blob: b822e158b319e8f2f02ecbfe76c31b6466be51f1 [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Johannes Weiner1575e682015-04-14 15:44:51 -070017 * Native page reclaim
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
22 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080023 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
27 *
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
32 */
33
Johannes Weiner3e32cb22014-12-10 15:42:31 -080034#include <linux/page_counter.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080035#include <linux/memcontrol.h>
36#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080037#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080038#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080039#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080040#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080041#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080043#include <linux/bit_spinlock.h>
44#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070045#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040046#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080047#include <linux/mutex.h>
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -070048#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070049#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080051#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080052#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080053#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050054#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080055#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080056#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080057#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070058#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070059#include <linux/mm_inline.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080060#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080061#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070062#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070063#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050064#include <linux/file.h>
Tejun Heob23afb92015-11-05 18:46:11 -080065#include <linux/tracehook.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080066#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000067#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070068#include <net/ip.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080069#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080070
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080071#include <linux/uaccess.h>
Balbir Singh8697d332008-02-07 00:13:59 -080072
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070073#include <trace/events/vmscan.h>
74
Tejun Heo073219e2014-02-08 10:36:58 -050075struct cgroup_subsys memory_cgrp_subsys __read_mostly;
76EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080077
Johannes Weiner7d828602016-01-14 15:20:56 -080078struct mem_cgroup *root_mem_cgroup __read_mostly;
79
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070080#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh8cdea7c2008-02-07 00:13:50 -080081
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080082/* Socket memory accounting disabled? */
83static bool cgroup_memory_nosocket;
84
Vladimir Davydov04823c82016-01-20 15:02:38 -080085/* Kernel memory accounting disabled? */
86static bool cgroup_memory_nokmem;
87
Johannes Weiner21afa382015-02-11 15:26:36 -080088/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070089#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080090int do_swap_account __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080091#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070092#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080093#endif
94
Johannes Weiner7941d212016-01-14 15:21:23 -080095/* Whether legacy memory+swap accounting is active */
96static bool do_memsw_account(void)
97{
98 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account;
99}
100
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700101static const char * const mem_cgroup_stat_names[] = {
102 "cache",
103 "rss",
David Rientjesb070e652013-05-07 16:18:09 -0700104 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700105 "mapped_file",
Greg Thelenc4843a72015-05-22 17:13:16 -0400106 "dirty",
Sha Zhengju3ea67d02013-09-12 15:13:53 -0700107 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700108 "swap",
109};
110
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700111static const char * const mem_cgroup_events_names[] = {
112 "pgpgin",
113 "pgpgout",
114 "pgfault",
115 "pgmajfault",
116};
117
Sha Zhengju58cf1882013-02-22 16:32:05 -0800118static const char * const mem_cgroup_lru_names[] = {
119 "inactive_anon",
120 "active_anon",
121 "inactive_file",
122 "active_file",
123 "unevictable",
124};
125
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700126#define THRESHOLDS_EVENTS_TARGET 128
127#define SOFTLIMIT_EVENTS_TARGET 1024
128#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700129
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700130/*
131 * Cgroups above their limits are maintained in a RB-Tree, independent of
132 * their hierarchy representation
133 */
134
Mel Gormanef8f2322016-07-28 15:46:05 -0700135struct mem_cgroup_tree_per_node {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700136 struct rb_root rb_root;
137 spinlock_t lock;
138};
139
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700140struct mem_cgroup_tree {
141 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
142};
143
144static struct mem_cgroup_tree soft_limit_tree __read_mostly;
145
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700146/* for OOM */
147struct mem_cgroup_eventfd_list {
148 struct list_head list;
149 struct eventfd_ctx *eventfd;
150};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800151
Tejun Heo79bd9812013-11-22 18:20:42 -0500152/*
153 * cgroup_event represents events which userspace want to receive.
154 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500155struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500156 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500157 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500158 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500159 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500160 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500161 * eventfd to signal userspace about the event.
162 */
163 struct eventfd_ctx *eventfd;
164 /*
165 * Each of these stored in a list by the cgroup.
166 */
167 struct list_head list;
168 /*
Tejun Heofba94802013-11-22 18:20:43 -0500169 * register_event() callback will be used to add new userspace
170 * waiter for changes related to this event. Use eventfd_signal()
171 * on eventfd to send notification to userspace.
172 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500173 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500174 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500175 /*
176 * unregister_event() callback will be called when userspace closes
177 * the eventfd or on cgroup removing. This callback must be set,
178 * if you want provide notification functionality.
179 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500180 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500181 struct eventfd_ctx *eventfd);
182 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500183 * All fields below needed to unregister event when
184 * userspace closes eventfd.
185 */
186 poll_table pt;
187 wait_queue_head_t *wqh;
188 wait_queue_t wait;
189 struct work_struct remove;
190};
191
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700192static void mem_cgroup_threshold(struct mem_cgroup *memcg);
193static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800194
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800195/* Stuffs for move charges at task migration. */
196/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800197 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800198 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800199#define MOVE_ANON 0x1U
200#define MOVE_FILE 0x2U
201#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800202
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800203/* "mc" and its members are protected by cgroup_mutex */
204static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800205 spinlock_t lock; /* for from, to */
Tejun Heo264a0ae2016-04-21 19:09:02 -0400206 struct mm_struct *mm;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800207 struct mem_cgroup *from;
208 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800209 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800210 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800211 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800212 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800213 struct task_struct *moving_task; /* a task moving charges */
214 wait_queue_head_t waitq; /* a waitq for other context */
215} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700216 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800217 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
218};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800219
Balbir Singh4e416952009-09-23 15:56:39 -0700220/*
221 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
222 * limit reclaim to prevent infinite loops, if they ever occur.
223 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700224#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700225#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700226
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800227enum charge_type {
228 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700229 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800230 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700231 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700232 NR_CHARGE_TYPE,
233};
234
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800235/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800236enum res_type {
237 _MEM,
238 _MEMSWAP,
239 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800240 _KMEM,
Vladimir Davydovd55f90b2016-01-20 15:02:44 -0800241 _TCP,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800242};
243
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700244#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
245#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800246#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700247/* Used for OOM nofiier */
248#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800249
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700250/* Some nice accessors for the vmpressure. */
251struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
252{
253 if (!memcg)
254 memcg = root_mem_cgroup;
255 return &memcg->vmpressure;
256}
257
258struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
259{
260 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
261}
262
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700263static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
264{
265 return (memcg == root_mem_cgroup);
266}
267
Johannes Weiner127424c2016-01-20 15:02:32 -0800268#ifndef CONFIG_SLOB
Glauber Costa55007d82012-12-18 14:22:38 -0800269/*
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800270 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800271 * The main reason for not using cgroup id for this:
272 * this works better in sparse environments, where we have a lot of memcgs,
273 * but only a few kmem-limited. Or also, if we have, for instance, 200
274 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
275 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800276 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800277 * The current size of the caches array is stored in memcg_nr_cache_ids. It
278 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800279 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800280static DEFINE_IDA(memcg_cache_ida);
281int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800282
Vladimir Davydov05257a12015-02-12 14:59:01 -0800283/* Protects memcg_nr_cache_ids */
284static DECLARE_RWSEM(memcg_cache_ids_sem);
285
286void memcg_get_cache_ids(void)
287{
288 down_read(&memcg_cache_ids_sem);
289}
290
291void memcg_put_cache_ids(void)
292{
293 up_read(&memcg_cache_ids_sem);
294}
295
Glauber Costa55007d82012-12-18 14:22:38 -0800296/*
297 * MIN_SIZE is different than 1, because we would like to avoid going through
298 * the alloc/free process all the time. In a small machine, 4 kmem-limited
299 * cgroups is a reasonable guess. In the future, it could be a parameter or
300 * tunable, but that is strictly not necessary.
301 *
Li Zefanb8627832013-09-23 16:56:47 +0800302 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800303 * this constant directly from cgroup, but it is understandable that this is
304 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800305 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800306 * increase ours as well if it increases.
307 */
308#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800309#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800310
Glauber Costad7f25f82012-12-18 14:22:40 -0800311/*
312 * A lot of the calls to the cache allocation functions are expected to be
313 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
314 * conditional to this static branch, we'll have to allow modules that does
315 * kmem_cache_alloc and the such to see this symbol as well
316 */
Johannes Weineref129472016-01-14 15:21:34 -0800317DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
Glauber Costad7f25f82012-12-18 14:22:40 -0800318EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800319
Johannes Weiner127424c2016-01-20 15:02:32 -0800320#endif /* !CONFIG_SLOB */
Glauber Costaa8964b92012-12-18 14:22:09 -0800321
Tejun Heoad7fa852015-05-27 20:00:02 -0400322/**
323 * mem_cgroup_css_from_page - css of the memcg associated with a page
324 * @page: page of interest
325 *
326 * If memcg is bound to the default hierarchy, css of the memcg associated
327 * with @page is returned. The returned css remains associated with @page
328 * until it is released.
329 *
330 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
331 * is returned.
Tejun Heoad7fa852015-05-27 20:00:02 -0400332 */
333struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
334{
335 struct mem_cgroup *memcg;
336
Tejun Heoad7fa852015-05-27 20:00:02 -0400337 memcg = page->mem_cgroup;
338
Tejun Heo9e10a132015-09-18 11:56:28 -0400339 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400340 memcg = root_mem_cgroup;
341
Tejun Heoad7fa852015-05-27 20:00:02 -0400342 return &memcg->css;
343}
344
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700345/**
346 * page_cgroup_ino - return inode number of the memcg a page is charged to
347 * @page: the page
348 *
349 * Look up the closest online ancestor of the memory cgroup @page is charged to
350 * and return its inode number or 0 if @page is not charged to any cgroup. It
351 * is safe to call this function without holding a reference to @page.
352 *
353 * Note, this function is inherently racy, because there is nothing to prevent
354 * the cgroup inode from getting torn down and potentially reallocated a moment
355 * after page_cgroup_ino() returns, so it only should be used by callers that
356 * do not care (such as procfs interfaces).
357 */
358ino_t page_cgroup_ino(struct page *page)
359{
360 struct mem_cgroup *memcg;
361 unsigned long ino = 0;
362
363 rcu_read_lock();
364 memcg = READ_ONCE(page->mem_cgroup);
365 while (memcg && !(memcg->css.flags & CSS_ONLINE))
366 memcg = parent_mem_cgroup(memcg);
367 if (memcg)
368 ino = cgroup_ino(memcg->css.cgroup);
369 rcu_read_unlock();
370 return ino;
371}
372
Mel Gormanef8f2322016-07-28 15:46:05 -0700373static struct mem_cgroup_per_node *
374mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700375{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700376 int nid = page_to_nid(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700377
Mel Gormanef8f2322016-07-28 15:46:05 -0700378 return memcg->nodeinfo[nid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700379}
380
Mel Gormanef8f2322016-07-28 15:46:05 -0700381static struct mem_cgroup_tree_per_node *
382soft_limit_tree_node(int nid)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700383{
Mel Gormanef8f2322016-07-28 15:46:05 -0700384 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700385}
386
Mel Gormanef8f2322016-07-28 15:46:05 -0700387static struct mem_cgroup_tree_per_node *
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700388soft_limit_tree_from_page(struct page *page)
389{
390 int nid = page_to_nid(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700391
Mel Gormanef8f2322016-07-28 15:46:05 -0700392 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700393}
394
Mel Gormanef8f2322016-07-28 15:46:05 -0700395static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz,
396 struct mem_cgroup_tree_per_node *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800397 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700398{
399 struct rb_node **p = &mctz->rb_root.rb_node;
400 struct rb_node *parent = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700401 struct mem_cgroup_per_node *mz_node;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700402
403 if (mz->on_tree)
404 return;
405
406 mz->usage_in_excess = new_usage_in_excess;
407 if (!mz->usage_in_excess)
408 return;
409 while (*p) {
410 parent = *p;
Mel Gormanef8f2322016-07-28 15:46:05 -0700411 mz_node = rb_entry(parent, struct mem_cgroup_per_node,
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700412 tree_node);
413 if (mz->usage_in_excess < mz_node->usage_in_excess)
414 p = &(*p)->rb_left;
415 /*
416 * We can't avoid mem cgroups that are over their soft
417 * limit by the same amount
418 */
419 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
420 p = &(*p)->rb_right;
421 }
422 rb_link_node(&mz->tree_node, parent, p);
423 rb_insert_color(&mz->tree_node, &mctz->rb_root);
424 mz->on_tree = true;
425}
426
Mel Gormanef8f2322016-07-28 15:46:05 -0700427static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
428 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700429{
430 if (!mz->on_tree)
431 return;
432 rb_erase(&mz->tree_node, &mctz->rb_root);
433 mz->on_tree = false;
434}
435
Mel Gormanef8f2322016-07-28 15:46:05 -0700436static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
437 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700438{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700439 unsigned long flags;
440
441 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700442 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700443 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700444}
445
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800446static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
447{
448 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700449 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800450 unsigned long excess = 0;
451
452 if (nr_pages > soft_limit)
453 excess = nr_pages - soft_limit;
454
455 return excess;
456}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700457
458static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
459{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800460 unsigned long excess;
Mel Gormanef8f2322016-07-28 15:46:05 -0700461 struct mem_cgroup_per_node *mz;
462 struct mem_cgroup_tree_per_node *mctz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700463
Jianyu Zhane2318752014-06-06 14:38:20 -0700464 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700465 /*
466 * Necessary to update all ancestors when hierarchy is used.
467 * because their event counter is not touched.
468 */
469 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700470 mz = mem_cgroup_page_nodeinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800471 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700472 /*
473 * We have to update the tree if mz is on RB-tree or
474 * mem is over its softlimit.
475 */
476 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700477 unsigned long flags;
478
479 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700480 /* if on-tree, remove it */
481 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700482 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700483 /*
484 * Insert again. mz->usage_in_excess will be updated.
485 * If excess is 0, no tree ops.
486 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700487 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700488 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700489 }
490 }
491}
492
493static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
494{
Mel Gormanef8f2322016-07-28 15:46:05 -0700495 struct mem_cgroup_tree_per_node *mctz;
496 struct mem_cgroup_per_node *mz;
497 int nid;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700498
Jianyu Zhane2318752014-06-06 14:38:20 -0700499 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700500 mz = mem_cgroup_nodeinfo(memcg, nid);
501 mctz = soft_limit_tree_node(nid);
502 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700503 }
504}
505
Mel Gormanef8f2322016-07-28 15:46:05 -0700506static struct mem_cgroup_per_node *
507__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700508{
509 struct rb_node *rightmost = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700510 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700511
512retry:
513 mz = NULL;
514 rightmost = rb_last(&mctz->rb_root);
515 if (!rightmost)
516 goto done; /* Nothing to reclaim from */
517
Mel Gormanef8f2322016-07-28 15:46:05 -0700518 mz = rb_entry(rightmost, struct mem_cgroup_per_node, tree_node);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700519 /*
520 * Remove the node now but someone else can add it back,
521 * we will to add it back at the end of reclaim to its correct
522 * position in the tree.
523 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700524 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800525 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400526 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700527 goto retry;
528done:
529 return mz;
530}
531
Mel Gormanef8f2322016-07-28 15:46:05 -0700532static struct mem_cgroup_per_node *
533mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700534{
Mel Gormanef8f2322016-07-28 15:46:05 -0700535 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700536
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700537 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700538 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700539 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700540 return mz;
541}
542
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700543/*
Greg Thelen484ebb32015-10-01 15:37:05 -0700544 * Return page count for single (non recursive) @memcg.
545 *
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700546 * Implementation Note: reading percpu statistics for memcg.
547 *
548 * Both of vmstat[] and percpu_counter has threshold and do periodic
549 * synchronization to implement "quick" read. There are trade-off between
550 * reading cost and precision of value. Then, we may have a chance to implement
Greg Thelen484ebb32015-10-01 15:37:05 -0700551 * a periodic synchronization of counter in memcg's counter.
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700552 *
553 * But this _read() function is used for user interface now. The user accounts
554 * memory usage by memory cgroup and he _always_ requires exact value because
555 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
556 * have to visit all online cpus and make sum. So, for now, unnecessary
557 * synchronization is not implemented. (just implemented for cpu hotplug)
558 *
559 * If there are kernel internal actions which can make use of some not-exact
560 * value, and reading all cpu value can be performance bottleneck in some
Greg Thelen484ebb32015-10-01 15:37:05 -0700561 * common workload, threshold and synchronization as vmstat[] should be
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700562 * implemented.
563 */
Greg Thelen484ebb32015-10-01 15:37:05 -0700564static unsigned long
565mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800566{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700567 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800568 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800569
Greg Thelen484ebb32015-10-01 15:37:05 -0700570 /* Per-cpu values can be negative, use a signed accumulator */
Tejun Heo733a5722015-05-22 18:23:18 -0400571 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700572 val += per_cpu(memcg->stat->count[idx], cpu);
Greg Thelen484ebb32015-10-01 15:37:05 -0700573 /*
574 * Summing races with updates, so val may be negative. Avoid exposing
575 * transient negative values.
576 */
577 if (val < 0)
578 val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800579 return val;
580}
581
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700582static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700583 enum mem_cgroup_events_index idx)
584{
585 unsigned long val = 0;
586 int cpu;
587
Tejun Heo733a5722015-05-22 18:23:18 -0400588 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700589 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700590 return val;
591}
592
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700593static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700594 struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800595 bool compound, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800596{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700597 /*
598 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
599 * counted as CACHE even if it's on ANON LRU.
600 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700601 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700602 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700603 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800604 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700605 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700606 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700607
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800608 if (compound) {
609 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
David Rientjesb070e652013-05-07 16:18:09 -0700610 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
611 nr_pages);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800612 }
David Rientjesb070e652013-05-07 16:18:09 -0700613
KAMEZAWA Hiroyukie401f1762011-01-20 14:44:23 -0800614 /* pagein of a big page is an event. So, ignore page size */
615 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700616 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800617 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700618 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800619 nr_pages = -nr_pages; /* for event */
620 }
KAMEZAWA Hiroyukie401f1762011-01-20 14:44:23 -0800621
Johannes Weiner13114712012-05-29 15:07:07 -0700622 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800623}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800624
Vladimir Davydov0a6b76d2016-03-17 14:18:42 -0700625unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
626 int nid, unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700627{
Michal Hockob4536f0c82017-01-10 16:58:04 -0800628 struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg);
Jianyu Zhane2318752014-06-06 14:38:20 -0700629 unsigned long nr = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -0700630 enum lru_list lru;
Ying Han889976d2011-05-26 16:25:33 -0700631
Jianyu Zhane2318752014-06-06 14:38:20 -0700632 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700633
Mel Gormanef8f2322016-07-28 15:46:05 -0700634 for_each_lru(lru) {
635 if (!(BIT(lru) & lru_mask))
636 continue;
Michal Hockob4536f0c82017-01-10 16:58:04 -0800637 nr += mem_cgroup_get_lru_size(lruvec, lru);
Jianyu Zhane2318752014-06-06 14:38:20 -0700638 }
639 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700640}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700641
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700642static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700643 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800644{
Jianyu Zhane2318752014-06-06 14:38:20 -0700645 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700646 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800647
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800648 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700649 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
650 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800651}
652
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800653static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
654 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800655{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700656 unsigned long val, next;
657
Johannes Weiner13114712012-05-29 15:07:07 -0700658 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700659 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700660 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800661 if ((long)next - (long)val < 0) {
662 switch (target) {
663 case MEM_CGROUP_TARGET_THRESH:
664 next = val + THRESHOLDS_EVENTS_TARGET;
665 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700666 case MEM_CGROUP_TARGET_SOFTLIMIT:
667 next = val + SOFTLIMIT_EVENTS_TARGET;
668 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800669 case MEM_CGROUP_TARGET_NUMAINFO:
670 next = val + NUMAINFO_EVENTS_TARGET;
671 break;
672 default:
673 break;
674 }
675 __this_cpu_write(memcg->stat->targets[target], next);
676 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700677 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800678 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800679}
680
681/*
682 * Check events in order.
683 *
684 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700685static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800686{
687 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800688 if (unlikely(mem_cgroup_event_ratelimit(memcg,
689 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700690 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800691 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800692
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700693 do_softlimit = mem_cgroup_event_ratelimit(memcg,
694 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700695#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800696 do_numainfo = mem_cgroup_event_ratelimit(memcg,
697 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700698#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800699 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700700 if (unlikely(do_softlimit))
701 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800702#if MAX_NUMNODES > 1
703 if (unlikely(do_numainfo))
704 atomic_inc(&memcg->numainfo_events);
705#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700706 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800707}
708
Balbir Singhcf475ad2008-04-29 01:00:16 -0700709struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800710{
Balbir Singh31a78f22008-09-28 23:09:31 +0100711 /*
712 * mm_update_next_owner() may clear mm->owner to NULL
713 * if it races with swapoff, page migration, etc.
714 * So this can be called with p == NULL.
715 */
716 if (unlikely(!p))
717 return NULL;
718
Tejun Heo073219e2014-02-08 10:36:58 -0500719 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800720}
Michal Hocko33398cf2015-09-08 15:01:02 -0700721EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800722
Johannes Weinerdf381972014-04-07 15:37:43 -0700723static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800724{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700725 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700726
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800727 rcu_read_lock();
728 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700729 /*
730 * Page cache insertions can happen withou an
731 * actual mm context, e.g. during disk probing
732 * on boot, loopback IO, acct() writes etc.
733 */
734 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700735 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700736 else {
737 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
738 if (unlikely(!memcg))
739 memcg = root_mem_cgroup;
740 }
Tejun Heoec903c02014-05-13 12:11:01 -0400741 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800742 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700743 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800744}
745
Johannes Weiner56600482012-01-12 17:17:59 -0800746/**
747 * mem_cgroup_iter - iterate over memory cgroup hierarchy
748 * @root: hierarchy root
749 * @prev: previously returned memcg, NULL on first invocation
750 * @reclaim: cookie for shared reclaim walks, NULL for full walks
751 *
752 * Returns references to children of the hierarchy below @root, or
753 * @root itself, or %NULL after a full round-trip.
754 *
755 * Caller must pass the return value in @prev on subsequent
756 * invocations for reference counting, or use mem_cgroup_iter_break()
757 * to cancel a hierarchy walk before the round-trip is complete.
758 *
759 * Reclaimers can specify a zone and a priority level in @reclaim to
760 * divide up the memcgs in the hierarchy among all concurrent
761 * reclaimers operating on the same zone and priority.
762 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700763struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800764 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700765 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700766{
Michal Hocko33398cf2015-09-08 15:01:02 -0700767 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800768 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800769 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800770 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700771
Andrew Morton694fbc02013-09-24 15:27:37 -0700772 if (mem_cgroup_disabled())
773 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800774
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700775 if (!root)
776 root = root_mem_cgroup;
777
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800778 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800779 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800780
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800781 if (!root->use_hierarchy && root != root_mem_cgroup) {
782 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800783 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700784 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800785 }
786
Michal Hocko542f85f2013-04-29 15:07:15 -0700787 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800788
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800789 if (reclaim) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700790 struct mem_cgroup_per_node *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800791
Mel Gormanef8f2322016-07-28 15:46:05 -0700792 mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800793 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700794
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800795 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700796 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800797
Vladimir Davydov6df38682015-12-29 14:54:10 -0800798 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -0700799 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800800 if (!pos || css_tryget(&pos->css))
801 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800802 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800803 * css reference reached zero, so iter->position will
804 * be cleared by ->css_released. However, we should not
805 * rely on this happening soon, because ->css_released
806 * is called from a work queue, and by busy-waiting we
807 * might block it. So we clear iter->position right
808 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800809 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800810 (void)cmpxchg(&iter->position, pos, NULL);
811 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800812 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800813
814 if (pos)
815 css = &pos->css;
816
817 for (;;) {
818 css = css_next_descendant_pre(css, &root->css);
819 if (!css) {
820 /*
821 * Reclaimers share the hierarchy walk, and a
822 * new one might jump in right at the end of
823 * the hierarchy - make sure they see at least
824 * one group and restart from the beginning.
825 */
826 if (!prev)
827 continue;
828 break;
829 }
830
831 /*
832 * Verify the css and acquire a reference. The root
833 * is provided by the caller, so we know it's alive
834 * and kicking, and don't take an extra reference.
835 */
836 memcg = mem_cgroup_from_css(css);
837
838 if (css == &root->css)
839 break;
840
Johannes Weiner0b8f73e2016-01-20 15:02:53 -0800841 if (css_tryget(css))
842 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800843
844 memcg = NULL;
845 }
846
847 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800848 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800849 * The position could have already been updated by a competing
850 * thread, so check that the value hasn't changed since we read
851 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800852 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800853 (void)cmpxchg(&iter->position, pos, memcg);
854
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800855 if (pos)
856 css_put(&pos->css);
857
858 if (!memcg)
859 iter->generation++;
860 else if (!prev)
861 reclaim->generation = iter->generation;
862 }
863
Michal Hocko542f85f2013-04-29 15:07:15 -0700864out_unlock:
865 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800866out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700867 if (prev && prev != root)
868 css_put(&prev->css);
869
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800870 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700871}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800872
Johannes Weiner56600482012-01-12 17:17:59 -0800873/**
874 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
875 * @root: hierarchy root
876 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
877 */
878void mem_cgroup_iter_break(struct mem_cgroup *root,
879 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800880{
881 if (!root)
882 root = root_mem_cgroup;
883 if (prev && prev != root)
884 css_put(&prev->css);
885}
886
Vladimir Davydov6df38682015-12-29 14:54:10 -0800887static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
888{
889 struct mem_cgroup *memcg = dead_memcg;
890 struct mem_cgroup_reclaim_iter *iter;
Mel Gormanef8f2322016-07-28 15:46:05 -0700891 struct mem_cgroup_per_node *mz;
892 int nid;
Vladimir Davydov6df38682015-12-29 14:54:10 -0800893 int i;
894
895 while ((memcg = parent_mem_cgroup(memcg))) {
896 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700897 mz = mem_cgroup_nodeinfo(memcg, nid);
898 for (i = 0; i <= DEF_PRIORITY; i++) {
899 iter = &mz->iter[i];
900 cmpxchg(&iter->position,
901 dead_memcg, NULL);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800902 }
903 }
904 }
905}
906
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700907/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800908 * Iteration constructs for visiting all cgroups (under a tree). If
909 * loops are exited prematurely (break), mem_cgroup_iter_break() must
910 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700911 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800912#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800913 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800914 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800915 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700916
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800917#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800918 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800919 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800920 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700921
Johannes Weiner925b7672012-01-12 17:18:15 -0800922/**
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700923 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
924 * @memcg: hierarchy root
925 * @fn: function to call for each task
926 * @arg: argument passed to @fn
927 *
928 * This function iterates over tasks attached to @memcg or to any of its
929 * descendants and calls @fn for each task. If @fn returns a non-zero
930 * value, the function breaks the iteration loop and returns the value.
931 * Otherwise, it will iterate over all tasks and return 0.
932 *
933 * This function must not be called for the root memory cgroup.
934 */
935int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
936 int (*fn)(struct task_struct *, void *), void *arg)
937{
938 struct mem_cgroup *iter;
939 int ret = 0;
940
941 BUG_ON(memcg == root_mem_cgroup);
942
943 for_each_mem_cgroup_tree(iter, memcg) {
944 struct css_task_iter it;
945 struct task_struct *task;
946
947 css_task_iter_start(&iter->css, &it);
948 while (!ret && (task = css_task_iter_next(&it)))
949 ret = fn(task, arg);
950 css_task_iter_end(&it);
951 if (ret) {
952 mem_cgroup_iter_break(memcg, iter);
953 break;
954 }
955 }
956 return ret;
957}
958
959/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800960 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -0800961 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700962 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800963 *
964 * This function is only safe when following the LRU page isolation
965 * and putback protocol: the LRU lock must be held, and the page must
966 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -0700967 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700968struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat)
Minchan Kim3f58a822011-03-22 16:32:53 -0700969{
Mel Gormanef8f2322016-07-28 15:46:05 -0700970 struct mem_cgroup_per_node *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -0800971 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800972 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800973
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800974 if (mem_cgroup_disabled()) {
Mel Gorman599d0c92016-07-28 15:45:31 -0700975 lruvec = &pgdat->lruvec;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800976 goto out;
977 }
Christoph Lameterb69408e2008-10-18 20:26:14 -0700978
Johannes Weiner1306a852014-12-10 15:44:52 -0800979 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800980 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800981 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -0800982 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -0800983 */
Johannes Weiner29833312014-12-10 15:44:02 -0800984 if (!memcg)
985 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800986
Mel Gormanef8f2322016-07-28 15:46:05 -0700987 mz = mem_cgroup_page_nodeinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800988 lruvec = &mz->lruvec;
989out:
990 /*
991 * Since a node can be onlined after the mem_cgroup was created,
992 * we have to be prepared to initialize lruvec->zone here;
993 * and if offlined then reonlined, we need to reinitialize it.
994 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700995 if (unlikely(lruvec->pgdat != pgdat))
996 lruvec->pgdat = pgdat;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800997 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -0800998}
999
1000/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001001 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1002 * @lruvec: mem_cgroup per zone lru vector
1003 * @lru: index of lru list the page is sitting on
Michal Hockob4536f0c82017-01-10 16:58:04 -08001004 * @zid: zone id of the accounted pages
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001005 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001006 *
Hugh Dickinsca707232016-05-19 17:12:35 -07001007 * This function must be called under lru_lock, just before a page is added
1008 * to or just after a page is removed from an lru list (that ordering being
1009 * so as to allow it to check that lru_size 0 is consistent with list_empty).
Johannes Weiner925b7672012-01-12 17:18:15 -08001010 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001011void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
Michal Hockob4536f0c82017-01-10 16:58:04 -08001012 int zid, int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001013{
Mel Gormanef8f2322016-07-28 15:46:05 -07001014 struct mem_cgroup_per_node *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001015 unsigned long *lru_size;
Hugh Dickinsca707232016-05-19 17:12:35 -07001016 long size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001017
1018 if (mem_cgroup_disabled())
1019 return;
1020
Mel Gormanef8f2322016-07-28 15:46:05 -07001021 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
Michal Hockob4536f0c82017-01-10 16:58:04 -08001022 lru_size = &mz->lru_zone_size[zid][lru];
Hugh Dickinsca707232016-05-19 17:12:35 -07001023
1024 if (nr_pages < 0)
1025 *lru_size += nr_pages;
1026
1027 size = *lru_size;
Michal Hockob4536f0c82017-01-10 16:58:04 -08001028 if (WARN_ONCE(size < 0,
1029 "%s(%p, %d, %d): lru_size %ld\n",
1030 __func__, lruvec, lru, nr_pages, size)) {
Hugh Dickinsca707232016-05-19 17:12:35 -07001031 VM_BUG_ON(1);
1032 *lru_size = 0;
1033 }
1034
1035 if (nr_pages > 0)
1036 *lru_size += nr_pages;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001037}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001038
Johannes Weiner2314b422014-12-10 15:44:33 -08001039bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001040{
Johannes Weiner2314b422014-12-10 15:44:33 -08001041 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001042 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001043 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001044
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001045 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001046 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001047 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001048 task_unlock(p);
1049 } else {
1050 /*
1051 * All threads may have already detached their mm's, but the oom
1052 * killer still needs to detect if they have already been oom
1053 * killed to prevent needlessly killing additional tasks.
1054 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001055 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001056 task_memcg = mem_cgroup_from_task(task);
1057 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001058 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001059 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001060 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1061 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001062 return ret;
1063}
1064
Johannes Weiner19942822011-02-01 15:52:43 -08001065/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001066 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001067 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001068 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001069 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001070 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001071 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001072static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001073{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001074 unsigned long margin = 0;
1075 unsigned long count;
1076 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001077
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001078 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001079 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001080 if (count < limit)
1081 margin = limit - count;
1082
Johannes Weiner7941d212016-01-14 15:21:23 -08001083 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001084 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001085 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001086 if (count <= limit)
1087 margin = min(margin, limit - count);
Li RongQingcbedbac2016-05-27 14:27:43 -07001088 else
1089 margin = 0;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001090 }
1091
1092 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001093}
1094
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001095/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001096 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001097 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001098 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1099 * moving cgroups. This is for waiting at high-memory pressure
1100 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001101 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001102static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001103{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001104 struct mem_cgroup *from;
1105 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001106 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001107 /*
1108 * Unlike task_move routines, we access mc.to, mc.from not under
1109 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1110 */
1111 spin_lock(&mc.lock);
1112 from = mc.from;
1113 to = mc.to;
1114 if (!from)
1115 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001116
Johannes Weiner2314b422014-12-10 15:44:33 -08001117 ret = mem_cgroup_is_descendant(from, memcg) ||
1118 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001119unlock:
1120 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001121 return ret;
1122}
1123
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001124static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001125{
1126 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001127 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001128 DEFINE_WAIT(wait);
1129 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1130 /* moving charge context might have finished. */
1131 if (mc.moving_task)
1132 schedule();
1133 finish_wait(&mc.waitq, &wait);
1134 return true;
1135 }
1136 }
1137 return false;
1138}
1139
Sha Zhengju58cf1882013-02-22 16:32:05 -08001140#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001141/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001142 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001143 * @memcg: The memory cgroup that went over limit
1144 * @p: Task that is going to be killed
1145 *
1146 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1147 * enabled
1148 */
1149void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1150{
Sha Zhengju58cf1882013-02-22 16:32:05 -08001151 struct mem_cgroup *iter;
1152 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001153
Balbir Singhe2224322009-04-02 16:57:39 -07001154 rcu_read_lock();
1155
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001156 if (p) {
1157 pr_info("Task in ");
1158 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1159 pr_cont(" killed as a result of limit of ");
1160 } else {
1161 pr_info("Memory limit reached of cgroup ");
1162 }
1163
Tejun Heoe61734c2014-02-12 09:29:50 -05001164 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001165 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001166
Balbir Singhe2224322009-04-02 16:57:39 -07001167 rcu_read_unlock();
1168
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001169 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1170 K((u64)page_counter_read(&memcg->memory)),
1171 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1172 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1173 K((u64)page_counter_read(&memcg->memsw)),
1174 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1175 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1176 K((u64)page_counter_read(&memcg->kmem)),
1177 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001178
1179 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001180 pr_info("Memory cgroup stats for ");
1181 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001182 pr_cont(":");
1183
1184 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08001185 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Sha Zhengju58cf1882013-02-22 16:32:05 -08001186 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07001187 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
Sha Zhengju58cf1882013-02-22 16:32:05 -08001188 K(mem_cgroup_read_stat(iter, i)));
1189 }
1190
1191 for (i = 0; i < NR_LRU_LISTS; i++)
1192 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1193 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1194
1195 pr_cont("\n");
1196 }
Balbir Singhe2224322009-04-02 16:57:39 -07001197}
1198
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001199/*
1200 * This function returns the number of memcg under hierarchy tree. Returns
1201 * 1(self count) if no children.
1202 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001203static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001204{
1205 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001206 struct mem_cgroup *iter;
1207
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001208 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001209 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001210 return num;
1211}
1212
Balbir Singh6d61ef42009-01-07 18:08:06 -08001213/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001214 * Return the memory (and swap, if configured) limit for a memcg.
1215 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001216unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001217{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001218 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001219
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001220 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001221 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001222 unsigned long memsw_limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001223 unsigned long swap_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001224
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001225 memsw_limit = memcg->memsw.limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001226 swap_limit = memcg->swap.limit;
1227 swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1228 limit = min(limit + swap_limit, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001229 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001230 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001231}
1232
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001233static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001234 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001235{
David Rientjes6e0fc462015-09-08 15:00:36 -07001236 struct oom_control oc = {
1237 .zonelist = NULL,
1238 .nodemask = NULL,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001239 .memcg = memcg,
David Rientjes6e0fc462015-09-08 15:00:36 -07001240 .gfp_mask = gfp_mask,
1241 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001242 };
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001243 bool ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001244
Johannes Weinerdc564012015-06-24 16:57:19 -07001245 mutex_lock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001246 ret = out_of_memory(&oc);
Johannes Weinerdc564012015-06-24 16:57:19 -07001247 mutex_unlock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001248 return ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001249}
1250
Michele Curtiae6e71d2014-12-12 16:56:35 -08001251#if MAX_NUMNODES > 1
1252
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001253/**
1254 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001255 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001256 * @nid: the node ID to be checked.
1257 * @noswap : specify true here if the user wants flle only information.
1258 *
1259 * This function returns whether the specified memcg contains any
1260 * reclaimable pages on a node. Returns true if there are any reclaimable
1261 * pages in the node.
1262 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001263static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001264 int nid, bool noswap)
1265{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001266 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001267 return true;
1268 if (noswap || !total_swap_pages)
1269 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001270 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001271 return true;
1272 return false;
1273
1274}
Ying Han889976d2011-05-26 16:25:33 -07001275
1276/*
1277 * Always updating the nodemask is not very good - even if we have an empty
1278 * list or the wrong list here, we can start from some node and traverse all
1279 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1280 *
1281 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001282static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001283{
1284 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001285 /*
1286 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1287 * pagein/pageout changes since the last update.
1288 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001289 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001290 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001291 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001292 return;
1293
Ying Han889976d2011-05-26 16:25:33 -07001294 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001295 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001296
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001297 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001298
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001299 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1300 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001301 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001302
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001303 atomic_set(&memcg->numainfo_events, 0);
1304 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001305}
1306
1307/*
1308 * Selecting a node where we start reclaim from. Because what we need is just
1309 * reducing usage counter, start from anywhere is O,K. Considering
1310 * memory reclaim from current node, there are pros. and cons.
1311 *
1312 * Freeing memory from current node means freeing memory from a node which
1313 * we'll use or we've used. So, it may make LRU bad. And if several threads
1314 * hit limits, it will see a contention on a node. But freeing from remote
1315 * node means more costs for memory reclaim because of memory latency.
1316 *
1317 * Now, we use round-robin. Better algorithm is welcomed.
1318 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001319int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001320{
1321 int node;
1322
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001323 mem_cgroup_may_update_nodemask(memcg);
1324 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001325
Andrew Morton0edaf862016-05-19 17:10:58 -07001326 node = next_node_in(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001327 /*
Michal Hockofda3d692016-05-19 17:11:34 -07001328 * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1329 * last time it really checked all the LRUs due to rate limiting.
1330 * Fallback to the current node in that case for simplicity.
Ying Han889976d2011-05-26 16:25:33 -07001331 */
1332 if (unlikely(node == MAX_NUMNODES))
1333 node = numa_node_id();
1334
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001335 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001336 return node;
1337}
Ying Han889976d2011-05-26 16:25:33 -07001338#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001339int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001340{
1341 return 0;
1342}
1343#endif
1344
Andrew Morton0608f432013-09-24 15:27:41 -07001345static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
Mel Gormanef8f2322016-07-28 15:46:05 -07001346 pg_data_t *pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001347 gfp_t gfp_mask,
1348 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001349{
Andrew Morton0608f432013-09-24 15:27:41 -07001350 struct mem_cgroup *victim = NULL;
1351 int total = 0;
1352 int loop = 0;
1353 unsigned long excess;
1354 unsigned long nr_scanned;
1355 struct mem_cgroup_reclaim_cookie reclaim = {
Mel Gormanef8f2322016-07-28 15:46:05 -07001356 .pgdat = pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001357 .priority = 0,
1358 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001359
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001360 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001361
Andrew Morton0608f432013-09-24 15:27:41 -07001362 while (1) {
1363 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1364 if (!victim) {
1365 loop++;
1366 if (loop >= 2) {
1367 /*
1368 * If we have not been able to reclaim
1369 * anything, it might because there are
1370 * no reclaimable pages under this hierarchy
1371 */
1372 if (!total)
1373 break;
1374 /*
1375 * We want to do more targeted reclaim.
1376 * excess >> 2 is not to excessive so as to
1377 * reclaim too much, nor too less that we keep
1378 * coming back to reclaim from this cgroup
1379 */
1380 if (total >= (excess >> 2) ||
1381 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1382 break;
1383 }
1384 continue;
1385 }
Mel Gormana9dd0a82016-07-28 15:46:02 -07001386 total += mem_cgroup_shrink_node(victim, gfp_mask, false,
Mel Gormanef8f2322016-07-28 15:46:05 -07001387 pgdat, &nr_scanned);
Andrew Morton0608f432013-09-24 15:27:41 -07001388 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001389 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001390 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001391 }
Andrew Morton0608f432013-09-24 15:27:41 -07001392 mem_cgroup_iter_break(root_memcg, victim);
1393 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001394}
1395
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001396#ifdef CONFIG_LOCKDEP
1397static struct lockdep_map memcg_oom_lock_dep_map = {
1398 .name = "memcg_oom_lock",
1399};
1400#endif
1401
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001402static DEFINE_SPINLOCK(memcg_oom_lock);
1403
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001404/*
1405 * Check OOM-Killer is already running under our hierarchy.
1406 * If someone is running, return false.
1407 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001408static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001409{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001410 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001411
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001412 spin_lock(&memcg_oom_lock);
1413
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001414 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001415 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001416 /*
1417 * this subtree of our hierarchy is already locked
1418 * so we cannot give a lock.
1419 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001420 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001421 mem_cgroup_iter_break(memcg, iter);
1422 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001423 } else
1424 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001425 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001426
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001427 if (failed) {
1428 /*
1429 * OK, we failed to lock the whole subtree so we have
1430 * to clean up what we set up to the failing subtree
1431 */
1432 for_each_mem_cgroup_tree(iter, memcg) {
1433 if (iter == failed) {
1434 mem_cgroup_iter_break(memcg, iter);
1435 break;
1436 }
1437 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001438 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001439 } else
1440 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001441
1442 spin_unlock(&memcg_oom_lock);
1443
1444 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001445}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001446
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001447static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001448{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001449 struct mem_cgroup *iter;
1450
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001451 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001452 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001453 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001454 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001455 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001456}
1457
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001458static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001459{
1460 struct mem_cgroup *iter;
1461
Tejun Heoc2b42d32015-06-24 16:58:23 -07001462 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001463 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001464 iter->under_oom++;
1465 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001466}
1467
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001468static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001469{
1470 struct mem_cgroup *iter;
1471
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001472 /*
1473 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001474 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001475 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001476 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001477 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001478 if (iter->under_oom > 0)
1479 iter->under_oom--;
1480 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001481}
1482
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001483static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1484
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001485struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001486 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001487 wait_queue_t wait;
1488};
1489
1490static int memcg_oom_wake_function(wait_queue_t *wait,
1491 unsigned mode, int sync, void *arg)
1492{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001493 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1494 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001495 struct oom_wait_info *oom_wait_info;
1496
1497 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001498 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001499
Johannes Weiner2314b422014-12-10 15:44:33 -08001500 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1501 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001502 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001503 return autoremove_wake_function(wait, mode, sync, arg);
1504}
1505
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001506static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001507{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001508 /*
1509 * For the following lockless ->under_oom test, the only required
1510 * guarantee is that it must see the state asserted by an OOM when
1511 * this function is called as a result of userland actions
1512 * triggered by the notification of the OOM. This is trivially
1513 * achieved by invoking mem_cgroup_mark_under_oom() before
1514 * triggering notification.
1515 */
1516 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001517 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001518}
1519
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001520static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001521{
Andrew Mortond0db7af2016-06-08 15:33:47 -07001522 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001523 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001524 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001525 * We are in the middle of the charge context here, so we
1526 * don't want to block when potentially sitting on a callstack
1527 * that holds all kinds of filesystem and mm locks.
1528 *
1529 * Also, the caller may handle a failed allocation gracefully
1530 * (like optional page cache readahead) and so an OOM killer
1531 * invocation might not even be necessary.
1532 *
1533 * That's why we don't do anything here except remember the
1534 * OOM context and then deal with it at the end of the page
1535 * fault when the stack is unwound, the locks are released,
1536 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001537 */
Johannes Weiner49426422013-10-16 13:46:59 -07001538 css_get(&memcg->css);
Tejun Heo626ebc42015-11-05 18:46:09 -08001539 current->memcg_in_oom = memcg;
1540 current->memcg_oom_gfp_mask = mask;
1541 current->memcg_oom_order = order;
Johannes Weiner49426422013-10-16 13:46:59 -07001542}
1543
1544/**
1545 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1546 * @handle: actually kill/wait or just clean up the OOM state
1547 *
1548 * This has to be called at the end of a page fault if the memcg OOM
1549 * handler was enabled.
1550 *
1551 * Memcg supports userspace OOM handling where failed allocations must
1552 * sleep on a waitqueue until the userspace task resolves the
1553 * situation. Sleeping directly in the charge context with all kinds
1554 * of locks held is not a good idea, instead we remember an OOM state
1555 * in the task and mem_cgroup_oom_synchronize() has to be called at
1556 * the end of the page fault to complete the OOM handling.
1557 *
1558 * Returns %true if an ongoing memcg OOM situation was detected and
1559 * completed, %false otherwise.
1560 */
1561bool mem_cgroup_oom_synchronize(bool handle)
1562{
Tejun Heo626ebc42015-11-05 18:46:09 -08001563 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001564 struct oom_wait_info owait;
1565 bool locked;
1566
1567 /* OOM is global, do not handle */
1568 if (!memcg)
1569 return false;
1570
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001571 if (!handle)
Johannes Weiner49426422013-10-16 13:46:59 -07001572 goto cleanup;
1573
1574 owait.memcg = memcg;
1575 owait.wait.flags = 0;
1576 owait.wait.func = memcg_oom_wake_function;
1577 owait.wait.private = current;
1578 INIT_LIST_HEAD(&owait.wait.task_list);
1579
1580 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001581 mem_cgroup_mark_under_oom(memcg);
1582
1583 locked = mem_cgroup_oom_trylock(memcg);
1584
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001585 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001586 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001587
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001588 if (locked && !memcg->oom_kill_disable) {
1589 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001590 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001591 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1592 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001593 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001594 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001595 mem_cgroup_unmark_under_oom(memcg);
1596 finish_wait(&memcg_oom_waitq, &owait.wait);
1597 }
1598
1599 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001600 mem_cgroup_oom_unlock(memcg);
1601 /*
1602 * There is no guarantee that an OOM-lock contender
1603 * sees the wakeups triggered by the OOM kill
1604 * uncharges. Wake any sleepers explicitely.
1605 */
1606 memcg_oom_recover(memcg);
1607 }
Johannes Weiner49426422013-10-16 13:46:59 -07001608cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001609 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001610 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001611 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001612}
1613
Johannes Weinerd7365e72014-10-29 14:50:48 -07001614/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001615 * lock_page_memcg - lock a page->mem_cgroup binding
1616 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001617 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001618 * This function protects unlocked LRU pages from being moved to
1619 * another cgroup and stabilizes their page->mem_cgroup binding.
Balbir Singhd69b0422009-06-17 16:26:34 -07001620 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001621void lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001622{
1623 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001624 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001625
Johannes Weiner6de22612015-02-11 15:25:01 -08001626 /*
1627 * The RCU lock is held throughout the transaction. The fast
1628 * path can get away without acquiring the memcg->move_lock
1629 * because page moving starts with an RCU grace period.
Johannes Weiner6de22612015-02-11 15:25:01 -08001630 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001631 rcu_read_lock();
1632
1633 if (mem_cgroup_disabled())
Johannes Weiner62cccb82016-03-15 14:57:22 -07001634 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001635again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001636 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001637 if (unlikely(!memcg))
Johannes Weiner62cccb82016-03-15 14:57:22 -07001638 return;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001639
Qiang Huangbdcbb652014-06-04 16:08:21 -07001640 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001641 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001642
Johannes Weiner6de22612015-02-11 15:25:01 -08001643 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001644 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001645 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001646 goto again;
1647 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001648
1649 /*
1650 * When charge migration first begins, we can have locked and
1651 * unlocked page stat updates happening concurrently. Track
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001652 * the task who has the lock for unlock_page_memcg().
Johannes Weiner6de22612015-02-11 15:25:01 -08001653 */
1654 memcg->move_lock_task = current;
1655 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001656
Johannes Weiner62cccb82016-03-15 14:57:22 -07001657 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001658}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001659EXPORT_SYMBOL(lock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001660
Johannes Weinerd7365e72014-10-29 14:50:48 -07001661/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001662 * unlock_page_memcg - unlock a page->mem_cgroup binding
Johannes Weiner62cccb82016-03-15 14:57:22 -07001663 * @page: the page
Johannes Weinerd7365e72014-10-29 14:50:48 -07001664 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001665void unlock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001666{
Johannes Weiner62cccb82016-03-15 14:57:22 -07001667 struct mem_cgroup *memcg = page->mem_cgroup;
1668
Johannes Weiner6de22612015-02-11 15:25:01 -08001669 if (memcg && memcg->move_lock_task == current) {
1670 unsigned long flags = memcg->move_lock_flags;
1671
1672 memcg->move_lock_task = NULL;
1673 memcg->move_lock_flags = 0;
1674
1675 spin_unlock_irqrestore(&memcg->move_lock, flags);
1676 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001677
Johannes Weinerd7365e72014-10-29 14:50:48 -07001678 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001679}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001680EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001681
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001682/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001683 * size of first charge trial. "32" comes from vmscan.c's magic value.
1684 * TODO: maybe necessary to use big numbers in big irons.
1685 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001686#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001687struct memcg_stock_pcp {
1688 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001689 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001690 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001691 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001692#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001693};
1694static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001695static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001696
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001697/**
1698 * consume_stock: Try to consume stocked charge on this cpu.
1699 * @memcg: memcg to consume from.
1700 * @nr_pages: how many pages to charge.
1701 *
1702 * The charges will only happen if @memcg matches the current cpu's memcg
1703 * stock, and at least @nr_pages are available in that stock. Failure to
1704 * service an allocation will refill the stock.
1705 *
1706 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001707 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001708static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001709{
1710 struct memcg_stock_pcp *stock;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001711 unsigned long flags;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001712 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001713
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001714 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001715 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001716
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001717 local_irq_save(flags);
1718
1719 stock = this_cpu_ptr(&memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001720 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001721 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001722 ret = true;
1723 }
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001724
1725 local_irq_restore(flags);
1726
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001727 return ret;
1728}
1729
1730/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001731 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001732 */
1733static void drain_stock(struct memcg_stock_pcp *stock)
1734{
1735 struct mem_cgroup *old = stock->cached;
1736
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001737 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001738 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001739 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001740 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001741 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001742 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001743 }
1744 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001745}
1746
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001747static void drain_local_stock(struct work_struct *dummy)
1748{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001749 struct memcg_stock_pcp *stock;
1750 unsigned long flags;
1751
1752 local_irq_save(flags);
1753
1754 stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001755 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001756 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001757
1758 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001759}
1760
1761/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001762 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001763 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001764 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001765static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001766{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001767 struct memcg_stock_pcp *stock;
1768 unsigned long flags;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001769
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001770 local_irq_save(flags);
1771
1772 stock = this_cpu_ptr(&memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001773 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001774 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001775 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001776 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001777 stock->nr_pages += nr_pages;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001778
1779 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001780}
1781
1782/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001783 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001784 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001785 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001786static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001787{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001788 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001789
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001790 /* If someone's already draining, avoid adding running more workers. */
1791 if (!mutex_trylock(&percpu_charge_mutex))
1792 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001793 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001794 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001795 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001796 for_each_online_cpu(cpu) {
1797 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001798 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001799
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001800 memcg = stock->cached;
1801 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001802 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001803 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001804 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001805 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1806 if (cpu == curcpu)
1807 drain_local_stock(&stock->work);
1808 else
1809 schedule_work_on(cpu, &stock->work);
1810 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001811 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001812 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001813 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001814 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001815}
1816
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01001817static int memcg_hotplug_cpu_dead(unsigned int cpu)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001818{
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001819 struct memcg_stock_pcp *stock;
1820
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001821 stock = &per_cpu(memcg_stock, cpu);
1822 drain_stock(stock);
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01001823 return 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001824}
1825
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001826static void reclaim_high(struct mem_cgroup *memcg,
1827 unsigned int nr_pages,
1828 gfp_t gfp_mask)
1829{
1830 do {
1831 if (page_counter_read(&memcg->memory) <= memcg->high)
1832 continue;
1833 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
1834 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1835 } while ((memcg = parent_mem_cgroup(memcg)));
1836}
1837
1838static void high_work_func(struct work_struct *work)
1839{
1840 struct mem_cgroup *memcg;
1841
1842 memcg = container_of(work, struct mem_cgroup, high_work);
1843 reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1844}
1845
Tejun Heob23afb92015-11-05 18:46:11 -08001846/*
1847 * Scheduled by try_charge() to be executed from the userland return path
1848 * and reclaims memory over the high limit.
1849 */
1850void mem_cgroup_handle_over_high(void)
1851{
1852 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001853 struct mem_cgroup *memcg;
Tejun Heob23afb92015-11-05 18:46:11 -08001854
1855 if (likely(!nr_pages))
1856 return;
1857
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001858 memcg = get_mem_cgroup_from_mm(current->mm);
1859 reclaim_high(memcg, nr_pages, GFP_KERNEL);
Tejun Heob23afb92015-11-05 18:46:11 -08001860 css_put(&memcg->css);
1861 current->memcg_nr_pages_over_high = 0;
1862}
1863
Johannes Weiner00501b52014-08-08 14:19:20 -07001864static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1865 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001866{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001867 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001868 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001869 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001870 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001871 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001872 bool may_swap = true;
1873 bool drained = false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001874
Johannes Weinerce00a962014-09-05 08:43:57 -04001875 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08001876 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001877retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001878 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08001879 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001880
Johannes Weiner7941d212016-01-14 15:21:23 -08001881 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08001882 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1883 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001884 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08001885 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001886 page_counter_uncharge(&memcg->memsw, batch);
1887 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001888 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001889 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001890 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001891 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001892
Johannes Weiner6539cc02014-08-06 16:05:42 -07001893 if (batch > nr_pages) {
1894 batch = nr_pages;
1895 goto retry;
1896 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001897
Johannes Weiner06b078f2014-08-06 16:05:44 -07001898 /*
1899 * Unlike in global OOM situations, memcg is not in a physical
1900 * memory shortage. Allow dying and OOM-killed tasks to
1901 * bypass the last charges so that they can exit quickly and
1902 * free their memory.
1903 */
1904 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1905 fatal_signal_pending(current) ||
1906 current->flags & PF_EXITING))
Tejun Heo10d53c72015-11-05 18:46:17 -08001907 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001908
Johannes Weiner89a28482016-10-27 17:46:56 -07001909 /*
1910 * Prevent unbounded recursion when reclaim operations need to
1911 * allocate memory. This might exceed the limits temporarily,
1912 * but we prefer facilitating memory reclaim and getting back
1913 * under the limit over triggering OOM kills in these cases.
1914 */
1915 if (unlikely(current->flags & PF_MEMALLOC))
1916 goto force;
1917
Johannes Weiner06b078f2014-08-06 16:05:44 -07001918 if (unlikely(task_in_memcg_oom(current)))
1919 goto nomem;
1920
Mel Gormand0164ad2015-11-06 16:28:21 -08001921 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001922 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001923
Johannes Weiner241994ed2015-02-11 15:26:06 -08001924 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
1925
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001926 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1927 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07001928
Johannes Weiner61e02c72014-08-06 16:08:16 -07001929 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07001930 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07001931
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001932 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001933 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001934 drained = true;
1935 goto retry;
1936 }
1937
Johannes Weiner28c34c22014-08-06 16:05:47 -07001938 if (gfp_mask & __GFP_NORETRY)
1939 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001940 /*
1941 * Even though the limit is exceeded at this point, reclaim
1942 * may have been able to free some pages. Retry the charge
1943 * before killing the task.
1944 *
1945 * Only for regular pages, though: huge pages are rather
1946 * unlikely to succeed so close to the limit, and we fall back
1947 * to regular pages anyway in case of failure.
1948 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07001949 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001950 goto retry;
1951 /*
1952 * At task move, charge accounts can be doubly counted. So, it's
1953 * better to wait until the end of task_move if something is going on.
1954 */
1955 if (mem_cgroup_wait_acct_move(mem_over_limit))
1956 goto retry;
1957
Johannes Weiner9b130612014-08-06 16:05:51 -07001958 if (nr_retries--)
1959 goto retry;
1960
Johannes Weiner06b078f2014-08-06 16:05:44 -07001961 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08001962 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001963
Johannes Weiner6539cc02014-08-06 16:05:42 -07001964 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08001965 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001966
Johannes Weiner241994ed2015-02-11 15:26:06 -08001967 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
1968
Jerome Marchand3608de02015-11-05 18:47:29 -08001969 mem_cgroup_oom(mem_over_limit, gfp_mask,
1970 get_order(nr_pages * PAGE_SIZE));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001971nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07001972 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07001973 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08001974force:
1975 /*
1976 * The allocation either can't fail or will lead to more memory
1977 * being freed very soon. Allow memory usage go over the limit
1978 * temporarily by force charging it.
1979 */
1980 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001981 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08001982 page_counter_charge(&memcg->memsw, nr_pages);
1983 css_get_many(&memcg->css, nr_pages);
1984
1985 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001986
1987done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001988 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07001989 if (batch > nr_pages)
1990 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08001991
Johannes Weiner241994ed2015-02-11 15:26:06 -08001992 /*
Tejun Heob23afb92015-11-05 18:46:11 -08001993 * If the hierarchy is above the normal consumption range, schedule
1994 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08001995 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08001996 * GFP_KERNEL can consistently be used during reclaim. @memcg is
1997 * not recorded as it most likely matches current's and won't
1998 * change in the meantime. As high limit is checked again before
1999 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994ed2015-02-11 15:26:06 -08002000 */
2001 do {
Tejun Heob23afb92015-11-05 18:46:11 -08002002 if (page_counter_read(&memcg->memory) > memcg->high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002003 /* Don't bother a random interrupted task */
2004 if (in_interrupt()) {
2005 schedule_work(&memcg->high_work);
2006 break;
2007 }
Vladimir Davydov9516a182015-12-11 13:40:24 -08002008 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002009 set_notify_resume(current);
2010 break;
2011 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08002012 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002013
2014 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002015}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002016
Johannes Weiner00501b52014-08-08 14:19:20 -07002017static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002018{
Johannes Weinerce00a962014-09-05 08:43:57 -04002019 if (mem_cgroup_is_root(memcg))
2020 return;
2021
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002022 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002023 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002024 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002025
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002026 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002027}
2028
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002029static void lock_page_lru(struct page *page, int *isolated)
2030{
2031 struct zone *zone = page_zone(page);
2032
Mel Gormana52633d2016-07-28 15:45:28 -07002033 spin_lock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002034 if (PageLRU(page)) {
2035 struct lruvec *lruvec;
2036
Mel Gorman599d0c92016-07-28 15:45:31 -07002037 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002038 ClearPageLRU(page);
2039 del_page_from_lru_list(page, lruvec, page_lru(page));
2040 *isolated = 1;
2041 } else
2042 *isolated = 0;
2043}
2044
2045static void unlock_page_lru(struct page *page, int isolated)
2046{
2047 struct zone *zone = page_zone(page);
2048
2049 if (isolated) {
2050 struct lruvec *lruvec;
2051
Mel Gorman599d0c92016-07-28 15:45:31 -07002052 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002053 VM_BUG_ON_PAGE(PageLRU(page), page);
2054 SetPageLRU(page);
2055 add_page_to_lru_list(page, lruvec, page_lru(page));
2056 }
Mel Gormana52633d2016-07-28 15:45:28 -07002057 spin_unlock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002058}
2059
Johannes Weiner00501b52014-08-08 14:19:20 -07002060static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002061 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002062{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002063 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002064
Johannes Weiner1306a852014-12-10 15:44:52 -08002065 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002066
2067 /*
2068 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2069 * may already be on some other mem_cgroup's LRU. Take care of it.
2070 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002071 if (lrucare)
2072 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002073
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002074 /*
2075 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002076 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002077 *
2078 * - the page is uncharged
2079 *
2080 * - the page is off-LRU
2081 *
2082 * - an anonymous fault has exclusive page access, except for
2083 * a locked page table
2084 *
2085 * - a page cache insertion, a swapin fault, or a migration
2086 * have the page locked
2087 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002088 page->mem_cgroup = memcg;
Hugh Dickins3be91272008-02-07 00:14:19 -08002089
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002090 if (lrucare)
2091 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002092}
2093
Johannes Weiner127424c2016-01-20 15:02:32 -08002094#ifndef CONFIG_SLOB
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002095static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002096{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002097 int id, size;
2098 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002099
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002100 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002101 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2102 if (id < 0)
2103 return id;
2104
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002105 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002106 return id;
2107
2108 /*
2109 * There's no space for the new id in memcg_caches arrays,
2110 * so we have to grow them.
2111 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002112 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002113
2114 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002115 if (size < MEMCG_CACHES_MIN_SIZE)
2116 size = MEMCG_CACHES_MIN_SIZE;
2117 else if (size > MEMCG_CACHES_MAX_SIZE)
2118 size = MEMCG_CACHES_MAX_SIZE;
2119
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002120 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002121 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002122 err = memcg_update_all_list_lrus(size);
2123 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002124 memcg_nr_cache_ids = size;
2125
2126 up_write(&memcg_cache_ids_sem);
2127
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002128 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002129 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002130 return err;
2131 }
2132 return id;
2133}
2134
2135static void memcg_free_cache_id(int id)
2136{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002137 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002138}
2139
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002140struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002141 struct mem_cgroup *memcg;
2142 struct kmem_cache *cachep;
2143 struct work_struct work;
2144};
2145
Vladimir Davydov13583c32016-12-12 16:41:29 -08002146static struct workqueue_struct *memcg_kmem_cache_create_wq;
2147
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002148static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002149{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002150 struct memcg_kmem_cache_create_work *cw =
2151 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002152 struct mem_cgroup *memcg = cw->memcg;
2153 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002154
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002155 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002156
Vladimir Davydov5722d092014-04-07 15:39:24 -07002157 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002158 kfree(cw);
2159}
2160
2161/*
2162 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002163 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002164static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2165 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002166{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002167 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002168
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002169 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002170 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002171 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002172
2173 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002174
2175 cw->memcg = memcg;
2176 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002177 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002178
Vladimir Davydov13583c32016-12-12 16:41:29 -08002179 queue_work(memcg_kmem_cache_create_wq, &cw->work);
Glauber Costad7f25f82012-12-18 14:22:40 -08002180}
2181
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002182static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2183 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002184{
2185 /*
2186 * We need to stop accounting when we kmalloc, because if the
2187 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002188 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002189 *
2190 * However, it is better to enclose the whole function. Depending on
2191 * the debugging options enabled, INIT_WORK(), for instance, can
2192 * trigger an allocation. This too, will make us recurse. Because at
2193 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2194 * the safest choice is to do it like this, wrapping the whole function.
2195 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002196 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002197 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002198 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002199}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002200
Vladimir Davydov45264772016-07-26 15:24:21 -07002201static inline bool memcg_kmem_bypass(void)
2202{
2203 if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
2204 return true;
2205 return false;
2206}
2207
2208/**
2209 * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2210 * @cachep: the original global kmem cache
2211 *
Glauber Costad7f25f82012-12-18 14:22:40 -08002212 * Return the kmem_cache we're supposed to use for a slab allocation.
2213 * We try to use the current memcg's version of the cache.
2214 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002215 * If the cache does not exist yet, if we are the first user of it, we
2216 * create it asynchronously in a workqueue and let the current allocation
2217 * go through with the original cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002218 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002219 * This function takes a reference to the cache it returns to assure it
2220 * won't get destroyed while we are working with it. Once the caller is
2221 * done with it, memcg_kmem_put_cache() must be called to release the
2222 * reference.
Glauber Costad7f25f82012-12-18 14:22:40 -08002223 */
Vladimir Davydov45264772016-07-26 15:24:21 -07002224struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002225{
2226 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002227 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002228 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002229
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002230 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002231
Vladimir Davydov45264772016-07-26 15:24:21 -07002232 if (memcg_kmem_bypass())
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002233 return cachep;
2234
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002235 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002236 return cachep;
2237
Vladimir Davydov8135be52014-12-12 16:56:38 -08002238 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002239 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002240 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002241 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002242
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002243 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002244 if (likely(memcg_cachep))
2245 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002246
2247 /*
2248 * If we are in a safe context (can wait, and not in interrupt
2249 * context), we could be be predictable and return right away.
2250 * This would guarantee that the allocation being performed
2251 * already belongs in the new cache.
2252 *
2253 * However, there are some clashes that can arrive from locking.
2254 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002255 * memcg_create_kmem_cache, this means no further allocation
2256 * could happen with the slab_mutex held. So it's better to
2257 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002258 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002259 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002260out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002261 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002262 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002263}
Glauber Costad7f25f82012-12-18 14:22:40 -08002264
Vladimir Davydov45264772016-07-26 15:24:21 -07002265/**
2266 * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2267 * @cachep: the cache returned by memcg_kmem_get_cache
2268 */
2269void memcg_kmem_put_cache(struct kmem_cache *cachep)
Vladimir Davydov8135be52014-12-12 16:56:38 -08002270{
2271 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002272 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002273}
2274
Vladimir Davydov45264772016-07-26 15:24:21 -07002275/**
2276 * memcg_kmem_charge: charge a kmem page
2277 * @page: page to charge
2278 * @gfp: reclaim mode
2279 * @order: allocation order
2280 * @memcg: memory cgroup to charge
2281 *
2282 * Returns 0 on success, an error code on failure.
2283 */
2284int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2285 struct mem_cgroup *memcg)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002286{
2287 unsigned int nr_pages = 1 << order;
2288 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002289 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002290
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002291 ret = try_charge(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08002292 if (ret)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002293 return ret;
Johannes Weiner52c29b02016-01-20 15:02:35 -08002294
2295 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2296 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2297 cancel_charge(memcg, nr_pages);
2298 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002299 }
2300
2301 page->mem_cgroup = memcg;
2302
2303 return 0;
2304}
2305
Vladimir Davydov45264772016-07-26 15:24:21 -07002306/**
2307 * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2308 * @page: page to charge
2309 * @gfp: reclaim mode
2310 * @order: allocation order
2311 *
2312 * Returns 0 on success, an error code on failure.
2313 */
2314int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002315{
2316 struct mem_cgroup *memcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002317 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002318
Vladimir Davydov45264772016-07-26 15:24:21 -07002319 if (memcg_kmem_bypass())
2320 return 0;
2321
Johannes Weinerdf381972014-04-07 15:37:43 -07002322 memcg = get_mem_cgroup_from_mm(current->mm);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002323 if (!mem_cgroup_is_root(memcg)) {
Vladimir Davydov45264772016-07-26 15:24:21 -07002324 ret = memcg_kmem_charge_memcg(page, gfp, order, memcg);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002325 if (!ret)
2326 __SetPageKmemcg(page);
2327 }
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002328 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002329 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002330}
Vladimir Davydov45264772016-07-26 15:24:21 -07002331/**
2332 * memcg_kmem_uncharge: uncharge a kmem page
2333 * @page: page to uncharge
2334 * @order: allocation order
2335 */
2336void memcg_kmem_uncharge(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002337{
Johannes Weiner1306a852014-12-10 15:44:52 -08002338 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002339 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002340
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002341 if (!memcg)
2342 return;
2343
Sasha Levin309381fea2014-01-23 15:52:54 -08002344 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002345
Johannes Weiner52c29b02016-01-20 15:02:35 -08002346 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2347 page_counter_uncharge(&memcg->kmem, nr_pages);
2348
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002349 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002350 if (do_memsw_account())
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002351 page_counter_uncharge(&memcg->memsw, nr_pages);
2352
Johannes Weiner1306a852014-12-10 15:44:52 -08002353 page->mem_cgroup = NULL;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002354
2355 /* slab pages do not have PageKmemcg flag set */
2356 if (PageKmemcg(page))
2357 __ClearPageKmemcg(page);
2358
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002359 css_put_many(&memcg->css, nr_pages);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002360}
Johannes Weiner127424c2016-01-20 15:02:32 -08002361#endif /* !CONFIG_SLOB */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002362
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002363#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2364
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002365/*
2366 * Because tail pages are not marked as "used", set it. We're under
Mel Gormana52633d2016-07-28 15:45:28 -07002367 * zone_lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002368 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002369void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002370{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002371 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002372
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002373 if (mem_cgroup_disabled())
2374 return;
David Rientjesb070e652013-05-07 16:18:09 -07002375
Johannes Weiner29833312014-12-10 15:44:02 -08002376 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002377 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002378
Johannes Weiner1306a852014-12-10 15:44:52 -08002379 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002380 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002381}
Hugh Dickins12d27102012-01-12 17:19:52 -08002382#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002383
Andrew Mortonc255a452012-07-31 16:43:02 -07002384#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002385static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2386 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002387{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002388 int val = (charge) ? 1 : -1;
2389 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002390}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002391
2392/**
2393 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2394 * @entry: swap entry to be moved
2395 * @from: mem_cgroup which the entry is moved from
2396 * @to: mem_cgroup which the entry is moved to
2397 *
2398 * It succeeds only when the swap_cgroup's record for this entry is the same
2399 * as the mem_cgroup's id of @from.
2400 *
2401 * Returns 0 on success, -EINVAL on failure.
2402 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002403 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002404 * both res and memsw, and called css_get().
2405 */
2406static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002407 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002408{
2409 unsigned short old_id, new_id;
2410
Li Zefan34c00c32013-09-23 16:56:01 +08002411 old_id = mem_cgroup_id(from);
2412 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002413
2414 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002415 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002416 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002417 return 0;
2418 }
2419 return -EINVAL;
2420}
2421#else
2422static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002423 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002424{
2425 return -EINVAL;
2426}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002427#endif
2428
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002429static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002430
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002431static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002432 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002433{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002434 unsigned long curusage;
2435 unsigned long oldusage;
2436 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002437 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002438 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002439
2440 /*
2441 * For keeping hierarchical_reclaim simple, how long we should retry
2442 * is depends on callers. We set our retry-count to be function
2443 * of # of children which we should visit in this loop.
2444 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002445 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2446 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002447
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002448 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002449
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002450 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002451 if (signal_pending(current)) {
2452 ret = -EINTR;
2453 break;
2454 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002455
2456 mutex_lock(&memcg_limit_mutex);
2457 if (limit > memcg->memsw.limit) {
2458 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002459 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002460 break;
2461 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002462 if (limit > memcg->memory.limit)
2463 enlarge = true;
2464 ret = page_counter_limit(&memcg->memory, limit);
2465 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002466
2467 if (!ret)
2468 break;
2469
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002470 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2471
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002472 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002473 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002474 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002475 retry_count--;
2476 else
2477 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002478 } while (retry_count);
2479
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002480 if (!ret && enlarge)
2481 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002482
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002483 return ret;
2484}
2485
Li Zefan338c8432009-06-17 16:27:15 -07002486static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002487 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002488{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002489 unsigned long curusage;
2490 unsigned long oldusage;
2491 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002492 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002493 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002494
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002495 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002496 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2497 mem_cgroup_count_children(memcg);
2498
2499 oldusage = page_counter_read(&memcg->memsw);
2500
2501 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002502 if (signal_pending(current)) {
2503 ret = -EINTR;
2504 break;
2505 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002506
2507 mutex_lock(&memcg_limit_mutex);
2508 if (limit < memcg->memory.limit) {
2509 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002510 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002511 break;
2512 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002513 if (limit > memcg->memsw.limit)
2514 enlarge = true;
2515 ret = page_counter_limit(&memcg->memsw, limit);
2516 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002517
2518 if (!ret)
2519 break;
2520
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002521 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2522
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002523 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002524 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002525 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002526 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002527 else
2528 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002529 } while (retry_count);
2530
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002531 if (!ret && enlarge)
2532 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002533
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002534 return ret;
2535}
2536
Mel Gormanef8f2322016-07-28 15:46:05 -07002537unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
Andrew Morton0608f432013-09-24 15:27:41 -07002538 gfp_t gfp_mask,
2539 unsigned long *total_scanned)
2540{
2541 unsigned long nr_reclaimed = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002542 struct mem_cgroup_per_node *mz, *next_mz = NULL;
Andrew Morton0608f432013-09-24 15:27:41 -07002543 unsigned long reclaimed;
2544 int loop = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002545 struct mem_cgroup_tree_per_node *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002546 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002547 unsigned long nr_scanned;
2548
2549 if (order > 0)
2550 return 0;
2551
Mel Gormanef8f2322016-07-28 15:46:05 -07002552 mctz = soft_limit_tree_node(pgdat->node_id);
Michal Hockod6507ff2016-08-02 14:02:37 -07002553
2554 /*
2555 * Do not even bother to check the largest node if the root
2556 * is empty. Do it lockless to prevent lock bouncing. Races
2557 * are acceptable as soft limit is best effort anyway.
2558 */
2559 if (RB_EMPTY_ROOT(&mctz->rb_root))
2560 return 0;
2561
Andrew Morton0608f432013-09-24 15:27:41 -07002562 /*
2563 * This loop can run a while, specially if mem_cgroup's continuously
2564 * keep exceeding their soft limit and putting the system under
2565 * pressure
2566 */
2567 do {
2568 if (next_mz)
2569 mz = next_mz;
2570 else
2571 mz = mem_cgroup_largest_soft_limit_node(mctz);
2572 if (!mz)
2573 break;
2574
2575 nr_scanned = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002576 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07002577 gfp_mask, &nr_scanned);
2578 nr_reclaimed += reclaimed;
2579 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002580 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002581 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002582
2583 /*
2584 * If we failed to reclaim anything from this memory cgroup
2585 * it is time to move on to the next cgroup
2586 */
2587 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002588 if (!reclaimed)
2589 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2590
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002591 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002592 /*
2593 * One school of thought says that we should not add
2594 * back the node to the tree if reclaim returns 0.
2595 * But our reclaim could return 0, simply because due
2596 * to priority we are exposing a smaller subset of
2597 * memory to reclaim from. Consider this as a longer
2598 * term TODO.
2599 */
2600 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002601 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002602 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002603 css_put(&mz->memcg->css);
2604 loop++;
2605 /*
2606 * Could not reclaim anything and there are no more
2607 * mem cgroups to try or we seem to be looping without
2608 * reclaiming anything.
2609 */
2610 if (!nr_reclaimed &&
2611 (next_mz == NULL ||
2612 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2613 break;
2614 } while (!nr_reclaimed);
2615 if (next_mz)
2616 css_put(&next_mz->memcg->css);
2617 return nr_reclaimed;
2618}
2619
Tejun Heoea280e72014-05-16 13:22:48 -04002620/*
2621 * Test whether @memcg has children, dead or alive. Note that this
2622 * function doesn't care whether @memcg has use_hierarchy enabled and
2623 * returns %true if there are child csses according to the cgroup
2624 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2625 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002626static inline bool memcg_has_children(struct mem_cgroup *memcg)
2627{
Tejun Heoea280e72014-05-16 13:22:48 -04002628 bool ret;
2629
Tejun Heoea280e72014-05-16 13:22:48 -04002630 rcu_read_lock();
2631 ret = css_next_child(NULL, &memcg->css);
2632 rcu_read_unlock();
2633 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002634}
2635
2636/*
Greg Thelen51038172016-05-20 16:58:18 -07002637 * Reclaims as many pages from the given memcg as possible.
Michal Hockoc26251f2012-10-26 13:37:28 +02002638 *
2639 * Caller is responsible for holding css reference for memcg.
2640 */
2641static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2642{
2643 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002644
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002645 /* we call try-to-free pages for make this cgroup empty */
2646 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002647 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002648 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002649 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002650
Michal Hockoc26251f2012-10-26 13:37:28 +02002651 if (signal_pending(current))
2652 return -EINTR;
2653
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002654 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2655 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002656 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002657 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002658 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002659 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002660 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002661
2662 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002663
2664 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002665}
2666
Tejun Heo6770c642014-05-13 12:16:21 -04002667static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2668 char *buf, size_t nbytes,
2669 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002670{
Tejun Heo6770c642014-05-13 12:16:21 -04002671 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002672
Michal Hockod8423012012-10-26 13:37:29 +02002673 if (mem_cgroup_is_root(memcg))
2674 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002675 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002676}
2677
Tejun Heo182446d2013-08-08 20:11:24 -04002678static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2679 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002680{
Tejun Heo182446d2013-08-08 20:11:24 -04002681 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002682}
2683
Tejun Heo182446d2013-08-08 20:11:24 -04002684static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2685 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002686{
2687 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002688 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002689 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002690
Glauber Costa567fb432012-07-31 16:43:07 -07002691 if (memcg->use_hierarchy == val)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002692 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07002693
Balbir Singh18f59ea2009-01-07 18:08:07 -08002694 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002695 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002696 * in the child subtrees. If it is unset, then the change can
2697 * occur, provided the current cgroup has no children.
2698 *
2699 * For the root cgroup, parent_mem is NULL, we allow value to be
2700 * set if there are no children.
2701 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002702 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002703 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002704 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002705 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002706 else
2707 retval = -EBUSY;
2708 } else
2709 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002710
Balbir Singh18f59ea2009-01-07 18:08:07 -08002711 return retval;
2712}
2713
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002714static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
Johannes Weinerce00a962014-09-05 08:43:57 -04002715{
2716 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002717 int i;
Johannes Weinerce00a962014-09-05 08:43:57 -04002718
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002719 memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
Johannes Weinerce00a962014-09-05 08:43:57 -04002720
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002721 for_each_mem_cgroup_tree(iter, memcg) {
2722 for (i = 0; i < MEMCG_NR_STAT; i++)
2723 stat[i] += mem_cgroup_read_stat(iter, i);
2724 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002725}
2726
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002727static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
Johannes Weiner587d9f72016-01-20 15:03:19 -08002728{
2729 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002730 int i;
Johannes Weiner587d9f72016-01-20 15:03:19 -08002731
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002732 memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
Johannes Weiner587d9f72016-01-20 15:03:19 -08002733
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002734 for_each_mem_cgroup_tree(iter, memcg) {
2735 for (i = 0; i < MEMCG_NR_EVENTS; i++)
2736 events[i] += mem_cgroup_read_events(iter, i);
2737 }
Johannes Weiner587d9f72016-01-20 15:03:19 -08002738}
2739
Andrew Morton6f646152015-11-06 16:28:58 -08002740static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04002741{
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002742 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002743
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002744 if (mem_cgroup_is_root(memcg)) {
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002745 struct mem_cgroup *iter;
2746
2747 for_each_mem_cgroup_tree(iter, memcg) {
2748 val += mem_cgroup_read_stat(iter,
2749 MEM_CGROUP_STAT_CACHE);
2750 val += mem_cgroup_read_stat(iter,
2751 MEM_CGROUP_STAT_RSS);
2752 if (swap)
2753 val += mem_cgroup_read_stat(iter,
2754 MEM_CGROUP_STAT_SWAP);
2755 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002756 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002757 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002758 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002759 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002760 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002761 }
Michal Hockoc12176d2015-11-05 18:50:29 -08002762 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002763}
2764
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002765enum {
2766 RES_USAGE,
2767 RES_LIMIT,
2768 RES_MAX_USAGE,
2769 RES_FAILCNT,
2770 RES_SOFT_LIMIT,
2771};
Johannes Weinerce00a962014-09-05 08:43:57 -04002772
Tejun Heo791badb2013-12-05 12:28:02 -05002773static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002774 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002775{
Tejun Heo182446d2013-08-08 20:11:24 -04002776 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002777 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002778
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002779 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002780 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002781 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002782 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002783 case _MEMSWAP:
2784 counter = &memcg->memsw;
2785 break;
2786 case _KMEM:
2787 counter = &memcg->kmem;
2788 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002789 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08002790 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002791 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002792 default:
2793 BUG();
2794 }
2795
2796 switch (MEMFILE_ATTR(cft->private)) {
2797 case RES_USAGE:
2798 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08002799 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002800 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08002801 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002802 return (u64)page_counter_read(counter) * PAGE_SIZE;
2803 case RES_LIMIT:
2804 return (u64)counter->limit * PAGE_SIZE;
2805 case RES_MAX_USAGE:
2806 return (u64)counter->watermark * PAGE_SIZE;
2807 case RES_FAILCNT:
2808 return counter->failcnt;
2809 case RES_SOFT_LIMIT:
2810 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002811 default:
2812 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002813 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002814}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002815
Johannes Weiner127424c2016-01-20 15:02:32 -08002816#ifndef CONFIG_SLOB
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002817static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002818{
Vladimir Davydovd6441632014-01-23 15:53:09 -08002819 int memcg_id;
2820
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002821 if (cgroup_memory_nokmem)
2822 return 0;
2823
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002824 BUG_ON(memcg->kmemcg_id >= 0);
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002825 BUG_ON(memcg->kmem_state);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002826
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002827 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002828 if (memcg_id < 0)
2829 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002830
Johannes Weineref129472016-01-14 15:21:34 -08002831 static_branch_inc(&memcg_kmem_enabled_key);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002832 /*
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002833 * A memory cgroup is considered kmem-online as soon as it gets
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002834 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002835 * guarantee no one starts accounting before all call sites are
2836 * patched.
2837 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002838 memcg->kmemcg_id = memcg_id;
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002839 memcg->kmem_state = KMEM_ONLINE;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002840
2841 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002842}
2843
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002844static void memcg_offline_kmem(struct mem_cgroup *memcg)
2845{
2846 struct cgroup_subsys_state *css;
2847 struct mem_cgroup *parent, *child;
2848 int kmemcg_id;
2849
2850 if (memcg->kmem_state != KMEM_ONLINE)
2851 return;
2852 /*
2853 * Clear the online state before clearing memcg_caches array
2854 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2855 * guarantees that no cache will be created for this cgroup
2856 * after we are done (see memcg_create_kmem_cache()).
2857 */
2858 memcg->kmem_state = KMEM_ALLOCATED;
2859
2860 memcg_deactivate_kmem_caches(memcg);
2861
2862 kmemcg_id = memcg->kmemcg_id;
2863 BUG_ON(kmemcg_id < 0);
2864
2865 parent = parent_mem_cgroup(memcg);
2866 if (!parent)
2867 parent = root_mem_cgroup;
2868
2869 /*
2870 * Change kmemcg_id of this cgroup and all its descendants to the
2871 * parent's id, and then move all entries from this cgroup's list_lrus
2872 * to ones of the parent. After we have finished, all list_lrus
2873 * corresponding to this cgroup are guaranteed to remain empty. The
2874 * ordering is imposed by list_lru_node->lock taken by
2875 * memcg_drain_all_list_lrus().
2876 */
Tejun Heo3a06bb72016-06-03 14:55:44 -07002877 rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002878 css_for_each_descendant_pre(css, &memcg->css) {
2879 child = mem_cgroup_from_css(css);
2880 BUG_ON(child->kmemcg_id != kmemcg_id);
2881 child->kmemcg_id = parent->kmemcg_id;
2882 if (!memcg->use_hierarchy)
2883 break;
2884 }
Tejun Heo3a06bb72016-06-03 14:55:44 -07002885 rcu_read_unlock();
2886
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002887 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2888
2889 memcg_free_cache_id(kmemcg_id);
2890}
2891
2892static void memcg_free_kmem(struct mem_cgroup *memcg)
2893{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002894 /* css_alloc() failed, offlining didn't happen */
2895 if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2896 memcg_offline_kmem(memcg);
2897
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002898 if (memcg->kmem_state == KMEM_ALLOCATED) {
2899 memcg_destroy_kmem_caches(memcg);
2900 static_branch_dec(&memcg_kmem_enabled_key);
2901 WARN_ON(page_counter_read(&memcg->kmem));
2902 }
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002903}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002904#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002905static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08002906{
2907 return 0;
2908}
2909static void memcg_offline_kmem(struct mem_cgroup *memcg)
2910{
2911}
2912static void memcg_free_kmem(struct mem_cgroup *memcg)
2913{
2914}
2915#endif /* !CONFIG_SLOB */
2916
Johannes Weiner127424c2016-01-20 15:02:32 -08002917static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2918 unsigned long limit)
2919{
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002920 int ret;
Johannes Weiner127424c2016-01-20 15:02:32 -08002921
2922 mutex_lock(&memcg_limit_mutex);
Johannes Weiner127424c2016-01-20 15:02:32 -08002923 ret = page_counter_limit(&memcg->kmem, limit);
Johannes Weiner127424c2016-01-20 15:02:32 -08002924 mutex_unlock(&memcg_limit_mutex);
2925 return ret;
2926}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002927
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002928static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2929{
2930 int ret;
2931
2932 mutex_lock(&memcg_limit_mutex);
2933
Johannes Weiner0db15292016-01-20 15:02:50 -08002934 ret = page_counter_limit(&memcg->tcpmem, limit);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002935 if (ret)
2936 goto out;
2937
Johannes Weiner0db15292016-01-20 15:02:50 -08002938 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002939 /*
2940 * The active flag needs to be written after the static_key
2941 * update. This is what guarantees that the socket activation
Johannes Weiner2d758072016-10-07 17:00:58 -07002942 * function is the last one to run. See mem_cgroup_sk_alloc()
2943 * for details, and note that we don't mark any socket as
2944 * belonging to this memcg until that flag is up.
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002945 *
2946 * We need to do this, because static_keys will span multiple
2947 * sites, but we can't control their order. If we mark a socket
2948 * as accounted, but the accounting functions are not patched in
2949 * yet, we'll lose accounting.
2950 *
Johannes Weiner2d758072016-10-07 17:00:58 -07002951 * We never race with the readers in mem_cgroup_sk_alloc(),
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002952 * because when this value change, the code to process it is not
2953 * patched in yet.
2954 */
2955 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08002956 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002957 }
2958out:
2959 mutex_unlock(&memcg_limit_mutex);
2960 return ret;
2961}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002962
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002963/*
2964 * The user of this function is...
2965 * RES_LIMIT.
2966 */
Tejun Heo451af502014-05-13 12:16:21 -04002967static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2968 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002969{
Tejun Heo451af502014-05-13 12:16:21 -04002970 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002971 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002972 int ret;
2973
Tejun Heo451af502014-05-13 12:16:21 -04002974 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08002975 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002976 if (ret)
2977 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07002978
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002979 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002980 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07002981 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2982 ret = -EINVAL;
2983 break;
2984 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002985 switch (MEMFILE_TYPE(of_cft(of)->private)) {
2986 case _MEM:
2987 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002988 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002989 case _MEMSWAP:
2990 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
2991 break;
2992 case _KMEM:
2993 ret = memcg_update_kmem_limit(memcg, nr_pages);
2994 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002995 case _TCP:
2996 ret = memcg_update_tcp_limit(memcg, nr_pages);
2997 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002998 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002999 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003000 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003001 memcg->soft_limit = nr_pages;
3002 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003003 break;
3004 }
Tejun Heo451af502014-05-13 12:16:21 -04003005 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003006}
3007
Tejun Heo6770c642014-05-13 12:16:21 -04003008static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3009 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003010{
Tejun Heo6770c642014-05-13 12:16:21 -04003011 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003012 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003013
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003014 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3015 case _MEM:
3016 counter = &memcg->memory;
3017 break;
3018 case _MEMSWAP:
3019 counter = &memcg->memsw;
3020 break;
3021 case _KMEM:
3022 counter = &memcg->kmem;
3023 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003024 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003025 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003026 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003027 default:
3028 BUG();
3029 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003030
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003031 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003032 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003033 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003034 break;
3035 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003036 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003037 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003038 default:
3039 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003040 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003041
Tejun Heo6770c642014-05-13 12:16:21 -04003042 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003043}
3044
Tejun Heo182446d2013-08-08 20:11:24 -04003045static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003046 struct cftype *cft)
3047{
Tejun Heo182446d2013-08-08 20:11:24 -04003048 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003049}
3050
Daisuke Nishimura02491442010-03-10 15:22:17 -08003051#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003052static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003053 struct cftype *cft, u64 val)
3054{
Tejun Heo182446d2013-08-08 20:11:24 -04003055 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003056
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003057 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003058 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003059
Glauber Costaee5e8472013-02-22 16:34:50 -08003060 /*
3061 * No kind of locking is needed in here, because ->can_attach() will
3062 * check this value once in the beginning of the process, and then carry
3063 * on with stale data. This means that changes to this value will only
3064 * affect task migrations starting after the change.
3065 */
3066 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003067 return 0;
3068}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003069#else
Tejun Heo182446d2013-08-08 20:11:24 -04003070static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003071 struct cftype *cft, u64 val)
3072{
3073 return -ENOSYS;
3074}
3075#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003076
Ying Han406eb0c2011-05-26 16:25:37 -07003077#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003078static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003079{
Greg Thelen25485de2013-11-12 15:07:40 -08003080 struct numa_stat {
3081 const char *name;
3082 unsigned int lru_mask;
3083 };
3084
3085 static const struct numa_stat stats[] = {
3086 { "total", LRU_ALL },
3087 { "file", LRU_ALL_FILE },
3088 { "anon", LRU_ALL_ANON },
3089 { "unevictable", BIT(LRU_UNEVICTABLE) },
3090 };
3091 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003092 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003093 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003094 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003095
Greg Thelen25485de2013-11-12 15:07:40 -08003096 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3097 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3098 seq_printf(m, "%s=%lu", stat->name, nr);
3099 for_each_node_state(nid, N_MEMORY) {
3100 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3101 stat->lru_mask);
3102 seq_printf(m, " N%d=%lu", nid, nr);
3103 }
3104 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003105 }
Ying Han406eb0c2011-05-26 16:25:37 -07003106
Ying Han071aee12013-11-12 15:07:41 -08003107 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3108 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003109
Ying Han071aee12013-11-12 15:07:41 -08003110 nr = 0;
3111 for_each_mem_cgroup_tree(iter, memcg)
3112 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3113 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3114 for_each_node_state(nid, N_MEMORY) {
3115 nr = 0;
3116 for_each_mem_cgroup_tree(iter, memcg)
3117 nr += mem_cgroup_node_nr_lru_pages(
3118 iter, nid, stat->lru_mask);
3119 seq_printf(m, " N%d=%lu", nid, nr);
3120 }
3121 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003122 }
Ying Han406eb0c2011-05-26 16:25:37 -07003123
Ying Han406eb0c2011-05-26 16:25:37 -07003124 return 0;
3125}
3126#endif /* CONFIG_NUMA */
3127
Tejun Heo2da8ca82013-12-05 12:28:04 -05003128static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003129{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003130 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003131 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003132 struct mem_cgroup *mi;
3133 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003134
Greg Thelen0ca44b12015-02-11 15:25:58 -08003135 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3136 MEM_CGROUP_STAT_NSTATS);
3137 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3138 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003139 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3140
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003141 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Johannes Weiner7941d212016-01-14 15:21:23 -08003142 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003143 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07003144 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003145 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003146 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003147
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003148 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3149 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3150 mem_cgroup_read_events(memcg, i));
3151
3152 for (i = 0; i < NR_LRU_LISTS; i++)
3153 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3154 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3155
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003156 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003157 memory = memsw = PAGE_COUNTER_MAX;
3158 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3159 memory = min(memory, mi->memory.limit);
3160 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003161 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003162 seq_printf(m, "hierarchical_memory_limit %llu\n",
3163 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08003164 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003165 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3166 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003167
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003168 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003169 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003170
Johannes Weiner7941d212016-01-14 15:21:23 -08003171 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003172 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003173 for_each_mem_cgroup_tree(mi, memcg)
3174 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
Greg Thelen484ebb32015-10-01 15:37:05 -07003175 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003176 }
3177
3178 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3179 unsigned long long val = 0;
3180
3181 for_each_mem_cgroup_tree(mi, memcg)
3182 val += mem_cgroup_read_events(mi, i);
3183 seq_printf(m, "total_%s %llu\n",
3184 mem_cgroup_events_names[i], val);
3185 }
3186
3187 for (i = 0; i < NR_LRU_LISTS; i++) {
3188 unsigned long long val = 0;
3189
3190 for_each_mem_cgroup_tree(mi, memcg)
3191 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3192 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003193 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003194
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003195#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003196 {
Mel Gormanef8f2322016-07-28 15:46:05 -07003197 pg_data_t *pgdat;
3198 struct mem_cgroup_per_node *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003199 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003200 unsigned long recent_rotated[2] = {0, 0};
3201 unsigned long recent_scanned[2] = {0, 0};
3202
Mel Gormanef8f2322016-07-28 15:46:05 -07003203 for_each_online_pgdat(pgdat) {
3204 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
3205 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003206
Mel Gormanef8f2322016-07-28 15:46:05 -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];
3211 }
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 Heo451af502014-05-13 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 Heo451af502014-05-13 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 Heo451af502014-05-13 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 Heo451af502014-05-13 12:16:21 -04003794 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003795
Tejun Heo451af502014-05-13 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 Heo451af502014-05-13 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 Heo451af502014-05-13 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 Heo451af502014-05-13 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 Heo451af502014-05-13 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 Heo451af502014-05-13 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 Heo451af502014-05-13 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 Heo451af502014-05-13 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
Johannes Weiner73f576c2016-07-20 15:44:57 -07004037/*
4038 * Private memory cgroup IDR
4039 *
4040 * Swap-out records and page cache shadow entries need to store memcg
4041 * references in constrained space, so we maintain an ID space that is
4042 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4043 * memory-controlled cgroups to 64k.
4044 *
4045 * However, there usually are many references to the oflline CSS after
4046 * the cgroup has been destroyed, such as page cache or reclaimable
4047 * slab objects, that don't need to hang on to the ID. We want to keep
4048 * those dead CSS from occupying IDs, or we might quickly exhaust the
4049 * relatively small ID space and prevent the creation of new cgroups
4050 * even when there are much fewer than 64k cgroups - possibly none.
4051 *
4052 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4053 * be freed and recycled when it's no longer needed, which is usually
4054 * when the CSS is offlined.
4055 *
4056 * The only exception to that are records of swapped out tmpfs/shmem
4057 * pages that need to be attributed to live ancestors on swapin. But
4058 * those references are manageable from userspace.
4059 */
4060
4061static DEFINE_IDR(mem_cgroup_idr);
4062
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004063static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004064{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004065 VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0);
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004066 atomic_add(n, &memcg->id.ref);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004067}
4068
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004069static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004070{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004071 VM_BUG_ON(atomic_read(&memcg->id.ref) < n);
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004072 if (atomic_sub_and_test(n, &memcg->id.ref)) {
Johannes Weiner73f576c2016-07-20 15:44:57 -07004073 idr_remove(&mem_cgroup_idr, memcg->id.id);
4074 memcg->id.id = 0;
4075
4076 /* Memcg ID pins CSS */
4077 css_put(&memcg->css);
4078 }
4079}
4080
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004081static inline void mem_cgroup_id_get(struct mem_cgroup *memcg)
4082{
4083 mem_cgroup_id_get_many(memcg, 1);
4084}
4085
4086static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
4087{
4088 mem_cgroup_id_put_many(memcg, 1);
4089}
4090
Johannes Weiner73f576c2016-07-20 15:44:57 -07004091/**
4092 * mem_cgroup_from_id - look up a memcg from a memcg id
4093 * @id: the memcg id to look up
4094 *
4095 * Caller must hold rcu_read_lock().
4096 */
4097struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
4098{
4099 WARN_ON_ONCE(!rcu_read_lock_held());
4100 return idr_find(&mem_cgroup_idr, id);
4101}
4102
Mel Gormanef8f2322016-07-28 15:46:05 -07004103static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004104{
4105 struct mem_cgroup_per_node *pn;
Mel Gormanef8f2322016-07-28 15:46:05 -07004106 int tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004107 /*
4108 * This routine is called against possible nodes.
4109 * But it's BUG to call kmalloc() against offline node.
4110 *
4111 * TODO: this routine can waste much memory for nodes which will
4112 * never be onlined. It's better to use memory hotplug callback
4113 * function.
4114 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004115 if (!node_state(node, N_NORMAL_MEMORY))
4116 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004117 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004118 if (!pn)
4119 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004120
Mel Gormanef8f2322016-07-28 15:46:05 -07004121 lruvec_init(&pn->lruvec);
4122 pn->usage_in_excess = 0;
4123 pn->on_tree = false;
4124 pn->memcg = memcg;
4125
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004126 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004127 return 0;
4128}
4129
Mel Gormanef8f2322016-07-28 15:46:05 -07004130static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004131{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004132 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004133}
4134
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004135static void mem_cgroup_free(struct mem_cgroup *memcg)
4136{
4137 int node;
4138
4139 memcg_wb_domain_exit(memcg);
4140 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004141 free_mem_cgroup_per_node_info(memcg, node);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004142 free_percpu(memcg->stat);
4143 kfree(memcg);
4144}
4145
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004146static struct mem_cgroup *mem_cgroup_alloc(void)
4147{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004148 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004149 size_t size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004150 int node;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004151
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004152 size = sizeof(struct mem_cgroup);
4153 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004154
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004155 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004156 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004157 return NULL;
4158
Johannes Weiner73f576c2016-07-20 15:44:57 -07004159 memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
4160 1, MEM_CGROUP_ID_MAX,
4161 GFP_KERNEL);
4162 if (memcg->id.id < 0)
4163 goto fail;
4164
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004165 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4166 if (!memcg->stat)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004167 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004168
Bob Liu3ed28fa2012-01-12 17:19:04 -08004169 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004170 if (alloc_mem_cgroup_per_node_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004171 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004172
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004173 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4174 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004175
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004176 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08004177 memcg->last_scanned_node = MAX_NUMNODES;
4178 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004179 mutex_init(&memcg->thresholds_lock);
4180 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004181 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004182 INIT_LIST_HEAD(&memcg->event_list);
4183 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004184 memcg->socket_pressure = jiffies;
Johannes Weiner127424c2016-01-20 15:02:32 -08004185#ifndef CONFIG_SLOB
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004186 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004187#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004188#ifdef CONFIG_CGROUP_WRITEBACK
4189 INIT_LIST_HEAD(&memcg->cgwb_list);
4190#endif
Johannes Weiner73f576c2016-07-20 15:44:57 -07004191 idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004192 return memcg;
4193fail:
Johannes Weiner73f576c2016-07-20 15:44:57 -07004194 if (memcg->id.id > 0)
4195 idr_remove(&mem_cgroup_idr, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004196 mem_cgroup_free(memcg);
4197 return NULL;
Glauber Costad142e3e2013-02-22 16:34:52 -08004198}
4199
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004200static struct cgroup_subsys_state * __ref
4201mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004202{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004203 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4204 struct mem_cgroup *memcg;
4205 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08004206
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004207 memcg = mem_cgroup_alloc();
4208 if (!memcg)
4209 return ERR_PTR(error);
Li Zefan4219b2d2013-09-23 16:56:29 +08004210
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004211 memcg->high = PAGE_COUNTER_MAX;
4212 memcg->soft_limit = PAGE_COUNTER_MAX;
4213 if (parent) {
4214 memcg->swappiness = mem_cgroup_swappiness(parent);
4215 memcg->oom_kill_disable = parent->oom_kill_disable;
4216 }
4217 if (parent && parent->use_hierarchy) {
4218 memcg->use_hierarchy = true;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004219 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004220 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004221 page_counter_init(&memcg->memsw, &parent->memsw);
4222 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08004223 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004224 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004225 page_counter_init(&memcg->memory, NULL);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004226 page_counter_init(&memcg->swap, NULL);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004227 page_counter_init(&memcg->memsw, NULL);
4228 page_counter_init(&memcg->kmem, NULL);
Johannes Weiner0db15292016-01-20 15:02:50 -08004229 page_counter_init(&memcg->tcpmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004230 /*
4231 * Deeper hierachy with use_hierarchy == false doesn't make
4232 * much sense so let cgroup subsystem know about this
4233 * unfortunate state in our controller.
4234 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004235 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004236 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004237 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08004238
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004239 /* The following stuff does not apply to the root */
4240 if (!parent) {
4241 root_mem_cgroup = memcg;
4242 return &memcg->css;
4243 }
4244
Vladimir Davydovb313aee2016-03-17 14:18:27 -07004245 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004246 if (error)
4247 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08004248
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004249 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004250 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004251
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004252 return &memcg->css;
4253fail:
4254 mem_cgroup_free(memcg);
Tejun Heoea3a9642016-06-24 14:49:58 -07004255 return ERR_PTR(-ENOMEM);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004256}
4257
Johannes Weiner73f576c2016-07-20 15:44:57 -07004258static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004259{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004260 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4261
Johannes Weiner73f576c2016-07-20 15:44:57 -07004262 /* Online state pins memcg ID, memcg ID pins CSS */
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004263 atomic_set(&memcg->id.ref, 1);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004264 css_get(css);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004265 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004266}
4267
Tejun Heoeb954192013-08-08 20:11:23 -04004268static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004269{
Tejun Heoeb954192013-08-08 20:11:23 -04004270 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004271 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004272
4273 /*
4274 * Unregister events and notify userspace.
4275 * Notify userspace about cgroup removing only after rmdir of cgroup
4276 * directory to avoid race between userspace and kernelspace.
4277 */
Tejun Heofba94802013-11-22 18:20:43 -05004278 spin_lock(&memcg->event_list_lock);
4279 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004280 list_del_init(&event->list);
4281 schedule_work(&event->remove);
4282 }
Tejun Heofba94802013-11-22 18:20:43 -05004283 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004284
Johannes Weiner567e9ab2016-01-20 15:02:24 -08004285 memcg_offline_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004286 wb_memcg_offline(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004287
4288 mem_cgroup_id_put(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004289}
4290
Vladimir Davydov6df38682015-12-29 14:54:10 -08004291static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4292{
4293 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4294
4295 invalidate_reclaim_iterators(memcg);
4296}
4297
Tejun Heoeb954192013-08-08 20:11:23 -04004298static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004299{
Tejun Heoeb954192013-08-08 20:11:23 -04004300 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004301
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004302 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004303 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004304
Johannes Weiner0db15292016-01-20 15:02:50 -08004305 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004306 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004307
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004308 vmpressure_cleanup(&memcg->vmpressure);
4309 cancel_work_sync(&memcg->high_work);
4310 mem_cgroup_remove_from_trees(memcg);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004311 memcg_free_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004312 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004313}
4314
Tejun Heo1ced9532014-07-08 18:02:57 -04004315/**
4316 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4317 * @css: the target css
4318 *
4319 * Reset the states of the mem_cgroup associated with @css. This is
4320 * invoked when the userland requests disabling on the default hierarchy
4321 * but the memcg is pinned through dependency. The memcg should stop
4322 * applying policies and should revert to the vanilla state as it may be
4323 * made visible again.
4324 *
4325 * The current implementation only resets the essential configurations.
4326 * This needs to be expanded to cover all the visible parts.
4327 */
4328static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4329{
4330 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4331
Vladimir Davydovd334c9bc2016-03-17 14:19:38 -07004332 page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4333 page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4334 page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4335 page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4336 page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004337 memcg->low = 0;
4338 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004339 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004340 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004341}
4342
Daisuke Nishimura02491442010-03-10 15:22:17 -08004343#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004344/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004345static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004346{
Johannes Weiner05b84302014-08-06 16:05:59 -07004347 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004348
Mel Gormand0164ad2015-11-06 16:28:21 -08004349 /* Try a single bulk charge without reclaim first, kswapd may wake */
4350 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004351 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004352 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004353 return ret;
4354 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004355
David Rientjes36745342017-01-24 15:18:10 -08004356 /* Try charges one by one with reclaim, but do not retry */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004357 while (count--) {
David Rientjes36745342017-01-24 15:18:10 -08004358 ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004359 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004360 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004361 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004362 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004363 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004364 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004365}
4366
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004367union mc_target {
4368 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004369 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004370};
4371
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004372enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004373 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004374 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004375 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004376};
4377
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004378static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4379 unsigned long addr, pte_t ptent)
4380{
4381 struct page *page = vm_normal_page(vma, addr, ptent);
4382
4383 if (!page || !page_mapped(page))
4384 return NULL;
4385 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004386 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004387 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004388 } else {
4389 if (!(mc.flags & MOVE_FILE))
4390 return NULL;
4391 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004392 if (!get_page_unless_zero(page))
4393 return NULL;
4394
4395 return page;
4396}
4397
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004398#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004399static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004400 pte_t ptent, swp_entry_t *entry)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004401{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004402 struct page *page = NULL;
4403 swp_entry_t ent = pte_to_swp_entry(ptent);
4404
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004405 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004406 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004407 /*
4408 * Because lookup_swap_cache() updates some statistics counter,
4409 * we call find_get_page() with swapper_space directly.
4410 */
Huang Yingf6ab1f72016-10-07 17:00:21 -07004411 page = find_get_page(swap_address_space(ent), swp_offset(ent));
Johannes Weiner7941d212016-01-14 15:21:23 -08004412 if (do_memsw_account())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004413 entry->val = ent.val;
4414
4415 return page;
4416}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004417#else
4418static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004419 pte_t ptent, swp_entry_t *entry)
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004420{
4421 return NULL;
4422}
4423#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004424
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004425static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4426 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4427{
4428 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004429 struct address_space *mapping;
4430 pgoff_t pgoff;
4431
4432 if (!vma->vm_file) /* anonymous vma */
4433 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004434 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004435 return NULL;
4436
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004437 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004438 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004439
4440 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004441#ifdef CONFIG_SWAP
4442 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004443 if (shmem_mapping(mapping)) {
4444 page = find_get_entry(mapping, pgoff);
4445 if (radix_tree_exceptional_entry(page)) {
4446 swp_entry_t swp = radix_to_swp_entry(page);
Johannes Weiner7941d212016-01-14 15:21:23 -08004447 if (do_memsw_account())
Johannes Weiner139b6a62014-05-06 12:50:05 -07004448 *entry = swp;
Huang Yingf6ab1f72016-10-07 17:00:21 -07004449 page = find_get_page(swap_address_space(swp),
4450 swp_offset(swp));
Johannes Weiner139b6a62014-05-06 12:50:05 -07004451 }
4452 } else
4453 page = find_get_page(mapping, pgoff);
4454#else
4455 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004456#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004457 return page;
4458}
4459
Chen Gangb1b0dea2015-04-14 15:47:35 -07004460/**
4461 * mem_cgroup_move_account - move account of the page
4462 * @page: the page
Li RongQing25843c22016-07-26 15:26:56 -07004463 * @compound: charge the page as compound or small page
Chen Gangb1b0dea2015-04-14 15:47:35 -07004464 * @from: mem_cgroup which the page is moved from.
4465 * @to: mem_cgroup which the page is moved to. @from != @to.
4466 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08004467 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07004468 *
4469 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4470 * from old cgroup.
4471 */
4472static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004473 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07004474 struct mem_cgroup *from,
4475 struct mem_cgroup *to)
4476{
4477 unsigned long flags;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004478 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004479 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004480 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004481
4482 VM_BUG_ON(from == to);
4483 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004484 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07004485
4486 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07004487 * Prevent mem_cgroup_migrate() from looking at
Hugh Dickins45637ba2015-11-05 18:49:40 -08004488 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07004489 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004490 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004491 if (!trylock_page(page))
4492 goto out;
4493
4494 ret = -EINVAL;
4495 if (page->mem_cgroup != from)
4496 goto out_unlock;
4497
Greg Thelenc4843a72015-05-22 17:13:16 -04004498 anon = PageAnon(page);
4499
Chen Gangb1b0dea2015-04-14 15:47:35 -07004500 spin_lock_irqsave(&from->move_lock, flags);
4501
Greg Thelenc4843a72015-05-22 17:13:16 -04004502 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004503 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4504 nr_pages);
4505 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4506 nr_pages);
4507 }
4508
Greg Thelenc4843a72015-05-22 17:13:16 -04004509 /*
4510 * move_lock grabbed above and caller set from->moving_account, so
4511 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4512 * So mapping should be stable for dirty pages.
4513 */
4514 if (!anon && PageDirty(page)) {
4515 struct address_space *mapping = page_mapping(page);
4516
4517 if (mapping_cap_account_dirty(mapping)) {
4518 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4519 nr_pages);
4520 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4521 nr_pages);
4522 }
4523 }
4524
Chen Gangb1b0dea2015-04-14 15:47:35 -07004525 if (PageWriteback(page)) {
4526 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4527 nr_pages);
4528 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4529 nr_pages);
4530 }
4531
4532 /*
4533 * It is safe to change page->mem_cgroup here because the page
4534 * is referenced, charged, and isolated - we can't race with
4535 * uncharging, charging, migration, or LRU putback.
4536 */
4537
4538 /* caller should have done css_get */
4539 page->mem_cgroup = to;
4540 spin_unlock_irqrestore(&from->move_lock, flags);
4541
4542 ret = 0;
4543
4544 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004545 mem_cgroup_charge_statistics(to, page, compound, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004546 memcg_check_events(to, page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004547 mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004548 memcg_check_events(from, page);
4549 local_irq_enable();
4550out_unlock:
4551 unlock_page(page);
4552out:
4553 return ret;
4554}
4555
Li RongQing7cf78062016-05-27 14:27:46 -07004556/**
4557 * get_mctgt_type - get target type of moving charge
4558 * @vma: the vma the pte to be checked belongs
4559 * @addr: the address corresponding to the pte to be checked
4560 * @ptent: the pte to be checked
4561 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4562 *
4563 * Returns
4564 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4565 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4566 * move charge. if @target is not NULL, the page is stored in target->page
4567 * with extra refcnt got(Callers should handle it).
4568 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4569 * target for charge migration. if @target is not NULL, the entry is stored
4570 * in target->ent.
4571 *
4572 * Called with pte lock held.
4573 */
4574
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004575static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004576 unsigned long addr, pte_t ptent, union mc_target *target)
4577{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004578 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004579 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004580 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004581
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004582 if (pte_present(ptent))
4583 page = mc_handle_present_pte(vma, addr, ptent);
4584 else if (is_swap_pte(ptent))
Li RongQing48406ef2016-07-26 15:22:14 -07004585 page = mc_handle_swap_pte(vma, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004586 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004587 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004588
4589 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004590 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004591 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004592 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004593 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004594 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004595 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004596 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004597 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004598 ret = MC_TARGET_PAGE;
4599 if (target)
4600 target->page = page;
4601 }
4602 if (!ret || !target)
4603 put_page(page);
4604 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004605 /* There is a swap entry and a page doesn't exist or isn't charged */
4606 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004607 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004608 ret = MC_TARGET_SWAP;
4609 if (target)
4610 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004611 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004612 return ret;
4613}
4614
Naoya Horiguchi12724852012-03-21 16:34:28 -07004615#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4616/*
4617 * We don't consider swapping or file mapped pages because THP does not
4618 * support them for now.
4619 * Caller should make sure that pmd_trans_huge(pmd) is true.
4620 */
4621static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4622 unsigned long addr, pmd_t pmd, union mc_target *target)
4623{
4624 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004625 enum mc_target_type ret = MC_TARGET_NONE;
4626
4627 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004628 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004629 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004630 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004631 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004632 ret = MC_TARGET_PAGE;
4633 if (target) {
4634 get_page(page);
4635 target->page = page;
4636 }
4637 }
4638 return ret;
4639}
4640#else
4641static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4642 unsigned long addr, pmd_t pmd, union mc_target *target)
4643{
4644 return MC_TARGET_NONE;
4645}
4646#endif
4647
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004648static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4649 unsigned long addr, unsigned long end,
4650 struct mm_walk *walk)
4651{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004652 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004653 pte_t *pte;
4654 spinlock_t *ptl;
4655
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004656 ptl = pmd_trans_huge_lock(pmd, vma);
4657 if (ptl) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004658 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4659 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004660 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004661 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004662 }
Dave Hansen03319322011-03-22 16:32:56 -07004663
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004664 if (pmd_trans_unstable(pmd))
4665 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004666 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4667 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004668 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004669 mc.precharge++; /* increment precharge temporarily */
4670 pte_unmap_unlock(pte - 1, ptl);
4671 cond_resched();
4672
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004673 return 0;
4674}
4675
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004676static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4677{
4678 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004679
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004680 struct mm_walk mem_cgroup_count_precharge_walk = {
4681 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4682 .mm = mm,
4683 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004684 down_read(&mm->mmap_sem);
James Morse0247f3f2016-10-07 17:00:12 -07004685 walk_page_range(0, mm->highest_vm_end,
4686 &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004687 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004688
4689 precharge = mc.precharge;
4690 mc.precharge = 0;
4691
4692 return precharge;
4693}
4694
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004695static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4696{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004697 unsigned long precharge = mem_cgroup_count_precharge(mm);
4698
4699 VM_BUG_ON(mc.moving_task);
4700 mc.moving_task = current;
4701 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004702}
4703
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004704/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4705static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004706{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004707 struct mem_cgroup *from = mc.from;
4708 struct mem_cgroup *to = mc.to;
4709
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004710 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004711 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004712 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004713 mc.precharge = 0;
4714 }
4715 /*
4716 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4717 * we must uncharge here.
4718 */
4719 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004720 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004721 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004722 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004723 /* we must fixup refcnts and charges */
4724 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004725 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004726 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004727 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004728
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004729 mem_cgroup_id_put_many(mc.from, mc.moved_swap);
4730
Johannes Weiner05b84302014-08-06 16:05:59 -07004731 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004732 * we charged both to->memory and to->memsw, so we
4733 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004734 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004735 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004736 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004737
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004738 mem_cgroup_id_get_many(mc.to, mc.moved_swap);
4739 css_put_many(&mc.to->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004740
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004741 mc.moved_swap = 0;
4742 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004743 memcg_oom_recover(from);
4744 memcg_oom_recover(to);
4745 wake_up_all(&mc.waitq);
4746}
4747
4748static void mem_cgroup_clear_mc(void)
4749{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004750 struct mm_struct *mm = mc.mm;
4751
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004752 /*
4753 * we must clear moving_task before waking up waiters at the end of
4754 * task migration.
4755 */
4756 mc.moving_task = NULL;
4757 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004758 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004759 mc.from = NULL;
4760 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04004761 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004762 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004763
4764 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004765}
4766
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004767static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004768{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004769 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07004770 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07004771 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04004772 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004773 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004774 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004775 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004776
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004777 /* charge immigration isn't supported on the default hierarchy */
4778 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07004779 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004780
Tejun Heo4530edd2015-09-11 15:00:19 -04004781 /*
4782 * Multi-process migrations only happen on the default hierarchy
4783 * where charge immigration is not used. Perform charge
4784 * immigration if @tset contains a leader and whine if there are
4785 * multiple.
4786 */
4787 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004788 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04004789 WARN_ON_ONCE(p);
4790 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004791 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04004792 }
4793 if (!p)
4794 return 0;
4795
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004796 /*
4797 * We are now commited to this value whatever it is. Changes in this
4798 * tunable will only affect upcoming migrations, not the current one.
4799 * So we need to save it, and keep it going.
4800 */
4801 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4802 if (!move_flags)
4803 return 0;
4804
Tejun Heo9f2115f2015-09-08 15:01:10 -07004805 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004806
Tejun Heo9f2115f2015-09-08 15:01:10 -07004807 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004808
Tejun Heo9f2115f2015-09-08 15:01:10 -07004809 mm = get_task_mm(p);
4810 if (!mm)
4811 return 0;
4812 /* We move charges only when we move a owner of the mm */
4813 if (mm->owner == p) {
4814 VM_BUG_ON(mc.from);
4815 VM_BUG_ON(mc.to);
4816 VM_BUG_ON(mc.precharge);
4817 VM_BUG_ON(mc.moved_charge);
4818 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004819
Tejun Heo9f2115f2015-09-08 15:01:10 -07004820 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004821 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004822 mc.from = from;
4823 mc.to = memcg;
4824 mc.flags = move_flags;
4825 spin_unlock(&mc.lock);
4826 /* We set mc.moving_task later */
4827
4828 ret = mem_cgroup_precharge_mc(mm);
4829 if (ret)
4830 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004831 } else {
4832 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004833 }
4834 return ret;
4835}
4836
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004837static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004838{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004839 if (mc.to)
4840 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004841}
4842
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004843static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4844 unsigned long addr, unsigned long end,
4845 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004846{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004847 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004848 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004849 pte_t *pte;
4850 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004851 enum mc_target_type target_type;
4852 union mc_target target;
4853 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004854
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004855 ptl = pmd_trans_huge_lock(pmd, vma);
4856 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004857 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004858 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004859 return 0;
4860 }
4861 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4862 if (target_type == MC_TARGET_PAGE) {
4863 page = target.page;
4864 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004865 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08004866 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004867 mc.precharge -= HPAGE_PMD_NR;
4868 mc.moved_charge += HPAGE_PMD_NR;
4869 }
4870 putback_lru_page(page);
4871 }
4872 put_page(page);
4873 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004874 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004875 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004876 }
4877
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004878 if (pmd_trans_unstable(pmd))
4879 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004880retry:
4881 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4882 for (; addr != end; addr += PAGE_SIZE) {
4883 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004884 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004885
4886 if (!mc.precharge)
4887 break;
4888
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004889 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004890 case MC_TARGET_PAGE:
4891 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08004892 /*
4893 * We can have a part of the split pmd here. Moving it
4894 * can be done but it would be too convoluted so simply
4895 * ignore such a partial THP and keep it in original
4896 * memcg. There should be somebody mapping the head.
4897 */
4898 if (PageTransCompound(page))
4899 goto put;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004900 if (isolate_lru_page(page))
4901 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004902 if (!mem_cgroup_move_account(page, false,
4903 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004904 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004905 /* we uncharge from mc.from later. */
4906 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004907 }
4908 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004909put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004910 put_page(page);
4911 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004912 case MC_TARGET_SWAP:
4913 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004914 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004915 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004916 /* we fixup refcnts and charges later. */
4917 mc.moved_swap++;
4918 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004919 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004920 default:
4921 break;
4922 }
4923 }
4924 pte_unmap_unlock(pte - 1, ptl);
4925 cond_resched();
4926
4927 if (addr != end) {
4928 /*
4929 * We have consumed all precharges we got in can_attach().
4930 * We try charge one by one, but don't do any additional
4931 * charges to mc.to if we have failed in charge once in attach()
4932 * phase.
4933 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004934 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004935 if (!ret)
4936 goto retry;
4937 }
4938
4939 return ret;
4940}
4941
Tejun Heo264a0ae2016-04-21 19:09:02 -04004942static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004943{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004944 struct mm_walk mem_cgroup_move_charge_walk = {
4945 .pmd_entry = mem_cgroup_move_charge_pte_range,
Tejun Heo264a0ae2016-04-21 19:09:02 -04004946 .mm = mc.mm,
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004947 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004948
4949 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004950 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07004951 * Signal lock_page_memcg() to take the memcg's move_lock
4952 * while we're moving its pages to another memcg. Then wait
4953 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08004954 */
4955 atomic_inc(&mc.from->moving_account);
4956 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004957retry:
Tejun Heo264a0ae2016-04-21 19:09:02 -04004958 if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004959 /*
4960 * Someone who are holding the mmap_sem might be waiting in
4961 * waitq. So we cancel all extra charges, wake up all waiters,
4962 * and retry. Because we cancel precharges, we might not be able
4963 * to move enough charges, but moving charge is a best-effort
4964 * feature anyway, so it wouldn't be a big problem.
4965 */
4966 __mem_cgroup_clear_mc();
4967 cond_resched();
4968 goto retry;
4969 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004970 /*
4971 * When we have consumed all precharges and failed in doing
4972 * additional charge, the page walk just aborts.
4973 */
James Morse0247f3f2016-10-07 17:00:12 -07004974 walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk);
4975
Tejun Heo264a0ae2016-04-21 19:09:02 -04004976 up_read(&mc.mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08004977 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004978}
4979
Tejun Heo264a0ae2016-04-21 19:09:02 -04004980static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08004981{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004982 if (mc.to) {
4983 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004984 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004985 }
Balbir Singh67e465a2008-02-07 00:13:54 -08004986}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004987#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004988static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004989{
4990 return 0;
4991}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004992static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004993{
4994}
Tejun Heo264a0ae2016-04-21 19:09:02 -04004995static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004996{
4997}
4998#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08004999
Tejun Heof00baae2013-04-15 13:41:15 -07005000/*
5001 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005002 * to verify whether we're attached to the default hierarchy on each mount
5003 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005004 */
Tejun Heoeb954192013-08-08 20:11:23 -04005005static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005006{
5007 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005008 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005009 * guarantees that @root doesn't have any children, so turning it
5010 * on for the root memcg is enough.
5011 */
Tejun Heo9e10a132015-09-18 11:56:28 -04005012 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee592015-03-12 16:26:19 -07005013 root_mem_cgroup->use_hierarchy = true;
5014 else
5015 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005016}
5017
Johannes Weiner241994ed2015-02-11 15:26:06 -08005018static u64 memory_current_read(struct cgroup_subsys_state *css,
5019 struct cftype *cft)
5020{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005021 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5022
5023 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005024}
5025
5026static int memory_low_show(struct seq_file *m, void *v)
5027{
5028 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005029 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005030
5031 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005032 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005033 else
5034 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5035
5036 return 0;
5037}
5038
5039static ssize_t memory_low_write(struct kernfs_open_file *of,
5040 char *buf, size_t nbytes, loff_t off)
5041{
5042 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5043 unsigned long low;
5044 int err;
5045
5046 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005047 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005048 if (err)
5049 return err;
5050
5051 memcg->low = low;
5052
5053 return nbytes;
5054}
5055
5056static int memory_high_show(struct seq_file *m, void *v)
5057{
5058 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005059 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005060
5061 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005062 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005063 else
5064 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5065
5066 return 0;
5067}
5068
5069static ssize_t memory_high_write(struct kernfs_open_file *of,
5070 char *buf, size_t nbytes, loff_t off)
5071{
5072 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner588083b2016-03-17 14:20:25 -07005073 unsigned long nr_pages;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005074 unsigned long high;
5075 int err;
5076
5077 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005078 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005079 if (err)
5080 return err;
5081
5082 memcg->high = high;
5083
Johannes Weiner588083b2016-03-17 14:20:25 -07005084 nr_pages = page_counter_read(&memcg->memory);
5085 if (nr_pages > high)
5086 try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5087 GFP_KERNEL, true);
5088
Tejun Heo2529bb32015-05-22 18:23:34 -04005089 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005090 return nbytes;
5091}
5092
5093static int memory_max_show(struct seq_file *m, void *v)
5094{
5095 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005096 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005097
5098 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005099 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005100 else
5101 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5102
5103 return 0;
5104}
5105
5106static ssize_t memory_max_write(struct kernfs_open_file *of,
5107 char *buf, size_t nbytes, loff_t off)
5108{
5109 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005110 unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5111 bool drained = false;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005112 unsigned long max;
5113 int err;
5114
5115 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005116 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005117 if (err)
5118 return err;
5119
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005120 xchg(&memcg->memory.limit, max);
5121
5122 for (;;) {
5123 unsigned long nr_pages = page_counter_read(&memcg->memory);
5124
5125 if (nr_pages <= max)
5126 break;
5127
5128 if (signal_pending(current)) {
5129 err = -EINTR;
5130 break;
5131 }
5132
5133 if (!drained) {
5134 drain_all_stock(memcg);
5135 drained = true;
5136 continue;
5137 }
5138
5139 if (nr_reclaims) {
5140 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5141 GFP_KERNEL, true))
5142 nr_reclaims--;
5143 continue;
5144 }
5145
5146 mem_cgroup_events(memcg, MEMCG_OOM, 1);
5147 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5148 break;
5149 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08005150
Tejun Heo2529bb32015-05-22 18:23:34 -04005151 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005152 return nbytes;
5153}
5154
5155static int memory_events_show(struct seq_file *m, void *v)
5156{
5157 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5158
5159 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5160 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5161 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5162 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5163
5164 return 0;
5165}
5166
Johannes Weiner587d9f72016-01-20 15:03:19 -08005167static int memory_stat_show(struct seq_file *m, void *v)
5168{
5169 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005170 unsigned long stat[MEMCG_NR_STAT];
5171 unsigned long events[MEMCG_NR_EVENTS];
Johannes Weiner587d9f72016-01-20 15:03:19 -08005172 int i;
5173
5174 /*
5175 * Provide statistics on the state of the memory subsystem as
5176 * well as cumulative event counters that show past behavior.
5177 *
5178 * This list is ordered following a combination of these gradients:
5179 * 1) generic big picture -> specifics and details
5180 * 2) reflecting userspace activity -> reflecting kernel heuristics
5181 *
5182 * Current memory state:
5183 */
5184
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005185 tree_stat(memcg, stat);
5186 tree_events(memcg, events);
5187
Johannes Weiner587d9f72016-01-20 15:03:19 -08005188 seq_printf(m, "anon %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005189 (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005190 seq_printf(m, "file %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005191 (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE);
Vladimir Davydov12580e42016-03-17 14:17:38 -07005192 seq_printf(m, "kernel_stack %llu\n",
Andy Lutomirskiefdc9492016-07-28 15:48:17 -07005193 (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005194 seq_printf(m, "slab %llu\n",
5195 (u64)(stat[MEMCG_SLAB_RECLAIMABLE] +
5196 stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
Johannes Weinerb2807f02016-01-20 15:03:22 -08005197 seq_printf(m, "sock %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005198 (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005199
5200 seq_printf(m, "file_mapped %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005201 (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005202 seq_printf(m, "file_dirty %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005203 (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005204 seq_printf(m, "file_writeback %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005205 (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005206
5207 for (i = 0; i < NR_LRU_LISTS; i++) {
5208 struct mem_cgroup *mi;
5209 unsigned long val = 0;
5210
5211 for_each_mem_cgroup_tree(mi, memcg)
5212 val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5213 seq_printf(m, "%s %llu\n",
5214 mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5215 }
5216
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005217 seq_printf(m, "slab_reclaimable %llu\n",
5218 (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE);
5219 seq_printf(m, "slab_unreclaimable %llu\n",
5220 (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5221
Johannes Weiner587d9f72016-01-20 15:03:19 -08005222 /* Accumulated memory events */
5223
5224 seq_printf(m, "pgfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005225 events[MEM_CGROUP_EVENTS_PGFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005226 seq_printf(m, "pgmajfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005227 events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005228
5229 return 0;
5230}
5231
Johannes Weiner241994ed2015-02-11 15:26:06 -08005232static struct cftype memory_files[] = {
5233 {
5234 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005235 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005236 .read_u64 = memory_current_read,
5237 },
5238 {
5239 .name = "low",
5240 .flags = CFTYPE_NOT_ON_ROOT,
5241 .seq_show = memory_low_show,
5242 .write = memory_low_write,
5243 },
5244 {
5245 .name = "high",
5246 .flags = CFTYPE_NOT_ON_ROOT,
5247 .seq_show = memory_high_show,
5248 .write = memory_high_write,
5249 },
5250 {
5251 .name = "max",
5252 .flags = CFTYPE_NOT_ON_ROOT,
5253 .seq_show = memory_max_show,
5254 .write = memory_max_write,
5255 },
5256 {
5257 .name = "events",
5258 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04005259 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994ed2015-02-11 15:26:06 -08005260 .seq_show = memory_events_show,
5261 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08005262 {
5263 .name = "stat",
5264 .flags = CFTYPE_NOT_ON_ROOT,
5265 .seq_show = memory_stat_show,
5266 },
Johannes Weiner241994ed2015-02-11 15:26:06 -08005267 { } /* terminate */
5268};
5269
Tejun Heo073219e2014-02-08 10:36:58 -05005270struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005271 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005272 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005273 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08005274 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08005275 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005276 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005277 .can_attach = mem_cgroup_can_attach,
5278 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04005279 .post_attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005280 .bind = mem_cgroup_bind,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005281 .dfl_cftypes = memory_files,
5282 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005283 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005284};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005285
Johannes Weiner241994ed2015-02-11 15:26:06 -08005286/**
Johannes Weiner241994ed2015-02-11 15:26:06 -08005287 * mem_cgroup_low - check if memory consumption is below the normal range
5288 * @root: the highest ancestor to consider
5289 * @memcg: the memory cgroup to check
5290 *
5291 * Returns %true if memory consumption of @memcg, and that of all
5292 * configurable ancestors up to @root, is below the normal range.
5293 */
5294bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5295{
5296 if (mem_cgroup_disabled())
5297 return false;
5298
5299 /*
5300 * The toplevel group doesn't have a configurable range, so
5301 * it's never low when looked at directly, and it is not
5302 * considered an ancestor when assessing the hierarchy.
5303 */
5304
5305 if (memcg == root_mem_cgroup)
5306 return false;
5307
Michal Hocko4e54ded2015-02-27 15:51:46 -08005308 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005309 return false;
5310
5311 while (memcg != root) {
5312 memcg = parent_mem_cgroup(memcg);
5313
5314 if (memcg == root_mem_cgroup)
5315 break;
5316
Michal Hocko4e54ded2015-02-27 15:51:46 -08005317 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005318 return false;
5319 }
5320 return true;
5321}
5322
Johannes Weiner00501b52014-08-08 14:19:20 -07005323/**
5324 * mem_cgroup_try_charge - try charging a page
5325 * @page: page to charge
5326 * @mm: mm context of the victim
5327 * @gfp_mask: reclaim mode
5328 * @memcgp: charged memcg return
Li RongQing25843c22016-07-26 15:26:56 -07005329 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005330 *
5331 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5332 * pages according to @gfp_mask if necessary.
5333 *
5334 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5335 * Otherwise, an error code is returned.
5336 *
5337 * After page->mapping has been set up, the caller must finalize the
5338 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5339 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5340 */
5341int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005342 gfp_t gfp_mask, struct mem_cgroup **memcgp,
5343 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005344{
5345 struct mem_cgroup *memcg = NULL;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005346 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005347 int ret = 0;
5348
5349 if (mem_cgroup_disabled())
5350 goto out;
5351
5352 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005353 /*
5354 * Every swap fault against a single page tries to charge the
5355 * page, bail as early as possible. shmem_unuse() encounters
5356 * already charged pages, too. The USED bit is protected by
5357 * the page lock, which serializes swap cache removal, which
5358 * in turn serializes uncharging.
5359 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005360 VM_BUG_ON_PAGE(!PageLocked(page), page);
Johannes Weiner1306a852014-12-10 15:44:52 -08005361 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005362 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005363
Vladimir Davydov37e84352016-01-20 15:02:56 -08005364 if (do_swap_account) {
Vladimir Davydove993d902015-09-09 15:35:35 -07005365 swp_entry_t ent = { .val = page_private(page), };
5366 unsigned short id = lookup_swap_cgroup_id(ent);
5367
5368 rcu_read_lock();
5369 memcg = mem_cgroup_from_id(id);
5370 if (memcg && !css_tryget_online(&memcg->css))
5371 memcg = NULL;
5372 rcu_read_unlock();
5373 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005374 }
5375
Johannes Weiner00501b52014-08-08 14:19:20 -07005376 if (!memcg)
5377 memcg = get_mem_cgroup_from_mm(mm);
5378
5379 ret = try_charge(memcg, gfp_mask, nr_pages);
5380
5381 css_put(&memcg->css);
Johannes Weiner00501b52014-08-08 14:19:20 -07005382out:
5383 *memcgp = memcg;
5384 return ret;
5385}
5386
5387/**
5388 * mem_cgroup_commit_charge - commit a page charge
5389 * @page: page to charge
5390 * @memcg: memcg to charge the page to
5391 * @lrucare: page might be on LRU already
Li RongQing25843c22016-07-26 15:26:56 -07005392 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005393 *
5394 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5395 * after page->mapping has been set up. This must happen atomically
5396 * as part of the page instantiation, i.e. under the page table lock
5397 * for anonymous pages, under the page lock for page and swap cache.
5398 *
5399 * In addition, the page must not be on the LRU during the commit, to
5400 * prevent racing with task migration. If it might be, use @lrucare.
5401 *
5402 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5403 */
5404void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005405 bool lrucare, bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005406{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005407 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005408
5409 VM_BUG_ON_PAGE(!page->mapping, page);
5410 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5411
5412 if (mem_cgroup_disabled())
5413 return;
5414 /*
5415 * Swap faults will attempt to charge the same page multiple
5416 * times. But reuse_swap_page() might have removed the page
5417 * from swapcache already, so we can't check PageSwapCache().
5418 */
5419 if (!memcg)
5420 return;
5421
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005422 commit_charge(page, memcg, lrucare);
5423
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005424 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005425 mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005426 memcg_check_events(memcg, page);
5427 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005428
Johannes Weiner7941d212016-01-14 15:21:23 -08005429 if (do_memsw_account() && PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005430 swp_entry_t entry = { .val = page_private(page) };
5431 /*
5432 * The swap entry might not get freed for a long time,
5433 * let's not wait for it. The page already received a
5434 * memory+swap charge, drop the swap entry duplicate.
5435 */
5436 mem_cgroup_uncharge_swap(entry);
5437 }
5438}
5439
5440/**
5441 * mem_cgroup_cancel_charge - cancel a page charge
5442 * @page: page to charge
5443 * @memcg: memcg to charge the page to
Li RongQing25843c22016-07-26 15:26:56 -07005444 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005445 *
5446 * Cancel a charge transaction started by mem_cgroup_try_charge().
5447 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005448void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5449 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005450{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005451 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005452
5453 if (mem_cgroup_disabled())
5454 return;
5455 /*
5456 * Swap faults will attempt to charge the same page multiple
5457 * times. But reuse_swap_page() might have removed the page
5458 * from swapcache already, so we can't check PageSwapCache().
5459 */
5460 if (!memcg)
5461 return;
5462
Johannes Weiner00501b52014-08-08 14:19:20 -07005463 cancel_charge(memcg, nr_pages);
5464}
5465
Johannes Weiner747db952014-08-08 14:19:24 -07005466static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005467 unsigned long nr_anon, unsigned long nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005468 unsigned long nr_huge, unsigned long nr_kmem,
5469 struct page *dummy_page)
Johannes Weiner747db952014-08-08 14:19:24 -07005470{
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005471 unsigned long nr_pages = nr_anon + nr_file + nr_kmem;
Johannes Weiner747db952014-08-08 14:19:24 -07005472 unsigned long flags;
5473
Johannes Weinerce00a962014-09-05 08:43:57 -04005474 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005475 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08005476 if (do_memsw_account())
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005477 page_counter_uncharge(&memcg->memsw, nr_pages);
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005478 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem)
5479 page_counter_uncharge(&memcg->kmem, nr_kmem);
Johannes Weinerce00a962014-09-05 08:43:57 -04005480 memcg_oom_recover(memcg);
5481 }
Johannes Weiner747db952014-08-08 14:19:24 -07005482
5483 local_irq_save(flags);
5484 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5485 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5486 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5487 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005488 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005489 memcg_check_events(memcg, dummy_page);
5490 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005491
5492 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005493 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005494}
5495
5496static void uncharge_list(struct list_head *page_list)
5497{
5498 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005499 unsigned long nr_anon = 0;
5500 unsigned long nr_file = 0;
5501 unsigned long nr_huge = 0;
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005502 unsigned long nr_kmem = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005503 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005504 struct list_head *next;
5505 struct page *page;
5506
Johannes Weiner8b592652016-03-17 14:20:31 -07005507 /*
5508 * Note that the list can be a single page->lru; hence the
5509 * do-while loop instead of a simple list_for_each_entry().
5510 */
Johannes Weiner747db952014-08-08 14:19:24 -07005511 next = page_list->next;
5512 do {
Johannes Weiner747db952014-08-08 14:19:24 -07005513 page = list_entry(next, struct page, lru);
5514 next = page->lru.next;
5515
5516 VM_BUG_ON_PAGE(PageLRU(page), page);
5517 VM_BUG_ON_PAGE(page_count(page), page);
5518
Johannes Weiner1306a852014-12-10 15:44:52 -08005519 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005520 continue;
5521
5522 /*
5523 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005524 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005525 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005526 */
5527
Johannes Weiner1306a852014-12-10 15:44:52 -08005528 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005529 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005530 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005531 nr_huge, nr_kmem, page);
5532 pgpgout = nr_anon = nr_file =
5533 nr_huge = nr_kmem = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005534 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005535 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005536 }
5537
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005538 if (!PageKmemcg(page)) {
5539 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005540
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005541 if (PageTransHuge(page)) {
5542 nr_pages <<= compound_order(page);
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005543 nr_huge += nr_pages;
5544 }
5545 if (PageAnon(page))
5546 nr_anon += nr_pages;
5547 else
5548 nr_file += nr_pages;
5549 pgpgout++;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03005550 } else {
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005551 nr_kmem += 1 << compound_order(page);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03005552 __ClearPageKmemcg(page);
5553 }
Johannes Weiner747db952014-08-08 14:19:24 -07005554
Johannes Weiner1306a852014-12-10 15:44:52 -08005555 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005556 } while (next != page_list);
5557
5558 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005559 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005560 nr_huge, nr_kmem, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005561}
5562
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005563/**
5564 * mem_cgroup_uncharge - uncharge a page
5565 * @page: page to uncharge
5566 *
5567 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5568 * mem_cgroup_commit_charge().
5569 */
5570void mem_cgroup_uncharge(struct page *page)
5571{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005572 if (mem_cgroup_disabled())
5573 return;
5574
Johannes Weiner747db952014-08-08 14:19:24 -07005575 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005576 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005577 return;
5578
Johannes Weiner747db952014-08-08 14:19:24 -07005579 INIT_LIST_HEAD(&page->lru);
5580 uncharge_list(&page->lru);
5581}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005582
Johannes Weiner747db952014-08-08 14:19:24 -07005583/**
5584 * mem_cgroup_uncharge_list - uncharge a list of page
5585 * @page_list: list of pages to uncharge
5586 *
5587 * Uncharge a list of pages previously charged with
5588 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5589 */
5590void mem_cgroup_uncharge_list(struct list_head *page_list)
5591{
5592 if (mem_cgroup_disabled())
5593 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005594
Johannes Weiner747db952014-08-08 14:19:24 -07005595 if (!list_empty(page_list))
5596 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005597}
5598
5599/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005600 * mem_cgroup_migrate - charge a page's replacement
5601 * @oldpage: currently circulating page
5602 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005603 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005604 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5605 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005606 *
5607 * Both pages must be locked, @newpage->mapping must be set up.
5608 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005609void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005610{
Johannes Weiner29833312014-12-10 15:44:02 -08005611 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005612 unsigned int nr_pages;
5613 bool compound;
Tejun Heod93c4132016-06-24 14:49:54 -07005614 unsigned long flags;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005615
5616 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5617 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005618 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005619 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5620 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005621
5622 if (mem_cgroup_disabled())
5623 return;
5624
5625 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005626 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005627 return;
5628
Hugh Dickins45637ba2015-11-05 18:49:40 -08005629 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08005630 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005631 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005632 return;
5633
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005634 /* Force-charge the new page. The old one will be freed soon */
5635 compound = PageTransHuge(newpage);
5636 nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5637
5638 page_counter_charge(&memcg->memory, nr_pages);
5639 if (do_memsw_account())
5640 page_counter_charge(&memcg->memsw, nr_pages);
5641 css_get_many(&memcg->css, nr_pages);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005642
Johannes Weiner9cf76662016-03-15 14:57:58 -07005643 commit_charge(newpage, memcg, false);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005644
Tejun Heod93c4132016-06-24 14:49:54 -07005645 local_irq_save(flags);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005646 mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5647 memcg_check_events(memcg, newpage);
Tejun Heod93c4132016-06-24 14:49:54 -07005648 local_irq_restore(flags);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005649}
5650
Johannes Weineref129472016-01-14 15:21:34 -08005651DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08005652EXPORT_SYMBOL(memcg_sockets_enabled_key);
5653
Johannes Weiner2d758072016-10-07 17:00:58 -07005654void mem_cgroup_sk_alloc(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08005655{
5656 struct mem_cgroup *memcg;
5657
Johannes Weiner2d758072016-10-07 17:00:58 -07005658 if (!mem_cgroup_sockets_enabled)
5659 return;
5660
5661 /*
5662 * Socket cloning can throw us here with sk_memcg already
Johannes Weiner11092082016-01-14 15:21:26 -08005663 * filled. It won't however, necessarily happen from
5664 * process context. So the test for root memcg given
5665 * the current task's memcg won't help us in this case.
5666 *
5667 * Respecting the original socket's memcg is a better
5668 * decision in this case.
5669 */
5670 if (sk->sk_memcg) {
5671 BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5672 css_get(&sk->sk_memcg->css);
5673 return;
5674 }
5675
5676 rcu_read_lock();
5677 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005678 if (memcg == root_mem_cgroup)
5679 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08005680 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005681 goto out;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005682 if (css_tryget_online(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08005683 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005684out:
Johannes Weiner11092082016-01-14 15:21:26 -08005685 rcu_read_unlock();
5686}
Johannes Weiner11092082016-01-14 15:21:26 -08005687
Johannes Weiner2d758072016-10-07 17:00:58 -07005688void mem_cgroup_sk_free(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08005689{
Johannes Weiner2d758072016-10-07 17:00:58 -07005690 if (sk->sk_memcg)
5691 css_put(&sk->sk_memcg->css);
Johannes Weiner11092082016-01-14 15:21:26 -08005692}
5693
5694/**
5695 * mem_cgroup_charge_skmem - charge socket memory
5696 * @memcg: memcg to charge
5697 * @nr_pages: number of pages to charge
5698 *
5699 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5700 * @memcg's configured limit, %false if the charge had to be forced.
5701 */
5702bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5703{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005704 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08005705
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005706 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005707 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005708
Johannes Weiner0db15292016-01-20 15:02:50 -08005709 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5710 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005711 return true;
5712 }
Johannes Weiner0db15292016-01-20 15:02:50 -08005713 page_counter_charge(&memcg->tcpmem, nr_pages);
5714 memcg->tcpmem_pressure = 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005715 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08005716 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005717
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005718 /* Don't block in the packet receive path */
5719 if (in_softirq())
5720 gfp_mask = GFP_NOWAIT;
5721
Johannes Weinerb2807f02016-01-20 15:03:22 -08005722 this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5723
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005724 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5725 return true;
5726
5727 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005728 return false;
5729}
5730
5731/**
5732 * mem_cgroup_uncharge_skmem - uncharge socket memory
5733 * @memcg - memcg to uncharge
5734 * @nr_pages - number of pages to uncharge
5735 */
5736void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5737{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005738 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005739 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005740 return;
5741 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005742
Johannes Weinerb2807f02016-01-20 15:03:22 -08005743 this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5744
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005745 page_counter_uncharge(&memcg->memory, nr_pages);
5746 css_put_many(&memcg->css, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005747}
5748
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005749static int __init cgroup_memory(char *s)
5750{
5751 char *token;
5752
5753 while ((token = strsep(&s, ",")) != NULL) {
5754 if (!*token)
5755 continue;
5756 if (!strcmp(token, "nosocket"))
5757 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08005758 if (!strcmp(token, "nokmem"))
5759 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005760 }
5761 return 0;
5762}
5763__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08005764
Michal Hocko2d110852013-02-22 16:34:43 -08005765/*
Michal Hocko10813122013-02-22 16:35:41 -08005766 * subsys_initcall() for memory controller.
5767 *
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01005768 * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this
5769 * context because of lock dependencies (cgroup_lock -> cpu hotplug) but
5770 * basically everything that doesn't depend on a specific mem_cgroup structure
5771 * should be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005772 */
5773static int __init mem_cgroup_init(void)
5774{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005775 int cpu, node;
5776
Vladimir Davydov13583c32016-12-12 16:41:29 -08005777#ifndef CONFIG_SLOB
5778 /*
5779 * Kmem cache creation is mostly done with the slab_mutex held,
5780 * so use a special workqueue to avoid stalling all worker
5781 * threads in case lots of cgroups are created simultaneously.
5782 */
5783 memcg_kmem_cache_create_wq =
5784 alloc_ordered_workqueue("memcg_kmem_cache_create", 0);
5785 BUG_ON(!memcg_kmem_cache_create_wq);
5786#endif
5787
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01005788 cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL,
5789 memcg_hotplug_cpu_dead);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005790
5791 for_each_possible_cpu(cpu)
5792 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5793 drain_local_stock);
5794
5795 for_each_node(node) {
5796 struct mem_cgroup_tree_per_node *rtpn;
Johannes Weiner95a045f2015-02-11 15:26:33 -08005797
5798 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5799 node_online(node) ? node : NUMA_NO_NODE);
5800
Mel Gormanef8f2322016-07-28 15:46:05 -07005801 rtpn->rb_root = RB_ROOT;
5802 spin_lock_init(&rtpn->lock);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005803 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5804 }
5805
Michal Hocko2d110852013-02-22 16:34:43 -08005806 return 0;
5807}
5808subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005809
5810#ifdef CONFIG_MEMCG_SWAP
Arnd Bergmann358c07f2016-08-25 15:17:08 -07005811static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
5812{
5813 while (!atomic_inc_not_zero(&memcg->id.ref)) {
5814 /*
5815 * The root cgroup cannot be destroyed, so it's refcount must
5816 * always be >= 1.
5817 */
5818 if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
5819 VM_BUG_ON(1);
5820 break;
5821 }
5822 memcg = parent_mem_cgroup(memcg);
5823 if (!memcg)
5824 memcg = root_mem_cgroup;
5825 }
5826 return memcg;
5827}
5828
Johannes Weiner21afa382015-02-11 15:26:36 -08005829/**
5830 * mem_cgroup_swapout - transfer a memsw charge to swap
5831 * @page: page whose memsw charge to transfer
5832 * @entry: swap entry to move the charge to
5833 *
5834 * Transfer the memsw charge of @page to @entry.
5835 */
5836void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5837{
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005838 struct mem_cgroup *memcg, *swap_memcg;
Johannes Weiner21afa382015-02-11 15:26:36 -08005839 unsigned short oldid;
5840
5841 VM_BUG_ON_PAGE(PageLRU(page), page);
5842 VM_BUG_ON_PAGE(page_count(page), page);
5843
Johannes Weiner7941d212016-01-14 15:21:23 -08005844 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08005845 return;
5846
5847 memcg = page->mem_cgroup;
5848
5849 /* Readahead page, never charged */
5850 if (!memcg)
5851 return;
5852
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005853 /*
5854 * In case the memcg owning these pages has been offlined and doesn't
5855 * have an ID allocated to it anymore, charge the closest online
5856 * ancestor for the swap instead and transfer the memory+swap charge.
5857 */
5858 swap_memcg = mem_cgroup_id_get_online(memcg);
5859 oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg));
Johannes Weiner21afa382015-02-11 15:26:36 -08005860 VM_BUG_ON_PAGE(oldid, page);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005861 mem_cgroup_swap_statistics(swap_memcg, true);
Johannes Weiner21afa382015-02-11 15:26:36 -08005862
5863 page->mem_cgroup = NULL;
5864
5865 if (!mem_cgroup_is_root(memcg))
5866 page_counter_uncharge(&memcg->memory, 1);
5867
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005868 if (memcg != swap_memcg) {
5869 if (!mem_cgroup_is_root(swap_memcg))
5870 page_counter_charge(&swap_memcg->memsw, 1);
5871 page_counter_uncharge(&memcg->memsw, 1);
5872 }
5873
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07005874 /*
5875 * Interrupts should be disabled here because the caller holds the
5876 * mapping->tree_lock lock which is taken with interrupts-off. It is
5877 * important here to have the interrupts disabled because it is the
5878 * only synchronisation we have for udpating the per-CPU variables.
5879 */
5880 VM_BUG_ON(!irqs_disabled());
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005881 mem_cgroup_charge_statistics(memcg, page, false, -1);
Johannes Weiner21afa382015-02-11 15:26:36 -08005882 memcg_check_events(memcg, page);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005883
5884 if (!mem_cgroup_is_root(memcg))
5885 css_put(&memcg->css);
Johannes Weiner21afa382015-02-11 15:26:36 -08005886}
5887
Vladimir Davydov37e84352016-01-20 15:02:56 -08005888/*
5889 * mem_cgroup_try_charge_swap - try charging a swap entry
5890 * @page: page being added to swap
5891 * @entry: swap entry to charge
5892 *
5893 * Try to charge @entry to the memcg that @page belongs to.
5894 *
5895 * Returns 0 on success, -ENOMEM on failure.
5896 */
5897int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5898{
5899 struct mem_cgroup *memcg;
5900 struct page_counter *counter;
5901 unsigned short oldid;
5902
5903 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5904 return 0;
5905
5906 memcg = page->mem_cgroup;
5907
5908 /* Readahead page, never charged */
5909 if (!memcg)
5910 return 0;
5911
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005912 memcg = mem_cgroup_id_get_online(memcg);
Vladimir Davydov37e84352016-01-20 15:02:56 -08005913
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005914 if (!mem_cgroup_is_root(memcg) &&
5915 !page_counter_try_charge(&memcg->swap, 1, &counter)) {
5916 mem_cgroup_id_put(memcg);
5917 return -ENOMEM;
5918 }
5919
Vladimir Davydov37e84352016-01-20 15:02:56 -08005920 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5921 VM_BUG_ON_PAGE(oldid, page);
5922 mem_cgroup_swap_statistics(memcg, true);
5923
Vladimir Davydov37e84352016-01-20 15:02:56 -08005924 return 0;
5925}
5926
Johannes Weiner21afa382015-02-11 15:26:36 -08005927/**
5928 * mem_cgroup_uncharge_swap - uncharge a swap entry
5929 * @entry: swap entry to uncharge
5930 *
Vladimir Davydov37e84352016-01-20 15:02:56 -08005931 * Drop the swap charge associated with @entry.
Johannes Weiner21afa382015-02-11 15:26:36 -08005932 */
5933void mem_cgroup_uncharge_swap(swp_entry_t entry)
5934{
5935 struct mem_cgroup *memcg;
5936 unsigned short id;
5937
Vladimir Davydov37e84352016-01-20 15:02:56 -08005938 if (!do_swap_account)
Johannes Weiner21afa382015-02-11 15:26:36 -08005939 return;
5940
5941 id = swap_cgroup_record(entry, 0);
5942 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005943 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005944 if (memcg) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08005945 if (!mem_cgroup_is_root(memcg)) {
5946 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5947 page_counter_uncharge(&memcg->swap, 1);
5948 else
5949 page_counter_uncharge(&memcg->memsw, 1);
5950 }
Johannes Weiner21afa382015-02-11 15:26:36 -08005951 mem_cgroup_swap_statistics(memcg, false);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005952 mem_cgroup_id_put(memcg);
Johannes Weiner21afa382015-02-11 15:26:36 -08005953 }
5954 rcu_read_unlock();
5955}
5956
Vladimir Davydovd8b38432016-01-20 15:03:07 -08005957long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
5958{
5959 long nr_swap_pages = get_nr_swap_pages();
5960
5961 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5962 return nr_swap_pages;
5963 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5964 nr_swap_pages = min_t(long, nr_swap_pages,
5965 READ_ONCE(memcg->swap.limit) -
5966 page_counter_read(&memcg->swap));
5967 return nr_swap_pages;
5968}
5969
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08005970bool mem_cgroup_swap_full(struct page *page)
5971{
5972 struct mem_cgroup *memcg;
5973
5974 VM_BUG_ON_PAGE(!PageLocked(page), page);
5975
5976 if (vm_swap_full())
5977 return true;
5978 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5979 return false;
5980
5981 memcg = page->mem_cgroup;
5982 if (!memcg)
5983 return false;
5984
5985 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5986 if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
5987 return true;
5988
5989 return false;
5990}
5991
Johannes Weiner21afa382015-02-11 15:26:36 -08005992/* for remember boot option*/
5993#ifdef CONFIG_MEMCG_SWAP_ENABLED
5994static int really_do_swap_account __initdata = 1;
5995#else
5996static int really_do_swap_account __initdata;
5997#endif
5998
5999static int __init enable_swap_account(char *s)
6000{
6001 if (!strcmp(s, "1"))
6002 really_do_swap_account = 1;
6003 else if (!strcmp(s, "0"))
6004 really_do_swap_account = 0;
6005 return 1;
6006}
6007__setup("swapaccount=", enable_swap_account);
6008
Vladimir Davydov37e84352016-01-20 15:02:56 -08006009static u64 swap_current_read(struct cgroup_subsys_state *css,
6010 struct cftype *cft)
6011{
6012 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6013
6014 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
6015}
6016
6017static int swap_max_show(struct seq_file *m, void *v)
6018{
6019 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
6020 unsigned long max = READ_ONCE(memcg->swap.limit);
6021
6022 if (max == PAGE_COUNTER_MAX)
6023 seq_puts(m, "max\n");
6024 else
6025 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
6026
6027 return 0;
6028}
6029
6030static ssize_t swap_max_write(struct kernfs_open_file *of,
6031 char *buf, size_t nbytes, loff_t off)
6032{
6033 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6034 unsigned long max;
6035 int err;
6036
6037 buf = strstrip(buf);
6038 err = page_counter_memparse(buf, "max", &max);
6039 if (err)
6040 return err;
6041
6042 mutex_lock(&memcg_limit_mutex);
6043 err = page_counter_limit(&memcg->swap, max);
6044 mutex_unlock(&memcg_limit_mutex);
6045 if (err)
6046 return err;
6047
6048 return nbytes;
6049}
6050
6051static struct cftype swap_files[] = {
6052 {
6053 .name = "swap.current",
6054 .flags = CFTYPE_NOT_ON_ROOT,
6055 .read_u64 = swap_current_read,
6056 },
6057 {
6058 .name = "swap.max",
6059 .flags = CFTYPE_NOT_ON_ROOT,
6060 .seq_show = swap_max_show,
6061 .write = swap_max_write,
6062 },
6063 { } /* terminate */
6064};
6065
Johannes Weiner21afa382015-02-11 15:26:36 -08006066static struct cftype memsw_cgroup_files[] = {
6067 {
6068 .name = "memsw.usage_in_bytes",
6069 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6070 .read_u64 = mem_cgroup_read_u64,
6071 },
6072 {
6073 .name = "memsw.max_usage_in_bytes",
6074 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6075 .write = mem_cgroup_reset,
6076 .read_u64 = mem_cgroup_read_u64,
6077 },
6078 {
6079 .name = "memsw.limit_in_bytes",
6080 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6081 .write = mem_cgroup_write,
6082 .read_u64 = mem_cgroup_read_u64,
6083 },
6084 {
6085 .name = "memsw.failcnt",
6086 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6087 .write = mem_cgroup_reset,
6088 .read_u64 = mem_cgroup_read_u64,
6089 },
6090 { }, /* terminate */
6091};
6092
6093static int __init mem_cgroup_swap_init(void)
6094{
6095 if (!mem_cgroup_disabled() && really_do_swap_account) {
6096 do_swap_account = 1;
Vladimir Davydov37e84352016-01-20 15:02:56 -08006097 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
6098 swap_files));
Johannes Weiner21afa382015-02-11 15:26:36 -08006099 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
6100 memsw_cgroup_files));
6101 }
6102 return 0;
6103}
6104subsys_initcall(mem_cgroup_swap_init);
6105
6106#endif /* CONFIG_MEMCG_SWAP */