blob: 36db05fa8acb8b49d95130a00de9b22233dac031 [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Johannes Weiner1575e682015-04-14 15:44:51 -070017 * Native page reclaim
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
22 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080023 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
27 *
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
32 */
33
Johannes Weiner3e32cb22014-12-10 15:42:31 -080034#include <linux/page_counter.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080035#include <linux/memcontrol.h>
36#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080037#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080038#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080039#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080040#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080041#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080043#include <linux/bit_spinlock.h>
44#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070045#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040046#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080047#include <linux/mutex.h>
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -070048#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070049#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080051#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080052#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080053#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050054#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080055#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080056#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080057#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070058#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070059#include <linux/mm_inline.h>
Johannes Weiner5d1ea48b2014-12-10 15:44:55 -080060#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080061#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070062#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070063#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050064#include <linux/file.h>
Tejun Heob23afb92015-11-05 18:46:11 -080065#include <linux/tracehook.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080066#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000067#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070068#include <net/ip.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080069#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080070
Balbir Singh8697d332008-02-07 00:13:59 -080071#include <asm/uaccess.h>
72
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070073#include <trace/events/vmscan.h>
74
Tejun Heo073219e2014-02-08 10:36:58 -050075struct cgroup_subsys memory_cgrp_subsys __read_mostly;
76EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080077
Johannes Weiner7d828602016-01-14 15:20:56 -080078struct mem_cgroup *root_mem_cgroup __read_mostly;
79
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070080#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh8cdea7c2008-02-07 00:13:50 -080081
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080082/* Socket memory accounting disabled? */
83static bool cgroup_memory_nosocket;
84
Vladimir Davydov04823c82016-01-20 15:02:38 -080085/* Kernel memory accounting disabled? */
86static bool cgroup_memory_nokmem;
87
Johannes Weiner21afa382015-02-11 15:26:36 -080088/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070089#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080090int do_swap_account __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080091#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070092#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080093#endif
94
Johannes Weiner7941d212016-01-14 15:21:23 -080095/* Whether legacy memory+swap accounting is active */
96static bool do_memsw_account(void)
97{
98 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account;
99}
100
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700101static const char * const mem_cgroup_stat_names[] = {
102 "cache",
103 "rss",
David Rientjesb070e652013-05-07 16:18:09 -0700104 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700105 "mapped_file",
Greg Thelenc4843a72015-05-22 17:13:16 -0400106 "dirty",
Sha Zhengju3ea67d02013-09-12 15:13:53 -0700107 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700108 "swap",
109};
110
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700111static const char * const mem_cgroup_events_names[] = {
112 "pgpgin",
113 "pgpgout",
114 "pgfault",
115 "pgmajfault",
116};
117
Sha Zhengju58cf1882013-02-22 16:32:05 -0800118static const char * const mem_cgroup_lru_names[] = {
119 "inactive_anon",
120 "active_anon",
121 "inactive_file",
122 "active_file",
123 "unevictable",
124};
125
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700126#define THRESHOLDS_EVENTS_TARGET 128
127#define SOFTLIMIT_EVENTS_TARGET 1024
128#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700129
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700130/*
131 * Cgroups above their limits are maintained in a RB-Tree, independent of
132 * their hierarchy representation
133 */
134
135struct mem_cgroup_tree_per_zone {
136 struct rb_root rb_root;
137 spinlock_t lock;
138};
139
140struct mem_cgroup_tree_per_node {
141 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
142};
143
144struct mem_cgroup_tree {
145 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
146};
147
148static struct mem_cgroup_tree soft_limit_tree __read_mostly;
149
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700150/* for OOM */
151struct mem_cgroup_eventfd_list {
152 struct list_head list;
153 struct eventfd_ctx *eventfd;
154};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800155
Tejun Heo79bd9812013-11-22 18:20:42 -0500156/*
157 * cgroup_event represents events which userspace want to receive.
158 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500159struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500160 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500161 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500162 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500163 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500164 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500165 * eventfd to signal userspace about the event.
166 */
167 struct eventfd_ctx *eventfd;
168 /*
169 * Each of these stored in a list by the cgroup.
170 */
171 struct list_head list;
172 /*
Tejun Heofba94802013-11-22 18:20:43 -0500173 * register_event() callback will be used to add new userspace
174 * waiter for changes related to this event. Use eventfd_signal()
175 * on eventfd to send notification to userspace.
176 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500177 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500178 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500179 /*
180 * unregister_event() callback will be called when userspace closes
181 * the eventfd or on cgroup removing. This callback must be set,
182 * if you want provide notification functionality.
183 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500184 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500185 struct eventfd_ctx *eventfd);
186 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500187 * All fields below needed to unregister event when
188 * userspace closes eventfd.
189 */
190 poll_table pt;
191 wait_queue_head_t *wqh;
192 wait_queue_t wait;
193 struct work_struct remove;
194};
195
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700196static void mem_cgroup_threshold(struct mem_cgroup *memcg);
197static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800198
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800199/* Stuffs for move charges at task migration. */
200/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800201 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800202 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800203#define MOVE_ANON 0x1U
204#define MOVE_FILE 0x2U
205#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800206
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800207/* "mc" and its members are protected by cgroup_mutex */
208static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800209 spinlock_t lock; /* for from, to */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800210 struct mem_cgroup *from;
211 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800212 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800213 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800214 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800215 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800216 struct task_struct *moving_task; /* a task moving charges */
217 wait_queue_head_t waitq; /* a waitq for other context */
218} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700219 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800220 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
221};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800222
Balbir Singh4e416952009-09-23 15:56:39 -0700223/*
224 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
225 * limit reclaim to prevent infinite loops, if they ever occur.
226 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700227#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700228#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700229
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800230enum charge_type {
231 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700232 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800233 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700234 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700235 NR_CHARGE_TYPE,
236};
237
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800238/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800239enum res_type {
240 _MEM,
241 _MEMSWAP,
242 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800243 _KMEM,
Vladimir Davydovd55f90b2016-01-20 15:02:44 -0800244 _TCP,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800245};
246
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700247#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
248#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800249#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700250/* Used for OOM nofiier */
251#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800252
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700253/* Some nice accessors for the vmpressure. */
254struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
255{
256 if (!memcg)
257 memcg = root_mem_cgroup;
258 return &memcg->vmpressure;
259}
260
261struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
262{
263 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
264}
265
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700266static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
267{
268 return (memcg == root_mem_cgroup);
269}
270
Johannes Weiner127424c2016-01-20 15:02:32 -0800271#ifndef CONFIG_SLOB
Glauber Costa55007d82012-12-18 14:22:38 -0800272/*
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800273 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800274 * The main reason for not using cgroup id for this:
275 * this works better in sparse environments, where we have a lot of memcgs,
276 * but only a few kmem-limited. Or also, if we have, for instance, 200
277 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
278 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800279 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800280 * The current size of the caches array is stored in memcg_nr_cache_ids. It
281 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800282 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800283static DEFINE_IDA(memcg_cache_ida);
284int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800285
Vladimir Davydov05257a12015-02-12 14:59:01 -0800286/* Protects memcg_nr_cache_ids */
287static DECLARE_RWSEM(memcg_cache_ids_sem);
288
289void memcg_get_cache_ids(void)
290{
291 down_read(&memcg_cache_ids_sem);
292}
293
294void memcg_put_cache_ids(void)
295{
296 up_read(&memcg_cache_ids_sem);
297}
298
Glauber Costa55007d82012-12-18 14:22:38 -0800299/*
300 * MIN_SIZE is different than 1, because we would like to avoid going through
301 * the alloc/free process all the time. In a small machine, 4 kmem-limited
302 * cgroups is a reasonable guess. In the future, it could be a parameter or
303 * tunable, but that is strictly not necessary.
304 *
Li Zefanb8627832013-09-23 16:56:47 +0800305 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800306 * this constant directly from cgroup, but it is understandable that this is
307 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800308 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800309 * increase ours as well if it increases.
310 */
311#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800312#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800313
Glauber Costad7f25f82012-12-18 14:22:40 -0800314/*
315 * A lot of the calls to the cache allocation functions are expected to be
316 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
317 * conditional to this static branch, we'll have to allow modules that does
318 * kmem_cache_alloc and the such to see this symbol as well
319 */
Johannes Weineref129472016-01-14 15:21:34 -0800320DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
Glauber Costad7f25f82012-12-18 14:22:40 -0800321EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800322
Johannes Weiner127424c2016-01-20 15:02:32 -0800323#endif /* !CONFIG_SLOB */
Glauber Costaa8964b92012-12-18 14:22:09 -0800324
Balbir Singhf64c3f52009-09-23 15:56:37 -0700325static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700326mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700327{
Jianyu Zhane2318752014-06-06 14:38:20 -0700328 int nid = zone_to_nid(zone);
329 int zid = zone_idx(zone);
330
Johannes Weiner54f72fe2013-07-08 15:59:49 -0700331 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700332}
333
Tejun Heoad7fa852015-05-27 20:00:02 -0400334/**
335 * mem_cgroup_css_from_page - css of the memcg associated with a page
336 * @page: page of interest
337 *
338 * If memcg is bound to the default hierarchy, css of the memcg associated
339 * with @page is returned. The returned css remains associated with @page
340 * until it is released.
341 *
342 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
343 * is returned.
Tejun Heoad7fa852015-05-27 20:00:02 -0400344 */
345struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
346{
347 struct mem_cgroup *memcg;
348
Tejun Heoad7fa852015-05-27 20:00:02 -0400349 memcg = page->mem_cgroup;
350
Tejun Heo9e10a132015-09-18 11:56:28 -0400351 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400352 memcg = root_mem_cgroup;
353
Tejun Heoad7fa852015-05-27 20:00:02 -0400354 return &memcg->css;
355}
356
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700357/**
358 * page_cgroup_ino - return inode number of the memcg a page is charged to
359 * @page: the page
360 *
361 * Look up the closest online ancestor of the memory cgroup @page is charged to
362 * and return its inode number or 0 if @page is not charged to any cgroup. It
363 * is safe to call this function without holding a reference to @page.
364 *
365 * Note, this function is inherently racy, because there is nothing to prevent
366 * the cgroup inode from getting torn down and potentially reallocated a moment
367 * after page_cgroup_ino() returns, so it only should be used by callers that
368 * do not care (such as procfs interfaces).
369 */
370ino_t page_cgroup_ino(struct page *page)
371{
372 struct mem_cgroup *memcg;
373 unsigned long ino = 0;
374
375 rcu_read_lock();
376 memcg = READ_ONCE(page->mem_cgroup);
377 while (memcg && !(memcg->css.flags & CSS_ONLINE))
378 memcg = parent_mem_cgroup(memcg);
379 if (memcg)
380 ino = cgroup_ino(memcg->css.cgroup);
381 rcu_read_unlock();
382 return ino;
383}
384
Balbir Singhf64c3f52009-09-23 15:56:37 -0700385static struct mem_cgroup_per_zone *
Jianyu Zhane2318752014-06-06 14:38:20 -0700386mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700387{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700388 int nid = page_to_nid(page);
389 int zid = page_zonenum(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700390
Jianyu Zhane2318752014-06-06 14:38:20 -0700391 return &memcg->nodeinfo[nid]->zoneinfo[zid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700392}
393
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700394static struct mem_cgroup_tree_per_zone *
395soft_limit_tree_node_zone(int nid, int zid)
396{
397 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
398}
399
400static struct mem_cgroup_tree_per_zone *
401soft_limit_tree_from_page(struct page *page)
402{
403 int nid = page_to_nid(page);
404 int zid = page_zonenum(page);
405
406 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
407}
408
Johannes Weinercf2c8122014-06-06 14:38:21 -0700409static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz,
410 struct mem_cgroup_tree_per_zone *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800411 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700412{
413 struct rb_node **p = &mctz->rb_root.rb_node;
414 struct rb_node *parent = NULL;
415 struct mem_cgroup_per_zone *mz_node;
416
417 if (mz->on_tree)
418 return;
419
420 mz->usage_in_excess = new_usage_in_excess;
421 if (!mz->usage_in_excess)
422 return;
423 while (*p) {
424 parent = *p;
425 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
426 tree_node);
427 if (mz->usage_in_excess < mz_node->usage_in_excess)
428 p = &(*p)->rb_left;
429 /*
430 * We can't avoid mem cgroups that are over their soft
431 * limit by the same amount
432 */
433 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
434 p = &(*p)->rb_right;
435 }
436 rb_link_node(&mz->tree_node, parent, p);
437 rb_insert_color(&mz->tree_node, &mctz->rb_root);
438 mz->on_tree = true;
439}
440
Johannes Weinercf2c8122014-06-06 14:38:21 -0700441static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
442 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700443{
444 if (!mz->on_tree)
445 return;
446 rb_erase(&mz->tree_node, &mctz->rb_root);
447 mz->on_tree = false;
448}
449
Johannes Weinercf2c8122014-06-06 14:38:21 -0700450static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
451 struct mem_cgroup_tree_per_zone *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700452{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700453 unsigned long flags;
454
455 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700456 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700457 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700458}
459
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800460static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
461{
462 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700463 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800464 unsigned long excess = 0;
465
466 if (nr_pages > soft_limit)
467 excess = nr_pages - soft_limit;
468
469 return excess;
470}
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700471
472static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
473{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800474 unsigned long excess;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700475 struct mem_cgroup_per_zone *mz;
476 struct mem_cgroup_tree_per_zone *mctz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700477
Jianyu Zhane2318752014-06-06 14:38:20 -0700478 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700479 /*
480 * Necessary to update all ancestors when hierarchy is used.
481 * because their event counter is not touched.
482 */
483 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Jianyu Zhane2318752014-06-06 14:38:20 -0700484 mz = mem_cgroup_page_zoneinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800485 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700486 /*
487 * We have to update the tree if mz is on RB-tree or
488 * mem is over its softlimit.
489 */
490 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700491 unsigned long flags;
492
493 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700494 /* if on-tree, remove it */
495 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700496 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700497 /*
498 * Insert again. mz->usage_in_excess will be updated.
499 * If excess is 0, no tree ops.
500 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700501 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700502 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700503 }
504 }
505}
506
507static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
508{
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700509 struct mem_cgroup_tree_per_zone *mctz;
Jianyu Zhane2318752014-06-06 14:38:20 -0700510 struct mem_cgroup_per_zone *mz;
511 int nid, zid;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700512
Jianyu Zhane2318752014-06-06 14:38:20 -0700513 for_each_node(nid) {
514 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
515 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
516 mctz = soft_limit_tree_node_zone(nid, zid);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700517 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700518 }
519 }
520}
521
522static struct mem_cgroup_per_zone *
523__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
524{
525 struct rb_node *rightmost = NULL;
526 struct mem_cgroup_per_zone *mz;
527
528retry:
529 mz = NULL;
530 rightmost = rb_last(&mctz->rb_root);
531 if (!rightmost)
532 goto done; /* Nothing to reclaim from */
533
534 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
535 /*
536 * Remove the node now but someone else can add it back,
537 * we will to add it back at the end of reclaim to its correct
538 * position in the tree.
539 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700540 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800541 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400542 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700543 goto retry;
544done:
545 return mz;
546}
547
548static struct mem_cgroup_per_zone *
549mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
550{
551 struct mem_cgroup_per_zone *mz;
552
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700553 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700554 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700555 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700556 return mz;
557}
558
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700559/*
Greg Thelen484ebb32015-10-01 15:37:05 -0700560 * Return page count for single (non recursive) @memcg.
561 *
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700562 * Implementation Note: reading percpu statistics for memcg.
563 *
564 * Both of vmstat[] and percpu_counter has threshold and do periodic
565 * synchronization to implement "quick" read. There are trade-off between
566 * reading cost and precision of value. Then, we may have a chance to implement
Greg Thelen484ebb32015-10-01 15:37:05 -0700567 * a periodic synchronization of counter in memcg's counter.
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700568 *
569 * But this _read() function is used for user interface now. The user accounts
570 * memory usage by memory cgroup and he _always_ requires exact value because
571 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
572 * have to visit all online cpus and make sum. So, for now, unnecessary
573 * synchronization is not implemented. (just implemented for cpu hotplug)
574 *
575 * If there are kernel internal actions which can make use of some not-exact
576 * value, and reading all cpu value can be performance bottleneck in some
Greg Thelen484ebb32015-10-01 15:37:05 -0700577 * common workload, threshold and synchronization as vmstat[] should be
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700578 * implemented.
579 */
Greg Thelen484ebb32015-10-01 15:37:05 -0700580static unsigned long
581mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800582{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700583 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800584 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800585
Greg Thelen484ebb32015-10-01 15:37:05 -0700586 /* Per-cpu values can be negative, use a signed accumulator */
Tejun Heo733a5722015-05-22 18:23:18 -0400587 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700588 val += per_cpu(memcg->stat->count[idx], cpu);
Greg Thelen484ebb32015-10-01 15:37:05 -0700589 /*
590 * Summing races with updates, so val may be negative. Avoid exposing
591 * transient negative values.
592 */
593 if (val < 0)
594 val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800595 return val;
596}
597
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700598static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700599 enum mem_cgroup_events_index idx)
600{
601 unsigned long val = 0;
602 int cpu;
603
Tejun Heo733a5722015-05-22 18:23:18 -0400604 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700605 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700606 return val;
607}
608
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700609static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700610 struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800611 bool compound, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800612{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700613 /*
614 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
615 * counted as CACHE even if it's on ANON LRU.
616 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700617 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700618 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700619 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800620 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700621 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700622 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700623
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800624 if (compound) {
625 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
David Rientjesb070e652013-05-07 16:18:09 -0700626 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
627 nr_pages);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800628 }
David Rientjesb070e652013-05-07 16:18:09 -0700629
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800630 /* pagein of a big page is an event. So, ignore page size */
631 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700632 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800633 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700634 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800635 nr_pages = -nr_pages; /* for event */
636 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800637
Johannes Weiner13114712012-05-29 15:07:07 -0700638 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800639}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800640
Vladimir Davydov0a6b76d2016-03-17 14:18:42 -0700641unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
642 int nid, unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700643{
Jianyu Zhane2318752014-06-06 14:38:20 -0700644 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700645 int zid;
646
Jianyu Zhane2318752014-06-06 14:38:20 -0700647 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700648
Jianyu Zhane2318752014-06-06 14:38:20 -0700649 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
650 struct mem_cgroup_per_zone *mz;
651 enum lru_list lru;
652
653 for_each_lru(lru) {
654 if (!(BIT(lru) & lru_mask))
655 continue;
656 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
657 nr += mz->lru_size[lru];
658 }
659 }
660 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700661}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700662
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700663static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700664 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800665{
Jianyu Zhane2318752014-06-06 14:38:20 -0700666 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700667 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800668
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800669 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700670 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
671 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800672}
673
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800674static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
675 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800676{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700677 unsigned long val, next;
678
Johannes Weiner13114712012-05-29 15:07:07 -0700679 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700680 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700681 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800682 if ((long)next - (long)val < 0) {
683 switch (target) {
684 case MEM_CGROUP_TARGET_THRESH:
685 next = val + THRESHOLDS_EVENTS_TARGET;
686 break;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700687 case MEM_CGROUP_TARGET_SOFTLIMIT:
688 next = val + SOFTLIMIT_EVENTS_TARGET;
689 break;
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800690 case MEM_CGROUP_TARGET_NUMAINFO:
691 next = val + NUMAINFO_EVENTS_TARGET;
692 break;
693 default:
694 break;
695 }
696 __this_cpu_write(memcg->stat->targets[target], next);
697 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700698 }
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800699 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800700}
701
702/*
703 * Check events in order.
704 *
705 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700706static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800707{
708 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800709 if (unlikely(mem_cgroup_event_ratelimit(memcg,
710 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700711 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800712 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800713
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700714 do_softlimit = mem_cgroup_event_ratelimit(memcg,
715 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -0700716#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800717 do_numainfo = mem_cgroup_event_ratelimit(memcg,
718 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -0700719#endif
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800720 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700721 if (unlikely(do_softlimit))
722 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800723#if MAX_NUMNODES > 1
724 if (unlikely(do_numainfo))
725 atomic_inc(&memcg->numainfo_events);
726#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700727 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800728}
729
Balbir Singhcf475ad2008-04-29 01:00:16 -0700730struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800731{
Balbir Singh31a78f22008-09-28 23:09:31 +0100732 /*
733 * mm_update_next_owner() may clear mm->owner to NULL
734 * if it races with swapoff, page migration, etc.
735 * So this can be called with p == NULL.
736 */
737 if (unlikely(!p))
738 return NULL;
739
Tejun Heo073219e2014-02-08 10:36:58 -0500740 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800741}
Michal Hocko33398cf2015-09-08 15:01:02 -0700742EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800743
Johannes Weinerdf381972014-04-07 15:37:43 -0700744static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800745{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700746 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700747
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800748 rcu_read_lock();
749 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700750 /*
751 * Page cache insertions can happen withou an
752 * actual mm context, e.g. during disk probing
753 * on boot, loopback IO, acct() writes etc.
754 */
755 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700756 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700757 else {
758 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
759 if (unlikely(!memcg))
760 memcg = root_mem_cgroup;
761 }
Tejun Heoec903c02014-05-13 12:11:01 -0400762 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800763 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700764 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800765}
766
Johannes Weiner56600482012-01-12 17:17:59 -0800767/**
768 * mem_cgroup_iter - iterate over memory cgroup hierarchy
769 * @root: hierarchy root
770 * @prev: previously returned memcg, NULL on first invocation
771 * @reclaim: cookie for shared reclaim walks, NULL for full walks
772 *
773 * Returns references to children of the hierarchy below @root, or
774 * @root itself, or %NULL after a full round-trip.
775 *
776 * Caller must pass the return value in @prev on subsequent
777 * invocations for reference counting, or use mem_cgroup_iter_break()
778 * to cancel a hierarchy walk before the round-trip is complete.
779 *
780 * Reclaimers can specify a zone and a priority level in @reclaim to
781 * divide up the memcgs in the hierarchy among all concurrent
782 * reclaimers operating on the same zone and priority.
783 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700784struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800785 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700786 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700787{
Michal Hocko33398cf2015-09-08 15:01:02 -0700788 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800789 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800790 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800791 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700792
Andrew Morton694fbc02013-09-24 15:27:37 -0700793 if (mem_cgroup_disabled())
794 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800795
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700796 if (!root)
797 root = root_mem_cgroup;
798
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800799 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800800 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800801
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800802 if (!root->use_hierarchy && root != root_mem_cgroup) {
803 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800804 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700805 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800806 }
807
Michal Hocko542f85f2013-04-29 15:07:15 -0700808 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800809
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800810 if (reclaim) {
811 struct mem_cgroup_per_zone *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800812
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800813 mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone);
814 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700815
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800816 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700817 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800818
Vladimir Davydov6df38682015-12-29 14:54:10 -0800819 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -0700820 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800821 if (!pos || css_tryget(&pos->css))
822 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800823 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800824 * css reference reached zero, so iter->position will
825 * be cleared by ->css_released. However, we should not
826 * rely on this happening soon, because ->css_released
827 * is called from a work queue, and by busy-waiting we
828 * might block it. So we clear iter->position right
829 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800830 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800831 (void)cmpxchg(&iter->position, pos, NULL);
832 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800833 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800834
835 if (pos)
836 css = &pos->css;
837
838 for (;;) {
839 css = css_next_descendant_pre(css, &root->css);
840 if (!css) {
841 /*
842 * Reclaimers share the hierarchy walk, and a
843 * new one might jump in right at the end of
844 * the hierarchy - make sure they see at least
845 * one group and restart from the beginning.
846 */
847 if (!prev)
848 continue;
849 break;
850 }
851
852 /*
853 * Verify the css and acquire a reference. The root
854 * is provided by the caller, so we know it's alive
855 * and kicking, and don't take an extra reference.
856 */
857 memcg = mem_cgroup_from_css(css);
858
859 if (css == &root->css)
860 break;
861
Johannes Weiner0b8f73e2016-01-20 15:02:53 -0800862 if (css_tryget(css))
863 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800864
865 memcg = NULL;
866 }
867
868 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800869 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800870 * The position could have already been updated by a competing
871 * thread, so check that the value hasn't changed since we read
872 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800873 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800874 (void)cmpxchg(&iter->position, pos, memcg);
875
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800876 if (pos)
877 css_put(&pos->css);
878
879 if (!memcg)
880 iter->generation++;
881 else if (!prev)
882 reclaim->generation = iter->generation;
883 }
884
Michal Hocko542f85f2013-04-29 15:07:15 -0700885out_unlock:
886 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800887out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700888 if (prev && prev != root)
889 css_put(&prev->css);
890
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800891 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700892}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800893
Johannes Weiner56600482012-01-12 17:17:59 -0800894/**
895 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
896 * @root: hierarchy root
897 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
898 */
899void mem_cgroup_iter_break(struct mem_cgroup *root,
900 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800901{
902 if (!root)
903 root = root_mem_cgroup;
904 if (prev && prev != root)
905 css_put(&prev->css);
906}
907
Vladimir Davydov6df38682015-12-29 14:54:10 -0800908static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
909{
910 struct mem_cgroup *memcg = dead_memcg;
911 struct mem_cgroup_reclaim_iter *iter;
912 struct mem_cgroup_per_zone *mz;
913 int nid, zid;
914 int i;
915
916 while ((memcg = parent_mem_cgroup(memcg))) {
917 for_each_node(nid) {
918 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
919 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
920 for (i = 0; i <= DEF_PRIORITY; i++) {
921 iter = &mz->iter[i];
922 cmpxchg(&iter->position,
923 dead_memcg, NULL);
924 }
925 }
926 }
927 }
928}
929
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700930/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800931 * Iteration constructs for visiting all cgroups (under a tree). If
932 * loops are exited prematurely (break), mem_cgroup_iter_break() must
933 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700934 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800935#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800936 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800937 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800938 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700939
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800940#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800941 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800942 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800943 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700944
Johannes Weiner925b7672012-01-12 17:18:15 -0800945/**
946 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
947 * @zone: zone of the wanted lruvec
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700948 * @memcg: memcg of the wanted lruvec
Johannes Weiner925b7672012-01-12 17:18:15 -0800949 *
950 * Returns the lru list vector holding pages for the given @zone and
951 * @mem. This can be the global zone lruvec, if the memory controller
952 * is disabled.
953 */
954struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
955 struct mem_cgroup *memcg)
956{
957 struct mem_cgroup_per_zone *mz;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800958 struct lruvec *lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -0800959
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800960 if (mem_cgroup_disabled()) {
961 lruvec = &zone->lruvec;
962 goto out;
963 }
Johannes Weiner925b7672012-01-12 17:18:15 -0800964
Jianyu Zhane2318752014-06-06 14:38:20 -0700965 mz = mem_cgroup_zone_zoneinfo(memcg, zone);
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800966 lruvec = &mz->lruvec;
967out:
968 /*
969 * Since a node can be onlined after the mem_cgroup was created,
970 * we have to be prepared to initialize lruvec->zone here;
971 * and if offlined then reonlined, we need to reinitialize it.
972 */
973 if (unlikely(lruvec->zone != zone))
974 lruvec->zone = zone;
975 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -0800976}
977
Johannes Weiner925b7672012-01-12 17:18:15 -0800978/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800979 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -0800980 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700981 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800982 *
983 * This function is only safe when following the LRU page isolation
984 * and putback protocol: the LRU lock must be held, and the page must
985 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -0700986 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700987struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
Minchan Kim3f58a822011-03-22 16:32:53 -0700988{
989 struct mem_cgroup_per_zone *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -0800990 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800991 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800992
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800993 if (mem_cgroup_disabled()) {
994 lruvec = &zone->lruvec;
995 goto out;
996 }
Christoph Lameterb69408e2008-10-18 20:26:14 -0700997
Johannes Weiner1306a852014-12-10 15:44:52 -0800998 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800999 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001000 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -08001001 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -08001002 */
Johannes Weiner29833312014-12-10 15:44:02 -08001003 if (!memcg)
1004 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001005
Jianyu Zhane2318752014-06-06 14:38:20 -07001006 mz = mem_cgroup_page_zoneinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001007 lruvec = &mz->lruvec;
1008out:
1009 /*
1010 * Since a node can be onlined after the mem_cgroup was created,
1011 * we have to be prepared to initialize lruvec->zone here;
1012 * and if offlined then reonlined, we need to reinitialize it.
1013 */
1014 if (unlikely(lruvec->zone != zone))
1015 lruvec->zone = zone;
1016 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001017}
1018
1019/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001020 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1021 * @lruvec: mem_cgroup per zone lru vector
1022 * @lru: index of lru list the page is sitting on
1023 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001024 *
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001025 * This function must be called when a page is added to or removed from an
1026 * lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001027 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001028void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1029 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001030{
1031 struct mem_cgroup_per_zone *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001032 unsigned long *lru_size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001033
1034 if (mem_cgroup_disabled())
1035 return;
1036
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001037 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1038 lru_size = mz->lru_size + lru;
1039 *lru_size += nr_pages;
1040 VM_BUG_ON((long)(*lru_size) < 0);
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001041}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001042
Johannes Weiner2314b422014-12-10 15:44:33 -08001043bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001044{
Johannes Weiner2314b422014-12-10 15:44:33 -08001045 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001046 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001047 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001048
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001049 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001050 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001051 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001052 task_unlock(p);
1053 } else {
1054 /*
1055 * All threads may have already detached their mm's, but the oom
1056 * killer still needs to detect if they have already been oom
1057 * killed to prevent needlessly killing additional tasks.
1058 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001059 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001060 task_memcg = mem_cgroup_from_task(task);
1061 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001062 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001063 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001064 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1065 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001066 return ret;
1067}
1068
Johannes Weiner19942822011-02-01 15:52:43 -08001069/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001070 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001071 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001072 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001073 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001074 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001075 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001076static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001077{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001078 unsigned long margin = 0;
1079 unsigned long count;
1080 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001081
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001082 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001083 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001084 if (count < limit)
1085 margin = limit - count;
1086
Johannes Weiner7941d212016-01-14 15:21:23 -08001087 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001088 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001089 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001090 if (count <= limit)
1091 margin = min(margin, limit - count);
1092 }
1093
1094 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001095}
1096
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001097/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001098 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001099 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001100 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1101 * moving cgroups. This is for waiting at high-memory pressure
1102 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001103 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001104static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001105{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001106 struct mem_cgroup *from;
1107 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001108 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001109 /*
1110 * Unlike task_move routines, we access mc.to, mc.from not under
1111 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1112 */
1113 spin_lock(&mc.lock);
1114 from = mc.from;
1115 to = mc.to;
1116 if (!from)
1117 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001118
Johannes Weiner2314b422014-12-10 15:44:33 -08001119 ret = mem_cgroup_is_descendant(from, memcg) ||
1120 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001121unlock:
1122 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001123 return ret;
1124}
1125
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001126static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001127{
1128 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001129 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001130 DEFINE_WAIT(wait);
1131 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1132 /* moving charge context might have finished. */
1133 if (mc.moving_task)
1134 schedule();
1135 finish_wait(&mc.waitq, &wait);
1136 return true;
1137 }
1138 }
1139 return false;
1140}
1141
Sha Zhengju58cf1882013-02-22 16:32:05 -08001142#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001143/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001144 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001145 * @memcg: The memory cgroup that went over limit
1146 * @p: Task that is going to be killed
1147 *
1148 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1149 * enabled
1150 */
1151void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1152{
Sha Zhengju58cf1882013-02-22 16:32:05 -08001153 struct mem_cgroup *iter;
1154 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001155
Balbir Singhe2224322009-04-02 16:57:39 -07001156 rcu_read_lock();
1157
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001158 if (p) {
1159 pr_info("Task in ");
1160 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1161 pr_cont(" killed as a result of limit of ");
1162 } else {
1163 pr_info("Memory limit reached of cgroup ");
1164 }
1165
Tejun Heoe61734c2014-02-12 09:29:50 -05001166 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001167 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001168
Balbir Singhe2224322009-04-02 16:57:39 -07001169 rcu_read_unlock();
1170
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001171 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1172 K((u64)page_counter_read(&memcg->memory)),
1173 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1174 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1175 K((u64)page_counter_read(&memcg->memsw)),
1176 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1177 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1178 K((u64)page_counter_read(&memcg->kmem)),
1179 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001180
1181 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001182 pr_info("Memory cgroup stats for ");
1183 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001184 pr_cont(":");
1185
1186 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08001187 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Sha Zhengju58cf1882013-02-22 16:32:05 -08001188 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07001189 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
Sha Zhengju58cf1882013-02-22 16:32:05 -08001190 K(mem_cgroup_read_stat(iter, i)));
1191 }
1192
1193 for (i = 0; i < NR_LRU_LISTS; i++)
1194 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1195 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1196
1197 pr_cont("\n");
1198 }
Balbir Singhe2224322009-04-02 16:57:39 -07001199}
1200
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001201/*
1202 * This function returns the number of memcg under hierarchy tree. Returns
1203 * 1(self count) if no children.
1204 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001205static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001206{
1207 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001208 struct mem_cgroup *iter;
1209
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001210 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001211 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001212 return num;
1213}
1214
Balbir Singh6d61ef42009-01-07 18:08:06 -08001215/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001216 * Return the memory (and swap, if configured) limit for a memcg.
1217 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001218static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001219{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001220 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001221
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001222 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001223 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001224 unsigned long memsw_limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001225 unsigned long swap_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001226
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001227 memsw_limit = memcg->memsw.limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001228 swap_limit = memcg->swap.limit;
1229 swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1230 limit = min(limit + swap_limit, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001231 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001232 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001233}
1234
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001235static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001236 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001237{
David Rientjes6e0fc462015-09-08 15:00:36 -07001238 struct oom_control oc = {
1239 .zonelist = NULL,
1240 .nodemask = NULL,
1241 .gfp_mask = gfp_mask,
1242 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001243 };
David Rientjes9cbb78b2012-07-31 16:43:44 -07001244 struct mem_cgroup *iter;
1245 unsigned long chosen_points = 0;
1246 unsigned long totalpages;
1247 unsigned int points = 0;
1248 struct task_struct *chosen = NULL;
1249
Johannes Weinerdc564012015-06-24 16:57:19 -07001250 mutex_lock(&oom_lock);
1251
David Rientjes876aafb2012-07-31 16:43:48 -07001252 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001253 * If current has a pending SIGKILL or is exiting, then automatically
1254 * select it. The goal is to allow it to allocate so that it may
1255 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001256 */
Oleg Nesterovd003f372014-12-12 16:56:24 -08001257 if (fatal_signal_pending(current) || task_will_free_mem(current)) {
Johannes Weiner16e95192015-06-24 16:57:07 -07001258 mark_oom_victim(current);
Johannes Weinerdc564012015-06-24 16:57:19 -07001259 goto unlock;
David Rientjes876aafb2012-07-31 16:43:48 -07001260 }
1261
David Rientjes6e0fc462015-09-08 15:00:36 -07001262 check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001263 totalpages = mem_cgroup_get_limit(memcg) ? : 1;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001264 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001265 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001266 struct task_struct *task;
1267
Tejun Heo72ec7022013-08-08 20:11:26 -04001268 css_task_iter_start(&iter->css, &it);
1269 while ((task = css_task_iter_next(&it))) {
David Rientjes6e0fc462015-09-08 15:00:36 -07001270 switch (oom_scan_process_thread(&oc, task, totalpages)) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001271 case OOM_SCAN_SELECT:
1272 if (chosen)
1273 put_task_struct(chosen);
1274 chosen = task;
1275 chosen_points = ULONG_MAX;
1276 get_task_struct(chosen);
1277 /* fall through */
1278 case OOM_SCAN_CONTINUE:
1279 continue;
1280 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001281 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001282 mem_cgroup_iter_break(memcg, iter);
1283 if (chosen)
1284 put_task_struct(chosen);
Johannes Weinerdc564012015-06-24 16:57:19 -07001285 goto unlock;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001286 case OOM_SCAN_OK:
1287 break;
1288 };
1289 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001290 if (!points || points < chosen_points)
1291 continue;
1292 /* Prefer thread group leaders for display purposes */
1293 if (points == chosen_points &&
1294 thread_group_leader(chosen))
1295 continue;
1296
1297 if (chosen)
1298 put_task_struct(chosen);
1299 chosen = task;
1300 chosen_points = points;
1301 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001302 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001303 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001304 }
1305
Johannes Weinerdc564012015-06-24 16:57:19 -07001306 if (chosen) {
1307 points = chosen_points * 1000 / totalpages;
David Rientjes6e0fc462015-09-08 15:00:36 -07001308 oom_kill_process(&oc, chosen, points, totalpages, memcg,
1309 "Memory cgroup out of memory");
Johannes Weinerdc564012015-06-24 16:57:19 -07001310 }
1311unlock:
1312 mutex_unlock(&oom_lock);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001313 return chosen;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001314}
1315
Michele Curtiae6e71d2014-12-12 16:56:35 -08001316#if MAX_NUMNODES > 1
1317
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001318/**
1319 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001320 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001321 * @nid: the node ID to be checked.
1322 * @noswap : specify true here if the user wants flle only information.
1323 *
1324 * This function returns whether the specified memcg contains any
1325 * reclaimable pages on a node. Returns true if there are any reclaimable
1326 * pages in the node.
1327 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001328static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001329 int nid, bool noswap)
1330{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001331 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001332 return true;
1333 if (noswap || !total_swap_pages)
1334 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001335 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001336 return true;
1337 return false;
1338
1339}
Ying Han889976d2011-05-26 16:25:33 -07001340
1341/*
1342 * Always updating the nodemask is not very good - even if we have an empty
1343 * list or the wrong list here, we can start from some node and traverse all
1344 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1345 *
1346 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001347static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001348{
1349 int nid;
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001350 /*
1351 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1352 * pagein/pageout changes since the last update.
1353 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001354 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001355 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001356 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001357 return;
1358
Ying Han889976d2011-05-26 16:25:33 -07001359 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001360 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001361
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001362 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001363
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001364 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1365 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001366 }
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001367
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001368 atomic_set(&memcg->numainfo_events, 0);
1369 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001370}
1371
1372/*
1373 * Selecting a node where we start reclaim from. Because what we need is just
1374 * reducing usage counter, start from anywhere is O,K. Considering
1375 * memory reclaim from current node, there are pros. and cons.
1376 *
1377 * Freeing memory from current node means freeing memory from a node which
1378 * we'll use or we've used. So, it may make LRU bad. And if several threads
1379 * hit limits, it will see a contention on a node. But freeing from remote
1380 * node means more costs for memory reclaim because of memory latency.
1381 *
1382 * Now, we use round-robin. Better algorithm is welcomed.
1383 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001384int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001385{
1386 int node;
1387
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001388 mem_cgroup_may_update_nodemask(memcg);
1389 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001390
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001391 node = next_node(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001392 if (node == MAX_NUMNODES)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001393 node = first_node(memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001394 /*
1395 * We call this when we hit limit, not when pages are added to LRU.
1396 * No LRU may hold pages because all pages are UNEVICTABLE or
1397 * memcg is too small and all pages are not on LRU. In that case,
1398 * we use curret node.
1399 */
1400 if (unlikely(node == MAX_NUMNODES))
1401 node = numa_node_id();
1402
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001403 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001404 return node;
1405}
Ying Han889976d2011-05-26 16:25:33 -07001406#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001407int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001408{
1409 return 0;
1410}
1411#endif
1412
Andrew Morton0608f432013-09-24 15:27:41 -07001413static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1414 struct zone *zone,
1415 gfp_t gfp_mask,
1416 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001417{
Andrew Morton0608f432013-09-24 15:27:41 -07001418 struct mem_cgroup *victim = NULL;
1419 int total = 0;
1420 int loop = 0;
1421 unsigned long excess;
1422 unsigned long nr_scanned;
1423 struct mem_cgroup_reclaim_cookie reclaim = {
1424 .zone = zone,
1425 .priority = 0,
1426 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001427
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001428 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001429
Andrew Morton0608f432013-09-24 15:27:41 -07001430 while (1) {
1431 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1432 if (!victim) {
1433 loop++;
1434 if (loop >= 2) {
1435 /*
1436 * If we have not been able to reclaim
1437 * anything, it might because there are
1438 * no reclaimable pages under this hierarchy
1439 */
1440 if (!total)
1441 break;
1442 /*
1443 * We want to do more targeted reclaim.
1444 * excess >> 2 is not to excessive so as to
1445 * reclaim too much, nor too less that we keep
1446 * coming back to reclaim from this cgroup
1447 */
1448 if (total >= (excess >> 2) ||
1449 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1450 break;
1451 }
1452 continue;
1453 }
Andrew Morton0608f432013-09-24 15:27:41 -07001454 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1455 zone, &nr_scanned);
1456 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001457 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001458 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001459 }
Andrew Morton0608f432013-09-24 15:27:41 -07001460 mem_cgroup_iter_break(root_memcg, victim);
1461 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001462}
1463
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001464#ifdef CONFIG_LOCKDEP
1465static struct lockdep_map memcg_oom_lock_dep_map = {
1466 .name = "memcg_oom_lock",
1467};
1468#endif
1469
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001470static DEFINE_SPINLOCK(memcg_oom_lock);
1471
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001472/*
1473 * Check OOM-Killer is already running under our hierarchy.
1474 * If someone is running, return false.
1475 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001476static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001477{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001478 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001479
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001480 spin_lock(&memcg_oom_lock);
1481
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001482 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001483 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001484 /*
1485 * this subtree of our hierarchy is already locked
1486 * so we cannot give a lock.
1487 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001488 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001489 mem_cgroup_iter_break(memcg, iter);
1490 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001491 } else
1492 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001493 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001494
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001495 if (failed) {
1496 /*
1497 * OK, we failed to lock the whole subtree so we have
1498 * to clean up what we set up to the failing subtree
1499 */
1500 for_each_mem_cgroup_tree(iter, memcg) {
1501 if (iter == failed) {
1502 mem_cgroup_iter_break(memcg, iter);
1503 break;
1504 }
1505 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001506 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001507 } else
1508 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001509
1510 spin_unlock(&memcg_oom_lock);
1511
1512 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001513}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001514
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001515static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001516{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001517 struct mem_cgroup *iter;
1518
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001519 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001520 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001521 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001522 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001523 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001524}
1525
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001526static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001527{
1528 struct mem_cgroup *iter;
1529
Tejun Heoc2b42d32015-06-24 16:58:23 -07001530 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001531 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001532 iter->under_oom++;
1533 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001534}
1535
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001536static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001537{
1538 struct mem_cgroup *iter;
1539
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001540 /*
1541 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001542 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001543 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001544 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001545 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001546 if (iter->under_oom > 0)
1547 iter->under_oom--;
1548 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001549}
1550
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001551static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1552
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001553struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001554 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001555 wait_queue_t wait;
1556};
1557
1558static int memcg_oom_wake_function(wait_queue_t *wait,
1559 unsigned mode, int sync, void *arg)
1560{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001561 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1562 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001563 struct oom_wait_info *oom_wait_info;
1564
1565 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001566 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001567
Johannes Weiner2314b422014-12-10 15:44:33 -08001568 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1569 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001570 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001571 return autoremove_wake_function(wait, mode, sync, arg);
1572}
1573
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001574static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001575{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001576 /*
1577 * For the following lockless ->under_oom test, the only required
1578 * guarantee is that it must see the state asserted by an OOM when
1579 * this function is called as a result of userland actions
1580 * triggered by the notification of the OOM. This is trivially
1581 * achieved by invoking mem_cgroup_mark_under_oom() before
1582 * triggering notification.
1583 */
1584 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001585 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001586}
1587
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001588static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001589{
Tejun Heo626ebc42015-11-05 18:46:09 -08001590 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001591 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001592 /*
Johannes Weiner494264202013-10-16 13:46:59 -07001593 * We are in the middle of the charge context here, so we
1594 * don't want to block when potentially sitting on a callstack
1595 * that holds all kinds of filesystem and mm locks.
1596 *
1597 * Also, the caller may handle a failed allocation gracefully
1598 * (like optional page cache readahead) and so an OOM killer
1599 * invocation might not even be necessary.
1600 *
1601 * That's why we don't do anything here except remember the
1602 * OOM context and then deal with it at the end of the page
1603 * fault when the stack is unwound, the locks are released,
1604 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001605 */
Johannes Weiner494264202013-10-16 13:46:59 -07001606 css_get(&memcg->css);
Tejun Heo626ebc42015-11-05 18:46:09 -08001607 current->memcg_in_oom = memcg;
1608 current->memcg_oom_gfp_mask = mask;
1609 current->memcg_oom_order = order;
Johannes Weiner494264202013-10-16 13:46:59 -07001610}
1611
1612/**
1613 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1614 * @handle: actually kill/wait or just clean up the OOM state
1615 *
1616 * This has to be called at the end of a page fault if the memcg OOM
1617 * handler was enabled.
1618 *
1619 * Memcg supports userspace OOM handling where failed allocations must
1620 * sleep on a waitqueue until the userspace task resolves the
1621 * situation. Sleeping directly in the charge context with all kinds
1622 * of locks held is not a good idea, instead we remember an OOM state
1623 * in the task and mem_cgroup_oom_synchronize() has to be called at
1624 * the end of the page fault to complete the OOM handling.
1625 *
1626 * Returns %true if an ongoing memcg OOM situation was detected and
1627 * completed, %false otherwise.
1628 */
1629bool mem_cgroup_oom_synchronize(bool handle)
1630{
Tejun Heo626ebc42015-11-05 18:46:09 -08001631 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner494264202013-10-16 13:46:59 -07001632 struct oom_wait_info owait;
1633 bool locked;
1634
1635 /* OOM is global, do not handle */
1636 if (!memcg)
1637 return false;
1638
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001639 if (!handle || oom_killer_disabled)
Johannes Weiner494264202013-10-16 13:46:59 -07001640 goto cleanup;
1641
1642 owait.memcg = memcg;
1643 owait.wait.flags = 0;
1644 owait.wait.func = memcg_oom_wake_function;
1645 owait.wait.private = current;
1646 INIT_LIST_HEAD(&owait.wait.task_list);
1647
1648 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001649 mem_cgroup_mark_under_oom(memcg);
1650
1651 locked = mem_cgroup_oom_trylock(memcg);
1652
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001653 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001654 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001655
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001656 if (locked && !memcg->oom_kill_disable) {
1657 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner494264202013-10-16 13:46:59 -07001658 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001659 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1660 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001661 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001662 schedule();
Johannes Weiner494264202013-10-16 13:46:59 -07001663 mem_cgroup_unmark_under_oom(memcg);
1664 finish_wait(&memcg_oom_waitq, &owait.wait);
1665 }
1666
1667 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001668 mem_cgroup_oom_unlock(memcg);
1669 /*
1670 * There is no guarantee that an OOM-lock contender
1671 * sees the wakeups triggered by the OOM kill
1672 * uncharges. Wake any sleepers explicitely.
1673 */
1674 memcg_oom_recover(memcg);
1675 }
Johannes Weiner494264202013-10-16 13:46:59 -07001676cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001677 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001678 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001679 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001680}
1681
Johannes Weinerd7365e72014-10-29 14:50:48 -07001682/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001683 * lock_page_memcg - lock a page->mem_cgroup binding
1684 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001685 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001686 * This function protects unlocked LRU pages from being moved to
1687 * another cgroup and stabilizes their page->mem_cgroup binding.
Balbir Singhd69b0422009-06-17 16:26:34 -07001688 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001689void lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001690{
1691 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001692 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001693
Johannes Weiner6de22612015-02-11 15:25:01 -08001694 /*
1695 * The RCU lock is held throughout the transaction. The fast
1696 * path can get away without acquiring the memcg->move_lock
1697 * because page moving starts with an RCU grace period.
Johannes Weiner6de22612015-02-11 15:25:01 -08001698 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001699 rcu_read_lock();
1700
1701 if (mem_cgroup_disabled())
Johannes Weiner62cccb82016-03-15 14:57:22 -07001702 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001703again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001704 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001705 if (unlikely(!memcg))
Johannes Weiner62cccb82016-03-15 14:57:22 -07001706 return;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001707
Qiang Huangbdcbb652014-06-04 16:08:21 -07001708 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001709 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001710
Johannes Weiner6de22612015-02-11 15:25:01 -08001711 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001712 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001713 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001714 goto again;
1715 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001716
1717 /*
1718 * When charge migration first begins, we can have locked and
1719 * unlocked page stat updates happening concurrently. Track
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001720 * the task who has the lock for unlock_page_memcg().
Johannes Weiner6de22612015-02-11 15:25:01 -08001721 */
1722 memcg->move_lock_task = current;
1723 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001724
Johannes Weiner62cccb82016-03-15 14:57:22 -07001725 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001726}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001727EXPORT_SYMBOL(lock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001728
Johannes Weinerd7365e72014-10-29 14:50:48 -07001729/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001730 * unlock_page_memcg - unlock a page->mem_cgroup binding
Johannes Weiner62cccb82016-03-15 14:57:22 -07001731 * @page: the page
Johannes Weinerd7365e72014-10-29 14:50:48 -07001732 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001733void unlock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001734{
Johannes Weiner62cccb82016-03-15 14:57:22 -07001735 struct mem_cgroup *memcg = page->mem_cgroup;
1736
Johannes Weiner6de22612015-02-11 15:25:01 -08001737 if (memcg && memcg->move_lock_task == current) {
1738 unsigned long flags = memcg->move_lock_flags;
1739
1740 memcg->move_lock_task = NULL;
1741 memcg->move_lock_flags = 0;
1742
1743 spin_unlock_irqrestore(&memcg->move_lock, flags);
1744 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001745
Johannes Weinerd7365e72014-10-29 14:50:48 -07001746 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001747}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001748EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001749
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001750/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001751 * size of first charge trial. "32" comes from vmscan.c's magic value.
1752 * TODO: maybe necessary to use big numbers in big irons.
1753 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001754#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001755struct memcg_stock_pcp {
1756 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001757 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001758 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001759 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001760#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001761};
1762static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001763static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001764
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001765/**
1766 * consume_stock: Try to consume stocked charge on this cpu.
1767 * @memcg: memcg to consume from.
1768 * @nr_pages: how many pages to charge.
1769 *
1770 * The charges will only happen if @memcg matches the current cpu's memcg
1771 * stock, and at least @nr_pages are available in that stock. Failure to
1772 * service an allocation will refill the stock.
1773 *
1774 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001775 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001776static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001777{
1778 struct memcg_stock_pcp *stock;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001779 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001780
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001781 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001782 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001783
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001784 stock = &get_cpu_var(memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001785 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001786 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001787 ret = true;
1788 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001789 put_cpu_var(memcg_stock);
1790 return ret;
1791}
1792
1793/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001794 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001795 */
1796static void drain_stock(struct memcg_stock_pcp *stock)
1797{
1798 struct mem_cgroup *old = stock->cached;
1799
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001800 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001801 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001802 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001803 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001804 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001805 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001806 }
1807 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001808}
1809
1810/*
1811 * This must be called under preempt disabled or must be called by
1812 * a thread which is pinned to local cpu.
1813 */
1814static void drain_local_stock(struct work_struct *dummy)
1815{
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001816 struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001817 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001818 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001819}
1820
1821/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001822 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001823 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001824 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001825static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001826{
1827 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1828
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001829 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001830 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001831 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001832 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001833 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001834 put_cpu_var(memcg_stock);
1835}
1836
1837/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001838 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001839 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001840 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001841static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001842{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001843 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001844
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001845 /* If someone's already draining, avoid adding running more workers. */
1846 if (!mutex_trylock(&percpu_charge_mutex))
1847 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001848 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001849 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001850 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001851 for_each_online_cpu(cpu) {
1852 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001853 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001854
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001855 memcg = stock->cached;
1856 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001857 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001858 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001859 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001860 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1861 if (cpu == curcpu)
1862 drain_local_stock(&stock->work);
1863 else
1864 schedule_work_on(cpu, &stock->work);
1865 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001866 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001867 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001868 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001869 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001870}
1871
Paul Gortmaker0db06282013-06-19 14:53:51 -04001872static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001873 unsigned long action,
1874 void *hcpu)
1875{
1876 int cpu = (unsigned long)hcpu;
1877 struct memcg_stock_pcp *stock;
1878
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001879 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001880 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001881
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07001882 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001883 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001884
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001885 stock = &per_cpu(memcg_stock, cpu);
1886 drain_stock(stock);
1887 return NOTIFY_OK;
1888}
1889
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001890static void reclaim_high(struct mem_cgroup *memcg,
1891 unsigned int nr_pages,
1892 gfp_t gfp_mask)
1893{
1894 do {
1895 if (page_counter_read(&memcg->memory) <= memcg->high)
1896 continue;
1897 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
1898 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1899 } while ((memcg = parent_mem_cgroup(memcg)));
1900}
1901
1902static void high_work_func(struct work_struct *work)
1903{
1904 struct mem_cgroup *memcg;
1905
1906 memcg = container_of(work, struct mem_cgroup, high_work);
1907 reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1908}
1909
Tejun Heob23afb92015-11-05 18:46:11 -08001910/*
1911 * Scheduled by try_charge() to be executed from the userland return path
1912 * and reclaims memory over the high limit.
1913 */
1914void mem_cgroup_handle_over_high(void)
1915{
1916 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001917 struct mem_cgroup *memcg;
Tejun Heob23afb92015-11-05 18:46:11 -08001918
1919 if (likely(!nr_pages))
1920 return;
1921
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001922 memcg = get_mem_cgroup_from_mm(current->mm);
1923 reclaim_high(memcg, nr_pages, GFP_KERNEL);
Tejun Heob23afb92015-11-05 18:46:11 -08001924 css_put(&memcg->css);
1925 current->memcg_nr_pages_over_high = 0;
1926}
1927
Johannes Weiner00501b52014-08-08 14:19:20 -07001928static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1929 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001930{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001931 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001932 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001933 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001934 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001935 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001936 bool may_swap = true;
1937 bool drained = false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001938
Johannes Weinerce00a962014-09-05 08:43:57 -04001939 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08001940 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001941retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001942 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08001943 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001944
Johannes Weiner7941d212016-01-14 15:21:23 -08001945 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08001946 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1947 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001948 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08001949 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001950 page_counter_uncharge(&memcg->memsw, batch);
1951 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001952 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001953 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001954 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001955 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001956
Johannes Weiner6539cc02014-08-06 16:05:42 -07001957 if (batch > nr_pages) {
1958 batch = nr_pages;
1959 goto retry;
1960 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001961
Johannes Weiner06b078fc02014-08-06 16:05:44 -07001962 /*
1963 * Unlike in global OOM situations, memcg is not in a physical
1964 * memory shortage. Allow dying and OOM-killed tasks to
1965 * bypass the last charges so that they can exit quickly and
1966 * free their memory.
1967 */
1968 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1969 fatal_signal_pending(current) ||
1970 current->flags & PF_EXITING))
Tejun Heo10d53c72015-11-05 18:46:17 -08001971 goto force;
Johannes Weiner06b078fc02014-08-06 16:05:44 -07001972
1973 if (unlikely(task_in_memcg_oom(current)))
1974 goto nomem;
1975
Mel Gormand0164ad2015-11-06 16:28:21 -08001976 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001977 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001978
Johannes Weiner241994ed2015-02-11 15:26:06 -08001979 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
1980
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001981 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1982 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07001983
Johannes Weiner61e02c72014-08-06 16:08:16 -07001984 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07001985 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07001986
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001987 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001988 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001989 drained = true;
1990 goto retry;
1991 }
1992
Johannes Weiner28c34c22014-08-06 16:05:47 -07001993 if (gfp_mask & __GFP_NORETRY)
1994 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001995 /*
1996 * Even though the limit is exceeded at this point, reclaim
1997 * may have been able to free some pages. Retry the charge
1998 * before killing the task.
1999 *
2000 * Only for regular pages, though: huge pages are rather
2001 * unlikely to succeed so close to the limit, and we fall back
2002 * to regular pages anyway in case of failure.
2003 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002004 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002005 goto retry;
2006 /*
2007 * At task move, charge accounts can be doubly counted. So, it's
2008 * better to wait until the end of task_move if something is going on.
2009 */
2010 if (mem_cgroup_wait_acct_move(mem_over_limit))
2011 goto retry;
2012
Johannes Weiner9b130612014-08-06 16:05:51 -07002013 if (nr_retries--)
2014 goto retry;
2015
Johannes Weiner06b078fc02014-08-06 16:05:44 -07002016 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08002017 goto force;
Johannes Weiner06b078fc02014-08-06 16:05:44 -07002018
Johannes Weiner6539cc02014-08-06 16:05:42 -07002019 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08002020 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002021
Johannes Weiner241994ed2015-02-11 15:26:06 -08002022 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
2023
Jerome Marchand3608de02015-11-05 18:47:29 -08002024 mem_cgroup_oom(mem_over_limit, gfp_mask,
2025 get_order(nr_pages * PAGE_SIZE));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002026nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002027 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002028 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08002029force:
2030 /*
2031 * The allocation either can't fail or will lead to more memory
2032 * being freed very soon. Allow memory usage go over the limit
2033 * temporarily by force charging it.
2034 */
2035 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002036 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08002037 page_counter_charge(&memcg->memsw, nr_pages);
2038 css_get_many(&memcg->css, nr_pages);
2039
2040 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002041
2042done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002043 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002044 if (batch > nr_pages)
2045 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002046
Johannes Weiner241994ed2015-02-11 15:26:06 -08002047 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002048 * If the hierarchy is above the normal consumption range, schedule
2049 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002050 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002051 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2052 * not recorded as it most likely matches current's and won't
2053 * change in the meantime. As high limit is checked again before
2054 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994ed2015-02-11 15:26:06 -08002055 */
2056 do {
Tejun Heob23afb92015-11-05 18:46:11 -08002057 if (page_counter_read(&memcg->memory) > memcg->high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002058 /* Don't bother a random interrupted task */
2059 if (in_interrupt()) {
2060 schedule_work(&memcg->high_work);
2061 break;
2062 }
Vladimir Davydov9516a182015-12-11 13:40:24 -08002063 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002064 set_notify_resume(current);
2065 break;
2066 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08002067 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002068
2069 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002070}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002071
Johannes Weiner00501b52014-08-08 14:19:20 -07002072static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002073{
Johannes Weinerce00a962014-09-05 08:43:57 -04002074 if (mem_cgroup_is_root(memcg))
2075 return;
2076
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002077 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002078 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002079 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002080
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002081 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002082}
2083
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002084static void lock_page_lru(struct page *page, int *isolated)
2085{
2086 struct zone *zone = page_zone(page);
2087
2088 spin_lock_irq(&zone->lru_lock);
2089 if (PageLRU(page)) {
2090 struct lruvec *lruvec;
2091
2092 lruvec = mem_cgroup_page_lruvec(page, zone);
2093 ClearPageLRU(page);
2094 del_page_from_lru_list(page, lruvec, page_lru(page));
2095 *isolated = 1;
2096 } else
2097 *isolated = 0;
2098}
2099
2100static void unlock_page_lru(struct page *page, int isolated)
2101{
2102 struct zone *zone = page_zone(page);
2103
2104 if (isolated) {
2105 struct lruvec *lruvec;
2106
2107 lruvec = mem_cgroup_page_lruvec(page, zone);
2108 VM_BUG_ON_PAGE(PageLRU(page), page);
2109 SetPageLRU(page);
2110 add_page_to_lru_list(page, lruvec, page_lru(page));
2111 }
2112 spin_unlock_irq(&zone->lru_lock);
2113}
2114
Johannes Weiner00501b52014-08-08 14:19:20 -07002115static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002116 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002117{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002118 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002119
Johannes Weiner1306a852014-12-10 15:44:52 -08002120 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002121
2122 /*
2123 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2124 * may already be on some other mem_cgroup's LRU. Take care of it.
2125 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002126 if (lrucare)
2127 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002128
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002129 /*
2130 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002131 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002132 *
2133 * - the page is uncharged
2134 *
2135 * - the page is off-LRU
2136 *
2137 * - an anonymous fault has exclusive page access, except for
2138 * a locked page table
2139 *
2140 * - a page cache insertion, a swapin fault, or a migration
2141 * have the page locked
2142 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002143 page->mem_cgroup = memcg;
Hugh Dickins3be912772008-02-07 00:14:19 -08002144
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002145 if (lrucare)
2146 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002147}
2148
Johannes Weiner127424c2016-01-20 15:02:32 -08002149#ifndef CONFIG_SLOB
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002150static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002151{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002152 int id, size;
2153 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002154
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002155 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002156 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2157 if (id < 0)
2158 return id;
2159
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002160 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002161 return id;
2162
2163 /*
2164 * There's no space for the new id in memcg_caches arrays,
2165 * so we have to grow them.
2166 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002167 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002168
2169 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002170 if (size < MEMCG_CACHES_MIN_SIZE)
2171 size = MEMCG_CACHES_MIN_SIZE;
2172 else if (size > MEMCG_CACHES_MAX_SIZE)
2173 size = MEMCG_CACHES_MAX_SIZE;
2174
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002175 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002176 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002177 err = memcg_update_all_list_lrus(size);
2178 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002179 memcg_nr_cache_ids = size;
2180
2181 up_write(&memcg_cache_ids_sem);
2182
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002183 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002184 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002185 return err;
2186 }
2187 return id;
2188}
2189
2190static void memcg_free_cache_id(int id)
2191{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002192 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002193}
2194
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002195struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002196 struct mem_cgroup *memcg;
2197 struct kmem_cache *cachep;
2198 struct work_struct work;
2199};
2200
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002201static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002202{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002203 struct memcg_kmem_cache_create_work *cw =
2204 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002205 struct mem_cgroup *memcg = cw->memcg;
2206 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002207
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002208 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002209
Vladimir Davydov5722d092014-04-07 15:39:24 -07002210 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002211 kfree(cw);
2212}
2213
2214/*
2215 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002216 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002217static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2218 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002219{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002220 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002221
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002222 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002223 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002224 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002225
2226 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002227
2228 cw->memcg = memcg;
2229 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002230 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002231
Glauber Costad7f25f82012-12-18 14:22:40 -08002232 schedule_work(&cw->work);
2233}
2234
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002235static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2236 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002237{
2238 /*
2239 * We need to stop accounting when we kmalloc, because if the
2240 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002241 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002242 *
2243 * However, it is better to enclose the whole function. Depending on
2244 * the debugging options enabled, INIT_WORK(), for instance, can
2245 * trigger an allocation. This too, will make us recurse. Because at
2246 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2247 * the safest choice is to do it like this, wrapping the whole function.
2248 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002249 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002250 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002251 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002252}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002253
Glauber Costad7f25f82012-12-18 14:22:40 -08002254/*
2255 * Return the kmem_cache we're supposed to use for a slab allocation.
2256 * We try to use the current memcg's version of the cache.
2257 *
2258 * If the cache does not exist yet, if we are the first user of it,
2259 * we either create it immediately, if possible, or create it asynchronously
2260 * in a workqueue.
2261 * In the latter case, we will let the current allocation go through with
2262 * the original cache.
2263 *
2264 * Can't be called in interrupt context or from kernel threads.
2265 * This function needs to be called with rcu_read_lock() held.
2266 */
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002267struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
Glauber Costad7f25f82012-12-18 14:22:40 -08002268{
2269 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002270 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002271 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002272
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002273 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002274
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002275 if (cachep->flags & SLAB_ACCOUNT)
2276 gfp |= __GFP_ACCOUNT;
2277
2278 if (!(gfp & __GFP_ACCOUNT))
2279 return cachep;
2280
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002281 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002282 return cachep;
2283
Vladimir Davydov8135be52014-12-12 16:56:38 -08002284 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002285 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002286 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002287 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002288
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002289 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002290 if (likely(memcg_cachep))
2291 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002292
2293 /*
2294 * If we are in a safe context (can wait, and not in interrupt
2295 * context), we could be be predictable and return right away.
2296 * This would guarantee that the allocation being performed
2297 * already belongs in the new cache.
2298 *
2299 * However, there are some clashes that can arrive from locking.
2300 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002301 * memcg_create_kmem_cache, this means no further allocation
2302 * could happen with the slab_mutex held. So it's better to
2303 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002304 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002305 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002306out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002307 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002308 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002309}
Glauber Costad7f25f82012-12-18 14:22:40 -08002310
Vladimir Davydov8135be52014-12-12 16:56:38 -08002311void __memcg_kmem_put_cache(struct kmem_cache *cachep)
2312{
2313 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002314 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002315}
2316
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002317int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2318 struct mem_cgroup *memcg)
2319{
2320 unsigned int nr_pages = 1 << order;
2321 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002322 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002323
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002324 ret = try_charge(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08002325 if (ret)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002326 return ret;
Johannes Weiner52c29b02016-01-20 15:02:35 -08002327
2328 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2329 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2330 cancel_charge(memcg, nr_pages);
2331 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002332 }
2333
2334 page->mem_cgroup = memcg;
2335
2336 return 0;
2337}
2338
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002339int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002340{
2341 struct mem_cgroup *memcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002342 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002343
Johannes Weinerdf381972014-04-07 15:37:43 -07002344 memcg = get_mem_cgroup_from_mm(current->mm);
Vladimir Davydovb6ecd2d2016-03-17 14:18:33 -07002345 if (!mem_cgroup_is_root(memcg))
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002346 ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg);
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002347 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002348 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002349}
2350
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002351void __memcg_kmem_uncharge(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002352{
Johannes Weiner1306a852014-12-10 15:44:52 -08002353 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002354 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002355
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002356 if (!memcg)
2357 return;
2358
Sasha Levin309381fea2014-01-23 15:52:54 -08002359 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002360
Johannes Weiner52c29b02016-01-20 15:02:35 -08002361 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2362 page_counter_uncharge(&memcg->kmem, nr_pages);
2363
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002364 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002365 if (do_memsw_account())
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002366 page_counter_uncharge(&memcg->memsw, nr_pages);
2367
Johannes Weiner1306a852014-12-10 15:44:52 -08002368 page->mem_cgroup = NULL;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002369 css_put_many(&memcg->css, nr_pages);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002370}
Johannes Weiner127424c2016-01-20 15:02:32 -08002371#endif /* !CONFIG_SLOB */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002372
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002373#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2374
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002375/*
2376 * Because tail pages are not marked as "used", set it. We're under
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08002377 * zone->lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002378 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002379void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002380{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002381 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002382
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002383 if (mem_cgroup_disabled())
2384 return;
David Rientjesb070e652013-05-07 16:18:09 -07002385
Johannes Weiner29833312014-12-10 15:44:02 -08002386 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002387 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002388
Johannes Weiner1306a852014-12-10 15:44:52 -08002389 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002390 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002391}
Hugh Dickins12d27102012-01-12 17:19:52 -08002392#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002393
Andrew Mortonc255a452012-07-31 16:43:02 -07002394#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002395static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2396 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002397{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002398 int val = (charge) ? 1 : -1;
2399 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002400}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002401
2402/**
2403 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2404 * @entry: swap entry to be moved
2405 * @from: mem_cgroup which the entry is moved from
2406 * @to: mem_cgroup which the entry is moved to
2407 *
2408 * It succeeds only when the swap_cgroup's record for this entry is the same
2409 * as the mem_cgroup's id of @from.
2410 *
2411 * Returns 0 on success, -EINVAL on failure.
2412 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002413 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002414 * both res and memsw, and called css_get().
2415 */
2416static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002417 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002418{
2419 unsigned short old_id, new_id;
2420
Li Zefan34c00c32013-09-23 16:56:01 +08002421 old_id = mem_cgroup_id(from);
2422 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002423
2424 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002425 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002426 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002427 return 0;
2428 }
2429 return -EINVAL;
2430}
2431#else
2432static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002433 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002434{
2435 return -EINVAL;
2436}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002437#endif
2438
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002439static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002440
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002441static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002442 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002443{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002444 unsigned long curusage;
2445 unsigned long oldusage;
2446 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002447 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002448 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002449
2450 /*
2451 * For keeping hierarchical_reclaim simple, how long we should retry
2452 * is depends on callers. We set our retry-count to be function
2453 * of # of children which we should visit in this loop.
2454 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002455 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2456 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002457
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002458 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002459
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002460 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002461 if (signal_pending(current)) {
2462 ret = -EINTR;
2463 break;
2464 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002465
2466 mutex_lock(&memcg_limit_mutex);
2467 if (limit > memcg->memsw.limit) {
2468 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002469 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002470 break;
2471 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002472 if (limit > memcg->memory.limit)
2473 enlarge = true;
2474 ret = page_counter_limit(&memcg->memory, limit);
2475 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002476
2477 if (!ret)
2478 break;
2479
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002480 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2481
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002482 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002483 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002484 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002485 retry_count--;
2486 else
2487 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002488 } while (retry_count);
2489
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002490 if (!ret && enlarge)
2491 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002492
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002493 return ret;
2494}
2495
Li Zefan338c8432009-06-17 16:27:15 -07002496static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002497 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002498{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002499 unsigned long curusage;
2500 unsigned long oldusage;
2501 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002502 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002503 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002504
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002505 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002506 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2507 mem_cgroup_count_children(memcg);
2508
2509 oldusage = page_counter_read(&memcg->memsw);
2510
2511 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002512 if (signal_pending(current)) {
2513 ret = -EINTR;
2514 break;
2515 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002516
2517 mutex_lock(&memcg_limit_mutex);
2518 if (limit < memcg->memory.limit) {
2519 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002520 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002521 break;
2522 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002523 if (limit > memcg->memsw.limit)
2524 enlarge = true;
2525 ret = page_counter_limit(&memcg->memsw, limit);
2526 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002527
2528 if (!ret)
2529 break;
2530
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002531 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2532
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002533 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002534 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002535 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002536 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002537 else
2538 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002539 } while (retry_count);
2540
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002541 if (!ret && enlarge)
2542 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002543
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002544 return ret;
2545}
2546
Andrew Morton0608f432013-09-24 15:27:41 -07002547unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2548 gfp_t gfp_mask,
2549 unsigned long *total_scanned)
2550{
2551 unsigned long nr_reclaimed = 0;
2552 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2553 unsigned long reclaimed;
2554 int loop = 0;
2555 struct mem_cgroup_tree_per_zone *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002556 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002557 unsigned long nr_scanned;
2558
2559 if (order > 0)
2560 return 0;
2561
2562 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
2563 /*
2564 * This loop can run a while, specially if mem_cgroup's continuously
2565 * keep exceeding their soft limit and putting the system under
2566 * pressure
2567 */
2568 do {
2569 if (next_mz)
2570 mz = next_mz;
2571 else
2572 mz = mem_cgroup_largest_soft_limit_node(mctz);
2573 if (!mz)
2574 break;
2575
2576 nr_scanned = 0;
2577 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
2578 gfp_mask, &nr_scanned);
2579 nr_reclaimed += reclaimed;
2580 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002581 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002582 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002583
2584 /*
2585 * If we failed to reclaim anything from this memory cgroup
2586 * it is time to move on to the next cgroup
2587 */
2588 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002589 if (!reclaimed)
2590 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2591
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002592 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002593 /*
2594 * One school of thought says that we should not add
2595 * back the node to the tree if reclaim returns 0.
2596 * But our reclaim could return 0, simply because due
2597 * to priority we are exposing a smaller subset of
2598 * memory to reclaim from. Consider this as a longer
2599 * term TODO.
2600 */
2601 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002602 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002603 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002604 css_put(&mz->memcg->css);
2605 loop++;
2606 /*
2607 * Could not reclaim anything and there are no more
2608 * mem cgroups to try or we seem to be looping without
2609 * reclaiming anything.
2610 */
2611 if (!nr_reclaimed &&
2612 (next_mz == NULL ||
2613 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2614 break;
2615 } while (!nr_reclaimed);
2616 if (next_mz)
2617 css_put(&next_mz->memcg->css);
2618 return nr_reclaimed;
2619}
2620
Tejun Heoea280e72014-05-16 13:22:48 -04002621/*
2622 * Test whether @memcg has children, dead or alive. Note that this
2623 * function doesn't care whether @memcg has use_hierarchy enabled and
2624 * returns %true if there are child csses according to the cgroup
2625 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2626 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002627static inline bool memcg_has_children(struct mem_cgroup *memcg)
2628{
Tejun Heoea280e72014-05-16 13:22:48 -04002629 bool ret;
2630
Tejun Heoea280e72014-05-16 13:22:48 -04002631 rcu_read_lock();
2632 ret = css_next_child(NULL, &memcg->css);
2633 rcu_read_unlock();
2634 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002635}
2636
2637/*
Michal Hockoc26251f2012-10-26 13:37:28 +02002638 * Reclaims as many pages from the given memcg as possible and moves
2639 * the rest to the parent.
2640 *
2641 * Caller is responsible for holding css reference for memcg.
2642 */
2643static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2644{
2645 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002646
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002647 /* we call try-to-free pages for make this cgroup empty */
2648 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002649 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002650 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002651 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002652
Michal Hockoc26251f2012-10-26 13:37:28 +02002653 if (signal_pending(current))
2654 return -EINTR;
2655
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002656 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2657 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002658 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002659 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002660 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002661 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002662 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002663
2664 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002665
2666 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002667}
2668
Tejun Heo6770c642014-05-13 12:16:21 -04002669static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2670 char *buf, size_t nbytes,
2671 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002672{
Tejun Heo6770c642014-05-13 12:16:21 -04002673 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002674
Michal Hockod8423012012-10-26 13:37:29 +02002675 if (mem_cgroup_is_root(memcg))
2676 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002677 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002678}
2679
Tejun Heo182446d2013-08-08 20:11:24 -04002680static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2681 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002682{
Tejun Heo182446d2013-08-08 20:11:24 -04002683 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002684}
2685
Tejun Heo182446d2013-08-08 20:11:24 -04002686static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2687 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002688{
2689 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002690 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002691 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002692
Glauber Costa567fb432012-07-31 16:43:07 -07002693 if (memcg->use_hierarchy == val)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002694 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07002695
Balbir Singh18f59ea2009-01-07 18:08:07 -08002696 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002697 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002698 * in the child subtrees. If it is unset, then the change can
2699 * occur, provided the current cgroup has no children.
2700 *
2701 * For the root cgroup, parent_mem is NULL, we allow value to be
2702 * set if there are no children.
2703 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002704 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002705 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002706 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002707 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002708 else
2709 retval = -EBUSY;
2710 } else
2711 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002712
Balbir Singh18f59ea2009-01-07 18:08:07 -08002713 return retval;
2714}
2715
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002716static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
Johannes Weinerce00a962014-09-05 08:43:57 -04002717{
2718 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002719 int i;
Johannes Weinerce00a962014-09-05 08:43:57 -04002720
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002721 memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
Johannes Weinerce00a962014-09-05 08:43:57 -04002722
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002723 for_each_mem_cgroup_tree(iter, memcg) {
2724 for (i = 0; i < MEMCG_NR_STAT; i++)
2725 stat[i] += mem_cgroup_read_stat(iter, i);
2726 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002727}
2728
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002729static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
Johannes Weiner587d9f72016-01-20 15:03:19 -08002730{
2731 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002732 int i;
Johannes Weiner587d9f72016-01-20 15:03:19 -08002733
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002734 memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
Johannes Weiner587d9f72016-01-20 15:03:19 -08002735
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002736 for_each_mem_cgroup_tree(iter, memcg) {
2737 for (i = 0; i < MEMCG_NR_EVENTS; i++)
2738 events[i] += mem_cgroup_read_events(iter, i);
2739 }
Johannes Weiner587d9f72016-01-20 15:03:19 -08002740}
2741
Andrew Morton6f646152015-11-06 16:28:58 -08002742static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04002743{
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002744 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002745
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002746 if (mem_cgroup_is_root(memcg)) {
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002747 struct mem_cgroup *iter;
2748
2749 for_each_mem_cgroup_tree(iter, memcg) {
2750 val += mem_cgroup_read_stat(iter,
2751 MEM_CGROUP_STAT_CACHE);
2752 val += mem_cgroup_read_stat(iter,
2753 MEM_CGROUP_STAT_RSS);
2754 if (swap)
2755 val += mem_cgroup_read_stat(iter,
2756 MEM_CGROUP_STAT_SWAP);
2757 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002758 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002759 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002760 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002761 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002762 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002763 }
Michal Hockoc12176d2015-11-05 18:50:29 -08002764 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002765}
2766
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002767enum {
2768 RES_USAGE,
2769 RES_LIMIT,
2770 RES_MAX_USAGE,
2771 RES_FAILCNT,
2772 RES_SOFT_LIMIT,
2773};
Johannes Weinerce00a962014-09-05 08:43:57 -04002774
Tejun Heo791badb2013-12-05 12:28:02 -05002775static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002776 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002777{
Tejun Heo182446d2013-08-08 20:11:24 -04002778 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002779 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002780
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002781 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002782 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002783 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002784 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002785 case _MEMSWAP:
2786 counter = &memcg->memsw;
2787 break;
2788 case _KMEM:
2789 counter = &memcg->kmem;
2790 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002791 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08002792 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002793 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002794 default:
2795 BUG();
2796 }
2797
2798 switch (MEMFILE_ATTR(cft->private)) {
2799 case RES_USAGE:
2800 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08002801 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002802 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08002803 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002804 return (u64)page_counter_read(counter) * PAGE_SIZE;
2805 case RES_LIMIT:
2806 return (u64)counter->limit * PAGE_SIZE;
2807 case RES_MAX_USAGE:
2808 return (u64)counter->watermark * PAGE_SIZE;
2809 case RES_FAILCNT:
2810 return counter->failcnt;
2811 case RES_SOFT_LIMIT:
2812 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002813 default:
2814 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002815 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002816}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002817
Johannes Weiner127424c2016-01-20 15:02:32 -08002818#ifndef CONFIG_SLOB
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002819static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002820{
Vladimir Davydovd6441632014-01-23 15:53:09 -08002821 int memcg_id;
2822
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002823 if (cgroup_memory_nokmem)
2824 return 0;
2825
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002826 BUG_ON(memcg->kmemcg_id >= 0);
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002827 BUG_ON(memcg->kmem_state);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002828
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002829 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002830 if (memcg_id < 0)
2831 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002832
Johannes Weineref129472016-01-14 15:21:34 -08002833 static_branch_inc(&memcg_kmem_enabled_key);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002834 /*
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002835 * A memory cgroup is considered kmem-online as soon as it gets
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002836 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002837 * guarantee no one starts accounting before all call sites are
2838 * patched.
2839 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002840 memcg->kmemcg_id = memcg_id;
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002841 memcg->kmem_state = KMEM_ONLINE;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002842
2843 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002844}
2845
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002846static void memcg_offline_kmem(struct mem_cgroup *memcg)
2847{
2848 struct cgroup_subsys_state *css;
2849 struct mem_cgroup *parent, *child;
2850 int kmemcg_id;
2851
2852 if (memcg->kmem_state != KMEM_ONLINE)
2853 return;
2854 /*
2855 * Clear the online state before clearing memcg_caches array
2856 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2857 * guarantees that no cache will be created for this cgroup
2858 * after we are done (see memcg_create_kmem_cache()).
2859 */
2860 memcg->kmem_state = KMEM_ALLOCATED;
2861
2862 memcg_deactivate_kmem_caches(memcg);
2863
2864 kmemcg_id = memcg->kmemcg_id;
2865 BUG_ON(kmemcg_id < 0);
2866
2867 parent = parent_mem_cgroup(memcg);
2868 if (!parent)
2869 parent = root_mem_cgroup;
2870
2871 /*
2872 * Change kmemcg_id of this cgroup and all its descendants to the
2873 * parent's id, and then move all entries from this cgroup's list_lrus
2874 * to ones of the parent. After we have finished, all list_lrus
2875 * corresponding to this cgroup are guaranteed to remain empty. The
2876 * ordering is imposed by list_lru_node->lock taken by
2877 * memcg_drain_all_list_lrus().
2878 */
2879 css_for_each_descendant_pre(css, &memcg->css) {
2880 child = mem_cgroup_from_css(css);
2881 BUG_ON(child->kmemcg_id != kmemcg_id);
2882 child->kmemcg_id = parent->kmemcg_id;
2883 if (!memcg->use_hierarchy)
2884 break;
2885 }
2886 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2887
2888 memcg_free_cache_id(kmemcg_id);
2889}
2890
2891static void memcg_free_kmem(struct mem_cgroup *memcg)
2892{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002893 /* css_alloc() failed, offlining didn't happen */
2894 if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2895 memcg_offline_kmem(memcg);
2896
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002897 if (memcg->kmem_state == KMEM_ALLOCATED) {
2898 memcg_destroy_kmem_caches(memcg);
2899 static_branch_dec(&memcg_kmem_enabled_key);
2900 WARN_ON(page_counter_read(&memcg->kmem));
2901 }
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002902}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002903#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002904static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08002905{
2906 return 0;
2907}
2908static void memcg_offline_kmem(struct mem_cgroup *memcg)
2909{
2910}
2911static void memcg_free_kmem(struct mem_cgroup *memcg)
2912{
2913}
2914#endif /* !CONFIG_SLOB */
2915
Johannes Weiner127424c2016-01-20 15:02:32 -08002916static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2917 unsigned long limit)
2918{
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002919 int ret;
Johannes Weiner127424c2016-01-20 15:02:32 -08002920
2921 mutex_lock(&memcg_limit_mutex);
Johannes Weiner127424c2016-01-20 15:02:32 -08002922 ret = page_counter_limit(&memcg->kmem, limit);
Johannes Weiner127424c2016-01-20 15:02:32 -08002923 mutex_unlock(&memcg_limit_mutex);
2924 return ret;
2925}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002926
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002927static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2928{
2929 int ret;
2930
2931 mutex_lock(&memcg_limit_mutex);
2932
Johannes Weiner0db15292016-01-20 15:02:50 -08002933 ret = page_counter_limit(&memcg->tcpmem, limit);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002934 if (ret)
2935 goto out;
2936
Johannes Weiner0db15292016-01-20 15:02:50 -08002937 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002938 /*
2939 * The active flag needs to be written after the static_key
2940 * update. This is what guarantees that the socket activation
2941 * function is the last one to run. See sock_update_memcg() for
2942 * details, and note that we don't mark any socket as belonging
2943 * to this memcg until that flag is up.
2944 *
2945 * We need to do this, because static_keys will span multiple
2946 * sites, but we can't control their order. If we mark a socket
2947 * as accounted, but the accounting functions are not patched in
2948 * yet, we'll lose accounting.
2949 *
2950 * We never race with the readers in sock_update_memcg(),
2951 * because when this value change, the code to process it is not
2952 * patched in yet.
2953 */
2954 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08002955 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002956 }
2957out:
2958 mutex_unlock(&memcg_limit_mutex);
2959 return ret;
2960}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002961
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002962/*
2963 * The user of this function is...
2964 * RES_LIMIT.
2965 */
Tejun Heo451af5042014-05-13 12:16:21 -04002966static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2967 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002968{
Tejun Heo451af5042014-05-13 12:16:21 -04002969 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002970 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002971 int ret;
2972
Tejun Heo451af5042014-05-13 12:16:21 -04002973 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08002974 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002975 if (ret)
2976 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07002977
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002978 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002979 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07002980 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2981 ret = -EINVAL;
2982 break;
2983 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002984 switch (MEMFILE_TYPE(of_cft(of)->private)) {
2985 case _MEM:
2986 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002987 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002988 case _MEMSWAP:
2989 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
2990 break;
2991 case _KMEM:
2992 ret = memcg_update_kmem_limit(memcg, nr_pages);
2993 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002994 case _TCP:
2995 ret = memcg_update_tcp_limit(memcg, nr_pages);
2996 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002997 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002998 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07002999 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003000 memcg->soft_limit = nr_pages;
3001 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003002 break;
3003 }
Tejun Heo451af5042014-05-13 12:16:21 -04003004 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003005}
3006
Tejun Heo6770c642014-05-13 12:16:21 -04003007static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3008 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003009{
Tejun Heo6770c642014-05-13 12:16:21 -04003010 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003011 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003012
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003013 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3014 case _MEM:
3015 counter = &memcg->memory;
3016 break;
3017 case _MEMSWAP:
3018 counter = &memcg->memsw;
3019 break;
3020 case _KMEM:
3021 counter = &memcg->kmem;
3022 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003023 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003024 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003025 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003026 default:
3027 BUG();
3028 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003029
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003030 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003031 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003032 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003033 break;
3034 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003035 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003036 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003037 default:
3038 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003039 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003040
Tejun Heo6770c642014-05-13 12:16:21 -04003041 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003042}
3043
Tejun Heo182446d2013-08-08 20:11:24 -04003044static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003045 struct cftype *cft)
3046{
Tejun Heo182446d2013-08-08 20:11:24 -04003047 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003048}
3049
Daisuke Nishimura02491442010-03-10 15:22:17 -08003050#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003051static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003052 struct cftype *cft, u64 val)
3053{
Tejun Heo182446d2013-08-08 20:11:24 -04003054 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003055
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003056 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003057 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003058
Glauber Costaee5e8472013-02-22 16:34:50 -08003059 /*
3060 * No kind of locking is needed in here, because ->can_attach() will
3061 * check this value once in the beginning of the process, and then carry
3062 * on with stale data. This means that changes to this value will only
3063 * affect task migrations starting after the change.
3064 */
3065 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003066 return 0;
3067}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003068#else
Tejun Heo182446d2013-08-08 20:11:24 -04003069static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003070 struct cftype *cft, u64 val)
3071{
3072 return -ENOSYS;
3073}
3074#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003075
Ying Han406eb0c2011-05-26 16:25:37 -07003076#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003077static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003078{
Greg Thelen25485de2013-11-12 15:07:40 -08003079 struct numa_stat {
3080 const char *name;
3081 unsigned int lru_mask;
3082 };
3083
3084 static const struct numa_stat stats[] = {
3085 { "total", LRU_ALL },
3086 { "file", LRU_ALL_FILE },
3087 { "anon", LRU_ALL_ANON },
3088 { "unevictable", BIT(LRU_UNEVICTABLE) },
3089 };
3090 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003091 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003092 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003093 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003094
Greg Thelen25485de2013-11-12 15:07:40 -08003095 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3096 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3097 seq_printf(m, "%s=%lu", stat->name, nr);
3098 for_each_node_state(nid, N_MEMORY) {
3099 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3100 stat->lru_mask);
3101 seq_printf(m, " N%d=%lu", nid, nr);
3102 }
3103 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003104 }
Ying Han406eb0c2011-05-26 16:25:37 -07003105
Ying Han071aee12013-11-12 15:07:41 -08003106 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3107 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003108
Ying Han071aee12013-11-12 15:07:41 -08003109 nr = 0;
3110 for_each_mem_cgroup_tree(iter, memcg)
3111 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3112 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3113 for_each_node_state(nid, N_MEMORY) {
3114 nr = 0;
3115 for_each_mem_cgroup_tree(iter, memcg)
3116 nr += mem_cgroup_node_nr_lru_pages(
3117 iter, nid, stat->lru_mask);
3118 seq_printf(m, " N%d=%lu", nid, nr);
3119 }
3120 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003121 }
Ying Han406eb0c2011-05-26 16:25:37 -07003122
Ying Han406eb0c2011-05-26 16:25:37 -07003123 return 0;
3124}
3125#endif /* CONFIG_NUMA */
3126
Tejun Heo2da8ca82013-12-05 12:28:04 -05003127static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003128{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003129 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003130 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003131 struct mem_cgroup *mi;
3132 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003133
Greg Thelen0ca44b12015-02-11 15:25:58 -08003134 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3135 MEM_CGROUP_STAT_NSTATS);
3136 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3137 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003138 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3139
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003140 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Johannes Weiner7941d212016-01-14 15:21:23 -08003141 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003142 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07003143 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003144 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003145 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003146
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003147 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3148 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3149 mem_cgroup_read_events(memcg, i));
3150
3151 for (i = 0; i < NR_LRU_LISTS; i++)
3152 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3153 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3154
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003155 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003156 memory = memsw = PAGE_COUNTER_MAX;
3157 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3158 memory = min(memory, mi->memory.limit);
3159 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003160 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003161 seq_printf(m, "hierarchical_memory_limit %llu\n",
3162 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08003163 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003164 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3165 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003166
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003167 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003168 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003169
Johannes Weiner7941d212016-01-14 15:21:23 -08003170 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003171 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003172 for_each_mem_cgroup_tree(mi, memcg)
3173 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
Greg Thelen484ebb32015-10-01 15:37:05 -07003174 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003175 }
3176
3177 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3178 unsigned long long val = 0;
3179
3180 for_each_mem_cgroup_tree(mi, memcg)
3181 val += mem_cgroup_read_events(mi, i);
3182 seq_printf(m, "total_%s %llu\n",
3183 mem_cgroup_events_names[i], val);
3184 }
3185
3186 for (i = 0; i < NR_LRU_LISTS; i++) {
3187 unsigned long long val = 0;
3188
3189 for_each_mem_cgroup_tree(mi, memcg)
3190 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3191 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003192 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003193
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003194#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003195 {
3196 int nid, zid;
3197 struct mem_cgroup_per_zone *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003198 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003199 unsigned long recent_rotated[2] = {0, 0};
3200 unsigned long recent_scanned[2] = {0, 0};
3201
3202 for_each_online_node(nid)
3203 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
Jianyu Zhane2318752014-06-06 14:38:20 -07003204 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
Hugh Dickins89abfab2012-05-29 15:06:53 -07003205 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003206
Hugh Dickins89abfab2012-05-29 15:06:53 -07003207 recent_rotated[0] += rstat->recent_rotated[0];
3208 recent_rotated[1] += rstat->recent_rotated[1];
3209 recent_scanned[0] += rstat->recent_scanned[0];
3210 recent_scanned[1] += rstat->recent_scanned[1];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003211 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003212 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3213 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3214 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3215 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003216 }
3217#endif
3218
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003219 return 0;
3220}
3221
Tejun Heo182446d2013-08-08 20:11:24 -04003222static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3223 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003224{
Tejun Heo182446d2013-08-08 20:11:24 -04003225 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003226
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003227 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003228}
3229
Tejun Heo182446d2013-08-08 20:11:24 -04003230static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3231 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003232{
Tejun Heo182446d2013-08-08 20:11:24 -04003233 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003234
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003235 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003236 return -EINVAL;
3237
Linus Torvalds14208b02014-06-09 15:03:33 -07003238 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003239 memcg->swappiness = val;
3240 else
3241 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003242
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003243 return 0;
3244}
3245
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003246static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3247{
3248 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003249 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003250 int i;
3251
3252 rcu_read_lock();
3253 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003254 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003255 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003256 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003257
3258 if (!t)
3259 goto unlock;
3260
Johannes Weinerce00a962014-09-05 08:43:57 -04003261 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003262
3263 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003264 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003265 * If it's not true, a threshold was crossed after last
3266 * call of __mem_cgroup_threshold().
3267 */
Phil Carmody5407a562010-05-26 14:42:42 -07003268 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003269
3270 /*
3271 * Iterate backward over array of thresholds starting from
3272 * current_threshold and check if a threshold is crossed.
3273 * If none of thresholds below usage is crossed, we read
3274 * only one element of the array here.
3275 */
3276 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3277 eventfd_signal(t->entries[i].eventfd, 1);
3278
3279 /* i = current_threshold + 1 */
3280 i++;
3281
3282 /*
3283 * Iterate forward over array of thresholds starting from
3284 * current_threshold+1 and check if a threshold is crossed.
3285 * If none of thresholds above usage is crossed, we read
3286 * only one element of the array here.
3287 */
3288 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3289 eventfd_signal(t->entries[i].eventfd, 1);
3290
3291 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003292 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003293unlock:
3294 rcu_read_unlock();
3295}
3296
3297static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3298{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003299 while (memcg) {
3300 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08003301 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003302 __mem_cgroup_threshold(memcg, true);
3303
3304 memcg = parent_mem_cgroup(memcg);
3305 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003306}
3307
3308static int compare_thresholds(const void *a, const void *b)
3309{
3310 const struct mem_cgroup_threshold *_a = a;
3311 const struct mem_cgroup_threshold *_b = b;
3312
Greg Thelen2bff24a2013-09-11 14:23:08 -07003313 if (_a->threshold > _b->threshold)
3314 return 1;
3315
3316 if (_a->threshold < _b->threshold)
3317 return -1;
3318
3319 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003320}
3321
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003322static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003323{
3324 struct mem_cgroup_eventfd_list *ev;
3325
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003326 spin_lock(&memcg_oom_lock);
3327
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003328 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003329 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003330
3331 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003332 return 0;
3333}
3334
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003335static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003336{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003337 struct mem_cgroup *iter;
3338
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003339 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003340 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003341}
3342
Tejun Heo59b6f872013-11-22 18:20:43 -05003343static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003344 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003345{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003346 struct mem_cgroup_thresholds *thresholds;
3347 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003348 unsigned long threshold;
3349 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003350 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003351
Johannes Weiner650c5e52015-02-11 15:26:03 -08003352 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003353 if (ret)
3354 return ret;
3355
3356 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003357
Johannes Weiner05b84302014-08-06 16:05:59 -07003358 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003359 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003360 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003361 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003362 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003363 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003364 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003365 BUG();
3366
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003367 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003368 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003369 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3370
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003371 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003372
3373 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003374 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003375 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003376 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003377 ret = -ENOMEM;
3378 goto unlock;
3379 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003380 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003381
3382 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003383 if (thresholds->primary) {
3384 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003385 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003386 }
3387
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003388 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003389 new->entries[size - 1].eventfd = eventfd;
3390 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003391
3392 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003393 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003394 compare_thresholds, NULL);
3395
3396 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003397 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003398 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003399 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003400 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003401 * new->current_threshold will not be used until
3402 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003403 * it here.
3404 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003405 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003406 } else
3407 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003408 }
3409
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003410 /* Free old spare buffer and save old primary buffer as spare */
3411 kfree(thresholds->spare);
3412 thresholds->spare = thresholds->primary;
3413
3414 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003415
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003416 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003417 synchronize_rcu();
3418
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003419unlock:
3420 mutex_unlock(&memcg->thresholds_lock);
3421
3422 return ret;
3423}
3424
Tejun Heo59b6f872013-11-22 18:20:43 -05003425static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003426 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003427{
Tejun Heo59b6f872013-11-22 18:20:43 -05003428 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003429}
3430
Tejun Heo59b6f872013-11-22 18:20:43 -05003431static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003432 struct eventfd_ctx *eventfd, const char *args)
3433{
Tejun Heo59b6f872013-11-22 18:20:43 -05003434 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003435}
3436
Tejun Heo59b6f872013-11-22 18:20:43 -05003437static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003438 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003439{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003440 struct mem_cgroup_thresholds *thresholds;
3441 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003442 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003443 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003444
3445 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003446
3447 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003448 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003449 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003450 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003451 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003452 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003453 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003454 BUG();
3455
Anton Vorontsov371528c2012-02-24 05:14:46 +04003456 if (!thresholds->primary)
3457 goto unlock;
3458
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003459 /* Check if a threshold crossed before removing */
3460 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3461
3462 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003463 size = 0;
3464 for (i = 0; i < thresholds->primary->size; i++) {
3465 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003466 size++;
3467 }
3468
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003469 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003470
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003471 /* Set thresholds array to NULL if we don't have thresholds */
3472 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003473 kfree(new);
3474 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003475 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003476 }
3477
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003478 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003479
3480 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003481 new->current_threshold = -1;
3482 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3483 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003484 continue;
3485
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003486 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003487 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003488 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003489 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003490 * until rcu_assign_pointer(), so it's safe to increment
3491 * it here.
3492 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003493 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003494 }
3495 j++;
3496 }
3497
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003498swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003499 /* Swap primary and spare array */
3500 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003501
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003502 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003503
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003504 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003505 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08003506
3507 /* If all events are unregistered, free the spare array */
3508 if (!new) {
3509 kfree(thresholds->spare);
3510 thresholds->spare = NULL;
3511 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04003512unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003513 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003514}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003515
Tejun Heo59b6f872013-11-22 18:20:43 -05003516static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003517 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003518{
Tejun Heo59b6f872013-11-22 18:20:43 -05003519 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003520}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003521
Tejun Heo59b6f872013-11-22 18:20:43 -05003522static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003523 struct eventfd_ctx *eventfd)
3524{
Tejun Heo59b6f872013-11-22 18:20:43 -05003525 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003526}
3527
Tejun Heo59b6f872013-11-22 18:20:43 -05003528static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003529 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003530{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003531 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003532
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003533 event = kmalloc(sizeof(*event), GFP_KERNEL);
3534 if (!event)
3535 return -ENOMEM;
3536
Michal Hocko1af8efe2011-07-26 16:08:24 -07003537 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003538
3539 event->eventfd = eventfd;
3540 list_add(&event->list, &memcg->oom_notify);
3541
3542 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07003543 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003544 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003545 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003546
3547 return 0;
3548}
3549
Tejun Heo59b6f872013-11-22 18:20:43 -05003550static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003551 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003552{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003553 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003554
Michal Hocko1af8efe2011-07-26 16:08:24 -07003555 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003556
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003557 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003558 if (ev->eventfd == eventfd) {
3559 list_del(&ev->list);
3560 kfree(ev);
3561 }
3562 }
3563
Michal Hocko1af8efe2011-07-26 16:08:24 -07003564 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003565}
3566
Tejun Heo2da8ca82013-12-05 12:28:04 -05003567static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003568{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003569 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003570
Tejun Heo791badb2013-12-05 12:28:02 -05003571 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07003572 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003573 return 0;
3574}
3575
Tejun Heo182446d2013-08-08 20:11:24 -04003576static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003577 struct cftype *cft, u64 val)
3578{
Tejun Heo182446d2013-08-08 20:11:24 -04003579 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003580
3581 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003582 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003583 return -EINVAL;
3584
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003585 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003586 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003587 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003588
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003589 return 0;
3590}
3591
Tejun Heo52ebea72015-05-22 17:13:37 -04003592#ifdef CONFIG_CGROUP_WRITEBACK
3593
3594struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3595{
3596 return &memcg->cgwb_list;
3597}
3598
Tejun Heo841710a2015-05-22 18:23:33 -04003599static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3600{
3601 return wb_domain_init(&memcg->cgwb_domain, gfp);
3602}
3603
3604static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3605{
3606 wb_domain_exit(&memcg->cgwb_domain);
3607}
3608
Tejun Heo2529bb32015-05-22 18:23:34 -04003609static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3610{
3611 wb_domain_size_changed(&memcg->cgwb_domain);
3612}
3613
Tejun Heo841710a2015-05-22 18:23:33 -04003614struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3615{
3616 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3617
3618 if (!memcg->css.parent)
3619 return NULL;
3620
3621 return &memcg->cgwb_domain;
3622}
3623
Tejun Heoc2aa7232015-05-22 18:23:35 -04003624/**
3625 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3626 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003627 * @pfilepages: out parameter for number of file pages
3628 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04003629 * @pdirty: out parameter for number of dirty pages
3630 * @pwriteback: out parameter for number of pages under writeback
3631 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003632 * Determine the numbers of file, headroom, dirty, and writeback pages in
3633 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3634 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003635 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003636 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3637 * headroom is calculated as the lowest headroom of itself and the
3638 * ancestors. Note that this doesn't consider the actual amount of
3639 * available memory in the system. The caller should further cap
3640 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003641 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003642void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3643 unsigned long *pheadroom, unsigned long *pdirty,
3644 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04003645{
3646 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3647 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003648
3649 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3650
3651 /* this should eventually include NR_UNSTABLE_NFS */
3652 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003653 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3654 (1 << LRU_ACTIVE_FILE));
3655 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003656
Tejun Heoc2aa7232015-05-22 18:23:35 -04003657 while ((parent = parent_mem_cgroup(memcg))) {
3658 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3659 unsigned long used = page_counter_read(&memcg->memory);
3660
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003661 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04003662 memcg = parent;
3663 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04003664}
3665
Tejun Heo841710a2015-05-22 18:23:33 -04003666#else /* CONFIG_CGROUP_WRITEBACK */
3667
3668static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3669{
3670 return 0;
3671}
3672
3673static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3674{
3675}
3676
Tejun Heo2529bb32015-05-22 18:23:34 -04003677static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3678{
3679}
3680
Tejun Heo52ebea72015-05-22 17:13:37 -04003681#endif /* CONFIG_CGROUP_WRITEBACK */
3682
Tejun Heo79bd9812013-11-22 18:20:42 -05003683/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003684 * DO NOT USE IN NEW FILES.
3685 *
3686 * "cgroup.event_control" implementation.
3687 *
3688 * This is way over-engineered. It tries to support fully configurable
3689 * events for each user. Such level of flexibility is completely
3690 * unnecessary especially in the light of the planned unified hierarchy.
3691 *
3692 * Please deprecate this and replace with something simpler if at all
3693 * possible.
3694 */
3695
3696/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003697 * Unregister event and free resources.
3698 *
3699 * Gets called from workqueue.
3700 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003701static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003702{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003703 struct mem_cgroup_event *event =
3704 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003705 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003706
3707 remove_wait_queue(event->wqh, &event->wait);
3708
Tejun Heo59b6f872013-11-22 18:20:43 -05003709 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003710
3711 /* Notify userspace the event is going away. */
3712 eventfd_signal(event->eventfd, 1);
3713
3714 eventfd_ctx_put(event->eventfd);
3715 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003716 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003717}
3718
3719/*
3720 * Gets called on POLLHUP on eventfd when user closes it.
3721 *
3722 * Called with wqh->lock held and interrupts disabled.
3723 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003724static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3725 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003726{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003727 struct mem_cgroup_event *event =
3728 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003729 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003730 unsigned long flags = (unsigned long)key;
3731
3732 if (flags & POLLHUP) {
3733 /*
3734 * If the event has been detached at cgroup removal, we
3735 * can simply return knowing the other side will cleanup
3736 * for us.
3737 *
3738 * We can't race against event freeing since the other
3739 * side will require wqh->lock via remove_wait_queue(),
3740 * which we hold.
3741 */
Tejun Heofba94802013-11-22 18:20:43 -05003742 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003743 if (!list_empty(&event->list)) {
3744 list_del_init(&event->list);
3745 /*
3746 * We are in atomic context, but cgroup_event_remove()
3747 * may sleep, so we have to call it in workqueue.
3748 */
3749 schedule_work(&event->remove);
3750 }
Tejun Heofba94802013-11-22 18:20:43 -05003751 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003752 }
3753
3754 return 0;
3755}
3756
Tejun Heo3bc942f2013-11-22 18:20:44 -05003757static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003758 wait_queue_head_t *wqh, poll_table *pt)
3759{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003760 struct mem_cgroup_event *event =
3761 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003762
3763 event->wqh = wqh;
3764 add_wait_queue(wqh, &event->wait);
3765}
3766
3767/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003768 * DO NOT USE IN NEW FILES.
3769 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003770 * Parse input and register new cgroup event handler.
3771 *
3772 * Input must be in format '<event_fd> <control_fd> <args>'.
3773 * Interpretation of args is defined by control file implementation.
3774 */
Tejun Heo451af5042014-05-13 12:16:21 -04003775static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3776 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003777{
Tejun Heo451af5042014-05-13 12:16:21 -04003778 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003779 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003780 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003781 struct cgroup_subsys_state *cfile_css;
3782 unsigned int efd, cfd;
3783 struct fd efile;
3784 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003785 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003786 char *endp;
3787 int ret;
3788
Tejun Heo451af5042014-05-13 12:16:21 -04003789 buf = strstrip(buf);
3790
3791 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003792 if (*endp != ' ')
3793 return -EINVAL;
Tejun Heo451af5042014-05-13 12:16:21 -04003794 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003795
Tejun Heo451af5042014-05-13 12:16:21 -04003796 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003797 if ((*endp != ' ') && (*endp != '\0'))
3798 return -EINVAL;
Tejun Heo451af5042014-05-13 12:16:21 -04003799 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003800
3801 event = kzalloc(sizeof(*event), GFP_KERNEL);
3802 if (!event)
3803 return -ENOMEM;
3804
Tejun Heo59b6f872013-11-22 18:20:43 -05003805 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003806 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003807 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3808 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3809 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003810
3811 efile = fdget(efd);
3812 if (!efile.file) {
3813 ret = -EBADF;
3814 goto out_kfree;
3815 }
3816
3817 event->eventfd = eventfd_ctx_fileget(efile.file);
3818 if (IS_ERR(event->eventfd)) {
3819 ret = PTR_ERR(event->eventfd);
3820 goto out_put_efile;
3821 }
3822
3823 cfile = fdget(cfd);
3824 if (!cfile.file) {
3825 ret = -EBADF;
3826 goto out_put_eventfd;
3827 }
3828
3829 /* the process need read permission on control file */
3830 /* AV: shouldn't we check that it's been opened for read instead? */
3831 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3832 if (ret < 0)
3833 goto out_put_cfile;
3834
Tejun Heo79bd9812013-11-22 18:20:42 -05003835 /*
Tejun Heofba94802013-11-22 18:20:43 -05003836 * Determine the event callbacks and set them in @event. This used
3837 * to be done via struct cftype but cgroup core no longer knows
3838 * about these events. The following is crude but the whole thing
3839 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003840 *
3841 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003842 */
Al Virob5830432014-10-31 01:22:04 -04003843 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003844
3845 if (!strcmp(name, "memory.usage_in_bytes")) {
3846 event->register_event = mem_cgroup_usage_register_event;
3847 event->unregister_event = mem_cgroup_usage_unregister_event;
3848 } else if (!strcmp(name, "memory.oom_control")) {
3849 event->register_event = mem_cgroup_oom_register_event;
3850 event->unregister_event = mem_cgroup_oom_unregister_event;
3851 } else if (!strcmp(name, "memory.pressure_level")) {
3852 event->register_event = vmpressure_register_event;
3853 event->unregister_event = vmpressure_unregister_event;
3854 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003855 event->register_event = memsw_cgroup_usage_register_event;
3856 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003857 } else {
3858 ret = -EINVAL;
3859 goto out_put_cfile;
3860 }
3861
3862 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003863 * Verify @cfile should belong to @css. Also, remaining events are
3864 * automatically removed on cgroup destruction but the removal is
3865 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003866 */
Al Virob5830432014-10-31 01:22:04 -04003867 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003868 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003869 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003870 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003871 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003872 if (cfile_css != css) {
3873 css_put(cfile_css);
3874 goto out_put_cfile;
3875 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003876
Tejun Heo451af5042014-05-13 12:16:21 -04003877 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003878 if (ret)
3879 goto out_put_css;
3880
3881 efile.file->f_op->poll(efile.file, &event->pt);
3882
Tejun Heofba94802013-11-22 18:20:43 -05003883 spin_lock(&memcg->event_list_lock);
3884 list_add(&event->list, &memcg->event_list);
3885 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003886
3887 fdput(cfile);
3888 fdput(efile);
3889
Tejun Heo451af5042014-05-13 12:16:21 -04003890 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05003891
3892out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05003893 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003894out_put_cfile:
3895 fdput(cfile);
3896out_put_eventfd:
3897 eventfd_ctx_put(event->eventfd);
3898out_put_efile:
3899 fdput(efile);
3900out_kfree:
3901 kfree(event);
3902
3903 return ret;
3904}
3905
Johannes Weiner241994ed2015-02-11 15:26:06 -08003906static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003907 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003908 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003909 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003910 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003911 },
3912 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003913 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003914 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04003915 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003916 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003917 },
3918 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003919 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003920 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af5042014-05-13 12:16:21 -04003921 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003922 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003923 },
3924 {
Balbir Singh296c81d2009-09-23 15:56:36 -07003925 .name = "soft_limit_in_bytes",
3926 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af5042014-05-13 12:16:21 -04003927 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003928 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07003929 },
3930 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003931 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003932 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04003933 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003934 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003935 },
Balbir Singh8697d332008-02-07 00:13:59 -08003936 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003937 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003938 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003939 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003940 {
3941 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04003942 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003943 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08003944 {
3945 .name = "use_hierarchy",
3946 .write_u64 = mem_cgroup_hierarchy_write,
3947 .read_u64 = mem_cgroup_hierarchy_read,
3948 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003949 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05003950 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af5042014-05-13 12:16:21 -04003951 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04003952 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05003953 },
3954 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003955 .name = "swappiness",
3956 .read_u64 = mem_cgroup_swappiness_read,
3957 .write_u64 = mem_cgroup_swappiness_write,
3958 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003959 {
3960 .name = "move_charge_at_immigrate",
3961 .read_u64 = mem_cgroup_move_charge_read,
3962 .write_u64 = mem_cgroup_move_charge_write,
3963 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003964 {
3965 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003966 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003967 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003968 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3969 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003970 {
3971 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003972 },
Ying Han406eb0c2011-05-26 16:25:37 -07003973#ifdef CONFIG_NUMA
3974 {
3975 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003976 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07003977 },
3978#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08003979 {
3980 .name = "kmem.limit_in_bytes",
3981 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af5042014-05-13 12:16:21 -04003982 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003983 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08003984 },
3985 {
3986 .name = "kmem.usage_in_bytes",
3987 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003988 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08003989 },
3990 {
3991 .name = "kmem.failcnt",
3992 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04003993 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003994 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08003995 },
3996 {
3997 .name = "kmem.max_usage_in_bytes",
3998 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04003999 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004000 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004001 },
Glauber Costa749c5412012-12-18 14:23:01 -08004002#ifdef CONFIG_SLABINFO
4003 {
4004 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004005 .seq_start = slab_start,
4006 .seq_next = slab_next,
4007 .seq_stop = slab_stop,
4008 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004009 },
4010#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004011 {
4012 .name = "kmem.tcp.limit_in_bytes",
4013 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4014 .write = mem_cgroup_write,
4015 .read_u64 = mem_cgroup_read_u64,
4016 },
4017 {
4018 .name = "kmem.tcp.usage_in_bytes",
4019 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4020 .read_u64 = mem_cgroup_read_u64,
4021 },
4022 {
4023 .name = "kmem.tcp.failcnt",
4024 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4025 .write = mem_cgroup_reset,
4026 .read_u64 = mem_cgroup_read_u64,
4027 },
4028 {
4029 .name = "kmem.tcp.max_usage_in_bytes",
4030 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4031 .write = mem_cgroup_reset,
4032 .read_u64 = mem_cgroup_read_u64,
4033 },
Tejun Heo6bc10342012-04-01 12:09:55 -07004034 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004035};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004036
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004037static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004038{
4039 struct mem_cgroup_per_node *pn;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004040 struct mem_cgroup_per_zone *mz;
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004041 int zone, tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004042 /*
4043 * This routine is called against possible nodes.
4044 * But it's BUG to call kmalloc() against offline node.
4045 *
4046 * TODO: this routine can waste much memory for nodes which will
4047 * never be onlined. It's better to use memory hotplug callback
4048 * function.
4049 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004050 if (!node_state(node, N_NORMAL_MEMORY))
4051 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004052 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004053 if (!pn)
4054 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004055
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004056 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4057 mz = &pn->zoneinfo[zone];
Hugh Dickinsbea8c152012-11-16 14:14:54 -08004058 lruvec_init(&mz->lruvec);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -07004059 mz->usage_in_excess = 0;
4060 mz->on_tree = false;
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004061 mz->memcg = memcg;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004062 }
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004063 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004064 return 0;
4065}
4066
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004067static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004068{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004069 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004070}
4071
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004072static void mem_cgroup_free(struct mem_cgroup *memcg)
4073{
4074 int node;
4075
4076 memcg_wb_domain_exit(memcg);
4077 for_each_node(node)
4078 free_mem_cgroup_per_zone_info(memcg, node);
4079 free_percpu(memcg->stat);
4080 kfree(memcg);
4081}
4082
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004083static struct mem_cgroup *mem_cgroup_alloc(void)
4084{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004085 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004086 size_t size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004087 int node;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004088
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004089 size = sizeof(struct mem_cgroup);
4090 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004091
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004092 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004093 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004094 return NULL;
4095
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004096 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4097 if (!memcg->stat)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004098 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004099
Bob Liu3ed28fa2012-01-12 17:19:04 -08004100 for_each_node(node)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004101 if (alloc_mem_cgroup_per_zone_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004102 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004103
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004104 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4105 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004106
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004107 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08004108 memcg->last_scanned_node = MAX_NUMNODES;
4109 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004110 mutex_init(&memcg->thresholds_lock);
4111 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004112 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004113 INIT_LIST_HEAD(&memcg->event_list);
4114 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004115 memcg->socket_pressure = jiffies;
Johannes Weiner127424c2016-01-20 15:02:32 -08004116#ifndef CONFIG_SLOB
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004117 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004118#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004119#ifdef CONFIG_CGROUP_WRITEBACK
4120 INIT_LIST_HEAD(&memcg->cgwb_list);
4121#endif
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004122 return memcg;
4123fail:
4124 mem_cgroup_free(memcg);
4125 return NULL;
Glauber Costad142e3e2013-02-22 16:34:52 -08004126}
4127
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004128static struct cgroup_subsys_state * __ref
4129mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004130{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004131 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4132 struct mem_cgroup *memcg;
4133 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08004134
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004135 memcg = mem_cgroup_alloc();
4136 if (!memcg)
4137 return ERR_PTR(error);
Li Zefan4219b2d2013-09-23 16:56:29 +08004138
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004139 memcg->high = PAGE_COUNTER_MAX;
4140 memcg->soft_limit = PAGE_COUNTER_MAX;
4141 if (parent) {
4142 memcg->swappiness = mem_cgroup_swappiness(parent);
4143 memcg->oom_kill_disable = parent->oom_kill_disable;
4144 }
4145 if (parent && parent->use_hierarchy) {
4146 memcg->use_hierarchy = true;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004147 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004148 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004149 page_counter_init(&memcg->memsw, &parent->memsw);
4150 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08004151 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004152 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004153 page_counter_init(&memcg->memory, NULL);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004154 page_counter_init(&memcg->swap, NULL);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004155 page_counter_init(&memcg->memsw, NULL);
4156 page_counter_init(&memcg->kmem, NULL);
Johannes Weiner0db15292016-01-20 15:02:50 -08004157 page_counter_init(&memcg->tcpmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004158 /*
4159 * Deeper hierachy with use_hierarchy == false doesn't make
4160 * much sense so let cgroup subsystem know about this
4161 * unfortunate state in our controller.
4162 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004163 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004164 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004165 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08004166
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004167 /* The following stuff does not apply to the root */
4168 if (!parent) {
4169 root_mem_cgroup = memcg;
4170 return &memcg->css;
4171 }
4172
Vladimir Davydovb313aee2016-03-17 14:18:27 -07004173 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004174 if (error)
4175 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08004176
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004177 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004178 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004179
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004180 return &memcg->css;
4181fail:
4182 mem_cgroup_free(memcg);
4183 return NULL;
4184}
4185
4186static int
4187mem_cgroup_css_online(struct cgroup_subsys_state *css)
4188{
4189 if (css->id > MEM_CGROUP_ID_MAX)
4190 return -ENOSPC;
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004191
4192 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004193}
4194
Tejun Heoeb954192013-08-08 20:11:23 -04004195static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004196{
Tejun Heoeb954192013-08-08 20:11:23 -04004197 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004198 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004199
4200 /*
4201 * Unregister events and notify userspace.
4202 * Notify userspace about cgroup removing only after rmdir of cgroup
4203 * directory to avoid race between userspace and kernelspace.
4204 */
Tejun Heofba94802013-11-22 18:20:43 -05004205 spin_lock(&memcg->event_list_lock);
4206 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004207 list_del_init(&event->list);
4208 schedule_work(&event->remove);
4209 }
Tejun Heofba94802013-11-22 18:20:43 -05004210 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004211
Johannes Weiner567e9ab2016-01-20 15:02:24 -08004212 memcg_offline_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004213 wb_memcg_offline(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004214}
4215
Vladimir Davydov6df38682015-12-29 14:54:10 -08004216static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4217{
4218 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4219
4220 invalidate_reclaim_iterators(memcg);
4221}
4222
Tejun Heoeb954192013-08-08 20:11:23 -04004223static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004224{
Tejun Heoeb954192013-08-08 20:11:23 -04004225 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004226
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004227 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004228 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004229
Johannes Weiner0db15292016-01-20 15:02:50 -08004230 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004231 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004232
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004233 vmpressure_cleanup(&memcg->vmpressure);
4234 cancel_work_sync(&memcg->high_work);
4235 mem_cgroup_remove_from_trees(memcg);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004236 memcg_free_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004237 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004238}
4239
Tejun Heo1ced953b2014-07-08 18:02:57 -04004240/**
4241 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4242 * @css: the target css
4243 *
4244 * Reset the states of the mem_cgroup associated with @css. This is
4245 * invoked when the userland requests disabling on the default hierarchy
4246 * but the memcg is pinned through dependency. The memcg should stop
4247 * applying policies and should revert to the vanilla state as it may be
4248 * made visible again.
4249 *
4250 * The current implementation only resets the essential configurations.
4251 * This needs to be expanded to cover all the visible parts.
4252 */
4253static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4254{
4255 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4256
Vladimir Davydovd334c9bc2016-03-17 14:19:38 -07004257 page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4258 page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4259 page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4260 page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4261 page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004262 memcg->low = 0;
4263 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004264 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004265 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced953b2014-07-08 18:02:57 -04004266}
4267
Daisuke Nishimura02491442010-03-10 15:22:17 -08004268#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004269/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004270static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004271{
Johannes Weiner05b84302014-08-06 16:05:59 -07004272 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004273
Mel Gormand0164ad2015-11-06 16:28:21 -08004274 /* Try a single bulk charge without reclaim first, kswapd may wake */
4275 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004276 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004277 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004278 return ret;
4279 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004280
4281 /* Try charges one by one with reclaim */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004282 while (count--) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004283 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004284 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004285 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004286 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004287 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004288 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004289 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004290}
4291
4292/**
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004293 * get_mctgt_type - get target type of moving charge
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004294 * @vma: the vma the pte to be checked belongs
4295 * @addr: the address corresponding to the pte to be checked
4296 * @ptent: the pte to be checked
Daisuke Nishimura02491442010-03-10 15:22:17 -08004297 * @target: the pointer the target page or swap ent will be stored(can be NULL)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004298 *
4299 * Returns
4300 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4301 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4302 * move charge. if @target is not NULL, the page is stored in target->page
4303 * with extra refcnt got(Callers should handle it).
Daisuke Nishimura02491442010-03-10 15:22:17 -08004304 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4305 * target for charge migration. if @target is not NULL, the entry is stored
4306 * in target->ent.
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004307 *
4308 * Called with pte lock held.
4309 */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004310union mc_target {
4311 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004312 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004313};
4314
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004315enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004316 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004317 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004318 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004319};
4320
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004321static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4322 unsigned long addr, pte_t ptent)
4323{
4324 struct page *page = vm_normal_page(vma, addr, ptent);
4325
4326 if (!page || !page_mapped(page))
4327 return NULL;
4328 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004329 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004330 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004331 } else {
4332 if (!(mc.flags & MOVE_FILE))
4333 return NULL;
4334 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004335 if (!get_page_unless_zero(page))
4336 return NULL;
4337
4338 return page;
4339}
4340
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004341#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004342static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4343 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4344{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004345 struct page *page = NULL;
4346 swp_entry_t ent = pte_to_swp_entry(ptent);
4347
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004348 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004349 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004350 /*
4351 * Because lookup_swap_cache() updates some statistics counter,
4352 * we call find_get_page() with swapper_space directly.
4353 */
Shaohua Li33806f02013-02-22 16:34:37 -08004354 page = find_get_page(swap_address_space(ent), ent.val);
Johannes Weiner7941d212016-01-14 15:21:23 -08004355 if (do_memsw_account())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004356 entry->val = ent.val;
4357
4358 return page;
4359}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004360#else
4361static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4362 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4363{
4364 return NULL;
4365}
4366#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004367
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004368static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4369 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4370{
4371 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004372 struct address_space *mapping;
4373 pgoff_t pgoff;
4374
4375 if (!vma->vm_file) /* anonymous vma */
4376 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004377 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004378 return NULL;
4379
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004380 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004381 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004382
4383 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004384#ifdef CONFIG_SWAP
4385 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004386 if (shmem_mapping(mapping)) {
4387 page = find_get_entry(mapping, pgoff);
4388 if (radix_tree_exceptional_entry(page)) {
4389 swp_entry_t swp = radix_to_swp_entry(page);
Johannes Weiner7941d212016-01-14 15:21:23 -08004390 if (do_memsw_account())
Johannes Weiner139b6a62014-05-06 12:50:05 -07004391 *entry = swp;
4392 page = find_get_page(swap_address_space(swp), swp.val);
4393 }
4394 } else
4395 page = find_get_page(mapping, pgoff);
4396#else
4397 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004398#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004399 return page;
4400}
4401
Chen Gangb1b0dea2015-04-14 15:47:35 -07004402/**
4403 * mem_cgroup_move_account - move account of the page
4404 * @page: the page
4405 * @nr_pages: number of regular pages (>1 for huge pages)
4406 * @from: mem_cgroup which the page is moved from.
4407 * @to: mem_cgroup which the page is moved to. @from != @to.
4408 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08004409 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07004410 *
4411 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4412 * from old cgroup.
4413 */
4414static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004415 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07004416 struct mem_cgroup *from,
4417 struct mem_cgroup *to)
4418{
4419 unsigned long flags;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004420 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004421 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004422 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004423
4424 VM_BUG_ON(from == to);
4425 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004426 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07004427
4428 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07004429 * Prevent mem_cgroup_migrate() from looking at
Hugh Dickins45637ba2015-11-05 18:49:40 -08004430 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07004431 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004432 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004433 if (!trylock_page(page))
4434 goto out;
4435
4436 ret = -EINVAL;
4437 if (page->mem_cgroup != from)
4438 goto out_unlock;
4439
Greg Thelenc4843a72015-05-22 17:13:16 -04004440 anon = PageAnon(page);
4441
Chen Gangb1b0dea2015-04-14 15:47:35 -07004442 spin_lock_irqsave(&from->move_lock, flags);
4443
Greg Thelenc4843a72015-05-22 17:13:16 -04004444 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004445 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4446 nr_pages);
4447 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4448 nr_pages);
4449 }
4450
Greg Thelenc4843a72015-05-22 17:13:16 -04004451 /*
4452 * move_lock grabbed above and caller set from->moving_account, so
4453 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4454 * So mapping should be stable for dirty pages.
4455 */
4456 if (!anon && PageDirty(page)) {
4457 struct address_space *mapping = page_mapping(page);
4458
4459 if (mapping_cap_account_dirty(mapping)) {
4460 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4461 nr_pages);
4462 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4463 nr_pages);
4464 }
4465 }
4466
Chen Gangb1b0dea2015-04-14 15:47:35 -07004467 if (PageWriteback(page)) {
4468 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4469 nr_pages);
4470 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4471 nr_pages);
4472 }
4473
4474 /*
4475 * It is safe to change page->mem_cgroup here because the page
4476 * is referenced, charged, and isolated - we can't race with
4477 * uncharging, charging, migration, or LRU putback.
4478 */
4479
4480 /* caller should have done css_get */
4481 page->mem_cgroup = to;
4482 spin_unlock_irqrestore(&from->move_lock, flags);
4483
4484 ret = 0;
4485
4486 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004487 mem_cgroup_charge_statistics(to, page, compound, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004488 memcg_check_events(to, page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004489 mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004490 memcg_check_events(from, page);
4491 local_irq_enable();
4492out_unlock:
4493 unlock_page(page);
4494out:
4495 return ret;
4496}
4497
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004498static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004499 unsigned long addr, pte_t ptent, union mc_target *target)
4500{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004501 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004502 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004503 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004504
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004505 if (pte_present(ptent))
4506 page = mc_handle_present_pte(vma, addr, ptent);
4507 else if (is_swap_pte(ptent))
4508 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004509 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004510 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004511
4512 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004513 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004514 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004515 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004516 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004517 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004518 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004519 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004520 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004521 ret = MC_TARGET_PAGE;
4522 if (target)
4523 target->page = page;
4524 }
4525 if (!ret || !target)
4526 put_page(page);
4527 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004528 /* There is a swap entry and a page doesn't exist or isn't charged */
4529 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004530 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004531 ret = MC_TARGET_SWAP;
4532 if (target)
4533 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004534 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004535 return ret;
4536}
4537
Naoya Horiguchi12724852012-03-21 16:34:28 -07004538#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4539/*
4540 * We don't consider swapping or file mapped pages because THP does not
4541 * support them for now.
4542 * Caller should make sure that pmd_trans_huge(pmd) is true.
4543 */
4544static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4545 unsigned long addr, pmd_t pmd, union mc_target *target)
4546{
4547 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004548 enum mc_target_type ret = MC_TARGET_NONE;
4549
4550 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004551 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004552 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004553 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004554 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004555 ret = MC_TARGET_PAGE;
4556 if (target) {
4557 get_page(page);
4558 target->page = page;
4559 }
4560 }
4561 return ret;
4562}
4563#else
4564static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4565 unsigned long addr, pmd_t pmd, union mc_target *target)
4566{
4567 return MC_TARGET_NONE;
4568}
4569#endif
4570
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004571static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4572 unsigned long addr, unsigned long end,
4573 struct mm_walk *walk)
4574{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004575 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004576 pte_t *pte;
4577 spinlock_t *ptl;
4578
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004579 ptl = pmd_trans_huge_lock(pmd, vma);
4580 if (ptl) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004581 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4582 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004583 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004584 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004585 }
Dave Hansen03319322011-03-22 16:32:56 -07004586
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004587 if (pmd_trans_unstable(pmd))
4588 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004589 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4590 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004591 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004592 mc.precharge++; /* increment precharge temporarily */
4593 pte_unmap_unlock(pte - 1, ptl);
4594 cond_resched();
4595
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004596 return 0;
4597}
4598
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004599static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4600{
4601 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004602
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004603 struct mm_walk mem_cgroup_count_precharge_walk = {
4604 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4605 .mm = mm,
4606 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004607 down_read(&mm->mmap_sem);
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004608 walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004609 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004610
4611 precharge = mc.precharge;
4612 mc.precharge = 0;
4613
4614 return precharge;
4615}
4616
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004617static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4618{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004619 unsigned long precharge = mem_cgroup_count_precharge(mm);
4620
4621 VM_BUG_ON(mc.moving_task);
4622 mc.moving_task = current;
4623 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004624}
4625
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004626/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4627static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004628{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004629 struct mem_cgroup *from = mc.from;
4630 struct mem_cgroup *to = mc.to;
4631
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004632 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004633 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004634 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004635 mc.precharge = 0;
4636 }
4637 /*
4638 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4639 * we must uncharge here.
4640 */
4641 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004642 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004643 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004644 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004645 /* we must fixup refcnts and charges */
4646 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004647 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004648 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004649 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004650
Johannes Weiner05b84302014-08-06 16:05:59 -07004651 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004652 * we charged both to->memory and to->memsw, so we
4653 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004654 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004655 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004656 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004657
Johannes Weinere8ea14c2014-12-10 15:42:42 -08004658 css_put_many(&mc.from->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004659
Li Zefan40503772013-07-08 16:00:34 -07004660 /* we've already done css_get(mc.to) */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004661 mc.moved_swap = 0;
4662 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004663 memcg_oom_recover(from);
4664 memcg_oom_recover(to);
4665 wake_up_all(&mc.waitq);
4666}
4667
4668static void mem_cgroup_clear_mc(void)
4669{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004670 /*
4671 * we must clear moving_task before waking up waiters at the end of
4672 * task migration.
4673 */
4674 mc.moving_task = NULL;
4675 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004676 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004677 mc.from = NULL;
4678 mc.to = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004679 spin_unlock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004680}
4681
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004682static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004683{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004684 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07004685 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07004686 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04004687 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004688 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004689 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004690 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004691
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004692 /* charge immigration isn't supported on the default hierarchy */
4693 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07004694 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004695
Tejun Heo4530edd2015-09-11 15:00:19 -04004696 /*
4697 * Multi-process migrations only happen on the default hierarchy
4698 * where charge immigration is not used. Perform charge
4699 * immigration if @tset contains a leader and whine if there are
4700 * multiple.
4701 */
4702 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004703 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04004704 WARN_ON_ONCE(p);
4705 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004706 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04004707 }
4708 if (!p)
4709 return 0;
4710
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004711 /*
4712 * We are now commited to this value whatever it is. Changes in this
4713 * tunable will only affect upcoming migrations, not the current one.
4714 * So we need to save it, and keep it going.
4715 */
4716 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4717 if (!move_flags)
4718 return 0;
4719
Tejun Heo9f2115f2015-09-08 15:01:10 -07004720 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004721
Tejun Heo9f2115f2015-09-08 15:01:10 -07004722 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004723
Tejun Heo9f2115f2015-09-08 15:01:10 -07004724 mm = get_task_mm(p);
4725 if (!mm)
4726 return 0;
4727 /* We move charges only when we move a owner of the mm */
4728 if (mm->owner == p) {
4729 VM_BUG_ON(mc.from);
4730 VM_BUG_ON(mc.to);
4731 VM_BUG_ON(mc.precharge);
4732 VM_BUG_ON(mc.moved_charge);
4733 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004734
Tejun Heo9f2115f2015-09-08 15:01:10 -07004735 spin_lock(&mc.lock);
4736 mc.from = from;
4737 mc.to = memcg;
4738 mc.flags = move_flags;
4739 spin_unlock(&mc.lock);
4740 /* We set mc.moving_task later */
4741
4742 ret = mem_cgroup_precharge_mc(mm);
4743 if (ret)
4744 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004745 }
Tejun Heo9f2115f2015-09-08 15:01:10 -07004746 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004747 return ret;
4748}
4749
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004750static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004751{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004752 if (mc.to)
4753 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004754}
4755
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004756static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4757 unsigned long addr, unsigned long end,
4758 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004759{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004760 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004761 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004762 pte_t *pte;
4763 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004764 enum mc_target_type target_type;
4765 union mc_target target;
4766 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004767
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004768 ptl = pmd_trans_huge_lock(pmd, vma);
4769 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004770 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004771 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004772 return 0;
4773 }
4774 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4775 if (target_type == MC_TARGET_PAGE) {
4776 page = target.page;
4777 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004778 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08004779 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004780 mc.precharge -= HPAGE_PMD_NR;
4781 mc.moved_charge += HPAGE_PMD_NR;
4782 }
4783 putback_lru_page(page);
4784 }
4785 put_page(page);
4786 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004787 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004788 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004789 }
4790
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004791 if (pmd_trans_unstable(pmd))
4792 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004793retry:
4794 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4795 for (; addr != end; addr += PAGE_SIZE) {
4796 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004797 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004798
4799 if (!mc.precharge)
4800 break;
4801
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004802 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004803 case MC_TARGET_PAGE:
4804 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08004805 /*
4806 * We can have a part of the split pmd here. Moving it
4807 * can be done but it would be too convoluted so simply
4808 * ignore such a partial THP and keep it in original
4809 * memcg. There should be somebody mapping the head.
4810 */
4811 if (PageTransCompound(page))
4812 goto put;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004813 if (isolate_lru_page(page))
4814 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004815 if (!mem_cgroup_move_account(page, false,
4816 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004817 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004818 /* we uncharge from mc.from later. */
4819 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004820 }
4821 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004822put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004823 put_page(page);
4824 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004825 case MC_TARGET_SWAP:
4826 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004827 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004828 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004829 /* we fixup refcnts and charges later. */
4830 mc.moved_swap++;
4831 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004832 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004833 default:
4834 break;
4835 }
4836 }
4837 pte_unmap_unlock(pte - 1, ptl);
4838 cond_resched();
4839
4840 if (addr != end) {
4841 /*
4842 * We have consumed all precharges we got in can_attach().
4843 * We try charge one by one, but don't do any additional
4844 * charges to mc.to if we have failed in charge once in attach()
4845 * phase.
4846 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004847 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004848 if (!ret)
4849 goto retry;
4850 }
4851
4852 return ret;
4853}
4854
4855static void mem_cgroup_move_charge(struct mm_struct *mm)
4856{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004857 struct mm_walk mem_cgroup_move_charge_walk = {
4858 .pmd_entry = mem_cgroup_move_charge_pte_range,
4859 .mm = mm,
4860 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004861
4862 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004863 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07004864 * Signal lock_page_memcg() to take the memcg's move_lock
4865 * while we're moving its pages to another memcg. Then wait
4866 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08004867 */
4868 atomic_inc(&mc.from->moving_account);
4869 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004870retry:
4871 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
4872 /*
4873 * Someone who are holding the mmap_sem might be waiting in
4874 * waitq. So we cancel all extra charges, wake up all waiters,
4875 * and retry. Because we cancel precharges, we might not be able
4876 * to move enough charges, but moving charge is a best-effort
4877 * feature anyway, so it wouldn't be a big problem.
4878 */
4879 __mem_cgroup_clear_mc();
4880 cond_resched();
4881 goto retry;
4882 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004883 /*
4884 * When we have consumed all precharges and failed in doing
4885 * additional charge, the page walk just aborts.
4886 */
4887 walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004888 up_read(&mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08004889 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004890}
4891
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004892static void mem_cgroup_move_task(struct cgroup_taskset *tset)
Balbir Singh67e465a2008-02-07 00:13:54 -08004893{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004894 struct cgroup_subsys_state *css;
4895 struct task_struct *p = cgroup_taskset_first(tset, &css);
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004896 struct mm_struct *mm = get_task_mm(p);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004897
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004898 if (mm) {
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004899 if (mc.to)
4900 mem_cgroup_move_charge(mm);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004901 mmput(mm);
4902 }
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004903 if (mc.to)
4904 mem_cgroup_clear_mc();
Balbir Singh67e465a2008-02-07 00:13:54 -08004905}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004906#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004907static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004908{
4909 return 0;
4910}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004911static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004912{
4913}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004914static void mem_cgroup_move_task(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004915{
4916}
4917#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08004918
Tejun Heof00baae2013-04-15 13:41:15 -07004919/*
4920 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04004921 * to verify whether we're attached to the default hierarchy on each mount
4922 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07004923 */
Tejun Heoeb954192013-08-08 20:11:23 -04004924static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07004925{
4926 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04004927 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07004928 * guarantees that @root doesn't have any children, so turning it
4929 * on for the root memcg is enough.
4930 */
Tejun Heo9e10a132015-09-18 11:56:28 -04004931 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee5902015-03-12 16:26:19 -07004932 root_mem_cgroup->use_hierarchy = true;
4933 else
4934 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07004935}
4936
Johannes Weiner241994ed2015-02-11 15:26:06 -08004937static u64 memory_current_read(struct cgroup_subsys_state *css,
4938 struct cftype *cft)
4939{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08004940 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4941
4942 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994ed2015-02-11 15:26:06 -08004943}
4944
4945static int memory_low_show(struct seq_file *m, void *v)
4946{
4947 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07004948 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004949
4950 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08004951 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08004952 else
4953 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
4954
4955 return 0;
4956}
4957
4958static ssize_t memory_low_write(struct kernfs_open_file *of,
4959 char *buf, size_t nbytes, loff_t off)
4960{
4961 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
4962 unsigned long low;
4963 int err;
4964
4965 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08004966 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004967 if (err)
4968 return err;
4969
4970 memcg->low = low;
4971
4972 return nbytes;
4973}
4974
4975static int memory_high_show(struct seq_file *m, void *v)
4976{
4977 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07004978 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004979
4980 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08004981 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08004982 else
4983 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
4984
4985 return 0;
4986}
4987
4988static ssize_t memory_high_write(struct kernfs_open_file *of,
4989 char *buf, size_t nbytes, loff_t off)
4990{
4991 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner588083b2016-03-17 14:20:25 -07004992 unsigned long nr_pages;
Johannes Weiner241994ed2015-02-11 15:26:06 -08004993 unsigned long high;
4994 int err;
4995
4996 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08004997 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004998 if (err)
4999 return err;
5000
5001 memcg->high = high;
5002
Johannes Weiner588083b2016-03-17 14:20:25 -07005003 nr_pages = page_counter_read(&memcg->memory);
5004 if (nr_pages > high)
5005 try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5006 GFP_KERNEL, true);
5007
Tejun Heo2529bb32015-05-22 18:23:34 -04005008 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005009 return nbytes;
5010}
5011
5012static int memory_max_show(struct seq_file *m, void *v)
5013{
5014 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005015 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005016
5017 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005018 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005019 else
5020 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5021
5022 return 0;
5023}
5024
5025static ssize_t memory_max_write(struct kernfs_open_file *of,
5026 char *buf, size_t nbytes, loff_t off)
5027{
5028 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005029 unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5030 bool drained = false;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005031 unsigned long max;
5032 int err;
5033
5034 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005035 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005036 if (err)
5037 return err;
5038
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005039 xchg(&memcg->memory.limit, max);
5040
5041 for (;;) {
5042 unsigned long nr_pages = page_counter_read(&memcg->memory);
5043
5044 if (nr_pages <= max)
5045 break;
5046
5047 if (signal_pending(current)) {
5048 err = -EINTR;
5049 break;
5050 }
5051
5052 if (!drained) {
5053 drain_all_stock(memcg);
5054 drained = true;
5055 continue;
5056 }
5057
5058 if (nr_reclaims) {
5059 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5060 GFP_KERNEL, true))
5061 nr_reclaims--;
5062 continue;
5063 }
5064
5065 mem_cgroup_events(memcg, MEMCG_OOM, 1);
5066 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5067 break;
5068 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08005069
Tejun Heo2529bb32015-05-22 18:23:34 -04005070 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005071 return nbytes;
5072}
5073
5074static int memory_events_show(struct seq_file *m, void *v)
5075{
5076 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5077
5078 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5079 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5080 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5081 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5082
5083 return 0;
5084}
5085
Johannes Weiner587d9f72016-01-20 15:03:19 -08005086static int memory_stat_show(struct seq_file *m, void *v)
5087{
5088 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005089 unsigned long stat[MEMCG_NR_STAT];
5090 unsigned long events[MEMCG_NR_EVENTS];
Johannes Weiner587d9f72016-01-20 15:03:19 -08005091 int i;
5092
5093 /*
5094 * Provide statistics on the state of the memory subsystem as
5095 * well as cumulative event counters that show past behavior.
5096 *
5097 * This list is ordered following a combination of these gradients:
5098 * 1) generic big picture -> specifics and details
5099 * 2) reflecting userspace activity -> reflecting kernel heuristics
5100 *
5101 * Current memory state:
5102 */
5103
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005104 tree_stat(memcg, stat);
5105 tree_events(memcg, events);
5106
Johannes Weiner587d9f72016-01-20 15:03:19 -08005107 seq_printf(m, "anon %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005108 (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005109 seq_printf(m, "file %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005110 (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE);
Vladimir Davydov12580e42016-03-17 14:17:38 -07005111 seq_printf(m, "kernel_stack %llu\n",
5112 (u64)stat[MEMCG_KERNEL_STACK] * PAGE_SIZE);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005113 seq_printf(m, "slab %llu\n",
5114 (u64)(stat[MEMCG_SLAB_RECLAIMABLE] +
5115 stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
Johannes Weinerb2807f02016-01-20 15:03:22 -08005116 seq_printf(m, "sock %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005117 (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005118
5119 seq_printf(m, "file_mapped %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005120 (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005121 seq_printf(m, "file_dirty %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005122 (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005123 seq_printf(m, "file_writeback %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005124 (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005125
5126 for (i = 0; i < NR_LRU_LISTS; i++) {
5127 struct mem_cgroup *mi;
5128 unsigned long val = 0;
5129
5130 for_each_mem_cgroup_tree(mi, memcg)
5131 val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5132 seq_printf(m, "%s %llu\n",
5133 mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5134 }
5135
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005136 seq_printf(m, "slab_reclaimable %llu\n",
5137 (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE);
5138 seq_printf(m, "slab_unreclaimable %llu\n",
5139 (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5140
Johannes Weiner587d9f72016-01-20 15:03:19 -08005141 /* Accumulated memory events */
5142
5143 seq_printf(m, "pgfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005144 events[MEM_CGROUP_EVENTS_PGFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005145 seq_printf(m, "pgmajfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005146 events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005147
5148 return 0;
5149}
5150
Johannes Weiner241994ed2015-02-11 15:26:06 -08005151static struct cftype memory_files[] = {
5152 {
5153 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005154 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005155 .read_u64 = memory_current_read,
5156 },
5157 {
5158 .name = "low",
5159 .flags = CFTYPE_NOT_ON_ROOT,
5160 .seq_show = memory_low_show,
5161 .write = memory_low_write,
5162 },
5163 {
5164 .name = "high",
5165 .flags = CFTYPE_NOT_ON_ROOT,
5166 .seq_show = memory_high_show,
5167 .write = memory_high_write,
5168 },
5169 {
5170 .name = "max",
5171 .flags = CFTYPE_NOT_ON_ROOT,
5172 .seq_show = memory_max_show,
5173 .write = memory_max_write,
5174 },
5175 {
5176 .name = "events",
5177 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04005178 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994ed2015-02-11 15:26:06 -08005179 .seq_show = memory_events_show,
5180 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08005181 {
5182 .name = "stat",
5183 .flags = CFTYPE_NOT_ON_ROOT,
5184 .seq_show = memory_stat_show,
5185 },
Johannes Weiner241994ed2015-02-11 15:26:06 -08005186 { } /* terminate */
5187};
5188
Tejun Heo073219e2014-02-08 10:36:58 -05005189struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005190 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005191 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005192 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08005193 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08005194 .css_free = mem_cgroup_css_free,
Tejun Heo1ced953b2014-07-08 18:02:57 -04005195 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005196 .can_attach = mem_cgroup_can_attach,
5197 .cancel_attach = mem_cgroup_cancel_attach,
Balbir Singh67e465a2008-02-07 00:13:54 -08005198 .attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005199 .bind = mem_cgroup_bind,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005200 .dfl_cftypes = memory_files,
5201 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005202 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005203};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005204
Johannes Weiner241994ed2015-02-11 15:26:06 -08005205/**
Johannes Weiner241994ed2015-02-11 15:26:06 -08005206 * mem_cgroup_low - check if memory consumption is below the normal range
5207 * @root: the highest ancestor to consider
5208 * @memcg: the memory cgroup to check
5209 *
5210 * Returns %true if memory consumption of @memcg, and that of all
5211 * configurable ancestors up to @root, is below the normal range.
5212 */
5213bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5214{
5215 if (mem_cgroup_disabled())
5216 return false;
5217
5218 /*
5219 * The toplevel group doesn't have a configurable range, so
5220 * it's never low when looked at directly, and it is not
5221 * considered an ancestor when assessing the hierarchy.
5222 */
5223
5224 if (memcg == root_mem_cgroup)
5225 return false;
5226
Michal Hocko4e54ded2015-02-27 15:51:46 -08005227 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005228 return false;
5229
5230 while (memcg != root) {
5231 memcg = parent_mem_cgroup(memcg);
5232
5233 if (memcg == root_mem_cgroup)
5234 break;
5235
Michal Hocko4e54ded2015-02-27 15:51:46 -08005236 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005237 return false;
5238 }
5239 return true;
5240}
5241
Johannes Weiner00501b52014-08-08 14:19:20 -07005242/**
5243 * mem_cgroup_try_charge - try charging a page
5244 * @page: page to charge
5245 * @mm: mm context of the victim
5246 * @gfp_mask: reclaim mode
5247 * @memcgp: charged memcg return
5248 *
5249 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5250 * pages according to @gfp_mask if necessary.
5251 *
5252 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5253 * Otherwise, an error code is returned.
5254 *
5255 * After page->mapping has been set up, the caller must finalize the
5256 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5257 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5258 */
5259int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005260 gfp_t gfp_mask, struct mem_cgroup **memcgp,
5261 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005262{
5263 struct mem_cgroup *memcg = NULL;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005264 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005265 int ret = 0;
5266
5267 if (mem_cgroup_disabled())
5268 goto out;
5269
5270 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005271 /*
5272 * Every swap fault against a single page tries to charge the
5273 * page, bail as early as possible. shmem_unuse() encounters
5274 * already charged pages, too. The USED bit is protected by
5275 * the page lock, which serializes swap cache removal, which
5276 * in turn serializes uncharging.
5277 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005278 VM_BUG_ON_PAGE(!PageLocked(page), page);
Johannes Weiner1306a852014-12-10 15:44:52 -08005279 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005280 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005281
Vladimir Davydov37e84352016-01-20 15:02:56 -08005282 if (do_swap_account) {
Vladimir Davydove993d902015-09-09 15:35:35 -07005283 swp_entry_t ent = { .val = page_private(page), };
5284 unsigned short id = lookup_swap_cgroup_id(ent);
5285
5286 rcu_read_lock();
5287 memcg = mem_cgroup_from_id(id);
5288 if (memcg && !css_tryget_online(&memcg->css))
5289 memcg = NULL;
5290 rcu_read_unlock();
5291 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005292 }
5293
Johannes Weiner00501b52014-08-08 14:19:20 -07005294 if (!memcg)
5295 memcg = get_mem_cgroup_from_mm(mm);
5296
5297 ret = try_charge(memcg, gfp_mask, nr_pages);
5298
5299 css_put(&memcg->css);
Johannes Weiner00501b52014-08-08 14:19:20 -07005300out:
5301 *memcgp = memcg;
5302 return ret;
5303}
5304
5305/**
5306 * mem_cgroup_commit_charge - commit a page charge
5307 * @page: page to charge
5308 * @memcg: memcg to charge the page to
5309 * @lrucare: page might be on LRU already
5310 *
5311 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5312 * after page->mapping has been set up. This must happen atomically
5313 * as part of the page instantiation, i.e. under the page table lock
5314 * for anonymous pages, under the page lock for page and swap cache.
5315 *
5316 * In addition, the page must not be on the LRU during the commit, to
5317 * prevent racing with task migration. If it might be, use @lrucare.
5318 *
5319 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5320 */
5321void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005322 bool lrucare, bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005323{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005324 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005325
5326 VM_BUG_ON_PAGE(!page->mapping, page);
5327 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5328
5329 if (mem_cgroup_disabled())
5330 return;
5331 /*
5332 * Swap faults will attempt to charge the same page multiple
5333 * times. But reuse_swap_page() might have removed the page
5334 * from swapcache already, so we can't check PageSwapCache().
5335 */
5336 if (!memcg)
5337 return;
5338
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005339 commit_charge(page, memcg, lrucare);
5340
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005341 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005342 mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005343 memcg_check_events(memcg, page);
5344 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005345
Johannes Weiner7941d212016-01-14 15:21:23 -08005346 if (do_memsw_account() && PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005347 swp_entry_t entry = { .val = page_private(page) };
5348 /*
5349 * The swap entry might not get freed for a long time,
5350 * let's not wait for it. The page already received a
5351 * memory+swap charge, drop the swap entry duplicate.
5352 */
5353 mem_cgroup_uncharge_swap(entry);
5354 }
5355}
5356
5357/**
5358 * mem_cgroup_cancel_charge - cancel a page charge
5359 * @page: page to charge
5360 * @memcg: memcg to charge the page to
5361 *
5362 * Cancel a charge transaction started by mem_cgroup_try_charge().
5363 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005364void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5365 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005366{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005367 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005368
5369 if (mem_cgroup_disabled())
5370 return;
5371 /*
5372 * Swap faults will attempt to charge the same page multiple
5373 * times. But reuse_swap_page() might have removed the page
5374 * from swapcache already, so we can't check PageSwapCache().
5375 */
5376 if (!memcg)
5377 return;
5378
Johannes Weiner00501b52014-08-08 14:19:20 -07005379 cancel_charge(memcg, nr_pages);
5380}
5381
Johannes Weiner747db952014-08-08 14:19:24 -07005382static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005383 unsigned long nr_anon, unsigned long nr_file,
5384 unsigned long nr_huge, struct page *dummy_page)
5385{
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005386 unsigned long nr_pages = nr_anon + nr_file;
Johannes Weiner747db952014-08-08 14:19:24 -07005387 unsigned long flags;
5388
Johannes Weinerce00a962014-09-05 08:43:57 -04005389 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005390 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08005391 if (do_memsw_account())
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005392 page_counter_uncharge(&memcg->memsw, nr_pages);
Johannes Weinerce00a962014-09-05 08:43:57 -04005393 memcg_oom_recover(memcg);
5394 }
Johannes Weiner747db952014-08-08 14:19:24 -07005395
5396 local_irq_save(flags);
5397 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5398 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5399 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5400 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005401 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005402 memcg_check_events(memcg, dummy_page);
5403 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005404
5405 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005406 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005407}
5408
5409static void uncharge_list(struct list_head *page_list)
5410{
5411 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005412 unsigned long nr_anon = 0;
5413 unsigned long nr_file = 0;
5414 unsigned long nr_huge = 0;
5415 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005416 struct list_head *next;
5417 struct page *page;
5418
Johannes Weiner8b592652016-03-17 14:20:31 -07005419 /*
5420 * Note that the list can be a single page->lru; hence the
5421 * do-while loop instead of a simple list_for_each_entry().
5422 */
Johannes Weiner747db952014-08-08 14:19:24 -07005423 next = page_list->next;
5424 do {
5425 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005426
5427 page = list_entry(next, struct page, lru);
5428 next = page->lru.next;
5429
5430 VM_BUG_ON_PAGE(PageLRU(page), page);
5431 VM_BUG_ON_PAGE(page_count(page), page);
5432
Johannes Weiner1306a852014-12-10 15:44:52 -08005433 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005434 continue;
5435
5436 /*
5437 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005438 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005439 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005440 */
5441
Johannes Weiner1306a852014-12-10 15:44:52 -08005442 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005443 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005444 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5445 nr_huge, page);
5446 pgpgout = nr_anon = nr_file = nr_huge = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005447 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005448 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005449 }
5450
5451 if (PageTransHuge(page)) {
5452 nr_pages <<= compound_order(page);
5453 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5454 nr_huge += nr_pages;
5455 }
5456
5457 if (PageAnon(page))
5458 nr_anon += nr_pages;
5459 else
5460 nr_file += nr_pages;
5461
Johannes Weiner1306a852014-12-10 15:44:52 -08005462 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005463
5464 pgpgout++;
5465 } while (next != page_list);
5466
5467 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005468 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5469 nr_huge, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005470}
5471
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005472/**
5473 * mem_cgroup_uncharge - uncharge a page
5474 * @page: page to uncharge
5475 *
5476 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5477 * mem_cgroup_commit_charge().
5478 */
5479void mem_cgroup_uncharge(struct page *page)
5480{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005481 if (mem_cgroup_disabled())
5482 return;
5483
Johannes Weiner747db952014-08-08 14:19:24 -07005484 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005485 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005486 return;
5487
Johannes Weiner747db952014-08-08 14:19:24 -07005488 INIT_LIST_HEAD(&page->lru);
5489 uncharge_list(&page->lru);
5490}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005491
Johannes Weiner747db952014-08-08 14:19:24 -07005492/**
5493 * mem_cgroup_uncharge_list - uncharge a list of page
5494 * @page_list: list of pages to uncharge
5495 *
5496 * Uncharge a list of pages previously charged with
5497 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5498 */
5499void mem_cgroup_uncharge_list(struct list_head *page_list)
5500{
5501 if (mem_cgroup_disabled())
5502 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005503
Johannes Weiner747db952014-08-08 14:19:24 -07005504 if (!list_empty(page_list))
5505 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005506}
5507
5508/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005509 * mem_cgroup_migrate - charge a page's replacement
5510 * @oldpage: currently circulating page
5511 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005512 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005513 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5514 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005515 *
5516 * Both pages must be locked, @newpage->mapping must be set up.
5517 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005518void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005519{
Johannes Weiner29833312014-12-10 15:44:02 -08005520 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005521 unsigned int nr_pages;
5522 bool compound;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005523
5524 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5525 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005526 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005527 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5528 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005529
5530 if (mem_cgroup_disabled())
5531 return;
5532
5533 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005534 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005535 return;
5536
Hugh Dickins45637ba2015-11-05 18:49:40 -08005537 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08005538 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005539 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005540 return;
5541
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005542 /* Force-charge the new page. The old one will be freed soon */
5543 compound = PageTransHuge(newpage);
5544 nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5545
5546 page_counter_charge(&memcg->memory, nr_pages);
5547 if (do_memsw_account())
5548 page_counter_charge(&memcg->memsw, nr_pages);
5549 css_get_many(&memcg->css, nr_pages);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005550
Johannes Weiner9cf76662016-03-15 14:57:58 -07005551 commit_charge(newpage, memcg, false);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005552
5553 local_irq_disable();
5554 mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5555 memcg_check_events(memcg, newpage);
5556 local_irq_enable();
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005557}
5558
Johannes Weineref129472016-01-14 15:21:34 -08005559DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08005560EXPORT_SYMBOL(memcg_sockets_enabled_key);
5561
5562void sock_update_memcg(struct sock *sk)
5563{
5564 struct mem_cgroup *memcg;
5565
5566 /* Socket cloning can throw us here with sk_cgrp already
5567 * filled. It won't however, necessarily happen from
5568 * process context. So the test for root memcg given
5569 * the current task's memcg won't help us in this case.
5570 *
5571 * Respecting the original socket's memcg is a better
5572 * decision in this case.
5573 */
5574 if (sk->sk_memcg) {
5575 BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5576 css_get(&sk->sk_memcg->css);
5577 return;
5578 }
5579
5580 rcu_read_lock();
5581 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005582 if (memcg == root_mem_cgroup)
5583 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08005584 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005585 goto out;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005586 if (css_tryget_online(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08005587 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005588out:
Johannes Weiner11092082016-01-14 15:21:26 -08005589 rcu_read_unlock();
5590}
5591EXPORT_SYMBOL(sock_update_memcg);
5592
5593void sock_release_memcg(struct sock *sk)
5594{
5595 WARN_ON(!sk->sk_memcg);
5596 css_put(&sk->sk_memcg->css);
5597}
5598
5599/**
5600 * mem_cgroup_charge_skmem - charge socket memory
5601 * @memcg: memcg to charge
5602 * @nr_pages: number of pages to charge
5603 *
5604 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5605 * @memcg's configured limit, %false if the charge had to be forced.
5606 */
5607bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5608{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005609 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08005610
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005611 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005612 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005613
Johannes Weiner0db15292016-01-20 15:02:50 -08005614 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5615 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005616 return true;
5617 }
Johannes Weiner0db15292016-01-20 15:02:50 -08005618 page_counter_charge(&memcg->tcpmem, nr_pages);
5619 memcg->tcpmem_pressure = 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005620 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08005621 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005622
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005623 /* Don't block in the packet receive path */
5624 if (in_softirq())
5625 gfp_mask = GFP_NOWAIT;
5626
Johannes Weinerb2807f02016-01-20 15:03:22 -08005627 this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5628
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005629 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5630 return true;
5631
5632 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005633 return false;
5634}
5635
5636/**
5637 * mem_cgroup_uncharge_skmem - uncharge socket memory
5638 * @memcg - memcg to uncharge
5639 * @nr_pages - number of pages to uncharge
5640 */
5641void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5642{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005643 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005644 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005645 return;
5646 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005647
Johannes Weinerb2807f02016-01-20 15:03:22 -08005648 this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5649
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005650 page_counter_uncharge(&memcg->memory, nr_pages);
5651 css_put_many(&memcg->css, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005652}
5653
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005654static int __init cgroup_memory(char *s)
5655{
5656 char *token;
5657
5658 while ((token = strsep(&s, ",")) != NULL) {
5659 if (!*token)
5660 continue;
5661 if (!strcmp(token, "nosocket"))
5662 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08005663 if (!strcmp(token, "nokmem"))
5664 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005665 }
5666 return 0;
5667}
5668__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08005669
Michal Hocko2d110852013-02-22 16:34:43 -08005670/*
Michal Hocko10813122013-02-22 16:35:41 -08005671 * subsys_initcall() for memory controller.
5672 *
5673 * Some parts like hotcpu_notifier() have to be initialized from this context
5674 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5675 * everything that doesn't depend on a specific mem_cgroup structure should
5676 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005677 */
5678static int __init mem_cgroup_init(void)
5679{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005680 int cpu, node;
5681
Michal Hocko2d110852013-02-22 16:34:43 -08005682 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005683
5684 for_each_possible_cpu(cpu)
5685 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5686 drain_local_stock);
5687
5688 for_each_node(node) {
5689 struct mem_cgroup_tree_per_node *rtpn;
5690 int zone;
5691
5692 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5693 node_online(node) ? node : NUMA_NO_NODE);
5694
5695 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5696 struct mem_cgroup_tree_per_zone *rtpz;
5697
5698 rtpz = &rtpn->rb_tree_per_zone[zone];
5699 rtpz->rb_root = RB_ROOT;
5700 spin_lock_init(&rtpz->lock);
5701 }
5702 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5703 }
5704
Michal Hocko2d110852013-02-22 16:34:43 -08005705 return 0;
5706}
5707subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005708
5709#ifdef CONFIG_MEMCG_SWAP
5710/**
5711 * mem_cgroup_swapout - transfer a memsw charge to swap
5712 * @page: page whose memsw charge to transfer
5713 * @entry: swap entry to move the charge to
5714 *
5715 * Transfer the memsw charge of @page to @entry.
5716 */
5717void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5718{
5719 struct mem_cgroup *memcg;
5720 unsigned short oldid;
5721
5722 VM_BUG_ON_PAGE(PageLRU(page), page);
5723 VM_BUG_ON_PAGE(page_count(page), page);
5724
Johannes Weiner7941d212016-01-14 15:21:23 -08005725 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08005726 return;
5727
5728 memcg = page->mem_cgroup;
5729
5730 /* Readahead page, never charged */
5731 if (!memcg)
5732 return;
5733
5734 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5735 VM_BUG_ON_PAGE(oldid, page);
5736 mem_cgroup_swap_statistics(memcg, true);
5737
5738 page->mem_cgroup = NULL;
5739
5740 if (!mem_cgroup_is_root(memcg))
5741 page_counter_uncharge(&memcg->memory, 1);
5742
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07005743 /*
5744 * Interrupts should be disabled here because the caller holds the
5745 * mapping->tree_lock lock which is taken with interrupts-off. It is
5746 * important here to have the interrupts disabled because it is the
5747 * only synchronisation we have for udpating the per-CPU variables.
5748 */
5749 VM_BUG_ON(!irqs_disabled());
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005750 mem_cgroup_charge_statistics(memcg, page, false, -1);
Johannes Weiner21afa382015-02-11 15:26:36 -08005751 memcg_check_events(memcg, page);
5752}
5753
Vladimir Davydov37e84352016-01-20 15:02:56 -08005754/*
5755 * mem_cgroup_try_charge_swap - try charging a swap entry
5756 * @page: page being added to swap
5757 * @entry: swap entry to charge
5758 *
5759 * Try to charge @entry to the memcg that @page belongs to.
5760 *
5761 * Returns 0 on success, -ENOMEM on failure.
5762 */
5763int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5764{
5765 struct mem_cgroup *memcg;
5766 struct page_counter *counter;
5767 unsigned short oldid;
5768
5769 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5770 return 0;
5771
5772 memcg = page->mem_cgroup;
5773
5774 /* Readahead page, never charged */
5775 if (!memcg)
5776 return 0;
5777
5778 if (!mem_cgroup_is_root(memcg) &&
5779 !page_counter_try_charge(&memcg->swap, 1, &counter))
5780 return -ENOMEM;
5781
5782 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5783 VM_BUG_ON_PAGE(oldid, page);
5784 mem_cgroup_swap_statistics(memcg, true);
5785
5786 css_get(&memcg->css);
5787 return 0;
5788}
5789
Johannes Weiner21afa382015-02-11 15:26:36 -08005790/**
5791 * mem_cgroup_uncharge_swap - uncharge a swap entry
5792 * @entry: swap entry to uncharge
5793 *
Vladimir Davydov37e84352016-01-20 15:02:56 -08005794 * Drop the swap charge associated with @entry.
Johannes Weiner21afa382015-02-11 15:26:36 -08005795 */
5796void mem_cgroup_uncharge_swap(swp_entry_t entry)
5797{
5798 struct mem_cgroup *memcg;
5799 unsigned short id;
5800
Vladimir Davydov37e84352016-01-20 15:02:56 -08005801 if (!do_swap_account)
Johannes Weiner21afa382015-02-11 15:26:36 -08005802 return;
5803
5804 id = swap_cgroup_record(entry, 0);
5805 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005806 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005807 if (memcg) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08005808 if (!mem_cgroup_is_root(memcg)) {
5809 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5810 page_counter_uncharge(&memcg->swap, 1);
5811 else
5812 page_counter_uncharge(&memcg->memsw, 1);
5813 }
Johannes Weiner21afa382015-02-11 15:26:36 -08005814 mem_cgroup_swap_statistics(memcg, false);
5815 css_put(&memcg->css);
5816 }
5817 rcu_read_unlock();
5818}
5819
Vladimir Davydovd8b38432016-01-20 15:03:07 -08005820long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
5821{
5822 long nr_swap_pages = get_nr_swap_pages();
5823
5824 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5825 return nr_swap_pages;
5826 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5827 nr_swap_pages = min_t(long, nr_swap_pages,
5828 READ_ONCE(memcg->swap.limit) -
5829 page_counter_read(&memcg->swap));
5830 return nr_swap_pages;
5831}
5832
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08005833bool mem_cgroup_swap_full(struct page *page)
5834{
5835 struct mem_cgroup *memcg;
5836
5837 VM_BUG_ON_PAGE(!PageLocked(page), page);
5838
5839 if (vm_swap_full())
5840 return true;
5841 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5842 return false;
5843
5844 memcg = page->mem_cgroup;
5845 if (!memcg)
5846 return false;
5847
5848 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5849 if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
5850 return true;
5851
5852 return false;
5853}
5854
Johannes Weiner21afa382015-02-11 15:26:36 -08005855/* for remember boot option*/
5856#ifdef CONFIG_MEMCG_SWAP_ENABLED
5857static int really_do_swap_account __initdata = 1;
5858#else
5859static int really_do_swap_account __initdata;
5860#endif
5861
5862static int __init enable_swap_account(char *s)
5863{
5864 if (!strcmp(s, "1"))
5865 really_do_swap_account = 1;
5866 else if (!strcmp(s, "0"))
5867 really_do_swap_account = 0;
5868 return 1;
5869}
5870__setup("swapaccount=", enable_swap_account);
5871
Vladimir Davydov37e84352016-01-20 15:02:56 -08005872static u64 swap_current_read(struct cgroup_subsys_state *css,
5873 struct cftype *cft)
5874{
5875 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5876
5877 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
5878}
5879
5880static int swap_max_show(struct seq_file *m, void *v)
5881{
5882 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5883 unsigned long max = READ_ONCE(memcg->swap.limit);
5884
5885 if (max == PAGE_COUNTER_MAX)
5886 seq_puts(m, "max\n");
5887 else
5888 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5889
5890 return 0;
5891}
5892
5893static ssize_t swap_max_write(struct kernfs_open_file *of,
5894 char *buf, size_t nbytes, loff_t off)
5895{
5896 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5897 unsigned long max;
5898 int err;
5899
5900 buf = strstrip(buf);
5901 err = page_counter_memparse(buf, "max", &max);
5902 if (err)
5903 return err;
5904
5905 mutex_lock(&memcg_limit_mutex);
5906 err = page_counter_limit(&memcg->swap, max);
5907 mutex_unlock(&memcg_limit_mutex);
5908 if (err)
5909 return err;
5910
5911 return nbytes;
5912}
5913
5914static struct cftype swap_files[] = {
5915 {
5916 .name = "swap.current",
5917 .flags = CFTYPE_NOT_ON_ROOT,
5918 .read_u64 = swap_current_read,
5919 },
5920 {
5921 .name = "swap.max",
5922 .flags = CFTYPE_NOT_ON_ROOT,
5923 .seq_show = swap_max_show,
5924 .write = swap_max_write,
5925 },
5926 { } /* terminate */
5927};
5928
Johannes Weiner21afa382015-02-11 15:26:36 -08005929static struct cftype memsw_cgroup_files[] = {
5930 {
5931 .name = "memsw.usage_in_bytes",
5932 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
5933 .read_u64 = mem_cgroup_read_u64,
5934 },
5935 {
5936 .name = "memsw.max_usage_in_bytes",
5937 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
5938 .write = mem_cgroup_reset,
5939 .read_u64 = mem_cgroup_read_u64,
5940 },
5941 {
5942 .name = "memsw.limit_in_bytes",
5943 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
5944 .write = mem_cgroup_write,
5945 .read_u64 = mem_cgroup_read_u64,
5946 },
5947 {
5948 .name = "memsw.failcnt",
5949 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
5950 .write = mem_cgroup_reset,
5951 .read_u64 = mem_cgroup_read_u64,
5952 },
5953 { }, /* terminate */
5954};
5955
5956static int __init mem_cgroup_swap_init(void)
5957{
5958 if (!mem_cgroup_disabled() && really_do_swap_account) {
5959 do_swap_account = 1;
Vladimir Davydov37e84352016-01-20 15:02:56 -08005960 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
5961 swap_files));
Johannes Weiner21afa382015-02-11 15:26:36 -08005962 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
5963 memsw_cgroup_files));
5964 }
5965 return 0;
5966}
5967subsys_initcall(mem_cgroup_swap_init);
5968
5969#endif /* CONFIG_MEMCG_SWAP */