blob: f60bd046413394af54c672aee8ab5f5623e38fcc [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Johannes Weiner1575e682015-04-14 15:44:51 -070017 * Native page reclaim
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
22 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080023 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
27 *
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
32 */
33
Johannes Weiner3e32cb22014-12-10 15:42:31 -080034#include <linux/page_counter.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080035#include <linux/memcontrol.h>
36#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080037#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080038#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080039#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080040#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080041#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080043#include <linux/bit_spinlock.h>
44#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070045#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040046#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080047#include <linux/mutex.h>
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -070048#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070049#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080051#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080052#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080053#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050054#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080055#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080056#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080057#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070058#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070059#include <linux/mm_inline.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080060#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080061#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070062#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070063#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050064#include <linux/file.h>
Tejun Heob23afb92015-11-05 18:46:11 -080065#include <linux/tracehook.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080066#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000067#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070068#include <net/ip.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080069#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080070
Balbir Singh8697d332008-02-07 00:13:59 -080071#include <asm/uaccess.h>
72
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070073#include <trace/events/vmscan.h>
74
Tejun Heo073219e2014-02-08 10:36:58 -050075struct cgroup_subsys memory_cgrp_subsys __read_mostly;
76EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080077
Johannes Weiner7d828602016-01-14 15:20:56 -080078struct mem_cgroup *root_mem_cgroup __read_mostly;
79
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070080#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh8cdea7c2008-02-07 00:13:50 -080081
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080082/* Socket memory accounting disabled? */
83static bool cgroup_memory_nosocket;
84
Vladimir Davydov04823c82016-01-20 15:02:38 -080085/* Kernel memory accounting disabled? */
86static bool cgroup_memory_nokmem;
87
Johannes Weiner21afa382015-02-11 15:26:36 -080088/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070089#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080090int do_swap_account __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080091#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070092#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080093#endif
94
Johannes Weiner7941d212016-01-14 15:21:23 -080095/* Whether legacy memory+swap accounting is active */
96static bool do_memsw_account(void)
97{
98 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account;
99}
100
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700101static const char * const mem_cgroup_stat_names[] = {
102 "cache",
103 "rss",
David Rientjesb070e652013-05-07 16:18:09 -0700104 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700105 "mapped_file",
Greg Thelenc4843a72015-05-22 17:13:16 -0400106 "dirty",
Sha Zhengju3ea67d02013-09-12 15:13:53 -0700107 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700108 "swap",
109};
110
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700111static const char * const mem_cgroup_events_names[] = {
112 "pgpgin",
113 "pgpgout",
114 "pgfault",
115 "pgmajfault",
116};
117
Sha Zhengju58cf1882013-02-22 16:32:05 -0800118static const char * const mem_cgroup_lru_names[] = {
119 "inactive_anon",
120 "active_anon",
121 "inactive_file",
122 "active_file",
123 "unevictable",
124};
125
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700126#define THRESHOLDS_EVENTS_TARGET 128
127#define SOFTLIMIT_EVENTS_TARGET 1024
128#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700129
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700130/*
131 * Cgroups above their limits are maintained in a RB-Tree, independent of
132 * their hierarchy representation
133 */
134
Mel Gormanef8f2322016-07-28 15:46:05 -0700135struct mem_cgroup_tree_per_node {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700136 struct rb_root rb_root;
137 spinlock_t lock;
138};
139
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700140struct mem_cgroup_tree {
141 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
142};
143
144static struct mem_cgroup_tree soft_limit_tree __read_mostly;
145
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700146/* for OOM */
147struct mem_cgroup_eventfd_list {
148 struct list_head list;
149 struct eventfd_ctx *eventfd;
150};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800151
Tejun Heo79bd9812013-11-22 18:20:42 -0500152/*
153 * cgroup_event represents events which userspace want to receive.
154 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500155struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500156 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500157 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500158 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500159 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500160 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500161 * eventfd to signal userspace about the event.
162 */
163 struct eventfd_ctx *eventfd;
164 /*
165 * Each of these stored in a list by the cgroup.
166 */
167 struct list_head list;
168 /*
Tejun Heofba94802013-11-22 18:20:43 -0500169 * register_event() callback will be used to add new userspace
170 * waiter for changes related to this event. Use eventfd_signal()
171 * on eventfd to send notification to userspace.
172 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500173 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500174 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500175 /*
176 * unregister_event() callback will be called when userspace closes
177 * the eventfd or on cgroup removing. This callback must be set,
178 * if you want provide notification functionality.
179 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500180 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500181 struct eventfd_ctx *eventfd);
182 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500183 * All fields below needed to unregister event when
184 * userspace closes eventfd.
185 */
186 poll_table pt;
187 wait_queue_head_t *wqh;
188 wait_queue_t wait;
189 struct work_struct remove;
190};
191
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700192static void mem_cgroup_threshold(struct mem_cgroup *memcg);
193static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800194
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800195/* Stuffs for move charges at task migration. */
196/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800197 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800198 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800199#define MOVE_ANON 0x1U
200#define MOVE_FILE 0x2U
201#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800202
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800203/* "mc" and its members are protected by cgroup_mutex */
204static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800205 spinlock_t lock; /* for from, to */
Tejun Heo264a0ae2016-04-21 19:09:02 -0400206 struct mm_struct *mm;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800207 struct mem_cgroup *from;
208 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800209 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800210 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800211 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800212 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800213 struct task_struct *moving_task; /* a task moving charges */
214 wait_queue_head_t waitq; /* a waitq for other context */
215} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700216 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800217 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
218};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800219
Balbir Singh4e416952009-09-23 15:56:39 -0700220/*
221 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
222 * limit reclaim to prevent infinite loops, if they ever occur.
223 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700224#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700225#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700226
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800227enum charge_type {
228 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700229 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800230 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700231 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700232 NR_CHARGE_TYPE,
233};
234
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800235/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800236enum res_type {
237 _MEM,
238 _MEMSWAP,
239 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800240 _KMEM,
Vladimir Davydovd55f90b2016-01-20 15:02:44 -0800241 _TCP,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800242};
243
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700244#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
245#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800246#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700247/* Used for OOM nofiier */
248#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800249
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700250/* Some nice accessors for the vmpressure. */
251struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
252{
253 if (!memcg)
254 memcg = root_mem_cgroup;
255 return &memcg->vmpressure;
256}
257
258struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
259{
260 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
261}
262
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700263static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
264{
265 return (memcg == root_mem_cgroup);
266}
267
Johannes Weiner127424c2016-01-20 15:02:32 -0800268#ifndef CONFIG_SLOB
Glauber Costa55007d82012-12-18 14:22:38 -0800269/*
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800270 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800271 * The main reason for not using cgroup id for this:
272 * this works better in sparse environments, where we have a lot of memcgs,
273 * but only a few kmem-limited. Or also, if we have, for instance, 200
274 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
275 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800276 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800277 * The current size of the caches array is stored in memcg_nr_cache_ids. It
278 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800279 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800280static DEFINE_IDA(memcg_cache_ida);
281int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800282
Vladimir Davydov05257a12015-02-12 14:59:01 -0800283/* Protects memcg_nr_cache_ids */
284static DECLARE_RWSEM(memcg_cache_ids_sem);
285
286void memcg_get_cache_ids(void)
287{
288 down_read(&memcg_cache_ids_sem);
289}
290
291void memcg_put_cache_ids(void)
292{
293 up_read(&memcg_cache_ids_sem);
294}
295
Glauber Costa55007d82012-12-18 14:22:38 -0800296/*
297 * MIN_SIZE is different than 1, because we would like to avoid going through
298 * the alloc/free process all the time. In a small machine, 4 kmem-limited
299 * cgroups is a reasonable guess. In the future, it could be a parameter or
300 * tunable, but that is strictly not necessary.
301 *
Li Zefanb8627832013-09-23 16:56:47 +0800302 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800303 * this constant directly from cgroup, but it is understandable that this is
304 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800305 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800306 * increase ours as well if it increases.
307 */
308#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800309#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800310
Glauber Costad7f25f82012-12-18 14:22:40 -0800311/*
312 * A lot of the calls to the cache allocation functions are expected to be
313 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
314 * conditional to this static branch, we'll have to allow modules that does
315 * kmem_cache_alloc and the such to see this symbol as well
316 */
Johannes Weineref129472016-01-14 15:21:34 -0800317DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
Glauber Costad7f25f82012-12-18 14:22:40 -0800318EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800319
Johannes Weiner127424c2016-01-20 15:02:32 -0800320#endif /* !CONFIG_SLOB */
Glauber Costaa8964b92012-12-18 14:22:09 -0800321
Tejun Heoad7fa852015-05-27 20:00:02 -0400322/**
323 * mem_cgroup_css_from_page - css of the memcg associated with a page
324 * @page: page of interest
325 *
326 * If memcg is bound to the default hierarchy, css of the memcg associated
327 * with @page is returned. The returned css remains associated with @page
328 * until it is released.
329 *
330 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
331 * is returned.
Tejun Heoad7fa852015-05-27 20:00:02 -0400332 */
333struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
334{
335 struct mem_cgroup *memcg;
336
Tejun Heoad7fa852015-05-27 20:00:02 -0400337 memcg = page->mem_cgroup;
338
Tejun Heo9e10a132015-09-18 11:56:28 -0400339 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400340 memcg = root_mem_cgroup;
341
Tejun Heoad7fa852015-05-27 20:00:02 -0400342 return &memcg->css;
343}
344
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700345/**
346 * page_cgroup_ino - return inode number of the memcg a page is charged to
347 * @page: the page
348 *
349 * Look up the closest online ancestor of the memory cgroup @page is charged to
350 * and return its inode number or 0 if @page is not charged to any cgroup. It
351 * is safe to call this function without holding a reference to @page.
352 *
353 * Note, this function is inherently racy, because there is nothing to prevent
354 * the cgroup inode from getting torn down and potentially reallocated a moment
355 * after page_cgroup_ino() returns, so it only should be used by callers that
356 * do not care (such as procfs interfaces).
357 */
358ino_t page_cgroup_ino(struct page *page)
359{
360 struct mem_cgroup *memcg;
361 unsigned long ino = 0;
362
363 rcu_read_lock();
364 memcg = READ_ONCE(page->mem_cgroup);
365 while (memcg && !(memcg->css.flags & CSS_ONLINE))
366 memcg = parent_mem_cgroup(memcg);
367 if (memcg)
368 ino = cgroup_ino(memcg->css.cgroup);
369 rcu_read_unlock();
370 return ino;
371}
372
Mel Gormanef8f2322016-07-28 15:46:05 -0700373static struct mem_cgroup_per_node *
374mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700375{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700376 int nid = page_to_nid(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700377
Mel Gormanef8f2322016-07-28 15:46:05 -0700378 return memcg->nodeinfo[nid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700379}
380
Mel Gormanef8f2322016-07-28 15:46:05 -0700381static struct mem_cgroup_tree_per_node *
382soft_limit_tree_node(int nid)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700383{
Mel Gormanef8f2322016-07-28 15:46:05 -0700384 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700385}
386
Mel Gormanef8f2322016-07-28 15:46:05 -0700387static struct mem_cgroup_tree_per_node *
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700388soft_limit_tree_from_page(struct page *page)
389{
390 int nid = page_to_nid(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700391
Mel Gormanef8f2322016-07-28 15:46:05 -0700392 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700393}
394
Mel Gormanef8f2322016-07-28 15:46:05 -0700395static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz,
396 struct mem_cgroup_tree_per_node *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800397 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700398{
399 struct rb_node **p = &mctz->rb_root.rb_node;
400 struct rb_node *parent = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700401 struct mem_cgroup_per_node *mz_node;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700402
403 if (mz->on_tree)
404 return;
405
406 mz->usage_in_excess = new_usage_in_excess;
407 if (!mz->usage_in_excess)
408 return;
409 while (*p) {
410 parent = *p;
Mel Gormanef8f2322016-07-28 15:46:05 -0700411 mz_node = rb_entry(parent, struct mem_cgroup_per_node,
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700412 tree_node);
413 if (mz->usage_in_excess < mz_node->usage_in_excess)
414 p = &(*p)->rb_left;
415 /*
416 * We can't avoid mem cgroups that are over their soft
417 * limit by the same amount
418 */
419 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
420 p = &(*p)->rb_right;
421 }
422 rb_link_node(&mz->tree_node, parent, p);
423 rb_insert_color(&mz->tree_node, &mctz->rb_root);
424 mz->on_tree = true;
425}
426
Mel Gormanef8f2322016-07-28 15:46:05 -0700427static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
428 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700429{
430 if (!mz->on_tree)
431 return;
432 rb_erase(&mz->tree_node, &mctz->rb_root);
433 mz->on_tree = false;
434}
435
Mel Gormanef8f2322016-07-28 15:46:05 -0700436static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
437 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700438{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700439 unsigned long flags;
440
441 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700442 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700443 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700444}
445
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800446static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
447{
448 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700449 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800450 unsigned long excess = 0;
451
452 if (nr_pages > soft_limit)
453 excess = nr_pages - soft_limit;
454
455 return excess;
456}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700457
458static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
459{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800460 unsigned long excess;
Mel Gormanef8f2322016-07-28 15:46:05 -0700461 struct mem_cgroup_per_node *mz;
462 struct mem_cgroup_tree_per_node *mctz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700463
Jianyu Zhane2318752014-06-06 14:38:20 -0700464 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700465 /*
466 * Necessary to update all ancestors when hierarchy is used.
467 * because their event counter is not touched.
468 */
469 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700470 mz = mem_cgroup_page_nodeinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800471 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700472 /*
473 * We have to update the tree if mz is on RB-tree or
474 * mem is over its softlimit.
475 */
476 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700477 unsigned long flags;
478
479 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700480 /* if on-tree, remove it */
481 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700482 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700483 /*
484 * Insert again. mz->usage_in_excess will be updated.
485 * If excess is 0, no tree ops.
486 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700487 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700488 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700489 }
490 }
491}
492
493static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
494{
Mel Gormanef8f2322016-07-28 15:46:05 -0700495 struct mem_cgroup_tree_per_node *mctz;
496 struct mem_cgroup_per_node *mz;
497 int nid;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700498
Jianyu Zhane2318752014-06-06 14:38:20 -0700499 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700500 mz = mem_cgroup_nodeinfo(memcg, nid);
501 mctz = soft_limit_tree_node(nid);
502 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700503 }
504}
505
Mel Gormanef8f2322016-07-28 15:46:05 -0700506static struct mem_cgroup_per_node *
507__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700508{
509 struct rb_node *rightmost = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700510 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700511
512retry:
513 mz = NULL;
514 rightmost = rb_last(&mctz->rb_root);
515 if (!rightmost)
516 goto done; /* Nothing to reclaim from */
517
Mel Gormanef8f2322016-07-28 15:46:05 -0700518 mz = rb_entry(rightmost, struct mem_cgroup_per_node, tree_node);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700519 /*
520 * Remove the node now but someone else can add it back,
521 * we will to add it back at the end of reclaim to its correct
522 * position in the tree.
523 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700524 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800525 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400526 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700527 goto retry;
528done:
529 return mz;
530}
531
Mel Gormanef8f2322016-07-28 15:46:05 -0700532static struct mem_cgroup_per_node *
533mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700534{
Mel Gormanef8f2322016-07-28 15:46:05 -0700535 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700536
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700537 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700538 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700539 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700540 return mz;
541}
542
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700543/*
Greg Thelen484ebb32015-10-01 15:37:05 -0700544 * Return page count for single (non recursive) @memcg.
545 *
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700546 * Implementation Note: reading percpu statistics for memcg.
547 *
548 * Both of vmstat[] and percpu_counter has threshold and do periodic
549 * synchronization to implement "quick" read. There are trade-off between
550 * reading cost and precision of value. Then, we may have a chance to implement
Greg Thelen484ebb32015-10-01 15:37:05 -0700551 * a periodic synchronization of counter in memcg's counter.
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700552 *
553 * But this _read() function is used for user interface now. The user accounts
554 * memory usage by memory cgroup and he _always_ requires exact value because
555 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
556 * have to visit all online cpus and make sum. So, for now, unnecessary
557 * synchronization is not implemented. (just implemented for cpu hotplug)
558 *
559 * If there are kernel internal actions which can make use of some not-exact
560 * value, and reading all cpu value can be performance bottleneck in some
Greg Thelen484ebb32015-10-01 15:37:05 -0700561 * common workload, threshold and synchronization as vmstat[] should be
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700562 * implemented.
563 */
Greg Thelen484ebb32015-10-01 15:37:05 -0700564static unsigned long
565mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800566{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700567 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800568 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800569
Greg Thelen484ebb32015-10-01 15:37:05 -0700570 /* Per-cpu values can be negative, use a signed accumulator */
Tejun Heo733a5722015-05-22 18:23:18 -0400571 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700572 val += per_cpu(memcg->stat->count[idx], cpu);
Greg Thelen484ebb32015-10-01 15:37:05 -0700573 /*
574 * Summing races with updates, so val may be negative. Avoid exposing
575 * transient negative values.
576 */
577 if (val < 0)
578 val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800579 return val;
580}
581
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700582static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700583 enum mem_cgroup_events_index idx)
584{
585 unsigned long val = 0;
586 int cpu;
587
Tejun Heo733a5722015-05-22 18:23:18 -0400588 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700589 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700590 return val;
591}
592
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700593static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700594 struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800595 bool compound, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800596{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700597 /*
598 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
599 * counted as CACHE even if it's on ANON LRU.
600 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700601 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700602 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700603 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800604 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700605 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700606 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700607
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800608 if (compound) {
609 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
David Rientjesb070e652013-05-07 16:18:09 -0700610 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
611 nr_pages);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800612 }
David Rientjesb070e652013-05-07 16:18:09 -0700613
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800614 /* pagein of a big page is an event. So, ignore page size */
615 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700616 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800617 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700618 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800619 nr_pages = -nr_pages; /* for event */
620 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800621
Johannes Weiner13114712012-05-29 15:07:07 -0700622 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800623}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800624
Vladimir Davydov0a6b76d2016-03-17 14:18:42 -0700625unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
626 int nid, unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700627{
Jianyu Zhane2318752014-06-06 14:38:20 -0700628 unsigned long nr = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -0700629 struct mem_cgroup_per_node *mz;
630 enum lru_list lru;
Ying Han889976d2011-05-26 16:25:33 -0700631
Jianyu Zhane2318752014-06-06 14:38:20 -0700632 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700633
Mel Gormanef8f2322016-07-28 15:46:05 -0700634 for_each_lru(lru) {
635 if (!(BIT(lru) & lru_mask))
636 continue;
637 mz = mem_cgroup_nodeinfo(memcg, nid);
638 nr += mz->lru_size[lru];
Jianyu Zhane2318752014-06-06 14:38:20 -0700639 }
640 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700641}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700642
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700643static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700644 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800645{
Jianyu Zhane2318752014-06-06 14:38:20 -0700646 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700647 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800648
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800649 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700650 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
651 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800652}
653
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800654static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
655 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800656{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700657 unsigned long val, next;
658
Johannes Weiner13114712012-05-29 15:07:07 -0700659 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700660 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700661 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800662 if ((long)next - (long)val < 0) {
663 switch (target) {
664 case MEM_CGROUP_TARGET_THRESH:
665 next = val + THRESHOLDS_EVENTS_TARGET;
666 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700667 case MEM_CGROUP_TARGET_SOFTLIMIT:
668 next = val + SOFTLIMIT_EVENTS_TARGET;
669 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800670 case MEM_CGROUP_TARGET_NUMAINFO:
671 next = val + NUMAINFO_EVENTS_TARGET;
672 break;
673 default:
674 break;
675 }
676 __this_cpu_write(memcg->stat->targets[target], next);
677 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700678 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800679 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800680}
681
682/*
683 * Check events in order.
684 *
685 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700686static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800687{
688 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800689 if (unlikely(mem_cgroup_event_ratelimit(memcg,
690 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700691 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800692 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800693
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700694 do_softlimit = mem_cgroup_event_ratelimit(memcg,
695 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700696#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800697 do_numainfo = mem_cgroup_event_ratelimit(memcg,
698 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700699#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800700 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700701 if (unlikely(do_softlimit))
702 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800703#if MAX_NUMNODES > 1
704 if (unlikely(do_numainfo))
705 atomic_inc(&memcg->numainfo_events);
706#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700707 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800708}
709
Balbir Singhcf475ad2008-04-29 01:00:16 -0700710struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800711{
Balbir Singh31a78f22008-09-28 23:09:31 +0100712 /*
713 * mm_update_next_owner() may clear mm->owner to NULL
714 * if it races with swapoff, page migration, etc.
715 * So this can be called with p == NULL.
716 */
717 if (unlikely(!p))
718 return NULL;
719
Tejun Heo073219e2014-02-08 10:36:58 -0500720 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800721}
Michal Hocko33398cf2015-09-08 15:01:02 -0700722EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800723
Johannes Weinerdf381972014-04-07 15:37:43 -0700724static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800725{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700726 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700727
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800728 rcu_read_lock();
729 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700730 /*
731 * Page cache insertions can happen withou an
732 * actual mm context, e.g. during disk probing
733 * on boot, loopback IO, acct() writes etc.
734 */
735 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700736 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700737 else {
738 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
739 if (unlikely(!memcg))
740 memcg = root_mem_cgroup;
741 }
Tejun Heoec903c02014-05-13 12:11:01 -0400742 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800743 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700744 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800745}
746
Johannes Weiner56600482012-01-12 17:17:59 -0800747/**
748 * mem_cgroup_iter - iterate over memory cgroup hierarchy
749 * @root: hierarchy root
750 * @prev: previously returned memcg, NULL on first invocation
751 * @reclaim: cookie for shared reclaim walks, NULL for full walks
752 *
753 * Returns references to children of the hierarchy below @root, or
754 * @root itself, or %NULL after a full round-trip.
755 *
756 * Caller must pass the return value in @prev on subsequent
757 * invocations for reference counting, or use mem_cgroup_iter_break()
758 * to cancel a hierarchy walk before the round-trip is complete.
759 *
760 * Reclaimers can specify a zone and a priority level in @reclaim to
761 * divide up the memcgs in the hierarchy among all concurrent
762 * reclaimers operating on the same zone and priority.
763 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700764struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800765 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700766 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700767{
Michal Hocko33398cf2015-09-08 15:01:02 -0700768 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800769 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800770 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800771 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700772
Andrew Morton694fbc02013-09-24 15:27:37 -0700773 if (mem_cgroup_disabled())
774 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800775
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700776 if (!root)
777 root = root_mem_cgroup;
778
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800779 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800780 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800781
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800782 if (!root->use_hierarchy && root != root_mem_cgroup) {
783 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800784 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700785 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800786 }
787
Michal Hocko542f85f2013-04-29 15:07:15 -0700788 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800789
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800790 if (reclaim) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700791 struct mem_cgroup_per_node *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800792
Mel Gormanef8f2322016-07-28 15:46:05 -0700793 mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800794 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700795
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800796 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700797 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800798
Vladimir Davydov6df38682015-12-29 14:54:10 -0800799 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -0700800 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800801 if (!pos || css_tryget(&pos->css))
802 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800803 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800804 * css reference reached zero, so iter->position will
805 * be cleared by ->css_released. However, we should not
806 * rely on this happening soon, because ->css_released
807 * is called from a work queue, and by busy-waiting we
808 * might block it. So we clear iter->position right
809 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800810 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800811 (void)cmpxchg(&iter->position, pos, NULL);
812 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800813 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800814
815 if (pos)
816 css = &pos->css;
817
818 for (;;) {
819 css = css_next_descendant_pre(css, &root->css);
820 if (!css) {
821 /*
822 * Reclaimers share the hierarchy walk, and a
823 * new one might jump in right at the end of
824 * the hierarchy - make sure they see at least
825 * one group and restart from the beginning.
826 */
827 if (!prev)
828 continue;
829 break;
830 }
831
832 /*
833 * Verify the css and acquire a reference. The root
834 * is provided by the caller, so we know it's alive
835 * and kicking, and don't take an extra reference.
836 */
837 memcg = mem_cgroup_from_css(css);
838
839 if (css == &root->css)
840 break;
841
Johannes Weiner0b8f73e2016-01-20 15:02:53 -0800842 if (css_tryget(css))
843 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800844
845 memcg = NULL;
846 }
847
848 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800849 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800850 * The position could have already been updated by a competing
851 * thread, so check that the value hasn't changed since we read
852 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800853 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800854 (void)cmpxchg(&iter->position, pos, memcg);
855
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800856 if (pos)
857 css_put(&pos->css);
858
859 if (!memcg)
860 iter->generation++;
861 else if (!prev)
862 reclaim->generation = iter->generation;
863 }
864
Michal Hocko542f85f2013-04-29 15:07:15 -0700865out_unlock:
866 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800867out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700868 if (prev && prev != root)
869 css_put(&prev->css);
870
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800871 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700872}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800873
Johannes Weiner56600482012-01-12 17:17:59 -0800874/**
875 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
876 * @root: hierarchy root
877 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
878 */
879void mem_cgroup_iter_break(struct mem_cgroup *root,
880 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800881{
882 if (!root)
883 root = root_mem_cgroup;
884 if (prev && prev != root)
885 css_put(&prev->css);
886}
887
Vladimir Davydov6df38682015-12-29 14:54:10 -0800888static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
889{
890 struct mem_cgroup *memcg = dead_memcg;
891 struct mem_cgroup_reclaim_iter *iter;
Mel Gormanef8f2322016-07-28 15:46:05 -0700892 struct mem_cgroup_per_node *mz;
893 int nid;
Vladimir Davydov6df38682015-12-29 14:54:10 -0800894 int i;
895
896 while ((memcg = parent_mem_cgroup(memcg))) {
897 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700898 mz = mem_cgroup_nodeinfo(memcg, nid);
899 for (i = 0; i <= DEF_PRIORITY; i++) {
900 iter = &mz->iter[i];
901 cmpxchg(&iter->position,
902 dead_memcg, NULL);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800903 }
904 }
905 }
906}
907
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700908/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800909 * Iteration constructs for visiting all cgroups (under a tree). If
910 * loops are exited prematurely (break), mem_cgroup_iter_break() must
911 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700912 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800913#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800914 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800915 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800916 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700917
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800918#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800919 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800920 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800921 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700922
Johannes Weiner925b7672012-01-12 17:18:15 -0800923/**
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700924 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
925 * @memcg: hierarchy root
926 * @fn: function to call for each task
927 * @arg: argument passed to @fn
928 *
929 * This function iterates over tasks attached to @memcg or to any of its
930 * descendants and calls @fn for each task. If @fn returns a non-zero
931 * value, the function breaks the iteration loop and returns the value.
932 * Otherwise, it will iterate over all tasks and return 0.
933 *
934 * This function must not be called for the root memory cgroup.
935 */
936int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
937 int (*fn)(struct task_struct *, void *), void *arg)
938{
939 struct mem_cgroup *iter;
940 int ret = 0;
941
942 BUG_ON(memcg == root_mem_cgroup);
943
944 for_each_mem_cgroup_tree(iter, memcg) {
945 struct css_task_iter it;
946 struct task_struct *task;
947
948 css_task_iter_start(&iter->css, &it);
949 while (!ret && (task = css_task_iter_next(&it)))
950 ret = fn(task, arg);
951 css_task_iter_end(&it);
952 if (ret) {
953 mem_cgroup_iter_break(memcg, iter);
954 break;
955 }
956 }
957 return ret;
958}
959
960/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800961 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -0800962 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700963 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800964 *
965 * This function is only safe when following the LRU page isolation
966 * and putback protocol: the LRU lock must be held, and the page must
967 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -0700968 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700969struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat)
Minchan Kim3f58a822011-03-22 16:32:53 -0700970{
Mel Gormanef8f2322016-07-28 15:46:05 -0700971 struct mem_cgroup_per_node *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -0800972 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800973 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800974
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800975 if (mem_cgroup_disabled()) {
Mel Gorman599d0c92016-07-28 15:45:31 -0700976 lruvec = &pgdat->lruvec;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800977 goto out;
978 }
Christoph Lameterb69408e2008-10-18 20:26:14 -0700979
Johannes Weiner1306a852014-12-10 15:44:52 -0800980 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800981 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800982 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -0800983 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -0800984 */
Johannes Weiner29833312014-12-10 15:44:02 -0800985 if (!memcg)
986 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800987
Mel Gormanef8f2322016-07-28 15:46:05 -0700988 mz = mem_cgroup_page_nodeinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800989 lruvec = &mz->lruvec;
990out:
991 /*
992 * Since a node can be onlined after the mem_cgroup was created,
993 * we have to be prepared to initialize lruvec->zone here;
994 * and if offlined then reonlined, we need to reinitialize it.
995 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700996 if (unlikely(lruvec->pgdat != pgdat))
997 lruvec->pgdat = pgdat;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800998 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -0800999}
1000
1001/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001002 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1003 * @lruvec: mem_cgroup per zone lru vector
1004 * @lru: index of lru list the page is sitting on
1005 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001006 *
Hugh Dickinsca707232016-05-19 17:12:35 -07001007 * This function must be called under lru_lock, just before a page is added
1008 * to or just after a page is removed from an lru list (that ordering being
1009 * so as to allow it to check that lru_size 0 is consistent with list_empty).
Johannes Weiner925b7672012-01-12 17:18:15 -08001010 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001011void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
Mel Gorman7ee36a12016-07-28 15:47:17 -07001012 int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001013{
Mel Gormanef8f2322016-07-28 15:46:05 -07001014 struct mem_cgroup_per_node *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001015 unsigned long *lru_size;
Hugh Dickinsca707232016-05-19 17:12:35 -07001016 long size;
1017 bool empty;
Johannes Weiner925b7672012-01-12 17:18:15 -08001018
1019 if (mem_cgroup_disabled())
1020 return;
1021
Mel Gormanef8f2322016-07-28 15:46:05 -07001022 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001023 lru_size = mz->lru_size + lru;
Hugh Dickinsca707232016-05-19 17:12:35 -07001024 empty = list_empty(lruvec->lists + lru);
1025
1026 if (nr_pages < 0)
1027 *lru_size += nr_pages;
1028
1029 size = *lru_size;
1030 if (WARN_ONCE(size < 0 || empty != !size,
1031 "%s(%p, %d, %d): lru_size %ld but %sempty\n",
1032 __func__, lruvec, lru, nr_pages, size, empty ? "" : "not ")) {
1033 VM_BUG_ON(1);
1034 *lru_size = 0;
1035 }
1036
1037 if (nr_pages > 0)
1038 *lru_size += nr_pages;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001039}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001040
Johannes Weiner2314b422014-12-10 15:44:33 -08001041bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001042{
Johannes Weiner2314b422014-12-10 15:44:33 -08001043 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001044 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001045 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001046
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001047 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001048 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001049 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001050 task_unlock(p);
1051 } else {
1052 /*
1053 * All threads may have already detached their mm's, but the oom
1054 * killer still needs to detect if they have already been oom
1055 * killed to prevent needlessly killing additional tasks.
1056 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001057 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001058 task_memcg = mem_cgroup_from_task(task);
1059 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001060 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001061 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001062 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1063 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001064 return ret;
1065}
1066
Johannes Weiner19942822011-02-01 15:52:43 -08001067/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001068 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001069 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001070 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001071 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001072 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001073 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001074static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001075{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001076 unsigned long margin = 0;
1077 unsigned long count;
1078 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001079
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001080 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001081 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001082 if (count < limit)
1083 margin = limit - count;
1084
Johannes Weiner7941d212016-01-14 15:21:23 -08001085 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001086 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001087 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001088 if (count <= limit)
1089 margin = min(margin, limit - count);
Li RongQingcbedbac2016-05-27 14:27:43 -07001090 else
1091 margin = 0;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001092 }
1093
1094 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001095}
1096
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001097/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001098 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001099 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001100 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1101 * moving cgroups. This is for waiting at high-memory pressure
1102 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001103 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001104static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001105{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001106 struct mem_cgroup *from;
1107 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001108 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001109 /*
1110 * Unlike task_move routines, we access mc.to, mc.from not under
1111 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1112 */
1113 spin_lock(&mc.lock);
1114 from = mc.from;
1115 to = mc.to;
1116 if (!from)
1117 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001118
Johannes Weiner2314b422014-12-10 15:44:33 -08001119 ret = mem_cgroup_is_descendant(from, memcg) ||
1120 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001121unlock:
1122 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001123 return ret;
1124}
1125
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001126static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001127{
1128 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001129 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001130 DEFINE_WAIT(wait);
1131 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1132 /* moving charge context might have finished. */
1133 if (mc.moving_task)
1134 schedule();
1135 finish_wait(&mc.waitq, &wait);
1136 return true;
1137 }
1138 }
1139 return false;
1140}
1141
Sha Zhengju58cf1882013-02-22 16:32:05 -08001142#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001143/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001144 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001145 * @memcg: The memory cgroup that went over limit
1146 * @p: Task that is going to be killed
1147 *
1148 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1149 * enabled
1150 */
1151void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1152{
Sha Zhengju58cf1882013-02-22 16:32:05 -08001153 struct mem_cgroup *iter;
1154 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001155
Balbir Singhe2224322009-04-02 16:57:39 -07001156 rcu_read_lock();
1157
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001158 if (p) {
1159 pr_info("Task in ");
1160 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1161 pr_cont(" killed as a result of limit of ");
1162 } else {
1163 pr_info("Memory limit reached of cgroup ");
1164 }
1165
Tejun Heoe61734c2014-02-12 09:29:50 -05001166 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001167 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001168
Balbir Singhe2224322009-04-02 16:57:39 -07001169 rcu_read_unlock();
1170
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001171 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1172 K((u64)page_counter_read(&memcg->memory)),
1173 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1174 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1175 K((u64)page_counter_read(&memcg->memsw)),
1176 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1177 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1178 K((u64)page_counter_read(&memcg->kmem)),
1179 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001180
1181 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001182 pr_info("Memory cgroup stats for ");
1183 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001184 pr_cont(":");
1185
1186 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08001187 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Sha Zhengju58cf1882013-02-22 16:32:05 -08001188 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07001189 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
Sha Zhengju58cf1882013-02-22 16:32:05 -08001190 K(mem_cgroup_read_stat(iter, i)));
1191 }
1192
1193 for (i = 0; i < NR_LRU_LISTS; i++)
1194 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1195 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1196
1197 pr_cont("\n");
1198 }
Balbir Singhe2224322009-04-02 16:57:39 -07001199}
1200
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001201/*
1202 * This function returns the number of memcg under hierarchy tree. Returns
1203 * 1(self count) if no children.
1204 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001205static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001206{
1207 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001208 struct mem_cgroup *iter;
1209
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001210 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001211 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001212 return num;
1213}
1214
Balbir Singh6d61ef42009-01-07 18:08:06 -08001215/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001216 * Return the memory (and swap, if configured) limit for a memcg.
1217 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001218unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001219{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001220 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001221
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001222 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001223 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001224 unsigned long memsw_limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001225 unsigned long swap_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001226
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001227 memsw_limit = memcg->memsw.limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001228 swap_limit = memcg->swap.limit;
1229 swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1230 limit = min(limit + swap_limit, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001231 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001232 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001233}
1234
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001235static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001236 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001237{
David Rientjes6e0fc462015-09-08 15:00:36 -07001238 struct oom_control oc = {
1239 .zonelist = NULL,
1240 .nodemask = NULL,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001241 .memcg = memcg,
David Rientjes6e0fc462015-09-08 15:00:36 -07001242 .gfp_mask = gfp_mask,
1243 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001244 };
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001245 bool ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001246
Johannes Weinerdc564012015-06-24 16:57:19 -07001247 mutex_lock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001248 ret = out_of_memory(&oc);
Johannes Weinerdc564012015-06-24 16:57:19 -07001249 mutex_unlock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001250 return ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001251}
1252
Michele Curtiae6e71d2014-12-12 16:56:35 -08001253#if MAX_NUMNODES > 1
1254
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001255/**
1256 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001257 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001258 * @nid: the node ID to be checked.
1259 * @noswap : specify true here if the user wants flle only information.
1260 *
1261 * This function returns whether the specified memcg contains any
1262 * reclaimable pages on a node. Returns true if there are any reclaimable
1263 * pages in the node.
1264 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001265static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001266 int nid, bool noswap)
1267{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001268 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001269 return true;
1270 if (noswap || !total_swap_pages)
1271 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001272 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001273 return true;
1274 return false;
1275
1276}
Ying Han889976d2011-05-26 16:25:33 -07001277
1278/*
1279 * Always updating the nodemask is not very good - even if we have an empty
1280 * list or the wrong list here, we can start from some node and traverse all
1281 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1282 *
1283 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001284static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001285{
1286 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001287 /*
1288 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1289 * pagein/pageout changes since the last update.
1290 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001291 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001292 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001293 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001294 return;
1295
Ying Han889976d2011-05-26 16:25:33 -07001296 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001297 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001298
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001299 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001300
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001301 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1302 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001303 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001304
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001305 atomic_set(&memcg->numainfo_events, 0);
1306 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001307}
1308
1309/*
1310 * Selecting a node where we start reclaim from. Because what we need is just
1311 * reducing usage counter, start from anywhere is O,K. Considering
1312 * memory reclaim from current node, there are pros. and cons.
1313 *
1314 * Freeing memory from current node means freeing memory from a node which
1315 * we'll use or we've used. So, it may make LRU bad. And if several threads
1316 * hit limits, it will see a contention on a node. But freeing from remote
1317 * node means more costs for memory reclaim because of memory latency.
1318 *
1319 * Now, we use round-robin. Better algorithm is welcomed.
1320 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001321int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001322{
1323 int node;
1324
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001325 mem_cgroup_may_update_nodemask(memcg);
1326 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001327
Andrew Morton0edaf862016-05-19 17:10:58 -07001328 node = next_node_in(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001329 /*
Michal Hockofda3d692016-05-19 17:11:34 -07001330 * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1331 * last time it really checked all the LRUs due to rate limiting.
1332 * Fallback to the current node in that case for simplicity.
Ying Han889976d2011-05-26 16:25:33 -07001333 */
1334 if (unlikely(node == MAX_NUMNODES))
1335 node = numa_node_id();
1336
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001337 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001338 return node;
1339}
Ying Han889976d2011-05-26 16:25:33 -07001340#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001341int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001342{
1343 return 0;
1344}
1345#endif
1346
Andrew Morton0608f432013-09-24 15:27:41 -07001347static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
Mel Gormanef8f2322016-07-28 15:46:05 -07001348 pg_data_t *pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001349 gfp_t gfp_mask,
1350 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001351{
Andrew Morton0608f432013-09-24 15:27:41 -07001352 struct mem_cgroup *victim = NULL;
1353 int total = 0;
1354 int loop = 0;
1355 unsigned long excess;
1356 unsigned long nr_scanned;
1357 struct mem_cgroup_reclaim_cookie reclaim = {
Mel Gormanef8f2322016-07-28 15:46:05 -07001358 .pgdat = pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001359 .priority = 0,
1360 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001361
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001362 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001363
Andrew Morton0608f432013-09-24 15:27:41 -07001364 while (1) {
1365 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1366 if (!victim) {
1367 loop++;
1368 if (loop >= 2) {
1369 /*
1370 * If we have not been able to reclaim
1371 * anything, it might because there are
1372 * no reclaimable pages under this hierarchy
1373 */
1374 if (!total)
1375 break;
1376 /*
1377 * We want to do more targeted reclaim.
1378 * excess >> 2 is not to excessive so as to
1379 * reclaim too much, nor too less that we keep
1380 * coming back to reclaim from this cgroup
1381 */
1382 if (total >= (excess >> 2) ||
1383 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1384 break;
1385 }
1386 continue;
1387 }
Mel Gormana9dd0a82016-07-28 15:46:02 -07001388 total += mem_cgroup_shrink_node(victim, gfp_mask, false,
Mel Gormanef8f2322016-07-28 15:46:05 -07001389 pgdat, &nr_scanned);
Andrew Morton0608f432013-09-24 15:27:41 -07001390 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001391 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001392 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001393 }
Andrew Morton0608f432013-09-24 15:27:41 -07001394 mem_cgroup_iter_break(root_memcg, victim);
1395 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001396}
1397
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001398#ifdef CONFIG_LOCKDEP
1399static struct lockdep_map memcg_oom_lock_dep_map = {
1400 .name = "memcg_oom_lock",
1401};
1402#endif
1403
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001404static DEFINE_SPINLOCK(memcg_oom_lock);
1405
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001406/*
1407 * Check OOM-Killer is already running under our hierarchy.
1408 * If someone is running, return false.
1409 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001410static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001411{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001412 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001413
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001414 spin_lock(&memcg_oom_lock);
1415
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001416 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001417 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001418 /*
1419 * this subtree of our hierarchy is already locked
1420 * so we cannot give a lock.
1421 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001422 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001423 mem_cgroup_iter_break(memcg, iter);
1424 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001425 } else
1426 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001427 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001428
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001429 if (failed) {
1430 /*
1431 * OK, we failed to lock the whole subtree so we have
1432 * to clean up what we set up to the failing subtree
1433 */
1434 for_each_mem_cgroup_tree(iter, memcg) {
1435 if (iter == failed) {
1436 mem_cgroup_iter_break(memcg, iter);
1437 break;
1438 }
1439 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001440 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001441 } else
1442 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001443
1444 spin_unlock(&memcg_oom_lock);
1445
1446 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001447}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001448
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001449static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001450{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001451 struct mem_cgroup *iter;
1452
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001453 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001454 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001455 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001456 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001457 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001458}
1459
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001460static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001461{
1462 struct mem_cgroup *iter;
1463
Tejun Heoc2b42d32015-06-24 16:58:23 -07001464 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001465 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001466 iter->under_oom++;
1467 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001468}
1469
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001470static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001471{
1472 struct mem_cgroup *iter;
1473
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001474 /*
1475 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001476 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001477 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001478 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001479 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001480 if (iter->under_oom > 0)
1481 iter->under_oom--;
1482 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001483}
1484
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001485static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1486
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001487struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001488 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001489 wait_queue_t wait;
1490};
1491
1492static int memcg_oom_wake_function(wait_queue_t *wait,
1493 unsigned mode, int sync, void *arg)
1494{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001495 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1496 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001497 struct oom_wait_info *oom_wait_info;
1498
1499 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001500 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001501
Johannes Weiner2314b422014-12-10 15:44:33 -08001502 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1503 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001504 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001505 return autoremove_wake_function(wait, mode, sync, arg);
1506}
1507
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001508static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001509{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001510 /*
1511 * For the following lockless ->under_oom test, the only required
1512 * guarantee is that it must see the state asserted by an OOM when
1513 * this function is called as a result of userland actions
1514 * triggered by the notification of the OOM. This is trivially
1515 * achieved by invoking mem_cgroup_mark_under_oom() before
1516 * triggering notification.
1517 */
1518 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001519 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001520}
1521
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001522static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001523{
Andrew Mortond0db7af2016-06-08 15:33:47 -07001524 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001525 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001526 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001527 * We are in the middle of the charge context here, so we
1528 * don't want to block when potentially sitting on a callstack
1529 * that holds all kinds of filesystem and mm locks.
1530 *
1531 * Also, the caller may handle a failed allocation gracefully
1532 * (like optional page cache readahead) and so an OOM killer
1533 * invocation might not even be necessary.
1534 *
1535 * That's why we don't do anything here except remember the
1536 * OOM context and then deal with it at the end of the page
1537 * fault when the stack is unwound, the locks are released,
1538 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001539 */
Johannes Weiner49426422013-10-16 13:46:59 -07001540 css_get(&memcg->css);
Tejun Heo626ebc42015-11-05 18:46:09 -08001541 current->memcg_in_oom = memcg;
1542 current->memcg_oom_gfp_mask = mask;
1543 current->memcg_oom_order = order;
Johannes Weiner49426422013-10-16 13:46:59 -07001544}
1545
1546/**
1547 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1548 * @handle: actually kill/wait or just clean up the OOM state
1549 *
1550 * This has to be called at the end of a page fault if the memcg OOM
1551 * handler was enabled.
1552 *
1553 * Memcg supports userspace OOM handling where failed allocations must
1554 * sleep on a waitqueue until the userspace task resolves the
1555 * situation. Sleeping directly in the charge context with all kinds
1556 * of locks held is not a good idea, instead we remember an OOM state
1557 * in the task and mem_cgroup_oom_synchronize() has to be called at
1558 * the end of the page fault to complete the OOM handling.
1559 *
1560 * Returns %true if an ongoing memcg OOM situation was detected and
1561 * completed, %false otherwise.
1562 */
1563bool mem_cgroup_oom_synchronize(bool handle)
1564{
Tejun Heo626ebc42015-11-05 18:46:09 -08001565 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001566 struct oom_wait_info owait;
1567 bool locked;
1568
1569 /* OOM is global, do not handle */
1570 if (!memcg)
1571 return false;
1572
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001573 if (!handle)
Johannes Weiner49426422013-10-16 13:46:59 -07001574 goto cleanup;
1575
1576 owait.memcg = memcg;
1577 owait.wait.flags = 0;
1578 owait.wait.func = memcg_oom_wake_function;
1579 owait.wait.private = current;
1580 INIT_LIST_HEAD(&owait.wait.task_list);
1581
1582 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001583 mem_cgroup_mark_under_oom(memcg);
1584
1585 locked = mem_cgroup_oom_trylock(memcg);
1586
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001587 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001588 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001589
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001590 if (locked && !memcg->oom_kill_disable) {
1591 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001592 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001593 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1594 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001595 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001596 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001597 mem_cgroup_unmark_under_oom(memcg);
1598 finish_wait(&memcg_oom_waitq, &owait.wait);
1599 }
1600
1601 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001602 mem_cgroup_oom_unlock(memcg);
1603 /*
1604 * There is no guarantee that an OOM-lock contender
1605 * sees the wakeups triggered by the OOM kill
1606 * uncharges. Wake any sleepers explicitely.
1607 */
1608 memcg_oom_recover(memcg);
1609 }
Johannes Weiner49426422013-10-16 13:46:59 -07001610cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001611 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001612 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001613 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001614}
1615
Johannes Weinerd7365e72014-10-29 14:50:48 -07001616/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001617 * lock_page_memcg - lock a page->mem_cgroup binding
1618 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001619 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001620 * This function protects unlocked LRU pages from being moved to
1621 * another cgroup and stabilizes their page->mem_cgroup binding.
Balbir Singhd69b0422009-06-17 16:26:34 -07001622 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001623void lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001624{
1625 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001626 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001627
Johannes Weiner6de22612015-02-11 15:25:01 -08001628 /*
1629 * The RCU lock is held throughout the transaction. The fast
1630 * path can get away without acquiring the memcg->move_lock
1631 * because page moving starts with an RCU grace period.
Johannes Weiner6de22612015-02-11 15:25:01 -08001632 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001633 rcu_read_lock();
1634
1635 if (mem_cgroup_disabled())
Johannes Weiner62cccb82016-03-15 14:57:22 -07001636 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001637again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001638 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001639 if (unlikely(!memcg))
Johannes Weiner62cccb82016-03-15 14:57:22 -07001640 return;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001641
Qiang Huangbdcbb652014-06-04 16:08:21 -07001642 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001643 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001644
Johannes Weiner6de22612015-02-11 15:25:01 -08001645 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001646 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001647 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001648 goto again;
1649 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001650
1651 /*
1652 * When charge migration first begins, we can have locked and
1653 * unlocked page stat updates happening concurrently. Track
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001654 * the task who has the lock for unlock_page_memcg().
Johannes Weiner6de22612015-02-11 15:25:01 -08001655 */
1656 memcg->move_lock_task = current;
1657 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001658
Johannes Weiner62cccb82016-03-15 14:57:22 -07001659 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001660}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001661EXPORT_SYMBOL(lock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001662
Johannes Weinerd7365e72014-10-29 14:50:48 -07001663/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001664 * unlock_page_memcg - unlock a page->mem_cgroup binding
Johannes Weiner62cccb82016-03-15 14:57:22 -07001665 * @page: the page
Johannes Weinerd7365e72014-10-29 14:50:48 -07001666 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001667void unlock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001668{
Johannes Weiner62cccb82016-03-15 14:57:22 -07001669 struct mem_cgroup *memcg = page->mem_cgroup;
1670
Johannes Weiner6de22612015-02-11 15:25:01 -08001671 if (memcg && memcg->move_lock_task == current) {
1672 unsigned long flags = memcg->move_lock_flags;
1673
1674 memcg->move_lock_task = NULL;
1675 memcg->move_lock_flags = 0;
1676
1677 spin_unlock_irqrestore(&memcg->move_lock, flags);
1678 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001679
Johannes Weinerd7365e72014-10-29 14:50:48 -07001680 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001681}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001682EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001683
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001684/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001685 * size of first charge trial. "32" comes from vmscan.c's magic value.
1686 * TODO: maybe necessary to use big numbers in big irons.
1687 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001688#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001689struct memcg_stock_pcp {
1690 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001691 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001692 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001693 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001694#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001695};
1696static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001697static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001698
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001699/**
1700 * consume_stock: Try to consume stocked charge on this cpu.
1701 * @memcg: memcg to consume from.
1702 * @nr_pages: how many pages to charge.
1703 *
1704 * The charges will only happen if @memcg matches the current cpu's memcg
1705 * stock, and at least @nr_pages are available in that stock. Failure to
1706 * service an allocation will refill the stock.
1707 *
1708 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001709 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001710static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001711{
1712 struct memcg_stock_pcp *stock;
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001713 unsigned long flags;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001714 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001715
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001716 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001717 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001718
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001719 local_irq_save(flags);
1720
1721 stock = this_cpu_ptr(&memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001722 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001723 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001724 ret = true;
1725 }
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001726
1727 local_irq_restore(flags);
1728
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001729 return ret;
1730}
1731
1732/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001733 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001734 */
1735static void drain_stock(struct memcg_stock_pcp *stock)
1736{
1737 struct mem_cgroup *old = stock->cached;
1738
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001739 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001740 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001741 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001742 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001743 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001744 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001745 }
1746 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001747}
1748
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001749static void drain_local_stock(struct work_struct *dummy)
1750{
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001751 struct memcg_stock_pcp *stock;
1752 unsigned long flags;
1753
1754 local_irq_save(flags);
1755
1756 stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001757 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001758 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001759
1760 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001761}
1762
1763/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001764 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001765 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001766 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001767static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001768{
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001769 struct memcg_stock_pcp *stock;
1770 unsigned long flags;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001771
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001772 local_irq_save(flags);
1773
1774 stock = this_cpu_ptr(&memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001775 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001776 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001777 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001778 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001779 stock->nr_pages += nr_pages;
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001780
1781 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001782}
1783
1784/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001785 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001786 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001787 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001788static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001789{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001790 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001791
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001792 /* If someone's already draining, avoid adding running more workers. */
1793 if (!mutex_trylock(&percpu_charge_mutex))
1794 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001795 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001796 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001797 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001798 for_each_online_cpu(cpu) {
1799 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001800 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001801
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001802 memcg = stock->cached;
1803 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001804 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001805 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001806 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001807 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1808 if (cpu == curcpu)
1809 drain_local_stock(&stock->work);
1810 else
1811 schedule_work_on(cpu, &stock->work);
1812 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001813 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001814 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001815 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001816 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001817}
1818
Paul Gortmaker0db06282013-06-19 14:53:51 -04001819static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001820 unsigned long action,
1821 void *hcpu)
1822{
1823 int cpu = (unsigned long)hcpu;
1824 struct memcg_stock_pcp *stock;
1825
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001826 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001827 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001828
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07001829 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001830 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001831
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001832 stock = &per_cpu(memcg_stock, cpu);
1833 drain_stock(stock);
1834 return NOTIFY_OK;
1835}
1836
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001837static void reclaim_high(struct mem_cgroup *memcg,
1838 unsigned int nr_pages,
1839 gfp_t gfp_mask)
1840{
1841 do {
1842 if (page_counter_read(&memcg->memory) <= memcg->high)
1843 continue;
1844 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
1845 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1846 } while ((memcg = parent_mem_cgroup(memcg)));
1847}
1848
1849static void high_work_func(struct work_struct *work)
1850{
1851 struct mem_cgroup *memcg;
1852
1853 memcg = container_of(work, struct mem_cgroup, high_work);
1854 reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1855}
1856
Tejun Heob23afb92015-11-05 18:46:11 -08001857/*
1858 * Scheduled by try_charge() to be executed from the userland return path
1859 * and reclaims memory over the high limit.
1860 */
1861void mem_cgroup_handle_over_high(void)
1862{
1863 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001864 struct mem_cgroup *memcg;
Tejun Heob23afb92015-11-05 18:46:11 -08001865
1866 if (likely(!nr_pages))
1867 return;
1868
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001869 memcg = get_mem_cgroup_from_mm(current->mm);
1870 reclaim_high(memcg, nr_pages, GFP_KERNEL);
Tejun Heob23afb92015-11-05 18:46:11 -08001871 css_put(&memcg->css);
1872 current->memcg_nr_pages_over_high = 0;
1873}
1874
Johannes Weiner00501b52014-08-08 14:19:20 -07001875static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1876 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001877{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001878 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001879 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001880 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001881 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001882 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001883 bool may_swap = true;
1884 bool drained = false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001885
Johannes Weinerce00a962014-09-05 08:43:57 -04001886 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08001887 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001888retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001889 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08001890 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001891
Johannes Weiner7941d212016-01-14 15:21:23 -08001892 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08001893 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1894 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001895 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08001896 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001897 page_counter_uncharge(&memcg->memsw, batch);
1898 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001899 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001900 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001901 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001902 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001903
Johannes Weiner6539cc02014-08-06 16:05:42 -07001904 if (batch > nr_pages) {
1905 batch = nr_pages;
1906 goto retry;
1907 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001908
Johannes Weiner06b078f2014-08-06 16:05:44 -07001909 /*
1910 * Unlike in global OOM situations, memcg is not in a physical
1911 * memory shortage. Allow dying and OOM-killed tasks to
1912 * bypass the last charges so that they can exit quickly and
1913 * free their memory.
1914 */
1915 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1916 fatal_signal_pending(current) ||
1917 current->flags & PF_EXITING))
Tejun Heo10d53c72015-11-05 18:46:17 -08001918 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001919
Johannes Weiner89a28482016-10-27 17:46:56 -07001920 /*
1921 * Prevent unbounded recursion when reclaim operations need to
1922 * allocate memory. This might exceed the limits temporarily,
1923 * but we prefer facilitating memory reclaim and getting back
1924 * under the limit over triggering OOM kills in these cases.
1925 */
1926 if (unlikely(current->flags & PF_MEMALLOC))
1927 goto force;
1928
Johannes Weiner06b078f2014-08-06 16:05:44 -07001929 if (unlikely(task_in_memcg_oom(current)))
1930 goto nomem;
1931
Mel Gormand0164ad2015-11-06 16:28:21 -08001932 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001933 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001934
Johannes Weiner241994e2015-02-11 15:26:06 -08001935 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
1936
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001937 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1938 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07001939
Johannes Weiner61e02c72014-08-06 16:08:16 -07001940 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07001941 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07001942
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001943 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001944 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001945 drained = true;
1946 goto retry;
1947 }
1948
Johannes Weiner28c34c22014-08-06 16:05:47 -07001949 if (gfp_mask & __GFP_NORETRY)
1950 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001951 /*
1952 * Even though the limit is exceeded at this point, reclaim
1953 * may have been able to free some pages. Retry the charge
1954 * before killing the task.
1955 *
1956 * Only for regular pages, though: huge pages are rather
1957 * unlikely to succeed so close to the limit, and we fall back
1958 * to regular pages anyway in case of failure.
1959 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07001960 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001961 goto retry;
1962 /*
1963 * At task move, charge accounts can be doubly counted. So, it's
1964 * better to wait until the end of task_move if something is going on.
1965 */
1966 if (mem_cgroup_wait_acct_move(mem_over_limit))
1967 goto retry;
1968
Johannes Weiner9b130612014-08-06 16:05:51 -07001969 if (nr_retries--)
1970 goto retry;
1971
Johannes Weiner06b078f2014-08-06 16:05:44 -07001972 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08001973 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001974
Johannes Weiner6539cc02014-08-06 16:05:42 -07001975 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08001976 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001977
Johannes Weiner241994e2015-02-11 15:26:06 -08001978 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
1979
Jerome Marchand3608de02015-11-05 18:47:29 -08001980 mem_cgroup_oom(mem_over_limit, gfp_mask,
1981 get_order(nr_pages * PAGE_SIZE));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001982nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07001983 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07001984 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08001985force:
1986 /*
1987 * The allocation either can't fail or will lead to more memory
1988 * being freed very soon. Allow memory usage go over the limit
1989 * temporarily by force charging it.
1990 */
1991 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001992 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08001993 page_counter_charge(&memcg->memsw, nr_pages);
1994 css_get_many(&memcg->css, nr_pages);
1995
1996 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001997
1998done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001999 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002000 if (batch > nr_pages)
2001 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002002
Johannes Weiner241994e2015-02-11 15:26:06 -08002003 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002004 * If the hierarchy is above the normal consumption range, schedule
2005 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002006 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002007 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2008 * not recorded as it most likely matches current's and won't
2009 * change in the meantime. As high limit is checked again before
2010 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994e2015-02-11 15:26:06 -08002011 */
2012 do {
Tejun Heob23afb92015-11-05 18:46:11 -08002013 if (page_counter_read(&memcg->memory) > memcg->high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002014 /* Don't bother a random interrupted task */
2015 if (in_interrupt()) {
2016 schedule_work(&memcg->high_work);
2017 break;
2018 }
Vladimir Davydov9516a182015-12-11 13:40:24 -08002019 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002020 set_notify_resume(current);
2021 break;
2022 }
Johannes Weiner241994e2015-02-11 15:26:06 -08002023 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002024
2025 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002026}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002027
Johannes Weiner00501b52014-08-08 14:19:20 -07002028static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002029{
Johannes Weinerce00a962014-09-05 08:43:57 -04002030 if (mem_cgroup_is_root(memcg))
2031 return;
2032
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002033 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002034 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002035 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002036
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002037 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002038}
2039
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002040static void lock_page_lru(struct page *page, int *isolated)
2041{
2042 struct zone *zone = page_zone(page);
2043
Mel Gormana52633d2016-07-28 15:45:28 -07002044 spin_lock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002045 if (PageLRU(page)) {
2046 struct lruvec *lruvec;
2047
Mel Gorman599d0c92016-07-28 15:45:31 -07002048 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002049 ClearPageLRU(page);
2050 del_page_from_lru_list(page, lruvec, page_lru(page));
2051 *isolated = 1;
2052 } else
2053 *isolated = 0;
2054}
2055
2056static void unlock_page_lru(struct page *page, int isolated)
2057{
2058 struct zone *zone = page_zone(page);
2059
2060 if (isolated) {
2061 struct lruvec *lruvec;
2062
Mel Gorman599d0c92016-07-28 15:45:31 -07002063 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002064 VM_BUG_ON_PAGE(PageLRU(page), page);
2065 SetPageLRU(page);
2066 add_page_to_lru_list(page, lruvec, page_lru(page));
2067 }
Mel Gormana52633d2016-07-28 15:45:28 -07002068 spin_unlock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002069}
2070
Johannes Weiner00501b52014-08-08 14:19:20 -07002071static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002072 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002073{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002074 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002075
Johannes Weiner1306a852014-12-10 15:44:52 -08002076 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002077
2078 /*
2079 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2080 * may already be on some other mem_cgroup's LRU. Take care of it.
2081 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002082 if (lrucare)
2083 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002084
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002085 /*
2086 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002087 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002088 *
2089 * - the page is uncharged
2090 *
2091 * - the page is off-LRU
2092 *
2093 * - an anonymous fault has exclusive page access, except for
2094 * a locked page table
2095 *
2096 * - a page cache insertion, a swapin fault, or a migration
2097 * have the page locked
2098 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002099 page->mem_cgroup = memcg;
Hugh Dickins3be91272008-02-07 00:14:19 -08002100
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002101 if (lrucare)
2102 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002103}
2104
Johannes Weiner127424c2016-01-20 15:02:32 -08002105#ifndef CONFIG_SLOB
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002106static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002107{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002108 int id, size;
2109 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002110
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002111 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002112 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2113 if (id < 0)
2114 return id;
2115
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002116 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002117 return id;
2118
2119 /*
2120 * There's no space for the new id in memcg_caches arrays,
2121 * so we have to grow them.
2122 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002123 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002124
2125 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002126 if (size < MEMCG_CACHES_MIN_SIZE)
2127 size = MEMCG_CACHES_MIN_SIZE;
2128 else if (size > MEMCG_CACHES_MAX_SIZE)
2129 size = MEMCG_CACHES_MAX_SIZE;
2130
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002131 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002132 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002133 err = memcg_update_all_list_lrus(size);
2134 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002135 memcg_nr_cache_ids = size;
2136
2137 up_write(&memcg_cache_ids_sem);
2138
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002139 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002140 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002141 return err;
2142 }
2143 return id;
2144}
2145
2146static void memcg_free_cache_id(int id)
2147{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002148 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002149}
2150
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002151struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002152 struct mem_cgroup *memcg;
2153 struct kmem_cache *cachep;
2154 struct work_struct work;
2155};
2156
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002157static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002158{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002159 struct memcg_kmem_cache_create_work *cw =
2160 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002161 struct mem_cgroup *memcg = cw->memcg;
2162 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002163
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002164 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002165
Vladimir Davydov5722d092014-04-07 15:39:24 -07002166 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002167 kfree(cw);
2168}
2169
2170/*
2171 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002172 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002173static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2174 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002175{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002176 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002177
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002178 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002179 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002180 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002181
2182 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002183
2184 cw->memcg = memcg;
2185 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002186 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002187
Glauber Costad7f25f82012-12-18 14:22:40 -08002188 schedule_work(&cw->work);
2189}
2190
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002191static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2192 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002193{
2194 /*
2195 * We need to stop accounting when we kmalloc, because if the
2196 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002197 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002198 *
2199 * However, it is better to enclose the whole function. Depending on
2200 * the debugging options enabled, INIT_WORK(), for instance, can
2201 * trigger an allocation. This too, will make us recurse. Because at
2202 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2203 * the safest choice is to do it like this, wrapping the whole function.
2204 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002205 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002206 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002207 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002208}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002209
Vladimir Davydov45264772016-07-26 15:24:21 -07002210static inline bool memcg_kmem_bypass(void)
2211{
2212 if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
2213 return true;
2214 return false;
2215}
2216
2217/**
2218 * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2219 * @cachep: the original global kmem cache
2220 *
Glauber Costad7f25f82012-12-18 14:22:40 -08002221 * Return the kmem_cache we're supposed to use for a slab allocation.
2222 * We try to use the current memcg's version of the cache.
2223 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002224 * If the cache does not exist yet, if we are the first user of it, we
2225 * create it asynchronously in a workqueue and let the current allocation
2226 * go through with the original cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002227 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002228 * This function takes a reference to the cache it returns to assure it
2229 * won't get destroyed while we are working with it. Once the caller is
2230 * done with it, memcg_kmem_put_cache() must be called to release the
2231 * reference.
Glauber Costad7f25f82012-12-18 14:22:40 -08002232 */
Vladimir Davydov45264772016-07-26 15:24:21 -07002233struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002234{
2235 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002236 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002237 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002238
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002239 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002240
Vladimir Davydov45264772016-07-26 15:24:21 -07002241 if (memcg_kmem_bypass())
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002242 return cachep;
2243
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002244 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002245 return cachep;
2246
Vladimir Davydov8135be52014-12-12 16:56:38 -08002247 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002248 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002249 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002250 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002251
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002252 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002253 if (likely(memcg_cachep))
2254 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002255
2256 /*
2257 * If we are in a safe context (can wait, and not in interrupt
2258 * context), we could be be predictable and return right away.
2259 * This would guarantee that the allocation being performed
2260 * already belongs in the new cache.
2261 *
2262 * However, there are some clashes that can arrive from locking.
2263 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002264 * memcg_create_kmem_cache, this means no further allocation
2265 * could happen with the slab_mutex held. So it's better to
2266 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002267 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002268 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002269out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002270 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002271 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002272}
Glauber Costad7f25f82012-12-18 14:22:40 -08002273
Vladimir Davydov45264772016-07-26 15:24:21 -07002274/**
2275 * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2276 * @cachep: the cache returned by memcg_kmem_get_cache
2277 */
2278void memcg_kmem_put_cache(struct kmem_cache *cachep)
Vladimir Davydov8135be52014-12-12 16:56:38 -08002279{
2280 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002281 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002282}
2283
Vladimir Davydov45264772016-07-26 15:24:21 -07002284/**
2285 * memcg_kmem_charge: charge a kmem page
2286 * @page: page to charge
2287 * @gfp: reclaim mode
2288 * @order: allocation order
2289 * @memcg: memory cgroup to charge
2290 *
2291 * Returns 0 on success, an error code on failure.
2292 */
2293int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2294 struct mem_cgroup *memcg)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002295{
2296 unsigned int nr_pages = 1 << order;
2297 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002298 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002299
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002300 ret = try_charge(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08002301 if (ret)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002302 return ret;
Johannes Weiner52c29b02016-01-20 15:02:35 -08002303
2304 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2305 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2306 cancel_charge(memcg, nr_pages);
2307 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002308 }
2309
2310 page->mem_cgroup = memcg;
2311
2312 return 0;
2313}
2314
Vladimir Davydov45264772016-07-26 15:24:21 -07002315/**
2316 * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2317 * @page: page to charge
2318 * @gfp: reclaim mode
2319 * @order: allocation order
2320 *
2321 * Returns 0 on success, an error code on failure.
2322 */
2323int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002324{
2325 struct mem_cgroup *memcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002326 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002327
Vladimir Davydov45264772016-07-26 15:24:21 -07002328 if (memcg_kmem_bypass())
2329 return 0;
2330
Johannes Weinerdf381972014-04-07 15:37:43 -07002331 memcg = get_mem_cgroup_from_mm(current->mm);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002332 if (!mem_cgroup_is_root(memcg)) {
Vladimir Davydov45264772016-07-26 15:24:21 -07002333 ret = memcg_kmem_charge_memcg(page, gfp, order, memcg);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002334 if (!ret)
2335 __SetPageKmemcg(page);
2336 }
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002337 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002338 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002339}
Vladimir Davydov45264772016-07-26 15:24:21 -07002340/**
2341 * memcg_kmem_uncharge: uncharge a kmem page
2342 * @page: page to uncharge
2343 * @order: allocation order
2344 */
2345void memcg_kmem_uncharge(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002346{
Johannes Weiner1306a852014-12-10 15:44:52 -08002347 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002348 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002349
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002350 if (!memcg)
2351 return;
2352
Sasha Levin309381fea2014-01-23 15:52:54 -08002353 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002354
Johannes Weiner52c29b02016-01-20 15:02:35 -08002355 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2356 page_counter_uncharge(&memcg->kmem, nr_pages);
2357
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002358 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002359 if (do_memsw_account())
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002360 page_counter_uncharge(&memcg->memsw, nr_pages);
2361
Johannes Weiner1306a852014-12-10 15:44:52 -08002362 page->mem_cgroup = NULL;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002363
2364 /* slab pages do not have PageKmemcg flag set */
2365 if (PageKmemcg(page))
2366 __ClearPageKmemcg(page);
2367
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002368 css_put_many(&memcg->css, nr_pages);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002369}
Johannes Weiner127424c2016-01-20 15:02:32 -08002370#endif /* !CONFIG_SLOB */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002371
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002372#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2373
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002374/*
2375 * Because tail pages are not marked as "used", set it. We're under
Mel Gormana52633d2016-07-28 15:45:28 -07002376 * zone_lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002377 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002378void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002379{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002380 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002381
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002382 if (mem_cgroup_disabled())
2383 return;
David Rientjesb070e652013-05-07 16:18:09 -07002384
Johannes Weiner29833312014-12-10 15:44:02 -08002385 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002386 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002387
Johannes Weiner1306a852014-12-10 15:44:52 -08002388 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002389 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002390}
Hugh Dickins12d27102012-01-12 17:19:52 -08002391#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002392
Andrew Mortonc255a452012-07-31 16:43:02 -07002393#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002394static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2395 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002396{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002397 int val = (charge) ? 1 : -1;
2398 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002399}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002400
2401/**
2402 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2403 * @entry: swap entry to be moved
2404 * @from: mem_cgroup which the entry is moved from
2405 * @to: mem_cgroup which the entry is moved to
2406 *
2407 * It succeeds only when the swap_cgroup's record for this entry is the same
2408 * as the mem_cgroup's id of @from.
2409 *
2410 * Returns 0 on success, -EINVAL on failure.
2411 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002412 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002413 * both res and memsw, and called css_get().
2414 */
2415static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002416 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002417{
2418 unsigned short old_id, new_id;
2419
Li Zefan34c00c32013-09-23 16:56:01 +08002420 old_id = mem_cgroup_id(from);
2421 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002422
2423 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002424 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002425 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002426 return 0;
2427 }
2428 return -EINVAL;
2429}
2430#else
2431static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002432 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002433{
2434 return -EINVAL;
2435}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002436#endif
2437
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002438static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002439
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002440static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002441 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002442{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002443 unsigned long curusage;
2444 unsigned long oldusage;
2445 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002446 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002447 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002448
2449 /*
2450 * For keeping hierarchical_reclaim simple, how long we should retry
2451 * is depends on callers. We set our retry-count to be function
2452 * of # of children which we should visit in this loop.
2453 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002454 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2455 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002456
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002457 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002458
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002459 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002460 if (signal_pending(current)) {
2461 ret = -EINTR;
2462 break;
2463 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002464
2465 mutex_lock(&memcg_limit_mutex);
2466 if (limit > memcg->memsw.limit) {
2467 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002468 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002469 break;
2470 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002471 if (limit > memcg->memory.limit)
2472 enlarge = true;
2473 ret = page_counter_limit(&memcg->memory, limit);
2474 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002475
2476 if (!ret)
2477 break;
2478
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002479 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2480
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002481 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002482 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002483 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002484 retry_count--;
2485 else
2486 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002487 } while (retry_count);
2488
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002489 if (!ret && enlarge)
2490 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002491
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002492 return ret;
2493}
2494
Li Zefan338c8432009-06-17 16:27:15 -07002495static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002496 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002497{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002498 unsigned long curusage;
2499 unsigned long oldusage;
2500 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002501 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002502 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002503
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002504 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002505 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2506 mem_cgroup_count_children(memcg);
2507
2508 oldusage = page_counter_read(&memcg->memsw);
2509
2510 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002511 if (signal_pending(current)) {
2512 ret = -EINTR;
2513 break;
2514 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002515
2516 mutex_lock(&memcg_limit_mutex);
2517 if (limit < memcg->memory.limit) {
2518 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002519 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002520 break;
2521 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002522 if (limit > memcg->memsw.limit)
2523 enlarge = true;
2524 ret = page_counter_limit(&memcg->memsw, limit);
2525 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002526
2527 if (!ret)
2528 break;
2529
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002530 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2531
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002532 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002533 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002534 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002535 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002536 else
2537 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002538 } while (retry_count);
2539
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002540 if (!ret && enlarge)
2541 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002542
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002543 return ret;
2544}
2545
Mel Gormanef8f2322016-07-28 15:46:05 -07002546unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
Andrew Morton0608f432013-09-24 15:27:41 -07002547 gfp_t gfp_mask,
2548 unsigned long *total_scanned)
2549{
2550 unsigned long nr_reclaimed = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002551 struct mem_cgroup_per_node *mz, *next_mz = NULL;
Andrew Morton0608f432013-09-24 15:27:41 -07002552 unsigned long reclaimed;
2553 int loop = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002554 struct mem_cgroup_tree_per_node *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002555 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002556 unsigned long nr_scanned;
2557
2558 if (order > 0)
2559 return 0;
2560
Mel Gormanef8f2322016-07-28 15:46:05 -07002561 mctz = soft_limit_tree_node(pgdat->node_id);
Michal Hockod6507ff2016-08-02 14:02:37 -07002562
2563 /*
2564 * Do not even bother to check the largest node if the root
2565 * is empty. Do it lockless to prevent lock bouncing. Races
2566 * are acceptable as soft limit is best effort anyway.
2567 */
2568 if (RB_EMPTY_ROOT(&mctz->rb_root))
2569 return 0;
2570
Andrew Morton0608f432013-09-24 15:27:41 -07002571 /*
2572 * This loop can run a while, specially if mem_cgroup's continuously
2573 * keep exceeding their soft limit and putting the system under
2574 * pressure
2575 */
2576 do {
2577 if (next_mz)
2578 mz = next_mz;
2579 else
2580 mz = mem_cgroup_largest_soft_limit_node(mctz);
2581 if (!mz)
2582 break;
2583
2584 nr_scanned = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002585 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07002586 gfp_mask, &nr_scanned);
2587 nr_reclaimed += reclaimed;
2588 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002589 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002590 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002591
2592 /*
2593 * If we failed to reclaim anything from this memory cgroup
2594 * it is time to move on to the next cgroup
2595 */
2596 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002597 if (!reclaimed)
2598 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2599
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002600 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002601 /*
2602 * One school of thought says that we should not add
2603 * back the node to the tree if reclaim returns 0.
2604 * But our reclaim could return 0, simply because due
2605 * to priority we are exposing a smaller subset of
2606 * memory to reclaim from. Consider this as a longer
2607 * term TODO.
2608 */
2609 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002610 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002611 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002612 css_put(&mz->memcg->css);
2613 loop++;
2614 /*
2615 * Could not reclaim anything and there are no more
2616 * mem cgroups to try or we seem to be looping without
2617 * reclaiming anything.
2618 */
2619 if (!nr_reclaimed &&
2620 (next_mz == NULL ||
2621 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2622 break;
2623 } while (!nr_reclaimed);
2624 if (next_mz)
2625 css_put(&next_mz->memcg->css);
2626 return nr_reclaimed;
2627}
2628
Tejun Heoea280e72014-05-16 13:22:48 -04002629/*
2630 * Test whether @memcg has children, dead or alive. Note that this
2631 * function doesn't care whether @memcg has use_hierarchy enabled and
2632 * returns %true if there are child csses according to the cgroup
2633 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2634 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002635static inline bool memcg_has_children(struct mem_cgroup *memcg)
2636{
Tejun Heoea280e72014-05-16 13:22:48 -04002637 bool ret;
2638
Tejun Heoea280e72014-05-16 13:22:48 -04002639 rcu_read_lock();
2640 ret = css_next_child(NULL, &memcg->css);
2641 rcu_read_unlock();
2642 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002643}
2644
2645/*
Greg Thelen51038172016-05-20 16:58:18 -07002646 * Reclaims as many pages from the given memcg as possible.
Michal Hockoc26251f2012-10-26 13:37:28 +02002647 *
2648 * Caller is responsible for holding css reference for memcg.
2649 */
2650static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2651{
2652 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002653
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002654 /* we call try-to-free pages for make this cgroup empty */
2655 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002656 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002657 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002658 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002659
Michal Hockoc26251f2012-10-26 13:37:28 +02002660 if (signal_pending(current))
2661 return -EINTR;
2662
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002663 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2664 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002665 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002666 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002667 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002668 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002669 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002670
2671 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002672
2673 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002674}
2675
Tejun Heo6770c642014-05-13 12:16:21 -04002676static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2677 char *buf, size_t nbytes,
2678 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002679{
Tejun Heo6770c642014-05-13 12:16:21 -04002680 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002681
Michal Hockod8423012012-10-26 13:37:29 +02002682 if (mem_cgroup_is_root(memcg))
2683 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002684 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002685}
2686
Tejun Heo182446d2013-08-08 20:11:24 -04002687static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2688 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002689{
Tejun Heo182446d2013-08-08 20:11:24 -04002690 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002691}
2692
Tejun Heo182446d2013-08-08 20:11:24 -04002693static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2694 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002695{
2696 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002697 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002698 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002699
Glauber Costa567fb432012-07-31 16:43:07 -07002700 if (memcg->use_hierarchy == val)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002701 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07002702
Balbir Singh18f59ea2009-01-07 18:08:07 -08002703 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002704 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002705 * in the child subtrees. If it is unset, then the change can
2706 * occur, provided the current cgroup has no children.
2707 *
2708 * For the root cgroup, parent_mem is NULL, we allow value to be
2709 * set if there are no children.
2710 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002711 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002712 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002713 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002714 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002715 else
2716 retval = -EBUSY;
2717 } else
2718 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002719
Balbir Singh18f59ea2009-01-07 18:08:07 -08002720 return retval;
2721}
2722
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002723static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
Johannes Weinerce00a962014-09-05 08:43:57 -04002724{
2725 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002726 int i;
Johannes Weinerce00a962014-09-05 08:43:57 -04002727
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002728 memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
Johannes Weinerce00a962014-09-05 08:43:57 -04002729
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002730 for_each_mem_cgroup_tree(iter, memcg) {
2731 for (i = 0; i < MEMCG_NR_STAT; i++)
2732 stat[i] += mem_cgroup_read_stat(iter, i);
2733 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002734}
2735
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002736static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
Johannes Weiner587d9f72016-01-20 15:03:19 -08002737{
2738 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002739 int i;
Johannes Weiner587d9f72016-01-20 15:03:19 -08002740
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002741 memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
Johannes Weiner587d9f72016-01-20 15:03:19 -08002742
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002743 for_each_mem_cgroup_tree(iter, memcg) {
2744 for (i = 0; i < MEMCG_NR_EVENTS; i++)
2745 events[i] += mem_cgroup_read_events(iter, i);
2746 }
Johannes Weiner587d9f72016-01-20 15:03:19 -08002747}
2748
Andrew Morton6f646152015-11-06 16:28:58 -08002749static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04002750{
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002751 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002752
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002753 if (mem_cgroup_is_root(memcg)) {
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002754 struct mem_cgroup *iter;
2755
2756 for_each_mem_cgroup_tree(iter, memcg) {
2757 val += mem_cgroup_read_stat(iter,
2758 MEM_CGROUP_STAT_CACHE);
2759 val += mem_cgroup_read_stat(iter,
2760 MEM_CGROUP_STAT_RSS);
2761 if (swap)
2762 val += mem_cgroup_read_stat(iter,
2763 MEM_CGROUP_STAT_SWAP);
2764 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002765 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002766 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002767 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002768 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002769 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002770 }
Michal Hockoc12176d2015-11-05 18:50:29 -08002771 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002772}
2773
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002774enum {
2775 RES_USAGE,
2776 RES_LIMIT,
2777 RES_MAX_USAGE,
2778 RES_FAILCNT,
2779 RES_SOFT_LIMIT,
2780};
Johannes Weinerce00a962014-09-05 08:43:57 -04002781
Tejun Heo791badb2013-12-05 12:28:02 -05002782static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002783 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002784{
Tejun Heo182446d2013-08-08 20:11:24 -04002785 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002786 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002787
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002788 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002789 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002790 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002791 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002792 case _MEMSWAP:
2793 counter = &memcg->memsw;
2794 break;
2795 case _KMEM:
2796 counter = &memcg->kmem;
2797 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002798 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08002799 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002800 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002801 default:
2802 BUG();
2803 }
2804
2805 switch (MEMFILE_ATTR(cft->private)) {
2806 case RES_USAGE:
2807 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08002808 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002809 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08002810 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002811 return (u64)page_counter_read(counter) * PAGE_SIZE;
2812 case RES_LIMIT:
2813 return (u64)counter->limit * PAGE_SIZE;
2814 case RES_MAX_USAGE:
2815 return (u64)counter->watermark * PAGE_SIZE;
2816 case RES_FAILCNT:
2817 return counter->failcnt;
2818 case RES_SOFT_LIMIT:
2819 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002820 default:
2821 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002822 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002823}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002824
Johannes Weiner127424c2016-01-20 15:02:32 -08002825#ifndef CONFIG_SLOB
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002826static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002827{
Vladimir Davydovd6441632014-01-23 15:53:09 -08002828 int memcg_id;
2829
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002830 if (cgroup_memory_nokmem)
2831 return 0;
2832
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002833 BUG_ON(memcg->kmemcg_id >= 0);
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002834 BUG_ON(memcg->kmem_state);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002835
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002836 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002837 if (memcg_id < 0)
2838 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002839
Johannes Weineref129472016-01-14 15:21:34 -08002840 static_branch_inc(&memcg_kmem_enabled_key);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002841 /*
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002842 * A memory cgroup is considered kmem-online as soon as it gets
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002843 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002844 * guarantee no one starts accounting before all call sites are
2845 * patched.
2846 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002847 memcg->kmemcg_id = memcg_id;
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002848 memcg->kmem_state = KMEM_ONLINE;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002849
2850 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002851}
2852
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002853static void memcg_offline_kmem(struct mem_cgroup *memcg)
2854{
2855 struct cgroup_subsys_state *css;
2856 struct mem_cgroup *parent, *child;
2857 int kmemcg_id;
2858
2859 if (memcg->kmem_state != KMEM_ONLINE)
2860 return;
2861 /*
2862 * Clear the online state before clearing memcg_caches array
2863 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2864 * guarantees that no cache will be created for this cgroup
2865 * after we are done (see memcg_create_kmem_cache()).
2866 */
2867 memcg->kmem_state = KMEM_ALLOCATED;
2868
2869 memcg_deactivate_kmem_caches(memcg);
2870
2871 kmemcg_id = memcg->kmemcg_id;
2872 BUG_ON(kmemcg_id < 0);
2873
2874 parent = parent_mem_cgroup(memcg);
2875 if (!parent)
2876 parent = root_mem_cgroup;
2877
2878 /*
2879 * Change kmemcg_id of this cgroup and all its descendants to the
2880 * parent's id, and then move all entries from this cgroup's list_lrus
2881 * to ones of the parent. After we have finished, all list_lrus
2882 * corresponding to this cgroup are guaranteed to remain empty. The
2883 * ordering is imposed by list_lru_node->lock taken by
2884 * memcg_drain_all_list_lrus().
2885 */
Tejun Heo3a06bb72016-06-03 14:55:44 -07002886 rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002887 css_for_each_descendant_pre(css, &memcg->css) {
2888 child = mem_cgroup_from_css(css);
2889 BUG_ON(child->kmemcg_id != kmemcg_id);
2890 child->kmemcg_id = parent->kmemcg_id;
2891 if (!memcg->use_hierarchy)
2892 break;
2893 }
Tejun Heo3a06bb72016-06-03 14:55:44 -07002894 rcu_read_unlock();
2895
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002896 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2897
2898 memcg_free_cache_id(kmemcg_id);
2899}
2900
2901static void memcg_free_kmem(struct mem_cgroup *memcg)
2902{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002903 /* css_alloc() failed, offlining didn't happen */
2904 if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2905 memcg_offline_kmem(memcg);
2906
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002907 if (memcg->kmem_state == KMEM_ALLOCATED) {
2908 memcg_destroy_kmem_caches(memcg);
2909 static_branch_dec(&memcg_kmem_enabled_key);
2910 WARN_ON(page_counter_read(&memcg->kmem));
2911 }
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002912}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002913#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002914static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08002915{
2916 return 0;
2917}
2918static void memcg_offline_kmem(struct mem_cgroup *memcg)
2919{
2920}
2921static void memcg_free_kmem(struct mem_cgroup *memcg)
2922{
2923}
2924#endif /* !CONFIG_SLOB */
2925
Johannes Weiner127424c2016-01-20 15:02:32 -08002926static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2927 unsigned long limit)
2928{
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002929 int ret;
Johannes Weiner127424c2016-01-20 15:02:32 -08002930
2931 mutex_lock(&memcg_limit_mutex);
Johannes Weiner127424c2016-01-20 15:02:32 -08002932 ret = page_counter_limit(&memcg->kmem, limit);
Johannes Weiner127424c2016-01-20 15:02:32 -08002933 mutex_unlock(&memcg_limit_mutex);
2934 return ret;
2935}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002936
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002937static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2938{
2939 int ret;
2940
2941 mutex_lock(&memcg_limit_mutex);
2942
Johannes Weiner0db15292016-01-20 15:02:50 -08002943 ret = page_counter_limit(&memcg->tcpmem, limit);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002944 if (ret)
2945 goto out;
2946
Johannes Weiner0db15292016-01-20 15:02:50 -08002947 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002948 /*
2949 * The active flag needs to be written after the static_key
2950 * update. This is what guarantees that the socket activation
Johannes Weiner2d758072016-10-07 17:00:58 -07002951 * function is the last one to run. See mem_cgroup_sk_alloc()
2952 * for details, and note that we don't mark any socket as
2953 * belonging to this memcg until that flag is up.
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002954 *
2955 * We need to do this, because static_keys will span multiple
2956 * sites, but we can't control their order. If we mark a socket
2957 * as accounted, but the accounting functions are not patched in
2958 * yet, we'll lose accounting.
2959 *
Johannes Weiner2d758072016-10-07 17:00:58 -07002960 * We never race with the readers in mem_cgroup_sk_alloc(),
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002961 * because when this value change, the code to process it is not
2962 * patched in yet.
2963 */
2964 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08002965 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002966 }
2967out:
2968 mutex_unlock(&memcg_limit_mutex);
2969 return ret;
2970}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002971
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002972/*
2973 * The user of this function is...
2974 * RES_LIMIT.
2975 */
Tejun Heo451af502014-05-13 12:16:21 -04002976static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2977 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002978{
Tejun Heo451af502014-05-13 12:16:21 -04002979 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002980 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002981 int ret;
2982
Tejun Heo451af502014-05-13 12:16:21 -04002983 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08002984 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002985 if (ret)
2986 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07002987
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002988 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002989 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07002990 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2991 ret = -EINVAL;
2992 break;
2993 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002994 switch (MEMFILE_TYPE(of_cft(of)->private)) {
2995 case _MEM:
2996 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002997 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002998 case _MEMSWAP:
2999 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3000 break;
3001 case _KMEM:
3002 ret = memcg_update_kmem_limit(memcg, nr_pages);
3003 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003004 case _TCP:
3005 ret = memcg_update_tcp_limit(memcg, nr_pages);
3006 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003007 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003008 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003009 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003010 memcg->soft_limit = nr_pages;
3011 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003012 break;
3013 }
Tejun Heo451af502014-05-13 12:16:21 -04003014 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003015}
3016
Tejun Heo6770c642014-05-13 12:16:21 -04003017static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3018 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003019{
Tejun Heo6770c642014-05-13 12:16:21 -04003020 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003021 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003022
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003023 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3024 case _MEM:
3025 counter = &memcg->memory;
3026 break;
3027 case _MEMSWAP:
3028 counter = &memcg->memsw;
3029 break;
3030 case _KMEM:
3031 counter = &memcg->kmem;
3032 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003033 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003034 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003035 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003036 default:
3037 BUG();
3038 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003039
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003040 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003041 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003042 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003043 break;
3044 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003045 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003046 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003047 default:
3048 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003049 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003050
Tejun Heo6770c642014-05-13 12:16:21 -04003051 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003052}
3053
Tejun Heo182446d2013-08-08 20:11:24 -04003054static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003055 struct cftype *cft)
3056{
Tejun Heo182446d2013-08-08 20:11:24 -04003057 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003058}
3059
Daisuke Nishimura02491442010-03-10 15:22:17 -08003060#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003061static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003062 struct cftype *cft, u64 val)
3063{
Tejun Heo182446d2013-08-08 20:11:24 -04003064 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003065
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003066 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003067 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003068
Glauber Costaee5e8472013-02-22 16:34:50 -08003069 /*
3070 * No kind of locking is needed in here, because ->can_attach() will
3071 * check this value once in the beginning of the process, and then carry
3072 * on with stale data. This means that changes to this value will only
3073 * affect task migrations starting after the change.
3074 */
3075 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003076 return 0;
3077}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003078#else
Tejun Heo182446d2013-08-08 20:11:24 -04003079static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003080 struct cftype *cft, u64 val)
3081{
3082 return -ENOSYS;
3083}
3084#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003085
Ying Han406eb0c2011-05-26 16:25:37 -07003086#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003087static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003088{
Greg Thelen25485de2013-11-12 15:07:40 -08003089 struct numa_stat {
3090 const char *name;
3091 unsigned int lru_mask;
3092 };
3093
3094 static const struct numa_stat stats[] = {
3095 { "total", LRU_ALL },
3096 { "file", LRU_ALL_FILE },
3097 { "anon", LRU_ALL_ANON },
3098 { "unevictable", BIT(LRU_UNEVICTABLE) },
3099 };
3100 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003101 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003102 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003103 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003104
Greg Thelen25485de2013-11-12 15:07:40 -08003105 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3106 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3107 seq_printf(m, "%s=%lu", stat->name, nr);
3108 for_each_node_state(nid, N_MEMORY) {
3109 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3110 stat->lru_mask);
3111 seq_printf(m, " N%d=%lu", nid, nr);
3112 }
3113 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003114 }
Ying Han406eb0c2011-05-26 16:25:37 -07003115
Ying Han071aee12013-11-12 15:07:41 -08003116 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3117 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003118
Ying Han071aee12013-11-12 15:07:41 -08003119 nr = 0;
3120 for_each_mem_cgroup_tree(iter, memcg)
3121 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3122 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3123 for_each_node_state(nid, N_MEMORY) {
3124 nr = 0;
3125 for_each_mem_cgroup_tree(iter, memcg)
3126 nr += mem_cgroup_node_nr_lru_pages(
3127 iter, nid, stat->lru_mask);
3128 seq_printf(m, " N%d=%lu", nid, nr);
3129 }
3130 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003131 }
Ying Han406eb0c2011-05-26 16:25:37 -07003132
Ying Han406eb0c2011-05-26 16:25:37 -07003133 return 0;
3134}
3135#endif /* CONFIG_NUMA */
3136
Tejun Heo2da8ca82013-12-05 12:28:04 -05003137static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003138{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003139 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003140 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003141 struct mem_cgroup *mi;
3142 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003143
Greg Thelen0ca44b12015-02-11 15:25:58 -08003144 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3145 MEM_CGROUP_STAT_NSTATS);
3146 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3147 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003148 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3149
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003150 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Johannes Weiner7941d212016-01-14 15:21:23 -08003151 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003152 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07003153 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003154 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003155 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003156
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003157 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3158 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3159 mem_cgroup_read_events(memcg, i));
3160
3161 for (i = 0; i < NR_LRU_LISTS; i++)
3162 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3163 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3164
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003165 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003166 memory = memsw = PAGE_COUNTER_MAX;
3167 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3168 memory = min(memory, mi->memory.limit);
3169 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003170 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003171 seq_printf(m, "hierarchical_memory_limit %llu\n",
3172 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08003173 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003174 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3175 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003176
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003177 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003178 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003179
Johannes Weiner7941d212016-01-14 15:21:23 -08003180 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003181 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003182 for_each_mem_cgroup_tree(mi, memcg)
3183 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
Greg Thelen484ebb32015-10-01 15:37:05 -07003184 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003185 }
3186
3187 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3188 unsigned long long val = 0;
3189
3190 for_each_mem_cgroup_tree(mi, memcg)
3191 val += mem_cgroup_read_events(mi, i);
3192 seq_printf(m, "total_%s %llu\n",
3193 mem_cgroup_events_names[i], val);
3194 }
3195
3196 for (i = 0; i < NR_LRU_LISTS; i++) {
3197 unsigned long long val = 0;
3198
3199 for_each_mem_cgroup_tree(mi, memcg)
3200 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3201 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003202 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003203
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003204#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003205 {
Mel Gormanef8f2322016-07-28 15:46:05 -07003206 pg_data_t *pgdat;
3207 struct mem_cgroup_per_node *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003208 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003209 unsigned long recent_rotated[2] = {0, 0};
3210 unsigned long recent_scanned[2] = {0, 0};
3211
Mel Gormanef8f2322016-07-28 15:46:05 -07003212 for_each_online_pgdat(pgdat) {
3213 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
3214 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003215
Mel Gormanef8f2322016-07-28 15:46:05 -07003216 recent_rotated[0] += rstat->recent_rotated[0];
3217 recent_rotated[1] += rstat->recent_rotated[1];
3218 recent_scanned[0] += rstat->recent_scanned[0];
3219 recent_scanned[1] += rstat->recent_scanned[1];
3220 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003221 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3222 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3223 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3224 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003225 }
3226#endif
3227
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003228 return 0;
3229}
3230
Tejun Heo182446d2013-08-08 20:11:24 -04003231static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3232 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003233{
Tejun Heo182446d2013-08-08 20:11:24 -04003234 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003235
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003236 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003237}
3238
Tejun Heo182446d2013-08-08 20:11:24 -04003239static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3240 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003241{
Tejun Heo182446d2013-08-08 20:11:24 -04003242 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003243
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003244 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003245 return -EINVAL;
3246
Linus Torvalds14208b02014-06-09 15:03:33 -07003247 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003248 memcg->swappiness = val;
3249 else
3250 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003251
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003252 return 0;
3253}
3254
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003255static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3256{
3257 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003258 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003259 int i;
3260
3261 rcu_read_lock();
3262 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003263 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003264 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003265 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003266
3267 if (!t)
3268 goto unlock;
3269
Johannes Weinerce00a962014-09-05 08:43:57 -04003270 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003271
3272 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003273 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003274 * If it's not true, a threshold was crossed after last
3275 * call of __mem_cgroup_threshold().
3276 */
Phil Carmody5407a562010-05-26 14:42:42 -07003277 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003278
3279 /*
3280 * Iterate backward over array of thresholds starting from
3281 * current_threshold and check if a threshold is crossed.
3282 * If none of thresholds below usage is crossed, we read
3283 * only one element of the array here.
3284 */
3285 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3286 eventfd_signal(t->entries[i].eventfd, 1);
3287
3288 /* i = current_threshold + 1 */
3289 i++;
3290
3291 /*
3292 * Iterate forward over array of thresholds starting from
3293 * current_threshold+1 and check if a threshold is crossed.
3294 * If none of thresholds above usage is crossed, we read
3295 * only one element of the array here.
3296 */
3297 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3298 eventfd_signal(t->entries[i].eventfd, 1);
3299
3300 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003301 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003302unlock:
3303 rcu_read_unlock();
3304}
3305
3306static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3307{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003308 while (memcg) {
3309 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08003310 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003311 __mem_cgroup_threshold(memcg, true);
3312
3313 memcg = parent_mem_cgroup(memcg);
3314 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003315}
3316
3317static int compare_thresholds(const void *a, const void *b)
3318{
3319 const struct mem_cgroup_threshold *_a = a;
3320 const struct mem_cgroup_threshold *_b = b;
3321
Greg Thelen2bff24a2013-09-11 14:23:08 -07003322 if (_a->threshold > _b->threshold)
3323 return 1;
3324
3325 if (_a->threshold < _b->threshold)
3326 return -1;
3327
3328 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003329}
3330
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003331static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003332{
3333 struct mem_cgroup_eventfd_list *ev;
3334
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003335 spin_lock(&memcg_oom_lock);
3336
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003337 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003338 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003339
3340 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003341 return 0;
3342}
3343
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003344static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003345{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003346 struct mem_cgroup *iter;
3347
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003348 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003349 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003350}
3351
Tejun Heo59b6f872013-11-22 18:20:43 -05003352static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003353 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003354{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003355 struct mem_cgroup_thresholds *thresholds;
3356 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003357 unsigned long threshold;
3358 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003359 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003360
Johannes Weiner650c5e52015-02-11 15:26:03 -08003361 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003362 if (ret)
3363 return ret;
3364
3365 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003366
Johannes Weiner05b84302014-08-06 16:05:59 -07003367 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003368 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003369 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003370 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003371 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003372 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003373 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003374 BUG();
3375
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003376 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003377 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003378 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3379
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003380 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003381
3382 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003383 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003384 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003385 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003386 ret = -ENOMEM;
3387 goto unlock;
3388 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003389 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003390
3391 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003392 if (thresholds->primary) {
3393 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003394 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003395 }
3396
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003397 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003398 new->entries[size - 1].eventfd = eventfd;
3399 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003400
3401 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003402 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003403 compare_thresholds, NULL);
3404
3405 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003406 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003407 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003408 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003409 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003410 * new->current_threshold will not be used until
3411 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003412 * it here.
3413 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003414 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003415 } else
3416 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003417 }
3418
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003419 /* Free old spare buffer and save old primary buffer as spare */
3420 kfree(thresholds->spare);
3421 thresholds->spare = thresholds->primary;
3422
3423 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003424
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003425 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003426 synchronize_rcu();
3427
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003428unlock:
3429 mutex_unlock(&memcg->thresholds_lock);
3430
3431 return ret;
3432}
3433
Tejun Heo59b6f872013-11-22 18:20:43 -05003434static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003435 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003436{
Tejun Heo59b6f872013-11-22 18:20:43 -05003437 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003438}
3439
Tejun Heo59b6f872013-11-22 18:20:43 -05003440static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003441 struct eventfd_ctx *eventfd, const char *args)
3442{
Tejun Heo59b6f872013-11-22 18:20:43 -05003443 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003444}
3445
Tejun Heo59b6f872013-11-22 18:20:43 -05003446static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003447 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003448{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003449 struct mem_cgroup_thresholds *thresholds;
3450 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003451 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003452 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003453
3454 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003455
3456 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003457 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003458 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003459 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003460 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003461 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003462 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003463 BUG();
3464
Anton Vorontsov371528c2012-02-24 05:14:46 +04003465 if (!thresholds->primary)
3466 goto unlock;
3467
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003468 /* Check if a threshold crossed before removing */
3469 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3470
3471 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003472 size = 0;
3473 for (i = 0; i < thresholds->primary->size; i++) {
3474 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003475 size++;
3476 }
3477
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003478 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003479
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003480 /* Set thresholds array to NULL if we don't have thresholds */
3481 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003482 kfree(new);
3483 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003484 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003485 }
3486
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003487 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003488
3489 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003490 new->current_threshold = -1;
3491 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3492 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003493 continue;
3494
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003495 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003496 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003497 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003498 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003499 * until rcu_assign_pointer(), so it's safe to increment
3500 * it here.
3501 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003502 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003503 }
3504 j++;
3505 }
3506
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003507swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003508 /* Swap primary and spare array */
3509 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003510
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003511 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003512
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003513 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003514 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08003515
3516 /* If all events are unregistered, free the spare array */
3517 if (!new) {
3518 kfree(thresholds->spare);
3519 thresholds->spare = NULL;
3520 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04003521unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003522 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003523}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003524
Tejun Heo59b6f872013-11-22 18:20:43 -05003525static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003526 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003527{
Tejun Heo59b6f872013-11-22 18:20:43 -05003528 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003529}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003530
Tejun Heo59b6f872013-11-22 18:20:43 -05003531static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003532 struct eventfd_ctx *eventfd)
3533{
Tejun Heo59b6f872013-11-22 18:20:43 -05003534 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003535}
3536
Tejun Heo59b6f872013-11-22 18:20:43 -05003537static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003538 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003539{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003540 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003541
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003542 event = kmalloc(sizeof(*event), GFP_KERNEL);
3543 if (!event)
3544 return -ENOMEM;
3545
Michal Hocko1af8efe2011-07-26 16:08:24 -07003546 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003547
3548 event->eventfd = eventfd;
3549 list_add(&event->list, &memcg->oom_notify);
3550
3551 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07003552 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003553 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003554 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003555
3556 return 0;
3557}
3558
Tejun Heo59b6f872013-11-22 18:20:43 -05003559static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003560 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003561{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003562 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003563
Michal Hocko1af8efe2011-07-26 16:08:24 -07003564 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003565
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003566 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003567 if (ev->eventfd == eventfd) {
3568 list_del(&ev->list);
3569 kfree(ev);
3570 }
3571 }
3572
Michal Hocko1af8efe2011-07-26 16:08:24 -07003573 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003574}
3575
Tejun Heo2da8ca82013-12-05 12:28:04 -05003576static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003577{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003578 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003579
Tejun Heo791badb2013-12-05 12:28:02 -05003580 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07003581 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003582 return 0;
3583}
3584
Tejun Heo182446d2013-08-08 20:11:24 -04003585static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003586 struct cftype *cft, u64 val)
3587{
Tejun Heo182446d2013-08-08 20:11:24 -04003588 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003589
3590 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003591 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003592 return -EINVAL;
3593
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003594 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003595 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003596 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003597
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003598 return 0;
3599}
3600
Tejun Heo52ebea72015-05-22 17:13:37 -04003601#ifdef CONFIG_CGROUP_WRITEBACK
3602
3603struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3604{
3605 return &memcg->cgwb_list;
3606}
3607
Tejun Heo841710a2015-05-22 18:23:33 -04003608static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3609{
3610 return wb_domain_init(&memcg->cgwb_domain, gfp);
3611}
3612
3613static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3614{
3615 wb_domain_exit(&memcg->cgwb_domain);
3616}
3617
Tejun Heo2529bb32015-05-22 18:23:34 -04003618static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3619{
3620 wb_domain_size_changed(&memcg->cgwb_domain);
3621}
3622
Tejun Heo841710a2015-05-22 18:23:33 -04003623struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3624{
3625 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3626
3627 if (!memcg->css.parent)
3628 return NULL;
3629
3630 return &memcg->cgwb_domain;
3631}
3632
Tejun Heoc2aa7232015-05-22 18:23:35 -04003633/**
3634 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3635 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003636 * @pfilepages: out parameter for number of file pages
3637 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04003638 * @pdirty: out parameter for number of dirty pages
3639 * @pwriteback: out parameter for number of pages under writeback
3640 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003641 * Determine the numbers of file, headroom, dirty, and writeback pages in
3642 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3643 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003644 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003645 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3646 * headroom is calculated as the lowest headroom of itself and the
3647 * ancestors. Note that this doesn't consider the actual amount of
3648 * available memory in the system. The caller should further cap
3649 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003650 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003651void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3652 unsigned long *pheadroom, unsigned long *pdirty,
3653 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04003654{
3655 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3656 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003657
3658 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3659
3660 /* this should eventually include NR_UNSTABLE_NFS */
3661 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003662 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3663 (1 << LRU_ACTIVE_FILE));
3664 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003665
Tejun Heoc2aa7232015-05-22 18:23:35 -04003666 while ((parent = parent_mem_cgroup(memcg))) {
3667 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3668 unsigned long used = page_counter_read(&memcg->memory);
3669
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003670 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04003671 memcg = parent;
3672 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04003673}
3674
Tejun Heo841710a2015-05-22 18:23:33 -04003675#else /* CONFIG_CGROUP_WRITEBACK */
3676
3677static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3678{
3679 return 0;
3680}
3681
3682static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3683{
3684}
3685
Tejun Heo2529bb32015-05-22 18:23:34 -04003686static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3687{
3688}
3689
Tejun Heo52ebea72015-05-22 17:13:37 -04003690#endif /* CONFIG_CGROUP_WRITEBACK */
3691
Tejun Heo79bd9812013-11-22 18:20:42 -05003692/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003693 * DO NOT USE IN NEW FILES.
3694 *
3695 * "cgroup.event_control" implementation.
3696 *
3697 * This is way over-engineered. It tries to support fully configurable
3698 * events for each user. Such level of flexibility is completely
3699 * unnecessary especially in the light of the planned unified hierarchy.
3700 *
3701 * Please deprecate this and replace with something simpler if at all
3702 * possible.
3703 */
3704
3705/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003706 * Unregister event and free resources.
3707 *
3708 * Gets called from workqueue.
3709 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003710static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003711{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003712 struct mem_cgroup_event *event =
3713 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003714 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003715
3716 remove_wait_queue(event->wqh, &event->wait);
3717
Tejun Heo59b6f872013-11-22 18:20:43 -05003718 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003719
3720 /* Notify userspace the event is going away. */
3721 eventfd_signal(event->eventfd, 1);
3722
3723 eventfd_ctx_put(event->eventfd);
3724 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003725 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003726}
3727
3728/*
3729 * Gets called on POLLHUP on eventfd when user closes it.
3730 *
3731 * Called with wqh->lock held and interrupts disabled.
3732 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003733static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3734 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003735{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003736 struct mem_cgroup_event *event =
3737 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003738 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003739 unsigned long flags = (unsigned long)key;
3740
3741 if (flags & POLLHUP) {
3742 /*
3743 * If the event has been detached at cgroup removal, we
3744 * can simply return knowing the other side will cleanup
3745 * for us.
3746 *
3747 * We can't race against event freeing since the other
3748 * side will require wqh->lock via remove_wait_queue(),
3749 * which we hold.
3750 */
Tejun Heofba94802013-11-22 18:20:43 -05003751 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003752 if (!list_empty(&event->list)) {
3753 list_del_init(&event->list);
3754 /*
3755 * We are in atomic context, but cgroup_event_remove()
3756 * may sleep, so we have to call it in workqueue.
3757 */
3758 schedule_work(&event->remove);
3759 }
Tejun Heofba94802013-11-22 18:20:43 -05003760 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003761 }
3762
3763 return 0;
3764}
3765
Tejun Heo3bc942f2013-11-22 18:20:44 -05003766static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003767 wait_queue_head_t *wqh, poll_table *pt)
3768{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003769 struct mem_cgroup_event *event =
3770 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003771
3772 event->wqh = wqh;
3773 add_wait_queue(wqh, &event->wait);
3774}
3775
3776/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003777 * DO NOT USE IN NEW FILES.
3778 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003779 * Parse input and register new cgroup event handler.
3780 *
3781 * Input must be in format '<event_fd> <control_fd> <args>'.
3782 * Interpretation of args is defined by control file implementation.
3783 */
Tejun Heo451af502014-05-13 12:16:21 -04003784static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3785 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003786{
Tejun Heo451af502014-05-13 12:16:21 -04003787 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003788 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003789 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003790 struct cgroup_subsys_state *cfile_css;
3791 unsigned int efd, cfd;
3792 struct fd efile;
3793 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003794 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003795 char *endp;
3796 int ret;
3797
Tejun Heo451af502014-05-13 12:16:21 -04003798 buf = strstrip(buf);
3799
3800 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003801 if (*endp != ' ')
3802 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003803 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003804
Tejun Heo451af502014-05-13 12:16:21 -04003805 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003806 if ((*endp != ' ') && (*endp != '\0'))
3807 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003808 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003809
3810 event = kzalloc(sizeof(*event), GFP_KERNEL);
3811 if (!event)
3812 return -ENOMEM;
3813
Tejun Heo59b6f872013-11-22 18:20:43 -05003814 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003815 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003816 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3817 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3818 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003819
3820 efile = fdget(efd);
3821 if (!efile.file) {
3822 ret = -EBADF;
3823 goto out_kfree;
3824 }
3825
3826 event->eventfd = eventfd_ctx_fileget(efile.file);
3827 if (IS_ERR(event->eventfd)) {
3828 ret = PTR_ERR(event->eventfd);
3829 goto out_put_efile;
3830 }
3831
3832 cfile = fdget(cfd);
3833 if (!cfile.file) {
3834 ret = -EBADF;
3835 goto out_put_eventfd;
3836 }
3837
3838 /* the process need read permission on control file */
3839 /* AV: shouldn't we check that it's been opened for read instead? */
3840 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3841 if (ret < 0)
3842 goto out_put_cfile;
3843
Tejun Heo79bd9812013-11-22 18:20:42 -05003844 /*
Tejun Heofba94802013-11-22 18:20:43 -05003845 * Determine the event callbacks and set them in @event. This used
3846 * to be done via struct cftype but cgroup core no longer knows
3847 * about these events. The following is crude but the whole thing
3848 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003849 *
3850 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003851 */
Al Virob5830432014-10-31 01:22:04 -04003852 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003853
3854 if (!strcmp(name, "memory.usage_in_bytes")) {
3855 event->register_event = mem_cgroup_usage_register_event;
3856 event->unregister_event = mem_cgroup_usage_unregister_event;
3857 } else if (!strcmp(name, "memory.oom_control")) {
3858 event->register_event = mem_cgroup_oom_register_event;
3859 event->unregister_event = mem_cgroup_oom_unregister_event;
3860 } else if (!strcmp(name, "memory.pressure_level")) {
3861 event->register_event = vmpressure_register_event;
3862 event->unregister_event = vmpressure_unregister_event;
3863 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003864 event->register_event = memsw_cgroup_usage_register_event;
3865 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003866 } else {
3867 ret = -EINVAL;
3868 goto out_put_cfile;
3869 }
3870
3871 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003872 * Verify @cfile should belong to @css. Also, remaining events are
3873 * automatically removed on cgroup destruction but the removal is
3874 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003875 */
Al Virob5830432014-10-31 01:22:04 -04003876 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003877 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003878 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003879 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003880 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003881 if (cfile_css != css) {
3882 css_put(cfile_css);
3883 goto out_put_cfile;
3884 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003885
Tejun Heo451af502014-05-13 12:16:21 -04003886 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003887 if (ret)
3888 goto out_put_css;
3889
3890 efile.file->f_op->poll(efile.file, &event->pt);
3891
Tejun Heofba94802013-11-22 18:20:43 -05003892 spin_lock(&memcg->event_list_lock);
3893 list_add(&event->list, &memcg->event_list);
3894 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003895
3896 fdput(cfile);
3897 fdput(efile);
3898
Tejun Heo451af502014-05-13 12:16:21 -04003899 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05003900
3901out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05003902 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003903out_put_cfile:
3904 fdput(cfile);
3905out_put_eventfd:
3906 eventfd_ctx_put(event->eventfd);
3907out_put_efile:
3908 fdput(efile);
3909out_kfree:
3910 kfree(event);
3911
3912 return ret;
3913}
3914
Johannes Weiner241994e2015-02-11 15:26:06 -08003915static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003916 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003917 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003918 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003919 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003920 },
3921 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003922 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003923 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04003924 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003925 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003926 },
3927 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003928 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003929 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003930 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003931 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003932 },
3933 {
Balbir Singh296c81d2009-09-23 15:56:36 -07003934 .name = "soft_limit_in_bytes",
3935 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003936 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003937 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07003938 },
3939 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003940 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003941 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04003942 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003943 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003944 },
Balbir Singh8697d332008-02-07 00:13:59 -08003945 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003946 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003947 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003948 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003949 {
3950 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04003951 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003952 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08003953 {
3954 .name = "use_hierarchy",
3955 .write_u64 = mem_cgroup_hierarchy_write,
3956 .read_u64 = mem_cgroup_hierarchy_read,
3957 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003958 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05003959 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04003960 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04003961 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05003962 },
3963 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003964 .name = "swappiness",
3965 .read_u64 = mem_cgroup_swappiness_read,
3966 .write_u64 = mem_cgroup_swappiness_write,
3967 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003968 {
3969 .name = "move_charge_at_immigrate",
3970 .read_u64 = mem_cgroup_move_charge_read,
3971 .write_u64 = mem_cgroup_move_charge_write,
3972 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003973 {
3974 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003975 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003976 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003977 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3978 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003979 {
3980 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003981 },
Ying Han406eb0c2011-05-26 16:25:37 -07003982#ifdef CONFIG_NUMA
3983 {
3984 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003985 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07003986 },
3987#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08003988 {
3989 .name = "kmem.limit_in_bytes",
3990 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003991 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003992 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08003993 },
3994 {
3995 .name = "kmem.usage_in_bytes",
3996 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003997 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08003998 },
3999 {
4000 .name = "kmem.failcnt",
4001 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004002 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004003 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004004 },
4005 {
4006 .name = "kmem.max_usage_in_bytes",
4007 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004008 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004009 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004010 },
Glauber Costa749c5412012-12-18 14:23:01 -08004011#ifdef CONFIG_SLABINFO
4012 {
4013 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004014 .seq_start = slab_start,
4015 .seq_next = slab_next,
4016 .seq_stop = slab_stop,
4017 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004018 },
4019#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004020 {
4021 .name = "kmem.tcp.limit_in_bytes",
4022 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4023 .write = mem_cgroup_write,
4024 .read_u64 = mem_cgroup_read_u64,
4025 },
4026 {
4027 .name = "kmem.tcp.usage_in_bytes",
4028 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4029 .read_u64 = mem_cgroup_read_u64,
4030 },
4031 {
4032 .name = "kmem.tcp.failcnt",
4033 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4034 .write = mem_cgroup_reset,
4035 .read_u64 = mem_cgroup_read_u64,
4036 },
4037 {
4038 .name = "kmem.tcp.max_usage_in_bytes",
4039 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4040 .write = mem_cgroup_reset,
4041 .read_u64 = mem_cgroup_read_u64,
4042 },
Tejun Heo6bc10342012-04-01 12:09:55 -07004043 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004044};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004045
Johannes Weiner73f576c2016-07-20 15:44:57 -07004046/*
4047 * Private memory cgroup IDR
4048 *
4049 * Swap-out records and page cache shadow entries need to store memcg
4050 * references in constrained space, so we maintain an ID space that is
4051 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4052 * memory-controlled cgroups to 64k.
4053 *
4054 * However, there usually are many references to the oflline CSS after
4055 * the cgroup has been destroyed, such as page cache or reclaimable
4056 * slab objects, that don't need to hang on to the ID. We want to keep
4057 * those dead CSS from occupying IDs, or we might quickly exhaust the
4058 * relatively small ID space and prevent the creation of new cgroups
4059 * even when there are much fewer than 64k cgroups - possibly none.
4060 *
4061 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4062 * be freed and recycled when it's no longer needed, which is usually
4063 * when the CSS is offlined.
4064 *
4065 * The only exception to that are records of swapped out tmpfs/shmem
4066 * pages that need to be attributed to live ancestors on swapin. But
4067 * those references are manageable from userspace.
4068 */
4069
4070static DEFINE_IDR(mem_cgroup_idr);
4071
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004072static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004073{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004074 VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0);
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004075 atomic_add(n, &memcg->id.ref);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004076}
4077
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004078static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004079{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004080 VM_BUG_ON(atomic_read(&memcg->id.ref) < n);
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004081 if (atomic_sub_and_test(n, &memcg->id.ref)) {
Johannes Weiner73f576c2016-07-20 15:44:57 -07004082 idr_remove(&mem_cgroup_idr, memcg->id.id);
4083 memcg->id.id = 0;
4084
4085 /* Memcg ID pins CSS */
4086 css_put(&memcg->css);
4087 }
4088}
4089
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004090static inline void mem_cgroup_id_get(struct mem_cgroup *memcg)
4091{
4092 mem_cgroup_id_get_many(memcg, 1);
4093}
4094
4095static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
4096{
4097 mem_cgroup_id_put_many(memcg, 1);
4098}
4099
Johannes Weiner73f576c2016-07-20 15:44:57 -07004100/**
4101 * mem_cgroup_from_id - look up a memcg from a memcg id
4102 * @id: the memcg id to look up
4103 *
4104 * Caller must hold rcu_read_lock().
4105 */
4106struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
4107{
4108 WARN_ON_ONCE(!rcu_read_lock_held());
4109 return idr_find(&mem_cgroup_idr, id);
4110}
4111
Mel Gormanef8f2322016-07-28 15:46:05 -07004112static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004113{
4114 struct mem_cgroup_per_node *pn;
Mel Gormanef8f2322016-07-28 15:46:05 -07004115 int tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004116 /*
4117 * This routine is called against possible nodes.
4118 * But it's BUG to call kmalloc() against offline node.
4119 *
4120 * TODO: this routine can waste much memory for nodes which will
4121 * never be onlined. It's better to use memory hotplug callback
4122 * function.
4123 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004124 if (!node_state(node, N_NORMAL_MEMORY))
4125 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004126 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004127 if (!pn)
4128 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004129
Mel Gormanef8f2322016-07-28 15:46:05 -07004130 lruvec_init(&pn->lruvec);
4131 pn->usage_in_excess = 0;
4132 pn->on_tree = false;
4133 pn->memcg = memcg;
4134
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004135 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004136 return 0;
4137}
4138
Mel Gormanef8f2322016-07-28 15:46:05 -07004139static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004140{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004141 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004142}
4143
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004144static void mem_cgroup_free(struct mem_cgroup *memcg)
4145{
4146 int node;
4147
4148 memcg_wb_domain_exit(memcg);
4149 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004150 free_mem_cgroup_per_node_info(memcg, node);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004151 free_percpu(memcg->stat);
4152 kfree(memcg);
4153}
4154
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004155static struct mem_cgroup *mem_cgroup_alloc(void)
4156{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004157 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004158 size_t size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004159 int node;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004160
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004161 size = sizeof(struct mem_cgroup);
4162 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004163
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004164 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004165 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004166 return NULL;
4167
Johannes Weiner73f576c2016-07-20 15:44:57 -07004168 memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
4169 1, MEM_CGROUP_ID_MAX,
4170 GFP_KERNEL);
4171 if (memcg->id.id < 0)
4172 goto fail;
4173
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004174 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4175 if (!memcg->stat)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004176 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004177
Bob Liu3ed28fa2012-01-12 17:19:04 -08004178 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004179 if (alloc_mem_cgroup_per_node_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004180 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004181
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004182 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4183 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004184
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004185 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08004186 memcg->last_scanned_node = MAX_NUMNODES;
4187 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004188 mutex_init(&memcg->thresholds_lock);
4189 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004190 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004191 INIT_LIST_HEAD(&memcg->event_list);
4192 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004193 memcg->socket_pressure = jiffies;
Johannes Weiner127424c2016-01-20 15:02:32 -08004194#ifndef CONFIG_SLOB
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004195 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004196#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004197#ifdef CONFIG_CGROUP_WRITEBACK
4198 INIT_LIST_HEAD(&memcg->cgwb_list);
4199#endif
Johannes Weiner73f576c2016-07-20 15:44:57 -07004200 idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004201 return memcg;
4202fail:
Johannes Weiner73f576c2016-07-20 15:44:57 -07004203 if (memcg->id.id > 0)
4204 idr_remove(&mem_cgroup_idr, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004205 mem_cgroup_free(memcg);
4206 return NULL;
Glauber Costad142e3e2013-02-22 16:34:52 -08004207}
4208
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004209static struct cgroup_subsys_state * __ref
4210mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004211{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004212 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4213 struct mem_cgroup *memcg;
4214 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08004215
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004216 memcg = mem_cgroup_alloc();
4217 if (!memcg)
4218 return ERR_PTR(error);
Li Zefan4219b2d2013-09-23 16:56:29 +08004219
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004220 memcg->high = PAGE_COUNTER_MAX;
4221 memcg->soft_limit = PAGE_COUNTER_MAX;
4222 if (parent) {
4223 memcg->swappiness = mem_cgroup_swappiness(parent);
4224 memcg->oom_kill_disable = parent->oom_kill_disable;
4225 }
4226 if (parent && parent->use_hierarchy) {
4227 memcg->use_hierarchy = true;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004228 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004229 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004230 page_counter_init(&memcg->memsw, &parent->memsw);
4231 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08004232 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004233 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004234 page_counter_init(&memcg->memory, NULL);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004235 page_counter_init(&memcg->swap, NULL);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004236 page_counter_init(&memcg->memsw, NULL);
4237 page_counter_init(&memcg->kmem, NULL);
Johannes Weiner0db15292016-01-20 15:02:50 -08004238 page_counter_init(&memcg->tcpmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004239 /*
4240 * Deeper hierachy with use_hierarchy == false doesn't make
4241 * much sense so let cgroup subsystem know about this
4242 * unfortunate state in our controller.
4243 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004244 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004245 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004246 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08004247
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004248 /* The following stuff does not apply to the root */
4249 if (!parent) {
4250 root_mem_cgroup = memcg;
4251 return &memcg->css;
4252 }
4253
Vladimir Davydovb313aee2016-03-17 14:18:27 -07004254 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004255 if (error)
4256 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08004257
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004258 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004259 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004260
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004261 return &memcg->css;
4262fail:
4263 mem_cgroup_free(memcg);
Tejun Heoea3a9642016-06-24 14:49:58 -07004264 return ERR_PTR(-ENOMEM);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004265}
4266
Johannes Weiner73f576c2016-07-20 15:44:57 -07004267static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004268{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004269 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4270
Johannes Weiner73f576c2016-07-20 15:44:57 -07004271 /* Online state pins memcg ID, memcg ID pins CSS */
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004272 atomic_set(&memcg->id.ref, 1);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004273 css_get(css);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004274 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004275}
4276
Tejun Heoeb954192013-08-08 20:11:23 -04004277static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004278{
Tejun Heoeb954192013-08-08 20:11:23 -04004279 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004280 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004281
4282 /*
4283 * Unregister events and notify userspace.
4284 * Notify userspace about cgroup removing only after rmdir of cgroup
4285 * directory to avoid race between userspace and kernelspace.
4286 */
Tejun Heofba94802013-11-22 18:20:43 -05004287 spin_lock(&memcg->event_list_lock);
4288 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004289 list_del_init(&event->list);
4290 schedule_work(&event->remove);
4291 }
Tejun Heofba94802013-11-22 18:20:43 -05004292 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004293
Johannes Weiner567e9ab2016-01-20 15:02:24 -08004294 memcg_offline_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004295 wb_memcg_offline(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004296
4297 mem_cgroup_id_put(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004298}
4299
Vladimir Davydov6df38682015-12-29 14:54:10 -08004300static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4301{
4302 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4303
4304 invalidate_reclaim_iterators(memcg);
4305}
4306
Tejun Heoeb954192013-08-08 20:11:23 -04004307static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004308{
Tejun Heoeb954192013-08-08 20:11:23 -04004309 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004310
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004311 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004312 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004313
Johannes Weiner0db15292016-01-20 15:02:50 -08004314 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004315 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004316
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004317 vmpressure_cleanup(&memcg->vmpressure);
4318 cancel_work_sync(&memcg->high_work);
4319 mem_cgroup_remove_from_trees(memcg);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004320 memcg_free_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004321 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004322}
4323
Tejun Heo1ced9532014-07-08 18:02:57 -04004324/**
4325 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4326 * @css: the target css
4327 *
4328 * Reset the states of the mem_cgroup associated with @css. This is
4329 * invoked when the userland requests disabling on the default hierarchy
4330 * but the memcg is pinned through dependency. The memcg should stop
4331 * applying policies and should revert to the vanilla state as it may be
4332 * made visible again.
4333 *
4334 * The current implementation only resets the essential configurations.
4335 * This needs to be expanded to cover all the visible parts.
4336 */
4337static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4338{
4339 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4340
Vladimir Davydovd334c9b2016-03-17 14:19:38 -07004341 page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4342 page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4343 page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4344 page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4345 page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
Johannes Weiner241994e2015-02-11 15:26:06 -08004346 memcg->low = 0;
4347 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004348 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004349 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004350}
4351
Daisuke Nishimura02491442010-03-10 15:22:17 -08004352#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004353/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004354static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004355{
Johannes Weiner05b84302014-08-06 16:05:59 -07004356 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004357
Mel Gormand0164ad2015-11-06 16:28:21 -08004358 /* Try a single bulk charge without reclaim first, kswapd may wake */
4359 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004360 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004361 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004362 return ret;
4363 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004364
4365 /* Try charges one by one with reclaim */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004366 while (count--) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004367 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004368 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004369 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004370 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004371 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004372 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004373 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004374}
4375
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004376union mc_target {
4377 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004378 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004379};
4380
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004381enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004382 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004383 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004384 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004385};
4386
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004387static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4388 unsigned long addr, pte_t ptent)
4389{
4390 struct page *page = vm_normal_page(vma, addr, ptent);
4391
4392 if (!page || !page_mapped(page))
4393 return NULL;
4394 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004395 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004396 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004397 } else {
4398 if (!(mc.flags & MOVE_FILE))
4399 return NULL;
4400 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004401 if (!get_page_unless_zero(page))
4402 return NULL;
4403
4404 return page;
4405}
4406
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004407#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004408static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004409 pte_t ptent, swp_entry_t *entry)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004410{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004411 struct page *page = NULL;
4412 swp_entry_t ent = pte_to_swp_entry(ptent);
4413
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004414 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004415 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004416 /*
4417 * Because lookup_swap_cache() updates some statistics counter,
4418 * we call find_get_page() with swapper_space directly.
4419 */
Huang Yingf6ab1f72016-10-07 17:00:21 -07004420 page = find_get_page(swap_address_space(ent), swp_offset(ent));
Johannes Weiner7941d212016-01-14 15:21:23 -08004421 if (do_memsw_account())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004422 entry->val = ent.val;
4423
4424 return page;
4425}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004426#else
4427static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004428 pte_t ptent, swp_entry_t *entry)
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004429{
4430 return NULL;
4431}
4432#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004433
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004434static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4435 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4436{
4437 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004438 struct address_space *mapping;
4439 pgoff_t pgoff;
4440
4441 if (!vma->vm_file) /* anonymous vma */
4442 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004443 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004444 return NULL;
4445
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004446 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004447 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004448
4449 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004450#ifdef CONFIG_SWAP
4451 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004452 if (shmem_mapping(mapping)) {
4453 page = find_get_entry(mapping, pgoff);
4454 if (radix_tree_exceptional_entry(page)) {
4455 swp_entry_t swp = radix_to_swp_entry(page);
Johannes Weiner7941d212016-01-14 15:21:23 -08004456 if (do_memsw_account())
Johannes Weiner139b6a62014-05-06 12:50:05 -07004457 *entry = swp;
Huang Yingf6ab1f72016-10-07 17:00:21 -07004458 page = find_get_page(swap_address_space(swp),
4459 swp_offset(swp));
Johannes Weiner139b6a62014-05-06 12:50:05 -07004460 }
4461 } else
4462 page = find_get_page(mapping, pgoff);
4463#else
4464 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004465#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004466 return page;
4467}
4468
Chen Gangb1b0dea2015-04-14 15:47:35 -07004469/**
4470 * mem_cgroup_move_account - move account of the page
4471 * @page: the page
Li RongQing25843c22016-07-26 15:26:56 -07004472 * @compound: charge the page as compound or small page
Chen Gangb1b0dea2015-04-14 15:47:35 -07004473 * @from: mem_cgroup which the page is moved from.
4474 * @to: mem_cgroup which the page is moved to. @from != @to.
4475 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08004476 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07004477 *
4478 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4479 * from old cgroup.
4480 */
4481static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004482 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07004483 struct mem_cgroup *from,
4484 struct mem_cgroup *to)
4485{
4486 unsigned long flags;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004487 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004488 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004489 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004490
4491 VM_BUG_ON(from == to);
4492 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004493 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07004494
4495 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07004496 * Prevent mem_cgroup_migrate() from looking at
Hugh Dickins45637ba2015-11-05 18:49:40 -08004497 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07004498 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004499 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004500 if (!trylock_page(page))
4501 goto out;
4502
4503 ret = -EINVAL;
4504 if (page->mem_cgroup != from)
4505 goto out_unlock;
4506
Greg Thelenc4843a72015-05-22 17:13:16 -04004507 anon = PageAnon(page);
4508
Chen Gangb1b0dea2015-04-14 15:47:35 -07004509 spin_lock_irqsave(&from->move_lock, flags);
4510
Greg Thelenc4843a72015-05-22 17:13:16 -04004511 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004512 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4513 nr_pages);
4514 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4515 nr_pages);
4516 }
4517
Greg Thelenc4843a72015-05-22 17:13:16 -04004518 /*
4519 * move_lock grabbed above and caller set from->moving_account, so
4520 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4521 * So mapping should be stable for dirty pages.
4522 */
4523 if (!anon && PageDirty(page)) {
4524 struct address_space *mapping = page_mapping(page);
4525
4526 if (mapping_cap_account_dirty(mapping)) {
4527 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4528 nr_pages);
4529 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4530 nr_pages);
4531 }
4532 }
4533
Chen Gangb1b0dea2015-04-14 15:47:35 -07004534 if (PageWriteback(page)) {
4535 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4536 nr_pages);
4537 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4538 nr_pages);
4539 }
4540
4541 /*
4542 * It is safe to change page->mem_cgroup here because the page
4543 * is referenced, charged, and isolated - we can't race with
4544 * uncharging, charging, migration, or LRU putback.
4545 */
4546
4547 /* caller should have done css_get */
4548 page->mem_cgroup = to;
4549 spin_unlock_irqrestore(&from->move_lock, flags);
4550
4551 ret = 0;
4552
4553 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004554 mem_cgroup_charge_statistics(to, page, compound, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004555 memcg_check_events(to, page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004556 mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004557 memcg_check_events(from, page);
4558 local_irq_enable();
4559out_unlock:
4560 unlock_page(page);
4561out:
4562 return ret;
4563}
4564
Li RongQing7cf78062016-05-27 14:27:46 -07004565/**
4566 * get_mctgt_type - get target type of moving charge
4567 * @vma: the vma the pte to be checked belongs
4568 * @addr: the address corresponding to the pte to be checked
4569 * @ptent: the pte to be checked
4570 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4571 *
4572 * Returns
4573 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4574 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4575 * move charge. if @target is not NULL, the page is stored in target->page
4576 * with extra refcnt got(Callers should handle it).
4577 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4578 * target for charge migration. if @target is not NULL, the entry is stored
4579 * in target->ent.
4580 *
4581 * Called with pte lock held.
4582 */
4583
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004584static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004585 unsigned long addr, pte_t ptent, union mc_target *target)
4586{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004587 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004588 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004589 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004590
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004591 if (pte_present(ptent))
4592 page = mc_handle_present_pte(vma, addr, ptent);
4593 else if (is_swap_pte(ptent))
Li RongQing48406ef2016-07-26 15:22:14 -07004594 page = mc_handle_swap_pte(vma, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004595 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004596 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004597
4598 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004599 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004600 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004601 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004602 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004603 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004604 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004605 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004606 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004607 ret = MC_TARGET_PAGE;
4608 if (target)
4609 target->page = page;
4610 }
4611 if (!ret || !target)
4612 put_page(page);
4613 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004614 /* There is a swap entry and a page doesn't exist or isn't charged */
4615 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004616 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004617 ret = MC_TARGET_SWAP;
4618 if (target)
4619 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004620 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004621 return ret;
4622}
4623
Naoya Horiguchi12724852012-03-21 16:34:28 -07004624#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4625/*
4626 * We don't consider swapping or file mapped pages because THP does not
4627 * support them for now.
4628 * Caller should make sure that pmd_trans_huge(pmd) is true.
4629 */
4630static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4631 unsigned long addr, pmd_t pmd, union mc_target *target)
4632{
4633 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004634 enum mc_target_type ret = MC_TARGET_NONE;
4635
4636 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004637 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004638 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004639 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004640 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004641 ret = MC_TARGET_PAGE;
4642 if (target) {
4643 get_page(page);
4644 target->page = page;
4645 }
4646 }
4647 return ret;
4648}
4649#else
4650static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4651 unsigned long addr, pmd_t pmd, union mc_target *target)
4652{
4653 return MC_TARGET_NONE;
4654}
4655#endif
4656
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004657static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4658 unsigned long addr, unsigned long end,
4659 struct mm_walk *walk)
4660{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004661 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004662 pte_t *pte;
4663 spinlock_t *ptl;
4664
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004665 ptl = pmd_trans_huge_lock(pmd, vma);
4666 if (ptl) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004667 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4668 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004669 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004670 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004671 }
Dave Hansen03319322011-03-22 16:32:56 -07004672
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004673 if (pmd_trans_unstable(pmd))
4674 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004675 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4676 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004677 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004678 mc.precharge++; /* increment precharge temporarily */
4679 pte_unmap_unlock(pte - 1, ptl);
4680 cond_resched();
4681
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004682 return 0;
4683}
4684
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004685static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4686{
4687 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004688
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004689 struct mm_walk mem_cgroup_count_precharge_walk = {
4690 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4691 .mm = mm,
4692 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004693 down_read(&mm->mmap_sem);
James Morse0247f3f2016-10-07 17:00:12 -07004694 walk_page_range(0, mm->highest_vm_end,
4695 &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004696 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004697
4698 precharge = mc.precharge;
4699 mc.precharge = 0;
4700
4701 return precharge;
4702}
4703
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004704static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4705{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004706 unsigned long precharge = mem_cgroup_count_precharge(mm);
4707
4708 VM_BUG_ON(mc.moving_task);
4709 mc.moving_task = current;
4710 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004711}
4712
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004713/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4714static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004715{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004716 struct mem_cgroup *from = mc.from;
4717 struct mem_cgroup *to = mc.to;
4718
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004719 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004720 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004721 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004722 mc.precharge = 0;
4723 }
4724 /*
4725 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4726 * we must uncharge here.
4727 */
4728 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004729 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004730 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004731 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004732 /* we must fixup refcnts and charges */
4733 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004734 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004735 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004736 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004737
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004738 mem_cgroup_id_put_many(mc.from, mc.moved_swap);
4739
Johannes Weiner05b84302014-08-06 16:05:59 -07004740 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004741 * we charged both to->memory and to->memsw, so we
4742 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004743 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004744 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004745 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004746
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004747 mem_cgroup_id_get_many(mc.to, mc.moved_swap);
4748 css_put_many(&mc.to->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004749
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004750 mc.moved_swap = 0;
4751 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004752 memcg_oom_recover(from);
4753 memcg_oom_recover(to);
4754 wake_up_all(&mc.waitq);
4755}
4756
4757static void mem_cgroup_clear_mc(void)
4758{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004759 struct mm_struct *mm = mc.mm;
4760
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004761 /*
4762 * we must clear moving_task before waking up waiters at the end of
4763 * task migration.
4764 */
4765 mc.moving_task = NULL;
4766 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004767 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004768 mc.from = NULL;
4769 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04004770 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004771 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004772
4773 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004774}
4775
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004776static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004777{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004778 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07004779 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07004780 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04004781 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004782 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004783 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004784 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004785
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004786 /* charge immigration isn't supported on the default hierarchy */
4787 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07004788 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004789
Tejun Heo4530edd2015-09-11 15:00:19 -04004790 /*
4791 * Multi-process migrations only happen on the default hierarchy
4792 * where charge immigration is not used. Perform charge
4793 * immigration if @tset contains a leader and whine if there are
4794 * multiple.
4795 */
4796 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004797 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04004798 WARN_ON_ONCE(p);
4799 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004800 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04004801 }
4802 if (!p)
4803 return 0;
4804
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004805 /*
4806 * We are now commited to this value whatever it is. Changes in this
4807 * tunable will only affect upcoming migrations, not the current one.
4808 * So we need to save it, and keep it going.
4809 */
4810 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4811 if (!move_flags)
4812 return 0;
4813
Tejun Heo9f2115f2015-09-08 15:01:10 -07004814 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004815
Tejun Heo9f2115f2015-09-08 15:01:10 -07004816 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004817
Tejun Heo9f2115f2015-09-08 15:01:10 -07004818 mm = get_task_mm(p);
4819 if (!mm)
4820 return 0;
4821 /* We move charges only when we move a owner of the mm */
4822 if (mm->owner == p) {
4823 VM_BUG_ON(mc.from);
4824 VM_BUG_ON(mc.to);
4825 VM_BUG_ON(mc.precharge);
4826 VM_BUG_ON(mc.moved_charge);
4827 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004828
Tejun Heo9f2115f2015-09-08 15:01:10 -07004829 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004830 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004831 mc.from = from;
4832 mc.to = memcg;
4833 mc.flags = move_flags;
4834 spin_unlock(&mc.lock);
4835 /* We set mc.moving_task later */
4836
4837 ret = mem_cgroup_precharge_mc(mm);
4838 if (ret)
4839 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004840 } else {
4841 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004842 }
4843 return ret;
4844}
4845
Amit Pundir3ba5a3b2015-12-21 16:01:10 +05304846static int mem_cgroup_allow_attach(struct cgroup_taskset *tset)
Rom Lemarchand5c85c742014-11-07 09:42:40 -08004847{
Amit Pundir3ba5a3b2015-12-21 16:01:10 +05304848 return subsys_cgroup_allow_attach(tset);
Rom Lemarchand5c85c742014-11-07 09:42:40 -08004849}
4850
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004851static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004852{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004853 if (mc.to)
4854 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004855}
4856
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004857static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4858 unsigned long addr, unsigned long end,
4859 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004860{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004861 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004862 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004863 pte_t *pte;
4864 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004865 enum mc_target_type target_type;
4866 union mc_target target;
4867 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004868
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004869 ptl = pmd_trans_huge_lock(pmd, vma);
4870 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004871 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004872 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004873 return 0;
4874 }
4875 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4876 if (target_type == MC_TARGET_PAGE) {
4877 page = target.page;
4878 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004879 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08004880 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004881 mc.precharge -= HPAGE_PMD_NR;
4882 mc.moved_charge += HPAGE_PMD_NR;
4883 }
4884 putback_lru_page(page);
4885 }
4886 put_page(page);
4887 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004888 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004889 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004890 }
4891
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004892 if (pmd_trans_unstable(pmd))
4893 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004894retry:
4895 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4896 for (; addr != end; addr += PAGE_SIZE) {
4897 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004898 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004899
4900 if (!mc.precharge)
4901 break;
4902
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004903 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004904 case MC_TARGET_PAGE:
4905 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08004906 /*
4907 * We can have a part of the split pmd here. Moving it
4908 * can be done but it would be too convoluted so simply
4909 * ignore such a partial THP and keep it in original
4910 * memcg. There should be somebody mapping the head.
4911 */
4912 if (PageTransCompound(page))
4913 goto put;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004914 if (isolate_lru_page(page))
4915 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004916 if (!mem_cgroup_move_account(page, false,
4917 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004918 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004919 /* we uncharge from mc.from later. */
4920 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004921 }
4922 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004923put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004924 put_page(page);
4925 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004926 case MC_TARGET_SWAP:
4927 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004928 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004929 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004930 /* we fixup refcnts and charges later. */
4931 mc.moved_swap++;
4932 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004933 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004934 default:
4935 break;
4936 }
4937 }
4938 pte_unmap_unlock(pte - 1, ptl);
4939 cond_resched();
4940
4941 if (addr != end) {
4942 /*
4943 * We have consumed all precharges we got in can_attach().
4944 * We try charge one by one, but don't do any additional
4945 * charges to mc.to if we have failed in charge once in attach()
4946 * phase.
4947 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004948 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004949 if (!ret)
4950 goto retry;
4951 }
4952
4953 return ret;
4954}
4955
Tejun Heo264a0ae2016-04-21 19:09:02 -04004956static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004957{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004958 struct mm_walk mem_cgroup_move_charge_walk = {
4959 .pmd_entry = mem_cgroup_move_charge_pte_range,
Tejun Heo264a0ae2016-04-21 19:09:02 -04004960 .mm = mc.mm,
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004961 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004962
4963 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004964 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07004965 * Signal lock_page_memcg() to take the memcg's move_lock
4966 * while we're moving its pages to another memcg. Then wait
4967 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08004968 */
4969 atomic_inc(&mc.from->moving_account);
4970 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004971retry:
Tejun Heo264a0ae2016-04-21 19:09:02 -04004972 if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004973 /*
4974 * Someone who are holding the mmap_sem might be waiting in
4975 * waitq. So we cancel all extra charges, wake up all waiters,
4976 * and retry. Because we cancel precharges, we might not be able
4977 * to move enough charges, but moving charge is a best-effort
4978 * feature anyway, so it wouldn't be a big problem.
4979 */
4980 __mem_cgroup_clear_mc();
4981 cond_resched();
4982 goto retry;
4983 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004984 /*
4985 * When we have consumed all precharges and failed in doing
4986 * additional charge, the page walk just aborts.
4987 */
James Morse0247f3f2016-10-07 17:00:12 -07004988 walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk);
4989
Tejun Heo264a0ae2016-04-21 19:09:02 -04004990 up_read(&mc.mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08004991 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004992}
4993
Tejun Heo264a0ae2016-04-21 19:09:02 -04004994static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08004995{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004996 if (mc.to) {
4997 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004998 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004999 }
Balbir Singh67e465a2008-02-07 00:13:54 -08005000}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005001#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005002static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005003{
5004 return 0;
5005}
Amit Pundir3ba5a3b2015-12-21 16:01:10 +05305006static int mem_cgroup_allow_attach(struct cgroup_taskset *tset)
Rom Lemarchand5c85c742014-11-07 09:42:40 -08005007{
5008 return 0;
5009}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005010static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005011{
5012}
Tejun Heo264a0ae2016-04-21 19:09:02 -04005013static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005014{
5015}
5016#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005017
Tejun Heof00baae2013-04-15 13:41:15 -07005018/*
5019 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005020 * to verify whether we're attached to the default hierarchy on each mount
5021 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005022 */
Tejun Heoeb954192013-08-08 20:11:23 -04005023static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005024{
5025 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005026 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005027 * guarantees that @root doesn't have any children, so turning it
5028 * on for the root memcg is enough.
5029 */
Tejun Heo9e10a132015-09-18 11:56:28 -04005030 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee592015-03-12 16:26:19 -07005031 root_mem_cgroup->use_hierarchy = true;
5032 else
5033 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005034}
5035
Johannes Weiner241994e2015-02-11 15:26:06 -08005036static u64 memory_current_read(struct cgroup_subsys_state *css,
5037 struct cftype *cft)
5038{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005039 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5040
5041 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994e2015-02-11 15:26:06 -08005042}
5043
5044static int memory_low_show(struct seq_file *m, void *v)
5045{
5046 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005047 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994e2015-02-11 15:26:06 -08005048
5049 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005050 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005051 else
5052 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5053
5054 return 0;
5055}
5056
5057static ssize_t memory_low_write(struct kernfs_open_file *of,
5058 char *buf, size_t nbytes, loff_t off)
5059{
5060 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5061 unsigned long low;
5062 int err;
5063
5064 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005065 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994e2015-02-11 15:26:06 -08005066 if (err)
5067 return err;
5068
5069 memcg->low = low;
5070
5071 return nbytes;
5072}
5073
5074static int memory_high_show(struct seq_file *m, void *v)
5075{
5076 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005077 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994e2015-02-11 15:26:06 -08005078
5079 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005080 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005081 else
5082 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5083
5084 return 0;
5085}
5086
5087static ssize_t memory_high_write(struct kernfs_open_file *of,
5088 char *buf, size_t nbytes, loff_t off)
5089{
5090 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner588083b2016-03-17 14:20:25 -07005091 unsigned long nr_pages;
Johannes Weiner241994e2015-02-11 15:26:06 -08005092 unsigned long high;
5093 int err;
5094
5095 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005096 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994e2015-02-11 15:26:06 -08005097 if (err)
5098 return err;
5099
5100 memcg->high = high;
5101
Johannes Weiner588083b2016-03-17 14:20:25 -07005102 nr_pages = page_counter_read(&memcg->memory);
5103 if (nr_pages > high)
5104 try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5105 GFP_KERNEL, true);
5106
Tejun Heo2529bb32015-05-22 18:23:34 -04005107 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005108 return nbytes;
5109}
5110
5111static int memory_max_show(struct seq_file *m, void *v)
5112{
5113 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005114 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994e2015-02-11 15:26:06 -08005115
5116 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005117 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005118 else
5119 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5120
5121 return 0;
5122}
5123
5124static ssize_t memory_max_write(struct kernfs_open_file *of,
5125 char *buf, size_t nbytes, loff_t off)
5126{
5127 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005128 unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5129 bool drained = false;
Johannes Weiner241994e2015-02-11 15:26:06 -08005130 unsigned long max;
5131 int err;
5132
5133 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005134 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994e2015-02-11 15:26:06 -08005135 if (err)
5136 return err;
5137
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005138 xchg(&memcg->memory.limit, max);
5139
5140 for (;;) {
5141 unsigned long nr_pages = page_counter_read(&memcg->memory);
5142
5143 if (nr_pages <= max)
5144 break;
5145
5146 if (signal_pending(current)) {
5147 err = -EINTR;
5148 break;
5149 }
5150
5151 if (!drained) {
5152 drain_all_stock(memcg);
5153 drained = true;
5154 continue;
5155 }
5156
5157 if (nr_reclaims) {
5158 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5159 GFP_KERNEL, true))
5160 nr_reclaims--;
5161 continue;
5162 }
5163
5164 mem_cgroup_events(memcg, MEMCG_OOM, 1);
5165 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5166 break;
5167 }
Johannes Weiner241994e2015-02-11 15:26:06 -08005168
Tejun Heo2529bb32015-05-22 18:23:34 -04005169 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005170 return nbytes;
5171}
5172
5173static int memory_events_show(struct seq_file *m, void *v)
5174{
5175 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5176
5177 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5178 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5179 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5180 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5181
5182 return 0;
5183}
5184
Johannes Weiner587d9f72016-01-20 15:03:19 -08005185static int memory_stat_show(struct seq_file *m, void *v)
5186{
5187 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005188 unsigned long stat[MEMCG_NR_STAT];
5189 unsigned long events[MEMCG_NR_EVENTS];
Johannes Weiner587d9f72016-01-20 15:03:19 -08005190 int i;
5191
5192 /*
5193 * Provide statistics on the state of the memory subsystem as
5194 * well as cumulative event counters that show past behavior.
5195 *
5196 * This list is ordered following a combination of these gradients:
5197 * 1) generic big picture -> specifics and details
5198 * 2) reflecting userspace activity -> reflecting kernel heuristics
5199 *
5200 * Current memory state:
5201 */
5202
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005203 tree_stat(memcg, stat);
5204 tree_events(memcg, events);
5205
Johannes Weiner587d9f72016-01-20 15:03:19 -08005206 seq_printf(m, "anon %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005207 (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005208 seq_printf(m, "file %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005209 (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE);
Vladimir Davydov12580e42016-03-17 14:17:38 -07005210 seq_printf(m, "kernel_stack %llu\n",
Andy Lutomirskiefdc9492016-07-28 15:48:17 -07005211 (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005212 seq_printf(m, "slab %llu\n",
5213 (u64)(stat[MEMCG_SLAB_RECLAIMABLE] +
5214 stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
Johannes Weinerb2807f02016-01-20 15:03:22 -08005215 seq_printf(m, "sock %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005216 (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005217
5218 seq_printf(m, "file_mapped %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005219 (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005220 seq_printf(m, "file_dirty %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005221 (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005222 seq_printf(m, "file_writeback %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005223 (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005224
5225 for (i = 0; i < NR_LRU_LISTS; i++) {
5226 struct mem_cgroup *mi;
5227 unsigned long val = 0;
5228
5229 for_each_mem_cgroup_tree(mi, memcg)
5230 val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5231 seq_printf(m, "%s %llu\n",
5232 mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5233 }
5234
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005235 seq_printf(m, "slab_reclaimable %llu\n",
5236 (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE);
5237 seq_printf(m, "slab_unreclaimable %llu\n",
5238 (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5239
Johannes Weiner587d9f72016-01-20 15:03:19 -08005240 /* Accumulated memory events */
5241
5242 seq_printf(m, "pgfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005243 events[MEM_CGROUP_EVENTS_PGFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005244 seq_printf(m, "pgmajfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005245 events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005246
5247 return 0;
5248}
5249
Johannes Weiner241994e2015-02-11 15:26:06 -08005250static struct cftype memory_files[] = {
5251 {
5252 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005253 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994e2015-02-11 15:26:06 -08005254 .read_u64 = memory_current_read,
5255 },
5256 {
5257 .name = "low",
5258 .flags = CFTYPE_NOT_ON_ROOT,
5259 .seq_show = memory_low_show,
5260 .write = memory_low_write,
5261 },
5262 {
5263 .name = "high",
5264 .flags = CFTYPE_NOT_ON_ROOT,
5265 .seq_show = memory_high_show,
5266 .write = memory_high_write,
5267 },
5268 {
5269 .name = "max",
5270 .flags = CFTYPE_NOT_ON_ROOT,
5271 .seq_show = memory_max_show,
5272 .write = memory_max_write,
5273 },
5274 {
5275 .name = "events",
5276 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04005277 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994e2015-02-11 15:26:06 -08005278 .seq_show = memory_events_show,
5279 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08005280 {
5281 .name = "stat",
5282 .flags = CFTYPE_NOT_ON_ROOT,
5283 .seq_show = memory_stat_show,
5284 },
Johannes Weiner241994e2015-02-11 15:26:06 -08005285 { } /* terminate */
5286};
5287
Tejun Heo073219e2014-02-08 10:36:58 -05005288struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005289 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005290 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005291 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08005292 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08005293 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005294 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005295 .can_attach = mem_cgroup_can_attach,
5296 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04005297 .post_attach = mem_cgroup_move_task,
Rom Lemarchand5c85c742014-11-07 09:42:40 -08005298 .allow_attach = mem_cgroup_allow_attach,
Tejun Heof00baae2013-04-15 13:41:15 -07005299 .bind = mem_cgroup_bind,
Johannes Weiner241994e2015-02-11 15:26:06 -08005300 .dfl_cftypes = memory_files,
5301 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005302 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005303};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005304
Johannes Weiner241994e2015-02-11 15:26:06 -08005305/**
Johannes Weiner241994e2015-02-11 15:26:06 -08005306 * mem_cgroup_low - check if memory consumption is below the normal range
5307 * @root: the highest ancestor to consider
5308 * @memcg: the memory cgroup to check
5309 *
5310 * Returns %true if memory consumption of @memcg, and that of all
5311 * configurable ancestors up to @root, is below the normal range.
5312 */
5313bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5314{
5315 if (mem_cgroup_disabled())
5316 return false;
5317
5318 /*
5319 * The toplevel group doesn't have a configurable range, so
5320 * it's never low when looked at directly, and it is not
5321 * considered an ancestor when assessing the hierarchy.
5322 */
5323
5324 if (memcg == root_mem_cgroup)
5325 return false;
5326
Michal Hocko4e54ded2015-02-27 15:51:46 -08005327 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005328 return false;
5329
5330 while (memcg != root) {
5331 memcg = parent_mem_cgroup(memcg);
5332
5333 if (memcg == root_mem_cgroup)
5334 break;
5335
Michal Hocko4e54ded2015-02-27 15:51:46 -08005336 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005337 return false;
5338 }
5339 return true;
5340}
5341
Johannes Weiner00501b52014-08-08 14:19:20 -07005342/**
5343 * mem_cgroup_try_charge - try charging a page
5344 * @page: page to charge
5345 * @mm: mm context of the victim
5346 * @gfp_mask: reclaim mode
5347 * @memcgp: charged memcg return
Li RongQing25843c22016-07-26 15:26:56 -07005348 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005349 *
5350 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5351 * pages according to @gfp_mask if necessary.
5352 *
5353 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5354 * Otherwise, an error code is returned.
5355 *
5356 * After page->mapping has been set up, the caller must finalize the
5357 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5358 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5359 */
5360int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005361 gfp_t gfp_mask, struct mem_cgroup **memcgp,
5362 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005363{
5364 struct mem_cgroup *memcg = NULL;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005365 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005366 int ret = 0;
5367
5368 if (mem_cgroup_disabled())
5369 goto out;
5370
5371 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005372 /*
5373 * Every swap fault against a single page tries to charge the
5374 * page, bail as early as possible. shmem_unuse() encounters
5375 * already charged pages, too. The USED bit is protected by
5376 * the page lock, which serializes swap cache removal, which
5377 * in turn serializes uncharging.
5378 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005379 VM_BUG_ON_PAGE(!PageLocked(page), page);
Johannes Weiner1306a852014-12-10 15:44:52 -08005380 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005381 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005382
Vladimir Davydov37e84352016-01-20 15:02:56 -08005383 if (do_swap_account) {
Vladimir Davydove993d902015-09-09 15:35:35 -07005384 swp_entry_t ent = { .val = page_private(page), };
5385 unsigned short id = lookup_swap_cgroup_id(ent);
5386
5387 rcu_read_lock();
5388 memcg = mem_cgroup_from_id(id);
5389 if (memcg && !css_tryget_online(&memcg->css))
5390 memcg = NULL;
5391 rcu_read_unlock();
5392 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005393 }
5394
Johannes Weiner00501b52014-08-08 14:19:20 -07005395 if (!memcg)
5396 memcg = get_mem_cgroup_from_mm(mm);
5397
5398 ret = try_charge(memcg, gfp_mask, nr_pages);
5399
5400 css_put(&memcg->css);
Johannes Weiner00501b52014-08-08 14:19:20 -07005401out:
5402 *memcgp = memcg;
5403 return ret;
5404}
5405
5406/**
5407 * mem_cgroup_commit_charge - commit a page charge
5408 * @page: page to charge
5409 * @memcg: memcg to charge the page to
5410 * @lrucare: page might be on LRU already
Li RongQing25843c22016-07-26 15:26:56 -07005411 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005412 *
5413 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5414 * after page->mapping has been set up. This must happen atomically
5415 * as part of the page instantiation, i.e. under the page table lock
5416 * for anonymous pages, under the page lock for page and swap cache.
5417 *
5418 * In addition, the page must not be on the LRU during the commit, to
5419 * prevent racing with task migration. If it might be, use @lrucare.
5420 *
5421 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5422 */
5423void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005424 bool lrucare, bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005425{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005426 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005427
5428 VM_BUG_ON_PAGE(!page->mapping, page);
5429 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5430
5431 if (mem_cgroup_disabled())
5432 return;
5433 /*
5434 * Swap faults will attempt to charge the same page multiple
5435 * times. But reuse_swap_page() might have removed the page
5436 * from swapcache already, so we can't check PageSwapCache().
5437 */
5438 if (!memcg)
5439 return;
5440
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005441 commit_charge(page, memcg, lrucare);
5442
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005443 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005444 mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005445 memcg_check_events(memcg, page);
5446 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005447
Johannes Weiner7941d212016-01-14 15:21:23 -08005448 if (do_memsw_account() && PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005449 swp_entry_t entry = { .val = page_private(page) };
5450 /*
5451 * The swap entry might not get freed for a long time,
5452 * let's not wait for it. The page already received a
5453 * memory+swap charge, drop the swap entry duplicate.
5454 */
5455 mem_cgroup_uncharge_swap(entry);
5456 }
5457}
5458
5459/**
5460 * mem_cgroup_cancel_charge - cancel a page charge
5461 * @page: page to charge
5462 * @memcg: memcg to charge the page to
Li RongQing25843c22016-07-26 15:26:56 -07005463 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005464 *
5465 * Cancel a charge transaction started by mem_cgroup_try_charge().
5466 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005467void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5468 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005469{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005470 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005471
5472 if (mem_cgroup_disabled())
5473 return;
5474 /*
5475 * Swap faults will attempt to charge the same page multiple
5476 * times. But reuse_swap_page() might have removed the page
5477 * from swapcache already, so we can't check PageSwapCache().
5478 */
5479 if (!memcg)
5480 return;
5481
Johannes Weiner00501b52014-08-08 14:19:20 -07005482 cancel_charge(memcg, nr_pages);
5483}
5484
Johannes Weiner747db952014-08-08 14:19:24 -07005485static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005486 unsigned long nr_anon, unsigned long nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005487 unsigned long nr_huge, unsigned long nr_kmem,
5488 struct page *dummy_page)
Johannes Weiner747db952014-08-08 14:19:24 -07005489{
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005490 unsigned long nr_pages = nr_anon + nr_file + nr_kmem;
Johannes Weiner747db952014-08-08 14:19:24 -07005491 unsigned long flags;
5492
Johannes Weinerce00a962014-09-05 08:43:57 -04005493 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005494 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08005495 if (do_memsw_account())
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005496 page_counter_uncharge(&memcg->memsw, nr_pages);
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005497 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem)
5498 page_counter_uncharge(&memcg->kmem, nr_kmem);
Johannes Weinerce00a962014-09-05 08:43:57 -04005499 memcg_oom_recover(memcg);
5500 }
Johannes Weiner747db952014-08-08 14:19:24 -07005501
5502 local_irq_save(flags);
5503 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5504 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5505 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5506 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005507 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005508 memcg_check_events(memcg, dummy_page);
5509 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005510
5511 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005512 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005513}
5514
5515static void uncharge_list(struct list_head *page_list)
5516{
5517 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005518 unsigned long nr_anon = 0;
5519 unsigned long nr_file = 0;
5520 unsigned long nr_huge = 0;
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005521 unsigned long nr_kmem = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005522 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005523 struct list_head *next;
5524 struct page *page;
5525
Johannes Weiner8b592652016-03-17 14:20:31 -07005526 /*
5527 * Note that the list can be a single page->lru; hence the
5528 * do-while loop instead of a simple list_for_each_entry().
5529 */
Johannes Weiner747db952014-08-08 14:19:24 -07005530 next = page_list->next;
5531 do {
Johannes Weiner747db952014-08-08 14:19:24 -07005532 page = list_entry(next, struct page, lru);
5533 next = page->lru.next;
5534
5535 VM_BUG_ON_PAGE(PageLRU(page), page);
5536 VM_BUG_ON_PAGE(page_count(page), page);
5537
Johannes Weiner1306a852014-12-10 15:44:52 -08005538 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005539 continue;
5540
5541 /*
5542 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005543 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005544 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005545 */
5546
Johannes Weiner1306a852014-12-10 15:44:52 -08005547 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005548 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005549 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005550 nr_huge, nr_kmem, page);
5551 pgpgout = nr_anon = nr_file =
5552 nr_huge = nr_kmem = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005553 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005554 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005555 }
5556
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005557 if (!PageKmemcg(page)) {
5558 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005559
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005560 if (PageTransHuge(page)) {
5561 nr_pages <<= compound_order(page);
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005562 nr_huge += nr_pages;
5563 }
5564 if (PageAnon(page))
5565 nr_anon += nr_pages;
5566 else
5567 nr_file += nr_pages;
5568 pgpgout++;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03005569 } else {
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005570 nr_kmem += 1 << compound_order(page);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03005571 __ClearPageKmemcg(page);
5572 }
Johannes Weiner747db952014-08-08 14:19:24 -07005573
Johannes Weiner1306a852014-12-10 15:44:52 -08005574 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005575 } while (next != page_list);
5576
5577 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005578 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005579 nr_huge, nr_kmem, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005580}
5581
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005582/**
5583 * mem_cgroup_uncharge - uncharge a page
5584 * @page: page to uncharge
5585 *
5586 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5587 * mem_cgroup_commit_charge().
5588 */
5589void mem_cgroup_uncharge(struct page *page)
5590{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005591 if (mem_cgroup_disabled())
5592 return;
5593
Johannes Weiner747db952014-08-08 14:19:24 -07005594 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005595 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005596 return;
5597
Johannes Weiner747db952014-08-08 14:19:24 -07005598 INIT_LIST_HEAD(&page->lru);
5599 uncharge_list(&page->lru);
5600}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005601
Johannes Weiner747db952014-08-08 14:19:24 -07005602/**
5603 * mem_cgroup_uncharge_list - uncharge a list of page
5604 * @page_list: list of pages to uncharge
5605 *
5606 * Uncharge a list of pages previously charged with
5607 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5608 */
5609void mem_cgroup_uncharge_list(struct list_head *page_list)
5610{
5611 if (mem_cgroup_disabled())
5612 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005613
Johannes Weiner747db952014-08-08 14:19:24 -07005614 if (!list_empty(page_list))
5615 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005616}
5617
5618/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005619 * mem_cgroup_migrate - charge a page's replacement
5620 * @oldpage: currently circulating page
5621 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005622 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005623 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5624 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005625 *
5626 * Both pages must be locked, @newpage->mapping must be set up.
5627 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005628void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005629{
Johannes Weiner29833312014-12-10 15:44:02 -08005630 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005631 unsigned int nr_pages;
5632 bool compound;
Tejun Heod93c4132016-06-24 14:49:54 -07005633 unsigned long flags;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005634
5635 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5636 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005637 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005638 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5639 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005640
5641 if (mem_cgroup_disabled())
5642 return;
5643
5644 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005645 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005646 return;
5647
Hugh Dickins45637ba2015-11-05 18:49:40 -08005648 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08005649 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005650 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005651 return;
5652
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005653 /* Force-charge the new page. The old one will be freed soon */
5654 compound = PageTransHuge(newpage);
5655 nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5656
5657 page_counter_charge(&memcg->memory, nr_pages);
5658 if (do_memsw_account())
5659 page_counter_charge(&memcg->memsw, nr_pages);
5660 css_get_many(&memcg->css, nr_pages);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005661
Johannes Weiner9cf76662016-03-15 14:57:58 -07005662 commit_charge(newpage, memcg, false);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005663
Tejun Heod93c4132016-06-24 14:49:54 -07005664 local_irq_save(flags);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005665 mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5666 memcg_check_events(memcg, newpage);
Tejun Heod93c4132016-06-24 14:49:54 -07005667 local_irq_restore(flags);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005668}
5669
Johannes Weineref129472016-01-14 15:21:34 -08005670DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08005671EXPORT_SYMBOL(memcg_sockets_enabled_key);
5672
Johannes Weiner2d758072016-10-07 17:00:58 -07005673void mem_cgroup_sk_alloc(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08005674{
5675 struct mem_cgroup *memcg;
5676
Johannes Weiner2d758072016-10-07 17:00:58 -07005677 if (!mem_cgroup_sockets_enabled)
5678 return;
5679
5680 /*
5681 * Socket cloning can throw us here with sk_memcg already
Johannes Weiner11092082016-01-14 15:21:26 -08005682 * filled. It won't however, necessarily happen from
5683 * process context. So the test for root memcg given
5684 * the current task's memcg won't help us in this case.
5685 *
5686 * Respecting the original socket's memcg is a better
5687 * decision in this case.
5688 */
5689 if (sk->sk_memcg) {
5690 BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5691 css_get(&sk->sk_memcg->css);
5692 return;
5693 }
5694
5695 rcu_read_lock();
5696 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005697 if (memcg == root_mem_cgroup)
5698 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08005699 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005700 goto out;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005701 if (css_tryget_online(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08005702 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005703out:
Johannes Weiner11092082016-01-14 15:21:26 -08005704 rcu_read_unlock();
5705}
Johannes Weiner11092082016-01-14 15:21:26 -08005706
Johannes Weiner2d758072016-10-07 17:00:58 -07005707void mem_cgroup_sk_free(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08005708{
Johannes Weiner2d758072016-10-07 17:00:58 -07005709 if (sk->sk_memcg)
5710 css_put(&sk->sk_memcg->css);
Johannes Weiner11092082016-01-14 15:21:26 -08005711}
5712
5713/**
5714 * mem_cgroup_charge_skmem - charge socket memory
5715 * @memcg: memcg to charge
5716 * @nr_pages: number of pages to charge
5717 *
5718 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5719 * @memcg's configured limit, %false if the charge had to be forced.
5720 */
5721bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5722{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005723 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08005724
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005725 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005726 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005727
Johannes Weiner0db15292016-01-20 15:02:50 -08005728 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5729 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005730 return true;
5731 }
Johannes Weiner0db15292016-01-20 15:02:50 -08005732 page_counter_charge(&memcg->tcpmem, nr_pages);
5733 memcg->tcpmem_pressure = 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005734 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08005735 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005736
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005737 /* Don't block in the packet receive path */
5738 if (in_softirq())
5739 gfp_mask = GFP_NOWAIT;
5740
Johannes Weinerb2807f02016-01-20 15:03:22 -08005741 this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5742
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005743 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5744 return true;
5745
5746 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005747 return false;
5748}
5749
5750/**
5751 * mem_cgroup_uncharge_skmem - uncharge socket memory
5752 * @memcg - memcg to uncharge
5753 * @nr_pages - number of pages to uncharge
5754 */
5755void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5756{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005757 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005758 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005759 return;
5760 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005761
Johannes Weinerb2807f02016-01-20 15:03:22 -08005762 this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5763
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005764 page_counter_uncharge(&memcg->memory, nr_pages);
5765 css_put_many(&memcg->css, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005766}
5767
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005768static int __init cgroup_memory(char *s)
5769{
5770 char *token;
5771
5772 while ((token = strsep(&s, ",")) != NULL) {
5773 if (!*token)
5774 continue;
5775 if (!strcmp(token, "nosocket"))
5776 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08005777 if (!strcmp(token, "nokmem"))
5778 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005779 }
5780 return 0;
5781}
5782__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08005783
Michal Hocko2d110852013-02-22 16:34:43 -08005784/*
Michal Hocko10813122013-02-22 16:35:41 -08005785 * subsys_initcall() for memory controller.
5786 *
5787 * Some parts like hotcpu_notifier() have to be initialized from this context
5788 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5789 * everything that doesn't depend on a specific mem_cgroup structure should
5790 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005791 */
5792static int __init mem_cgroup_init(void)
5793{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005794 int cpu, node;
5795
Michal Hocko2d110852013-02-22 16:34:43 -08005796 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005797
5798 for_each_possible_cpu(cpu)
5799 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5800 drain_local_stock);
5801
5802 for_each_node(node) {
5803 struct mem_cgroup_tree_per_node *rtpn;
Johannes Weiner95a045f2015-02-11 15:26:33 -08005804
5805 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5806 node_online(node) ? node : NUMA_NO_NODE);
5807
Mel Gormanef8f2322016-07-28 15:46:05 -07005808 rtpn->rb_root = RB_ROOT;
5809 spin_lock_init(&rtpn->lock);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005810 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5811 }
5812
Michal Hocko2d110852013-02-22 16:34:43 -08005813 return 0;
5814}
5815subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005816
5817#ifdef CONFIG_MEMCG_SWAP
Arnd Bergmann358c07f2016-08-25 15:17:08 -07005818static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
5819{
5820 while (!atomic_inc_not_zero(&memcg->id.ref)) {
5821 /*
5822 * The root cgroup cannot be destroyed, so it's refcount must
5823 * always be >= 1.
5824 */
5825 if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
5826 VM_BUG_ON(1);
5827 break;
5828 }
5829 memcg = parent_mem_cgroup(memcg);
5830 if (!memcg)
5831 memcg = root_mem_cgroup;
5832 }
5833 return memcg;
5834}
5835
Johannes Weiner21afa382015-02-11 15:26:36 -08005836/**
5837 * mem_cgroup_swapout - transfer a memsw charge to swap
5838 * @page: page whose memsw charge to transfer
5839 * @entry: swap entry to move the charge to
5840 *
5841 * Transfer the memsw charge of @page to @entry.
5842 */
5843void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5844{
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005845 struct mem_cgroup *memcg, *swap_memcg;
Johannes Weiner21afa382015-02-11 15:26:36 -08005846 unsigned short oldid;
5847
5848 VM_BUG_ON_PAGE(PageLRU(page), page);
5849 VM_BUG_ON_PAGE(page_count(page), page);
5850
Johannes Weiner7941d212016-01-14 15:21:23 -08005851 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08005852 return;
5853
5854 memcg = page->mem_cgroup;
5855
5856 /* Readahead page, never charged */
5857 if (!memcg)
5858 return;
5859
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005860 /*
5861 * In case the memcg owning these pages has been offlined and doesn't
5862 * have an ID allocated to it anymore, charge the closest online
5863 * ancestor for the swap instead and transfer the memory+swap charge.
5864 */
5865 swap_memcg = mem_cgroup_id_get_online(memcg);
5866 oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg));
Johannes Weiner21afa382015-02-11 15:26:36 -08005867 VM_BUG_ON_PAGE(oldid, page);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005868 mem_cgroup_swap_statistics(swap_memcg, true);
Johannes Weiner21afa382015-02-11 15:26:36 -08005869
5870 page->mem_cgroup = NULL;
5871
5872 if (!mem_cgroup_is_root(memcg))
5873 page_counter_uncharge(&memcg->memory, 1);
5874
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005875 if (memcg != swap_memcg) {
5876 if (!mem_cgroup_is_root(swap_memcg))
5877 page_counter_charge(&swap_memcg->memsw, 1);
5878 page_counter_uncharge(&memcg->memsw, 1);
5879 }
5880
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07005881 /*
5882 * Interrupts should be disabled here because the caller holds the
5883 * mapping->tree_lock lock which is taken with interrupts-off. It is
5884 * important here to have the interrupts disabled because it is the
5885 * only synchronisation we have for udpating the per-CPU variables.
5886 */
5887 VM_BUG_ON(!irqs_disabled());
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005888 mem_cgroup_charge_statistics(memcg, page, false, -1);
Johannes Weiner21afa382015-02-11 15:26:36 -08005889 memcg_check_events(memcg, page);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005890
5891 if (!mem_cgroup_is_root(memcg))
5892 css_put(&memcg->css);
Johannes Weiner21afa382015-02-11 15:26:36 -08005893}
5894
Vladimir Davydov37e84352016-01-20 15:02:56 -08005895/*
5896 * mem_cgroup_try_charge_swap - try charging a swap entry
5897 * @page: page being added to swap
5898 * @entry: swap entry to charge
5899 *
5900 * Try to charge @entry to the memcg that @page belongs to.
5901 *
5902 * Returns 0 on success, -ENOMEM on failure.
5903 */
5904int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5905{
5906 struct mem_cgroup *memcg;
5907 struct page_counter *counter;
5908 unsigned short oldid;
5909
5910 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5911 return 0;
5912
5913 memcg = page->mem_cgroup;
5914
5915 /* Readahead page, never charged */
5916 if (!memcg)
5917 return 0;
5918
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005919 memcg = mem_cgroup_id_get_online(memcg);
Vladimir Davydov37e84352016-01-20 15:02:56 -08005920
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005921 if (!mem_cgroup_is_root(memcg) &&
5922 !page_counter_try_charge(&memcg->swap, 1, &counter)) {
5923 mem_cgroup_id_put(memcg);
5924 return -ENOMEM;
5925 }
5926
Vladimir Davydov37e84352016-01-20 15:02:56 -08005927 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5928 VM_BUG_ON_PAGE(oldid, page);
5929 mem_cgroup_swap_statistics(memcg, true);
5930
Vladimir Davydov37e84352016-01-20 15:02:56 -08005931 return 0;
5932}
5933
Johannes Weiner21afa382015-02-11 15:26:36 -08005934/**
5935 * mem_cgroup_uncharge_swap - uncharge a swap entry
5936 * @entry: swap entry to uncharge
5937 *
Vladimir Davydov37e84352016-01-20 15:02:56 -08005938 * Drop the swap charge associated with @entry.
Johannes Weiner21afa382015-02-11 15:26:36 -08005939 */
5940void mem_cgroup_uncharge_swap(swp_entry_t entry)
5941{
5942 struct mem_cgroup *memcg;
5943 unsigned short id;
5944
Vladimir Davydov37e84352016-01-20 15:02:56 -08005945 if (!do_swap_account)
Johannes Weiner21afa382015-02-11 15:26:36 -08005946 return;
5947
5948 id = swap_cgroup_record(entry, 0);
5949 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005950 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005951 if (memcg) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08005952 if (!mem_cgroup_is_root(memcg)) {
5953 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5954 page_counter_uncharge(&memcg->swap, 1);
5955 else
5956 page_counter_uncharge(&memcg->memsw, 1);
5957 }
Johannes Weiner21afa382015-02-11 15:26:36 -08005958 mem_cgroup_swap_statistics(memcg, false);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005959 mem_cgroup_id_put(memcg);
Johannes Weiner21afa382015-02-11 15:26:36 -08005960 }
5961 rcu_read_unlock();
5962}
5963
Vladimir Davydovd8b38432016-01-20 15:03:07 -08005964long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
5965{
5966 long nr_swap_pages = get_nr_swap_pages();
5967
5968 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5969 return nr_swap_pages;
5970 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5971 nr_swap_pages = min_t(long, nr_swap_pages,
5972 READ_ONCE(memcg->swap.limit) -
5973 page_counter_read(&memcg->swap));
5974 return nr_swap_pages;
5975}
5976
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08005977bool mem_cgroup_swap_full(struct page *page)
5978{
5979 struct mem_cgroup *memcg;
5980
5981 VM_BUG_ON_PAGE(!PageLocked(page), page);
5982
5983 if (vm_swap_full())
5984 return true;
5985 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5986 return false;
5987
5988 memcg = page->mem_cgroup;
5989 if (!memcg)
5990 return false;
5991
5992 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5993 if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
5994 return true;
5995
5996 return false;
5997}
5998
Johannes Weiner21afa382015-02-11 15:26:36 -08005999/* for remember boot option*/
6000#ifdef CONFIG_MEMCG_SWAP_ENABLED
6001static int really_do_swap_account __initdata = 1;
6002#else
6003static int really_do_swap_account __initdata;
6004#endif
6005
6006static int __init enable_swap_account(char *s)
6007{
6008 if (!strcmp(s, "1"))
6009 really_do_swap_account = 1;
6010 else if (!strcmp(s, "0"))
6011 really_do_swap_account = 0;
6012 return 1;
6013}
6014__setup("swapaccount=", enable_swap_account);
6015
Vladimir Davydov37e84352016-01-20 15:02:56 -08006016static u64 swap_current_read(struct cgroup_subsys_state *css,
6017 struct cftype *cft)
6018{
6019 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6020
6021 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
6022}
6023
6024static int swap_max_show(struct seq_file *m, void *v)
6025{
6026 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
6027 unsigned long max = READ_ONCE(memcg->swap.limit);
6028
6029 if (max == PAGE_COUNTER_MAX)
6030 seq_puts(m, "max\n");
6031 else
6032 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
6033
6034 return 0;
6035}
6036
6037static ssize_t swap_max_write(struct kernfs_open_file *of,
6038 char *buf, size_t nbytes, loff_t off)
6039{
6040 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6041 unsigned long max;
6042 int err;
6043
6044 buf = strstrip(buf);
6045 err = page_counter_memparse(buf, "max", &max);
6046 if (err)
6047 return err;
6048
6049 mutex_lock(&memcg_limit_mutex);
6050 err = page_counter_limit(&memcg->swap, max);
6051 mutex_unlock(&memcg_limit_mutex);
6052 if (err)
6053 return err;
6054
6055 return nbytes;
6056}
6057
6058static struct cftype swap_files[] = {
6059 {
6060 .name = "swap.current",
6061 .flags = CFTYPE_NOT_ON_ROOT,
6062 .read_u64 = swap_current_read,
6063 },
6064 {
6065 .name = "swap.max",
6066 .flags = CFTYPE_NOT_ON_ROOT,
6067 .seq_show = swap_max_show,
6068 .write = swap_max_write,
6069 },
6070 { } /* terminate */
6071};
6072
Johannes Weiner21afa382015-02-11 15:26:36 -08006073static struct cftype memsw_cgroup_files[] = {
6074 {
6075 .name = "memsw.usage_in_bytes",
6076 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6077 .read_u64 = mem_cgroup_read_u64,
6078 },
6079 {
6080 .name = "memsw.max_usage_in_bytes",
6081 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6082 .write = mem_cgroup_reset,
6083 .read_u64 = mem_cgroup_read_u64,
6084 },
6085 {
6086 .name = "memsw.limit_in_bytes",
6087 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6088 .write = mem_cgroup_write,
6089 .read_u64 = mem_cgroup_read_u64,
6090 },
6091 {
6092 .name = "memsw.failcnt",
6093 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6094 .write = mem_cgroup_reset,
6095 .read_u64 = mem_cgroup_read_u64,
6096 },
6097 { }, /* terminate */
6098};
6099
6100static int __init mem_cgroup_swap_init(void)
6101{
6102 if (!mem_cgroup_disabled() && really_do_swap_account) {
6103 do_swap_account = 1;
Vladimir Davydov37e84352016-01-20 15:02:56 -08006104 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
6105 swap_files));
Johannes Weiner21afa382015-02-11 15:26:36 -08006106 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
6107 memsw_cgroup_files));
6108 }
6109 return 0;
6110}
6111subsys_initcall(mem_cgroup_swap_init);
6112
6113#endif /* CONFIG_MEMCG_SWAP */