_cgroup_try_charge_swap - try charging a swap entry
5849 * @page: page being added to swap
5850 * @entry: swap entry to charge
5851 *
5852 * Try to charge @entry to the memcg that @page belongs to.
5853 *
5854 * Returns 0 on success, -ENOMEM on failure.
5855 */
5856int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5857{
5858 struct mem_cgroup *memcg;
5859 struct page_counter *counter;
5860 unsigned short oldid;
5861
5862 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5863 return 0;
5864
5865 memcg = page->mem_cgroup;
5866
5867 /* Readahead page, never charged */
5868 if (!memcg)
5869 return 0;
5870
5871 if (!mem_cgroup_is_root(memcg) &&
5872 !page_counter_try_charge(&memcg->swap, 1, &counter))
5873 return -ENOMEM;
5874
Johannes Weiner73f576c2016-07-20 15:44:57 -07005875 mem_cgroup_id_get(memcg);
Vladimir Davydov37e84352016-01-20 15:02:56 -08005876 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5877 VM_BUG_ON_PAGE(oldid, page);
5878 mem_cgroup_swap_statistics(memcg, true);
5879
Vladimir Davydov37e84352016-01-20 15:02:56 -08005880 return 0;
5881}
5882
Johannes Weiner21afa382015-02-11 15:26:36 -08005883/**
5884 * mem_cgroup_uncharge_swap - uncharge a swap entry
5885 * @entry: swap entry to uncharge
5886 *
Vladimir Davydov37e84352016-01-20 15:02:56 -08005887 * Drop the swap charge associated with @entry.
Johannes Weiner21afa382015-02-11 15:26:36 -08005888 */
5889void mem_cgroup_uncharge_swap(swp_entry_t entry)
5890{
5891 struct mem_cgroup *memcg;
5892 unsigned short id;
5893
Vladimir Davydov37e84352016-01-20 15:02:56 -08005894 if (!do_swap_account)
Johannes Weiner21afa382015-02-11 15:26:36 -08005895 return;
5896
5897 id = swap_cgroup_record(entry, 0);
5898 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005899 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005900 if (memcg) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08005901 if (!mem_cgroup_is_root(memcg)) {
5902 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5903 page_counter_uncharge(&memcg->swap, 1);
5904 else
5905 page_counter_uncharge(&memcg->memsw, 1);
5906 }
Johannes Weiner21afa382015-02-11 15:26:36 -08005907 mem_cgroup_swap_statistics(memcg, false);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005908 mem_cgroup_id_put(memcg);
Johannes Weiner21afa382015-02-11 15:26:36 -08005909 }
5910 rcu_read_unlock();
5911}
5912
Vladimir Davydovd8b38432016-01-20 15:03:07 -08005913long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
5914{
5915 long nr_swap_pages = get_nr_swap_pages();
5916
5917 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5918 return nr_swap_pages;
5919 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5920 nr_swap_pages = min_t(long, nr_swap_pages,
5921 READ_ONCE(memcg->swap.limit) -
5922 page_counter_read(&memcg->swap));
5923 return nr_swap_pages;
5924}
5925
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08005926bool mem_cgroup_swap_full(struct page *page)
5927{
5928 struct mem_cgroup *memcg;
5929
5930 VM_BUG_ON_PAGE(!PageLocked(page), page);
5931
5932 if (vm_swap_full())
5933 return true;
5934 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5935 return false;
5936
5937 memcg = page->mem_cgroup;
5938 if (!memcg)
5939 return false;
5940
5941 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5942 if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
5943 return true;
5944
5945 return false;
5946}
5947
Johannes Weiner21afa382015-02-11 15:26:36 -08005948/* for remember boot option*/
5949#ifdef CONFIG_MEMCG_SWAP_ENABLED
5950static int really_do_swap_account __initdata = 1;
5951#else
5952static int really_do_swap_account __initdata;
5953#endif
5954
5955static int __init enable_swap_account(char *s)
5956{
5957 if (!strcmp(s, "1"))
5958 really_do_swap_account = 1;
5959 else if (!strcmp(s, "0"))
5960 really_do_swap_account = 0;
5961 return 1;
5962}
5963__setup("swapaccount=", enable_swap_account);
5964
Vladimir Davydov37e84352016-01-20 15:02:56 -08005965static u64 swap_current_read(struct cgroup_subsys_state *css,
5966 struct cftype *cft)
5967{
5968 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5969
5970 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
5971}
5972
5973static int swap_max_show(struct seq_file *m, void *v)
5974{
5975 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5976 unsigned long max = READ_ONCE(memcg->swap.limit);
5977
5978 if (max == PAGE_COUNTER_MAX)
5979 seq_puts(m, "max\n");
5980 else
5981 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5982
5983 return 0;
5984}
5985
5986static ssize_t swap_max_write(struct kernfs_open_file *of,
5987 char *buf, size_t nbytes, loff_t off)
5988{
5989 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5990 unsigned long max;
5991 int err;
5992
5993 buf = strstrip(buf);
5994 err = page_counter_memparse(buf, "max", &max);
5995 if (err)
5996 return err;
5997
5998 mutex_lock(&memcg_limit_mutex);
5999 err = page_counter_limit(&memcg->swap, max);
6000 mutex_unlock(&memcg_limit_mutex);
6001 if (err)
6002 return err;
6003
6004 return nbytes;
6005}
6006
6007static struct cftype swap_files[] = {
6008 {
6009 .name = "swap.current",
6010 .flags = CFTYPE_NOT_ON_ROOT,
6011 .read_u64 = swap_current_read,
6012 },
6013 {
6014 .name = "swap.max",
6015 .flags = CFTYPE_NOT_ON_ROOT,
6016 .seq_show = swap_max_show,
6017 .write = swap_max_write,
6018 },
6019 { } /* terminate */
6020};
6021
Johannes Weiner21afa382015-02-11 15:26:36 -08006022static struct cftype memsw_cgroup_files[] = {
6023 {
6024 .name = "memsw.usage_in_bytes",
6025 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6026 .read_u64 = mem_cgroup_read_u64,
6027 },
6028 {
6029 .name = "memsw.max_usage_in_bytes",
6030 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6031 .write = mem_cgroup_reset,
6032 .read_u64 = mem_cgroup_read_u64,
6033 },
6034 {
6035 .name = "memsw.limit_in_bytes",
6036 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6037 .write = mem_cgroup_write,
6038 .read_u64 = mem_cgroup_read_u64,
6039 },
6040 {
6041 .name = "memsw.failcnt",
6042 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6043 .write = mem_cgroup_reset,
6044 .read_u64 = mem_cgroup_read_u64,
6045 },
6046 { }, /* terminate */
6047};
6048
6049static int __init mem_cgroup_swap_init(void)
6050{
6051 if (!mem_cgroup_disabled() && really_do_swap_account) {
6052 do_swap_account = 1;
Vladimir Davydov37e84352016-01-20 15:02:56 -08006053 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
6054 swap_files));
Johannes Weiner21afa382015-02-11 15:26:36 -08006055 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
6056 memsw_cgroup_files));
6057 }
6058 return 0;
6059}
6060subsys_initcall(mem_cgroup_swap_init);
6061
6062#endif /* CONFIG_MEMCG_SWAP */