blob: 9a6a51a7c416074ccd5d974ac9f17a54c85d14db [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Johannes Weiner1575e682015-04-14 15:44:51 -070017 * Native page reclaim
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
22 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080023 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
27 *
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
32 */
33
Johannes Weiner3e32cb22014-12-10 15:42:31 -080034#include <linux/page_counter.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080035#include <linux/memcontrol.h>
36#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080037#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080038#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080039#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080040#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080041#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080043#include <linux/bit_spinlock.h>
44#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070045#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040046#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080047#include <linux/mutex.h>
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -070048#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070049#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080051#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080052#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080053#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050054#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080055#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080056#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080057#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070058#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070059#include <linux/mm_inline.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080060#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080061#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070062#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070063#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050064#include <linux/file.h>
Tejun Heob23afb92015-11-05 18:46:11 -080065#include <linux/tracehook.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080066#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000067#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070068#include <net/ip.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080069#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080070
Balbir Singh8697d332008-02-07 00:13:59 -080071#include <asm/uaccess.h>
72
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070073#include <trace/events/vmscan.h>
74
Tejun Heo073219e2014-02-08 10:36:58 -050075struct cgroup_subsys memory_cgrp_subsys __read_mostly;
76EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080077
Johannes Weiner7d828602016-01-14 15:20:56 -080078struct mem_cgroup *root_mem_cgroup __read_mostly;
79
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070080#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh8cdea7c2008-02-07 00:13:50 -080081
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080082/* Socket memory accounting disabled? */
83static bool cgroup_memory_nosocket;
84
Vladimir Davydov04823c82016-01-20 15:02:38 -080085/* Kernel memory accounting disabled? */
86static bool cgroup_memory_nokmem;
87
Johannes Weiner21afa382015-02-11 15:26:36 -080088/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070089#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080090int do_swap_account __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080091#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070092#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080093#endif
94
Johannes Weiner7941d212016-01-14 15:21:23 -080095/* Whether legacy memory+swap accounting is active */
96static bool do_memsw_account(void)
97{
98 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account;
99}
100
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700101static const char * const mem_cgroup_stat_names[] = {
102 "cache",
103 "rss",
David Rientjesb070e652013-05-07 16:18:09 -0700104 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700105 "mapped_file",
Greg Thelenc4843a72015-05-22 17:13:16 -0400106 "dirty",
Sha Zhengju3ea67d02013-09-12 15:13:53 -0700107 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700108 "swap",
109};
110
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700111static const char * const mem_cgroup_events_names[] = {
112 "pgpgin",
113 "pgpgout",
114 "pgfault",
115 "pgmajfault",
116};
117
Sha Zhengju58cf1882013-02-22 16:32:05 -0800118static const char * const mem_cgroup_lru_names[] = {
119 "inactive_anon",
120 "active_anon",
121 "inactive_file",
122 "active_file",
123 "unevictable",
124};
125
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700126#define THRESHOLDS_EVENTS_TARGET 128
127#define SOFTLIMIT_EVENTS_TARGET 1024
128#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700129
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700130/*
131 * Cgroups above their limits are maintained in a RB-Tree, independent of
132 * their hierarchy representation
133 */
134
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 Hiroyukie401f172011-01-20 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 Hiroyukie401f172011-01-20 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{
Jianyu Zhane2318752014-06-06 14:38:20 -0700628 unsigned long nr = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -0700629 struct mem_cgroup_per_node *mz;
630 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;
637 mz = mem_cgroup_nodeinfo(memcg, nid);
638 nr += mz->lru_size[lru];
Jianyu Zhane2318752014-06-06 14:38:20 -0700639 }
640 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700641}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700642
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700643static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700644 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800645{
Jianyu Zhane2318752014-06-06 14:38:20 -0700646 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700647 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800648
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800649 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700650 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
651 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800652}
653
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800654static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
655 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800656{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700657 unsigned long val, next;
658
Johannes Weiner13114712012-05-29 15:07:07 -0700659 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700660 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700661 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800662 if ((long)next - (long)val < 0) {
663 switch (target) {
664 case MEM_CGROUP_TARGET_THRESH:
665 next = val + THRESHOLDS_EVENTS_TARGET;
666 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700667 case MEM_CGROUP_TARGET_SOFTLIMIT:
668 next = val + SOFTLIMIT_EVENTS_TARGET;
669 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800670 case MEM_CGROUP_TARGET_NUMAINFO:
671 next = val + NUMAINFO_EVENTS_TARGET;
672 break;
673 default:
674 break;
675 }
676 __this_cpu_write(memcg->stat->targets[target], next);
677 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700678 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800679 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800680}
681
682/*
683 * Check events in order.
684 *
685 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700686static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800687{
688 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800689 if (unlikely(mem_cgroup_event_ratelimit(memcg,
690 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700691 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800692 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800693
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700694 do_softlimit = mem_cgroup_event_ratelimit(memcg,
695 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700696#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800697 do_numainfo = mem_cgroup_event_ratelimit(memcg,
698 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700699#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800700 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700701 if (unlikely(do_softlimit))
702 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800703#if MAX_NUMNODES > 1
704 if (unlikely(do_numainfo))
705 atomic_inc(&memcg->numainfo_events);
706#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700707 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800708}
709
Balbir Singhcf475ad2008-04-29 01:00:16 -0700710struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800711{
Balbir Singh31a78f22008-09-28 23:09:31 +0100712 /*
713 * mm_update_next_owner() may clear mm->owner to NULL
714 * if it races with swapoff, page migration, etc.
715 * So this can be called with p == NULL.
716 */
717 if (unlikely(!p))
718 return NULL;
719
Tejun Heo073219e2014-02-08 10:36:58 -0500720 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800721}
Michal Hocko33398cf2015-09-08 15:01:02 -0700722EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800723
Johannes Weinerdf381972014-04-07 15:37:43 -0700724static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800725{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700726 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700727
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800728 rcu_read_lock();
729 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700730 /*
731 * Page cache insertions can happen withou an
732 * actual mm context, e.g. during disk probing
733 * on boot, loopback IO, acct() writes etc.
734 */
735 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700736 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700737 else {
738 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
739 if (unlikely(!memcg))
740 memcg = root_mem_cgroup;
741 }
Tejun Heoec903c02014-05-13 12:11:01 -0400742 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800743 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700744 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800745}
746
Johannes Weiner56600482012-01-12 17:17:59 -0800747/**
748 * mem_cgroup_iter - iterate over memory cgroup hierarchy
749 * @root: hierarchy root
750 * @prev: previously returned memcg, NULL on first invocation
751 * @reclaim: cookie for shared reclaim walks, NULL for full walks
752 *
753 * Returns references to children of the hierarchy below @root, or
754 * @root itself, or %NULL after a full round-trip.
755 *
756 * Caller must pass the return value in @prev on subsequent
757 * invocations for reference counting, or use mem_cgroup_iter_break()
758 * to cancel a hierarchy walk before the round-trip is complete.
759 *
760 * Reclaimers can specify a zone and a priority level in @reclaim to
761 * divide up the memcgs in the hierarchy among all concurrent
762 * reclaimers operating on the same zone and priority.
763 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700764struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800765 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700766 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700767{
Michal Hocko33398cf2015-09-08 15:01:02 -0700768 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800769 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800770 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800771 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700772
Andrew Morton694fbc02013-09-24 15:27:37 -0700773 if (mem_cgroup_disabled())
774 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800775
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700776 if (!root)
777 root = root_mem_cgroup;
778
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800779 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800780 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800781
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800782 if (!root->use_hierarchy && root != root_mem_cgroup) {
783 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800784 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700785 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800786 }
787
Michal Hocko542f85f2013-04-29 15:07:15 -0700788 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800789
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800790 if (reclaim) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700791 struct mem_cgroup_per_node *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800792
Mel Gormanef8f2322016-07-28 15:46:05 -0700793 mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800794 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700795
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800796 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700797 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800798
Vladimir Davydov6df38682015-12-29 14:54:10 -0800799 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -0700800 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800801 if (!pos || css_tryget(&pos->css))
802 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800803 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800804 * css reference reached zero, so iter->position will
805 * be cleared by ->css_released. However, we should not
806 * rely on this happening soon, because ->css_released
807 * is called from a work queue, and by busy-waiting we
808 * might block it. So we clear iter->position right
809 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800810 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800811 (void)cmpxchg(&iter->position, pos, NULL);
812 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800813 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800814
815 if (pos)
816 css = &pos->css;
817
818 for (;;) {
819 css = css_next_descendant_pre(css, &root->css);
820 if (!css) {
821 /*
822 * Reclaimers share the hierarchy walk, and a
823 * new one might jump in right at the end of
824 * the hierarchy - make sure they see at least
825 * one group and restart from the beginning.
826 */
827 if (!prev)
828 continue;
829 break;
830 }
831
832 /*
833 * Verify the css and acquire a reference. The root
834 * is provided by the caller, so we know it's alive
835 * and kicking, and don't take an extra reference.
836 */
837 memcg = mem_cgroup_from_css(css);
838
839 if (css == &root->css)
840 break;
841
Johannes Weiner0b8f73e2016-01-20 15:02:53 -0800842 if (css_tryget(css))
843 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800844
845 memcg = NULL;
846 }
847
848 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800849 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800850 * The position could have already been updated by a competing
851 * thread, so check that the value hasn't changed since we read
852 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800853 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800854 (void)cmpxchg(&iter->position, pos, memcg);
855
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800856 if (pos)
857 css_put(&pos->css);
858
859 if (!memcg)
860 iter->generation++;
861 else if (!prev)
862 reclaim->generation = iter->generation;
863 }
864
Michal Hocko542f85f2013-04-29 15:07:15 -0700865out_unlock:
866 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800867out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700868 if (prev && prev != root)
869 css_put(&prev->css);
870
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800871 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700872}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800873
Johannes Weiner56600482012-01-12 17:17:59 -0800874/**
875 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
876 * @root: hierarchy root
877 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
878 */
879void mem_cgroup_iter_break(struct mem_cgroup *root,
880 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800881{
882 if (!root)
883 root = root_mem_cgroup;
884 if (prev && prev != root)
885 css_put(&prev->css);
886}
887
Vladimir Davydov6df38682015-12-29 14:54:10 -0800888static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
889{
890 struct mem_cgroup *memcg = dead_memcg;
891 struct mem_cgroup_reclaim_iter *iter;
Mel Gormanef8f2322016-07-28 15:46:05 -0700892 struct mem_cgroup_per_node *mz;
893 int nid;
Vladimir Davydov6df38682015-12-29 14:54:10 -0800894 int i;
895
896 while ((memcg = parent_mem_cgroup(memcg))) {
897 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700898 mz = mem_cgroup_nodeinfo(memcg, nid);
899 for (i = 0; i <= DEF_PRIORITY; i++) {
900 iter = &mz->iter[i];
901 cmpxchg(&iter->position,
902 dead_memcg, NULL);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800903 }
904 }
905 }
906}
907
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700908/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800909 * Iteration constructs for visiting all cgroups (under a tree). If
910 * loops are exited prematurely (break), mem_cgroup_iter_break() must
911 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700912 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800913#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800914 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800915 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800916 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700917
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800918#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800919 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800920 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800921 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700922
Johannes Weiner925b7672012-01-12 17:18:15 -0800923/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800924 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -0800925 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700926 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800927 *
928 * This function is only safe when following the LRU page isolation
929 * and putback protocol: the LRU lock must be held, and the page must
930 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -0700931 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700932struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat)
Minchan Kim3f58a822011-03-22 16:32:53 -0700933{
Mel Gormanef8f2322016-07-28 15:46:05 -0700934 struct mem_cgroup_per_node *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -0800935 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800936 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800937
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800938 if (mem_cgroup_disabled()) {
Mel Gorman599d0c92016-07-28 15:45:31 -0700939 lruvec = &pgdat->lruvec;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800940 goto out;
941 }
Christoph Lameterb69408e2008-10-18 20:26:14 -0700942
Johannes Weiner1306a852014-12-10 15:44:52 -0800943 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800944 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800945 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -0800946 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -0800947 */
Johannes Weiner29833312014-12-10 15:44:02 -0800948 if (!memcg)
949 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800950
Mel Gormanef8f2322016-07-28 15:46:05 -0700951 mz = mem_cgroup_page_nodeinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800952 lruvec = &mz->lruvec;
953out:
954 /*
955 * Since a node can be onlined after the mem_cgroup was created,
956 * we have to be prepared to initialize lruvec->zone here;
957 * and if offlined then reonlined, we need to reinitialize it.
958 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700959 if (unlikely(lruvec->pgdat != pgdat))
960 lruvec->pgdat = pgdat;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800961 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -0800962}
963
964/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700965 * mem_cgroup_update_lru_size - account for adding or removing an lru page
966 * @lruvec: mem_cgroup per zone lru vector
967 * @lru: index of lru list the page is sitting on
968 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -0800969 *
Hugh Dickinsca707232016-05-19 17:12:35 -0700970 * This function must be called under lru_lock, just before a page is added
971 * to or just after a page is removed from an lru list (that ordering being
972 * so as to allow it to check that lru_size 0 is consistent with list_empty).
Johannes Weiner925b7672012-01-12 17:18:15 -0800973 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700974void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
Mel Gorman7ee36a12016-07-28 15:47:17 -0700975 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -0800976{
Mel Gormanef8f2322016-07-28 15:46:05 -0700977 struct mem_cgroup_per_node *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700978 unsigned long *lru_size;
Hugh Dickinsca707232016-05-19 17:12:35 -0700979 long size;
980 bool empty;
Johannes Weiner925b7672012-01-12 17:18:15 -0800981
982 if (mem_cgroup_disabled())
983 return;
984
Mel Gormanef8f2322016-07-28 15:46:05 -0700985 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700986 lru_size = mz->lru_size + lru;
Hugh Dickinsca707232016-05-19 17:12:35 -0700987 empty = list_empty(lruvec->lists + lru);
988
989 if (nr_pages < 0)
990 *lru_size += nr_pages;
991
992 size = *lru_size;
993 if (WARN_ONCE(size < 0 || empty != !size,
994 "%s(%p, %d, %d): lru_size %ld but %sempty\n",
995 __func__, lruvec, lru, nr_pages, size, empty ? "" : "not ")) {
996 VM_BUG_ON(1);
997 *lru_size = 0;
998 }
999
1000 if (nr_pages > 0)
1001 *lru_size += nr_pages;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001002}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001003
Johannes Weiner2314b422014-12-10 15:44:33 -08001004bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001005{
Johannes Weiner2314b422014-12-10 15:44:33 -08001006 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001007 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001008 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001009
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001010 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001011 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001012 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001013 task_unlock(p);
1014 } else {
1015 /*
1016 * All threads may have already detached their mm's, but the oom
1017 * killer still needs to detect if they have already been oom
1018 * killed to prevent needlessly killing additional tasks.
1019 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001020 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001021 task_memcg = mem_cgroup_from_task(task);
1022 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001023 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001024 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001025 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1026 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001027 return ret;
1028}
1029
Johannes Weiner19942822011-02-01 15:52:43 -08001030/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001031 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001032 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001033 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001034 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001035 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001036 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001037static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001038{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001039 unsigned long margin = 0;
1040 unsigned long count;
1041 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001042
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001043 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001044 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001045 if (count < limit)
1046 margin = limit - count;
1047
Johannes Weiner7941d212016-01-14 15:21:23 -08001048 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001049 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001050 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001051 if (count <= limit)
1052 margin = min(margin, limit - count);
Li RongQingcbedbac2016-05-27 14:27:43 -07001053 else
1054 margin = 0;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001055 }
1056
1057 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001058}
1059
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001060/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001061 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001062 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001063 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1064 * moving cgroups. This is for waiting at high-memory pressure
1065 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001066 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001067static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001068{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001069 struct mem_cgroup *from;
1070 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001071 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001072 /*
1073 * Unlike task_move routines, we access mc.to, mc.from not under
1074 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1075 */
1076 spin_lock(&mc.lock);
1077 from = mc.from;
1078 to = mc.to;
1079 if (!from)
1080 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001081
Johannes Weiner2314b422014-12-10 15:44:33 -08001082 ret = mem_cgroup_is_descendant(from, memcg) ||
1083 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001084unlock:
1085 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001086 return ret;
1087}
1088
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001089static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001090{
1091 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001092 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001093 DEFINE_WAIT(wait);
1094 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1095 /* moving charge context might have finished. */
1096 if (mc.moving_task)
1097 schedule();
1098 finish_wait(&mc.waitq, &wait);
1099 return true;
1100 }
1101 }
1102 return false;
1103}
1104
Sha Zhengju58cf1882013-02-22 16:32:05 -08001105#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001106/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001107 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001108 * @memcg: The memory cgroup that went over limit
1109 * @p: Task that is going to be killed
1110 *
1111 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1112 * enabled
1113 */
1114void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1115{
Sha Zhengju58cf1882013-02-22 16:32:05 -08001116 struct mem_cgroup *iter;
1117 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001118
Balbir Singhe2224322009-04-02 16:57:39 -07001119 rcu_read_lock();
1120
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001121 if (p) {
1122 pr_info("Task in ");
1123 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1124 pr_cont(" killed as a result of limit of ");
1125 } else {
1126 pr_info("Memory limit reached of cgroup ");
1127 }
1128
Tejun Heoe61734c2014-02-12 09:29:50 -05001129 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001130 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001131
Balbir Singhe2224322009-04-02 16:57:39 -07001132 rcu_read_unlock();
1133
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001134 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1135 K((u64)page_counter_read(&memcg->memory)),
1136 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1137 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1138 K((u64)page_counter_read(&memcg->memsw)),
1139 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1140 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1141 K((u64)page_counter_read(&memcg->kmem)),
1142 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001143
1144 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001145 pr_info("Memory cgroup stats for ");
1146 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001147 pr_cont(":");
1148
1149 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08001150 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Sha Zhengju58cf1882013-02-22 16:32:05 -08001151 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07001152 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
Sha Zhengju58cf1882013-02-22 16:32:05 -08001153 K(mem_cgroup_read_stat(iter, i)));
1154 }
1155
1156 for (i = 0; i < NR_LRU_LISTS; i++)
1157 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1158 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1159
1160 pr_cont("\n");
1161 }
Balbir Singhe2224322009-04-02 16:57:39 -07001162}
1163
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001164/*
1165 * This function returns the number of memcg under hierarchy tree. Returns
1166 * 1(self count) if no children.
1167 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001168static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001169{
1170 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001171 struct mem_cgroup *iter;
1172
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001173 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001174 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001175 return num;
1176}
1177
Balbir Singh6d61ef42009-01-07 18:08:06 -08001178/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001179 * Return the memory (and swap, if configured) limit for a memcg.
1180 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001181static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001182{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001183 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001184
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001185 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001186 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001187 unsigned long memsw_limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001188 unsigned long swap_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001189
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001190 memsw_limit = memcg->memsw.limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001191 swap_limit = memcg->swap.limit;
1192 swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1193 limit = min(limit + swap_limit, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001194 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001195 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001196}
1197
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001198static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001199 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001200{
David Rientjes6e0fc462015-09-08 15:00:36 -07001201 struct oom_control oc = {
1202 .zonelist = NULL,
1203 .nodemask = NULL,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001204 .memcg = memcg,
David Rientjes6e0fc462015-09-08 15:00:36 -07001205 .gfp_mask = gfp_mask,
1206 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001207 };
David Rientjes9cbb78b2012-07-31 16:43:44 -07001208 struct mem_cgroup *iter;
1209 unsigned long chosen_points = 0;
1210 unsigned long totalpages;
1211 unsigned int points = 0;
1212 struct task_struct *chosen = NULL;
1213
Johannes Weinerdc564012015-06-24 16:57:19 -07001214 mutex_lock(&oom_lock);
1215
David Rientjes876aafb2012-07-31 16:43:48 -07001216 /*
David Rientjes465adcf2013-04-29 15:08:45 -07001217 * If current has a pending SIGKILL or is exiting, then automatically
1218 * select it. The goal is to allow it to allocate so that it may
1219 * quickly exit and free its memory.
David Rientjes876aafb2012-07-31 16:43:48 -07001220 */
Michal Hocko1af8bb42016-07-28 15:44:52 -07001221 if (task_will_free_mem(current)) {
Johannes Weiner16e95192015-06-24 16:57:07 -07001222 mark_oom_victim(current);
Michal Hocko1af8bb42016-07-28 15:44:52 -07001223 wake_oom_reaper(current);
Johannes Weinerdc564012015-06-24 16:57:19 -07001224 goto unlock;
David Rientjes876aafb2012-07-31 16:43:48 -07001225 }
1226
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001227 check_panic_on_oom(&oc, CONSTRAINT_MEMCG);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001228 totalpages = mem_cgroup_get_limit(memcg) ? : 1;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001229 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heo72ec7022013-08-08 20:11:26 -04001230 struct css_task_iter it;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001231 struct task_struct *task;
1232
Tejun Heo72ec7022013-08-08 20:11:26 -04001233 css_task_iter_start(&iter->css, &it);
1234 while ((task = css_task_iter_next(&it))) {
Tetsuo Handafbe84a02016-07-26 15:24:39 -07001235 switch (oom_scan_process_thread(&oc, task)) {
David Rientjes9cbb78b2012-07-31 16:43:44 -07001236 case OOM_SCAN_SELECT:
1237 if (chosen)
1238 put_task_struct(chosen);
1239 chosen = task;
1240 chosen_points = ULONG_MAX;
1241 get_task_struct(chosen);
1242 /* fall through */
1243 case OOM_SCAN_CONTINUE:
1244 continue;
1245 case OOM_SCAN_ABORT:
Tejun Heo72ec7022013-08-08 20:11:26 -04001246 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001247 mem_cgroup_iter_break(memcg, iter);
1248 if (chosen)
1249 put_task_struct(chosen);
Tetsuo Handa1ebab2d2016-05-26 15:16:19 -07001250 /* Set a dummy value to return "true". */
1251 chosen = (void *) 1;
Johannes Weinerdc564012015-06-24 16:57:19 -07001252 goto unlock;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001253 case OOM_SCAN_OK:
1254 break;
1255 };
1256 points = oom_badness(task, memcg, NULL, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -08001257 if (!points || points < chosen_points)
1258 continue;
1259 /* Prefer thread group leaders for display purposes */
1260 if (points == chosen_points &&
1261 thread_group_leader(chosen))
1262 continue;
1263
1264 if (chosen)
1265 put_task_struct(chosen);
1266 chosen = task;
1267 chosen_points = points;
1268 get_task_struct(chosen);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001269 }
Tejun Heo72ec7022013-08-08 20:11:26 -04001270 css_task_iter_end(&it);
David Rientjes9cbb78b2012-07-31 16:43:44 -07001271 }
1272
Johannes Weinerdc564012015-06-24 16:57:19 -07001273 if (chosen) {
1274 points = chosen_points * 1000 / totalpages;
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001275 oom_kill_process(&oc, chosen, points, totalpages,
David Rientjes6e0fc462015-09-08 15:00:36 -07001276 "Memory cgroup out of memory");
Johannes Weinerdc564012015-06-24 16:57:19 -07001277 }
1278unlock:
1279 mutex_unlock(&oom_lock);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001280 return chosen;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001281}
1282
Michele Curtiae6e71d2014-12-12 16:56:35 -08001283#if MAX_NUMNODES > 1
1284
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001285/**
1286 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001287 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001288 * @nid: the node ID to be checked.
1289 * @noswap : specify true here if the user wants flle only information.
1290 *
1291 * This function returns whether the specified memcg contains any
1292 * reclaimable pages on a node. Returns true if there are any reclaimable
1293 * pages in the node.
1294 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001295static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001296 int nid, bool noswap)
1297{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001298 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001299 return true;
1300 if (noswap || !total_swap_pages)
1301 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001302 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001303 return true;
1304 return false;
1305
1306}
Ying Han889976d2011-05-26 16:25:33 -07001307
1308/*
1309 * Always updating the nodemask is not very good - even if we have an empty
1310 * list or the wrong list here, we can start from some node and traverse all
1311 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1312 *
1313 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001314static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001315{
1316 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001317 /*
1318 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1319 * pagein/pageout changes since the last update.
1320 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001321 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001322 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001323 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001324 return;
1325
Ying Han889976d2011-05-26 16:25:33 -07001326 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001327 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001328
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001329 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001330
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001331 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1332 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001333 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001334
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001335 atomic_set(&memcg->numainfo_events, 0);
1336 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001337}
1338
1339/*
1340 * Selecting a node where we start reclaim from. Because what we need is just
1341 * reducing usage counter, start from anywhere is O,K. Considering
1342 * memory reclaim from current node, there are pros. and cons.
1343 *
1344 * Freeing memory from current node means freeing memory from a node which
1345 * we'll use or we've used. So, it may make LRU bad. And if several threads
1346 * hit limits, it will see a contention on a node. But freeing from remote
1347 * node means more costs for memory reclaim because of memory latency.
1348 *
1349 * Now, we use round-robin. Better algorithm is welcomed.
1350 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001351int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001352{
1353 int node;
1354
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001355 mem_cgroup_may_update_nodemask(memcg);
1356 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001357
Andrew Morton0edaf862016-05-19 17:10:58 -07001358 node = next_node_in(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001359 /*
Michal Hockofda3d692016-05-19 17:11:34 -07001360 * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1361 * last time it really checked all the LRUs due to rate limiting.
1362 * Fallback to the current node in that case for simplicity.
Ying Han889976d2011-05-26 16:25:33 -07001363 */
1364 if (unlikely(node == MAX_NUMNODES))
1365 node = numa_node_id();
1366
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001367 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001368 return node;
1369}
Ying Han889976d2011-05-26 16:25:33 -07001370#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001371int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001372{
1373 return 0;
1374}
1375#endif
1376
Andrew Morton0608f432013-09-24 15:27:41 -07001377static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
Mel Gormanef8f2322016-07-28 15:46:05 -07001378 pg_data_t *pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001379 gfp_t gfp_mask,
1380 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001381{
Andrew Morton0608f432013-09-24 15:27:41 -07001382 struct mem_cgroup *victim = NULL;
1383 int total = 0;
1384 int loop = 0;
1385 unsigned long excess;
1386 unsigned long nr_scanned;
1387 struct mem_cgroup_reclaim_cookie reclaim = {
Mel Gormanef8f2322016-07-28 15:46:05 -07001388 .pgdat = pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001389 .priority = 0,
1390 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001391
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001392 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001393
Andrew Morton0608f432013-09-24 15:27:41 -07001394 while (1) {
1395 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1396 if (!victim) {
1397 loop++;
1398 if (loop >= 2) {
1399 /*
1400 * If we have not been able to reclaim
1401 * anything, it might because there are
1402 * no reclaimable pages under this hierarchy
1403 */
1404 if (!total)
1405 break;
1406 /*
1407 * We want to do more targeted reclaim.
1408 * excess >> 2 is not to excessive so as to
1409 * reclaim too much, nor too less that we keep
1410 * coming back to reclaim from this cgroup
1411 */
1412 if (total >= (excess >> 2) ||
1413 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1414 break;
1415 }
1416 continue;
1417 }
Mel Gormana9dd0a82016-07-28 15:46:02 -07001418 total += mem_cgroup_shrink_node(victim, gfp_mask, false,
Mel Gormanef8f2322016-07-28 15:46:05 -07001419 pgdat, &nr_scanned);
Andrew Morton0608f432013-09-24 15:27:41 -07001420 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001421 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001422 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001423 }
Andrew Morton0608f432013-09-24 15:27:41 -07001424 mem_cgroup_iter_break(root_memcg, victim);
1425 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001426}
1427
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001428#ifdef CONFIG_LOCKDEP
1429static struct lockdep_map memcg_oom_lock_dep_map = {
1430 .name = "memcg_oom_lock",
1431};
1432#endif
1433
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001434static DEFINE_SPINLOCK(memcg_oom_lock);
1435
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001436/*
1437 * Check OOM-Killer is already running under our hierarchy.
1438 * If someone is running, return false.
1439 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001440static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001441{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001442 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001443
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001444 spin_lock(&memcg_oom_lock);
1445
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001446 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001447 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001448 /*
1449 * this subtree of our hierarchy is already locked
1450 * so we cannot give a lock.
1451 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001452 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001453 mem_cgroup_iter_break(memcg, iter);
1454 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001455 } else
1456 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001457 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001458
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001459 if (failed) {
1460 /*
1461 * OK, we failed to lock the whole subtree so we have
1462 * to clean up what we set up to the failing subtree
1463 */
1464 for_each_mem_cgroup_tree(iter, memcg) {
1465 if (iter == failed) {
1466 mem_cgroup_iter_break(memcg, iter);
1467 break;
1468 }
1469 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001470 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001471 } else
1472 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001473
1474 spin_unlock(&memcg_oom_lock);
1475
1476 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001477}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001478
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001479static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001480{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001481 struct mem_cgroup *iter;
1482
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001483 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001484 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001485 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001486 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001487 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001488}
1489
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001490static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001491{
1492 struct mem_cgroup *iter;
1493
Tejun Heoc2b42d32015-06-24 16:58:23 -07001494 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001495 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001496 iter->under_oom++;
1497 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001498}
1499
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001500static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001501{
1502 struct mem_cgroup *iter;
1503
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001504 /*
1505 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001506 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001507 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001508 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001509 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001510 if (iter->under_oom > 0)
1511 iter->under_oom--;
1512 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001513}
1514
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001515static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1516
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001517struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001518 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001519 wait_queue_t wait;
1520};
1521
1522static int memcg_oom_wake_function(wait_queue_t *wait,
1523 unsigned mode, int sync, void *arg)
1524{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001525 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1526 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001527 struct oom_wait_info *oom_wait_info;
1528
1529 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001530 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001531
Johannes Weiner2314b422014-12-10 15:44:33 -08001532 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1533 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001534 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001535 return autoremove_wake_function(wait, mode, sync, arg);
1536}
1537
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001538static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001539{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001540 /*
1541 * For the following lockless ->under_oom test, the only required
1542 * guarantee is that it must see the state asserted by an OOM when
1543 * this function is called as a result of userland actions
1544 * triggered by the notification of the OOM. This is trivially
1545 * achieved by invoking mem_cgroup_mark_under_oom() before
1546 * triggering notification.
1547 */
1548 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001549 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001550}
1551
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001552static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001553{
Andrew Mortond0db7af2016-06-08 15:33:47 -07001554 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001555 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001556 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001557 * We are in the middle of the charge context here, so we
1558 * don't want to block when potentially sitting on a callstack
1559 * that holds all kinds of filesystem and mm locks.
1560 *
1561 * Also, the caller may handle a failed allocation gracefully
1562 * (like optional page cache readahead) and so an OOM killer
1563 * invocation might not even be necessary.
1564 *
1565 * That's why we don't do anything here except remember the
1566 * OOM context and then deal with it at the end of the page
1567 * fault when the stack is unwound, the locks are released,
1568 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001569 */
Johannes Weiner49426422013-10-16 13:46:59 -07001570 css_get(&memcg->css);
Tejun Heo626ebc42015-11-05 18:46:09 -08001571 current->memcg_in_oom = memcg;
1572 current->memcg_oom_gfp_mask = mask;
1573 current->memcg_oom_order = order;
Johannes Weiner49426422013-10-16 13:46:59 -07001574}
1575
1576/**
1577 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1578 * @handle: actually kill/wait or just clean up the OOM state
1579 *
1580 * This has to be called at the end of a page fault if the memcg OOM
1581 * handler was enabled.
1582 *
1583 * Memcg supports userspace OOM handling where failed allocations must
1584 * sleep on a waitqueue until the userspace task resolves the
1585 * situation. Sleeping directly in the charge context with all kinds
1586 * of locks held is not a good idea, instead we remember an OOM state
1587 * in the task and mem_cgroup_oom_synchronize() has to be called at
1588 * the end of the page fault to complete the OOM handling.
1589 *
1590 * Returns %true if an ongoing memcg OOM situation was detected and
1591 * completed, %false otherwise.
1592 */
1593bool mem_cgroup_oom_synchronize(bool handle)
1594{
Tejun Heo626ebc42015-11-05 18:46:09 -08001595 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001596 struct oom_wait_info owait;
1597 bool locked;
1598
1599 /* OOM is global, do not handle */
1600 if (!memcg)
1601 return false;
1602
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001603 if (!handle || oom_killer_disabled)
Johannes Weiner49426422013-10-16 13:46:59 -07001604 goto cleanup;
1605
1606 owait.memcg = memcg;
1607 owait.wait.flags = 0;
1608 owait.wait.func = memcg_oom_wake_function;
1609 owait.wait.private = current;
1610 INIT_LIST_HEAD(&owait.wait.task_list);
1611
1612 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001613 mem_cgroup_mark_under_oom(memcg);
1614
1615 locked = mem_cgroup_oom_trylock(memcg);
1616
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001617 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001618 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001619
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001620 if (locked && !memcg->oom_kill_disable) {
1621 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001622 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001623 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1624 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001625 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001626 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001627 mem_cgroup_unmark_under_oom(memcg);
1628 finish_wait(&memcg_oom_waitq, &owait.wait);
1629 }
1630
1631 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001632 mem_cgroup_oom_unlock(memcg);
1633 /*
1634 * There is no guarantee that an OOM-lock contender
1635 * sees the wakeups triggered by the OOM kill
1636 * uncharges. Wake any sleepers explicitely.
1637 */
1638 memcg_oom_recover(memcg);
1639 }
Johannes Weiner49426422013-10-16 13:46:59 -07001640cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001641 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001642 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001643 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001644}
1645
Johannes Weinerd7365e72014-10-29 14:50:48 -07001646/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001647 * lock_page_memcg - lock a page->mem_cgroup binding
1648 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001649 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001650 * This function protects unlocked LRU pages from being moved to
1651 * another cgroup and stabilizes their page->mem_cgroup binding.
Balbir Singhd69b0422009-06-17 16:26:34 -07001652 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001653void lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001654{
1655 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001656 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001657
Johannes Weiner6de22612015-02-11 15:25:01 -08001658 /*
1659 * The RCU lock is held throughout the transaction. The fast
1660 * path can get away without acquiring the memcg->move_lock
1661 * because page moving starts with an RCU grace period.
Johannes Weiner6de22612015-02-11 15:25:01 -08001662 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001663 rcu_read_lock();
1664
1665 if (mem_cgroup_disabled())
Johannes Weiner62cccb82016-03-15 14:57:22 -07001666 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001667again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001668 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001669 if (unlikely(!memcg))
Johannes Weiner62cccb82016-03-15 14:57:22 -07001670 return;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001671
Qiang Huangbdcbb652014-06-04 16:08:21 -07001672 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001673 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001674
Johannes Weiner6de22612015-02-11 15:25:01 -08001675 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001676 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001677 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001678 goto again;
1679 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001680
1681 /*
1682 * When charge migration first begins, we can have locked and
1683 * unlocked page stat updates happening concurrently. Track
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001684 * the task who has the lock for unlock_page_memcg().
Johannes Weiner6de22612015-02-11 15:25:01 -08001685 */
1686 memcg->move_lock_task = current;
1687 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001688
Johannes Weiner62cccb82016-03-15 14:57:22 -07001689 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001690}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001691EXPORT_SYMBOL(lock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001692
Johannes Weinerd7365e72014-10-29 14:50:48 -07001693/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001694 * unlock_page_memcg - unlock a page->mem_cgroup binding
Johannes Weiner62cccb82016-03-15 14:57:22 -07001695 * @page: the page
Johannes Weinerd7365e72014-10-29 14:50:48 -07001696 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001697void unlock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001698{
Johannes Weiner62cccb82016-03-15 14:57:22 -07001699 struct mem_cgroup *memcg = page->mem_cgroup;
1700
Johannes Weiner6de22612015-02-11 15:25:01 -08001701 if (memcg && memcg->move_lock_task == current) {
1702 unsigned long flags = memcg->move_lock_flags;
1703
1704 memcg->move_lock_task = NULL;
1705 memcg->move_lock_flags = 0;
1706
1707 spin_unlock_irqrestore(&memcg->move_lock, flags);
1708 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001709
Johannes Weinerd7365e72014-10-29 14:50:48 -07001710 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001711}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001712EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001713
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001714/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001715 * size of first charge trial. "32" comes from vmscan.c's magic value.
1716 * TODO: maybe necessary to use big numbers in big irons.
1717 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001718#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001719struct memcg_stock_pcp {
1720 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001721 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001722 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001723 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001724#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001725};
1726static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001727static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001728
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001729/**
1730 * consume_stock: Try to consume stocked charge on this cpu.
1731 * @memcg: memcg to consume from.
1732 * @nr_pages: how many pages to charge.
1733 *
1734 * The charges will only happen if @memcg matches the current cpu's memcg
1735 * stock, and at least @nr_pages are available in that stock. Failure to
1736 * service an allocation will refill the stock.
1737 *
1738 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001739 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001740static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001741{
1742 struct memcg_stock_pcp *stock;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001743 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001744
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001745 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001746 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001747
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001748 stock = &get_cpu_var(memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001749 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001750 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001751 ret = true;
1752 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001753 put_cpu_var(memcg_stock);
1754 return ret;
1755}
1756
1757/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001758 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001759 */
1760static void drain_stock(struct memcg_stock_pcp *stock)
1761{
1762 struct mem_cgroup *old = stock->cached;
1763
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001764 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001765 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001766 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001767 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001768 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001769 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001770 }
1771 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001772}
1773
1774/*
1775 * This must be called under preempt disabled or must be called by
1776 * a thread which is pinned to local cpu.
1777 */
1778static void drain_local_stock(struct work_struct *dummy)
1779{
Christoph Lameter7c8e0182014-06-04 16:07:56 -07001780 struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001781 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001782 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001783}
1784
1785/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001786 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001787 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001788 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001789static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001790{
1791 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1792
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001793 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001794 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001795 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001796 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001797 stock->nr_pages += nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001798 put_cpu_var(memcg_stock);
1799}
1800
1801/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001802 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001803 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001804 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001805static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001806{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001807 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001808
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001809 /* If someone's already draining, avoid adding running more workers. */
1810 if (!mutex_trylock(&percpu_charge_mutex))
1811 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001812 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001813 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001814 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001815 for_each_online_cpu(cpu) {
1816 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001817 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001818
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001819 memcg = stock->cached;
1820 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001821 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001822 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001823 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001824 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1825 if (cpu == curcpu)
1826 drain_local_stock(&stock->work);
1827 else
1828 schedule_work_on(cpu, &stock->work);
1829 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001830 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001831 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001832 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001833 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001834}
1835
Paul Gortmaker0db06282013-06-19 14:53:51 -04001836static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001837 unsigned long action,
1838 void *hcpu)
1839{
1840 int cpu = (unsigned long)hcpu;
1841 struct memcg_stock_pcp *stock;
1842
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001843 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001844 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001845
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07001846 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001847 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001848
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001849 stock = &per_cpu(memcg_stock, cpu);
1850 drain_stock(stock);
1851 return NOTIFY_OK;
1852}
1853
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001854static void reclaim_high(struct mem_cgroup *memcg,
1855 unsigned int nr_pages,
1856 gfp_t gfp_mask)
1857{
1858 do {
1859 if (page_counter_read(&memcg->memory) <= memcg->high)
1860 continue;
1861 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
1862 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1863 } while ((memcg = parent_mem_cgroup(memcg)));
1864}
1865
1866static void high_work_func(struct work_struct *work)
1867{
1868 struct mem_cgroup *memcg;
1869
1870 memcg = container_of(work, struct mem_cgroup, high_work);
1871 reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1872}
1873
Tejun Heob23afb92015-11-05 18:46:11 -08001874/*
1875 * Scheduled by try_charge() to be executed from the userland return path
1876 * and reclaims memory over the high limit.
1877 */
1878void mem_cgroup_handle_over_high(void)
1879{
1880 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001881 struct mem_cgroup *memcg;
Tejun Heob23afb92015-11-05 18:46:11 -08001882
1883 if (likely(!nr_pages))
1884 return;
1885
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001886 memcg = get_mem_cgroup_from_mm(current->mm);
1887 reclaim_high(memcg, nr_pages, GFP_KERNEL);
Tejun Heob23afb92015-11-05 18:46:11 -08001888 css_put(&memcg->css);
1889 current->memcg_nr_pages_over_high = 0;
1890}
1891
Johannes Weiner00501b52014-08-08 14:19:20 -07001892static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1893 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001894{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001895 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001896 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001897 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001898 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001899 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001900 bool may_swap = true;
1901 bool drained = false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001902
Johannes Weinerce00a962014-09-05 08:43:57 -04001903 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08001904 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001905retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001906 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08001907 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001908
Johannes Weiner7941d212016-01-14 15:21:23 -08001909 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08001910 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1911 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001912 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08001913 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001914 page_counter_uncharge(&memcg->memsw, batch);
1915 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001916 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001917 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001918 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001919 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001920
Johannes Weiner6539cc02014-08-06 16:05:42 -07001921 if (batch > nr_pages) {
1922 batch = nr_pages;
1923 goto retry;
1924 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001925
Johannes Weiner06b078f2014-08-06 16:05:44 -07001926 /*
1927 * Unlike in global OOM situations, memcg is not in a physical
1928 * memory shortage. Allow dying and OOM-killed tasks to
1929 * bypass the last charges so that they can exit quickly and
1930 * free their memory.
1931 */
1932 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1933 fatal_signal_pending(current) ||
1934 current->flags & PF_EXITING))
Tejun Heo10d53c72015-11-05 18:46:17 -08001935 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001936
1937 if (unlikely(task_in_memcg_oom(current)))
1938 goto nomem;
1939
Mel Gormand0164ad2015-11-06 16:28:21 -08001940 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001941 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001942
Johannes Weiner241994e2015-02-11 15:26:06 -08001943 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
1944
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001945 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1946 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07001947
Johannes Weiner61e02c72014-08-06 16:08:16 -07001948 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07001949 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07001950
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001951 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001952 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001953 drained = true;
1954 goto retry;
1955 }
1956
Johannes Weiner28c34c22014-08-06 16:05:47 -07001957 if (gfp_mask & __GFP_NORETRY)
1958 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001959 /*
1960 * Even though the limit is exceeded at this point, reclaim
1961 * may have been able to free some pages. Retry the charge
1962 * before killing the task.
1963 *
1964 * Only for regular pages, though: huge pages are rather
1965 * unlikely to succeed so close to the limit, and we fall back
1966 * to regular pages anyway in case of failure.
1967 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07001968 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001969 goto retry;
1970 /*
1971 * At task move, charge accounts can be doubly counted. So, it's
1972 * better to wait until the end of task_move if something is going on.
1973 */
1974 if (mem_cgroup_wait_acct_move(mem_over_limit))
1975 goto retry;
1976
Johannes Weiner9b130612014-08-06 16:05:51 -07001977 if (nr_retries--)
1978 goto retry;
1979
Johannes Weiner06b078f2014-08-06 16:05:44 -07001980 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08001981 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001982
Johannes Weiner6539cc02014-08-06 16:05:42 -07001983 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08001984 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001985
Johannes Weiner241994e2015-02-11 15:26:06 -08001986 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
1987
Jerome Marchand3608de02015-11-05 18:47:29 -08001988 mem_cgroup_oom(mem_over_limit, gfp_mask,
1989 get_order(nr_pages * PAGE_SIZE));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001990nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07001991 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07001992 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08001993force:
1994 /*
1995 * The allocation either can't fail or will lead to more memory
1996 * being freed very soon. Allow memory usage go over the limit
1997 * temporarily by force charging it.
1998 */
1999 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002000 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08002001 page_counter_charge(&memcg->memsw, nr_pages);
2002 css_get_many(&memcg->css, nr_pages);
2003
2004 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002005
2006done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002007 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002008 if (batch > nr_pages)
2009 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002010
Johannes Weiner241994e2015-02-11 15:26:06 -08002011 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002012 * If the hierarchy is above the normal consumption range, schedule
2013 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002014 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002015 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2016 * not recorded as it most likely matches current's and won't
2017 * change in the meantime. As high limit is checked again before
2018 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994e2015-02-11 15:26:06 -08002019 */
2020 do {
Tejun Heob23afb92015-11-05 18:46:11 -08002021 if (page_counter_read(&memcg->memory) > memcg->high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002022 /* Don't bother a random interrupted task */
2023 if (in_interrupt()) {
2024 schedule_work(&memcg->high_work);
2025 break;
2026 }
Vladimir Davydov9516a182015-12-11 13:40:24 -08002027 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002028 set_notify_resume(current);
2029 break;
2030 }
Johannes Weiner241994e2015-02-11 15:26:06 -08002031 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002032
2033 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002034}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002035
Johannes Weiner00501b52014-08-08 14:19:20 -07002036static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002037{
Johannes Weinerce00a962014-09-05 08:43:57 -04002038 if (mem_cgroup_is_root(memcg))
2039 return;
2040
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002041 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002042 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002043 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002044
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002045 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002046}
2047
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002048static void lock_page_lru(struct page *page, int *isolated)
2049{
2050 struct zone *zone = page_zone(page);
2051
Mel Gormana52633d2016-07-28 15:45:28 -07002052 spin_lock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002053 if (PageLRU(page)) {
2054 struct lruvec *lruvec;
2055
Mel Gorman599d0c92016-07-28 15:45:31 -07002056 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002057 ClearPageLRU(page);
2058 del_page_from_lru_list(page, lruvec, page_lru(page));
2059 *isolated = 1;
2060 } else
2061 *isolated = 0;
2062}
2063
2064static void unlock_page_lru(struct page *page, int isolated)
2065{
2066 struct zone *zone = page_zone(page);
2067
2068 if (isolated) {
2069 struct lruvec *lruvec;
2070
Mel Gorman599d0c92016-07-28 15:45:31 -07002071 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002072 VM_BUG_ON_PAGE(PageLRU(page), page);
2073 SetPageLRU(page);
2074 add_page_to_lru_list(page, lruvec, page_lru(page));
2075 }
Mel Gormana52633d2016-07-28 15:45:28 -07002076 spin_unlock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002077}
2078
Johannes Weiner00501b52014-08-08 14:19:20 -07002079static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002080 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002081{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002082 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002083
Johannes Weiner1306a852014-12-10 15:44:52 -08002084 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002085
2086 /*
2087 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2088 * may already be on some other mem_cgroup's LRU. Take care of it.
2089 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002090 if (lrucare)
2091 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002092
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002093 /*
2094 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002095 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002096 *
2097 * - the page is uncharged
2098 *
2099 * - the page is off-LRU
2100 *
2101 * - an anonymous fault has exclusive page access, except for
2102 * a locked page table
2103 *
2104 * - a page cache insertion, a swapin fault, or a migration
2105 * have the page locked
2106 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002107 page->mem_cgroup = memcg;
Hugh Dickins3be91272008-02-07 00:14:19 -08002108
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002109 if (lrucare)
2110 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002111}
2112
Johannes Weiner127424c2016-01-20 15:02:32 -08002113#ifndef CONFIG_SLOB
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002114static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002115{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002116 int id, size;
2117 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002118
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002119 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002120 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2121 if (id < 0)
2122 return id;
2123
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002124 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002125 return id;
2126
2127 /*
2128 * There's no space for the new id in memcg_caches arrays,
2129 * so we have to grow them.
2130 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002131 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002132
2133 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002134 if (size < MEMCG_CACHES_MIN_SIZE)
2135 size = MEMCG_CACHES_MIN_SIZE;
2136 else if (size > MEMCG_CACHES_MAX_SIZE)
2137 size = MEMCG_CACHES_MAX_SIZE;
2138
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002139 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002140 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002141 err = memcg_update_all_list_lrus(size);
2142 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002143 memcg_nr_cache_ids = size;
2144
2145 up_write(&memcg_cache_ids_sem);
2146
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002147 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002148 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002149 return err;
2150 }
2151 return id;
2152}
2153
2154static void memcg_free_cache_id(int id)
2155{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002156 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002157}
2158
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002159struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002160 struct mem_cgroup *memcg;
2161 struct kmem_cache *cachep;
2162 struct work_struct work;
2163};
2164
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002165static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002166{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002167 struct memcg_kmem_cache_create_work *cw =
2168 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002169 struct mem_cgroup *memcg = cw->memcg;
2170 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002171
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002172 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002173
Vladimir Davydov5722d092014-04-07 15:39:24 -07002174 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002175 kfree(cw);
2176}
2177
2178/*
2179 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002180 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002181static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2182 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002183{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002184 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002185
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002186 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002187 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002188 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002189
2190 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002191
2192 cw->memcg = memcg;
2193 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002194 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002195
Glauber Costad7f25f82012-12-18 14:22:40 -08002196 schedule_work(&cw->work);
2197}
2198
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002199static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2200 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002201{
2202 /*
2203 * We need to stop accounting when we kmalloc, because if the
2204 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002205 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002206 *
2207 * However, it is better to enclose the whole function. Depending on
2208 * the debugging options enabled, INIT_WORK(), for instance, can
2209 * trigger an allocation. This too, will make us recurse. Because at
2210 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2211 * the safest choice is to do it like this, wrapping the whole function.
2212 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002213 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002214 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002215 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002216}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002217
Vladimir Davydov45264772016-07-26 15:24:21 -07002218static inline bool memcg_kmem_bypass(void)
2219{
2220 if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
2221 return true;
2222 return false;
2223}
2224
2225/**
2226 * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2227 * @cachep: the original global kmem cache
2228 *
Glauber Costad7f25f82012-12-18 14:22:40 -08002229 * Return the kmem_cache we're supposed to use for a slab allocation.
2230 * We try to use the current memcg's version of the cache.
2231 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002232 * If the cache does not exist yet, if we are the first user of it, we
2233 * create it asynchronously in a workqueue and let the current allocation
2234 * go through with the original cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002235 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002236 * This function takes a reference to the cache it returns to assure it
2237 * won't get destroyed while we are working with it. Once the caller is
2238 * done with it, memcg_kmem_put_cache() must be called to release the
2239 * reference.
Glauber Costad7f25f82012-12-18 14:22:40 -08002240 */
Vladimir Davydov45264772016-07-26 15:24:21 -07002241struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002242{
2243 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002244 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002245 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002246
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002247 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002248
Vladimir Davydov45264772016-07-26 15:24:21 -07002249 if (memcg_kmem_bypass())
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002250 return cachep;
2251
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002252 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002253 return cachep;
2254
Vladimir Davydov8135be52014-12-12 16:56:38 -08002255 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002256 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002257 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002258 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002259
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002260 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002261 if (likely(memcg_cachep))
2262 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002263
2264 /*
2265 * If we are in a safe context (can wait, and not in interrupt
2266 * context), we could be be predictable and return right away.
2267 * This would guarantee that the allocation being performed
2268 * already belongs in the new cache.
2269 *
2270 * However, there are some clashes that can arrive from locking.
2271 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002272 * memcg_create_kmem_cache, this means no further allocation
2273 * could happen with the slab_mutex held. So it's better to
2274 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002275 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002276 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002277out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002278 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002279 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002280}
Glauber Costad7f25f82012-12-18 14:22:40 -08002281
Vladimir Davydov45264772016-07-26 15:24:21 -07002282/**
2283 * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2284 * @cachep: the cache returned by memcg_kmem_get_cache
2285 */
2286void memcg_kmem_put_cache(struct kmem_cache *cachep)
Vladimir Davydov8135be52014-12-12 16:56:38 -08002287{
2288 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002289 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002290}
2291
Vladimir Davydov45264772016-07-26 15:24:21 -07002292/**
2293 * memcg_kmem_charge: charge a kmem page
2294 * @page: page to charge
2295 * @gfp: reclaim mode
2296 * @order: allocation order
2297 * @memcg: memory cgroup to charge
2298 *
2299 * Returns 0 on success, an error code on failure.
2300 */
2301int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2302 struct mem_cgroup *memcg)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002303{
2304 unsigned int nr_pages = 1 << order;
2305 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002306 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002307
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002308 ret = try_charge(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08002309 if (ret)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002310 return ret;
Johannes Weiner52c29b02016-01-20 15:02:35 -08002311
2312 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2313 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2314 cancel_charge(memcg, nr_pages);
2315 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002316 }
2317
2318 page->mem_cgroup = memcg;
2319
2320 return 0;
2321}
2322
Vladimir Davydov45264772016-07-26 15:24:21 -07002323/**
2324 * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2325 * @page: page to charge
2326 * @gfp: reclaim mode
2327 * @order: allocation order
2328 *
2329 * Returns 0 on success, an error code on failure.
2330 */
2331int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002332{
2333 struct mem_cgroup *memcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002334 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002335
Vladimir Davydov45264772016-07-26 15:24:21 -07002336 if (memcg_kmem_bypass())
2337 return 0;
2338
Johannes Weinerdf381972014-04-07 15:37:43 -07002339 memcg = get_mem_cgroup_from_mm(current->mm);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002340 if (!mem_cgroup_is_root(memcg)) {
Vladimir Davydov45264772016-07-26 15:24:21 -07002341 ret = memcg_kmem_charge_memcg(page, gfp, order, memcg);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002342 if (!ret)
2343 __SetPageKmemcg(page);
2344 }
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002345 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002346 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002347}
Vladimir Davydov45264772016-07-26 15:24:21 -07002348/**
2349 * memcg_kmem_uncharge: uncharge a kmem page
2350 * @page: page to uncharge
2351 * @order: allocation order
2352 */
2353void memcg_kmem_uncharge(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002354{
Johannes Weiner1306a852014-12-10 15:44:52 -08002355 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002356 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002357
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002358 if (!memcg)
2359 return;
2360
Sasha Levin309381fea2014-01-23 15:52:54 -08002361 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002362
Johannes Weiner52c29b02016-01-20 15:02:35 -08002363 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2364 page_counter_uncharge(&memcg->kmem, nr_pages);
2365
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002366 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002367 if (do_memsw_account())
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002368 page_counter_uncharge(&memcg->memsw, nr_pages);
2369
Johannes Weiner1306a852014-12-10 15:44:52 -08002370 page->mem_cgroup = NULL;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002371
2372 /* slab pages do not have PageKmemcg flag set */
2373 if (PageKmemcg(page))
2374 __ClearPageKmemcg(page);
2375
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002376 css_put_many(&memcg->css, nr_pages);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002377}
Johannes Weiner127424c2016-01-20 15:02:32 -08002378#endif /* !CONFIG_SLOB */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002379
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002380#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2381
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002382/*
2383 * Because tail pages are not marked as "used", set it. We're under
Mel Gormana52633d2016-07-28 15:45:28 -07002384 * zone_lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002385 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002386void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002387{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002388 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002389
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002390 if (mem_cgroup_disabled())
2391 return;
David Rientjesb070e652013-05-07 16:18:09 -07002392
Johannes Weiner29833312014-12-10 15:44:02 -08002393 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002394 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002395
Johannes Weiner1306a852014-12-10 15:44:52 -08002396 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002397 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002398}
Hugh Dickins12d27102012-01-12 17:19:52 -08002399#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002400
Andrew Mortonc255a452012-07-31 16:43:02 -07002401#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002402static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2403 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002404{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002405 int val = (charge) ? 1 : -1;
2406 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002407}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002408
2409/**
2410 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2411 * @entry: swap entry to be moved
2412 * @from: mem_cgroup which the entry is moved from
2413 * @to: mem_cgroup which the entry is moved to
2414 *
2415 * It succeeds only when the swap_cgroup's record for this entry is the same
2416 * as the mem_cgroup's id of @from.
2417 *
2418 * Returns 0 on success, -EINVAL on failure.
2419 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002420 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002421 * both res and memsw, and called css_get().
2422 */
2423static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002424 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002425{
2426 unsigned short old_id, new_id;
2427
Li Zefan34c00c32013-09-23 16:56:01 +08002428 old_id = mem_cgroup_id(from);
2429 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002430
2431 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002432 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002433 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002434 return 0;
2435 }
2436 return -EINVAL;
2437}
2438#else
2439static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002440 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002441{
2442 return -EINVAL;
2443}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002444#endif
2445
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002446static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002447
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002448static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002449 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002450{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002451 unsigned long curusage;
2452 unsigned long oldusage;
2453 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002454 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002455 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002456
2457 /*
2458 * For keeping hierarchical_reclaim simple, how long we should retry
2459 * is depends on callers. We set our retry-count to be function
2460 * of # of children which we should visit in this loop.
2461 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002462 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2463 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002464
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002465 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002466
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002467 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002468 if (signal_pending(current)) {
2469 ret = -EINTR;
2470 break;
2471 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002472
2473 mutex_lock(&memcg_limit_mutex);
2474 if (limit > memcg->memsw.limit) {
2475 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002476 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002477 break;
2478 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002479 if (limit > memcg->memory.limit)
2480 enlarge = true;
2481 ret = page_counter_limit(&memcg->memory, limit);
2482 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002483
2484 if (!ret)
2485 break;
2486
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002487 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2488
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002489 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002490 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002491 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002492 retry_count--;
2493 else
2494 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002495 } while (retry_count);
2496
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002497 if (!ret && enlarge)
2498 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002499
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002500 return ret;
2501}
2502
Li Zefan338c8432009-06-17 16:27:15 -07002503static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002504 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002505{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002506 unsigned long curusage;
2507 unsigned long oldusage;
2508 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002509 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002510 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002511
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002512 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002513 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2514 mem_cgroup_count_children(memcg);
2515
2516 oldusage = page_counter_read(&memcg->memsw);
2517
2518 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002519 if (signal_pending(current)) {
2520 ret = -EINTR;
2521 break;
2522 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002523
2524 mutex_lock(&memcg_limit_mutex);
2525 if (limit < memcg->memory.limit) {
2526 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002527 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002528 break;
2529 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002530 if (limit > memcg->memsw.limit)
2531 enlarge = true;
2532 ret = page_counter_limit(&memcg->memsw, limit);
2533 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002534
2535 if (!ret)
2536 break;
2537
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002538 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2539
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002540 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002541 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002542 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002543 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002544 else
2545 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002546 } while (retry_count);
2547
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002548 if (!ret && enlarge)
2549 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002550
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002551 return ret;
2552}
2553
Mel Gormanef8f2322016-07-28 15:46:05 -07002554unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
Andrew Morton0608f432013-09-24 15:27:41 -07002555 gfp_t gfp_mask,
2556 unsigned long *total_scanned)
2557{
2558 unsigned long nr_reclaimed = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002559 struct mem_cgroup_per_node *mz, *next_mz = NULL;
Andrew Morton0608f432013-09-24 15:27:41 -07002560 unsigned long reclaimed;
2561 int loop = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002562 struct mem_cgroup_tree_per_node *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002563 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002564 unsigned long nr_scanned;
2565
2566 if (order > 0)
2567 return 0;
2568
Mel Gormanef8f2322016-07-28 15:46:05 -07002569 mctz = soft_limit_tree_node(pgdat->node_id);
Michal Hockod6507ff2016-08-02 14:02:37 -07002570
2571 /*
2572 * Do not even bother to check the largest node if the root
2573 * is empty. Do it lockless to prevent lock bouncing. Races
2574 * are acceptable as soft limit is best effort anyway.
2575 */
2576 if (RB_EMPTY_ROOT(&mctz->rb_root))
2577 return 0;
2578
Andrew Morton0608f432013-09-24 15:27:41 -07002579 /*
2580 * This loop can run a while, specially if mem_cgroup's continuously
2581 * keep exceeding their soft limit and putting the system under
2582 * pressure
2583 */
2584 do {
2585 if (next_mz)
2586 mz = next_mz;
2587 else
2588 mz = mem_cgroup_largest_soft_limit_node(mctz);
2589 if (!mz)
2590 break;
2591
2592 nr_scanned = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002593 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07002594 gfp_mask, &nr_scanned);
2595 nr_reclaimed += reclaimed;
2596 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002597 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002598 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002599
2600 /*
2601 * If we failed to reclaim anything from this memory cgroup
2602 * it is time to move on to the next cgroup
2603 */
2604 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002605 if (!reclaimed)
2606 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2607
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002608 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002609 /*
2610 * One school of thought says that we should not add
2611 * back the node to the tree if reclaim returns 0.
2612 * But our reclaim could return 0, simply because due
2613 * to priority we are exposing a smaller subset of
2614 * memory to reclaim from. Consider this as a longer
2615 * term TODO.
2616 */
2617 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002618 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002619 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002620 css_put(&mz->memcg->css);
2621 loop++;
2622 /*
2623 * Could not reclaim anything and there are no more
2624 * mem cgroups to try or we seem to be looping without
2625 * reclaiming anything.
2626 */
2627 if (!nr_reclaimed &&
2628 (next_mz == NULL ||
2629 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2630 break;
2631 } while (!nr_reclaimed);
2632 if (next_mz)
2633 css_put(&next_mz->memcg->css);
2634 return nr_reclaimed;
2635}
2636
Tejun Heoea280e72014-05-16 13:22:48 -04002637/*
2638 * Test whether @memcg has children, dead or alive. Note that this
2639 * function doesn't care whether @memcg has use_hierarchy enabled and
2640 * returns %true if there are child csses according to the cgroup
2641 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2642 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002643static inline bool memcg_has_children(struct mem_cgroup *memcg)
2644{
Tejun Heoea280e72014-05-16 13:22:48 -04002645 bool ret;
2646
Tejun Heoea280e72014-05-16 13:22:48 -04002647 rcu_read_lock();
2648 ret = css_next_child(NULL, &memcg->css);
2649 rcu_read_unlock();
2650 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002651}
2652
2653/*
Greg Thelen51038172016-05-20 16:58:18 -07002654 * Reclaims as many pages from the given memcg as possible.
Michal Hockoc26251f2012-10-26 13:37:28 +02002655 *
2656 * Caller is responsible for holding css reference for memcg.
2657 */
2658static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2659{
2660 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002661
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002662 /* we call try-to-free pages for make this cgroup empty */
2663 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002664 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002665 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002666 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002667
Michal Hockoc26251f2012-10-26 13:37:28 +02002668 if (signal_pending(current))
2669 return -EINTR;
2670
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002671 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2672 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002673 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002674 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002675 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002676 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002677 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002678
2679 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002680
2681 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002682}
2683
Tejun Heo6770c642014-05-13 12:16:21 -04002684static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2685 char *buf, size_t nbytes,
2686 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002687{
Tejun Heo6770c642014-05-13 12:16:21 -04002688 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002689
Michal Hockod8423012012-10-26 13:37:29 +02002690 if (mem_cgroup_is_root(memcg))
2691 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002692 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002693}
2694
Tejun Heo182446d2013-08-08 20:11:24 -04002695static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2696 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002697{
Tejun Heo182446d2013-08-08 20:11:24 -04002698 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002699}
2700
Tejun Heo182446d2013-08-08 20:11:24 -04002701static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2702 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002703{
2704 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002705 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002706 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002707
Glauber Costa567fb432012-07-31 16:43:07 -07002708 if (memcg->use_hierarchy == val)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002709 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07002710
Balbir Singh18f59ea2009-01-07 18:08:07 -08002711 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002712 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002713 * in the child subtrees. If it is unset, then the change can
2714 * occur, provided the current cgroup has no children.
2715 *
2716 * For the root cgroup, parent_mem is NULL, we allow value to be
2717 * set if there are no children.
2718 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002719 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002720 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002721 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002722 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002723 else
2724 retval = -EBUSY;
2725 } else
2726 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002727
Balbir Singh18f59ea2009-01-07 18:08:07 -08002728 return retval;
2729}
2730
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002731static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
Johannes Weinerce00a962014-09-05 08:43:57 -04002732{
2733 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002734 int i;
Johannes Weinerce00a962014-09-05 08:43:57 -04002735
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002736 memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
Johannes Weinerce00a962014-09-05 08:43:57 -04002737
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002738 for_each_mem_cgroup_tree(iter, memcg) {
2739 for (i = 0; i < MEMCG_NR_STAT; i++)
2740 stat[i] += mem_cgroup_read_stat(iter, i);
2741 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002742}
2743
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002744static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
Johannes Weiner587d9f72016-01-20 15:03:19 -08002745{
2746 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002747 int i;
Johannes Weiner587d9f72016-01-20 15:03:19 -08002748
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002749 memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
Johannes Weiner587d9f72016-01-20 15:03:19 -08002750
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002751 for_each_mem_cgroup_tree(iter, memcg) {
2752 for (i = 0; i < MEMCG_NR_EVENTS; i++)
2753 events[i] += mem_cgroup_read_events(iter, i);
2754 }
Johannes Weiner587d9f72016-01-20 15:03:19 -08002755}
2756
Andrew Morton6f646152015-11-06 16:28:58 -08002757static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04002758{
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002759 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002760
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002761 if (mem_cgroup_is_root(memcg)) {
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002762 struct mem_cgroup *iter;
2763
2764 for_each_mem_cgroup_tree(iter, memcg) {
2765 val += mem_cgroup_read_stat(iter,
2766 MEM_CGROUP_STAT_CACHE);
2767 val += mem_cgroup_read_stat(iter,
2768 MEM_CGROUP_STAT_RSS);
2769 if (swap)
2770 val += mem_cgroup_read_stat(iter,
2771 MEM_CGROUP_STAT_SWAP);
2772 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002773 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002774 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002775 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002776 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002777 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002778 }
Michal Hockoc12176d2015-11-05 18:50:29 -08002779 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002780}
2781
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002782enum {
2783 RES_USAGE,
2784 RES_LIMIT,
2785 RES_MAX_USAGE,
2786 RES_FAILCNT,
2787 RES_SOFT_LIMIT,
2788};
Johannes Weinerce00a962014-09-05 08:43:57 -04002789
Tejun Heo791badb2013-12-05 12:28:02 -05002790static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002791 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002792{
Tejun Heo182446d2013-08-08 20:11:24 -04002793 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002794 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002795
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002796 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002797 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002798 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002799 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002800 case _MEMSWAP:
2801 counter = &memcg->memsw;
2802 break;
2803 case _KMEM:
2804 counter = &memcg->kmem;
2805 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002806 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08002807 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002808 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002809 default:
2810 BUG();
2811 }
2812
2813 switch (MEMFILE_ATTR(cft->private)) {
2814 case RES_USAGE:
2815 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08002816 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002817 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08002818 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002819 return (u64)page_counter_read(counter) * PAGE_SIZE;
2820 case RES_LIMIT:
2821 return (u64)counter->limit * PAGE_SIZE;
2822 case RES_MAX_USAGE:
2823 return (u64)counter->watermark * PAGE_SIZE;
2824 case RES_FAILCNT:
2825 return counter->failcnt;
2826 case RES_SOFT_LIMIT:
2827 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002828 default:
2829 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002830 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002831}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002832
Johannes Weiner127424c2016-01-20 15:02:32 -08002833#ifndef CONFIG_SLOB
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002834static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002835{
Vladimir Davydovd6441632014-01-23 15:53:09 -08002836 int memcg_id;
2837
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002838 if (cgroup_memory_nokmem)
2839 return 0;
2840
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002841 BUG_ON(memcg->kmemcg_id >= 0);
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002842 BUG_ON(memcg->kmem_state);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002843
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002844 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002845 if (memcg_id < 0)
2846 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002847
Johannes Weineref129472016-01-14 15:21:34 -08002848 static_branch_inc(&memcg_kmem_enabled_key);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002849 /*
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002850 * A memory cgroup is considered kmem-online as soon as it gets
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002851 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002852 * guarantee no one starts accounting before all call sites are
2853 * patched.
2854 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002855 memcg->kmemcg_id = memcg_id;
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002856 memcg->kmem_state = KMEM_ONLINE;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002857
2858 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002859}
2860
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002861static void memcg_offline_kmem(struct mem_cgroup *memcg)
2862{
2863 struct cgroup_subsys_state *css;
2864 struct mem_cgroup *parent, *child;
2865 int kmemcg_id;
2866
2867 if (memcg->kmem_state != KMEM_ONLINE)
2868 return;
2869 /*
2870 * Clear the online state before clearing memcg_caches array
2871 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2872 * guarantees that no cache will be created for this cgroup
2873 * after we are done (see memcg_create_kmem_cache()).
2874 */
2875 memcg->kmem_state = KMEM_ALLOCATED;
2876
2877 memcg_deactivate_kmem_caches(memcg);
2878
2879 kmemcg_id = memcg->kmemcg_id;
2880 BUG_ON(kmemcg_id < 0);
2881
2882 parent = parent_mem_cgroup(memcg);
2883 if (!parent)
2884 parent = root_mem_cgroup;
2885
2886 /*
2887 * Change kmemcg_id of this cgroup and all its descendants to the
2888 * parent's id, and then move all entries from this cgroup's list_lrus
2889 * to ones of the parent. After we have finished, all list_lrus
2890 * corresponding to this cgroup are guaranteed to remain empty. The
2891 * ordering is imposed by list_lru_node->lock taken by
2892 * memcg_drain_all_list_lrus().
2893 */
Tejun Heo3a06bb72016-06-03 14:55:44 -07002894 rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002895 css_for_each_descendant_pre(css, &memcg->css) {
2896 child = mem_cgroup_from_css(css);
2897 BUG_ON(child->kmemcg_id != kmemcg_id);
2898 child->kmemcg_id = parent->kmemcg_id;
2899 if (!memcg->use_hierarchy)
2900 break;
2901 }
Tejun Heo3a06bb72016-06-03 14:55:44 -07002902 rcu_read_unlock();
2903
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002904 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2905
2906 memcg_free_cache_id(kmemcg_id);
2907}
2908
2909static void memcg_free_kmem(struct mem_cgroup *memcg)
2910{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002911 /* css_alloc() failed, offlining didn't happen */
2912 if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2913 memcg_offline_kmem(memcg);
2914
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002915 if (memcg->kmem_state == KMEM_ALLOCATED) {
2916 memcg_destroy_kmem_caches(memcg);
2917 static_branch_dec(&memcg_kmem_enabled_key);
2918 WARN_ON(page_counter_read(&memcg->kmem));
2919 }
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002920}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002921#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002922static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08002923{
2924 return 0;
2925}
2926static void memcg_offline_kmem(struct mem_cgroup *memcg)
2927{
2928}
2929static void memcg_free_kmem(struct mem_cgroup *memcg)
2930{
2931}
2932#endif /* !CONFIG_SLOB */
2933
Johannes Weiner127424c2016-01-20 15:02:32 -08002934static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2935 unsigned long limit)
2936{
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002937 int ret;
Johannes Weiner127424c2016-01-20 15:02:32 -08002938
2939 mutex_lock(&memcg_limit_mutex);
Johannes Weiner127424c2016-01-20 15:02:32 -08002940 ret = page_counter_limit(&memcg->kmem, limit);
Johannes Weiner127424c2016-01-20 15:02:32 -08002941 mutex_unlock(&memcg_limit_mutex);
2942 return ret;
2943}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002944
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002945static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2946{
2947 int ret;
2948
2949 mutex_lock(&memcg_limit_mutex);
2950
Johannes Weiner0db15292016-01-20 15:02:50 -08002951 ret = page_counter_limit(&memcg->tcpmem, limit);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002952 if (ret)
2953 goto out;
2954
Johannes Weiner0db15292016-01-20 15:02:50 -08002955 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002956 /*
2957 * The active flag needs to be written after the static_key
2958 * update. This is what guarantees that the socket activation
2959 * function is the last one to run. See sock_update_memcg() for
2960 * details, and note that we don't mark any socket as belonging
2961 * to this memcg until that flag is up.
2962 *
2963 * We need to do this, because static_keys will span multiple
2964 * sites, but we can't control their order. If we mark a socket
2965 * as accounted, but the accounting functions are not patched in
2966 * yet, we'll lose accounting.
2967 *
2968 * We never race with the readers in sock_update_memcg(),
2969 * because when this value change, the code to process it is not
2970 * patched in yet.
2971 */
2972 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08002973 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002974 }
2975out:
2976 mutex_unlock(&memcg_limit_mutex);
2977 return ret;
2978}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002979
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002980/*
2981 * The user of this function is...
2982 * RES_LIMIT.
2983 */
Tejun Heo451af502014-05-13 12:16:21 -04002984static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2985 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002986{
Tejun Heo451af502014-05-13 12:16:21 -04002987 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002988 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002989 int ret;
2990
Tejun Heo451af502014-05-13 12:16:21 -04002991 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08002992 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002993 if (ret)
2994 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07002995
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002996 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002997 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07002998 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2999 ret = -EINVAL;
3000 break;
3001 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003002 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3003 case _MEM:
3004 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003005 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003006 case _MEMSWAP:
3007 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3008 break;
3009 case _KMEM:
3010 ret = memcg_update_kmem_limit(memcg, nr_pages);
3011 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003012 case _TCP:
3013 ret = memcg_update_tcp_limit(memcg, nr_pages);
3014 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003015 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003016 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003017 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003018 memcg->soft_limit = nr_pages;
3019 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003020 break;
3021 }
Tejun Heo451af502014-05-13 12:16:21 -04003022 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003023}
3024
Tejun Heo6770c642014-05-13 12:16:21 -04003025static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3026 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003027{
Tejun Heo6770c642014-05-13 12:16:21 -04003028 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003029 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003030
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003031 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3032 case _MEM:
3033 counter = &memcg->memory;
3034 break;
3035 case _MEMSWAP:
3036 counter = &memcg->memsw;
3037 break;
3038 case _KMEM:
3039 counter = &memcg->kmem;
3040 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003041 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003042 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003043 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003044 default:
3045 BUG();
3046 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003047
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003048 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003049 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003050 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003051 break;
3052 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003053 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003054 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003055 default:
3056 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003057 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003058
Tejun Heo6770c642014-05-13 12:16:21 -04003059 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003060}
3061
Tejun Heo182446d2013-08-08 20:11:24 -04003062static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003063 struct cftype *cft)
3064{
Tejun Heo182446d2013-08-08 20:11:24 -04003065 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003066}
3067
Daisuke Nishimura02491442010-03-10 15:22:17 -08003068#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003069static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003070 struct cftype *cft, u64 val)
3071{
Tejun Heo182446d2013-08-08 20:11:24 -04003072 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003073
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003074 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003075 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003076
Glauber Costaee5e8472013-02-22 16:34:50 -08003077 /*
3078 * No kind of locking is needed in here, because ->can_attach() will
3079 * check this value once in the beginning of the process, and then carry
3080 * on with stale data. This means that changes to this value will only
3081 * affect task migrations starting after the change.
3082 */
3083 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003084 return 0;
3085}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003086#else
Tejun Heo182446d2013-08-08 20:11:24 -04003087static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003088 struct cftype *cft, u64 val)
3089{
3090 return -ENOSYS;
3091}
3092#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003093
Ying Han406eb0c2011-05-26 16:25:37 -07003094#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003095static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003096{
Greg Thelen25485de2013-11-12 15:07:40 -08003097 struct numa_stat {
3098 const char *name;
3099 unsigned int lru_mask;
3100 };
3101
3102 static const struct numa_stat stats[] = {
3103 { "total", LRU_ALL },
3104 { "file", LRU_ALL_FILE },
3105 { "anon", LRU_ALL_ANON },
3106 { "unevictable", BIT(LRU_UNEVICTABLE) },
3107 };
3108 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003109 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003110 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003111 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003112
Greg Thelen25485de2013-11-12 15:07:40 -08003113 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3114 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3115 seq_printf(m, "%s=%lu", stat->name, nr);
3116 for_each_node_state(nid, N_MEMORY) {
3117 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3118 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 Han071aee12013-11-12 15:07:41 -08003124 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3125 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003126
Ying Han071aee12013-11-12 15:07:41 -08003127 nr = 0;
3128 for_each_mem_cgroup_tree(iter, memcg)
3129 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3130 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3131 for_each_node_state(nid, N_MEMORY) {
3132 nr = 0;
3133 for_each_mem_cgroup_tree(iter, memcg)
3134 nr += mem_cgroup_node_nr_lru_pages(
3135 iter, nid, stat->lru_mask);
3136 seq_printf(m, " N%d=%lu", nid, nr);
3137 }
3138 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003139 }
Ying Han406eb0c2011-05-26 16:25:37 -07003140
Ying Han406eb0c2011-05-26 16:25:37 -07003141 return 0;
3142}
3143#endif /* CONFIG_NUMA */
3144
Tejun Heo2da8ca82013-12-05 12:28:04 -05003145static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003146{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003147 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003148 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003149 struct mem_cgroup *mi;
3150 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003151
Greg Thelen0ca44b12015-02-11 15:25:58 -08003152 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3153 MEM_CGROUP_STAT_NSTATS);
3154 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3155 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003156 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3157
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003158 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Johannes Weiner7941d212016-01-14 15:21:23 -08003159 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003160 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07003161 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003162 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003163 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003164
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003165 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3166 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3167 mem_cgroup_read_events(memcg, i));
3168
3169 for (i = 0; i < NR_LRU_LISTS; i++)
3170 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3171 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3172
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003173 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003174 memory = memsw = PAGE_COUNTER_MAX;
3175 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3176 memory = min(memory, mi->memory.limit);
3177 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003178 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003179 seq_printf(m, "hierarchical_memory_limit %llu\n",
3180 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08003181 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003182 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3183 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003184
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003185 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003186 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003187
Johannes Weiner7941d212016-01-14 15:21:23 -08003188 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003189 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003190 for_each_mem_cgroup_tree(mi, memcg)
3191 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
Greg Thelen484ebb32015-10-01 15:37:05 -07003192 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003193 }
3194
3195 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3196 unsigned long long val = 0;
3197
3198 for_each_mem_cgroup_tree(mi, memcg)
3199 val += mem_cgroup_read_events(mi, i);
3200 seq_printf(m, "total_%s %llu\n",
3201 mem_cgroup_events_names[i], val);
3202 }
3203
3204 for (i = 0; i < NR_LRU_LISTS; i++) {
3205 unsigned long long val = 0;
3206
3207 for_each_mem_cgroup_tree(mi, memcg)
3208 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3209 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003210 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003211
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003212#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003213 {
Mel Gormanef8f2322016-07-28 15:46:05 -07003214 pg_data_t *pgdat;
3215 struct mem_cgroup_per_node *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003216 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003217 unsigned long recent_rotated[2] = {0, 0};
3218 unsigned long recent_scanned[2] = {0, 0};
3219
Mel Gormanef8f2322016-07-28 15:46:05 -07003220 for_each_online_pgdat(pgdat) {
3221 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
3222 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003223
Mel Gormanef8f2322016-07-28 15:46:05 -07003224 recent_rotated[0] += rstat->recent_rotated[0];
3225 recent_rotated[1] += rstat->recent_rotated[1];
3226 recent_scanned[0] += rstat->recent_scanned[0];
3227 recent_scanned[1] += rstat->recent_scanned[1];
3228 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003229 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3230 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3231 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3232 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003233 }
3234#endif
3235
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003236 return 0;
3237}
3238
Tejun Heo182446d2013-08-08 20:11:24 -04003239static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3240 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003241{
Tejun Heo182446d2013-08-08 20:11:24 -04003242 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003243
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003244 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003245}
3246
Tejun Heo182446d2013-08-08 20:11:24 -04003247static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3248 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003249{
Tejun Heo182446d2013-08-08 20:11:24 -04003250 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003251
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003252 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003253 return -EINVAL;
3254
Linus Torvalds14208b02014-06-09 15:03:33 -07003255 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003256 memcg->swappiness = val;
3257 else
3258 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003259
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003260 return 0;
3261}
3262
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003263static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3264{
3265 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003266 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003267 int i;
3268
3269 rcu_read_lock();
3270 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003271 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003272 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003273 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003274
3275 if (!t)
3276 goto unlock;
3277
Johannes Weinerce00a962014-09-05 08:43:57 -04003278 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003279
3280 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003281 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003282 * If it's not true, a threshold was crossed after last
3283 * call of __mem_cgroup_threshold().
3284 */
Phil Carmody5407a562010-05-26 14:42:42 -07003285 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003286
3287 /*
3288 * Iterate backward over array of thresholds starting from
3289 * current_threshold and check if a threshold is crossed.
3290 * If none of thresholds below usage is crossed, we read
3291 * only one element of the array here.
3292 */
3293 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3294 eventfd_signal(t->entries[i].eventfd, 1);
3295
3296 /* i = current_threshold + 1 */
3297 i++;
3298
3299 /*
3300 * Iterate forward over array of thresholds starting from
3301 * current_threshold+1 and check if a threshold is crossed.
3302 * If none of thresholds above usage is crossed, we read
3303 * only one element of the array here.
3304 */
3305 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3306 eventfd_signal(t->entries[i].eventfd, 1);
3307
3308 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003309 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003310unlock:
3311 rcu_read_unlock();
3312}
3313
3314static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3315{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003316 while (memcg) {
3317 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08003318 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003319 __mem_cgroup_threshold(memcg, true);
3320
3321 memcg = parent_mem_cgroup(memcg);
3322 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003323}
3324
3325static int compare_thresholds(const void *a, const void *b)
3326{
3327 const struct mem_cgroup_threshold *_a = a;
3328 const struct mem_cgroup_threshold *_b = b;
3329
Greg Thelen2bff24a2013-09-11 14:23:08 -07003330 if (_a->threshold > _b->threshold)
3331 return 1;
3332
3333 if (_a->threshold < _b->threshold)
3334 return -1;
3335
3336 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003337}
3338
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003339static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003340{
3341 struct mem_cgroup_eventfd_list *ev;
3342
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003343 spin_lock(&memcg_oom_lock);
3344
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003345 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003346 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003347
3348 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003349 return 0;
3350}
3351
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003352static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003353{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003354 struct mem_cgroup *iter;
3355
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003356 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003357 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003358}
3359
Tejun Heo59b6f872013-11-22 18:20:43 -05003360static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003361 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003362{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003363 struct mem_cgroup_thresholds *thresholds;
3364 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003365 unsigned long threshold;
3366 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003367 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003368
Johannes Weiner650c5e52015-02-11 15:26:03 -08003369 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003370 if (ret)
3371 return ret;
3372
3373 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003374
Johannes Weiner05b84302014-08-06 16:05:59 -07003375 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003376 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003377 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003378 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003379 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003380 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003381 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003382 BUG();
3383
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003384 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003385 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003386 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3387
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003388 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003389
3390 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003391 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003392 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003393 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003394 ret = -ENOMEM;
3395 goto unlock;
3396 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003397 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003398
3399 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003400 if (thresholds->primary) {
3401 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003402 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003403 }
3404
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003405 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003406 new->entries[size - 1].eventfd = eventfd;
3407 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003408
3409 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003410 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003411 compare_thresholds, NULL);
3412
3413 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003414 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003415 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003416 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003417 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003418 * new->current_threshold will not be used until
3419 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003420 * it here.
3421 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003422 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003423 } else
3424 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003425 }
3426
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003427 /* Free old spare buffer and save old primary buffer as spare */
3428 kfree(thresholds->spare);
3429 thresholds->spare = thresholds->primary;
3430
3431 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003432
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003433 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003434 synchronize_rcu();
3435
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003436unlock:
3437 mutex_unlock(&memcg->thresholds_lock);
3438
3439 return ret;
3440}
3441
Tejun Heo59b6f872013-11-22 18:20:43 -05003442static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003443 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003444{
Tejun Heo59b6f872013-11-22 18:20:43 -05003445 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003446}
3447
Tejun Heo59b6f872013-11-22 18:20:43 -05003448static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003449 struct eventfd_ctx *eventfd, const char *args)
3450{
Tejun Heo59b6f872013-11-22 18:20:43 -05003451 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003452}
3453
Tejun Heo59b6f872013-11-22 18:20:43 -05003454static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003455 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003456{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003457 struct mem_cgroup_thresholds *thresholds;
3458 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003459 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003460 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003461
3462 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003463
3464 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003465 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003466 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003467 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003468 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003469 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003470 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003471 BUG();
3472
Anton Vorontsov371528c2012-02-24 05:14:46 +04003473 if (!thresholds->primary)
3474 goto unlock;
3475
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003476 /* Check if a threshold crossed before removing */
3477 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3478
3479 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003480 size = 0;
3481 for (i = 0; i < thresholds->primary->size; i++) {
3482 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003483 size++;
3484 }
3485
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003486 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003487
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003488 /* Set thresholds array to NULL if we don't have thresholds */
3489 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003490 kfree(new);
3491 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003492 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003493 }
3494
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003495 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003496
3497 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003498 new->current_threshold = -1;
3499 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3500 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003501 continue;
3502
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003503 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003504 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003505 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003506 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003507 * until rcu_assign_pointer(), so it's safe to increment
3508 * it here.
3509 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003510 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003511 }
3512 j++;
3513 }
3514
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003515swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003516 /* Swap primary and spare array */
3517 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003518
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003519 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003520
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003521 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003522 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08003523
3524 /* If all events are unregistered, free the spare array */
3525 if (!new) {
3526 kfree(thresholds->spare);
3527 thresholds->spare = NULL;
3528 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04003529unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003530 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003531}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003532
Tejun Heo59b6f872013-11-22 18:20:43 -05003533static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003534 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003535{
Tejun Heo59b6f872013-11-22 18:20:43 -05003536 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003537}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003538
Tejun Heo59b6f872013-11-22 18:20:43 -05003539static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003540 struct eventfd_ctx *eventfd)
3541{
Tejun Heo59b6f872013-11-22 18:20:43 -05003542 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003543}
3544
Tejun Heo59b6f872013-11-22 18:20:43 -05003545static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003546 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003547{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003548 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003549
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003550 event = kmalloc(sizeof(*event), GFP_KERNEL);
3551 if (!event)
3552 return -ENOMEM;
3553
Michal Hocko1af8efe2011-07-26 16:08:24 -07003554 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003555
3556 event->eventfd = eventfd;
3557 list_add(&event->list, &memcg->oom_notify);
3558
3559 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07003560 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003561 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003562 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003563
3564 return 0;
3565}
3566
Tejun Heo59b6f872013-11-22 18:20:43 -05003567static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003568 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003569{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003570 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003571
Michal Hocko1af8efe2011-07-26 16:08:24 -07003572 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003573
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003574 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003575 if (ev->eventfd == eventfd) {
3576 list_del(&ev->list);
3577 kfree(ev);
3578 }
3579 }
3580
Michal Hocko1af8efe2011-07-26 16:08:24 -07003581 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003582}
3583
Tejun Heo2da8ca82013-12-05 12:28:04 -05003584static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003585{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003586 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003587
Tejun Heo791badb2013-12-05 12:28:02 -05003588 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07003589 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003590 return 0;
3591}
3592
Tejun Heo182446d2013-08-08 20:11:24 -04003593static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003594 struct cftype *cft, u64 val)
3595{
Tejun Heo182446d2013-08-08 20:11:24 -04003596 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003597
3598 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003599 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003600 return -EINVAL;
3601
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003602 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003603 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003604 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003605
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003606 return 0;
3607}
3608
Tejun Heo52ebea72015-05-22 17:13:37 -04003609#ifdef CONFIG_CGROUP_WRITEBACK
3610
3611struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3612{
3613 return &memcg->cgwb_list;
3614}
3615
Tejun Heo841710a2015-05-22 18:23:33 -04003616static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3617{
3618 return wb_domain_init(&memcg->cgwb_domain, gfp);
3619}
3620
3621static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3622{
3623 wb_domain_exit(&memcg->cgwb_domain);
3624}
3625
Tejun Heo2529bb32015-05-22 18:23:34 -04003626static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3627{
3628 wb_domain_size_changed(&memcg->cgwb_domain);
3629}
3630
Tejun Heo841710a2015-05-22 18:23:33 -04003631struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3632{
3633 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3634
3635 if (!memcg->css.parent)
3636 return NULL;
3637
3638 return &memcg->cgwb_domain;
3639}
3640
Tejun Heoc2aa7232015-05-22 18:23:35 -04003641/**
3642 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3643 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003644 * @pfilepages: out parameter for number of file pages
3645 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04003646 * @pdirty: out parameter for number of dirty pages
3647 * @pwriteback: out parameter for number of pages under writeback
3648 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003649 * Determine the numbers of file, headroom, dirty, and writeback pages in
3650 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3651 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003652 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003653 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3654 * headroom is calculated as the lowest headroom of itself and the
3655 * ancestors. Note that this doesn't consider the actual amount of
3656 * available memory in the system. The caller should further cap
3657 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003658 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003659void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3660 unsigned long *pheadroom, unsigned long *pdirty,
3661 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04003662{
3663 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3664 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003665
3666 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3667
3668 /* this should eventually include NR_UNSTABLE_NFS */
3669 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003670 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3671 (1 << LRU_ACTIVE_FILE));
3672 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003673
Tejun Heoc2aa7232015-05-22 18:23:35 -04003674 while ((parent = parent_mem_cgroup(memcg))) {
3675 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3676 unsigned long used = page_counter_read(&memcg->memory);
3677
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003678 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04003679 memcg = parent;
3680 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04003681}
3682
Tejun Heo841710a2015-05-22 18:23:33 -04003683#else /* CONFIG_CGROUP_WRITEBACK */
3684
3685static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3686{
3687 return 0;
3688}
3689
3690static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3691{
3692}
3693
Tejun Heo2529bb32015-05-22 18:23:34 -04003694static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3695{
3696}
3697
Tejun Heo52ebea72015-05-22 17:13:37 -04003698#endif /* CONFIG_CGROUP_WRITEBACK */
3699
Tejun Heo79bd9812013-11-22 18:20:42 -05003700/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003701 * DO NOT USE IN NEW FILES.
3702 *
3703 * "cgroup.event_control" implementation.
3704 *
3705 * This is way over-engineered. It tries to support fully configurable
3706 * events for each user. Such level of flexibility is completely
3707 * unnecessary especially in the light of the planned unified hierarchy.
3708 *
3709 * Please deprecate this and replace with something simpler if at all
3710 * possible.
3711 */
3712
3713/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003714 * Unregister event and free resources.
3715 *
3716 * Gets called from workqueue.
3717 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003718static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003719{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003720 struct mem_cgroup_event *event =
3721 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003722 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003723
3724 remove_wait_queue(event->wqh, &event->wait);
3725
Tejun Heo59b6f872013-11-22 18:20:43 -05003726 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003727
3728 /* Notify userspace the event is going away. */
3729 eventfd_signal(event->eventfd, 1);
3730
3731 eventfd_ctx_put(event->eventfd);
3732 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003733 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003734}
3735
3736/*
3737 * Gets called on POLLHUP on eventfd when user closes it.
3738 *
3739 * Called with wqh->lock held and interrupts disabled.
3740 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003741static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3742 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003743{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003744 struct mem_cgroup_event *event =
3745 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003746 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003747 unsigned long flags = (unsigned long)key;
3748
3749 if (flags & POLLHUP) {
3750 /*
3751 * If the event has been detached at cgroup removal, we
3752 * can simply return knowing the other side will cleanup
3753 * for us.
3754 *
3755 * We can't race against event freeing since the other
3756 * side will require wqh->lock via remove_wait_queue(),
3757 * which we hold.
3758 */
Tejun Heofba94802013-11-22 18:20:43 -05003759 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003760 if (!list_empty(&event->list)) {
3761 list_del_init(&event->list);
3762 /*
3763 * We are in atomic context, but cgroup_event_remove()
3764 * may sleep, so we have to call it in workqueue.
3765 */
3766 schedule_work(&event->remove);
3767 }
Tejun Heofba94802013-11-22 18:20:43 -05003768 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003769 }
3770
3771 return 0;
3772}
3773
Tejun Heo3bc942f2013-11-22 18:20:44 -05003774static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003775 wait_queue_head_t *wqh, poll_table *pt)
3776{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003777 struct mem_cgroup_event *event =
3778 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003779
3780 event->wqh = wqh;
3781 add_wait_queue(wqh, &event->wait);
3782}
3783
3784/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003785 * DO NOT USE IN NEW FILES.
3786 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003787 * Parse input and register new cgroup event handler.
3788 *
3789 * Input must be in format '<event_fd> <control_fd> <args>'.
3790 * Interpretation of args is defined by control file implementation.
3791 */
Tejun Heo451af502014-05-13 12:16:21 -04003792static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3793 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003794{
Tejun Heo451af502014-05-13 12:16:21 -04003795 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003796 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003797 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003798 struct cgroup_subsys_state *cfile_css;
3799 unsigned int efd, cfd;
3800 struct fd efile;
3801 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003802 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003803 char *endp;
3804 int ret;
3805
Tejun Heo451af502014-05-13 12:16:21 -04003806 buf = strstrip(buf);
3807
3808 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003809 if (*endp != ' ')
3810 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003811 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003812
Tejun Heo451af502014-05-13 12:16:21 -04003813 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003814 if ((*endp != ' ') && (*endp != '\0'))
3815 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003816 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003817
3818 event = kzalloc(sizeof(*event), GFP_KERNEL);
3819 if (!event)
3820 return -ENOMEM;
3821
Tejun Heo59b6f872013-11-22 18:20:43 -05003822 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003823 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003824 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3825 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3826 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003827
3828 efile = fdget(efd);
3829 if (!efile.file) {
3830 ret = -EBADF;
3831 goto out_kfree;
3832 }
3833
3834 event->eventfd = eventfd_ctx_fileget(efile.file);
3835 if (IS_ERR(event->eventfd)) {
3836 ret = PTR_ERR(event->eventfd);
3837 goto out_put_efile;
3838 }
3839
3840 cfile = fdget(cfd);
3841 if (!cfile.file) {
3842 ret = -EBADF;
3843 goto out_put_eventfd;
3844 }
3845
3846 /* the process need read permission on control file */
3847 /* AV: shouldn't we check that it's been opened for read instead? */
3848 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3849 if (ret < 0)
3850 goto out_put_cfile;
3851
Tejun Heo79bd9812013-11-22 18:20:42 -05003852 /*
Tejun Heofba94802013-11-22 18:20:43 -05003853 * Determine the event callbacks and set them in @event. This used
3854 * to be done via struct cftype but cgroup core no longer knows
3855 * about these events. The following is crude but the whole thing
3856 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003857 *
3858 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003859 */
Al Virob5830432014-10-31 01:22:04 -04003860 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003861
3862 if (!strcmp(name, "memory.usage_in_bytes")) {
3863 event->register_event = mem_cgroup_usage_register_event;
3864 event->unregister_event = mem_cgroup_usage_unregister_event;
3865 } else if (!strcmp(name, "memory.oom_control")) {
3866 event->register_event = mem_cgroup_oom_register_event;
3867 event->unregister_event = mem_cgroup_oom_unregister_event;
3868 } else if (!strcmp(name, "memory.pressure_level")) {
3869 event->register_event = vmpressure_register_event;
3870 event->unregister_event = vmpressure_unregister_event;
3871 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003872 event->register_event = memsw_cgroup_usage_register_event;
3873 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003874 } else {
3875 ret = -EINVAL;
3876 goto out_put_cfile;
3877 }
3878
3879 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003880 * Verify @cfile should belong to @css. Also, remaining events are
3881 * automatically removed on cgroup destruction but the removal is
3882 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003883 */
Al Virob5830432014-10-31 01:22:04 -04003884 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003885 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003886 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003887 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003888 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003889 if (cfile_css != css) {
3890 css_put(cfile_css);
3891 goto out_put_cfile;
3892 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003893
Tejun Heo451af502014-05-13 12:16:21 -04003894 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003895 if (ret)
3896 goto out_put_css;
3897
3898 efile.file->f_op->poll(efile.file, &event->pt);
3899
Tejun Heofba94802013-11-22 18:20:43 -05003900 spin_lock(&memcg->event_list_lock);
3901 list_add(&event->list, &memcg->event_list);
3902 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003903
3904 fdput(cfile);
3905 fdput(efile);
3906
Tejun Heo451af502014-05-13 12:16:21 -04003907 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05003908
3909out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05003910 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003911out_put_cfile:
3912 fdput(cfile);
3913out_put_eventfd:
3914 eventfd_ctx_put(event->eventfd);
3915out_put_efile:
3916 fdput(efile);
3917out_kfree:
3918 kfree(event);
3919
3920 return ret;
3921}
3922
Johannes Weiner241994e2015-02-11 15:26:06 -08003923static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003924 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003925 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003926 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003927 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003928 },
3929 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003930 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003931 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04003932 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003933 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003934 },
3935 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003936 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003937 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003938 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003939 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003940 },
3941 {
Balbir Singh296c81d2009-09-23 15:56:36 -07003942 .name = "soft_limit_in_bytes",
3943 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003944 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003945 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07003946 },
3947 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003948 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003949 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04003950 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003951 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003952 },
Balbir Singh8697d332008-02-07 00:13:59 -08003953 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003954 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003955 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003956 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003957 {
3958 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04003959 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003960 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08003961 {
3962 .name = "use_hierarchy",
3963 .write_u64 = mem_cgroup_hierarchy_write,
3964 .read_u64 = mem_cgroup_hierarchy_read,
3965 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003966 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05003967 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04003968 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04003969 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05003970 },
3971 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003972 .name = "swappiness",
3973 .read_u64 = mem_cgroup_swappiness_read,
3974 .write_u64 = mem_cgroup_swappiness_write,
3975 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003976 {
3977 .name = "move_charge_at_immigrate",
3978 .read_u64 = mem_cgroup_move_charge_read,
3979 .write_u64 = mem_cgroup_move_charge_write,
3980 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003981 {
3982 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003983 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003984 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003985 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3986 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003987 {
3988 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003989 },
Ying Han406eb0c2011-05-26 16:25:37 -07003990#ifdef CONFIG_NUMA
3991 {
3992 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003993 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07003994 },
3995#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08003996 {
3997 .name = "kmem.limit_in_bytes",
3998 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003999 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004000 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004001 },
4002 {
4003 .name = "kmem.usage_in_bytes",
4004 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004005 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004006 },
4007 {
4008 .name = "kmem.failcnt",
4009 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004010 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004011 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004012 },
4013 {
4014 .name = "kmem.max_usage_in_bytes",
4015 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004016 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004017 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004018 },
Glauber Costa749c5412012-12-18 14:23:01 -08004019#ifdef CONFIG_SLABINFO
4020 {
4021 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004022 .seq_start = slab_start,
4023 .seq_next = slab_next,
4024 .seq_stop = slab_stop,
4025 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004026 },
4027#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004028 {
4029 .name = "kmem.tcp.limit_in_bytes",
4030 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4031 .write = mem_cgroup_write,
4032 .read_u64 = mem_cgroup_read_u64,
4033 },
4034 {
4035 .name = "kmem.tcp.usage_in_bytes",
4036 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4037 .read_u64 = mem_cgroup_read_u64,
4038 },
4039 {
4040 .name = "kmem.tcp.failcnt",
4041 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4042 .write = mem_cgroup_reset,
4043 .read_u64 = mem_cgroup_read_u64,
4044 },
4045 {
4046 .name = "kmem.tcp.max_usage_in_bytes",
4047 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4048 .write = mem_cgroup_reset,
4049 .read_u64 = mem_cgroup_read_u64,
4050 },
Tejun Heo6bc10342012-04-01 12:09:55 -07004051 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004052};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004053
Johannes Weiner73f576c2016-07-20 15:44:57 -07004054/*
4055 * Private memory cgroup IDR
4056 *
4057 * Swap-out records and page cache shadow entries need to store memcg
4058 * references in constrained space, so we maintain an ID space that is
4059 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4060 * memory-controlled cgroups to 64k.
4061 *
4062 * However, there usually are many references to the oflline CSS after
4063 * the cgroup has been destroyed, such as page cache or reclaimable
4064 * slab objects, that don't need to hang on to the ID. We want to keep
4065 * those dead CSS from occupying IDs, or we might quickly exhaust the
4066 * relatively small ID space and prevent the creation of new cgroups
4067 * even when there are much fewer than 64k cgroups - possibly none.
4068 *
4069 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4070 * be freed and recycled when it's no longer needed, which is usually
4071 * when the CSS is offlined.
4072 *
4073 * The only exception to that are records of swapped out tmpfs/shmem
4074 * pages that need to be attributed to live ancestors on swapin. But
4075 * those references are manageable from userspace.
4076 */
4077
4078static DEFINE_IDR(mem_cgroup_idr);
4079
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004080static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004081{
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004082 atomic_add(n, &memcg->id.ref);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004083}
4084
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004085static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004086{
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004087 if (atomic_sub_and_test(n, &memcg->id.ref)) {
Johannes Weiner73f576c2016-07-20 15:44:57 -07004088 idr_remove(&mem_cgroup_idr, memcg->id.id);
4089 memcg->id.id = 0;
4090
4091 /* Memcg ID pins CSS */
4092 css_put(&memcg->css);
4093 }
4094}
4095
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004096static inline void mem_cgroup_id_get(struct mem_cgroup *memcg)
4097{
4098 mem_cgroup_id_get_many(memcg, 1);
4099}
4100
4101static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
4102{
4103 mem_cgroup_id_put_many(memcg, 1);
4104}
4105
Johannes Weiner73f576c2016-07-20 15:44:57 -07004106/**
4107 * mem_cgroup_from_id - look up a memcg from a memcg id
4108 * @id: the memcg id to look up
4109 *
4110 * Caller must hold rcu_read_lock().
4111 */
4112struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
4113{
4114 WARN_ON_ONCE(!rcu_read_lock_held());
4115 return idr_find(&mem_cgroup_idr, id);
4116}
4117
Mel Gormanef8f2322016-07-28 15:46:05 -07004118static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004119{
4120 struct mem_cgroup_per_node *pn;
Mel Gormanef8f2322016-07-28 15:46:05 -07004121 int tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004122 /*
4123 * This routine is called against possible nodes.
4124 * But it's BUG to call kmalloc() against offline node.
4125 *
4126 * TODO: this routine can waste much memory for nodes which will
4127 * never be onlined. It's better to use memory hotplug callback
4128 * function.
4129 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004130 if (!node_state(node, N_NORMAL_MEMORY))
4131 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004132 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004133 if (!pn)
4134 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004135
Mel Gormanef8f2322016-07-28 15:46:05 -07004136 lruvec_init(&pn->lruvec);
4137 pn->usage_in_excess = 0;
4138 pn->on_tree = false;
4139 pn->memcg = memcg;
4140
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004141 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004142 return 0;
4143}
4144
Mel Gormanef8f2322016-07-28 15:46:05 -07004145static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004146{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004147 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004148}
4149
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004150static void mem_cgroup_free(struct mem_cgroup *memcg)
4151{
4152 int node;
4153
4154 memcg_wb_domain_exit(memcg);
4155 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004156 free_mem_cgroup_per_node_info(memcg, node);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004157 free_percpu(memcg->stat);
4158 kfree(memcg);
4159}
4160
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004161static struct mem_cgroup *mem_cgroup_alloc(void)
4162{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004163 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004164 size_t size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004165 int node;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004166
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004167 size = sizeof(struct mem_cgroup);
4168 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004169
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004170 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004171 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004172 return NULL;
4173
Johannes Weiner73f576c2016-07-20 15:44:57 -07004174 memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
4175 1, MEM_CGROUP_ID_MAX,
4176 GFP_KERNEL);
4177 if (memcg->id.id < 0)
4178 goto fail;
4179
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004180 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4181 if (!memcg->stat)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004182 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004183
Bob Liu3ed28fa2012-01-12 17:19:04 -08004184 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004185 if (alloc_mem_cgroup_per_node_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004186 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004187
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004188 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4189 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004190
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004191 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08004192 memcg->last_scanned_node = MAX_NUMNODES;
4193 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004194 mutex_init(&memcg->thresholds_lock);
4195 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004196 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004197 INIT_LIST_HEAD(&memcg->event_list);
4198 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004199 memcg->socket_pressure = jiffies;
Johannes Weiner127424c2016-01-20 15:02:32 -08004200#ifndef CONFIG_SLOB
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004201 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004202#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004203#ifdef CONFIG_CGROUP_WRITEBACK
4204 INIT_LIST_HEAD(&memcg->cgwb_list);
4205#endif
Johannes Weiner73f576c2016-07-20 15:44:57 -07004206 idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004207 return memcg;
4208fail:
Johannes Weiner73f576c2016-07-20 15:44:57 -07004209 if (memcg->id.id > 0)
4210 idr_remove(&mem_cgroup_idr, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004211 mem_cgroup_free(memcg);
4212 return NULL;
Glauber Costad142e3e2013-02-22 16:34:52 -08004213}
4214
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004215static struct cgroup_subsys_state * __ref
4216mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004217{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004218 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4219 struct mem_cgroup *memcg;
4220 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08004221
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004222 memcg = mem_cgroup_alloc();
4223 if (!memcg)
4224 return ERR_PTR(error);
Li Zefan4219b2d2013-09-23 16:56:29 +08004225
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004226 memcg->high = PAGE_COUNTER_MAX;
4227 memcg->soft_limit = PAGE_COUNTER_MAX;
4228 if (parent) {
4229 memcg->swappiness = mem_cgroup_swappiness(parent);
4230 memcg->oom_kill_disable = parent->oom_kill_disable;
4231 }
4232 if (parent && parent->use_hierarchy) {
4233 memcg->use_hierarchy = true;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004234 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004235 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004236 page_counter_init(&memcg->memsw, &parent->memsw);
4237 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08004238 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004239 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004240 page_counter_init(&memcg->memory, NULL);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004241 page_counter_init(&memcg->swap, NULL);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004242 page_counter_init(&memcg->memsw, NULL);
4243 page_counter_init(&memcg->kmem, NULL);
Johannes Weiner0db15292016-01-20 15:02:50 -08004244 page_counter_init(&memcg->tcpmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004245 /*
4246 * Deeper hierachy with use_hierarchy == false doesn't make
4247 * much sense so let cgroup subsystem know about this
4248 * unfortunate state in our controller.
4249 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004250 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004251 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004252 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08004253
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004254 /* The following stuff does not apply to the root */
4255 if (!parent) {
4256 root_mem_cgroup = memcg;
4257 return &memcg->css;
4258 }
4259
Vladimir Davydovb313aee2016-03-17 14:18:27 -07004260 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004261 if (error)
4262 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08004263
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004264 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004265 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004266
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004267 return &memcg->css;
4268fail:
4269 mem_cgroup_free(memcg);
Tejun Heoea3a9642016-06-24 14:49:58 -07004270 return ERR_PTR(-ENOMEM);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004271}
4272
Johannes Weiner73f576c2016-07-20 15:44:57 -07004273static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004274{
Johannes Weiner73f576c2016-07-20 15:44:57 -07004275 /* Online state pins memcg ID, memcg ID pins CSS */
4276 mem_cgroup_id_get(mem_cgroup_from_css(css));
4277 css_get(css);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004278 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004279}
4280
Tejun Heoeb954192013-08-08 20:11:23 -04004281static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004282{
Tejun Heoeb954192013-08-08 20:11:23 -04004283 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004284 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004285
4286 /*
4287 * Unregister events and notify userspace.
4288 * Notify userspace about cgroup removing only after rmdir of cgroup
4289 * directory to avoid race between userspace and kernelspace.
4290 */
Tejun Heofba94802013-11-22 18:20:43 -05004291 spin_lock(&memcg->event_list_lock);
4292 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004293 list_del_init(&event->list);
4294 schedule_work(&event->remove);
4295 }
Tejun Heofba94802013-11-22 18:20:43 -05004296 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004297
Johannes Weiner567e9ab2016-01-20 15:02:24 -08004298 memcg_offline_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004299 wb_memcg_offline(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004300
4301 mem_cgroup_id_put(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004302}
4303
Vladimir Davydov6df38682015-12-29 14:54:10 -08004304static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4305{
4306 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4307
4308 invalidate_reclaim_iterators(memcg);
4309}
4310
Tejun Heoeb954192013-08-08 20:11:23 -04004311static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004312{
Tejun Heoeb954192013-08-08 20:11:23 -04004313 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004314
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004315 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004316 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004317
Johannes Weiner0db15292016-01-20 15:02:50 -08004318 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004319 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004320
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004321 vmpressure_cleanup(&memcg->vmpressure);
4322 cancel_work_sync(&memcg->high_work);
4323 mem_cgroup_remove_from_trees(memcg);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004324 memcg_free_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004325 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004326}
4327
Tejun Heo1ced9532014-07-08 18:02:57 -04004328/**
4329 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4330 * @css: the target css
4331 *
4332 * Reset the states of the mem_cgroup associated with @css. This is
4333 * invoked when the userland requests disabling on the default hierarchy
4334 * but the memcg is pinned through dependency. The memcg should stop
4335 * applying policies and should revert to the vanilla state as it may be
4336 * made visible again.
4337 *
4338 * The current implementation only resets the essential configurations.
4339 * This needs to be expanded to cover all the visible parts.
4340 */
4341static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4342{
4343 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4344
Vladimir Davydovd334c9b2016-03-17 14:19:38 -07004345 page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4346 page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4347 page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4348 page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4349 page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
Johannes Weiner241994e2015-02-11 15:26:06 -08004350 memcg->low = 0;
4351 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004352 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004353 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004354}
4355
Daisuke Nishimura02491442010-03-10 15:22:17 -08004356#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004357/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004358static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004359{
Johannes Weiner05b84302014-08-06 16:05:59 -07004360 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004361
Mel Gormand0164ad2015-11-06 16:28:21 -08004362 /* Try a single bulk charge without reclaim first, kswapd may wake */
4363 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004364 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004365 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004366 return ret;
4367 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004368
4369 /* Try charges one by one with reclaim */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004370 while (count--) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004371 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004372 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004373 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004374 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004375 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004376 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004377 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004378}
4379
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004380union mc_target {
4381 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004382 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004383};
4384
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004385enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004386 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004387 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004388 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004389};
4390
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004391static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4392 unsigned long addr, pte_t ptent)
4393{
4394 struct page *page = vm_normal_page(vma, addr, ptent);
4395
4396 if (!page || !page_mapped(page))
4397 return NULL;
4398 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004399 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004400 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004401 } else {
4402 if (!(mc.flags & MOVE_FILE))
4403 return NULL;
4404 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004405 if (!get_page_unless_zero(page))
4406 return NULL;
4407
4408 return page;
4409}
4410
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004411#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004412static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004413 pte_t ptent, swp_entry_t *entry)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004414{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004415 struct page *page = NULL;
4416 swp_entry_t ent = pte_to_swp_entry(ptent);
4417
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004418 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004419 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004420 /*
4421 * Because lookup_swap_cache() updates some statistics counter,
4422 * we call find_get_page() with swapper_space directly.
4423 */
Shaohua Li33806f02013-02-22 16:34:37 -08004424 page = find_get_page(swap_address_space(ent), ent.val);
Johannes Weiner7941d212016-01-14 15:21:23 -08004425 if (do_memsw_account())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004426 entry->val = ent.val;
4427
4428 return page;
4429}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004430#else
4431static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004432 pte_t ptent, swp_entry_t *entry)
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004433{
4434 return NULL;
4435}
4436#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004437
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004438static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4439 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4440{
4441 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004442 struct address_space *mapping;
4443 pgoff_t pgoff;
4444
4445 if (!vma->vm_file) /* anonymous vma */
4446 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004447 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004448 return NULL;
4449
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004450 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004451 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004452
4453 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004454#ifdef CONFIG_SWAP
4455 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004456 if (shmem_mapping(mapping)) {
4457 page = find_get_entry(mapping, pgoff);
4458 if (radix_tree_exceptional_entry(page)) {
4459 swp_entry_t swp = radix_to_swp_entry(page);
Johannes Weiner7941d212016-01-14 15:21:23 -08004460 if (do_memsw_account())
Johannes Weiner139b6a62014-05-06 12:50:05 -07004461 *entry = swp;
4462 page = find_get_page(swap_address_space(swp), swp.val);
4463 }
4464 } else
4465 page = find_get_page(mapping, pgoff);
4466#else
4467 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004468#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004469 return page;
4470}
4471
Chen Gangb1b0dea2015-04-14 15:47:35 -07004472/**
4473 * mem_cgroup_move_account - move account of the page
4474 * @page: the page
Li RongQing25843c22016-07-26 15:26:56 -07004475 * @compound: charge the page as compound or small page
Chen Gangb1b0dea2015-04-14 15:47:35 -07004476 * @from: mem_cgroup which the page is moved from.
4477 * @to: mem_cgroup which the page is moved to. @from != @to.
4478 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08004479 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07004480 *
4481 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4482 * from old cgroup.
4483 */
4484static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004485 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07004486 struct mem_cgroup *from,
4487 struct mem_cgroup *to)
4488{
4489 unsigned long flags;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004490 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004491 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004492 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004493
4494 VM_BUG_ON(from == to);
4495 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004496 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07004497
4498 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07004499 * Prevent mem_cgroup_migrate() from looking at
Hugh Dickins45637ba2015-11-05 18:49:40 -08004500 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07004501 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004502 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004503 if (!trylock_page(page))
4504 goto out;
4505
4506 ret = -EINVAL;
4507 if (page->mem_cgroup != from)
4508 goto out_unlock;
4509
Greg Thelenc4843a72015-05-22 17:13:16 -04004510 anon = PageAnon(page);
4511
Chen Gangb1b0dea2015-04-14 15:47:35 -07004512 spin_lock_irqsave(&from->move_lock, flags);
4513
Greg Thelenc4843a72015-05-22 17:13:16 -04004514 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004515 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4516 nr_pages);
4517 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4518 nr_pages);
4519 }
4520
Greg Thelenc4843a72015-05-22 17:13:16 -04004521 /*
4522 * move_lock grabbed above and caller set from->moving_account, so
4523 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4524 * So mapping should be stable for dirty pages.
4525 */
4526 if (!anon && PageDirty(page)) {
4527 struct address_space *mapping = page_mapping(page);
4528
4529 if (mapping_cap_account_dirty(mapping)) {
4530 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4531 nr_pages);
4532 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4533 nr_pages);
4534 }
4535 }
4536
Chen Gangb1b0dea2015-04-14 15:47:35 -07004537 if (PageWriteback(page)) {
4538 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4539 nr_pages);
4540 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4541 nr_pages);
4542 }
4543
4544 /*
4545 * It is safe to change page->mem_cgroup here because the page
4546 * is referenced, charged, and isolated - we can't race with
4547 * uncharging, charging, migration, or LRU putback.
4548 */
4549
4550 /* caller should have done css_get */
4551 page->mem_cgroup = to;
4552 spin_unlock_irqrestore(&from->move_lock, flags);
4553
4554 ret = 0;
4555
4556 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004557 mem_cgroup_charge_statistics(to, page, compound, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004558 memcg_check_events(to, page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004559 mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004560 memcg_check_events(from, page);
4561 local_irq_enable();
4562out_unlock:
4563 unlock_page(page);
4564out:
4565 return ret;
4566}
4567
Li RongQing7cf78062016-05-27 14:27:46 -07004568/**
4569 * get_mctgt_type - get target type of moving charge
4570 * @vma: the vma the pte to be checked belongs
4571 * @addr: the address corresponding to the pte to be checked
4572 * @ptent: the pte to be checked
4573 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4574 *
4575 * Returns
4576 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4577 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4578 * move charge. if @target is not NULL, the page is stored in target->page
4579 * with extra refcnt got(Callers should handle it).
4580 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4581 * target for charge migration. if @target is not NULL, the entry is stored
4582 * in target->ent.
4583 *
4584 * Called with pte lock held.
4585 */
4586
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004587static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004588 unsigned long addr, pte_t ptent, union mc_target *target)
4589{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004590 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004591 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004592 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004593
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004594 if (pte_present(ptent))
4595 page = mc_handle_present_pte(vma, addr, ptent);
4596 else if (is_swap_pte(ptent))
Li RongQing48406ef2016-07-26 15:22:14 -07004597 page = mc_handle_swap_pte(vma, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004598 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004599 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004600
4601 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004602 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004603 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004604 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004605 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004606 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004607 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004608 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004609 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004610 ret = MC_TARGET_PAGE;
4611 if (target)
4612 target->page = page;
4613 }
4614 if (!ret || !target)
4615 put_page(page);
4616 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004617 /* There is a swap entry and a page doesn't exist or isn't charged */
4618 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004619 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004620 ret = MC_TARGET_SWAP;
4621 if (target)
4622 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004623 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004624 return ret;
4625}
4626
Naoya Horiguchi12724852012-03-21 16:34:28 -07004627#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4628/*
4629 * We don't consider swapping or file mapped pages because THP does not
4630 * support them for now.
4631 * Caller should make sure that pmd_trans_huge(pmd) is true.
4632 */
4633static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4634 unsigned long addr, pmd_t pmd, union mc_target *target)
4635{
4636 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004637 enum mc_target_type ret = MC_TARGET_NONE;
4638
4639 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004640 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004641 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004642 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004643 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004644 ret = MC_TARGET_PAGE;
4645 if (target) {
4646 get_page(page);
4647 target->page = page;
4648 }
4649 }
4650 return ret;
4651}
4652#else
4653static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4654 unsigned long addr, pmd_t pmd, union mc_target *target)
4655{
4656 return MC_TARGET_NONE;
4657}
4658#endif
4659
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004660static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4661 unsigned long addr, unsigned long end,
4662 struct mm_walk *walk)
4663{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004664 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004665 pte_t *pte;
4666 spinlock_t *ptl;
4667
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004668 ptl = pmd_trans_huge_lock(pmd, vma);
4669 if (ptl) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004670 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4671 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004672 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004673 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004674 }
Dave Hansen03319322011-03-22 16:32:56 -07004675
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004676 if (pmd_trans_unstable(pmd))
4677 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004678 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4679 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004680 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004681 mc.precharge++; /* increment precharge temporarily */
4682 pte_unmap_unlock(pte - 1, ptl);
4683 cond_resched();
4684
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004685 return 0;
4686}
4687
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004688static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4689{
4690 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004691
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004692 struct mm_walk mem_cgroup_count_precharge_walk = {
4693 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4694 .mm = mm,
4695 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004696 down_read(&mm->mmap_sem);
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004697 walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004698 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004699
4700 precharge = mc.precharge;
4701 mc.precharge = 0;
4702
4703 return precharge;
4704}
4705
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004706static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4707{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004708 unsigned long precharge = mem_cgroup_count_precharge(mm);
4709
4710 VM_BUG_ON(mc.moving_task);
4711 mc.moving_task = current;
4712 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004713}
4714
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004715/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4716static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004717{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004718 struct mem_cgroup *from = mc.from;
4719 struct mem_cgroup *to = mc.to;
4720
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004721 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004722 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004723 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004724 mc.precharge = 0;
4725 }
4726 /*
4727 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4728 * we must uncharge here.
4729 */
4730 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004731 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004732 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004733 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004734 /* we must fixup refcnts and charges */
4735 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004736 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004737 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004738 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004739
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004740 mem_cgroup_id_put_many(mc.from, mc.moved_swap);
4741
Johannes Weiner05b84302014-08-06 16:05:59 -07004742 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004743 * we charged both to->memory and to->memsw, so we
4744 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004745 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004746 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004747 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004748
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004749 mem_cgroup_id_get_many(mc.to, mc.moved_swap);
4750 css_put_many(&mc.to->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004751
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004752 mc.moved_swap = 0;
4753 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004754 memcg_oom_recover(from);
4755 memcg_oom_recover(to);
4756 wake_up_all(&mc.waitq);
4757}
4758
4759static void mem_cgroup_clear_mc(void)
4760{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004761 struct mm_struct *mm = mc.mm;
4762
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004763 /*
4764 * we must clear moving_task before waking up waiters at the end of
4765 * task migration.
4766 */
4767 mc.moving_task = NULL;
4768 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004769 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004770 mc.from = NULL;
4771 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04004772 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004773 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004774
4775 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004776}
4777
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004778static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004779{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004780 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07004781 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07004782 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04004783 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004784 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004785 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004786 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004787
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004788 /* charge immigration isn't supported on the default hierarchy */
4789 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07004790 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004791
Tejun Heo4530edd2015-09-11 15:00:19 -04004792 /*
4793 * Multi-process migrations only happen on the default hierarchy
4794 * where charge immigration is not used. Perform charge
4795 * immigration if @tset contains a leader and whine if there are
4796 * multiple.
4797 */
4798 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004799 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04004800 WARN_ON_ONCE(p);
4801 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004802 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04004803 }
4804 if (!p)
4805 return 0;
4806
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004807 /*
4808 * We are now commited to this value whatever it is. Changes in this
4809 * tunable will only affect upcoming migrations, not the current one.
4810 * So we need to save it, and keep it going.
4811 */
4812 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4813 if (!move_flags)
4814 return 0;
4815
Tejun Heo9f2115f2015-09-08 15:01:10 -07004816 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004817
Tejun Heo9f2115f2015-09-08 15:01:10 -07004818 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004819
Tejun Heo9f2115f2015-09-08 15:01:10 -07004820 mm = get_task_mm(p);
4821 if (!mm)
4822 return 0;
4823 /* We move charges only when we move a owner of the mm */
4824 if (mm->owner == p) {
4825 VM_BUG_ON(mc.from);
4826 VM_BUG_ON(mc.to);
4827 VM_BUG_ON(mc.precharge);
4828 VM_BUG_ON(mc.moved_charge);
4829 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004830
Tejun Heo9f2115f2015-09-08 15:01:10 -07004831 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004832 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004833 mc.from = from;
4834 mc.to = memcg;
4835 mc.flags = move_flags;
4836 spin_unlock(&mc.lock);
4837 /* We set mc.moving_task later */
4838
4839 ret = mem_cgroup_precharge_mc(mm);
4840 if (ret)
4841 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004842 } else {
4843 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004844 }
4845 return ret;
4846}
4847
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004848static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004849{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004850 if (mc.to)
4851 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004852}
4853
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004854static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4855 unsigned long addr, unsigned long end,
4856 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004857{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004858 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004859 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004860 pte_t *pte;
4861 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004862 enum mc_target_type target_type;
4863 union mc_target target;
4864 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004865
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004866 ptl = pmd_trans_huge_lock(pmd, vma);
4867 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004868 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004869 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004870 return 0;
4871 }
4872 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4873 if (target_type == MC_TARGET_PAGE) {
4874 page = target.page;
4875 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004876 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08004877 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004878 mc.precharge -= HPAGE_PMD_NR;
4879 mc.moved_charge += HPAGE_PMD_NR;
4880 }
4881 putback_lru_page(page);
4882 }
4883 put_page(page);
4884 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004885 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004886 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004887 }
4888
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004889 if (pmd_trans_unstable(pmd))
4890 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004891retry:
4892 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4893 for (; addr != end; addr += PAGE_SIZE) {
4894 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004895 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004896
4897 if (!mc.precharge)
4898 break;
4899
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004900 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004901 case MC_TARGET_PAGE:
4902 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08004903 /*
4904 * We can have a part of the split pmd here. Moving it
4905 * can be done but it would be too convoluted so simply
4906 * ignore such a partial THP and keep it in original
4907 * memcg. There should be somebody mapping the head.
4908 */
4909 if (PageTransCompound(page))
4910 goto put;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004911 if (isolate_lru_page(page))
4912 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004913 if (!mem_cgroup_move_account(page, false,
4914 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004915 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004916 /* we uncharge from mc.from later. */
4917 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004918 }
4919 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004920put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004921 put_page(page);
4922 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004923 case MC_TARGET_SWAP:
4924 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004925 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004926 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004927 /* we fixup refcnts and charges later. */
4928 mc.moved_swap++;
4929 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004930 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004931 default:
4932 break;
4933 }
4934 }
4935 pte_unmap_unlock(pte - 1, ptl);
4936 cond_resched();
4937
4938 if (addr != end) {
4939 /*
4940 * We have consumed all precharges we got in can_attach().
4941 * We try charge one by one, but don't do any additional
4942 * charges to mc.to if we have failed in charge once in attach()
4943 * phase.
4944 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004945 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004946 if (!ret)
4947 goto retry;
4948 }
4949
4950 return ret;
4951}
4952
Tejun Heo264a0ae2016-04-21 19:09:02 -04004953static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004954{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004955 struct mm_walk mem_cgroup_move_charge_walk = {
4956 .pmd_entry = mem_cgroup_move_charge_pte_range,
Tejun Heo264a0ae2016-04-21 19:09:02 -04004957 .mm = mc.mm,
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004958 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004959
4960 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004961 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07004962 * Signal lock_page_memcg() to take the memcg's move_lock
4963 * while we're moving its pages to another memcg. Then wait
4964 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08004965 */
4966 atomic_inc(&mc.from->moving_account);
4967 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004968retry:
Tejun Heo264a0ae2016-04-21 19:09:02 -04004969 if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004970 /*
4971 * Someone who are holding the mmap_sem might be waiting in
4972 * waitq. So we cancel all extra charges, wake up all waiters,
4973 * and retry. Because we cancel precharges, we might not be able
4974 * to move enough charges, but moving charge is a best-effort
4975 * feature anyway, so it wouldn't be a big problem.
4976 */
4977 __mem_cgroup_clear_mc();
4978 cond_resched();
4979 goto retry;
4980 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004981 /*
4982 * When we have consumed all precharges and failed in doing
4983 * additional charge, the page walk just aborts.
4984 */
4985 walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004986 up_read(&mc.mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08004987 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004988}
4989
Tejun Heo264a0ae2016-04-21 19:09:02 -04004990static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08004991{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004992 if (mc.to) {
4993 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004994 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004995 }
Balbir Singh67e465a2008-02-07 00:13:54 -08004996}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004997#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004998static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004999{
5000 return 0;
5001}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005002static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005003{
5004}
Tejun Heo264a0ae2016-04-21 19:09:02 -04005005static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005006{
5007}
5008#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005009
Tejun Heof00baae2013-04-15 13:41:15 -07005010/*
5011 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005012 * to verify whether we're attached to the default hierarchy on each mount
5013 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005014 */
Tejun Heoeb954192013-08-08 20:11:23 -04005015static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005016{
5017 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005018 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005019 * guarantees that @root doesn't have any children, so turning it
5020 * on for the root memcg is enough.
5021 */
Tejun Heo9e10a132015-09-18 11:56:28 -04005022 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee592015-03-12 16:26:19 -07005023 root_mem_cgroup->use_hierarchy = true;
5024 else
5025 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005026}
5027
Johannes Weiner241994e2015-02-11 15:26:06 -08005028static u64 memory_current_read(struct cgroup_subsys_state *css,
5029 struct cftype *cft)
5030{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005031 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5032
5033 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994e2015-02-11 15:26:06 -08005034}
5035
5036static int memory_low_show(struct seq_file *m, void *v)
5037{
5038 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005039 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994e2015-02-11 15:26:06 -08005040
5041 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005042 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005043 else
5044 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5045
5046 return 0;
5047}
5048
5049static ssize_t memory_low_write(struct kernfs_open_file *of,
5050 char *buf, size_t nbytes, loff_t off)
5051{
5052 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5053 unsigned long low;
5054 int err;
5055
5056 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005057 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994e2015-02-11 15:26:06 -08005058 if (err)
5059 return err;
5060
5061 memcg->low = low;
5062
5063 return nbytes;
5064}
5065
5066static int memory_high_show(struct seq_file *m, void *v)
5067{
5068 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005069 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994e2015-02-11 15:26:06 -08005070
5071 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005072 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005073 else
5074 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5075
5076 return 0;
5077}
5078
5079static ssize_t memory_high_write(struct kernfs_open_file *of,
5080 char *buf, size_t nbytes, loff_t off)
5081{
5082 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner588083b2016-03-17 14:20:25 -07005083 unsigned long nr_pages;
Johannes Weiner241994e2015-02-11 15:26:06 -08005084 unsigned long high;
5085 int err;
5086
5087 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005088 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994e2015-02-11 15:26:06 -08005089 if (err)
5090 return err;
5091
5092 memcg->high = high;
5093
Johannes Weiner588083b2016-03-17 14:20:25 -07005094 nr_pages = page_counter_read(&memcg->memory);
5095 if (nr_pages > high)
5096 try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5097 GFP_KERNEL, true);
5098
Tejun Heo2529bb32015-05-22 18:23:34 -04005099 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005100 return nbytes;
5101}
5102
5103static int memory_max_show(struct seq_file *m, void *v)
5104{
5105 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005106 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994e2015-02-11 15:26:06 -08005107
5108 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005109 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005110 else
5111 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5112
5113 return 0;
5114}
5115
5116static ssize_t memory_max_write(struct kernfs_open_file *of,
5117 char *buf, size_t nbytes, loff_t off)
5118{
5119 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005120 unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5121 bool drained = false;
Johannes Weiner241994e2015-02-11 15:26:06 -08005122 unsigned long max;
5123 int err;
5124
5125 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005126 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994e2015-02-11 15:26:06 -08005127 if (err)
5128 return err;
5129
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005130 xchg(&memcg->memory.limit, max);
5131
5132 for (;;) {
5133 unsigned long nr_pages = page_counter_read(&memcg->memory);
5134
5135 if (nr_pages <= max)
5136 break;
5137
5138 if (signal_pending(current)) {
5139 err = -EINTR;
5140 break;
5141 }
5142
5143 if (!drained) {
5144 drain_all_stock(memcg);
5145 drained = true;
5146 continue;
5147 }
5148
5149 if (nr_reclaims) {
5150 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5151 GFP_KERNEL, true))
5152 nr_reclaims--;
5153 continue;
5154 }
5155
5156 mem_cgroup_events(memcg, MEMCG_OOM, 1);
5157 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5158 break;
5159 }
Johannes Weiner241994e2015-02-11 15:26:06 -08005160
Tejun Heo2529bb32015-05-22 18:23:34 -04005161 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005162 return nbytes;
5163}
5164
5165static int memory_events_show(struct seq_file *m, void *v)
5166{
5167 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5168
5169 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5170 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5171 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5172 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5173
5174 return 0;
5175}
5176
Johannes Weiner587d9f72016-01-20 15:03:19 -08005177static int memory_stat_show(struct seq_file *m, void *v)
5178{
5179 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005180 unsigned long stat[MEMCG_NR_STAT];
5181 unsigned long events[MEMCG_NR_EVENTS];
Johannes Weiner587d9f72016-01-20 15:03:19 -08005182 int i;
5183
5184 /*
5185 * Provide statistics on the state of the memory subsystem as
5186 * well as cumulative event counters that show past behavior.
5187 *
5188 * This list is ordered following a combination of these gradients:
5189 * 1) generic big picture -> specifics and details
5190 * 2) reflecting userspace activity -> reflecting kernel heuristics
5191 *
5192 * Current memory state:
5193 */
5194
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005195 tree_stat(memcg, stat);
5196 tree_events(memcg, events);
5197
Johannes Weiner587d9f72016-01-20 15:03:19 -08005198 seq_printf(m, "anon %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005199 (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005200 seq_printf(m, "file %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005201 (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE);
Vladimir Davydov12580e42016-03-17 14:17:38 -07005202 seq_printf(m, "kernel_stack %llu\n",
Andy Lutomirskiefdc9492016-07-28 15:48:17 -07005203 (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005204 seq_printf(m, "slab %llu\n",
5205 (u64)(stat[MEMCG_SLAB_RECLAIMABLE] +
5206 stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
Johannes Weinerb2807f02016-01-20 15:03:22 -08005207 seq_printf(m, "sock %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005208 (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005209
5210 seq_printf(m, "file_mapped %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005211 (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005212 seq_printf(m, "file_dirty %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005213 (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005214 seq_printf(m, "file_writeback %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005215 (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005216
5217 for (i = 0; i < NR_LRU_LISTS; i++) {
5218 struct mem_cgroup *mi;
5219 unsigned long val = 0;
5220
5221 for_each_mem_cgroup_tree(mi, memcg)
5222 val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5223 seq_printf(m, "%s %llu\n",
5224 mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5225 }
5226
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005227 seq_printf(m, "slab_reclaimable %llu\n",
5228 (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE);
5229 seq_printf(m, "slab_unreclaimable %llu\n",
5230 (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5231
Johannes Weiner587d9f72016-01-20 15:03:19 -08005232 /* Accumulated memory events */
5233
5234 seq_printf(m, "pgfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005235 events[MEM_CGROUP_EVENTS_PGFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005236 seq_printf(m, "pgmajfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005237 events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005238
5239 return 0;
5240}
5241
Johannes Weiner241994e2015-02-11 15:26:06 -08005242static struct cftype memory_files[] = {
5243 {
5244 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005245 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994e2015-02-11 15:26:06 -08005246 .read_u64 = memory_current_read,
5247 },
5248 {
5249 .name = "low",
5250 .flags = CFTYPE_NOT_ON_ROOT,
5251 .seq_show = memory_low_show,
5252 .write = memory_low_write,
5253 },
5254 {
5255 .name = "high",
5256 .flags = CFTYPE_NOT_ON_ROOT,
5257 .seq_show = memory_high_show,
5258 .write = memory_high_write,
5259 },
5260 {
5261 .name = "max",
5262 .flags = CFTYPE_NOT_ON_ROOT,
5263 .seq_show = memory_max_show,
5264 .write = memory_max_write,
5265 },
5266 {
5267 .name = "events",
5268 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04005269 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994e2015-02-11 15:26:06 -08005270 .seq_show = memory_events_show,
5271 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08005272 {
5273 .name = "stat",
5274 .flags = CFTYPE_NOT_ON_ROOT,
5275 .seq_show = memory_stat_show,
5276 },
Johannes Weiner241994e2015-02-11 15:26:06 -08005277 { } /* terminate */
5278};
5279
Tejun Heo073219e2014-02-08 10:36:58 -05005280struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005281 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005282 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005283 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08005284 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08005285 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005286 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005287 .can_attach = mem_cgroup_can_attach,
5288 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04005289 .post_attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005290 .bind = mem_cgroup_bind,
Johannes Weiner241994e2015-02-11 15:26:06 -08005291 .dfl_cftypes = memory_files,
5292 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005293 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005294};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005295
Johannes Weiner241994e2015-02-11 15:26:06 -08005296/**
Johannes Weiner241994e2015-02-11 15:26:06 -08005297 * mem_cgroup_low - check if memory consumption is below the normal range
5298 * @root: the highest ancestor to consider
5299 * @memcg: the memory cgroup to check
5300 *
5301 * Returns %true if memory consumption of @memcg, and that of all
5302 * configurable ancestors up to @root, is below the normal range.
5303 */
5304bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5305{
5306 if (mem_cgroup_disabled())
5307 return false;
5308
5309 /*
5310 * The toplevel group doesn't have a configurable range, so
5311 * it's never low when looked at directly, and it is not
5312 * considered an ancestor when assessing the hierarchy.
5313 */
5314
5315 if (memcg == root_mem_cgroup)
5316 return false;
5317
Michal Hocko4e54ded2015-02-27 15:51:46 -08005318 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005319 return false;
5320
5321 while (memcg != root) {
5322 memcg = parent_mem_cgroup(memcg);
5323
5324 if (memcg == root_mem_cgroup)
5325 break;
5326
Michal Hocko4e54ded2015-02-27 15:51:46 -08005327 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005328 return false;
5329 }
5330 return true;
5331}
5332
Johannes Weiner00501b52014-08-08 14:19:20 -07005333/**
5334 * mem_cgroup_try_charge - try charging a page
5335 * @page: page to charge
5336 * @mm: mm context of the victim
5337 * @gfp_mask: reclaim mode
5338 * @memcgp: charged memcg return
Li RongQing25843c22016-07-26 15:26:56 -07005339 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005340 *
5341 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5342 * pages according to @gfp_mask if necessary.
5343 *
5344 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5345 * Otherwise, an error code is returned.
5346 *
5347 * After page->mapping has been set up, the caller must finalize the
5348 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5349 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5350 */
5351int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005352 gfp_t gfp_mask, struct mem_cgroup **memcgp,
5353 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005354{
5355 struct mem_cgroup *memcg = NULL;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005356 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005357 int ret = 0;
5358
5359 if (mem_cgroup_disabled())
5360 goto out;
5361
5362 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005363 /*
5364 * Every swap fault against a single page tries to charge the
5365 * page, bail as early as possible. shmem_unuse() encounters
5366 * already charged pages, too. The USED bit is protected by
5367 * the page lock, which serializes swap cache removal, which
5368 * in turn serializes uncharging.
5369 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005370 VM_BUG_ON_PAGE(!PageLocked(page), page);
Johannes Weiner1306a852014-12-10 15:44:52 -08005371 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005372 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005373
Vladimir Davydov37e84352016-01-20 15:02:56 -08005374 if (do_swap_account) {
Vladimir Davydove993d902015-09-09 15:35:35 -07005375 swp_entry_t ent = { .val = page_private(page), };
5376 unsigned short id = lookup_swap_cgroup_id(ent);
5377
5378 rcu_read_lock();
5379 memcg = mem_cgroup_from_id(id);
5380 if (memcg && !css_tryget_online(&memcg->css))
5381 memcg = NULL;
5382 rcu_read_unlock();
5383 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005384 }
5385
Johannes Weiner00501b52014-08-08 14:19:20 -07005386 if (!memcg)
5387 memcg = get_mem_cgroup_from_mm(mm);
5388
5389 ret = try_charge(memcg, gfp_mask, nr_pages);
5390
5391 css_put(&memcg->css);
Johannes Weiner00501b52014-08-08 14:19:20 -07005392out:
5393 *memcgp = memcg;
5394 return ret;
5395}
5396
5397/**
5398 * mem_cgroup_commit_charge - commit a page charge
5399 * @page: page to charge
5400 * @memcg: memcg to charge the page to
5401 * @lrucare: page might be on LRU already
Li RongQing25843c22016-07-26 15:26:56 -07005402 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005403 *
5404 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5405 * after page->mapping has been set up. This must happen atomically
5406 * as part of the page instantiation, i.e. under the page table lock
5407 * for anonymous pages, under the page lock for page and swap cache.
5408 *
5409 * In addition, the page must not be on the LRU during the commit, to
5410 * prevent racing with task migration. If it might be, use @lrucare.
5411 *
5412 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5413 */
5414void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005415 bool lrucare, bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005416{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005417 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005418
5419 VM_BUG_ON_PAGE(!page->mapping, page);
5420 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5421
5422 if (mem_cgroup_disabled())
5423 return;
5424 /*
5425 * Swap faults will attempt to charge the same page multiple
5426 * times. But reuse_swap_page() might have removed the page
5427 * from swapcache already, so we can't check PageSwapCache().
5428 */
5429 if (!memcg)
5430 return;
5431
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005432 commit_charge(page, memcg, lrucare);
5433
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005434 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005435 mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005436 memcg_check_events(memcg, page);
5437 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005438
Johannes Weiner7941d212016-01-14 15:21:23 -08005439 if (do_memsw_account() && PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005440 swp_entry_t entry = { .val = page_private(page) };
5441 /*
5442 * The swap entry might not get freed for a long time,
5443 * let's not wait for it. The page already received a
5444 * memory+swap charge, drop the swap entry duplicate.
5445 */
5446 mem_cgroup_uncharge_swap(entry);
5447 }
5448}
5449
5450/**
5451 * mem_cgroup_cancel_charge - cancel a page charge
5452 * @page: page to charge
5453 * @memcg: memcg to charge the page to
Li RongQing25843c22016-07-26 15:26:56 -07005454 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005455 *
5456 * Cancel a charge transaction started by mem_cgroup_try_charge().
5457 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005458void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5459 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005460{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005461 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005462
5463 if (mem_cgroup_disabled())
5464 return;
5465 /*
5466 * Swap faults will attempt to charge the same page multiple
5467 * times. But reuse_swap_page() might have removed the page
5468 * from swapcache already, so we can't check PageSwapCache().
5469 */
5470 if (!memcg)
5471 return;
5472
Johannes Weiner00501b52014-08-08 14:19:20 -07005473 cancel_charge(memcg, nr_pages);
5474}
5475
Johannes Weiner747db952014-08-08 14:19:24 -07005476static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005477 unsigned long nr_anon, unsigned long nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005478 unsigned long nr_huge, unsigned long nr_kmem,
5479 struct page *dummy_page)
Johannes Weiner747db952014-08-08 14:19:24 -07005480{
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005481 unsigned long nr_pages = nr_anon + nr_file + nr_kmem;
Johannes Weiner747db952014-08-08 14:19:24 -07005482 unsigned long flags;
5483
Johannes Weinerce00a962014-09-05 08:43:57 -04005484 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005485 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08005486 if (do_memsw_account())
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005487 page_counter_uncharge(&memcg->memsw, nr_pages);
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005488 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem)
5489 page_counter_uncharge(&memcg->kmem, nr_kmem);
Johannes Weinerce00a962014-09-05 08:43:57 -04005490 memcg_oom_recover(memcg);
5491 }
Johannes Weiner747db952014-08-08 14:19:24 -07005492
5493 local_irq_save(flags);
5494 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5495 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5496 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5497 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005498 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005499 memcg_check_events(memcg, dummy_page);
5500 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005501
5502 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005503 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005504}
5505
5506static void uncharge_list(struct list_head *page_list)
5507{
5508 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005509 unsigned long nr_anon = 0;
5510 unsigned long nr_file = 0;
5511 unsigned long nr_huge = 0;
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005512 unsigned long nr_kmem = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005513 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005514 struct list_head *next;
5515 struct page *page;
5516
Johannes Weiner8b592652016-03-17 14:20:31 -07005517 /*
5518 * Note that the list can be a single page->lru; hence the
5519 * do-while loop instead of a simple list_for_each_entry().
5520 */
Johannes Weiner747db952014-08-08 14:19:24 -07005521 next = page_list->next;
5522 do {
Johannes Weiner747db952014-08-08 14:19:24 -07005523 page = list_entry(next, struct page, lru);
5524 next = page->lru.next;
5525
5526 VM_BUG_ON_PAGE(PageLRU(page), page);
5527 VM_BUG_ON_PAGE(page_count(page), page);
5528
Johannes Weiner1306a852014-12-10 15:44:52 -08005529 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005530 continue;
5531
5532 /*
5533 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005534 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005535 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005536 */
5537
Johannes Weiner1306a852014-12-10 15:44:52 -08005538 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005539 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005540 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005541 nr_huge, nr_kmem, page);
5542 pgpgout = nr_anon = nr_file =
5543 nr_huge = nr_kmem = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005544 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005545 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005546 }
5547
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005548 if (!PageKmemcg(page)) {
5549 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005550
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005551 if (PageTransHuge(page)) {
5552 nr_pages <<= compound_order(page);
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005553 nr_huge += nr_pages;
5554 }
5555 if (PageAnon(page))
5556 nr_anon += nr_pages;
5557 else
5558 nr_file += nr_pages;
5559 pgpgout++;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03005560 } else {
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005561 nr_kmem += 1 << compound_order(page);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03005562 __ClearPageKmemcg(page);
5563 }
Johannes Weiner747db952014-08-08 14:19:24 -07005564
Johannes Weiner1306a852014-12-10 15:44:52 -08005565 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005566 } while (next != page_list);
5567
5568 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005569 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005570 nr_huge, nr_kmem, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005571}
5572
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005573/**
5574 * mem_cgroup_uncharge - uncharge a page
5575 * @page: page to uncharge
5576 *
5577 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5578 * mem_cgroup_commit_charge().
5579 */
5580void mem_cgroup_uncharge(struct page *page)
5581{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005582 if (mem_cgroup_disabled())
5583 return;
5584
Johannes Weiner747db952014-08-08 14:19:24 -07005585 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005586 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005587 return;
5588
Johannes Weiner747db952014-08-08 14:19:24 -07005589 INIT_LIST_HEAD(&page->lru);
5590 uncharge_list(&page->lru);
5591}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005592
Johannes Weiner747db952014-08-08 14:19:24 -07005593/**
5594 * mem_cgroup_uncharge_list - uncharge a list of page
5595 * @page_list: list of pages to uncharge
5596 *
5597 * Uncharge a list of pages previously charged with
5598 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5599 */
5600void mem_cgroup_uncharge_list(struct list_head *page_list)
5601{
5602 if (mem_cgroup_disabled())
5603 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005604
Johannes Weiner747db952014-08-08 14:19:24 -07005605 if (!list_empty(page_list))
5606 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005607}
5608
5609/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005610 * mem_cgroup_migrate - charge a page's replacement
5611 * @oldpage: currently circulating page
5612 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005613 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005614 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5615 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005616 *
5617 * Both pages must be locked, @newpage->mapping must be set up.
5618 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005619void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005620{
Johannes Weiner29833312014-12-10 15:44:02 -08005621 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005622 unsigned int nr_pages;
5623 bool compound;
Tejun Heod93c4132016-06-24 14:49:54 -07005624 unsigned long flags;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005625
5626 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5627 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005628 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005629 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5630 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005631
5632 if (mem_cgroup_disabled())
5633 return;
5634
5635 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005636 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005637 return;
5638
Hugh Dickins45637ba2015-11-05 18:49:40 -08005639 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08005640 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005641 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005642 return;
5643
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005644 /* Force-charge the new page. The old one will be freed soon */
5645 compound = PageTransHuge(newpage);
5646 nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5647
5648 page_counter_charge(&memcg->memory, nr_pages);
5649 if (do_memsw_account())
5650 page_counter_charge(&memcg->memsw, nr_pages);
5651 css_get_many(&memcg->css, nr_pages);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005652
Johannes Weiner9cf76662016-03-15 14:57:58 -07005653 commit_charge(newpage, memcg, false);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005654
Tejun Heod93c4132016-06-24 14:49:54 -07005655 local_irq_save(flags);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005656 mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5657 memcg_check_events(memcg, newpage);
Tejun Heod93c4132016-06-24 14:49:54 -07005658 local_irq_restore(flags);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005659}
5660
Johannes Weineref129472016-01-14 15:21:34 -08005661DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08005662EXPORT_SYMBOL(memcg_sockets_enabled_key);
5663
5664void sock_update_memcg(struct sock *sk)
5665{
5666 struct mem_cgroup *memcg;
5667
5668 /* Socket cloning can throw us here with sk_cgrp already
5669 * filled. It won't however, necessarily happen from
5670 * process context. So the test for root memcg given
5671 * the current task's memcg won't help us in this case.
5672 *
5673 * Respecting the original socket's memcg is a better
5674 * decision in this case.
5675 */
5676 if (sk->sk_memcg) {
5677 BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5678 css_get(&sk->sk_memcg->css);
5679 return;
5680 }
5681
5682 rcu_read_lock();
5683 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005684 if (memcg == root_mem_cgroup)
5685 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08005686 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005687 goto out;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005688 if (css_tryget_online(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08005689 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005690out:
Johannes Weiner11092082016-01-14 15:21:26 -08005691 rcu_read_unlock();
5692}
5693EXPORT_SYMBOL(sock_update_memcg);
5694
5695void sock_release_memcg(struct sock *sk)
5696{
5697 WARN_ON(!sk->sk_memcg);
5698 css_put(&sk->sk_memcg->css);
5699}
5700
5701/**
5702 * mem_cgroup_charge_skmem - charge socket memory
5703 * @memcg: memcg to charge
5704 * @nr_pages: number of pages to charge
5705 *
5706 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5707 * @memcg's configured limit, %false if the charge had to be forced.
5708 */
5709bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5710{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005711 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08005712
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005713 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005714 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005715
Johannes Weiner0db15292016-01-20 15:02:50 -08005716 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5717 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005718 return true;
5719 }
Johannes Weiner0db15292016-01-20 15:02:50 -08005720 page_counter_charge(&memcg->tcpmem, nr_pages);
5721 memcg->tcpmem_pressure = 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005722 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08005723 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005724
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005725 /* Don't block in the packet receive path */
5726 if (in_softirq())
5727 gfp_mask = GFP_NOWAIT;
5728
Johannes Weinerb2807f02016-01-20 15:03:22 -08005729 this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5730
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005731 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5732 return true;
5733
5734 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005735 return false;
5736}
5737
5738/**
5739 * mem_cgroup_uncharge_skmem - uncharge socket memory
5740 * @memcg - memcg to uncharge
5741 * @nr_pages - number of pages to uncharge
5742 */
5743void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5744{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005745 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005746 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005747 return;
5748 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005749
Johannes Weinerb2807f02016-01-20 15:03:22 -08005750 this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5751
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005752 page_counter_uncharge(&memcg->memory, nr_pages);
5753 css_put_many(&memcg->css, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005754}
5755
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005756static int __init cgroup_memory(char *s)
5757{
5758 char *token;
5759
5760 while ((token = strsep(&s, ",")) != NULL) {
5761 if (!*token)
5762 continue;
5763 if (!strcmp(token, "nosocket"))
5764 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08005765 if (!strcmp(token, "nokmem"))
5766 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005767 }
5768 return 0;
5769}
5770__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08005771
Michal Hocko2d110852013-02-22 16:34:43 -08005772/*
Michal Hocko10813122013-02-22 16:35:41 -08005773 * subsys_initcall() for memory controller.
5774 *
5775 * Some parts like hotcpu_notifier() have to be initialized from this context
5776 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5777 * everything that doesn't depend on a specific mem_cgroup structure should
5778 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005779 */
5780static int __init mem_cgroup_init(void)
5781{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005782 int cpu, node;
5783
Michal Hocko2d110852013-02-22 16:34:43 -08005784 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005785
5786 for_each_possible_cpu(cpu)
5787 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5788 drain_local_stock);
5789
5790 for_each_node(node) {
5791 struct mem_cgroup_tree_per_node *rtpn;
Johannes Weiner95a045f2015-02-11 15:26:33 -08005792
5793 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5794 node_online(node) ? node : NUMA_NO_NODE);
5795
Mel Gormanef8f2322016-07-28 15:46:05 -07005796 rtpn->rb_root = RB_ROOT;
5797 spin_lock_init(&rtpn->lock);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005798 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5799 }
5800
Michal Hocko2d110852013-02-22 16:34:43 -08005801 return 0;
5802}
5803subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005804
5805#ifdef CONFIG_MEMCG_SWAP
Arnd Bergmann358c07f2016-08-25 15:17:08 -07005806static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
5807{
5808 while (!atomic_inc_not_zero(&memcg->id.ref)) {
5809 /*
5810 * The root cgroup cannot be destroyed, so it's refcount must
5811 * always be >= 1.
5812 */
5813 if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
5814 VM_BUG_ON(1);
5815 break;
5816 }
5817 memcg = parent_mem_cgroup(memcg);
5818 if (!memcg)
5819 memcg = root_mem_cgroup;
5820 }
5821 return memcg;
5822}
5823
Johannes Weiner21afa382015-02-11 15:26:36 -08005824/**
5825 * mem_cgroup_swapout - transfer a memsw charge to swap
5826 * @page: page whose memsw charge to transfer
5827 * @entry: swap entry to move the charge to
5828 *
5829 * Transfer the memsw charge of @page to @entry.
5830 */
5831void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5832{
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005833 struct mem_cgroup *memcg, *swap_memcg;
Johannes Weiner21afa382015-02-11 15:26:36 -08005834 unsigned short oldid;
5835
5836 VM_BUG_ON_PAGE(PageLRU(page), page);
5837 VM_BUG_ON_PAGE(page_count(page), page);
5838
Johannes Weiner7941d212016-01-14 15:21:23 -08005839 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08005840 return;
5841
5842 memcg = page->mem_cgroup;
5843
5844 /* Readahead page, never charged */
5845 if (!memcg)
5846 return;
5847
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005848 /*
5849 * In case the memcg owning these pages has been offlined and doesn't
5850 * have an ID allocated to it anymore, charge the closest online
5851 * ancestor for the swap instead and transfer the memory+swap charge.
5852 */
5853 swap_memcg = mem_cgroup_id_get_online(memcg);
5854 oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg));
Johannes Weiner21afa382015-02-11 15:26:36 -08005855 VM_BUG_ON_PAGE(oldid, page);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005856 mem_cgroup_swap_statistics(swap_memcg, true);
Johannes Weiner21afa382015-02-11 15:26:36 -08005857
5858 page->mem_cgroup = NULL;
5859
5860 if (!mem_cgroup_is_root(memcg))
5861 page_counter_uncharge(&memcg->memory, 1);
5862
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005863 if (memcg != swap_memcg) {
5864 if (!mem_cgroup_is_root(swap_memcg))
5865 page_counter_charge(&swap_memcg->memsw, 1);
5866 page_counter_uncharge(&memcg->memsw, 1);
5867 }
5868
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07005869 /*
5870 * Interrupts should be disabled here because the caller holds the
5871 * mapping->tree_lock lock which is taken with interrupts-off. It is
5872 * important here to have the interrupts disabled because it is the
5873 * only synchronisation we have for udpating the per-CPU variables.
5874 */
5875 VM_BUG_ON(!irqs_disabled());
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005876 mem_cgroup_charge_statistics(memcg, page, false, -1);
Johannes Weiner21afa382015-02-11 15:26:36 -08005877 memcg_check_events(memcg, page);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005878
5879 if (!mem_cgroup_is_root(memcg))
5880 css_put(&memcg->css);
Johannes Weiner21afa382015-02-11 15:26:36 -08005881}
5882
Vladimir Davydov37e84352016-01-20 15:02:56 -08005883/*
5884 * mem_cgroup_try_charge_swap - try charging a swap entry
5885 * @page: page being added to swap
5886 * @entry: swap entry to charge
5887 *
5888 * Try to charge @entry to the memcg that @page belongs to.
5889 *
5890 * Returns 0 on success, -ENOMEM on failure.
5891 */
5892int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5893{
5894 struct mem_cgroup *memcg;
5895 struct page_counter *counter;
5896 unsigned short oldid;
5897
5898 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5899 return 0;
5900
5901 memcg = page->mem_cgroup;
5902
5903 /* Readahead page, never charged */
5904 if (!memcg)
5905 return 0;
5906
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005907 memcg = mem_cgroup_id_get_online(memcg);
Vladimir Davydov37e84352016-01-20 15:02:56 -08005908
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005909 if (!mem_cgroup_is_root(memcg) &&
5910 !page_counter_try_charge(&memcg->swap, 1, &counter)) {
5911 mem_cgroup_id_put(memcg);
5912 return -ENOMEM;
5913 }
5914
Vladimir Davydov37e84352016-01-20 15:02:56 -08005915 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5916 VM_BUG_ON_PAGE(oldid, page);
5917 mem_cgroup_swap_statistics(memcg, true);
5918
Vladimir Davydov37e84352016-01-20 15:02:56 -08005919 return 0;
5920}
5921
Johannes Weiner21afa382015-02-11 15:26:36 -08005922/**
5923 * mem_cgroup_uncharge_swap - uncharge a swap entry
5924 * @entry: swap entry to uncharge
5925 *
Vladimir Davydov37e84352016-01-20 15:02:56 -08005926 * Drop the swap charge associated with @entry.
Johannes Weiner21afa382015-02-11 15:26:36 -08005927 */
5928void mem_cgroup_uncharge_swap(swp_entry_t entry)
5929{
5930 struct mem_cgroup *memcg;
5931 unsigned short id;
5932
Vladimir Davydov37e84352016-01-20 15:02:56 -08005933 if (!do_swap_account)
Johannes Weiner21afa382015-02-11 15:26:36 -08005934 return;
5935
5936 id = swap_cgroup_record(entry, 0);
5937 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005938 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005939 if (memcg) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08005940 if (!mem_cgroup_is_root(memcg)) {
5941 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5942 page_counter_uncharge(&memcg->swap, 1);
5943 else
5944 page_counter_uncharge(&memcg->memsw, 1);
5945 }
Johannes Weiner21afa382015-02-11 15:26:36 -08005946 mem_cgroup_swap_statistics(memcg, false);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005947 mem_cgroup_id_put(memcg);
Johannes Weiner21afa382015-02-11 15:26:36 -08005948 }
5949 rcu_read_unlock();
5950}
5951
Vladimir Davydovd8b38432016-01-20 15:03:07 -08005952long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
5953{
5954 long nr_swap_pages = get_nr_swap_pages();
5955
5956 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5957 return nr_swap_pages;
5958 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5959 nr_swap_pages = min_t(long, nr_swap_pages,
5960 READ_ONCE(memcg->swap.limit) -
5961 page_counter_read(&memcg->swap));
5962 return nr_swap_pages;
5963}
5964
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08005965bool mem_cgroup_swap_full(struct page *page)
5966{
5967 struct mem_cgroup *memcg;
5968
5969 VM_BUG_ON_PAGE(!PageLocked(page), page);
5970
5971 if (vm_swap_full())
5972 return true;
5973 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5974 return false;
5975
5976 memcg = page->mem_cgroup;
5977 if (!memcg)
5978 return false;
5979
5980 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5981 if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
5982 return true;
5983
5984 return false;
5985}
5986
Johannes Weiner21afa382015-02-11 15:26:36 -08005987/* for remember boot option*/
5988#ifdef CONFIG_MEMCG_SWAP_ENABLED
5989static int really_do_swap_account __initdata = 1;
5990#else
5991static int really_do_swap_account __initdata;
5992#endif
5993
5994static int __init enable_swap_account(char *s)
5995{
5996 if (!strcmp(s, "1"))
5997 really_do_swap_account = 1;
5998 else if (!strcmp(s, "0"))
5999 really_do_swap_account = 0;
6000 return 1;
6001}
6002__setup("swapaccount=", enable_swap_account);
6003
Vladimir Davydov37e84352016-01-20 15:02:56 -08006004static u64 swap_current_read(struct cgroup_subsys_state *css,
6005 struct cftype *cft)
6006{
6007 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6008
6009 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
6010}
6011
6012static int swap_max_show(struct seq_file *m, void *v)
6013{
6014 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
6015 unsigned long max = READ_ONCE(memcg->swap.limit);
6016
6017 if (max == PAGE_COUNTER_MAX)
6018 seq_puts(m, "max\n");
6019 else
6020 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
6021
6022 return 0;
6023}
6024
6025static ssize_t swap_max_write(struct kernfs_open_file *of,
6026 char *buf, size_t nbytes, loff_t off)
6027{
6028 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6029 unsigned long max;
6030 int err;
6031
6032 buf = strstrip(buf);
6033 err = page_counter_memparse(buf, "max", &max);
6034 if (err)
6035 return err;
6036
6037 mutex_lock(&memcg_limit_mutex);
6038 err = page_counter_limit(&memcg->swap, max);
6039 mutex_unlock(&memcg_limit_mutex);
6040 if (err)
6041 return err;
6042
6043 return nbytes;
6044}
6045
6046static struct cftype swap_files[] = {
6047 {
6048 .name = "swap.current",
6049 .flags = CFTYPE_NOT_ON_ROOT,
6050 .read_u64 = swap_current_read,
6051 },
6052 {
6053 .name = "swap.max",
6054 .flags = CFTYPE_NOT_ON_ROOT,
6055 .seq_show = swap_max_show,
6056 .write = swap_max_write,
6057 },
6058 { } /* terminate */
6059};
6060
Johannes Weiner21afa382015-02-11 15:26:36 -08006061static struct cftype memsw_cgroup_files[] = {
6062 {
6063 .name = "memsw.usage_in_bytes",
6064 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6065 .read_u64 = mem_cgroup_read_u64,
6066 },
6067 {
6068 .name = "memsw.max_usage_in_bytes",
6069 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6070 .write = mem_cgroup_reset,
6071 .read_u64 = mem_cgroup_read_u64,
6072 },
6073 {
6074 .name = "memsw.limit_in_bytes",
6075 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6076 .write = mem_cgroup_write,
6077 .read_u64 = mem_cgroup_read_u64,
6078 },
6079 {
6080 .name = "memsw.failcnt",
6081 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6082 .write = mem_cgroup_reset,
6083 .read_u64 = mem_cgroup_read_u64,
6084 },
6085 { }, /* terminate */
6086};
6087
6088static int __init mem_cgroup_swap_init(void)
6089{
6090 if (!mem_cgroup_disabled() && really_do_swap_account) {
6091 do_swap_account = 1;
Vladimir Davydov37e84352016-01-20 15:02:56 -08006092 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
6093 swap_files));
Johannes Weiner21afa382015-02-11 15:26:36 -08006094 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
6095 memsw_cgroup_files));
6096 }
6097 return 0;
6098}
6099subsys_initcall(mem_cgroup_swap_init);
6100
6101#endif /* CONFIG_MEMCG_SWAP */