blob: 86a6b331b96488d05c4bd1b3fa245e0a25a4c1b3 [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);
Laurent Dufoura495f722017-03-09 16:17:06 -0800465 if (!mctz)
466 return;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700467 /*
468 * Necessary to update all ancestors when hierarchy is used.
469 * because their event counter is not touched.
470 */
471 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700472 mz = mem_cgroup_page_nodeinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800473 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700474 /*
475 * We have to update the tree if mz is on RB-tree or
476 * mem is over its softlimit.
477 */
478 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700479 unsigned long flags;
480
481 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700482 /* if on-tree, remove it */
483 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700484 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700485 /*
486 * Insert again. mz->usage_in_excess will be updated.
487 * If excess is 0, no tree ops.
488 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700489 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700490 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700491 }
492 }
493}
494
495static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
496{
Mel Gormanef8f2322016-07-28 15:46:05 -0700497 struct mem_cgroup_tree_per_node *mctz;
498 struct mem_cgroup_per_node *mz;
499 int nid;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700500
Jianyu Zhane2318752014-06-06 14:38:20 -0700501 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700502 mz = mem_cgroup_nodeinfo(memcg, nid);
503 mctz = soft_limit_tree_node(nid);
Laurent Dufoura495f722017-03-09 16:17:06 -0800504 if (mctz)
505 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700506 }
507}
508
Mel Gormanef8f2322016-07-28 15:46:05 -0700509static struct mem_cgroup_per_node *
510__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700511{
512 struct rb_node *rightmost = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700513 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700514
515retry:
516 mz = NULL;
517 rightmost = rb_last(&mctz->rb_root);
518 if (!rightmost)
519 goto done; /* Nothing to reclaim from */
520
Mel Gormanef8f2322016-07-28 15:46:05 -0700521 mz = rb_entry(rightmost, struct mem_cgroup_per_node, tree_node);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700522 /*
523 * Remove the node now but someone else can add it back,
524 * we will to add it back at the end of reclaim to its correct
525 * position in the tree.
526 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700527 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800528 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400529 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700530 goto retry;
531done:
532 return mz;
533}
534
Mel Gormanef8f2322016-07-28 15:46:05 -0700535static struct mem_cgroup_per_node *
536mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700537{
Mel Gormanef8f2322016-07-28 15:46:05 -0700538 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700539
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700540 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700541 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700542 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700543 return mz;
544}
545
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700546/*
Greg Thelen484ebb32015-10-01 15:37:05 -0700547 * Return page count for single (non recursive) @memcg.
548 *
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700549 * Implementation Note: reading percpu statistics for memcg.
550 *
551 * Both of vmstat[] and percpu_counter has threshold and do periodic
552 * synchronization to implement "quick" read. There are trade-off between
553 * reading cost and precision of value. Then, we may have a chance to implement
Greg Thelen484ebb32015-10-01 15:37:05 -0700554 * a periodic synchronization of counter in memcg's counter.
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700555 *
556 * But this _read() function is used for user interface now. The user accounts
557 * memory usage by memory cgroup and he _always_ requires exact value because
558 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
559 * have to visit all online cpus and make sum. So, for now, unnecessary
560 * synchronization is not implemented. (just implemented for cpu hotplug)
561 *
562 * If there are kernel internal actions which can make use of some not-exact
563 * value, and reading all cpu value can be performance bottleneck in some
Greg Thelen484ebb32015-10-01 15:37:05 -0700564 * common workload, threshold and synchronization as vmstat[] should be
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700565 * implemented.
566 */
Greg Thelen484ebb32015-10-01 15:37:05 -0700567static unsigned long
568mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800569{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700570 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800571 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800572
Greg Thelen484ebb32015-10-01 15:37:05 -0700573 /* Per-cpu values can be negative, use a signed accumulator */
Tejun Heo733a5722015-05-22 18:23:18 -0400574 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700575 val += per_cpu(memcg->stat->count[idx], cpu);
Greg Thelen484ebb32015-10-01 15:37:05 -0700576 /*
577 * Summing races with updates, so val may be negative. Avoid exposing
578 * transient negative values.
579 */
580 if (val < 0)
581 val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800582 return val;
583}
584
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700585static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700586 enum mem_cgroup_events_index idx)
587{
588 unsigned long val = 0;
589 int cpu;
590
Tejun Heo733a5722015-05-22 18:23:18 -0400591 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700592 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700593 return val;
594}
595
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700596static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700597 struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800598 bool compound, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800599{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700600 /*
601 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
602 * counted as CACHE even if it's on ANON LRU.
603 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700604 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700605 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700606 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800607 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700608 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700609 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700610
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800611 if (compound) {
612 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
David Rientjesb070e652013-05-07 16:18:09 -0700613 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
614 nr_pages);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800615 }
David Rientjesb070e652013-05-07 16:18:09 -0700616
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800617 /* pagein of a big page is an event. So, ignore page size */
618 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700619 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800620 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700621 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800622 nr_pages = -nr_pages; /* for event */
623 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800624
Johannes Weiner13114712012-05-29 15:07:07 -0700625 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800626}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800627
Vladimir Davydov0a6b76d2016-03-17 14:18:42 -0700628unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
629 int nid, unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700630{
Michal Hocko07fc9572017-01-10 16:58:04 -0800631 struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg);
Jianyu Zhane2318752014-06-06 14:38:20 -0700632 unsigned long nr = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -0700633 enum lru_list lru;
Ying Han889976d2011-05-26 16:25:33 -0700634
Jianyu Zhane2318752014-06-06 14:38:20 -0700635 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700636
Mel Gormanef8f2322016-07-28 15:46:05 -0700637 for_each_lru(lru) {
638 if (!(BIT(lru) & lru_mask))
639 continue;
Michal Hocko07fc9572017-01-10 16:58:04 -0800640 nr += mem_cgroup_get_lru_size(lruvec, lru);
Jianyu Zhane2318752014-06-06 14:38:20 -0700641 }
642 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700643}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700644
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700645static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700646 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800647{
Jianyu Zhane2318752014-06-06 14:38:20 -0700648 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700649 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800650
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800651 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700652 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
653 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800654}
655
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800656static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
657 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800658{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700659 unsigned long val, next;
660
Johannes Weiner13114712012-05-29 15:07:07 -0700661 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700662 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700663 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800664 if ((long)next - (long)val < 0) {
665 switch (target) {
666 case MEM_CGROUP_TARGET_THRESH:
667 next = val + THRESHOLDS_EVENTS_TARGET;
668 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700669 case MEM_CGROUP_TARGET_SOFTLIMIT:
670 next = val + SOFTLIMIT_EVENTS_TARGET;
671 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800672 case MEM_CGROUP_TARGET_NUMAINFO:
673 next = val + NUMAINFO_EVENTS_TARGET;
674 break;
675 default:
676 break;
677 }
678 __this_cpu_write(memcg->stat->targets[target], next);
679 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700680 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800681 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800682}
683
684/*
685 * Check events in order.
686 *
687 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700688static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800689{
690 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800691 if (unlikely(mem_cgroup_event_ratelimit(memcg,
692 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700693 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800694 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800695
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700696 do_softlimit = mem_cgroup_event_ratelimit(memcg,
697 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700698#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800699 do_numainfo = mem_cgroup_event_ratelimit(memcg,
700 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700701#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800702 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700703 if (unlikely(do_softlimit))
704 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800705#if MAX_NUMNODES > 1
706 if (unlikely(do_numainfo))
707 atomic_inc(&memcg->numainfo_events);
708#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700709 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800710}
711
Balbir Singhcf475ad2008-04-29 01:00:16 -0700712struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800713{
Balbir Singh31a78f22008-09-28 23:09:31 +0100714 /*
715 * mm_update_next_owner() may clear mm->owner to NULL
716 * if it races with swapoff, page migration, etc.
717 * So this can be called with p == NULL.
718 */
719 if (unlikely(!p))
720 return NULL;
721
Tejun Heo073219e2014-02-08 10:36:58 -0500722 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800723}
Michal Hocko33398cf2015-09-08 15:01:02 -0700724EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800725
Johannes Weinerdf381972014-04-07 15:37:43 -0700726static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800727{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700728 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700729
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800730 rcu_read_lock();
731 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700732 /*
733 * Page cache insertions can happen withou an
734 * actual mm context, e.g. during disk probing
735 * on boot, loopback IO, acct() writes etc.
736 */
737 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700738 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700739 else {
740 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
741 if (unlikely(!memcg))
742 memcg = root_mem_cgroup;
743 }
Tejun Heoec903c02014-05-13 12:11:01 -0400744 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800745 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700746 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800747}
748
Johannes Weiner56600482012-01-12 17:17:59 -0800749/**
750 * mem_cgroup_iter - iterate over memory cgroup hierarchy
751 * @root: hierarchy root
752 * @prev: previously returned memcg, NULL on first invocation
753 * @reclaim: cookie for shared reclaim walks, NULL for full walks
754 *
755 * Returns references to children of the hierarchy below @root, or
756 * @root itself, or %NULL after a full round-trip.
757 *
758 * Caller must pass the return value in @prev on subsequent
759 * invocations for reference counting, or use mem_cgroup_iter_break()
760 * to cancel a hierarchy walk before the round-trip is complete.
761 *
762 * Reclaimers can specify a zone and a priority level in @reclaim to
763 * divide up the memcgs in the hierarchy among all concurrent
764 * reclaimers operating on the same zone and priority.
765 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700766struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800767 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700768 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700769{
Michal Hocko33398cf2015-09-08 15:01:02 -0700770 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800771 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800772 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800773 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700774
Andrew Morton694fbc02013-09-24 15:27:37 -0700775 if (mem_cgroup_disabled())
776 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800777
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700778 if (!root)
779 root = root_mem_cgroup;
780
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800781 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800782 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800783
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800784 if (!root->use_hierarchy && root != root_mem_cgroup) {
785 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800786 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700787 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800788 }
789
Michal Hocko542f85f2013-04-29 15:07:15 -0700790 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800791
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800792 if (reclaim) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700793 struct mem_cgroup_per_node *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800794
Mel Gormanef8f2322016-07-28 15:46:05 -0700795 mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800796 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700797
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800798 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700799 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800800
Vladimir Davydov6df38682015-12-29 14:54:10 -0800801 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -0700802 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800803 if (!pos || css_tryget(&pos->css))
804 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800805 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800806 * css reference reached zero, so iter->position will
807 * be cleared by ->css_released. However, we should not
808 * rely on this happening soon, because ->css_released
809 * is called from a work queue, and by busy-waiting we
810 * might block it. So we clear iter->position right
811 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800812 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800813 (void)cmpxchg(&iter->position, pos, NULL);
814 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800815 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800816
817 if (pos)
818 css = &pos->css;
819
820 for (;;) {
821 css = css_next_descendant_pre(css, &root->css);
822 if (!css) {
823 /*
824 * Reclaimers share the hierarchy walk, and a
825 * new one might jump in right at the end of
826 * the hierarchy - make sure they see at least
827 * one group and restart from the beginning.
828 */
829 if (!prev)
830 continue;
831 break;
832 }
833
834 /*
835 * Verify the css and acquire a reference. The root
836 * is provided by the caller, so we know it's alive
837 * and kicking, and don't take an extra reference.
838 */
839 memcg = mem_cgroup_from_css(css);
840
841 if (css == &root->css)
842 break;
843
Johannes Weiner0b8f73e2016-01-20 15:02:53 -0800844 if (css_tryget(css))
845 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800846
847 memcg = NULL;
848 }
849
850 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800851 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800852 * The position could have already been updated by a competing
853 * thread, so check that the value hasn't changed since we read
854 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800855 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800856 (void)cmpxchg(&iter->position, pos, memcg);
857
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800858 if (pos)
859 css_put(&pos->css);
860
861 if (!memcg)
862 iter->generation++;
863 else if (!prev)
864 reclaim->generation = iter->generation;
865 }
866
Michal Hocko542f85f2013-04-29 15:07:15 -0700867out_unlock:
868 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800869out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700870 if (prev && prev != root)
871 css_put(&prev->css);
872
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800873 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700874}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800875
Johannes Weiner56600482012-01-12 17:17:59 -0800876/**
877 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
878 * @root: hierarchy root
879 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
880 */
881void mem_cgroup_iter_break(struct mem_cgroup *root,
882 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800883{
884 if (!root)
885 root = root_mem_cgroup;
886 if (prev && prev != root)
887 css_put(&prev->css);
888}
889
Vladimir Davydov6df38682015-12-29 14:54:10 -0800890static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
891{
892 struct mem_cgroup *memcg = dead_memcg;
893 struct mem_cgroup_reclaim_iter *iter;
Mel Gormanef8f2322016-07-28 15:46:05 -0700894 struct mem_cgroup_per_node *mz;
895 int nid;
Vladimir Davydov6df38682015-12-29 14:54:10 -0800896 int i;
897
Jing Xiaf46b0542018-07-20 17:53:48 -0700898 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Vladimir Davydov6df38682015-12-29 14:54:10 -0800899 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700900 mz = mem_cgroup_nodeinfo(memcg, nid);
901 for (i = 0; i <= DEF_PRIORITY; i++) {
902 iter = &mz->iter[i];
903 cmpxchg(&iter->position,
904 dead_memcg, NULL);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800905 }
906 }
907 }
908}
909
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700910/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800911 * Iteration constructs for visiting all cgroups (under a tree). If
912 * loops are exited prematurely (break), mem_cgroup_iter_break() must
913 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700914 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800915#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800916 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800917 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800918 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700919
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800920#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800921 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800922 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800923 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700924
Johannes Weiner925b7672012-01-12 17:18:15 -0800925/**
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700926 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
927 * @memcg: hierarchy root
928 * @fn: function to call for each task
929 * @arg: argument passed to @fn
930 *
931 * This function iterates over tasks attached to @memcg or to any of its
932 * descendants and calls @fn for each task. If @fn returns a non-zero
933 * value, the function breaks the iteration loop and returns the value.
934 * Otherwise, it will iterate over all tasks and return 0.
935 *
936 * This function must not be called for the root memory cgroup.
937 */
938int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
939 int (*fn)(struct task_struct *, void *), void *arg)
940{
941 struct mem_cgroup *iter;
942 int ret = 0;
943
944 BUG_ON(memcg == root_mem_cgroup);
945
946 for_each_mem_cgroup_tree(iter, memcg) {
947 struct css_task_iter it;
948 struct task_struct *task;
949
950 css_task_iter_start(&iter->css, &it);
951 while (!ret && (task = css_task_iter_next(&it)))
952 ret = fn(task, arg);
953 css_task_iter_end(&it);
954 if (ret) {
955 mem_cgroup_iter_break(memcg, iter);
956 break;
957 }
958 }
959 return ret;
960}
961
962/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800963 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -0800964 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700965 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800966 *
967 * This function is only safe when following the LRU page isolation
968 * and putback protocol: the LRU lock must be held, and the page must
969 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -0700970 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700971struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat)
Minchan Kim3f58a822011-03-22 16:32:53 -0700972{
Mel Gormanef8f2322016-07-28 15:46:05 -0700973 struct mem_cgroup_per_node *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -0800974 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800975 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800976
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800977 if (mem_cgroup_disabled()) {
Mel Gorman599d0c92016-07-28 15:45:31 -0700978 lruvec = &pgdat->lruvec;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800979 goto out;
980 }
Christoph Lameterb69408e2008-10-18 20:26:14 -0700981
Johannes Weiner1306a852014-12-10 15:44:52 -0800982 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800983 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800984 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -0800985 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -0800986 */
Johannes Weiner29833312014-12-10 15:44:02 -0800987 if (!memcg)
988 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800989
Mel Gormanef8f2322016-07-28 15:46:05 -0700990 mz = mem_cgroup_page_nodeinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800991 lruvec = &mz->lruvec;
992out:
993 /*
994 * Since a node can be onlined after the mem_cgroup was created,
995 * we have to be prepared to initialize lruvec->zone here;
996 * and if offlined then reonlined, we need to reinitialize it.
997 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700998 if (unlikely(lruvec->pgdat != pgdat))
999 lruvec->pgdat = pgdat;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001000 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001001}
1002
1003/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001004 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1005 * @lruvec: mem_cgroup per zone lru vector
1006 * @lru: index of lru list the page is sitting on
Michal Hocko07fc9572017-01-10 16:58:04 -08001007 * @zid: zone id of the accounted pages
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001008 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001009 *
Hugh Dickinsca707232016-05-19 17:12:35 -07001010 * This function must be called under lru_lock, just before a page is added
1011 * to or just after a page is removed from an lru list (that ordering being
1012 * so as to allow it to check that lru_size 0 is consistent with list_empty).
Johannes Weiner925b7672012-01-12 17:18:15 -08001013 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001014void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
Michal Hocko07fc9572017-01-10 16:58:04 -08001015 int zid, int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001016{
Mel Gormanef8f2322016-07-28 15:46:05 -07001017 struct mem_cgroup_per_node *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001018 unsigned long *lru_size;
Hugh Dickinsca707232016-05-19 17:12:35 -07001019 long size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001020
1021 if (mem_cgroup_disabled())
1022 return;
1023
Mel Gormanef8f2322016-07-28 15:46:05 -07001024 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
Michal Hocko07fc9572017-01-10 16:58:04 -08001025 lru_size = &mz->lru_zone_size[zid][lru];
Hugh Dickinsca707232016-05-19 17:12:35 -07001026
1027 if (nr_pages < 0)
1028 *lru_size += nr_pages;
1029
1030 size = *lru_size;
Michal Hocko07fc9572017-01-10 16:58:04 -08001031 if (WARN_ONCE(size < 0,
1032 "%s(%p, %d, %d): lru_size %ld\n",
1033 __func__, lruvec, lru, nr_pages, size)) {
Hugh Dickinsca707232016-05-19 17:12:35 -07001034 VM_BUG_ON(1);
1035 *lru_size = 0;
1036 }
1037
1038 if (nr_pages > 0)
1039 *lru_size += nr_pages;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001040}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001041
Johannes Weiner2314b422014-12-10 15:44:33 -08001042bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001043{
Johannes Weiner2314b422014-12-10 15:44:33 -08001044 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001045 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001046 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001047
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001048 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001049 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001050 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001051 task_unlock(p);
1052 } else {
1053 /*
1054 * All threads may have already detached their mm's, but the oom
1055 * killer still needs to detect if they have already been oom
1056 * killed to prevent needlessly killing additional tasks.
1057 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001058 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001059 task_memcg = mem_cgroup_from_task(task);
1060 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001061 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001062 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001063 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1064 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001065 return ret;
1066}
1067
Johannes Weiner19942822011-02-01 15:52:43 -08001068/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001069 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001070 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001071 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001072 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001073 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001074 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001075static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001076{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001077 unsigned long margin = 0;
1078 unsigned long count;
1079 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001080
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001081 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001082 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001083 if (count < limit)
1084 margin = limit - count;
1085
Johannes Weiner7941d212016-01-14 15:21:23 -08001086 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001087 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001088 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001089 if (count <= limit)
1090 margin = min(margin, limit - count);
Li RongQingcbedbac2016-05-27 14:27:43 -07001091 else
1092 margin = 0;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001093 }
1094
1095 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001096}
1097
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001098/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001099 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001100 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001101 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1102 * moving cgroups. This is for waiting at high-memory pressure
1103 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001104 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001105static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001106{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001107 struct mem_cgroup *from;
1108 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001109 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001110 /*
1111 * Unlike task_move routines, we access mc.to, mc.from not under
1112 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1113 */
1114 spin_lock(&mc.lock);
1115 from = mc.from;
1116 to = mc.to;
1117 if (!from)
1118 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001119
Johannes Weiner2314b422014-12-10 15:44:33 -08001120 ret = mem_cgroup_is_descendant(from, memcg) ||
1121 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001122unlock:
1123 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001124 return ret;
1125}
1126
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001127static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001128{
1129 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001130 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001131 DEFINE_WAIT(wait);
1132 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1133 /* moving charge context might have finished. */
1134 if (mc.moving_task)
1135 schedule();
1136 finish_wait(&mc.waitq, &wait);
1137 return true;
1138 }
1139 }
1140 return false;
1141}
1142
Sha Zhengju58cf1882013-02-22 16:32:05 -08001143#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001144/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001145 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001146 * @memcg: The memory cgroup that went over limit
1147 * @p: Task that is going to be killed
1148 *
1149 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1150 * enabled
1151 */
1152void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1153{
Sha Zhengju58cf1882013-02-22 16:32:05 -08001154 struct mem_cgroup *iter;
1155 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001156
Balbir Singhe2224322009-04-02 16:57:39 -07001157 rcu_read_lock();
1158
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001159 if (p) {
1160 pr_info("Task in ");
1161 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1162 pr_cont(" killed as a result of limit of ");
1163 } else {
1164 pr_info("Memory limit reached of cgroup ");
1165 }
1166
Tejun Heoe61734c2014-02-12 09:29:50 -05001167 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001168 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001169
Balbir Singhe2224322009-04-02 16:57:39 -07001170 rcu_read_unlock();
1171
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001172 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1173 K((u64)page_counter_read(&memcg->memory)),
1174 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1175 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1176 K((u64)page_counter_read(&memcg->memsw)),
1177 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1178 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1179 K((u64)page_counter_read(&memcg->kmem)),
1180 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001181
1182 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001183 pr_info("Memory cgroup stats for ");
1184 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001185 pr_cont(":");
1186
1187 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08001188 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Sha Zhengju58cf1882013-02-22 16:32:05 -08001189 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07001190 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
Sha Zhengju58cf1882013-02-22 16:32:05 -08001191 K(mem_cgroup_read_stat(iter, i)));
1192 }
1193
1194 for (i = 0; i < NR_LRU_LISTS; i++)
1195 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1196 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1197
1198 pr_cont("\n");
1199 }
Balbir Singhe2224322009-04-02 16:57:39 -07001200}
1201
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001202/*
1203 * This function returns the number of memcg under hierarchy tree. Returns
1204 * 1(self count) if no children.
1205 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001206static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001207{
1208 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001209 struct mem_cgroup *iter;
1210
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001211 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001212 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001213 return num;
1214}
1215
Balbir Singh6d61ef42009-01-07 18:08:06 -08001216/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001217 * Return the memory (and swap, if configured) limit for a memcg.
1218 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001219unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001220{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001221 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001222
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001223 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001224 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001225 unsigned long memsw_limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001226 unsigned long swap_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001227
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001228 memsw_limit = memcg->memsw.limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001229 swap_limit = memcg->swap.limit;
1230 swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1231 limit = min(limit + swap_limit, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001232 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001233 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001234}
1235
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001236static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001237 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001238{
David Rientjes6e0fc462015-09-08 15:00:36 -07001239 struct oom_control oc = {
1240 .zonelist = NULL,
1241 .nodemask = NULL,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001242 .memcg = memcg,
David Rientjes6e0fc462015-09-08 15:00:36 -07001243 .gfp_mask = gfp_mask,
1244 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001245 };
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001246 bool ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001247
Johannes Weinerdc564012015-06-24 16:57:19 -07001248 mutex_lock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001249 ret = out_of_memory(&oc);
Johannes Weinerdc564012015-06-24 16:57:19 -07001250 mutex_unlock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001251 return ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001252}
1253
Michele Curtiae6e71d2014-12-12 16:56:35 -08001254#if MAX_NUMNODES > 1
1255
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001256/**
1257 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001258 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001259 * @nid: the node ID to be checked.
1260 * @noswap : specify true here if the user wants flle only information.
1261 *
1262 * This function returns whether the specified memcg contains any
1263 * reclaimable pages on a node. Returns true if there are any reclaimable
1264 * pages in the node.
1265 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001266static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001267 int nid, bool noswap)
1268{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001269 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001270 return true;
1271 if (noswap || !total_swap_pages)
1272 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001273 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001274 return true;
1275 return false;
1276
1277}
Ying Han889976d2011-05-26 16:25:33 -07001278
1279/*
1280 * Always updating the nodemask is not very good - even if we have an empty
1281 * list or the wrong list here, we can start from some node and traverse all
1282 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1283 *
1284 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001285static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001286{
1287 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001288 /*
1289 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1290 * pagein/pageout changes since the last update.
1291 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001292 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001293 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001294 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001295 return;
1296
Ying Han889976d2011-05-26 16:25:33 -07001297 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001298 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001299
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001300 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001301
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001302 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1303 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001304 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001305
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001306 atomic_set(&memcg->numainfo_events, 0);
1307 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001308}
1309
1310/*
1311 * Selecting a node where we start reclaim from. Because what we need is just
1312 * reducing usage counter, start from anywhere is O,K. Considering
1313 * memory reclaim from current node, there are pros. and cons.
1314 *
1315 * Freeing memory from current node means freeing memory from a node which
1316 * we'll use or we've used. So, it may make LRU bad. And if several threads
1317 * hit limits, it will see a contention on a node. But freeing from remote
1318 * node means more costs for memory reclaim because of memory latency.
1319 *
1320 * Now, we use round-robin. Better algorithm is welcomed.
1321 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001322int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001323{
1324 int node;
1325
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001326 mem_cgroup_may_update_nodemask(memcg);
1327 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001328
Andrew Morton0edaf862016-05-19 17:10:58 -07001329 node = next_node_in(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001330 /*
Michal Hockofda3d692016-05-19 17:11:34 -07001331 * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1332 * last time it really checked all the LRUs due to rate limiting.
1333 * Fallback to the current node in that case for simplicity.
Ying Han889976d2011-05-26 16:25:33 -07001334 */
1335 if (unlikely(node == MAX_NUMNODES))
1336 node = numa_node_id();
1337
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001338 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001339 return node;
1340}
Ying Han889976d2011-05-26 16:25:33 -07001341#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001342int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001343{
1344 return 0;
1345}
1346#endif
1347
Andrew Morton0608f432013-09-24 15:27:41 -07001348static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
Mel Gormanef8f2322016-07-28 15:46:05 -07001349 pg_data_t *pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001350 gfp_t gfp_mask,
1351 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001352{
Andrew Morton0608f432013-09-24 15:27:41 -07001353 struct mem_cgroup *victim = NULL;
1354 int total = 0;
1355 int loop = 0;
1356 unsigned long excess;
1357 unsigned long nr_scanned;
1358 struct mem_cgroup_reclaim_cookie reclaim = {
Mel Gormanef8f2322016-07-28 15:46:05 -07001359 .pgdat = pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001360 .priority = 0,
1361 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001362
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001363 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001364
Andrew Morton0608f432013-09-24 15:27:41 -07001365 while (1) {
1366 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1367 if (!victim) {
1368 loop++;
1369 if (loop >= 2) {
1370 /*
1371 * If we have not been able to reclaim
1372 * anything, it might because there are
1373 * no reclaimable pages under this hierarchy
1374 */
1375 if (!total)
1376 break;
1377 /*
1378 * We want to do more targeted reclaim.
1379 * excess >> 2 is not to excessive so as to
1380 * reclaim too much, nor too less that we keep
1381 * coming back to reclaim from this cgroup
1382 */
1383 if (total >= (excess >> 2) ||
1384 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1385 break;
1386 }
1387 continue;
1388 }
Mel Gormana9dd0a82016-07-28 15:46:02 -07001389 total += mem_cgroup_shrink_node(victim, gfp_mask, false,
Mel Gormanef8f2322016-07-28 15:46:05 -07001390 pgdat, &nr_scanned);
Andrew Morton0608f432013-09-24 15:27:41 -07001391 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001392 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001393 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001394 }
Andrew Morton0608f432013-09-24 15:27:41 -07001395 mem_cgroup_iter_break(root_memcg, victim);
1396 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001397}
1398
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001399#ifdef CONFIG_LOCKDEP
1400static struct lockdep_map memcg_oom_lock_dep_map = {
1401 .name = "memcg_oom_lock",
1402};
1403#endif
1404
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001405static DEFINE_SPINLOCK(memcg_oom_lock);
1406
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001407/*
1408 * Check OOM-Killer is already running under our hierarchy.
1409 * If someone is running, return false.
1410 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001411static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001412{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001413 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001414
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001415 spin_lock(&memcg_oom_lock);
1416
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001417 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001418 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001419 /*
1420 * this subtree of our hierarchy is already locked
1421 * so we cannot give a lock.
1422 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001423 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001424 mem_cgroup_iter_break(memcg, iter);
1425 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001426 } else
1427 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001428 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001429
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001430 if (failed) {
1431 /*
1432 * OK, we failed to lock the whole subtree so we have
1433 * to clean up what we set up to the failing subtree
1434 */
1435 for_each_mem_cgroup_tree(iter, memcg) {
1436 if (iter == failed) {
1437 mem_cgroup_iter_break(memcg, iter);
1438 break;
1439 }
1440 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001441 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001442 } else
1443 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001444
1445 spin_unlock(&memcg_oom_lock);
1446
1447 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001448}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001449
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001450static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001451{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001452 struct mem_cgroup *iter;
1453
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001454 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001455 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001456 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001457 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001458 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001459}
1460
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001461static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001462{
1463 struct mem_cgroup *iter;
1464
Tejun Heoc2b42d32015-06-24 16:58:23 -07001465 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001466 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001467 iter->under_oom++;
1468 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001469}
1470
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001471static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001472{
1473 struct mem_cgroup *iter;
1474
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001475 /*
1476 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001477 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001478 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001479 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001480 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001481 if (iter->under_oom > 0)
1482 iter->under_oom--;
1483 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001484}
1485
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001486static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1487
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001488struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001489 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001490 wait_queue_t wait;
1491};
1492
1493static int memcg_oom_wake_function(wait_queue_t *wait,
1494 unsigned mode, int sync, void *arg)
1495{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001496 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1497 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001498 struct oom_wait_info *oom_wait_info;
1499
1500 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001501 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001502
Johannes Weiner2314b422014-12-10 15:44:33 -08001503 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1504 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001505 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001506 return autoremove_wake_function(wait, mode, sync, arg);
1507}
1508
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001509static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001510{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001511 /*
1512 * For the following lockless ->under_oom test, the only required
1513 * guarantee is that it must see the state asserted by an OOM when
1514 * this function is called as a result of userland actions
1515 * triggered by the notification of the OOM. This is trivially
1516 * achieved by invoking mem_cgroup_mark_under_oom() before
1517 * triggering notification.
1518 */
1519 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001520 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001521}
1522
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001523static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001524{
Andrew Mortond0db7af2016-06-08 15:33:47 -07001525 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001526 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001527 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001528 * We are in the middle of the charge context here, so we
1529 * don't want to block when potentially sitting on a callstack
1530 * that holds all kinds of filesystem and mm locks.
1531 *
1532 * Also, the caller may handle a failed allocation gracefully
1533 * (like optional page cache readahead) and so an OOM killer
1534 * invocation might not even be necessary.
1535 *
1536 * That's why we don't do anything here except remember the
1537 * OOM context and then deal with it at the end of the page
1538 * fault when the stack is unwound, the locks are released,
1539 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001540 */
Johannes Weiner49426422013-10-16 13:46:59 -07001541 css_get(&memcg->css);
Tejun Heo626ebc42015-11-05 18:46:09 -08001542 current->memcg_in_oom = memcg;
1543 current->memcg_oom_gfp_mask = mask;
1544 current->memcg_oom_order = order;
Johannes Weiner49426422013-10-16 13:46:59 -07001545}
1546
1547/**
1548 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1549 * @handle: actually kill/wait or just clean up the OOM state
1550 *
1551 * This has to be called at the end of a page fault if the memcg OOM
1552 * handler was enabled.
1553 *
1554 * Memcg supports userspace OOM handling where failed allocations must
1555 * sleep on a waitqueue until the userspace task resolves the
1556 * situation. Sleeping directly in the charge context with all kinds
1557 * of locks held is not a good idea, instead we remember an OOM state
1558 * in the task and mem_cgroup_oom_synchronize() has to be called at
1559 * the end of the page fault to complete the OOM handling.
1560 *
1561 * Returns %true if an ongoing memcg OOM situation was detected and
1562 * completed, %false otherwise.
1563 */
1564bool mem_cgroup_oom_synchronize(bool handle)
1565{
Tejun Heo626ebc42015-11-05 18:46:09 -08001566 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001567 struct oom_wait_info owait;
1568 bool locked;
1569
1570 /* OOM is global, do not handle */
1571 if (!memcg)
1572 return false;
1573
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001574 if (!handle)
Johannes Weiner49426422013-10-16 13:46:59 -07001575 goto cleanup;
1576
1577 owait.memcg = memcg;
1578 owait.wait.flags = 0;
1579 owait.wait.func = memcg_oom_wake_function;
1580 owait.wait.private = current;
1581 INIT_LIST_HEAD(&owait.wait.task_list);
1582
1583 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001584 mem_cgroup_mark_under_oom(memcg);
1585
1586 locked = mem_cgroup_oom_trylock(memcg);
1587
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001588 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001589 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001590
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001591 if (locked && !memcg->oom_kill_disable) {
1592 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001593 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001594 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1595 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001596 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001597 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001598 mem_cgroup_unmark_under_oom(memcg);
1599 finish_wait(&memcg_oom_waitq, &owait.wait);
1600 }
1601
1602 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001603 mem_cgroup_oom_unlock(memcg);
1604 /*
1605 * There is no guarantee that an OOM-lock contender
1606 * sees the wakeups triggered by the OOM kill
1607 * uncharges. Wake any sleepers explicitely.
1608 */
1609 memcg_oom_recover(memcg);
1610 }
Johannes Weiner49426422013-10-16 13:46:59 -07001611cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001612 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001613 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001614 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001615}
1616
Johannes Weinerd7365e72014-10-29 14:50:48 -07001617/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001618 * lock_page_memcg - lock a page->mem_cgroup binding
1619 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001620 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001621 * This function protects unlocked LRU pages from being moved to
1622 * another cgroup and stabilizes their page->mem_cgroup binding.
Balbir Singhd69b0422009-06-17 16:26:34 -07001623 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001624void lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001625{
1626 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001627 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001628
Johannes Weiner6de22612015-02-11 15:25:01 -08001629 /*
1630 * The RCU lock is held throughout the transaction. The fast
1631 * path can get away without acquiring the memcg->move_lock
1632 * because page moving starts with an RCU grace period.
Johannes Weiner6de22612015-02-11 15:25:01 -08001633 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001634 rcu_read_lock();
1635
1636 if (mem_cgroup_disabled())
Johannes Weiner62cccb82016-03-15 14:57:22 -07001637 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001638again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001639 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001640 if (unlikely(!memcg))
Johannes Weiner62cccb82016-03-15 14:57:22 -07001641 return;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001642
Qiang Huangbdcbb652014-06-04 16:08:21 -07001643 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001644 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001645
Johannes Weiner6de22612015-02-11 15:25:01 -08001646 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001647 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001648 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001649 goto again;
1650 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001651
1652 /*
1653 * When charge migration first begins, we can have locked and
1654 * unlocked page stat updates happening concurrently. Track
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001655 * the task who has the lock for unlock_page_memcg().
Johannes Weiner6de22612015-02-11 15:25:01 -08001656 */
1657 memcg->move_lock_task = current;
1658 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001659
Johannes Weiner62cccb82016-03-15 14:57:22 -07001660 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001661}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001662EXPORT_SYMBOL(lock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001663
Johannes Weinerd7365e72014-10-29 14:50:48 -07001664/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001665 * unlock_page_memcg - unlock a page->mem_cgroup binding
Johannes Weiner62cccb82016-03-15 14:57:22 -07001666 * @page: the page
Johannes Weinerd7365e72014-10-29 14:50:48 -07001667 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001668void unlock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001669{
Johannes Weiner62cccb82016-03-15 14:57:22 -07001670 struct mem_cgroup *memcg = page->mem_cgroup;
1671
Johannes Weiner6de22612015-02-11 15:25:01 -08001672 if (memcg && memcg->move_lock_task == current) {
1673 unsigned long flags = memcg->move_lock_flags;
1674
1675 memcg->move_lock_task = NULL;
1676 memcg->move_lock_flags = 0;
1677
1678 spin_unlock_irqrestore(&memcg->move_lock, flags);
1679 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001680
Johannes Weinerd7365e72014-10-29 14:50:48 -07001681 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001682}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001683EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001684
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001685/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001686 * size of first charge trial. "32" comes from vmscan.c's magic value.
1687 * TODO: maybe necessary to use big numbers in big irons.
1688 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001689#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001690struct memcg_stock_pcp {
1691 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001692 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001693 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001694 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001695#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001696};
1697static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001698static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001699
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001700/**
1701 * consume_stock: Try to consume stocked charge on this cpu.
1702 * @memcg: memcg to consume from.
1703 * @nr_pages: how many pages to charge.
1704 *
1705 * The charges will only happen if @memcg matches the current cpu's memcg
1706 * stock, and at least @nr_pages are available in that stock. Failure to
1707 * service an allocation will refill the stock.
1708 *
1709 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001710 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001711static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001712{
1713 struct memcg_stock_pcp *stock;
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001714 unsigned long flags;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001715 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001716
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001717 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001718 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001719
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001720 local_irq_save(flags);
1721
1722 stock = this_cpu_ptr(&memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001723 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001724 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001725 ret = true;
1726 }
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001727
1728 local_irq_restore(flags);
1729
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001730 return ret;
1731}
1732
1733/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001734 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001735 */
1736static void drain_stock(struct memcg_stock_pcp *stock)
1737{
1738 struct mem_cgroup *old = stock->cached;
1739
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001740 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001741 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001742 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001743 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001744 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001745 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001746 }
1747 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001748}
1749
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001750static void drain_local_stock(struct work_struct *dummy)
1751{
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001752 struct memcg_stock_pcp *stock;
1753 unsigned long flags;
1754
1755 local_irq_save(flags);
1756
1757 stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001758 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001759 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001760
1761 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001762}
1763
1764/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001765 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001766 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001767 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001768static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001769{
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001770 struct memcg_stock_pcp *stock;
1771 unsigned long flags;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001772
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001773 local_irq_save(flags);
1774
1775 stock = this_cpu_ptr(&memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001776 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001777 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001778 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001779 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001780 stock->nr_pages += nr_pages;
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001781
1782 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001783}
1784
1785/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001786 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001787 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001788 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001789static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001790{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001791 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001792
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001793 /* If someone's already draining, avoid adding running more workers. */
1794 if (!mutex_trylock(&percpu_charge_mutex))
1795 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001796 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001797 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001798 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001799 for_each_online_cpu(cpu) {
1800 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001801 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001802
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001803 memcg = stock->cached;
1804 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001805 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001806 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001807 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001808 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1809 if (cpu == curcpu)
1810 drain_local_stock(&stock->work);
1811 else
1812 schedule_work_on(cpu, &stock->work);
1813 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001814 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001815 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001816 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001817 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001818}
1819
Paul Gortmaker0db06282013-06-19 14:53:51 -04001820static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001821 unsigned long action,
1822 void *hcpu)
1823{
1824 int cpu = (unsigned long)hcpu;
1825 struct memcg_stock_pcp *stock;
1826
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001827 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001828 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001829
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07001830 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001831 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001832
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001833 stock = &per_cpu(memcg_stock, cpu);
1834 drain_stock(stock);
1835 return NOTIFY_OK;
1836}
1837
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001838static void reclaim_high(struct mem_cgroup *memcg,
1839 unsigned int nr_pages,
1840 gfp_t gfp_mask)
1841{
1842 do {
1843 if (page_counter_read(&memcg->memory) <= memcg->high)
1844 continue;
1845 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
1846 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1847 } while ((memcg = parent_mem_cgroup(memcg)));
1848}
1849
1850static void high_work_func(struct work_struct *work)
1851{
1852 struct mem_cgroup *memcg;
1853
1854 memcg = container_of(work, struct mem_cgroup, high_work);
1855 reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1856}
1857
Tejun Heob23afb92015-11-05 18:46:11 -08001858/*
1859 * Scheduled by try_charge() to be executed from the userland return path
1860 * and reclaims memory over the high limit.
1861 */
1862void mem_cgroup_handle_over_high(void)
1863{
1864 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001865 struct mem_cgroup *memcg;
Tejun Heob23afb92015-11-05 18:46:11 -08001866
1867 if (likely(!nr_pages))
1868 return;
1869
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001870 memcg = get_mem_cgroup_from_mm(current->mm);
1871 reclaim_high(memcg, nr_pages, GFP_KERNEL);
Tejun Heob23afb92015-11-05 18:46:11 -08001872 css_put(&memcg->css);
1873 current->memcg_nr_pages_over_high = 0;
1874}
1875
Johannes Weiner00501b52014-08-08 14:19:20 -07001876static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1877 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001878{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001879 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001880 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001881 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001882 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001883 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001884 bool may_swap = true;
1885 bool drained = false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001886
Johannes Weinerce00a962014-09-05 08:43:57 -04001887 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08001888 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001889retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001890 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08001891 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001892
Johannes Weiner7941d212016-01-14 15:21:23 -08001893 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08001894 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1895 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001896 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08001897 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001898 page_counter_uncharge(&memcg->memsw, batch);
1899 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001900 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001901 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001902 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001903 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001904
Johannes Weiner6539cc02014-08-06 16:05:42 -07001905 if (batch > nr_pages) {
1906 batch = nr_pages;
1907 goto retry;
1908 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001909
Johannes Weiner06b078f2014-08-06 16:05:44 -07001910 /*
1911 * Unlike in global OOM situations, memcg is not in a physical
1912 * memory shortage. Allow dying and OOM-killed tasks to
1913 * bypass the last charges so that they can exit quickly and
1914 * free their memory.
1915 */
1916 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1917 fatal_signal_pending(current) ||
1918 current->flags & PF_EXITING))
Tejun Heo10d53c72015-11-05 18:46:17 -08001919 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001920
Johannes Weiner89a28482016-10-27 17:46:56 -07001921 /*
1922 * Prevent unbounded recursion when reclaim operations need to
1923 * allocate memory. This might exceed the limits temporarily,
1924 * but we prefer facilitating memory reclaim and getting back
1925 * under the limit over triggering OOM kills in these cases.
1926 */
1927 if (unlikely(current->flags & PF_MEMALLOC))
1928 goto force;
1929
Johannes Weiner06b078f2014-08-06 16:05:44 -07001930 if (unlikely(task_in_memcg_oom(current)))
1931 goto nomem;
1932
Mel Gormand0164ad2015-11-06 16:28:21 -08001933 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001934 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001935
Johannes Weiner241994e2015-02-11 15:26:06 -08001936 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
1937
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001938 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1939 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07001940
Johannes Weiner61e02c72014-08-06 16:08:16 -07001941 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07001942 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07001943
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001944 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001945 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001946 drained = true;
1947 goto retry;
1948 }
1949
Johannes Weiner28c34c22014-08-06 16:05:47 -07001950 if (gfp_mask & __GFP_NORETRY)
1951 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001952 /*
1953 * Even though the limit is exceeded at this point, reclaim
1954 * may have been able to free some pages. Retry the charge
1955 * before killing the task.
1956 *
1957 * Only for regular pages, though: huge pages are rather
1958 * unlikely to succeed so close to the limit, and we fall back
1959 * to regular pages anyway in case of failure.
1960 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07001961 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001962 goto retry;
1963 /*
1964 * At task move, charge accounts can be doubly counted. So, it's
1965 * better to wait until the end of task_move if something is going on.
1966 */
1967 if (mem_cgroup_wait_acct_move(mem_over_limit))
1968 goto retry;
1969
Johannes Weiner9b130612014-08-06 16:05:51 -07001970 if (nr_retries--)
1971 goto retry;
1972
Johannes Weiner06b078f2014-08-06 16:05:44 -07001973 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08001974 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001975
Johannes Weiner6539cc02014-08-06 16:05:42 -07001976 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08001977 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001978
Johannes Weiner241994e2015-02-11 15:26:06 -08001979 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
1980
Jerome Marchand3608de02015-11-05 18:47:29 -08001981 mem_cgroup_oom(mem_over_limit, gfp_mask,
1982 get_order(nr_pages * PAGE_SIZE));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001983nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07001984 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07001985 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08001986force:
1987 /*
1988 * The allocation either can't fail or will lead to more memory
1989 * being freed very soon. Allow memory usage go over the limit
1990 * temporarily by force charging it.
1991 */
1992 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001993 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08001994 page_counter_charge(&memcg->memsw, nr_pages);
1995 css_get_many(&memcg->css, nr_pages);
1996
1997 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001998
1999done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002000 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002001 if (batch > nr_pages)
2002 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002003
Johannes Weiner241994e2015-02-11 15:26:06 -08002004 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002005 * If the hierarchy is above the normal consumption range, schedule
2006 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002007 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002008 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2009 * not recorded as it most likely matches current's and won't
2010 * change in the meantime. As high limit is checked again before
2011 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994e2015-02-11 15:26:06 -08002012 */
2013 do {
Tejun Heob23afb92015-11-05 18:46:11 -08002014 if (page_counter_read(&memcg->memory) > memcg->high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002015 /* Don't bother a random interrupted task */
2016 if (in_interrupt()) {
2017 schedule_work(&memcg->high_work);
2018 break;
2019 }
Vladimir Davydov9516a182015-12-11 13:40:24 -08002020 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002021 set_notify_resume(current);
2022 break;
2023 }
Johannes Weiner241994e2015-02-11 15:26:06 -08002024 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002025
2026 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002027}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002028
Johannes Weiner00501b52014-08-08 14:19:20 -07002029static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002030{
Johannes Weinerce00a962014-09-05 08:43:57 -04002031 if (mem_cgroup_is_root(memcg))
2032 return;
2033
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002034 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002035 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002036 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002037
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002038 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002039}
2040
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002041static void lock_page_lru(struct page *page, int *isolated)
2042{
2043 struct zone *zone = page_zone(page);
2044
Mel Gormana52633d2016-07-28 15:45:28 -07002045 spin_lock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002046 if (PageLRU(page)) {
2047 struct lruvec *lruvec;
2048
Mel Gorman599d0c92016-07-28 15:45:31 -07002049 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002050 ClearPageLRU(page);
2051 del_page_from_lru_list(page, lruvec, page_lru(page));
2052 *isolated = 1;
2053 } else
2054 *isolated = 0;
2055}
2056
2057static void unlock_page_lru(struct page *page, int isolated)
2058{
2059 struct zone *zone = page_zone(page);
2060
2061 if (isolated) {
2062 struct lruvec *lruvec;
2063
Mel Gorman599d0c92016-07-28 15:45:31 -07002064 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002065 VM_BUG_ON_PAGE(PageLRU(page), page);
2066 SetPageLRU(page);
2067 add_page_to_lru_list(page, lruvec, page_lru(page));
2068 }
Mel Gormana52633d2016-07-28 15:45:28 -07002069 spin_unlock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002070}
2071
Johannes Weiner00501b52014-08-08 14:19:20 -07002072static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002073 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002074{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002075 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002076
Johannes Weiner1306a852014-12-10 15:44:52 -08002077 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002078
2079 /*
2080 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2081 * may already be on some other mem_cgroup's LRU. Take care of it.
2082 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002083 if (lrucare)
2084 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002085
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002086 /*
2087 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002088 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002089 *
2090 * - the page is uncharged
2091 *
2092 * - the page is off-LRU
2093 *
2094 * - an anonymous fault has exclusive page access, except for
2095 * a locked page table
2096 *
2097 * - a page cache insertion, a swapin fault, or a migration
2098 * have the page locked
2099 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002100 page->mem_cgroup = memcg;
Hugh Dickins3be91272008-02-07 00:14:19 -08002101
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002102 if (lrucare)
2103 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002104}
2105
Johannes Weiner127424c2016-01-20 15:02:32 -08002106#ifndef CONFIG_SLOB
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002107static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002108{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002109 int id, size;
2110 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002111
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002112 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002113 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2114 if (id < 0)
2115 return id;
2116
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002117 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002118 return id;
2119
2120 /*
2121 * There's no space for the new id in memcg_caches arrays,
2122 * so we have to grow them.
2123 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002124 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002125
2126 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002127 if (size < MEMCG_CACHES_MIN_SIZE)
2128 size = MEMCG_CACHES_MIN_SIZE;
2129 else if (size > MEMCG_CACHES_MAX_SIZE)
2130 size = MEMCG_CACHES_MAX_SIZE;
2131
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002132 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002133 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002134 err = memcg_update_all_list_lrus(size);
2135 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002136 memcg_nr_cache_ids = size;
2137
2138 up_write(&memcg_cache_ids_sem);
2139
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002140 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002141 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002142 return err;
2143 }
2144 return id;
2145}
2146
2147static void memcg_free_cache_id(int id)
2148{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002149 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002150}
2151
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002152struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002153 struct mem_cgroup *memcg;
2154 struct kmem_cache *cachep;
2155 struct work_struct work;
2156};
2157
Vladimir Davydov754ae7e2016-12-12 16:41:29 -08002158static struct workqueue_struct *memcg_kmem_cache_create_wq;
2159
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002160static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002161{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002162 struct memcg_kmem_cache_create_work *cw =
2163 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002164 struct mem_cgroup *memcg = cw->memcg;
2165 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002166
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002167 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002168
Vladimir Davydov5722d092014-04-07 15:39:24 -07002169 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002170 kfree(cw);
2171}
2172
2173/*
2174 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002175 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002176static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2177 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002178{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002179 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002180
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002181 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002182 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002183 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002184
2185 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002186
2187 cw->memcg = memcg;
2188 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002189 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002190
Vladimir Davydov754ae7e2016-12-12 16:41:29 -08002191 queue_work(memcg_kmem_cache_create_wq, &cw->work);
Glauber Costad7f25f82012-12-18 14:22:40 -08002192}
2193
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002194static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2195 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002196{
2197 /*
2198 * We need to stop accounting when we kmalloc, because if the
2199 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002200 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002201 *
2202 * However, it is better to enclose the whole function. Depending on
2203 * the debugging options enabled, INIT_WORK(), for instance, can
2204 * trigger an allocation. This too, will make us recurse. Because at
2205 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2206 * the safest choice is to do it like this, wrapping the whole function.
2207 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002208 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002209 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002210 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002211}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002212
Vladimir Davydov45264772016-07-26 15:24:21 -07002213static inline bool memcg_kmem_bypass(void)
2214{
2215 if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
2216 return true;
2217 return false;
2218}
2219
2220/**
2221 * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2222 * @cachep: the original global kmem cache
2223 *
Glauber Costad7f25f82012-12-18 14:22:40 -08002224 * Return the kmem_cache we're supposed to use for a slab allocation.
2225 * We try to use the current memcg's version of the cache.
2226 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002227 * If the cache does not exist yet, if we are the first user of it, we
2228 * create it asynchronously in a workqueue and let the current allocation
2229 * go through with the original cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002230 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002231 * This function takes a reference to the cache it returns to assure it
2232 * won't get destroyed while we are working with it. Once the caller is
2233 * done with it, memcg_kmem_put_cache() must be called to release the
2234 * reference.
Glauber Costad7f25f82012-12-18 14:22:40 -08002235 */
Vladimir Davydov45264772016-07-26 15:24:21 -07002236struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002237{
2238 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002239 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002240 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002241
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002242 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002243
Vladimir Davydov45264772016-07-26 15:24:21 -07002244 if (memcg_kmem_bypass())
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002245 return cachep;
2246
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002247 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002248 return cachep;
2249
Vladimir Davydov8135be52014-12-12 16:56:38 -08002250 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002251 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002252 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002253 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002254
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002255 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002256 if (likely(memcg_cachep))
2257 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002258
2259 /*
2260 * If we are in a safe context (can wait, and not in interrupt
2261 * context), we could be be predictable and return right away.
2262 * This would guarantee that the allocation being performed
2263 * already belongs in the new cache.
2264 *
2265 * However, there are some clashes that can arrive from locking.
2266 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002267 * memcg_create_kmem_cache, this means no further allocation
2268 * could happen with the slab_mutex held. So it's better to
2269 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002270 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002271 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002272out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002273 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002274 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002275}
Glauber Costad7f25f82012-12-18 14:22:40 -08002276
Vladimir Davydov45264772016-07-26 15:24:21 -07002277/**
2278 * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2279 * @cachep: the cache returned by memcg_kmem_get_cache
2280 */
2281void memcg_kmem_put_cache(struct kmem_cache *cachep)
Vladimir Davydov8135be52014-12-12 16:56:38 -08002282{
2283 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002284 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002285}
2286
Vladimir Davydov45264772016-07-26 15:24:21 -07002287/**
2288 * memcg_kmem_charge: charge a kmem page
2289 * @page: page to charge
2290 * @gfp: reclaim mode
2291 * @order: allocation order
2292 * @memcg: memory cgroup to charge
2293 *
2294 * Returns 0 on success, an error code on failure.
2295 */
2296int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2297 struct mem_cgroup *memcg)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002298{
2299 unsigned int nr_pages = 1 << order;
2300 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002301 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002302
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002303 ret = try_charge(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08002304 if (ret)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002305 return ret;
Johannes Weiner52c29b02016-01-20 15:02:35 -08002306
2307 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2308 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2309 cancel_charge(memcg, nr_pages);
2310 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002311 }
2312
2313 page->mem_cgroup = memcg;
2314
2315 return 0;
2316}
2317
Vladimir Davydov45264772016-07-26 15:24:21 -07002318/**
2319 * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2320 * @page: page to charge
2321 * @gfp: reclaim mode
2322 * @order: allocation order
2323 *
2324 * Returns 0 on success, an error code on failure.
2325 */
2326int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002327{
2328 struct mem_cgroup *memcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002329 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002330
Vladimir Davydov45264772016-07-26 15:24:21 -07002331 if (memcg_kmem_bypass())
2332 return 0;
2333
Johannes Weinerdf381972014-04-07 15:37:43 -07002334 memcg = get_mem_cgroup_from_mm(current->mm);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002335 if (!mem_cgroup_is_root(memcg)) {
Vladimir Davydov45264772016-07-26 15:24:21 -07002336 ret = memcg_kmem_charge_memcg(page, gfp, order, memcg);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002337 if (!ret)
2338 __SetPageKmemcg(page);
2339 }
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002340 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002341 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002342}
Vladimir Davydov45264772016-07-26 15:24:21 -07002343/**
2344 * memcg_kmem_uncharge: uncharge a kmem page
2345 * @page: page to uncharge
2346 * @order: allocation order
2347 */
2348void memcg_kmem_uncharge(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002349{
Johannes Weiner1306a852014-12-10 15:44:52 -08002350 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002351 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002352
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002353 if (!memcg)
2354 return;
2355
Sasha Levin309381fea2014-01-23 15:52:54 -08002356 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002357
Johannes Weiner52c29b02016-01-20 15:02:35 -08002358 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2359 page_counter_uncharge(&memcg->kmem, nr_pages);
2360
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002361 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002362 if (do_memsw_account())
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002363 page_counter_uncharge(&memcg->memsw, nr_pages);
2364
Johannes Weiner1306a852014-12-10 15:44:52 -08002365 page->mem_cgroup = NULL;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002366
2367 /* slab pages do not have PageKmemcg flag set */
2368 if (PageKmemcg(page))
2369 __ClearPageKmemcg(page);
2370
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002371 css_put_many(&memcg->css, nr_pages);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002372}
Johannes Weiner127424c2016-01-20 15:02:32 -08002373#endif /* !CONFIG_SLOB */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002374
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002375#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2376
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002377/*
2378 * Because tail pages are not marked as "used", set it. We're under
Mel Gormana52633d2016-07-28 15:45:28 -07002379 * zone_lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002380 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002381void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002382{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002383 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002384
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002385 if (mem_cgroup_disabled())
2386 return;
David Rientjesb070e652013-05-07 16:18:09 -07002387
Johannes Weiner29833312014-12-10 15:44:02 -08002388 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002389 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002390
Johannes Weiner1306a852014-12-10 15:44:52 -08002391 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002392 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002393}
Hugh Dickins12d27102012-01-12 17:19:52 -08002394#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002395
Andrew Mortonc255a452012-07-31 16:43:02 -07002396#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002397static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2398 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002399{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002400 int val = (charge) ? 1 : -1;
2401 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002402}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002403
2404/**
2405 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2406 * @entry: swap entry to be moved
2407 * @from: mem_cgroup which the entry is moved from
2408 * @to: mem_cgroup which the entry is moved to
2409 *
2410 * It succeeds only when the swap_cgroup's record for this entry is the same
2411 * as the mem_cgroup's id of @from.
2412 *
2413 * Returns 0 on success, -EINVAL on failure.
2414 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002415 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002416 * both res and memsw, and called css_get().
2417 */
2418static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002419 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002420{
2421 unsigned short old_id, new_id;
2422
Li Zefan34c00c32013-09-23 16:56:01 +08002423 old_id = mem_cgroup_id(from);
2424 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002425
2426 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002427 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002428 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002429 return 0;
2430 }
2431 return -EINVAL;
2432}
2433#else
2434static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002435 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002436{
2437 return -EINVAL;
2438}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002439#endif
2440
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002441static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002442
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002443static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002444 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002445{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002446 unsigned long curusage;
2447 unsigned long oldusage;
2448 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002449 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002450 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002451
2452 /*
2453 * For keeping hierarchical_reclaim simple, how long we should retry
2454 * is depends on callers. We set our retry-count to be function
2455 * of # of children which we should visit in this loop.
2456 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002457 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2458 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002459
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002460 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002461
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002462 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002463 if (signal_pending(current)) {
2464 ret = -EINTR;
2465 break;
2466 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002467
2468 mutex_lock(&memcg_limit_mutex);
2469 if (limit > memcg->memsw.limit) {
2470 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002471 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002472 break;
2473 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002474 if (limit > memcg->memory.limit)
2475 enlarge = true;
2476 ret = page_counter_limit(&memcg->memory, limit);
2477 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002478
2479 if (!ret)
2480 break;
2481
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002482 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2483
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002484 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002485 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002486 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002487 retry_count--;
2488 else
2489 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002490 } while (retry_count);
2491
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002492 if (!ret && enlarge)
2493 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002494
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002495 return ret;
2496}
2497
Li Zefan338c8432009-06-17 16:27:15 -07002498static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002499 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002500{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002501 unsigned long curusage;
2502 unsigned long oldusage;
2503 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002504 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002505 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002506
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002507 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002508 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2509 mem_cgroup_count_children(memcg);
2510
2511 oldusage = page_counter_read(&memcg->memsw);
2512
2513 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002514 if (signal_pending(current)) {
2515 ret = -EINTR;
2516 break;
2517 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002518
2519 mutex_lock(&memcg_limit_mutex);
2520 if (limit < memcg->memory.limit) {
2521 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002522 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002523 break;
2524 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002525 if (limit > memcg->memsw.limit)
2526 enlarge = true;
2527 ret = page_counter_limit(&memcg->memsw, limit);
2528 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002529
2530 if (!ret)
2531 break;
2532
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002533 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2534
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002535 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002536 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002537 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002538 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002539 else
2540 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002541 } while (retry_count);
2542
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002543 if (!ret && enlarge)
2544 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002545
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002546 return ret;
2547}
2548
Mel Gormanef8f2322016-07-28 15:46:05 -07002549unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
Andrew Morton0608f432013-09-24 15:27:41 -07002550 gfp_t gfp_mask,
2551 unsigned long *total_scanned)
2552{
2553 unsigned long nr_reclaimed = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002554 struct mem_cgroup_per_node *mz, *next_mz = NULL;
Andrew Morton0608f432013-09-24 15:27:41 -07002555 unsigned long reclaimed;
2556 int loop = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002557 struct mem_cgroup_tree_per_node *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002558 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002559 unsigned long nr_scanned;
2560
2561 if (order > 0)
2562 return 0;
2563
Mel Gormanef8f2322016-07-28 15:46:05 -07002564 mctz = soft_limit_tree_node(pgdat->node_id);
Michal Hockod6507ff2016-08-02 14:02:37 -07002565
2566 /*
2567 * Do not even bother to check the largest node if the root
2568 * is empty. Do it lockless to prevent lock bouncing. Races
2569 * are acceptable as soft limit is best effort anyway.
2570 */
Laurent Dufoura495f722017-03-09 16:17:06 -08002571 if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root))
Michal Hockod6507ff2016-08-02 14:02:37 -07002572 return 0;
2573
Andrew Morton0608f432013-09-24 15:27:41 -07002574 /*
2575 * This loop can run a while, specially if mem_cgroup's continuously
2576 * keep exceeding their soft limit and putting the system under
2577 * pressure
2578 */
2579 do {
2580 if (next_mz)
2581 mz = next_mz;
2582 else
2583 mz = mem_cgroup_largest_soft_limit_node(mctz);
2584 if (!mz)
2585 break;
2586
2587 nr_scanned = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002588 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07002589 gfp_mask, &nr_scanned);
2590 nr_reclaimed += reclaimed;
2591 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002592 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002593 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002594
2595 /*
2596 * If we failed to reclaim anything from this memory cgroup
2597 * it is time to move on to the next cgroup
2598 */
2599 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002600 if (!reclaimed)
2601 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2602
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002603 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002604 /*
2605 * One school of thought says that we should not add
2606 * back the node to the tree if reclaim returns 0.
2607 * But our reclaim could return 0, simply because due
2608 * to priority we are exposing a smaller subset of
2609 * memory to reclaim from. Consider this as a longer
2610 * term TODO.
2611 */
2612 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002613 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002614 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002615 css_put(&mz->memcg->css);
2616 loop++;
2617 /*
2618 * Could not reclaim anything and there are no more
2619 * mem cgroups to try or we seem to be looping without
2620 * reclaiming anything.
2621 */
2622 if (!nr_reclaimed &&
2623 (next_mz == NULL ||
2624 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2625 break;
2626 } while (!nr_reclaimed);
2627 if (next_mz)
2628 css_put(&next_mz->memcg->css);
2629 return nr_reclaimed;
2630}
2631
Tejun Heoea280e72014-05-16 13:22:48 -04002632/*
2633 * Test whether @memcg has children, dead or alive. Note that this
2634 * function doesn't care whether @memcg has use_hierarchy enabled and
2635 * returns %true if there are child csses according to the cgroup
2636 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2637 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002638static inline bool memcg_has_children(struct mem_cgroup *memcg)
2639{
Tejun Heoea280e72014-05-16 13:22:48 -04002640 bool ret;
2641
Tejun Heoea280e72014-05-16 13:22:48 -04002642 rcu_read_lock();
2643 ret = css_next_child(NULL, &memcg->css);
2644 rcu_read_unlock();
2645 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002646}
2647
2648/*
Greg Thelen51038172016-05-20 16:58:18 -07002649 * Reclaims as many pages from the given memcg as possible.
Michal Hockoc26251f2012-10-26 13:37:28 +02002650 *
2651 * Caller is responsible for holding css reference for memcg.
2652 */
2653static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2654{
2655 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002656
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002657 /* we call try-to-free pages for make this cgroup empty */
2658 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002659 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002660 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002661 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002662
Michal Hockoc26251f2012-10-26 13:37:28 +02002663 if (signal_pending(current))
2664 return -EINTR;
2665
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002666 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2667 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002668 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002669 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002670 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002671 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002672 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002673
2674 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002675
2676 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002677}
2678
Tejun Heo6770c642014-05-13 12:16:21 -04002679static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2680 char *buf, size_t nbytes,
2681 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002682{
Tejun Heo6770c642014-05-13 12:16:21 -04002683 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002684
Michal Hockod8423012012-10-26 13:37:29 +02002685 if (mem_cgroup_is_root(memcg))
2686 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002687 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002688}
2689
Tejun Heo182446d2013-08-08 20:11:24 -04002690static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2691 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002692{
Tejun Heo182446d2013-08-08 20:11:24 -04002693 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002694}
2695
Tejun Heo182446d2013-08-08 20:11:24 -04002696static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2697 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002698{
2699 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002700 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002701 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002702
Glauber Costa567fb432012-07-31 16:43:07 -07002703 if (memcg->use_hierarchy == val)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002704 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07002705
Balbir Singh18f59ea2009-01-07 18:08:07 -08002706 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002707 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002708 * in the child subtrees. If it is unset, then the change can
2709 * occur, provided the current cgroup has no children.
2710 *
2711 * For the root cgroup, parent_mem is NULL, we allow value to be
2712 * set if there are no children.
2713 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002714 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002715 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002716 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002717 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002718 else
2719 retval = -EBUSY;
2720 } else
2721 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002722
Balbir Singh18f59ea2009-01-07 18:08:07 -08002723 return retval;
2724}
2725
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002726static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
Johannes Weinerce00a962014-09-05 08:43:57 -04002727{
2728 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002729 int i;
Johannes Weinerce00a962014-09-05 08:43:57 -04002730
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002731 memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
Johannes Weinerce00a962014-09-05 08:43:57 -04002732
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002733 for_each_mem_cgroup_tree(iter, memcg) {
2734 for (i = 0; i < MEMCG_NR_STAT; i++)
2735 stat[i] += mem_cgroup_read_stat(iter, i);
2736 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002737}
2738
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002739static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
Johannes Weiner587d9f72016-01-20 15:03:19 -08002740{
2741 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002742 int i;
Johannes Weiner587d9f72016-01-20 15:03:19 -08002743
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002744 memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
Johannes Weiner587d9f72016-01-20 15:03:19 -08002745
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002746 for_each_mem_cgroup_tree(iter, memcg) {
2747 for (i = 0; i < MEMCG_NR_EVENTS; i++)
2748 events[i] += mem_cgroup_read_events(iter, i);
2749 }
Johannes Weiner587d9f72016-01-20 15:03:19 -08002750}
2751
Andrew Morton6f646152015-11-06 16:28:58 -08002752static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04002753{
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002754 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002755
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002756 if (mem_cgroup_is_root(memcg)) {
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002757 struct mem_cgroup *iter;
2758
2759 for_each_mem_cgroup_tree(iter, memcg) {
2760 val += mem_cgroup_read_stat(iter,
2761 MEM_CGROUP_STAT_CACHE);
2762 val += mem_cgroup_read_stat(iter,
2763 MEM_CGROUP_STAT_RSS);
2764 if (swap)
2765 val += mem_cgroup_read_stat(iter,
2766 MEM_CGROUP_STAT_SWAP);
2767 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002768 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002769 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002770 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002771 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002772 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002773 }
Michal Hockoc12176d2015-11-05 18:50:29 -08002774 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002775}
2776
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002777enum {
2778 RES_USAGE,
2779 RES_LIMIT,
2780 RES_MAX_USAGE,
2781 RES_FAILCNT,
2782 RES_SOFT_LIMIT,
2783};
Johannes Weinerce00a962014-09-05 08:43:57 -04002784
Tejun Heo791badb2013-12-05 12:28:02 -05002785static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002786 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002787{
Tejun Heo182446d2013-08-08 20:11:24 -04002788 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002789 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002790
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002791 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002792 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002793 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002794 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002795 case _MEMSWAP:
2796 counter = &memcg->memsw;
2797 break;
2798 case _KMEM:
2799 counter = &memcg->kmem;
2800 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002801 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08002802 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002803 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002804 default:
2805 BUG();
2806 }
2807
2808 switch (MEMFILE_ATTR(cft->private)) {
2809 case RES_USAGE:
2810 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08002811 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002812 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08002813 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002814 return (u64)page_counter_read(counter) * PAGE_SIZE;
2815 case RES_LIMIT:
2816 return (u64)counter->limit * PAGE_SIZE;
2817 case RES_MAX_USAGE:
2818 return (u64)counter->watermark * PAGE_SIZE;
2819 case RES_FAILCNT:
2820 return counter->failcnt;
2821 case RES_SOFT_LIMIT:
2822 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002823 default:
2824 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002825 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002826}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002827
Johannes Weiner127424c2016-01-20 15:02:32 -08002828#ifndef CONFIG_SLOB
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002829static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002830{
Vladimir Davydovd6441632014-01-23 15:53:09 -08002831 int memcg_id;
2832
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002833 if (cgroup_memory_nokmem)
2834 return 0;
2835
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002836 BUG_ON(memcg->kmemcg_id >= 0);
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002837 BUG_ON(memcg->kmem_state);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002838
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002839 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002840 if (memcg_id < 0)
2841 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002842
Johannes Weineref129472016-01-14 15:21:34 -08002843 static_branch_inc(&memcg_kmem_enabled_key);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002844 /*
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002845 * A memory cgroup is considered kmem-online as soon as it gets
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002846 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002847 * guarantee no one starts accounting before all call sites are
2848 * patched.
2849 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002850 memcg->kmemcg_id = memcg_id;
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002851 memcg->kmem_state = KMEM_ONLINE;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002852
2853 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002854}
2855
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002856static void memcg_offline_kmem(struct mem_cgroup *memcg)
2857{
2858 struct cgroup_subsys_state *css;
2859 struct mem_cgroup *parent, *child;
2860 int kmemcg_id;
2861
2862 if (memcg->kmem_state != KMEM_ONLINE)
2863 return;
2864 /*
2865 * Clear the online state before clearing memcg_caches array
2866 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2867 * guarantees that no cache will be created for this cgroup
2868 * after we are done (see memcg_create_kmem_cache()).
2869 */
2870 memcg->kmem_state = KMEM_ALLOCATED;
2871
2872 memcg_deactivate_kmem_caches(memcg);
2873
2874 kmemcg_id = memcg->kmemcg_id;
2875 BUG_ON(kmemcg_id < 0);
2876
2877 parent = parent_mem_cgroup(memcg);
2878 if (!parent)
2879 parent = root_mem_cgroup;
2880
2881 /*
2882 * Change kmemcg_id of this cgroup and all its descendants to the
2883 * parent's id, and then move all entries from this cgroup's list_lrus
2884 * to ones of the parent. After we have finished, all list_lrus
2885 * corresponding to this cgroup are guaranteed to remain empty. The
2886 * ordering is imposed by list_lru_node->lock taken by
2887 * memcg_drain_all_list_lrus().
2888 */
Tejun Heo3a06bb72016-06-03 14:55:44 -07002889 rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002890 css_for_each_descendant_pre(css, &memcg->css) {
2891 child = mem_cgroup_from_css(css);
2892 BUG_ON(child->kmemcg_id != kmemcg_id);
2893 child->kmemcg_id = parent->kmemcg_id;
2894 if (!memcg->use_hierarchy)
2895 break;
2896 }
Tejun Heo3a06bb72016-06-03 14:55:44 -07002897 rcu_read_unlock();
2898
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002899 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2900
2901 memcg_free_cache_id(kmemcg_id);
2902}
2903
2904static void memcg_free_kmem(struct mem_cgroup *memcg)
2905{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002906 /* css_alloc() failed, offlining didn't happen */
2907 if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2908 memcg_offline_kmem(memcg);
2909
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002910 if (memcg->kmem_state == KMEM_ALLOCATED) {
2911 memcg_destroy_kmem_caches(memcg);
2912 static_branch_dec(&memcg_kmem_enabled_key);
2913 WARN_ON(page_counter_read(&memcg->kmem));
2914 }
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002915}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002916#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002917static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08002918{
2919 return 0;
2920}
2921static void memcg_offline_kmem(struct mem_cgroup *memcg)
2922{
2923}
2924static void memcg_free_kmem(struct mem_cgroup *memcg)
2925{
2926}
2927#endif /* !CONFIG_SLOB */
2928
Johannes Weiner127424c2016-01-20 15:02:32 -08002929static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2930 unsigned long limit)
2931{
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002932 int ret;
Johannes Weiner127424c2016-01-20 15:02:32 -08002933
2934 mutex_lock(&memcg_limit_mutex);
Johannes Weiner127424c2016-01-20 15:02:32 -08002935 ret = page_counter_limit(&memcg->kmem, limit);
Johannes Weiner127424c2016-01-20 15:02:32 -08002936 mutex_unlock(&memcg_limit_mutex);
2937 return ret;
2938}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002939
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002940static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2941{
2942 int ret;
2943
2944 mutex_lock(&memcg_limit_mutex);
2945
Johannes Weiner0db15292016-01-20 15:02:50 -08002946 ret = page_counter_limit(&memcg->tcpmem, limit);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002947 if (ret)
2948 goto out;
2949
Johannes Weiner0db15292016-01-20 15:02:50 -08002950 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002951 /*
2952 * The active flag needs to be written after the static_key
2953 * update. This is what guarantees that the socket activation
Johannes Weiner2d758072016-10-07 17:00:58 -07002954 * function is the last one to run. See mem_cgroup_sk_alloc()
2955 * for details, and note that we don't mark any socket as
2956 * belonging to this memcg until that flag is up.
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002957 *
2958 * We need to do this, because static_keys will span multiple
2959 * sites, but we can't control their order. If we mark a socket
2960 * as accounted, but the accounting functions are not patched in
2961 * yet, we'll lose accounting.
2962 *
Johannes Weiner2d758072016-10-07 17:00:58 -07002963 * We never race with the readers in mem_cgroup_sk_alloc(),
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002964 * because when this value change, the code to process it is not
2965 * patched in yet.
2966 */
2967 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08002968 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002969 }
2970out:
2971 mutex_unlock(&memcg_limit_mutex);
2972 return ret;
2973}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002974
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002975/*
2976 * The user of this function is...
2977 * RES_LIMIT.
2978 */
Tejun Heo451af502014-05-13 12:16:21 -04002979static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2980 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002981{
Tejun Heo451af502014-05-13 12:16:21 -04002982 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002983 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002984 int ret;
2985
Tejun Heo451af502014-05-13 12:16:21 -04002986 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08002987 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002988 if (ret)
2989 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07002990
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002991 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002992 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07002993 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2994 ret = -EINVAL;
2995 break;
2996 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002997 switch (MEMFILE_TYPE(of_cft(of)->private)) {
2998 case _MEM:
2999 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003000 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003001 case _MEMSWAP:
3002 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3003 break;
3004 case _KMEM:
3005 ret = memcg_update_kmem_limit(memcg, nr_pages);
3006 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003007 case _TCP:
3008 ret = memcg_update_tcp_limit(memcg, nr_pages);
3009 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003010 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003011 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003012 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003013 memcg->soft_limit = nr_pages;
3014 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003015 break;
3016 }
Tejun Heo451af502014-05-13 12:16:21 -04003017 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003018}
3019
Tejun Heo6770c642014-05-13 12:16:21 -04003020static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3021 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003022{
Tejun Heo6770c642014-05-13 12:16:21 -04003023 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003024 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003025
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003026 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3027 case _MEM:
3028 counter = &memcg->memory;
3029 break;
3030 case _MEMSWAP:
3031 counter = &memcg->memsw;
3032 break;
3033 case _KMEM:
3034 counter = &memcg->kmem;
3035 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003036 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003037 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003038 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003039 default:
3040 BUG();
3041 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003042
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003043 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003044 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003045 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003046 break;
3047 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003048 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003049 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003050 default:
3051 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003052 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003053
Tejun Heo6770c642014-05-13 12:16:21 -04003054 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003055}
3056
Tejun Heo182446d2013-08-08 20:11:24 -04003057static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003058 struct cftype *cft)
3059{
Tejun Heo182446d2013-08-08 20:11:24 -04003060 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003061}
3062
Daisuke Nishimura02491442010-03-10 15:22:17 -08003063#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003064static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003065 struct cftype *cft, u64 val)
3066{
Tejun Heo182446d2013-08-08 20:11:24 -04003067 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003068
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003069 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003070 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003071
Glauber Costaee5e8472013-02-22 16:34:50 -08003072 /*
3073 * No kind of locking is needed in here, because ->can_attach() will
3074 * check this value once in the beginning of the process, and then carry
3075 * on with stale data. This means that changes to this value will only
3076 * affect task migrations starting after the change.
3077 */
3078 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003079 return 0;
3080}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003081#else
Tejun Heo182446d2013-08-08 20:11:24 -04003082static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003083 struct cftype *cft, u64 val)
3084{
3085 return -ENOSYS;
3086}
3087#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003088
Ying Han406eb0c2011-05-26 16:25:37 -07003089#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003090static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003091{
Greg Thelen25485de2013-11-12 15:07:40 -08003092 struct numa_stat {
3093 const char *name;
3094 unsigned int lru_mask;
3095 };
3096
3097 static const struct numa_stat stats[] = {
3098 { "total", LRU_ALL },
3099 { "file", LRU_ALL_FILE },
3100 { "anon", LRU_ALL_ANON },
3101 { "unevictable", BIT(LRU_UNEVICTABLE) },
3102 };
3103 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003104 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003105 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003106 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003107
Greg Thelen25485de2013-11-12 15:07:40 -08003108 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3109 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3110 seq_printf(m, "%s=%lu", stat->name, nr);
3111 for_each_node_state(nid, N_MEMORY) {
3112 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3113 stat->lru_mask);
3114 seq_printf(m, " N%d=%lu", nid, nr);
3115 }
3116 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003117 }
Ying Han406eb0c2011-05-26 16:25:37 -07003118
Ying Han071aee12013-11-12 15:07:41 -08003119 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3120 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003121
Ying Han071aee12013-11-12 15:07:41 -08003122 nr = 0;
3123 for_each_mem_cgroup_tree(iter, memcg)
3124 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3125 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3126 for_each_node_state(nid, N_MEMORY) {
3127 nr = 0;
3128 for_each_mem_cgroup_tree(iter, memcg)
3129 nr += mem_cgroup_node_nr_lru_pages(
3130 iter, nid, stat->lru_mask);
3131 seq_printf(m, " N%d=%lu", nid, nr);
3132 }
3133 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003134 }
Ying Han406eb0c2011-05-26 16:25:37 -07003135
Ying Han406eb0c2011-05-26 16:25:37 -07003136 return 0;
3137}
3138#endif /* CONFIG_NUMA */
3139
Tejun Heo2da8ca82013-12-05 12:28:04 -05003140static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003141{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003142 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003143 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003144 struct mem_cgroup *mi;
3145 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003146
Greg Thelen0ca44b12015-02-11 15:25:58 -08003147 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3148 MEM_CGROUP_STAT_NSTATS);
3149 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3150 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003151 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3152
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003153 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Johannes Weiner7941d212016-01-14 15:21:23 -08003154 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003155 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07003156 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003157 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003158 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003159
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003160 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3161 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3162 mem_cgroup_read_events(memcg, i));
3163
3164 for (i = 0; i < NR_LRU_LISTS; i++)
3165 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3166 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3167
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003168 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003169 memory = memsw = PAGE_COUNTER_MAX;
3170 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3171 memory = min(memory, mi->memory.limit);
3172 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003173 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003174 seq_printf(m, "hierarchical_memory_limit %llu\n",
3175 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08003176 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003177 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3178 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003179
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003180 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003181 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003182
Johannes Weiner7941d212016-01-14 15:21:23 -08003183 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003184 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003185 for_each_mem_cgroup_tree(mi, memcg)
3186 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
Greg Thelen484ebb32015-10-01 15:37:05 -07003187 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003188 }
3189
3190 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3191 unsigned long long val = 0;
3192
3193 for_each_mem_cgroup_tree(mi, memcg)
3194 val += mem_cgroup_read_events(mi, i);
3195 seq_printf(m, "total_%s %llu\n",
3196 mem_cgroup_events_names[i], val);
3197 }
3198
3199 for (i = 0; i < NR_LRU_LISTS; i++) {
3200 unsigned long long val = 0;
3201
3202 for_each_mem_cgroup_tree(mi, memcg)
3203 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3204 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003205 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003206
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003207#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003208 {
Mel Gormanef8f2322016-07-28 15:46:05 -07003209 pg_data_t *pgdat;
3210 struct mem_cgroup_per_node *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003211 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003212 unsigned long recent_rotated[2] = {0, 0};
3213 unsigned long recent_scanned[2] = {0, 0};
3214
Mel Gormanef8f2322016-07-28 15:46:05 -07003215 for_each_online_pgdat(pgdat) {
3216 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
3217 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003218
Mel Gormanef8f2322016-07-28 15:46:05 -07003219 recent_rotated[0] += rstat->recent_rotated[0];
3220 recent_rotated[1] += rstat->recent_rotated[1];
3221 recent_scanned[0] += rstat->recent_scanned[0];
3222 recent_scanned[1] += rstat->recent_scanned[1];
3223 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003224 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3225 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3226 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3227 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003228 }
3229#endif
3230
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003231 return 0;
3232}
3233
Tejun Heo182446d2013-08-08 20:11:24 -04003234static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3235 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003236{
Tejun Heo182446d2013-08-08 20:11:24 -04003237 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003238
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003239 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003240}
3241
Tejun Heo182446d2013-08-08 20:11:24 -04003242static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3243 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003244{
Tejun Heo182446d2013-08-08 20:11:24 -04003245 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003246
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003247 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003248 return -EINVAL;
3249
Linus Torvalds14208b02014-06-09 15:03:33 -07003250 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003251 memcg->swappiness = val;
3252 else
3253 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003254
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003255 return 0;
3256}
3257
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003258static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3259{
3260 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003261 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003262 int i;
3263
3264 rcu_read_lock();
3265 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003266 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003267 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003268 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003269
3270 if (!t)
3271 goto unlock;
3272
Johannes Weinerce00a962014-09-05 08:43:57 -04003273 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003274
3275 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003276 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003277 * If it's not true, a threshold was crossed after last
3278 * call of __mem_cgroup_threshold().
3279 */
Phil Carmody5407a562010-05-26 14:42:42 -07003280 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003281
3282 /*
3283 * Iterate backward over array of thresholds starting from
3284 * current_threshold and check if a threshold is crossed.
3285 * If none of thresholds below usage is crossed, we read
3286 * only one element of the array here.
3287 */
3288 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3289 eventfd_signal(t->entries[i].eventfd, 1);
3290
3291 /* i = current_threshold + 1 */
3292 i++;
3293
3294 /*
3295 * Iterate forward over array of thresholds starting from
3296 * current_threshold+1 and check if a threshold is crossed.
3297 * If none of thresholds above usage is crossed, we read
3298 * only one element of the array here.
3299 */
3300 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3301 eventfd_signal(t->entries[i].eventfd, 1);
3302
3303 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003304 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003305unlock:
3306 rcu_read_unlock();
3307}
3308
3309static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3310{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003311 while (memcg) {
3312 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08003313 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003314 __mem_cgroup_threshold(memcg, true);
3315
3316 memcg = parent_mem_cgroup(memcg);
3317 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003318}
3319
3320static int compare_thresholds(const void *a, const void *b)
3321{
3322 const struct mem_cgroup_threshold *_a = a;
3323 const struct mem_cgroup_threshold *_b = b;
3324
Greg Thelen2bff24a2013-09-11 14:23:08 -07003325 if (_a->threshold > _b->threshold)
3326 return 1;
3327
3328 if (_a->threshold < _b->threshold)
3329 return -1;
3330
3331 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003332}
3333
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003334static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003335{
3336 struct mem_cgroup_eventfd_list *ev;
3337
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003338 spin_lock(&memcg_oom_lock);
3339
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003340 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003341 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003342
3343 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003344 return 0;
3345}
3346
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003347static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003348{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003349 struct mem_cgroup *iter;
3350
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003351 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003352 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003353}
3354
Tejun Heo59b6f872013-11-22 18:20:43 -05003355static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003356 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003357{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003358 struct mem_cgroup_thresholds *thresholds;
3359 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003360 unsigned long threshold;
3361 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003362 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003363
Johannes Weiner650c5e52015-02-11 15:26:03 -08003364 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003365 if (ret)
3366 return ret;
3367
3368 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003369
Johannes Weiner05b84302014-08-06 16:05:59 -07003370 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003371 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003372 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003373 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003374 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003375 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003376 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003377 BUG();
3378
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003379 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003380 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003381 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3382
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003383 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003384
3385 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003386 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003387 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003388 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003389 ret = -ENOMEM;
3390 goto unlock;
3391 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003392 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003393
3394 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003395 if (thresholds->primary) {
3396 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003397 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003398 }
3399
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003400 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003401 new->entries[size - 1].eventfd = eventfd;
3402 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003403
3404 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003405 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003406 compare_thresholds, NULL);
3407
3408 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003409 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003410 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003411 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003412 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003413 * new->current_threshold will not be used until
3414 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003415 * it here.
3416 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003417 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003418 } else
3419 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003420 }
3421
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003422 /* Free old spare buffer and save old primary buffer as spare */
3423 kfree(thresholds->spare);
3424 thresholds->spare = thresholds->primary;
3425
3426 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003427
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003428 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003429 synchronize_rcu();
3430
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003431unlock:
3432 mutex_unlock(&memcg->thresholds_lock);
3433
3434 return ret;
3435}
3436
Tejun Heo59b6f872013-11-22 18:20:43 -05003437static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003438 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003439{
Tejun Heo59b6f872013-11-22 18:20:43 -05003440 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003441}
3442
Tejun Heo59b6f872013-11-22 18:20:43 -05003443static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003444 struct eventfd_ctx *eventfd, const char *args)
3445{
Tejun Heo59b6f872013-11-22 18:20:43 -05003446 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003447}
3448
Tejun Heo59b6f872013-11-22 18:20:43 -05003449static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003450 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003451{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003452 struct mem_cgroup_thresholds *thresholds;
3453 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003454 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003455 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003456
3457 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003458
3459 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003460 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003461 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003462 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003463 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003464 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003465 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003466 BUG();
3467
Anton Vorontsov371528c2012-02-24 05:14:46 +04003468 if (!thresholds->primary)
3469 goto unlock;
3470
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003471 /* Check if a threshold crossed before removing */
3472 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3473
3474 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003475 size = 0;
3476 for (i = 0; i < thresholds->primary->size; i++) {
3477 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003478 size++;
3479 }
3480
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003481 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003482
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003483 /* Set thresholds array to NULL if we don't have thresholds */
3484 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003485 kfree(new);
3486 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003487 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003488 }
3489
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003490 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003491
3492 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003493 new->current_threshold = -1;
3494 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3495 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003496 continue;
3497
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003498 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003499 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003500 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003501 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003502 * until rcu_assign_pointer(), so it's safe to increment
3503 * it here.
3504 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003505 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003506 }
3507 j++;
3508 }
3509
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003510swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003511 /* Swap primary and spare array */
3512 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003513
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003514 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003515
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003516 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003517 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08003518
3519 /* If all events are unregistered, free the spare array */
3520 if (!new) {
3521 kfree(thresholds->spare);
3522 thresholds->spare = NULL;
3523 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04003524unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003525 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003526}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003527
Tejun Heo59b6f872013-11-22 18:20:43 -05003528static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003529 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003530{
Tejun Heo59b6f872013-11-22 18:20:43 -05003531 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003532}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003533
Tejun Heo59b6f872013-11-22 18:20:43 -05003534static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003535 struct eventfd_ctx *eventfd)
3536{
Tejun Heo59b6f872013-11-22 18:20:43 -05003537 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003538}
3539
Tejun Heo59b6f872013-11-22 18:20:43 -05003540static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003541 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003542{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003543 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003544
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003545 event = kmalloc(sizeof(*event), GFP_KERNEL);
3546 if (!event)
3547 return -ENOMEM;
3548
Michal Hocko1af8efe2011-07-26 16:08:24 -07003549 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003550
3551 event->eventfd = eventfd;
3552 list_add(&event->list, &memcg->oom_notify);
3553
3554 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07003555 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003556 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003557 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003558
3559 return 0;
3560}
3561
Tejun Heo59b6f872013-11-22 18:20:43 -05003562static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003563 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003564{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003565 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003566
Michal Hocko1af8efe2011-07-26 16:08:24 -07003567 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003568
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003569 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003570 if (ev->eventfd == eventfd) {
3571 list_del(&ev->list);
3572 kfree(ev);
3573 }
3574 }
3575
Michal Hocko1af8efe2011-07-26 16:08:24 -07003576 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003577}
3578
Tejun Heo2da8ca82013-12-05 12:28:04 -05003579static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003580{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003581 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003582
Tejun Heo791badb2013-12-05 12:28:02 -05003583 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07003584 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003585 return 0;
3586}
3587
Tejun Heo182446d2013-08-08 20:11:24 -04003588static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003589 struct cftype *cft, u64 val)
3590{
Tejun Heo182446d2013-08-08 20:11:24 -04003591 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003592
3593 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003594 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003595 return -EINVAL;
3596
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003597 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003598 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003599 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003600
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003601 return 0;
3602}
3603
Tejun Heo52ebea72015-05-22 17:13:37 -04003604#ifdef CONFIG_CGROUP_WRITEBACK
3605
3606struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3607{
3608 return &memcg->cgwb_list;
3609}
3610
Tejun Heo841710a2015-05-22 18:23:33 -04003611static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3612{
3613 return wb_domain_init(&memcg->cgwb_domain, gfp);
3614}
3615
3616static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3617{
3618 wb_domain_exit(&memcg->cgwb_domain);
3619}
3620
Tejun Heo2529bb32015-05-22 18:23:34 -04003621static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3622{
3623 wb_domain_size_changed(&memcg->cgwb_domain);
3624}
3625
Tejun Heo841710a2015-05-22 18:23:33 -04003626struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3627{
3628 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3629
3630 if (!memcg->css.parent)
3631 return NULL;
3632
3633 return &memcg->cgwb_domain;
3634}
3635
Tejun Heoc2aa7232015-05-22 18:23:35 -04003636/**
3637 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3638 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003639 * @pfilepages: out parameter for number of file pages
3640 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04003641 * @pdirty: out parameter for number of dirty pages
3642 * @pwriteback: out parameter for number of pages under writeback
3643 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003644 * Determine the numbers of file, headroom, dirty, and writeback pages in
3645 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3646 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003647 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003648 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3649 * headroom is calculated as the lowest headroom of itself and the
3650 * ancestors. Note that this doesn't consider the actual amount of
3651 * available memory in the system. The caller should further cap
3652 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003653 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003654void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3655 unsigned long *pheadroom, unsigned long *pdirty,
3656 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04003657{
3658 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3659 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003660
3661 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3662
3663 /* this should eventually include NR_UNSTABLE_NFS */
3664 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003665 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3666 (1 << LRU_ACTIVE_FILE));
3667 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003668
Tejun Heoc2aa7232015-05-22 18:23:35 -04003669 while ((parent = parent_mem_cgroup(memcg))) {
3670 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3671 unsigned long used = page_counter_read(&memcg->memory);
3672
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003673 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04003674 memcg = parent;
3675 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04003676}
3677
Tejun Heo841710a2015-05-22 18:23:33 -04003678#else /* CONFIG_CGROUP_WRITEBACK */
3679
3680static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3681{
3682 return 0;
3683}
3684
3685static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3686{
3687}
3688
Tejun Heo2529bb32015-05-22 18:23:34 -04003689static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3690{
3691}
3692
Tejun Heo52ebea72015-05-22 17:13:37 -04003693#endif /* CONFIG_CGROUP_WRITEBACK */
3694
Tejun Heo79bd9812013-11-22 18:20:42 -05003695/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003696 * DO NOT USE IN NEW FILES.
3697 *
3698 * "cgroup.event_control" implementation.
3699 *
3700 * This is way over-engineered. It tries to support fully configurable
3701 * events for each user. Such level of flexibility is completely
3702 * unnecessary especially in the light of the planned unified hierarchy.
3703 *
3704 * Please deprecate this and replace with something simpler if at all
3705 * possible.
3706 */
3707
3708/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003709 * Unregister event and free resources.
3710 *
3711 * Gets called from workqueue.
3712 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003713static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003714{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003715 struct mem_cgroup_event *event =
3716 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003717 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003718
3719 remove_wait_queue(event->wqh, &event->wait);
3720
Tejun Heo59b6f872013-11-22 18:20:43 -05003721 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003722
3723 /* Notify userspace the event is going away. */
3724 eventfd_signal(event->eventfd, 1);
3725
3726 eventfd_ctx_put(event->eventfd);
3727 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003728 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003729}
3730
3731/*
3732 * Gets called on POLLHUP on eventfd when user closes it.
3733 *
3734 * Called with wqh->lock held and interrupts disabled.
3735 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003736static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3737 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003738{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003739 struct mem_cgroup_event *event =
3740 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003741 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003742 unsigned long flags = (unsigned long)key;
3743
3744 if (flags & POLLHUP) {
3745 /*
3746 * If the event has been detached at cgroup removal, we
3747 * can simply return knowing the other side will cleanup
3748 * for us.
3749 *
3750 * We can't race against event freeing since the other
3751 * side will require wqh->lock via remove_wait_queue(),
3752 * which we hold.
3753 */
Tejun Heofba94802013-11-22 18:20:43 -05003754 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003755 if (!list_empty(&event->list)) {
3756 list_del_init(&event->list);
3757 /*
3758 * We are in atomic context, but cgroup_event_remove()
3759 * may sleep, so we have to call it in workqueue.
3760 */
3761 schedule_work(&event->remove);
3762 }
Tejun Heofba94802013-11-22 18:20:43 -05003763 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003764 }
3765
3766 return 0;
3767}
3768
Tejun Heo3bc942f2013-11-22 18:20:44 -05003769static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003770 wait_queue_head_t *wqh, poll_table *pt)
3771{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003772 struct mem_cgroup_event *event =
3773 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003774
3775 event->wqh = wqh;
3776 add_wait_queue(wqh, &event->wait);
3777}
3778
3779/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003780 * DO NOT USE IN NEW FILES.
3781 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003782 * Parse input and register new cgroup event handler.
3783 *
3784 * Input must be in format '<event_fd> <control_fd> <args>'.
3785 * Interpretation of args is defined by control file implementation.
3786 */
Tejun Heo451af502014-05-13 12:16:21 -04003787static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3788 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003789{
Tejun Heo451af502014-05-13 12:16:21 -04003790 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003791 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003792 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003793 struct cgroup_subsys_state *cfile_css;
3794 unsigned int efd, cfd;
3795 struct fd efile;
3796 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003797 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003798 char *endp;
3799 int ret;
3800
Tejun Heo451af502014-05-13 12:16:21 -04003801 buf = strstrip(buf);
3802
3803 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003804 if (*endp != ' ')
3805 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003806 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003807
Tejun Heo451af502014-05-13 12:16:21 -04003808 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003809 if ((*endp != ' ') && (*endp != '\0'))
3810 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003811 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003812
3813 event = kzalloc(sizeof(*event), GFP_KERNEL);
3814 if (!event)
3815 return -ENOMEM;
3816
Tejun Heo59b6f872013-11-22 18:20:43 -05003817 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003818 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003819 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3820 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3821 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003822
3823 efile = fdget(efd);
3824 if (!efile.file) {
3825 ret = -EBADF;
3826 goto out_kfree;
3827 }
3828
3829 event->eventfd = eventfd_ctx_fileget(efile.file);
3830 if (IS_ERR(event->eventfd)) {
3831 ret = PTR_ERR(event->eventfd);
3832 goto out_put_efile;
3833 }
3834
3835 cfile = fdget(cfd);
3836 if (!cfile.file) {
3837 ret = -EBADF;
3838 goto out_put_eventfd;
3839 }
3840
3841 /* the process need read permission on control file */
3842 /* AV: shouldn't we check that it's been opened for read instead? */
3843 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3844 if (ret < 0)
3845 goto out_put_cfile;
3846
Tejun Heo79bd9812013-11-22 18:20:42 -05003847 /*
Tejun Heofba94802013-11-22 18:20:43 -05003848 * Determine the event callbacks and set them in @event. This used
3849 * to be done via struct cftype but cgroup core no longer knows
3850 * about these events. The following is crude but the whole thing
3851 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003852 *
3853 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003854 */
Al Virob5830432014-10-31 01:22:04 -04003855 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003856
3857 if (!strcmp(name, "memory.usage_in_bytes")) {
3858 event->register_event = mem_cgroup_usage_register_event;
3859 event->unregister_event = mem_cgroup_usage_unregister_event;
3860 } else if (!strcmp(name, "memory.oom_control")) {
3861 event->register_event = mem_cgroup_oom_register_event;
3862 event->unregister_event = mem_cgroup_oom_unregister_event;
3863 } else if (!strcmp(name, "memory.pressure_level")) {
3864 event->register_event = vmpressure_register_event;
3865 event->unregister_event = vmpressure_unregister_event;
3866 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003867 event->register_event = memsw_cgroup_usage_register_event;
3868 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003869 } else {
3870 ret = -EINVAL;
3871 goto out_put_cfile;
3872 }
3873
3874 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003875 * Verify @cfile should belong to @css. Also, remaining events are
3876 * automatically removed on cgroup destruction but the removal is
3877 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003878 */
Al Virob5830432014-10-31 01:22:04 -04003879 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003880 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003881 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003882 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003883 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003884 if (cfile_css != css) {
3885 css_put(cfile_css);
3886 goto out_put_cfile;
3887 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003888
Tejun Heo451af502014-05-13 12:16:21 -04003889 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003890 if (ret)
3891 goto out_put_css;
3892
3893 efile.file->f_op->poll(efile.file, &event->pt);
3894
Tejun Heofba94802013-11-22 18:20:43 -05003895 spin_lock(&memcg->event_list_lock);
3896 list_add(&event->list, &memcg->event_list);
3897 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003898
3899 fdput(cfile);
3900 fdput(efile);
3901
Tejun Heo451af502014-05-13 12:16:21 -04003902 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05003903
3904out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05003905 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003906out_put_cfile:
3907 fdput(cfile);
3908out_put_eventfd:
3909 eventfd_ctx_put(event->eventfd);
3910out_put_efile:
3911 fdput(efile);
3912out_kfree:
3913 kfree(event);
3914
3915 return ret;
3916}
3917
Johannes Weiner241994e2015-02-11 15:26:06 -08003918static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003919 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003920 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003921 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003922 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003923 },
3924 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003925 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003926 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04003927 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003928 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003929 },
3930 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003931 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003932 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003933 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003934 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003935 },
3936 {
Balbir Singh296c81d2009-09-23 15:56:36 -07003937 .name = "soft_limit_in_bytes",
3938 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003939 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003940 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07003941 },
3942 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003943 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003944 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04003945 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003946 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003947 },
Balbir Singh8697d332008-02-07 00:13:59 -08003948 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003949 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003950 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003951 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003952 {
3953 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04003954 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003955 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08003956 {
3957 .name = "use_hierarchy",
3958 .write_u64 = mem_cgroup_hierarchy_write,
3959 .read_u64 = mem_cgroup_hierarchy_read,
3960 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003961 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05003962 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04003963 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04003964 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05003965 },
3966 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003967 .name = "swappiness",
3968 .read_u64 = mem_cgroup_swappiness_read,
3969 .write_u64 = mem_cgroup_swappiness_write,
3970 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003971 {
3972 .name = "move_charge_at_immigrate",
3973 .read_u64 = mem_cgroup_move_charge_read,
3974 .write_u64 = mem_cgroup_move_charge_write,
3975 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003976 {
3977 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003978 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003979 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003980 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3981 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003982 {
3983 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003984 },
Ying Han406eb0c2011-05-26 16:25:37 -07003985#ifdef CONFIG_NUMA
3986 {
3987 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003988 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07003989 },
3990#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08003991 {
3992 .name = "kmem.limit_in_bytes",
3993 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003994 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003995 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08003996 },
3997 {
3998 .name = "kmem.usage_in_bytes",
3999 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
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.failcnt",
4004 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004005 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004006 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004007 },
4008 {
4009 .name = "kmem.max_usage_in_bytes",
4010 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004011 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004012 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004013 },
Glauber Costa749c5412012-12-18 14:23:01 -08004014#ifdef CONFIG_SLABINFO
4015 {
4016 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004017 .seq_start = slab_start,
4018 .seq_next = slab_next,
4019 .seq_stop = slab_stop,
4020 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004021 },
4022#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004023 {
4024 .name = "kmem.tcp.limit_in_bytes",
4025 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4026 .write = mem_cgroup_write,
4027 .read_u64 = mem_cgroup_read_u64,
4028 },
4029 {
4030 .name = "kmem.tcp.usage_in_bytes",
4031 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4032 .read_u64 = mem_cgroup_read_u64,
4033 },
4034 {
4035 .name = "kmem.tcp.failcnt",
4036 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4037 .write = mem_cgroup_reset,
4038 .read_u64 = mem_cgroup_read_u64,
4039 },
4040 {
4041 .name = "kmem.tcp.max_usage_in_bytes",
4042 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4043 .write = mem_cgroup_reset,
4044 .read_u64 = mem_cgroup_read_u64,
4045 },
Tejun Heo6bc10342012-04-01 12:09:55 -07004046 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004047};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004048
Johannes Weiner73f576c2016-07-20 15:44:57 -07004049/*
4050 * Private memory cgroup IDR
4051 *
4052 * Swap-out records and page cache shadow entries need to store memcg
4053 * references in constrained space, so we maintain an ID space that is
4054 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4055 * memory-controlled cgroups to 64k.
4056 *
4057 * However, there usually are many references to the oflline CSS after
4058 * the cgroup has been destroyed, such as page cache or reclaimable
4059 * slab objects, that don't need to hang on to the ID. We want to keep
4060 * those dead CSS from occupying IDs, or we might quickly exhaust the
4061 * relatively small ID space and prevent the creation of new cgroups
4062 * even when there are much fewer than 64k cgroups - possibly none.
4063 *
4064 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4065 * be freed and recycled when it's no longer needed, which is usually
4066 * when the CSS is offlined.
4067 *
4068 * The only exception to that are records of swapped out tmpfs/shmem
4069 * pages that need to be attributed to live ancestors on swapin. But
4070 * those references are manageable from userspace.
4071 */
4072
4073static DEFINE_IDR(mem_cgroup_idr);
4074
Kirill Tkhaif108e462018-08-02 15:36:01 -07004075static void mem_cgroup_id_remove(struct mem_cgroup *memcg)
4076{
4077 if (memcg->id.id > 0) {
4078 idr_remove(&mem_cgroup_idr, memcg->id.id);
4079 memcg->id.id = 0;
4080 }
4081}
4082
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004083static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004084{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004085 VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0);
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004086 atomic_add(n, &memcg->id.ref);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004087}
4088
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004089static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004090{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004091 VM_BUG_ON(atomic_read(&memcg->id.ref) < n);
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004092 if (atomic_sub_and_test(n, &memcg->id.ref)) {
Kirill Tkhaif108e462018-08-02 15:36:01 -07004093 mem_cgroup_id_remove(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004094
4095 /* Memcg ID pins CSS */
4096 css_put(&memcg->css);
4097 }
4098}
4099
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004100static inline void mem_cgroup_id_get(struct mem_cgroup *memcg)
4101{
4102 mem_cgroup_id_get_many(memcg, 1);
4103}
4104
4105static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
4106{
4107 mem_cgroup_id_put_many(memcg, 1);
4108}
4109
Johannes Weiner73f576c2016-07-20 15:44:57 -07004110/**
4111 * mem_cgroup_from_id - look up a memcg from a memcg id
4112 * @id: the memcg id to look up
4113 *
4114 * Caller must hold rcu_read_lock().
4115 */
4116struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
4117{
4118 WARN_ON_ONCE(!rcu_read_lock_held());
4119 return idr_find(&mem_cgroup_idr, id);
4120}
4121
Mel Gormanef8f2322016-07-28 15:46:05 -07004122static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004123{
4124 struct mem_cgroup_per_node *pn;
Mel Gormanef8f2322016-07-28 15:46:05 -07004125 int tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004126 /*
4127 * This routine is called against possible nodes.
4128 * But it's BUG to call kmalloc() against offline node.
4129 *
4130 * TODO: this routine can waste much memory for nodes which will
4131 * never be onlined. It's better to use memory hotplug callback
4132 * function.
4133 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004134 if (!node_state(node, N_NORMAL_MEMORY))
4135 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004136 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004137 if (!pn)
4138 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004139
Mel Gormanef8f2322016-07-28 15:46:05 -07004140 lruvec_init(&pn->lruvec);
4141 pn->usage_in_excess = 0;
4142 pn->on_tree = false;
4143 pn->memcg = memcg;
4144
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004145 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004146 return 0;
4147}
4148
Mel Gormanef8f2322016-07-28 15:46:05 -07004149static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004150{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004151 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004152}
4153
Tahsin Erdogan16ace912017-03-09 16:17:26 -08004154static void __mem_cgroup_free(struct mem_cgroup *memcg)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004155{
4156 int node;
4157
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004158 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004159 free_mem_cgroup_per_node_info(memcg, node);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004160 free_percpu(memcg->stat);
4161 kfree(memcg);
4162}
4163
Tahsin Erdogan16ace912017-03-09 16:17:26 -08004164static void mem_cgroup_free(struct mem_cgroup *memcg)
4165{
4166 memcg_wb_domain_exit(memcg);
4167 __mem_cgroup_free(memcg);
4168}
4169
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004170static struct mem_cgroup *mem_cgroup_alloc(void)
4171{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004172 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004173 size_t size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004174 int node;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004175
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004176 size = sizeof(struct mem_cgroup);
4177 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004178
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004179 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004180 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004181 return NULL;
4182
Johannes Weiner73f576c2016-07-20 15:44:57 -07004183 memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
4184 1, MEM_CGROUP_ID_MAX,
4185 GFP_KERNEL);
4186 if (memcg->id.id < 0)
4187 goto fail;
4188
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004189 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4190 if (!memcg->stat)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004191 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004192
Bob Liu3ed28fa2012-01-12 17:19:04 -08004193 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004194 if (alloc_mem_cgroup_per_node_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004195 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004196
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004197 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4198 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004199
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004200 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08004201 memcg->last_scanned_node = MAX_NUMNODES;
4202 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004203 mutex_init(&memcg->thresholds_lock);
4204 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004205 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004206 INIT_LIST_HEAD(&memcg->event_list);
4207 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004208 memcg->socket_pressure = jiffies;
Johannes Weiner127424c2016-01-20 15:02:32 -08004209#ifndef CONFIG_SLOB
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004210 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004211#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004212#ifdef CONFIG_CGROUP_WRITEBACK
4213 INIT_LIST_HEAD(&memcg->cgwb_list);
4214#endif
Johannes Weiner73f576c2016-07-20 15:44:57 -07004215 idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004216 return memcg;
4217fail:
Kirill Tkhaif108e462018-08-02 15:36:01 -07004218 mem_cgroup_id_remove(memcg);
Tahsin Erdogan16ace912017-03-09 16:17:26 -08004219 __mem_cgroup_free(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004220 return NULL;
Glauber Costad142e3e2013-02-22 16:34:52 -08004221}
4222
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004223static struct cgroup_subsys_state * __ref
4224mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004225{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004226 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4227 struct mem_cgroup *memcg;
4228 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08004229
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004230 memcg = mem_cgroup_alloc();
4231 if (!memcg)
4232 return ERR_PTR(error);
Li Zefan4219b2d2013-09-23 16:56:29 +08004233
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004234 memcg->high = PAGE_COUNTER_MAX;
4235 memcg->soft_limit = PAGE_COUNTER_MAX;
4236 if (parent) {
4237 memcg->swappiness = mem_cgroup_swappiness(parent);
4238 memcg->oom_kill_disable = parent->oom_kill_disable;
4239 }
4240 if (parent && parent->use_hierarchy) {
4241 memcg->use_hierarchy = true;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004242 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004243 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004244 page_counter_init(&memcg->memsw, &parent->memsw);
4245 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08004246 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004247 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004248 page_counter_init(&memcg->memory, NULL);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004249 page_counter_init(&memcg->swap, NULL);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004250 page_counter_init(&memcg->memsw, NULL);
4251 page_counter_init(&memcg->kmem, NULL);
Johannes Weiner0db15292016-01-20 15:02:50 -08004252 page_counter_init(&memcg->tcpmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004253 /*
4254 * Deeper hierachy with use_hierarchy == false doesn't make
4255 * much sense so let cgroup subsystem know about this
4256 * unfortunate state in our controller.
4257 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004258 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004259 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004260 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08004261
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004262 /* The following stuff does not apply to the root */
4263 if (!parent) {
4264 root_mem_cgroup = memcg;
4265 return &memcg->css;
4266 }
4267
Vladimir Davydovb313aee2016-03-17 14:18:27 -07004268 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004269 if (error)
4270 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08004271
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004272 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004273 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004274
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004275 return &memcg->css;
4276fail:
Kirill Tkhaif108e462018-08-02 15:36:01 -07004277 mem_cgroup_id_remove(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004278 mem_cgroup_free(memcg);
Tejun Heoea3a9642016-06-24 14:49:58 -07004279 return ERR_PTR(-ENOMEM);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004280}
4281
Johannes Weiner73f576c2016-07-20 15:44:57 -07004282static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004283{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004284 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4285
Johannes Weiner73f576c2016-07-20 15:44:57 -07004286 /* Online state pins memcg ID, memcg ID pins CSS */
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004287 atomic_set(&memcg->id.ref, 1);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004288 css_get(css);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004289 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004290}
4291
Tejun Heoeb954192013-08-08 20:11:23 -04004292static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004293{
Tejun Heoeb954192013-08-08 20:11:23 -04004294 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004295 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004296
4297 /*
4298 * Unregister events and notify userspace.
4299 * Notify userspace about cgroup removing only after rmdir of cgroup
4300 * directory to avoid race between userspace and kernelspace.
4301 */
Tejun Heofba94802013-11-22 18:20:43 -05004302 spin_lock(&memcg->event_list_lock);
4303 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004304 list_del_init(&event->list);
4305 schedule_work(&event->remove);
4306 }
Tejun Heofba94802013-11-22 18:20:43 -05004307 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004308
Johannes Weiner567e9ab2016-01-20 15:02:24 -08004309 memcg_offline_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004310 wb_memcg_offline(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004311
4312 mem_cgroup_id_put(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004313}
4314
Vladimir Davydov6df38682015-12-29 14:54:10 -08004315static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4316{
4317 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4318
4319 invalidate_reclaim_iterators(memcg);
4320}
4321
Tejun Heoeb954192013-08-08 20:11:23 -04004322static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004323{
Tejun Heoeb954192013-08-08 20:11:23 -04004324 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004325
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004326 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004327 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004328
Johannes Weiner0db15292016-01-20 15:02:50 -08004329 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004330 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004331
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004332 vmpressure_cleanup(&memcg->vmpressure);
4333 cancel_work_sync(&memcg->high_work);
4334 mem_cgroup_remove_from_trees(memcg);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004335 memcg_free_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004336 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004337}
4338
Tejun Heo1ced9532014-07-08 18:02:57 -04004339/**
4340 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4341 * @css: the target css
4342 *
4343 * Reset the states of the mem_cgroup associated with @css. This is
4344 * invoked when the userland requests disabling on the default hierarchy
4345 * but the memcg is pinned through dependency. The memcg should stop
4346 * applying policies and should revert to the vanilla state as it may be
4347 * made visible again.
4348 *
4349 * The current implementation only resets the essential configurations.
4350 * This needs to be expanded to cover all the visible parts.
4351 */
4352static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4353{
4354 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4355
Vladimir Davydovd334c9b2016-03-17 14:19:38 -07004356 page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4357 page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4358 page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4359 page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4360 page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
Johannes Weiner241994e2015-02-11 15:26:06 -08004361 memcg->low = 0;
4362 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004363 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004364 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004365}
4366
Daisuke Nishimura02491442010-03-10 15:22:17 -08004367#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004368/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004369static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004370{
Johannes Weiner05b84302014-08-06 16:05:59 -07004371 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004372
Mel Gormand0164ad2015-11-06 16:28:21 -08004373 /* Try a single bulk charge without reclaim first, kswapd may wake */
4374 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004375 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004376 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004377 return ret;
4378 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004379
David Rientjesf5f415c2017-01-24 15:18:10 -08004380 /* Try charges one by one with reclaim, but do not retry */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004381 while (count--) {
David Rientjesf5f415c2017-01-24 15:18:10 -08004382 ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004383 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004384 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004385 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004386 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004387 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004388 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004389}
4390
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004391union mc_target {
4392 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004393 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004394};
4395
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004396enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004397 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004398 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004399 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004400};
4401
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004402static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4403 unsigned long addr, pte_t ptent)
4404{
4405 struct page *page = vm_normal_page(vma, addr, ptent);
4406
4407 if (!page || !page_mapped(page))
4408 return NULL;
4409 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004410 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004411 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004412 } else {
4413 if (!(mc.flags & MOVE_FILE))
4414 return NULL;
4415 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004416 if (!get_page_unless_zero(page))
4417 return NULL;
4418
4419 return page;
4420}
4421
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004422#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004423static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004424 pte_t ptent, swp_entry_t *entry)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004425{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004426 struct page *page = NULL;
4427 swp_entry_t ent = pte_to_swp_entry(ptent);
4428
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004429 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004430 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004431 /*
4432 * Because lookup_swap_cache() updates some statistics counter,
4433 * we call find_get_page() with swapper_space directly.
4434 */
Huang Yingf6ab1f72016-10-07 17:00:21 -07004435 page = find_get_page(swap_address_space(ent), swp_offset(ent));
Johannes Weiner7941d212016-01-14 15:21:23 -08004436 if (do_memsw_account())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004437 entry->val = ent.val;
4438
4439 return page;
4440}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004441#else
4442static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004443 pte_t ptent, swp_entry_t *entry)
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004444{
4445 return NULL;
4446}
4447#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004448
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004449static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4450 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4451{
4452 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004453 struct address_space *mapping;
4454 pgoff_t pgoff;
4455
4456 if (!vma->vm_file) /* anonymous vma */
4457 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004458 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004459 return NULL;
4460
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004461 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004462 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004463
4464 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004465#ifdef CONFIG_SWAP
4466 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004467 if (shmem_mapping(mapping)) {
4468 page = find_get_entry(mapping, pgoff);
4469 if (radix_tree_exceptional_entry(page)) {
4470 swp_entry_t swp = radix_to_swp_entry(page);
Johannes Weiner7941d212016-01-14 15:21:23 -08004471 if (do_memsw_account())
Johannes Weiner139b6a62014-05-06 12:50:05 -07004472 *entry = swp;
Huang Yingf6ab1f72016-10-07 17:00:21 -07004473 page = find_get_page(swap_address_space(swp),
4474 swp_offset(swp));
Johannes Weiner139b6a62014-05-06 12:50:05 -07004475 }
4476 } else
4477 page = find_get_page(mapping, pgoff);
4478#else
4479 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004480#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004481 return page;
4482}
4483
Chen Gangb1b0dea2015-04-14 15:47:35 -07004484/**
4485 * mem_cgroup_move_account - move account of the page
4486 * @page: the page
Li RongQing25843c22016-07-26 15:26:56 -07004487 * @compound: charge the page as compound or small page
Chen Gangb1b0dea2015-04-14 15:47:35 -07004488 * @from: mem_cgroup which the page is moved from.
4489 * @to: mem_cgroup which the page is moved to. @from != @to.
4490 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08004491 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07004492 *
4493 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4494 * from old cgroup.
4495 */
4496static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004497 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07004498 struct mem_cgroup *from,
4499 struct mem_cgroup *to)
4500{
4501 unsigned long flags;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004502 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004503 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004504 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004505
4506 VM_BUG_ON(from == to);
4507 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004508 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07004509
4510 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07004511 * Prevent mem_cgroup_migrate() from looking at
Hugh Dickins45637ba2015-11-05 18:49:40 -08004512 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07004513 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004514 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004515 if (!trylock_page(page))
4516 goto out;
4517
4518 ret = -EINVAL;
4519 if (page->mem_cgroup != from)
4520 goto out_unlock;
4521
Greg Thelenc4843a72015-05-22 17:13:16 -04004522 anon = PageAnon(page);
4523
Chen Gangb1b0dea2015-04-14 15:47:35 -07004524 spin_lock_irqsave(&from->move_lock, flags);
4525
Greg Thelenc4843a72015-05-22 17:13:16 -04004526 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004527 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4528 nr_pages);
4529 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4530 nr_pages);
4531 }
4532
Greg Thelenc4843a72015-05-22 17:13:16 -04004533 /*
4534 * move_lock grabbed above and caller set from->moving_account, so
4535 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4536 * So mapping should be stable for dirty pages.
4537 */
4538 if (!anon && PageDirty(page)) {
4539 struct address_space *mapping = page_mapping(page);
4540
4541 if (mapping_cap_account_dirty(mapping)) {
4542 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4543 nr_pages);
4544 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4545 nr_pages);
4546 }
4547 }
4548
Chen Gangb1b0dea2015-04-14 15:47:35 -07004549 if (PageWriteback(page)) {
4550 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4551 nr_pages);
4552 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4553 nr_pages);
4554 }
4555
4556 /*
4557 * It is safe to change page->mem_cgroup here because the page
4558 * is referenced, charged, and isolated - we can't race with
4559 * uncharging, charging, migration, or LRU putback.
4560 */
4561
4562 /* caller should have done css_get */
4563 page->mem_cgroup = to;
4564 spin_unlock_irqrestore(&from->move_lock, flags);
4565
4566 ret = 0;
4567
4568 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004569 mem_cgroup_charge_statistics(to, page, compound, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004570 memcg_check_events(to, page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004571 mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004572 memcg_check_events(from, page);
4573 local_irq_enable();
4574out_unlock:
4575 unlock_page(page);
4576out:
4577 return ret;
4578}
4579
Li RongQing7cf78062016-05-27 14:27:46 -07004580/**
4581 * get_mctgt_type - get target type of moving charge
4582 * @vma: the vma the pte to be checked belongs
4583 * @addr: the address corresponding to the pte to be checked
4584 * @ptent: the pte to be checked
4585 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4586 *
4587 * Returns
4588 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4589 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4590 * move charge. if @target is not NULL, the page is stored in target->page
4591 * with extra refcnt got(Callers should handle it).
4592 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4593 * target for charge migration. if @target is not NULL, the entry is stored
4594 * in target->ent.
4595 *
4596 * Called with pte lock held.
4597 */
4598
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004599static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004600 unsigned long addr, pte_t ptent, union mc_target *target)
4601{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004602 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004603 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004604 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004605
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004606 if (pte_present(ptent))
4607 page = mc_handle_present_pte(vma, addr, ptent);
4608 else if (is_swap_pte(ptent))
Li RongQing48406ef2016-07-26 15:22:14 -07004609 page = mc_handle_swap_pte(vma, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004610 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004611 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004612
4613 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004614 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004615 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004616 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004617 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004618 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004619 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004620 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004621 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004622 ret = MC_TARGET_PAGE;
4623 if (target)
4624 target->page = page;
4625 }
4626 if (!ret || !target)
4627 put_page(page);
4628 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004629 /* There is a swap entry and a page doesn't exist or isn't charged */
4630 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004631 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004632 ret = MC_TARGET_SWAP;
4633 if (target)
4634 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004635 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004636 return ret;
4637}
4638
Naoya Horiguchi12724852012-03-21 16:34:28 -07004639#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4640/*
4641 * We don't consider swapping or file mapped pages because THP does not
4642 * support them for now.
4643 * Caller should make sure that pmd_trans_huge(pmd) is true.
4644 */
4645static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4646 unsigned long addr, pmd_t pmd, union mc_target *target)
4647{
4648 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004649 enum mc_target_type ret = MC_TARGET_NONE;
4650
4651 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004652 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004653 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004654 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004655 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004656 ret = MC_TARGET_PAGE;
4657 if (target) {
4658 get_page(page);
4659 target->page = page;
4660 }
4661 }
4662 return ret;
4663}
4664#else
4665static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4666 unsigned long addr, pmd_t pmd, union mc_target *target)
4667{
4668 return MC_TARGET_NONE;
4669}
4670#endif
4671
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004672static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4673 unsigned long addr, unsigned long end,
4674 struct mm_walk *walk)
4675{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004676 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004677 pte_t *pte;
4678 spinlock_t *ptl;
4679
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004680 ptl = pmd_trans_huge_lock(pmd, vma);
4681 if (ptl) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004682 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4683 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004684 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004685 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004686 }
Dave Hansen03319322011-03-22 16:32:56 -07004687
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004688 if (pmd_trans_unstable(pmd))
4689 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004690 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4691 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004692 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004693 mc.precharge++; /* increment precharge temporarily */
4694 pte_unmap_unlock(pte - 1, ptl);
4695 cond_resched();
4696
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004697 return 0;
4698}
4699
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004700static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4701{
4702 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004703
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004704 struct mm_walk mem_cgroup_count_precharge_walk = {
4705 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4706 .mm = mm,
4707 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004708 down_read(&mm->mmap_sem);
James Morse0247f3f2016-10-07 17:00:12 -07004709 walk_page_range(0, mm->highest_vm_end,
4710 &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004711 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004712
4713 precharge = mc.precharge;
4714 mc.precharge = 0;
4715
4716 return precharge;
4717}
4718
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004719static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4720{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004721 unsigned long precharge = mem_cgroup_count_precharge(mm);
4722
4723 VM_BUG_ON(mc.moving_task);
4724 mc.moving_task = current;
4725 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004726}
4727
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004728/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4729static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004730{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004731 struct mem_cgroup *from = mc.from;
4732 struct mem_cgroup *to = mc.to;
4733
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004734 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004735 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004736 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004737 mc.precharge = 0;
4738 }
4739 /*
4740 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4741 * we must uncharge here.
4742 */
4743 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004744 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004745 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004746 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004747 /* we must fixup refcnts and charges */
4748 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004749 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004750 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004751 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004752
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004753 mem_cgroup_id_put_many(mc.from, mc.moved_swap);
4754
Johannes Weiner05b84302014-08-06 16:05:59 -07004755 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004756 * we charged both to->memory and to->memsw, so we
4757 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004758 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004759 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004760 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004761
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004762 mem_cgroup_id_get_many(mc.to, mc.moved_swap);
4763 css_put_many(&mc.to->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004764
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004765 mc.moved_swap = 0;
4766 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004767 memcg_oom_recover(from);
4768 memcg_oom_recover(to);
4769 wake_up_all(&mc.waitq);
4770}
4771
4772static void mem_cgroup_clear_mc(void)
4773{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004774 struct mm_struct *mm = mc.mm;
4775
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004776 /*
4777 * we must clear moving_task before waking up waiters at the end of
4778 * task migration.
4779 */
4780 mc.moving_task = NULL;
4781 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004782 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004783 mc.from = NULL;
4784 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04004785 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004786 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004787
4788 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004789}
4790
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004791static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004792{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004793 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07004794 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07004795 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04004796 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004797 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004798 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004799 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004800
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004801 /* charge immigration isn't supported on the default hierarchy */
4802 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07004803 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004804
Tejun Heo4530edd2015-09-11 15:00:19 -04004805 /*
4806 * Multi-process migrations only happen on the default hierarchy
4807 * where charge immigration is not used. Perform charge
4808 * immigration if @tset contains a leader and whine if there are
4809 * multiple.
4810 */
4811 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004812 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04004813 WARN_ON_ONCE(p);
4814 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004815 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04004816 }
4817 if (!p)
4818 return 0;
4819
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004820 /*
4821 * We are now commited to this value whatever it is. Changes in this
4822 * tunable will only affect upcoming migrations, not the current one.
4823 * So we need to save it, and keep it going.
4824 */
4825 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4826 if (!move_flags)
4827 return 0;
4828
Tejun Heo9f2115f2015-09-08 15:01:10 -07004829 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004830
Tejun Heo9f2115f2015-09-08 15:01:10 -07004831 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004832
Tejun Heo9f2115f2015-09-08 15:01:10 -07004833 mm = get_task_mm(p);
4834 if (!mm)
4835 return 0;
4836 /* We move charges only when we move a owner of the mm */
4837 if (mm->owner == p) {
4838 VM_BUG_ON(mc.from);
4839 VM_BUG_ON(mc.to);
4840 VM_BUG_ON(mc.precharge);
4841 VM_BUG_ON(mc.moved_charge);
4842 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004843
Tejun Heo9f2115f2015-09-08 15:01:10 -07004844 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004845 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004846 mc.from = from;
4847 mc.to = memcg;
4848 mc.flags = move_flags;
4849 spin_unlock(&mc.lock);
4850 /* We set mc.moving_task later */
4851
4852 ret = mem_cgroup_precharge_mc(mm);
4853 if (ret)
4854 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004855 } else {
4856 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004857 }
4858 return ret;
4859}
4860
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004861static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004862{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004863 if (mc.to)
4864 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004865}
4866
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004867static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4868 unsigned long addr, unsigned long end,
4869 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004870{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004871 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004872 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004873 pte_t *pte;
4874 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004875 enum mc_target_type target_type;
4876 union mc_target target;
4877 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004878
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004879 ptl = pmd_trans_huge_lock(pmd, vma);
4880 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004881 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004882 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004883 return 0;
4884 }
4885 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4886 if (target_type == MC_TARGET_PAGE) {
4887 page = target.page;
4888 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004889 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08004890 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004891 mc.precharge -= HPAGE_PMD_NR;
4892 mc.moved_charge += HPAGE_PMD_NR;
4893 }
4894 putback_lru_page(page);
4895 }
4896 put_page(page);
4897 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004898 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004899 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004900 }
4901
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004902 if (pmd_trans_unstable(pmd))
4903 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004904retry:
4905 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4906 for (; addr != end; addr += PAGE_SIZE) {
4907 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004908 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004909
4910 if (!mc.precharge)
4911 break;
4912
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004913 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004914 case MC_TARGET_PAGE:
4915 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08004916 /*
4917 * We can have a part of the split pmd here. Moving it
4918 * can be done but it would be too convoluted so simply
4919 * ignore such a partial THP and keep it in original
4920 * memcg. There should be somebody mapping the head.
4921 */
4922 if (PageTransCompound(page))
4923 goto put;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004924 if (isolate_lru_page(page))
4925 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004926 if (!mem_cgroup_move_account(page, false,
4927 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004928 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004929 /* we uncharge from mc.from later. */
4930 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004931 }
4932 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004933put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004934 put_page(page);
4935 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004936 case MC_TARGET_SWAP:
4937 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004938 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004939 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004940 /* we fixup refcnts and charges later. */
4941 mc.moved_swap++;
4942 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004943 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004944 default:
4945 break;
4946 }
4947 }
4948 pte_unmap_unlock(pte - 1, ptl);
4949 cond_resched();
4950
4951 if (addr != end) {
4952 /*
4953 * We have consumed all precharges we got in can_attach().
4954 * We try charge one by one, but don't do any additional
4955 * charges to mc.to if we have failed in charge once in attach()
4956 * phase.
4957 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004958 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004959 if (!ret)
4960 goto retry;
4961 }
4962
4963 return ret;
4964}
4965
Tejun Heo264a0ae2016-04-21 19:09:02 -04004966static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004967{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004968 struct mm_walk mem_cgroup_move_charge_walk = {
4969 .pmd_entry = mem_cgroup_move_charge_pte_range,
Tejun Heo264a0ae2016-04-21 19:09:02 -04004970 .mm = mc.mm,
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004971 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004972
4973 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004974 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07004975 * Signal lock_page_memcg() to take the memcg's move_lock
4976 * while we're moving its pages to another memcg. Then wait
4977 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08004978 */
4979 atomic_inc(&mc.from->moving_account);
4980 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004981retry:
Tejun Heo264a0ae2016-04-21 19:09:02 -04004982 if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004983 /*
4984 * Someone who are holding the mmap_sem might be waiting in
4985 * waitq. So we cancel all extra charges, wake up all waiters,
4986 * and retry. Because we cancel precharges, we might not be able
4987 * to move enough charges, but moving charge is a best-effort
4988 * feature anyway, so it wouldn't be a big problem.
4989 */
4990 __mem_cgroup_clear_mc();
4991 cond_resched();
4992 goto retry;
4993 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004994 /*
4995 * When we have consumed all precharges and failed in doing
4996 * additional charge, the page walk just aborts.
4997 */
James Morse0247f3f2016-10-07 17:00:12 -07004998 walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk);
4999
Tejun Heo264a0ae2016-04-21 19:09:02 -04005000 up_read(&mc.mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08005001 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005002}
5003
Tejun Heo264a0ae2016-04-21 19:09:02 -04005004static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08005005{
Tejun Heo264a0ae2016-04-21 19:09:02 -04005006 if (mc.to) {
5007 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005008 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04005009 }
Balbir Singh67e465a2008-02-07 00:13:54 -08005010}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005011#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005012static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005013{
5014 return 0;
5015}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005016static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005017{
5018}
Tejun Heo264a0ae2016-04-21 19:09:02 -04005019static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005020{
5021}
5022#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005023
Tejun Heof00baae2013-04-15 13:41:15 -07005024/*
5025 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005026 * to verify whether we're attached to the default hierarchy on each mount
5027 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005028 */
Tejun Heoeb954192013-08-08 20:11:23 -04005029static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005030{
5031 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005032 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005033 * guarantees that @root doesn't have any children, so turning it
5034 * on for the root memcg is enough.
5035 */
Tejun Heo9e10a132015-09-18 11:56:28 -04005036 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee592015-03-12 16:26:19 -07005037 root_mem_cgroup->use_hierarchy = true;
5038 else
5039 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005040}
5041
Johannes Weiner241994e2015-02-11 15:26:06 -08005042static u64 memory_current_read(struct cgroup_subsys_state *css,
5043 struct cftype *cft)
5044{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005045 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5046
5047 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994e2015-02-11 15:26:06 -08005048}
5049
5050static int memory_low_show(struct seq_file *m, void *v)
5051{
5052 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005053 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994e2015-02-11 15:26:06 -08005054
5055 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005056 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005057 else
5058 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5059
5060 return 0;
5061}
5062
5063static ssize_t memory_low_write(struct kernfs_open_file *of,
5064 char *buf, size_t nbytes, loff_t off)
5065{
5066 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5067 unsigned long low;
5068 int err;
5069
5070 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005071 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994e2015-02-11 15:26:06 -08005072 if (err)
5073 return err;
5074
5075 memcg->low = low;
5076
5077 return nbytes;
5078}
5079
5080static int memory_high_show(struct seq_file *m, void *v)
5081{
5082 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005083 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994e2015-02-11 15:26:06 -08005084
5085 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005086 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005087 else
5088 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5089
5090 return 0;
5091}
5092
5093static ssize_t memory_high_write(struct kernfs_open_file *of,
5094 char *buf, size_t nbytes, loff_t off)
5095{
5096 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner588083b2016-03-17 14:20:25 -07005097 unsigned long nr_pages;
Johannes Weiner241994e2015-02-11 15:26:06 -08005098 unsigned long high;
5099 int err;
5100
5101 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005102 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994e2015-02-11 15:26:06 -08005103 if (err)
5104 return err;
5105
5106 memcg->high = high;
5107
Johannes Weiner588083b2016-03-17 14:20:25 -07005108 nr_pages = page_counter_read(&memcg->memory);
5109 if (nr_pages > high)
5110 try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5111 GFP_KERNEL, true);
5112
Tejun Heo2529bb32015-05-22 18:23:34 -04005113 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005114 return nbytes;
5115}
5116
5117static int memory_max_show(struct seq_file *m, void *v)
5118{
5119 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005120 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994e2015-02-11 15:26:06 -08005121
5122 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005123 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005124 else
5125 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5126
5127 return 0;
5128}
5129
5130static ssize_t memory_max_write(struct kernfs_open_file *of,
5131 char *buf, size_t nbytes, loff_t off)
5132{
5133 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005134 unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5135 bool drained = false;
Johannes Weiner241994e2015-02-11 15:26:06 -08005136 unsigned long max;
5137 int err;
5138
5139 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005140 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994e2015-02-11 15:26:06 -08005141 if (err)
5142 return err;
5143
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005144 xchg(&memcg->memory.limit, max);
5145
5146 for (;;) {
5147 unsigned long nr_pages = page_counter_read(&memcg->memory);
5148
5149 if (nr_pages <= max)
5150 break;
5151
5152 if (signal_pending(current)) {
5153 err = -EINTR;
5154 break;
5155 }
5156
5157 if (!drained) {
5158 drain_all_stock(memcg);
5159 drained = true;
5160 continue;
5161 }
5162
5163 if (nr_reclaims) {
5164 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5165 GFP_KERNEL, true))
5166 nr_reclaims--;
5167 continue;
5168 }
5169
5170 mem_cgroup_events(memcg, MEMCG_OOM, 1);
5171 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5172 break;
5173 }
Johannes Weiner241994e2015-02-11 15:26:06 -08005174
Tejun Heo2529bb32015-05-22 18:23:34 -04005175 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005176 return nbytes;
5177}
5178
5179static int memory_events_show(struct seq_file *m, void *v)
5180{
5181 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5182
5183 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5184 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5185 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5186 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5187
5188 return 0;
5189}
5190
Johannes Weiner587d9f72016-01-20 15:03:19 -08005191static int memory_stat_show(struct seq_file *m, void *v)
5192{
5193 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005194 unsigned long stat[MEMCG_NR_STAT];
5195 unsigned long events[MEMCG_NR_EVENTS];
Johannes Weiner587d9f72016-01-20 15:03:19 -08005196 int i;
5197
5198 /*
5199 * Provide statistics on the state of the memory subsystem as
5200 * well as cumulative event counters that show past behavior.
5201 *
5202 * This list is ordered following a combination of these gradients:
5203 * 1) generic big picture -> specifics and details
5204 * 2) reflecting userspace activity -> reflecting kernel heuristics
5205 *
5206 * Current memory state:
5207 */
5208
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005209 tree_stat(memcg, stat);
5210 tree_events(memcg, events);
5211
Johannes Weiner587d9f72016-01-20 15:03:19 -08005212 seq_printf(m, "anon %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005213 (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005214 seq_printf(m, "file %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005215 (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE);
Vladimir Davydov12580e42016-03-17 14:17:38 -07005216 seq_printf(m, "kernel_stack %llu\n",
Andy Lutomirskiefdc9492016-07-28 15:48:17 -07005217 (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005218 seq_printf(m, "slab %llu\n",
5219 (u64)(stat[MEMCG_SLAB_RECLAIMABLE] +
5220 stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
Johannes Weinerb2807f02016-01-20 15:03:22 -08005221 seq_printf(m, "sock %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005222 (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005223
5224 seq_printf(m, "file_mapped %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005225 (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005226 seq_printf(m, "file_dirty %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005227 (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005228 seq_printf(m, "file_writeback %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005229 (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005230
5231 for (i = 0; i < NR_LRU_LISTS; i++) {
5232 struct mem_cgroup *mi;
5233 unsigned long val = 0;
5234
5235 for_each_mem_cgroup_tree(mi, memcg)
5236 val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5237 seq_printf(m, "%s %llu\n",
5238 mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5239 }
5240
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005241 seq_printf(m, "slab_reclaimable %llu\n",
5242 (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE);
5243 seq_printf(m, "slab_unreclaimable %llu\n",
5244 (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5245
Johannes Weiner587d9f72016-01-20 15:03:19 -08005246 /* Accumulated memory events */
5247
5248 seq_printf(m, "pgfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005249 events[MEM_CGROUP_EVENTS_PGFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005250 seq_printf(m, "pgmajfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005251 events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005252
5253 return 0;
5254}
5255
Johannes Weiner241994e2015-02-11 15:26:06 -08005256static struct cftype memory_files[] = {
5257 {
5258 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005259 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994e2015-02-11 15:26:06 -08005260 .read_u64 = memory_current_read,
5261 },
5262 {
5263 .name = "low",
5264 .flags = CFTYPE_NOT_ON_ROOT,
5265 .seq_show = memory_low_show,
5266 .write = memory_low_write,
5267 },
5268 {
5269 .name = "high",
5270 .flags = CFTYPE_NOT_ON_ROOT,
5271 .seq_show = memory_high_show,
5272 .write = memory_high_write,
5273 },
5274 {
5275 .name = "max",
5276 .flags = CFTYPE_NOT_ON_ROOT,
5277 .seq_show = memory_max_show,
5278 .write = memory_max_write,
5279 },
5280 {
5281 .name = "events",
5282 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04005283 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994e2015-02-11 15:26:06 -08005284 .seq_show = memory_events_show,
5285 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08005286 {
5287 .name = "stat",
5288 .flags = CFTYPE_NOT_ON_ROOT,
5289 .seq_show = memory_stat_show,
5290 },
Johannes Weiner241994e2015-02-11 15:26:06 -08005291 { } /* terminate */
5292};
5293
Tejun Heo073219e2014-02-08 10:36:58 -05005294struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005295 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005296 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005297 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08005298 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08005299 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005300 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005301 .can_attach = mem_cgroup_can_attach,
5302 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04005303 .post_attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005304 .bind = mem_cgroup_bind,
Johannes Weiner241994e2015-02-11 15:26:06 -08005305 .dfl_cftypes = memory_files,
5306 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005307 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005308};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005309
Johannes Weiner241994e2015-02-11 15:26:06 -08005310/**
Johannes Weiner241994e2015-02-11 15:26:06 -08005311 * mem_cgroup_low - check if memory consumption is below the normal range
5312 * @root: the highest ancestor to consider
5313 * @memcg: the memory cgroup to check
5314 *
5315 * Returns %true if memory consumption of @memcg, and that of all
5316 * configurable ancestors up to @root, is below the normal range.
5317 */
5318bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5319{
5320 if (mem_cgroup_disabled())
5321 return false;
5322
5323 /*
5324 * The toplevel group doesn't have a configurable range, so
5325 * it's never low when looked at directly, and it is not
5326 * considered an ancestor when assessing the hierarchy.
5327 */
5328
5329 if (memcg == root_mem_cgroup)
5330 return false;
5331
Michal Hocko4e54ded2015-02-27 15:51:46 -08005332 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005333 return false;
5334
5335 while (memcg != root) {
5336 memcg = parent_mem_cgroup(memcg);
5337
5338 if (memcg == root_mem_cgroup)
5339 break;
5340
Michal Hocko4e54ded2015-02-27 15:51:46 -08005341 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005342 return false;
5343 }
5344 return true;
5345}
5346
Johannes Weiner00501b52014-08-08 14:19:20 -07005347/**
5348 * mem_cgroup_try_charge - try charging a page
5349 * @page: page to charge
5350 * @mm: mm context of the victim
5351 * @gfp_mask: reclaim mode
5352 * @memcgp: charged memcg return
Li RongQing25843c22016-07-26 15:26:56 -07005353 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005354 *
5355 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5356 * pages according to @gfp_mask if necessary.
5357 *
5358 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5359 * Otherwise, an error code is returned.
5360 *
5361 * After page->mapping has been set up, the caller must finalize the
5362 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5363 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5364 */
5365int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005366 gfp_t gfp_mask, struct mem_cgroup **memcgp,
5367 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005368{
5369 struct mem_cgroup *memcg = NULL;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005370 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005371 int ret = 0;
5372
5373 if (mem_cgroup_disabled())
5374 goto out;
5375
5376 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005377 /*
5378 * Every swap fault against a single page tries to charge the
5379 * page, bail as early as possible. shmem_unuse() encounters
5380 * already charged pages, too. The USED bit is protected by
5381 * the page lock, which serializes swap cache removal, which
5382 * in turn serializes uncharging.
5383 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005384 VM_BUG_ON_PAGE(!PageLocked(page), page);
Johannes Weiner1306a852014-12-10 15:44:52 -08005385 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005386 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005387
Vladimir Davydov37e84352016-01-20 15:02:56 -08005388 if (do_swap_account) {
Vladimir Davydove993d902015-09-09 15:35:35 -07005389 swp_entry_t ent = { .val = page_private(page), };
5390 unsigned short id = lookup_swap_cgroup_id(ent);
5391
5392 rcu_read_lock();
5393 memcg = mem_cgroup_from_id(id);
5394 if (memcg && !css_tryget_online(&memcg->css))
5395 memcg = NULL;
5396 rcu_read_unlock();
5397 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005398 }
5399
Johannes Weiner00501b52014-08-08 14:19:20 -07005400 if (!memcg)
5401 memcg = get_mem_cgroup_from_mm(mm);
5402
5403 ret = try_charge(memcg, gfp_mask, nr_pages);
5404
5405 css_put(&memcg->css);
Johannes Weiner00501b52014-08-08 14:19:20 -07005406out:
5407 *memcgp = memcg;
5408 return ret;
5409}
5410
5411/**
5412 * mem_cgroup_commit_charge - commit a page charge
5413 * @page: page to charge
5414 * @memcg: memcg to charge the page to
5415 * @lrucare: page might be on LRU already
Li RongQing25843c22016-07-26 15:26:56 -07005416 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005417 *
5418 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5419 * after page->mapping has been set up. This must happen atomically
5420 * as part of the page instantiation, i.e. under the page table lock
5421 * for anonymous pages, under the page lock for page and swap cache.
5422 *
5423 * In addition, the page must not be on the LRU during the commit, to
5424 * prevent racing with task migration. If it might be, use @lrucare.
5425 *
5426 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5427 */
5428void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005429 bool lrucare, bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005430{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005431 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005432
5433 VM_BUG_ON_PAGE(!page->mapping, page);
5434 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5435
5436 if (mem_cgroup_disabled())
5437 return;
5438 /*
5439 * Swap faults will attempt to charge the same page multiple
5440 * times. But reuse_swap_page() might have removed the page
5441 * from swapcache already, so we can't check PageSwapCache().
5442 */
5443 if (!memcg)
5444 return;
5445
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005446 commit_charge(page, memcg, lrucare);
5447
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005448 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005449 mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005450 memcg_check_events(memcg, page);
5451 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005452
Johannes Weiner7941d212016-01-14 15:21:23 -08005453 if (do_memsw_account() && PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005454 swp_entry_t entry = { .val = page_private(page) };
5455 /*
5456 * The swap entry might not get freed for a long time,
5457 * let's not wait for it. The page already received a
5458 * memory+swap charge, drop the swap entry duplicate.
5459 */
5460 mem_cgroup_uncharge_swap(entry);
5461 }
5462}
5463
5464/**
5465 * mem_cgroup_cancel_charge - cancel a page charge
5466 * @page: page to charge
5467 * @memcg: memcg to charge the page to
Li RongQing25843c22016-07-26 15:26:56 -07005468 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005469 *
5470 * Cancel a charge transaction started by mem_cgroup_try_charge().
5471 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005472void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5473 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005474{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005475 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005476
5477 if (mem_cgroup_disabled())
5478 return;
5479 /*
5480 * Swap faults will attempt to charge the same page multiple
5481 * times. But reuse_swap_page() might have removed the page
5482 * from swapcache already, so we can't check PageSwapCache().
5483 */
5484 if (!memcg)
5485 return;
5486
Johannes Weiner00501b52014-08-08 14:19:20 -07005487 cancel_charge(memcg, nr_pages);
5488}
5489
Johannes Weiner747db952014-08-08 14:19:24 -07005490static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005491 unsigned long nr_anon, unsigned long nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005492 unsigned long nr_huge, unsigned long nr_kmem,
5493 struct page *dummy_page)
Johannes Weiner747db952014-08-08 14:19:24 -07005494{
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005495 unsigned long nr_pages = nr_anon + nr_file + nr_kmem;
Johannes Weiner747db952014-08-08 14:19:24 -07005496 unsigned long flags;
5497
Johannes Weinerce00a962014-09-05 08:43:57 -04005498 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005499 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08005500 if (do_memsw_account())
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005501 page_counter_uncharge(&memcg->memsw, nr_pages);
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005502 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem)
5503 page_counter_uncharge(&memcg->kmem, nr_kmem);
Johannes Weinerce00a962014-09-05 08:43:57 -04005504 memcg_oom_recover(memcg);
5505 }
Johannes Weiner747db952014-08-08 14:19:24 -07005506
5507 local_irq_save(flags);
5508 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5509 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5510 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5511 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005512 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005513 memcg_check_events(memcg, dummy_page);
5514 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005515
5516 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005517 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005518}
5519
5520static void uncharge_list(struct list_head *page_list)
5521{
5522 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005523 unsigned long nr_anon = 0;
5524 unsigned long nr_file = 0;
5525 unsigned long nr_huge = 0;
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005526 unsigned long nr_kmem = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005527 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005528 struct list_head *next;
5529 struct page *page;
5530
Johannes Weiner8b592652016-03-17 14:20:31 -07005531 /*
5532 * Note that the list can be a single page->lru; hence the
5533 * do-while loop instead of a simple list_for_each_entry().
5534 */
Johannes Weiner747db952014-08-08 14:19:24 -07005535 next = page_list->next;
5536 do {
Johannes Weiner747db952014-08-08 14:19:24 -07005537 page = list_entry(next, struct page, lru);
5538 next = page->lru.next;
5539
5540 VM_BUG_ON_PAGE(PageLRU(page), page);
Michal Hockobc0e2172017-05-12 15:46:26 -07005541 VM_BUG_ON_PAGE(!PageHWPoison(page) && page_count(page), page);
Johannes Weiner747db952014-08-08 14:19:24 -07005542
Johannes Weiner1306a852014-12-10 15:44:52 -08005543 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005544 continue;
5545
5546 /*
5547 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005548 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005549 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005550 */
5551
Johannes Weiner1306a852014-12-10 15:44:52 -08005552 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005553 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005554 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005555 nr_huge, nr_kmem, page);
5556 pgpgout = nr_anon = nr_file =
5557 nr_huge = nr_kmem = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005558 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005559 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005560 }
5561
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005562 if (!PageKmemcg(page)) {
5563 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005564
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005565 if (PageTransHuge(page)) {
5566 nr_pages <<= compound_order(page);
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005567 nr_huge += nr_pages;
5568 }
5569 if (PageAnon(page))
5570 nr_anon += nr_pages;
5571 else
5572 nr_file += nr_pages;
5573 pgpgout++;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03005574 } else {
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005575 nr_kmem += 1 << compound_order(page);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03005576 __ClearPageKmemcg(page);
5577 }
Johannes Weiner747db952014-08-08 14:19:24 -07005578
Johannes Weiner1306a852014-12-10 15:44:52 -08005579 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005580 } while (next != page_list);
5581
5582 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005583 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005584 nr_huge, nr_kmem, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005585}
5586
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005587/**
5588 * mem_cgroup_uncharge - uncharge a page
5589 * @page: page to uncharge
5590 *
5591 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5592 * mem_cgroup_commit_charge().
5593 */
5594void mem_cgroup_uncharge(struct page *page)
5595{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005596 if (mem_cgroup_disabled())
5597 return;
5598
Johannes Weiner747db952014-08-08 14:19:24 -07005599 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005600 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005601 return;
5602
Johannes Weiner747db952014-08-08 14:19:24 -07005603 INIT_LIST_HEAD(&page->lru);
5604 uncharge_list(&page->lru);
5605}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005606
Johannes Weiner747db952014-08-08 14:19:24 -07005607/**
5608 * mem_cgroup_uncharge_list - uncharge a list of page
5609 * @page_list: list of pages to uncharge
5610 *
5611 * Uncharge a list of pages previously charged with
5612 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5613 */
5614void mem_cgroup_uncharge_list(struct list_head *page_list)
5615{
5616 if (mem_cgroup_disabled())
5617 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005618
Johannes Weiner747db952014-08-08 14:19:24 -07005619 if (!list_empty(page_list))
5620 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005621}
5622
5623/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005624 * mem_cgroup_migrate - charge a page's replacement
5625 * @oldpage: currently circulating page
5626 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005627 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005628 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5629 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005630 *
5631 * Both pages must be locked, @newpage->mapping must be set up.
5632 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005633void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005634{
Johannes Weiner29833312014-12-10 15:44:02 -08005635 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005636 unsigned int nr_pages;
5637 bool compound;
Tejun Heod93c4132016-06-24 14:49:54 -07005638 unsigned long flags;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005639
5640 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5641 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005642 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005643 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5644 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005645
5646 if (mem_cgroup_disabled())
5647 return;
5648
5649 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005650 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005651 return;
5652
Hugh Dickins45637ba2015-11-05 18:49:40 -08005653 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08005654 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005655 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005656 return;
5657
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005658 /* Force-charge the new page. The old one will be freed soon */
5659 compound = PageTransHuge(newpage);
5660 nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5661
5662 page_counter_charge(&memcg->memory, nr_pages);
5663 if (do_memsw_account())
5664 page_counter_charge(&memcg->memsw, nr_pages);
5665 css_get_many(&memcg->css, nr_pages);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005666
Johannes Weiner9cf76662016-03-15 14:57:58 -07005667 commit_charge(newpage, memcg, false);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005668
Tejun Heod93c4132016-06-24 14:49:54 -07005669 local_irq_save(flags);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005670 mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5671 memcg_check_events(memcg, newpage);
Tejun Heod93c4132016-06-24 14:49:54 -07005672 local_irq_restore(flags);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005673}
5674
Johannes Weineref129472016-01-14 15:21:34 -08005675DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08005676EXPORT_SYMBOL(memcg_sockets_enabled_key);
5677
Johannes Weiner2d758072016-10-07 17:00:58 -07005678void mem_cgroup_sk_alloc(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08005679{
5680 struct mem_cgroup *memcg;
5681
Johannes Weiner2d758072016-10-07 17:00:58 -07005682 if (!mem_cgroup_sockets_enabled)
5683 return;
5684
5685 /*
5686 * Socket cloning can throw us here with sk_memcg already
Johannes Weiner11092082016-01-14 15:21:26 -08005687 * filled. It won't however, necessarily happen from
5688 * process context. So the test for root memcg given
5689 * the current task's memcg won't help us in this case.
5690 *
5691 * Respecting the original socket's memcg is a better
5692 * decision in this case.
5693 */
5694 if (sk->sk_memcg) {
5695 BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5696 css_get(&sk->sk_memcg->css);
5697 return;
5698 }
5699
5700 rcu_read_lock();
5701 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005702 if (memcg == root_mem_cgroup)
5703 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08005704 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005705 goto out;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005706 if (css_tryget_online(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08005707 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005708out:
Johannes Weiner11092082016-01-14 15:21:26 -08005709 rcu_read_unlock();
5710}
Johannes Weiner11092082016-01-14 15:21:26 -08005711
Johannes Weiner2d758072016-10-07 17:00:58 -07005712void mem_cgroup_sk_free(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08005713{
Johannes Weiner2d758072016-10-07 17:00:58 -07005714 if (sk->sk_memcg)
5715 css_put(&sk->sk_memcg->css);
Johannes Weiner11092082016-01-14 15:21:26 -08005716}
5717
5718/**
5719 * mem_cgroup_charge_skmem - charge socket memory
5720 * @memcg: memcg to charge
5721 * @nr_pages: number of pages to charge
5722 *
5723 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5724 * @memcg's configured limit, %false if the charge had to be forced.
5725 */
5726bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5727{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005728 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08005729
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005730 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005731 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005732
Johannes Weiner0db15292016-01-20 15:02:50 -08005733 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5734 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005735 return true;
5736 }
Johannes Weiner0db15292016-01-20 15:02:50 -08005737 page_counter_charge(&memcg->tcpmem, nr_pages);
5738 memcg->tcpmem_pressure = 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005739 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08005740 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005741
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005742 /* Don't block in the packet receive path */
5743 if (in_softirq())
5744 gfp_mask = GFP_NOWAIT;
5745
Johannes Weinerb2807f02016-01-20 15:03:22 -08005746 this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5747
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005748 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5749 return true;
5750
5751 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005752 return false;
5753}
5754
5755/**
5756 * mem_cgroup_uncharge_skmem - uncharge socket memory
5757 * @memcg - memcg to uncharge
5758 * @nr_pages - number of pages to uncharge
5759 */
5760void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5761{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005762 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005763 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005764 return;
5765 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005766
Johannes Weinerb2807f02016-01-20 15:03:22 -08005767 this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5768
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005769 page_counter_uncharge(&memcg->memory, nr_pages);
5770 css_put_many(&memcg->css, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005771}
5772
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005773static int __init cgroup_memory(char *s)
5774{
5775 char *token;
5776
5777 while ((token = strsep(&s, ",")) != NULL) {
5778 if (!*token)
5779 continue;
5780 if (!strcmp(token, "nosocket"))
5781 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08005782 if (!strcmp(token, "nokmem"))
5783 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005784 }
5785 return 0;
5786}
5787__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08005788
Michal Hocko2d110852013-02-22 16:34:43 -08005789/*
Michal Hocko10813122013-02-22 16:35:41 -08005790 * subsys_initcall() for memory controller.
5791 *
5792 * Some parts like hotcpu_notifier() have to be initialized from this context
5793 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5794 * everything that doesn't depend on a specific mem_cgroup structure should
5795 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005796 */
5797static int __init mem_cgroup_init(void)
5798{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005799 int cpu, node;
5800
Vladimir Davydov754ae7e2016-12-12 16:41:29 -08005801#ifndef CONFIG_SLOB
5802 /*
5803 * Kmem cache creation is mostly done with the slab_mutex held,
5804 * so use a special workqueue to avoid stalling all worker
5805 * threads in case lots of cgroups are created simultaneously.
5806 */
5807 memcg_kmem_cache_create_wq =
5808 alloc_ordered_workqueue("memcg_kmem_cache_create", 0);
5809 BUG_ON(!memcg_kmem_cache_create_wq);
5810#endif
5811
Michal Hocko2d110852013-02-22 16:34:43 -08005812 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005813
5814 for_each_possible_cpu(cpu)
5815 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5816 drain_local_stock);
5817
5818 for_each_node(node) {
5819 struct mem_cgroup_tree_per_node *rtpn;
Johannes Weiner95a045f2015-02-11 15:26:33 -08005820
5821 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5822 node_online(node) ? node : NUMA_NO_NODE);
5823
Mel Gormanef8f2322016-07-28 15:46:05 -07005824 rtpn->rb_root = RB_ROOT;
5825 spin_lock_init(&rtpn->lock);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005826 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5827 }
5828
Michal Hocko2d110852013-02-22 16:34:43 -08005829 return 0;
5830}
5831subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005832
5833#ifdef CONFIG_MEMCG_SWAP
Arnd Bergmann358c07f2016-08-25 15:17:08 -07005834static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
5835{
5836 while (!atomic_inc_not_zero(&memcg->id.ref)) {
5837 /*
5838 * The root cgroup cannot be destroyed, so it's refcount must
5839 * always be >= 1.
5840 */
5841 if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
5842 VM_BUG_ON(1);
5843 break;
5844 }
5845 memcg = parent_mem_cgroup(memcg);
5846 if (!memcg)
5847 memcg = root_mem_cgroup;
5848 }
5849 return memcg;
5850}
5851
Johannes Weiner21afa382015-02-11 15:26:36 -08005852/**
5853 * mem_cgroup_swapout - transfer a memsw charge to swap
5854 * @page: page whose memsw charge to transfer
5855 * @entry: swap entry to move the charge to
5856 *
5857 * Transfer the memsw charge of @page to @entry.
5858 */
5859void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5860{
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005861 struct mem_cgroup *memcg, *swap_memcg;
Johannes Weiner21afa382015-02-11 15:26:36 -08005862 unsigned short oldid;
5863
5864 VM_BUG_ON_PAGE(PageLRU(page), page);
5865 VM_BUG_ON_PAGE(page_count(page), page);
5866
Johannes Weiner7941d212016-01-14 15:21:23 -08005867 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08005868 return;
5869
5870 memcg = page->mem_cgroup;
5871
5872 /* Readahead page, never charged */
5873 if (!memcg)
5874 return;
5875
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005876 /*
5877 * In case the memcg owning these pages has been offlined and doesn't
5878 * have an ID allocated to it anymore, charge the closest online
5879 * ancestor for the swap instead and transfer the memory+swap charge.
5880 */
5881 swap_memcg = mem_cgroup_id_get_online(memcg);
5882 oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg));
Johannes Weiner21afa382015-02-11 15:26:36 -08005883 VM_BUG_ON_PAGE(oldid, page);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005884 mem_cgroup_swap_statistics(swap_memcg, true);
Johannes Weiner21afa382015-02-11 15:26:36 -08005885
5886 page->mem_cgroup = NULL;
5887
5888 if (!mem_cgroup_is_root(memcg))
5889 page_counter_uncharge(&memcg->memory, 1);
5890
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005891 if (memcg != swap_memcg) {
5892 if (!mem_cgroup_is_root(swap_memcg))
5893 page_counter_charge(&swap_memcg->memsw, 1);
5894 page_counter_uncharge(&memcg->memsw, 1);
5895 }
5896
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07005897 /*
5898 * Interrupts should be disabled here because the caller holds the
5899 * mapping->tree_lock lock which is taken with interrupts-off. It is
5900 * important here to have the interrupts disabled because it is the
5901 * only synchronisation we have for udpating the per-CPU variables.
5902 */
5903 VM_BUG_ON(!irqs_disabled());
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005904 mem_cgroup_charge_statistics(memcg, page, false, -1);
Johannes Weiner21afa382015-02-11 15:26:36 -08005905 memcg_check_events(memcg, page);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005906
5907 if (!mem_cgroup_is_root(memcg))
5908 css_put(&memcg->css);
Johannes Weiner21afa382015-02-11 15:26:36 -08005909}
5910
Vladimir Davydov37e84352016-01-20 15:02:56 -08005911/*
5912 * mem_cgroup_try_charge_swap - try charging a swap entry
5913 * @page: page being added to swap
5914 * @entry: swap entry to charge
5915 *
5916 * Try to charge @entry to the memcg that @page belongs to.
5917 *
5918 * Returns 0 on success, -ENOMEM on failure.
5919 */
5920int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5921{
5922 struct mem_cgroup *memcg;
5923 struct page_counter *counter;
5924 unsigned short oldid;
5925
5926 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5927 return 0;
5928
5929 memcg = page->mem_cgroup;
5930
5931 /* Readahead page, never charged */
5932 if (!memcg)
5933 return 0;
5934
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005935 memcg = mem_cgroup_id_get_online(memcg);
Vladimir Davydov37e84352016-01-20 15:02:56 -08005936
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005937 if (!mem_cgroup_is_root(memcg) &&
5938 !page_counter_try_charge(&memcg->swap, 1, &counter)) {
5939 mem_cgroup_id_put(memcg);
5940 return -ENOMEM;
5941 }
5942
Vladimir Davydov37e84352016-01-20 15:02:56 -08005943 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5944 VM_BUG_ON_PAGE(oldid, page);
5945 mem_cgroup_swap_statistics(memcg, true);
5946
Vladimir Davydov37e84352016-01-20 15:02:56 -08005947 return 0;
5948}
5949
Johannes Weiner21afa382015-02-11 15:26:36 -08005950/**
5951 * mem_cgroup_uncharge_swap - uncharge a swap entry
5952 * @entry: swap entry to uncharge
5953 *
Vladimir Davydov37e84352016-01-20 15:02:56 -08005954 * Drop the swap charge associated with @entry.
Johannes Weiner21afa382015-02-11 15:26:36 -08005955 */
5956void mem_cgroup_uncharge_swap(swp_entry_t entry)
5957{
5958 struct mem_cgroup *memcg;
5959 unsigned short id;
5960
Vladimir Davydov37e84352016-01-20 15:02:56 -08005961 if (!do_swap_account)
Johannes Weiner21afa382015-02-11 15:26:36 -08005962 return;
5963
5964 id = swap_cgroup_record(entry, 0);
5965 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005966 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005967 if (memcg) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08005968 if (!mem_cgroup_is_root(memcg)) {
5969 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5970 page_counter_uncharge(&memcg->swap, 1);
5971 else
5972 page_counter_uncharge(&memcg->memsw, 1);
5973 }
Johannes Weiner21afa382015-02-11 15:26:36 -08005974 mem_cgroup_swap_statistics(memcg, false);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005975 mem_cgroup_id_put(memcg);
Johannes Weiner21afa382015-02-11 15:26:36 -08005976 }
5977 rcu_read_unlock();
5978}
5979
Vladimir Davydovd8b38432016-01-20 15:03:07 -08005980long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
5981{
5982 long nr_swap_pages = get_nr_swap_pages();
5983
5984 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5985 return nr_swap_pages;
5986 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5987 nr_swap_pages = min_t(long, nr_swap_pages,
5988 READ_ONCE(memcg->swap.limit) -
5989 page_counter_read(&memcg->swap));
5990 return nr_swap_pages;
5991}
5992
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08005993bool mem_cgroup_swap_full(struct page *page)
5994{
5995 struct mem_cgroup *memcg;
5996
5997 VM_BUG_ON_PAGE(!PageLocked(page), page);
5998
5999 if (vm_swap_full())
6000 return true;
6001 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
6002 return false;
6003
6004 memcg = page->mem_cgroup;
6005 if (!memcg)
6006 return false;
6007
6008 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
6009 if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
6010 return true;
6011
6012 return false;
6013}
6014
Johannes Weiner21afa382015-02-11 15:26:36 -08006015/* for remember boot option*/
6016#ifdef CONFIG_MEMCG_SWAP_ENABLED
6017static int really_do_swap_account __initdata = 1;
6018#else
6019static int really_do_swap_account __initdata;
6020#endif
6021
6022static int __init enable_swap_account(char *s)
6023{
6024 if (!strcmp(s, "1"))
6025 really_do_swap_account = 1;
6026 else if (!strcmp(s, "0"))
6027 really_do_swap_account = 0;
6028 return 1;
6029}
6030__setup("swapaccount=", enable_swap_account);
6031
Vladimir Davydov37e84352016-01-20 15:02:56 -08006032static u64 swap_current_read(struct cgroup_subsys_state *css,
6033 struct cftype *cft)
6034{
6035 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6036
6037 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
6038}
6039
6040static int swap_max_show(struct seq_file *m, void *v)
6041{
6042 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
6043 unsigned long max = READ_ONCE(memcg->swap.limit);
6044
6045 if (max == PAGE_COUNTER_MAX)
6046 seq_puts(m, "max\n");
6047 else
6048 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
6049
6050 return 0;
6051}
6052
6053static ssize_t swap_max_write(struct kernfs_open_file *of,
6054 char *buf, size_t nbytes, loff_t off)
6055{
6056 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6057 unsigned long max;
6058 int err;
6059
6060 buf = strstrip(buf);
6061 err = page_counter_memparse(buf, "max", &max);
6062 if (err)
6063 return err;
6064
6065 mutex_lock(&memcg_limit_mutex);
6066 err = page_counter_limit(&memcg->swap, max);
6067 mutex_unlock(&memcg_limit_mutex);
6068 if (err)
6069 return err;
6070
6071 return nbytes;
6072}
6073
6074static struct cftype swap_files[] = {
6075 {
6076 .name = "swap.current",
6077 .flags = CFTYPE_NOT_ON_ROOT,
6078 .read_u64 = swap_current_read,
6079 },
6080 {
6081 .name = "swap.max",
6082 .flags = CFTYPE_NOT_ON_ROOT,
6083 .seq_show = swap_max_show,
6084 .write = swap_max_write,
6085 },
6086 { } /* terminate */
6087};
6088
Johannes Weiner21afa382015-02-11 15:26:36 -08006089static struct cftype memsw_cgroup_files[] = {
6090 {
6091 .name = "memsw.usage_in_bytes",
6092 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6093 .read_u64 = mem_cgroup_read_u64,
6094 },
6095 {
6096 .name = "memsw.max_usage_in_bytes",
6097 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6098 .write = mem_cgroup_reset,
6099 .read_u64 = mem_cgroup_read_u64,
6100 },
6101 {
6102 .name = "memsw.limit_in_bytes",
6103 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6104 .write = mem_cgroup_write,
6105 .read_u64 = mem_cgroup_read_u64,
6106 },
6107 {
6108 .name = "memsw.failcnt",
6109 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6110 .write = mem_cgroup_reset,
6111 .read_u64 = mem_cgroup_read_u64,
6112 },
6113 { }, /* terminate */
6114};
6115
6116static int __init mem_cgroup_swap_init(void)
6117{
6118 if (!mem_cgroup_disabled() && really_do_swap_account) {
6119 do_swap_account = 1;
Vladimir Davydov37e84352016-01-20 15:02:56 -08006120 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
6121 swap_files));
Johannes Weiner21afa382015-02-11 15:26:36 -08006122 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
6123 memsw_cgroup_files));
6124 }
6125 return 0;
6126}
6127subsys_initcall(mem_cgroup_swap_init);
6128
6129#endif /* CONFIG_MEMCG_SWAP */