blob: 2bd7541d7c11231431c060ca6cfe84a89f096fe3 [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>
Ingo Molnar6e84f312017-02-08 18:51:29 +010038#include <linux/sched/mm.h>
Hugh Dickins3a4f8a02017-02-24 14:59:36 -080039#include <linux/shmem_fs.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080040#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080041#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080042#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080043#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080044#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080045#include <linux/bit_spinlock.h>
46#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070047#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040048#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080049#include <linux/mutex.h>
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -070050#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070051#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080052#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080053#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080054#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080055#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050056#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080057#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080058#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080059#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070060#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070061#include <linux/mm_inline.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080062#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080063#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070064#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070065#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050066#include <linux/file.h>
Tejun Heob23afb92015-11-05 18:46:11 -080067#include <linux/tracehook.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080068#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000069#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070070#include <net/ip.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080071#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080072
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080073#include <linux/uaccess.h>
Balbir Singh8697d332008-02-07 00:13:59 -080074
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070075#include <trace/events/vmscan.h>
76
Tejun Heo073219e2014-02-08 10:36:58 -050077struct cgroup_subsys memory_cgrp_subsys __read_mostly;
78EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080079
Johannes Weiner7d828602016-01-14 15:20:56 -080080struct mem_cgroup *root_mem_cgroup __read_mostly;
81
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070082#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh8cdea7c2008-02-07 00:13:50 -080083
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080084/* Socket memory accounting disabled? */
85static bool cgroup_memory_nosocket;
86
Vladimir Davydov04823c82016-01-20 15:02:38 -080087/* Kernel memory accounting disabled? */
88static bool cgroup_memory_nokmem;
89
Johannes Weiner21afa382015-02-11 15:26:36 -080090/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070091#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080092int do_swap_account __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080093#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070094#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080095#endif
96
Johannes Weiner7941d212016-01-14 15:21:23 -080097/* Whether legacy memory+swap accounting is active */
98static bool do_memsw_account(void)
99{
100 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account;
101}
102
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700103static const char * const mem_cgroup_stat_names[] = {
104 "cache",
105 "rss",
David Rientjesb070e652013-05-07 16:18:09 -0700106 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700107 "mapped_file",
Greg Thelenc4843a72015-05-22 17:13:16 -0400108 "dirty",
Sha Zhengju3ea67d02013-09-12 15:13:53 -0700109 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700110 "swap",
111};
112
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700113static const char * const mem_cgroup_events_names[] = {
114 "pgpgin",
115 "pgpgout",
116 "pgfault",
117 "pgmajfault",
118};
119
Sha Zhengju58cf1882013-02-22 16:32:05 -0800120static const char * const mem_cgroup_lru_names[] = {
121 "inactive_anon",
122 "active_anon",
123 "inactive_file",
124 "active_file",
125 "unevictable",
126};
127
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700128#define THRESHOLDS_EVENTS_TARGET 128
129#define SOFTLIMIT_EVENTS_TARGET 1024
130#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700131
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700132/*
133 * Cgroups above their limits are maintained in a RB-Tree, independent of
134 * their hierarchy representation
135 */
136
Mel Gormanef8f2322016-07-28 15:46:05 -0700137struct mem_cgroup_tree_per_node {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700138 struct rb_root rb_root;
139 spinlock_t lock;
140};
141
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700142struct mem_cgroup_tree {
143 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
144};
145
146static struct mem_cgroup_tree soft_limit_tree __read_mostly;
147
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700148/* for OOM */
149struct mem_cgroup_eventfd_list {
150 struct list_head list;
151 struct eventfd_ctx *eventfd;
152};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800153
Tejun Heo79bd9812013-11-22 18:20:42 -0500154/*
155 * cgroup_event represents events which userspace want to receive.
156 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500157struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500158 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500159 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500160 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500161 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500162 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500163 * eventfd to signal userspace about the event.
164 */
165 struct eventfd_ctx *eventfd;
166 /*
167 * Each of these stored in a list by the cgroup.
168 */
169 struct list_head list;
170 /*
Tejun Heofba94802013-11-22 18:20:43 -0500171 * register_event() callback will be used to add new userspace
172 * waiter for changes related to this event. Use eventfd_signal()
173 * on eventfd to send notification to userspace.
174 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500175 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500176 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500177 /*
178 * unregister_event() callback will be called when userspace closes
179 * the eventfd or on cgroup removing. This callback must be set,
180 * if you want provide notification functionality.
181 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500182 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500183 struct eventfd_ctx *eventfd);
184 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500185 * All fields below needed to unregister event when
186 * userspace closes eventfd.
187 */
188 poll_table pt;
189 wait_queue_head_t *wqh;
190 wait_queue_t wait;
191 struct work_struct remove;
192};
193
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700194static void mem_cgroup_threshold(struct mem_cgroup *memcg);
195static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800196
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800197/* Stuffs for move charges at task migration. */
198/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800199 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800200 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800201#define MOVE_ANON 0x1U
202#define MOVE_FILE 0x2U
203#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800204
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800205/* "mc" and its members are protected by cgroup_mutex */
206static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800207 spinlock_t lock; /* for from, to */
Tejun Heo264a0ae2016-04-21 19:09:02 -0400208 struct mm_struct *mm;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800209 struct mem_cgroup *from;
210 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800211 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800212 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800213 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800214 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800215 struct task_struct *moving_task; /* a task moving charges */
216 wait_queue_head_t waitq; /* a waitq for other context */
217} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700218 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800219 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
220};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800221
Balbir Singh4e416952009-09-23 15:56:39 -0700222/*
223 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
224 * limit reclaim to prevent infinite loops, if they ever occur.
225 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700226#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700227#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700228
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800229enum charge_type {
230 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700231 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800232 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700233 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700234 NR_CHARGE_TYPE,
235};
236
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800237/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800238enum res_type {
239 _MEM,
240 _MEMSWAP,
241 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800242 _KMEM,
Vladimir Davydovd55f90b2016-01-20 15:02:44 -0800243 _TCP,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800244};
245
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700246#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
247#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800248#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700249/* Used for OOM nofiier */
250#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800251
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700252/* Some nice accessors for the vmpressure. */
253struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
254{
255 if (!memcg)
256 memcg = root_mem_cgroup;
257 return &memcg->vmpressure;
258}
259
260struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
261{
262 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
263}
264
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700265static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
266{
267 return (memcg == root_mem_cgroup);
268}
269
Johannes Weiner127424c2016-01-20 15:02:32 -0800270#ifndef CONFIG_SLOB
Glauber Costa55007d82012-12-18 14:22:38 -0800271/*
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800272 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800273 * The main reason for not using cgroup id for this:
274 * this works better in sparse environments, where we have a lot of memcgs,
275 * but only a few kmem-limited. Or also, if we have, for instance, 200
276 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
277 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800278 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800279 * The current size of the caches array is stored in memcg_nr_cache_ids. It
280 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800281 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800282static DEFINE_IDA(memcg_cache_ida);
283int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800284
Vladimir Davydov05257a12015-02-12 14:59:01 -0800285/* Protects memcg_nr_cache_ids */
286static DECLARE_RWSEM(memcg_cache_ids_sem);
287
288void memcg_get_cache_ids(void)
289{
290 down_read(&memcg_cache_ids_sem);
291}
292
293void memcg_put_cache_ids(void)
294{
295 up_read(&memcg_cache_ids_sem);
296}
297
Glauber Costa55007d82012-12-18 14:22:38 -0800298/*
299 * MIN_SIZE is different than 1, because we would like to avoid going through
300 * the alloc/free process all the time. In a small machine, 4 kmem-limited
301 * cgroups is a reasonable guess. In the future, it could be a parameter or
302 * tunable, but that is strictly not necessary.
303 *
Li Zefanb8627832013-09-23 16:56:47 +0800304 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800305 * this constant directly from cgroup, but it is understandable that this is
306 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800307 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800308 * increase ours as well if it increases.
309 */
310#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800311#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800312
Glauber Costad7f25f82012-12-18 14:22:40 -0800313/*
314 * A lot of the calls to the cache allocation functions are expected to be
315 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
316 * conditional to this static branch, we'll have to allow modules that does
317 * kmem_cache_alloc and the such to see this symbol as well
318 */
Johannes Weineref129472016-01-14 15:21:34 -0800319DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
Glauber Costad7f25f82012-12-18 14:22:40 -0800320EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800321
Tejun Heo17cc4df2017-02-22 15:41:36 -0800322struct workqueue_struct *memcg_kmem_cache_wq;
323
Johannes Weiner127424c2016-01-20 15:02:32 -0800324#endif /* !CONFIG_SLOB */
Glauber Costaa8964b92012-12-18 14:22:09 -0800325
Tejun Heoad7fa852015-05-27 20:00:02 -0400326/**
327 * mem_cgroup_css_from_page - css of the memcg associated with a page
328 * @page: page of interest
329 *
330 * If memcg is bound to the default hierarchy, css of the memcg associated
331 * with @page is returned. The returned css remains associated with @page
332 * until it is released.
333 *
334 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
335 * is returned.
Tejun Heoad7fa852015-05-27 20:00:02 -0400336 */
337struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
338{
339 struct mem_cgroup *memcg;
340
Tejun Heoad7fa852015-05-27 20:00:02 -0400341 memcg = page->mem_cgroup;
342
Tejun Heo9e10a132015-09-18 11:56:28 -0400343 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400344 memcg = root_mem_cgroup;
345
Tejun Heoad7fa852015-05-27 20:00:02 -0400346 return &memcg->css;
347}
348
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700349/**
350 * page_cgroup_ino - return inode number of the memcg a page is charged to
351 * @page: the page
352 *
353 * Look up the closest online ancestor of the memory cgroup @page is charged to
354 * and return its inode number or 0 if @page is not charged to any cgroup. It
355 * is safe to call this function without holding a reference to @page.
356 *
357 * Note, this function is inherently racy, because there is nothing to prevent
358 * the cgroup inode from getting torn down and potentially reallocated a moment
359 * after page_cgroup_ino() returns, so it only should be used by callers that
360 * do not care (such as procfs interfaces).
361 */
362ino_t page_cgroup_ino(struct page *page)
363{
364 struct mem_cgroup *memcg;
365 unsigned long ino = 0;
366
367 rcu_read_lock();
368 memcg = READ_ONCE(page->mem_cgroup);
369 while (memcg && !(memcg->css.flags & CSS_ONLINE))
370 memcg = parent_mem_cgroup(memcg);
371 if (memcg)
372 ino = cgroup_ino(memcg->css.cgroup);
373 rcu_read_unlock();
374 return ino;
375}
376
Mel Gormanef8f2322016-07-28 15:46:05 -0700377static struct mem_cgroup_per_node *
378mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700379{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700380 int nid = page_to_nid(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700381
Mel Gormanef8f2322016-07-28 15:46:05 -0700382 return memcg->nodeinfo[nid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700383}
384
Mel Gormanef8f2322016-07-28 15:46:05 -0700385static struct mem_cgroup_tree_per_node *
386soft_limit_tree_node(int nid)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700387{
Mel Gormanef8f2322016-07-28 15:46:05 -0700388 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700389}
390
Mel Gormanef8f2322016-07-28 15:46:05 -0700391static struct mem_cgroup_tree_per_node *
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700392soft_limit_tree_from_page(struct page *page)
393{
394 int nid = page_to_nid(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700395
Mel Gormanef8f2322016-07-28 15:46:05 -0700396 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700397}
398
Mel Gormanef8f2322016-07-28 15:46:05 -0700399static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz,
400 struct mem_cgroup_tree_per_node *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800401 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700402{
403 struct rb_node **p = &mctz->rb_root.rb_node;
404 struct rb_node *parent = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700405 struct mem_cgroup_per_node *mz_node;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700406
407 if (mz->on_tree)
408 return;
409
410 mz->usage_in_excess = new_usage_in_excess;
411 if (!mz->usage_in_excess)
412 return;
413 while (*p) {
414 parent = *p;
Mel Gormanef8f2322016-07-28 15:46:05 -0700415 mz_node = rb_entry(parent, struct mem_cgroup_per_node,
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700416 tree_node);
417 if (mz->usage_in_excess < mz_node->usage_in_excess)
418 p = &(*p)->rb_left;
419 /*
420 * We can't avoid mem cgroups that are over their soft
421 * limit by the same amount
422 */
423 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
424 p = &(*p)->rb_right;
425 }
426 rb_link_node(&mz->tree_node, parent, p);
427 rb_insert_color(&mz->tree_node, &mctz->rb_root);
428 mz->on_tree = true;
429}
430
Mel Gormanef8f2322016-07-28 15:46:05 -0700431static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
432 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700433{
434 if (!mz->on_tree)
435 return;
436 rb_erase(&mz->tree_node, &mctz->rb_root);
437 mz->on_tree = false;
438}
439
Mel Gormanef8f2322016-07-28 15:46:05 -0700440static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
441 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700442{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700443 unsigned long flags;
444
445 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700446 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700447 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700448}
449
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800450static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
451{
452 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700453 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800454 unsigned long excess = 0;
455
456 if (nr_pages > soft_limit)
457 excess = nr_pages - soft_limit;
458
459 return excess;
460}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700461
462static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
463{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800464 unsigned long excess;
Mel Gormanef8f2322016-07-28 15:46:05 -0700465 struct mem_cgroup_per_node *mz;
466 struct mem_cgroup_tree_per_node *mctz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700467
Jianyu Zhane2318752014-06-06 14:38:20 -0700468 mctz = soft_limit_tree_from_page(page);
Laurent Dufourbfc72282017-03-09 16:17:06 -0800469 if (!mctz)
470 return;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700471 /*
472 * Necessary to update all ancestors when hierarchy is used.
473 * because their event counter is not touched.
474 */
475 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700476 mz = mem_cgroup_page_nodeinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800477 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700478 /*
479 * We have to update the tree if mz is on RB-tree or
480 * mem is over its softlimit.
481 */
482 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700483 unsigned long flags;
484
485 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700486 /* if on-tree, remove it */
487 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700488 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700489 /*
490 * Insert again. mz->usage_in_excess will be updated.
491 * If excess is 0, no tree ops.
492 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700493 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700494 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700495 }
496 }
497}
498
499static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
500{
Mel Gormanef8f2322016-07-28 15:46:05 -0700501 struct mem_cgroup_tree_per_node *mctz;
502 struct mem_cgroup_per_node *mz;
503 int nid;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700504
Jianyu Zhane2318752014-06-06 14:38:20 -0700505 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700506 mz = mem_cgroup_nodeinfo(memcg, nid);
507 mctz = soft_limit_tree_node(nid);
Laurent Dufourbfc72282017-03-09 16:17:06 -0800508 if (mctz)
509 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700510 }
511}
512
Mel Gormanef8f2322016-07-28 15:46:05 -0700513static struct mem_cgroup_per_node *
514__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700515{
516 struct rb_node *rightmost = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700517 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700518
519retry:
520 mz = NULL;
521 rightmost = rb_last(&mctz->rb_root);
522 if (!rightmost)
523 goto done; /* Nothing to reclaim from */
524
Mel Gormanef8f2322016-07-28 15:46:05 -0700525 mz = rb_entry(rightmost, struct mem_cgroup_per_node, tree_node);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700526 /*
527 * Remove the node now but someone else can add it back,
528 * we will to add it back at the end of reclaim to its correct
529 * position in the tree.
530 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700531 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800532 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400533 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700534 goto retry;
535done:
536 return mz;
537}
538
Mel Gormanef8f2322016-07-28 15:46:05 -0700539static struct mem_cgroup_per_node *
540mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700541{
Mel Gormanef8f2322016-07-28 15:46:05 -0700542 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700543
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700544 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700545 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700546 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700547 return mz;
548}
549
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700550/*
Greg Thelen484ebb32015-10-01 15:37:05 -0700551 * Return page count for single (non recursive) @memcg.
552 *
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700553 * Implementation Note: reading percpu statistics for memcg.
554 *
555 * Both of vmstat[] and percpu_counter has threshold and do periodic
556 * synchronization to implement "quick" read. There are trade-off between
557 * reading cost and precision of value. Then, we may have a chance to implement
Greg Thelen484ebb32015-10-01 15:37:05 -0700558 * a periodic synchronization of counter in memcg's counter.
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700559 *
560 * But this _read() function is used for user interface now. The user accounts
561 * memory usage by memory cgroup and he _always_ requires exact value because
562 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
563 * have to visit all online cpus and make sum. So, for now, unnecessary
564 * synchronization is not implemented. (just implemented for cpu hotplug)
565 *
566 * If there are kernel internal actions which can make use of some not-exact
567 * value, and reading all cpu value can be performance bottleneck in some
Greg Thelen484ebb32015-10-01 15:37:05 -0700568 * common workload, threshold and synchronization as vmstat[] should be
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700569 * implemented.
570 */
Greg Thelen484ebb32015-10-01 15:37:05 -0700571static unsigned long
572mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800573{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700574 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800575 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800576
Greg Thelen484ebb32015-10-01 15:37:05 -0700577 /* Per-cpu values can be negative, use a signed accumulator */
Tejun Heo733a5722015-05-22 18:23:18 -0400578 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700579 val += per_cpu(memcg->stat->count[idx], cpu);
Greg Thelen484ebb32015-10-01 15:37:05 -0700580 /*
581 * Summing races with updates, so val may be negative. Avoid exposing
582 * transient negative values.
583 */
584 if (val < 0)
585 val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800586 return val;
587}
588
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700589static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700590 enum mem_cgroup_events_index idx)
591{
592 unsigned long val = 0;
593 int cpu;
594
Tejun Heo733a5722015-05-22 18:23:18 -0400595 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700596 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700597 return val;
598}
599
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700600static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700601 struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800602 bool compound, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800603{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700604 /*
605 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
606 * counted as CACHE even if it's on ANON LRU.
607 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700608 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700609 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700610 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800611 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700612 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700613 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700614
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800615 if (compound) {
616 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
David Rientjesb070e652013-05-07 16:18:09 -0700617 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
618 nr_pages);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800619 }
David Rientjesb070e652013-05-07 16:18:09 -0700620
KAMEZAWA Hiroyukie401f1762011-01-20 14:44:23 -0800621 /* pagein of a big page is an event. So, ignore page size */
622 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700623 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800624 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700625 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800626 nr_pages = -nr_pages; /* for event */
627 }
KAMEZAWA Hiroyukie401f1762011-01-20 14:44:23 -0800628
Johannes Weiner13114712012-05-29 15:07:07 -0700629 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800630}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800631
Vladimir Davydov0a6b76d2016-03-17 14:18:42 -0700632unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
633 int nid, unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700634{
Michal Hockob4536f0c82017-01-10 16:58:04 -0800635 struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg);
Jianyu Zhane2318752014-06-06 14:38:20 -0700636 unsigned long nr = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -0700637 enum lru_list lru;
Ying Han889976d2011-05-26 16:25:33 -0700638
Jianyu Zhane2318752014-06-06 14:38:20 -0700639 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700640
Mel Gormanef8f2322016-07-28 15:46:05 -0700641 for_each_lru(lru) {
642 if (!(BIT(lru) & lru_mask))
643 continue;
Michal Hockob4536f0c82017-01-10 16:58:04 -0800644 nr += mem_cgroup_get_lru_size(lruvec, lru);
Jianyu Zhane2318752014-06-06 14:38:20 -0700645 }
646 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700647}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700648
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700649static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700650 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800651{
Jianyu Zhane2318752014-06-06 14:38:20 -0700652 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700653 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800654
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800655 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700656 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
657 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800658}
659
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800660static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
661 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800662{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700663 unsigned long val, next;
664
Johannes Weiner13114712012-05-29 15:07:07 -0700665 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700666 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700667 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800668 if ((long)next - (long)val < 0) {
669 switch (target) {
670 case MEM_CGROUP_TARGET_THRESH:
671 next = val + THRESHOLDS_EVENTS_TARGET;
672 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700673 case MEM_CGROUP_TARGET_SOFTLIMIT:
674 next = val + SOFTLIMIT_EVENTS_TARGET;
675 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800676 case MEM_CGROUP_TARGET_NUMAINFO:
677 next = val + NUMAINFO_EVENTS_TARGET;
678 break;
679 default:
680 break;
681 }
682 __this_cpu_write(memcg->stat->targets[target], next);
683 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700684 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800685 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800686}
687
688/*
689 * Check events in order.
690 *
691 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700692static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800693{
694 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800695 if (unlikely(mem_cgroup_event_ratelimit(memcg,
696 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700697 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800698 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800699
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700700 do_softlimit = mem_cgroup_event_ratelimit(memcg,
701 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700702#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800703 do_numainfo = mem_cgroup_event_ratelimit(memcg,
704 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700705#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800706 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700707 if (unlikely(do_softlimit))
708 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800709#if MAX_NUMNODES > 1
710 if (unlikely(do_numainfo))
711 atomic_inc(&memcg->numainfo_events);
712#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700713 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800714}
715
Balbir Singhcf475ad2008-04-29 01:00:16 -0700716struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800717{
Balbir Singh31a78f22008-09-28 23:09:31 +0100718 /*
719 * mm_update_next_owner() may clear mm->owner to NULL
720 * if it races with swapoff, page migration, etc.
721 * So this can be called with p == NULL.
722 */
723 if (unlikely(!p))
724 return NULL;
725
Tejun Heo073219e2014-02-08 10:36:58 -0500726 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800727}
Michal Hocko33398cf2015-09-08 15:01:02 -0700728EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800729
Johannes Weinerdf381972014-04-07 15:37:43 -0700730static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800731{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700732 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700733
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800734 rcu_read_lock();
735 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700736 /*
737 * Page cache insertions can happen withou an
738 * actual mm context, e.g. during disk probing
739 * on boot, loopback IO, acct() writes etc.
740 */
741 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700742 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700743 else {
744 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
745 if (unlikely(!memcg))
746 memcg = root_mem_cgroup;
747 }
Tejun Heoec903c02014-05-13 12:11:01 -0400748 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800749 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700750 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800751}
752
Johannes Weiner56600482012-01-12 17:17:59 -0800753/**
754 * mem_cgroup_iter - iterate over memory cgroup hierarchy
755 * @root: hierarchy root
756 * @prev: previously returned memcg, NULL on first invocation
757 * @reclaim: cookie for shared reclaim walks, NULL for full walks
758 *
759 * Returns references to children of the hierarchy below @root, or
760 * @root itself, or %NULL after a full round-trip.
761 *
762 * Caller must pass the return value in @prev on subsequent
763 * invocations for reference counting, or use mem_cgroup_iter_break()
764 * to cancel a hierarchy walk before the round-trip is complete.
765 *
766 * Reclaimers can specify a zone and a priority level in @reclaim to
767 * divide up the memcgs in the hierarchy among all concurrent
768 * reclaimers operating on the same zone and priority.
769 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700770struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800771 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700772 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700773{
Michal Hocko33398cf2015-09-08 15:01:02 -0700774 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800775 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800776 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800777 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700778
Andrew Morton694fbc02013-09-24 15:27:37 -0700779 if (mem_cgroup_disabled())
780 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800781
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700782 if (!root)
783 root = root_mem_cgroup;
784
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800785 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800786 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800787
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800788 if (!root->use_hierarchy && root != root_mem_cgroup) {
789 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800790 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700791 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800792 }
793
Michal Hocko542f85f2013-04-29 15:07:15 -0700794 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800795
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800796 if (reclaim) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700797 struct mem_cgroup_per_node *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800798
Mel Gormanef8f2322016-07-28 15:46:05 -0700799 mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800800 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700801
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800802 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700803 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800804
Vladimir Davydov6df38682015-12-29 14:54:10 -0800805 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -0700806 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800807 if (!pos || css_tryget(&pos->css))
808 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800809 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800810 * css reference reached zero, so iter->position will
811 * be cleared by ->css_released. However, we should not
812 * rely on this happening soon, because ->css_released
813 * is called from a work queue, and by busy-waiting we
814 * might block it. So we clear iter->position right
815 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800816 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800817 (void)cmpxchg(&iter->position, pos, NULL);
818 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800819 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800820
821 if (pos)
822 css = &pos->css;
823
824 for (;;) {
825 css = css_next_descendant_pre(css, &root->css);
826 if (!css) {
827 /*
828 * Reclaimers share the hierarchy walk, and a
829 * new one might jump in right at the end of
830 * the hierarchy - make sure they see at least
831 * one group and restart from the beginning.
832 */
833 if (!prev)
834 continue;
835 break;
836 }
837
838 /*
839 * Verify the css and acquire a reference. The root
840 * is provided by the caller, so we know it's alive
841 * and kicking, and don't take an extra reference.
842 */
843 memcg = mem_cgroup_from_css(css);
844
845 if (css == &root->css)
846 break;
847
Johannes Weiner0b8f73e2016-01-20 15:02:53 -0800848 if (css_tryget(css))
849 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800850
851 memcg = NULL;
852 }
853
854 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800855 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800856 * The position could have already been updated by a competing
857 * thread, so check that the value hasn't changed since we read
858 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800859 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800860 (void)cmpxchg(&iter->position, pos, memcg);
861
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800862 if (pos)
863 css_put(&pos->css);
864
865 if (!memcg)
866 iter->generation++;
867 else if (!prev)
868 reclaim->generation = iter->generation;
869 }
870
Michal Hocko542f85f2013-04-29 15:07:15 -0700871out_unlock:
872 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800873out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700874 if (prev && prev != root)
875 css_put(&prev->css);
876
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800877 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700878}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800879
Johannes Weiner56600482012-01-12 17:17:59 -0800880/**
881 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
882 * @root: hierarchy root
883 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
884 */
885void mem_cgroup_iter_break(struct mem_cgroup *root,
886 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800887{
888 if (!root)
889 root = root_mem_cgroup;
890 if (prev && prev != root)
891 css_put(&prev->css);
892}
893
Vladimir Davydov6df38682015-12-29 14:54:10 -0800894static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
895{
896 struct mem_cgroup *memcg = dead_memcg;
897 struct mem_cgroup_reclaim_iter *iter;
Mel Gormanef8f2322016-07-28 15:46:05 -0700898 struct mem_cgroup_per_node *mz;
899 int nid;
Vladimir Davydov6df38682015-12-29 14:54:10 -0800900 int i;
901
902 while ((memcg = parent_mem_cgroup(memcg))) {
903 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700904 mz = mem_cgroup_nodeinfo(memcg, nid);
905 for (i = 0; i <= DEF_PRIORITY; i++) {
906 iter = &mz->iter[i];
907 cmpxchg(&iter->position,
908 dead_memcg, NULL);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800909 }
910 }
911 }
912}
913
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700914/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800915 * Iteration constructs for visiting all cgroups (under a tree). If
916 * loops are exited prematurely (break), mem_cgroup_iter_break() must
917 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700918 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800919#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800920 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800921 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800922 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700923
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800924#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800925 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800926 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800927 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700928
Johannes Weiner925b7672012-01-12 17:18:15 -0800929/**
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700930 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
931 * @memcg: hierarchy root
932 * @fn: function to call for each task
933 * @arg: argument passed to @fn
934 *
935 * This function iterates over tasks attached to @memcg or to any of its
936 * descendants and calls @fn for each task. If @fn returns a non-zero
937 * value, the function breaks the iteration loop and returns the value.
938 * Otherwise, it will iterate over all tasks and return 0.
939 *
940 * This function must not be called for the root memory cgroup.
941 */
942int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
943 int (*fn)(struct task_struct *, void *), void *arg)
944{
945 struct mem_cgroup *iter;
946 int ret = 0;
947
948 BUG_ON(memcg == root_mem_cgroup);
949
950 for_each_mem_cgroup_tree(iter, memcg) {
951 struct css_task_iter it;
952 struct task_struct *task;
953
954 css_task_iter_start(&iter->css, &it);
955 while (!ret && (task = css_task_iter_next(&it)))
956 ret = fn(task, arg);
957 css_task_iter_end(&it);
958 if (ret) {
959 mem_cgroup_iter_break(memcg, iter);
960 break;
961 }
962 }
963 return ret;
964}
965
966/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800967 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -0800968 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700969 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800970 *
971 * This function is only safe when following the LRU page isolation
972 * and putback protocol: the LRU lock must be held, and the page must
973 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -0700974 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700975struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat)
Minchan Kim3f58a822011-03-22 16:32:53 -0700976{
Mel Gormanef8f2322016-07-28 15:46:05 -0700977 struct mem_cgroup_per_node *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -0800978 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800979 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800980
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800981 if (mem_cgroup_disabled()) {
Mel Gorman599d0c92016-07-28 15:45:31 -0700982 lruvec = &pgdat->lruvec;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800983 goto out;
984 }
Christoph Lameterb69408e2008-10-18 20:26:14 -0700985
Johannes Weiner1306a852014-12-10 15:44:52 -0800986 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800987 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800988 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -0800989 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -0800990 */
Johannes Weiner29833312014-12-10 15:44:02 -0800991 if (!memcg)
992 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800993
Mel Gormanef8f2322016-07-28 15:46:05 -0700994 mz = mem_cgroup_page_nodeinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800995 lruvec = &mz->lruvec;
996out:
997 /*
998 * Since a node can be onlined after the mem_cgroup was created,
999 * we have to be prepared to initialize lruvec->zone here;
1000 * and if offlined then reonlined, we need to reinitialize it.
1001 */
Mel Gorman599d0c92016-07-28 15:45:31 -07001002 if (unlikely(lruvec->pgdat != pgdat))
1003 lruvec->pgdat = pgdat;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001004 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001005}
1006
1007/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001008 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1009 * @lruvec: mem_cgroup per zone lru vector
1010 * @lru: index of lru list the page is sitting on
Michal Hockob4536f0c82017-01-10 16:58:04 -08001011 * @zid: zone id of the accounted pages
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001012 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001013 *
Hugh Dickinsca707232016-05-19 17:12:35 -07001014 * This function must be called under lru_lock, just before a page is added
1015 * to or just after a page is removed from an lru list (that ordering being
1016 * so as to allow it to check that lru_size 0 is consistent with list_empty).
Johannes Weiner925b7672012-01-12 17:18:15 -08001017 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001018void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
Michal Hockob4536f0c82017-01-10 16:58:04 -08001019 int zid, int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001020{
Mel Gormanef8f2322016-07-28 15:46:05 -07001021 struct mem_cgroup_per_node *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001022 unsigned long *lru_size;
Hugh Dickinsca707232016-05-19 17:12:35 -07001023 long size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001024
1025 if (mem_cgroup_disabled())
1026 return;
1027
Mel Gormanef8f2322016-07-28 15:46:05 -07001028 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
Michal Hockob4536f0c82017-01-10 16:58:04 -08001029 lru_size = &mz->lru_zone_size[zid][lru];
Hugh Dickinsca707232016-05-19 17:12:35 -07001030
1031 if (nr_pages < 0)
1032 *lru_size += nr_pages;
1033
1034 size = *lru_size;
Michal Hockob4536f0c82017-01-10 16:58:04 -08001035 if (WARN_ONCE(size < 0,
1036 "%s(%p, %d, %d): lru_size %ld\n",
1037 __func__, lruvec, lru, nr_pages, size)) {
Hugh Dickinsca707232016-05-19 17:12:35 -07001038 VM_BUG_ON(1);
1039 *lru_size = 0;
1040 }
1041
1042 if (nr_pages > 0)
1043 *lru_size += nr_pages;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001044}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001045
Johannes Weiner2314b422014-12-10 15:44:33 -08001046bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001047{
Johannes Weiner2314b422014-12-10 15:44:33 -08001048 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001049 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001050 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001051
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001052 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001053 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001054 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001055 task_unlock(p);
1056 } else {
1057 /*
1058 * All threads may have already detached their mm's, but the oom
1059 * killer still needs to detect if they have already been oom
1060 * killed to prevent needlessly killing additional tasks.
1061 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001062 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001063 task_memcg = mem_cgroup_from_task(task);
1064 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001065 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001066 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001067 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1068 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001069 return ret;
1070}
1071
Johannes Weiner19942822011-02-01 15:52:43 -08001072/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001073 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001074 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001075 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001076 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001077 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001078 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001079static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001080{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001081 unsigned long margin = 0;
1082 unsigned long count;
1083 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001084
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001085 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001086 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001087 if (count < limit)
1088 margin = limit - count;
1089
Johannes Weiner7941d212016-01-14 15:21:23 -08001090 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001091 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001092 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001093 if (count <= limit)
1094 margin = min(margin, limit - count);
Li RongQingcbedbac2016-05-27 14:27:43 -07001095 else
1096 margin = 0;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001097 }
1098
1099 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001100}
1101
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001102/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001103 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001104 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001105 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1106 * moving cgroups. This is for waiting at high-memory pressure
1107 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001108 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001109static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001110{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001111 struct mem_cgroup *from;
1112 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001113 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001114 /*
1115 * Unlike task_move routines, we access mc.to, mc.from not under
1116 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1117 */
1118 spin_lock(&mc.lock);
1119 from = mc.from;
1120 to = mc.to;
1121 if (!from)
1122 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001123
Johannes Weiner2314b422014-12-10 15:44:33 -08001124 ret = mem_cgroup_is_descendant(from, memcg) ||
1125 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001126unlock:
1127 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001128 return ret;
1129}
1130
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001131static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001132{
1133 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001134 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001135 DEFINE_WAIT(wait);
1136 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1137 /* moving charge context might have finished. */
1138 if (mc.moving_task)
1139 schedule();
1140 finish_wait(&mc.waitq, &wait);
1141 return true;
1142 }
1143 }
1144 return false;
1145}
1146
Sha Zhengju58cf1882013-02-22 16:32:05 -08001147#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001148/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001149 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001150 * @memcg: The memory cgroup that went over limit
1151 * @p: Task that is going to be killed
1152 *
1153 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1154 * enabled
1155 */
1156void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1157{
Sha Zhengju58cf1882013-02-22 16:32:05 -08001158 struct mem_cgroup *iter;
1159 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001160
Balbir Singhe2224322009-04-02 16:57:39 -07001161 rcu_read_lock();
1162
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001163 if (p) {
1164 pr_info("Task in ");
1165 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1166 pr_cont(" killed as a result of limit of ");
1167 } else {
1168 pr_info("Memory limit reached of cgroup ");
1169 }
1170
Tejun Heoe61734c2014-02-12 09:29:50 -05001171 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001172 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001173
Balbir Singhe2224322009-04-02 16:57:39 -07001174 rcu_read_unlock();
1175
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001176 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1177 K((u64)page_counter_read(&memcg->memory)),
1178 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1179 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1180 K((u64)page_counter_read(&memcg->memsw)),
1181 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1182 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1183 K((u64)page_counter_read(&memcg->kmem)),
1184 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001185
1186 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001187 pr_info("Memory cgroup stats for ");
1188 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001189 pr_cont(":");
1190
1191 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08001192 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Sha Zhengju58cf1882013-02-22 16:32:05 -08001193 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07001194 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
Sha Zhengju58cf1882013-02-22 16:32:05 -08001195 K(mem_cgroup_read_stat(iter, i)));
1196 }
1197
1198 for (i = 0; i < NR_LRU_LISTS; i++)
1199 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1200 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1201
1202 pr_cont("\n");
1203 }
Balbir Singhe2224322009-04-02 16:57:39 -07001204}
1205
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001206/*
1207 * This function returns the number of memcg under hierarchy tree. Returns
1208 * 1(self count) if no children.
1209 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001210static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001211{
1212 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001213 struct mem_cgroup *iter;
1214
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001215 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001216 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001217 return num;
1218}
1219
Balbir Singh6d61ef42009-01-07 18:08:06 -08001220/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001221 * Return the memory (and swap, if configured) limit for a memcg.
1222 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001223unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001224{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001225 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001226
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001227 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001228 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001229 unsigned long memsw_limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001230 unsigned long swap_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001231
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001232 memsw_limit = memcg->memsw.limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001233 swap_limit = memcg->swap.limit;
1234 swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1235 limit = min(limit + swap_limit, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001236 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001237 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001238}
1239
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001240static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001241 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001242{
David Rientjes6e0fc462015-09-08 15:00:36 -07001243 struct oom_control oc = {
1244 .zonelist = NULL,
1245 .nodemask = NULL,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001246 .memcg = memcg,
David Rientjes6e0fc462015-09-08 15:00:36 -07001247 .gfp_mask = gfp_mask,
1248 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001249 };
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001250 bool ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001251
Johannes Weinerdc564012015-06-24 16:57:19 -07001252 mutex_lock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001253 ret = out_of_memory(&oc);
Johannes Weinerdc564012015-06-24 16:57:19 -07001254 mutex_unlock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001255 return ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001256}
1257
Michele Curtiae6e71d2014-12-12 16:56:35 -08001258#if MAX_NUMNODES > 1
1259
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001260/**
1261 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001262 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001263 * @nid: the node ID to be checked.
1264 * @noswap : specify true here if the user wants flle only information.
1265 *
1266 * This function returns whether the specified memcg contains any
1267 * reclaimable pages on a node. Returns true if there are any reclaimable
1268 * pages in the node.
1269 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001270static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001271 int nid, bool noswap)
1272{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001273 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001274 return true;
1275 if (noswap || !total_swap_pages)
1276 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001277 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001278 return true;
1279 return false;
1280
1281}
Ying Han889976d2011-05-26 16:25:33 -07001282
1283/*
1284 * Always updating the nodemask is not very good - even if we have an empty
1285 * list or the wrong list here, we can start from some node and traverse all
1286 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1287 *
1288 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001289static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001290{
1291 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001292 /*
1293 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1294 * pagein/pageout changes since the last update.
1295 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001296 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001297 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001298 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001299 return;
1300
Ying Han889976d2011-05-26 16:25:33 -07001301 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001302 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001303
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001304 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001305
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001306 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1307 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001308 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001309
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001310 atomic_set(&memcg->numainfo_events, 0);
1311 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001312}
1313
1314/*
1315 * Selecting a node where we start reclaim from. Because what we need is just
1316 * reducing usage counter, start from anywhere is O,K. Considering
1317 * memory reclaim from current node, there are pros. and cons.
1318 *
1319 * Freeing memory from current node means freeing memory from a node which
1320 * we'll use or we've used. So, it may make LRU bad. And if several threads
1321 * hit limits, it will see a contention on a node. But freeing from remote
1322 * node means more costs for memory reclaim because of memory latency.
1323 *
1324 * Now, we use round-robin. Better algorithm is welcomed.
1325 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001326int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001327{
1328 int node;
1329
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001330 mem_cgroup_may_update_nodemask(memcg);
1331 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001332
Andrew Morton0edaf862016-05-19 17:10:58 -07001333 node = next_node_in(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001334 /*
Michal Hockofda3d692016-05-19 17:11:34 -07001335 * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1336 * last time it really checked all the LRUs due to rate limiting.
1337 * Fallback to the current node in that case for simplicity.
Ying Han889976d2011-05-26 16:25:33 -07001338 */
1339 if (unlikely(node == MAX_NUMNODES))
1340 node = numa_node_id();
1341
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001342 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001343 return node;
1344}
Ying Han889976d2011-05-26 16:25:33 -07001345#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001346int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001347{
1348 return 0;
1349}
1350#endif
1351
Andrew Morton0608f432013-09-24 15:27:41 -07001352static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
Mel Gormanef8f2322016-07-28 15:46:05 -07001353 pg_data_t *pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001354 gfp_t gfp_mask,
1355 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001356{
Andrew Morton0608f432013-09-24 15:27:41 -07001357 struct mem_cgroup *victim = NULL;
1358 int total = 0;
1359 int loop = 0;
1360 unsigned long excess;
1361 unsigned long nr_scanned;
1362 struct mem_cgroup_reclaim_cookie reclaim = {
Mel Gormanef8f2322016-07-28 15:46:05 -07001363 .pgdat = pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001364 .priority = 0,
1365 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001366
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001367 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001368
Andrew Morton0608f432013-09-24 15:27:41 -07001369 while (1) {
1370 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1371 if (!victim) {
1372 loop++;
1373 if (loop >= 2) {
1374 /*
1375 * If we have not been able to reclaim
1376 * anything, it might because there are
1377 * no reclaimable pages under this hierarchy
1378 */
1379 if (!total)
1380 break;
1381 /*
1382 * We want to do more targeted reclaim.
1383 * excess >> 2 is not to excessive so as to
1384 * reclaim too much, nor too less that we keep
1385 * coming back to reclaim from this cgroup
1386 */
1387 if (total >= (excess >> 2) ||
1388 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1389 break;
1390 }
1391 continue;
1392 }
Mel Gormana9dd0a82016-07-28 15:46:02 -07001393 total += mem_cgroup_shrink_node(victim, gfp_mask, false,
Mel Gormanef8f2322016-07-28 15:46:05 -07001394 pgdat, &nr_scanned);
Andrew Morton0608f432013-09-24 15:27:41 -07001395 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001396 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001397 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001398 }
Andrew Morton0608f432013-09-24 15:27:41 -07001399 mem_cgroup_iter_break(root_memcg, victim);
1400 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001401}
1402
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001403#ifdef CONFIG_LOCKDEP
1404static struct lockdep_map memcg_oom_lock_dep_map = {
1405 .name = "memcg_oom_lock",
1406};
1407#endif
1408
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001409static DEFINE_SPINLOCK(memcg_oom_lock);
1410
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001411/*
1412 * Check OOM-Killer is already running under our hierarchy.
1413 * If someone is running, return false.
1414 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001415static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001416{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001417 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001418
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001419 spin_lock(&memcg_oom_lock);
1420
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001421 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001422 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001423 /*
1424 * this subtree of our hierarchy is already locked
1425 * so we cannot give a lock.
1426 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001427 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001428 mem_cgroup_iter_break(memcg, iter);
1429 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001430 } else
1431 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001432 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001433
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001434 if (failed) {
1435 /*
1436 * OK, we failed to lock the whole subtree so we have
1437 * to clean up what we set up to the failing subtree
1438 */
1439 for_each_mem_cgroup_tree(iter, memcg) {
1440 if (iter == failed) {
1441 mem_cgroup_iter_break(memcg, iter);
1442 break;
1443 }
1444 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001445 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001446 } else
1447 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001448
1449 spin_unlock(&memcg_oom_lock);
1450
1451 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001452}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001453
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001454static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001455{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001456 struct mem_cgroup *iter;
1457
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001458 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001459 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001460 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001461 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001462 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001463}
1464
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001465static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001466{
1467 struct mem_cgroup *iter;
1468
Tejun Heoc2b42d32015-06-24 16:58:23 -07001469 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001470 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001471 iter->under_oom++;
1472 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001473}
1474
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001475static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001476{
1477 struct mem_cgroup *iter;
1478
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001479 /*
1480 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001481 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001482 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001483 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001484 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001485 if (iter->under_oom > 0)
1486 iter->under_oom--;
1487 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001488}
1489
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001490static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1491
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001492struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001493 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001494 wait_queue_t wait;
1495};
1496
1497static int memcg_oom_wake_function(wait_queue_t *wait,
1498 unsigned mode, int sync, void *arg)
1499{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001500 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1501 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001502 struct oom_wait_info *oom_wait_info;
1503
1504 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001505 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001506
Johannes Weiner2314b422014-12-10 15:44:33 -08001507 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1508 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001509 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001510 return autoremove_wake_function(wait, mode, sync, arg);
1511}
1512
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001513static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001514{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001515 /*
1516 * For the following lockless ->under_oom test, the only required
1517 * guarantee is that it must see the state asserted by an OOM when
1518 * this function is called as a result of userland actions
1519 * triggered by the notification of the OOM. This is trivially
1520 * achieved by invoking mem_cgroup_mark_under_oom() before
1521 * triggering notification.
1522 */
1523 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001524 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001525}
1526
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001527static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001528{
Andrew Mortond0db7af2016-06-08 15:33:47 -07001529 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001530 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001531 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001532 * We are in the middle of the charge context here, so we
1533 * don't want to block when potentially sitting on a callstack
1534 * that holds all kinds of filesystem and mm locks.
1535 *
1536 * Also, the caller may handle a failed allocation gracefully
1537 * (like optional page cache readahead) and so an OOM killer
1538 * invocation might not even be necessary.
1539 *
1540 * That's why we don't do anything here except remember the
1541 * OOM context and then deal with it at the end of the page
1542 * fault when the stack is unwound, the locks are released,
1543 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001544 */
Johannes Weiner49426422013-10-16 13:46:59 -07001545 css_get(&memcg->css);
Tejun Heo626ebc42015-11-05 18:46:09 -08001546 current->memcg_in_oom = memcg;
1547 current->memcg_oom_gfp_mask = mask;
1548 current->memcg_oom_order = order;
Johannes Weiner49426422013-10-16 13:46:59 -07001549}
1550
1551/**
1552 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1553 * @handle: actually kill/wait or just clean up the OOM state
1554 *
1555 * This has to be called at the end of a page fault if the memcg OOM
1556 * handler was enabled.
1557 *
1558 * Memcg supports userspace OOM handling where failed allocations must
1559 * sleep on a waitqueue until the userspace task resolves the
1560 * situation. Sleeping directly in the charge context with all kinds
1561 * of locks held is not a good idea, instead we remember an OOM state
1562 * in the task and mem_cgroup_oom_synchronize() has to be called at
1563 * the end of the page fault to complete the OOM handling.
1564 *
1565 * Returns %true if an ongoing memcg OOM situation was detected and
1566 * completed, %false otherwise.
1567 */
1568bool mem_cgroup_oom_synchronize(bool handle)
1569{
Tejun Heo626ebc42015-11-05 18:46:09 -08001570 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001571 struct oom_wait_info owait;
1572 bool locked;
1573
1574 /* OOM is global, do not handle */
1575 if (!memcg)
1576 return false;
1577
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001578 if (!handle)
Johannes Weiner49426422013-10-16 13:46:59 -07001579 goto cleanup;
1580
1581 owait.memcg = memcg;
1582 owait.wait.flags = 0;
1583 owait.wait.func = memcg_oom_wake_function;
1584 owait.wait.private = current;
1585 INIT_LIST_HEAD(&owait.wait.task_list);
1586
1587 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001588 mem_cgroup_mark_under_oom(memcg);
1589
1590 locked = mem_cgroup_oom_trylock(memcg);
1591
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001592 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001593 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001594
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001595 if (locked && !memcg->oom_kill_disable) {
1596 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001597 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001598 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1599 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001600 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001601 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001602 mem_cgroup_unmark_under_oom(memcg);
1603 finish_wait(&memcg_oom_waitq, &owait.wait);
1604 }
1605
1606 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001607 mem_cgroup_oom_unlock(memcg);
1608 /*
1609 * There is no guarantee that an OOM-lock contender
1610 * sees the wakeups triggered by the OOM kill
1611 * uncharges. Wake any sleepers explicitely.
1612 */
1613 memcg_oom_recover(memcg);
1614 }
Johannes Weiner49426422013-10-16 13:46:59 -07001615cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001616 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001617 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001618 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001619}
1620
Johannes Weinerd7365e72014-10-29 14:50:48 -07001621/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001622 * lock_page_memcg - lock a page->mem_cgroup binding
1623 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001624 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001625 * This function protects unlocked LRU pages from being moved to
1626 * another cgroup and stabilizes their page->mem_cgroup binding.
Balbir Singhd69b0422009-06-17 16:26:34 -07001627 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001628void lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001629{
1630 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001631 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001632
Johannes Weiner6de22612015-02-11 15:25:01 -08001633 /*
1634 * The RCU lock is held throughout the transaction. The fast
1635 * path can get away without acquiring the memcg->move_lock
1636 * because page moving starts with an RCU grace period.
Johannes Weiner6de22612015-02-11 15:25:01 -08001637 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001638 rcu_read_lock();
1639
1640 if (mem_cgroup_disabled())
Johannes Weiner62cccb82016-03-15 14:57:22 -07001641 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001642again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001643 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001644 if (unlikely(!memcg))
Johannes Weiner62cccb82016-03-15 14:57:22 -07001645 return;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001646
Qiang Huangbdcbb652014-06-04 16:08:21 -07001647 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001648 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001649
Johannes Weiner6de22612015-02-11 15:25:01 -08001650 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001651 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001652 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001653 goto again;
1654 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001655
1656 /*
1657 * When charge migration first begins, we can have locked and
1658 * unlocked page stat updates happening concurrently. Track
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001659 * the task who has the lock for unlock_page_memcg().
Johannes Weiner6de22612015-02-11 15:25:01 -08001660 */
1661 memcg->move_lock_task = current;
1662 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001663
Johannes Weiner62cccb82016-03-15 14:57:22 -07001664 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001665}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001666EXPORT_SYMBOL(lock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001667
Johannes Weinerd7365e72014-10-29 14:50:48 -07001668/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001669 * unlock_page_memcg - unlock a page->mem_cgroup binding
Johannes Weiner62cccb82016-03-15 14:57:22 -07001670 * @page: the page
Johannes Weinerd7365e72014-10-29 14:50:48 -07001671 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001672void unlock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001673{
Johannes Weiner62cccb82016-03-15 14:57:22 -07001674 struct mem_cgroup *memcg = page->mem_cgroup;
1675
Johannes Weiner6de22612015-02-11 15:25:01 -08001676 if (memcg && memcg->move_lock_task == current) {
1677 unsigned long flags = memcg->move_lock_flags;
1678
1679 memcg->move_lock_task = NULL;
1680 memcg->move_lock_flags = 0;
1681
1682 spin_unlock_irqrestore(&memcg->move_lock, flags);
1683 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001684
Johannes Weinerd7365e72014-10-29 14:50:48 -07001685 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001686}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001687EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001688
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001689/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001690 * size of first charge trial. "32" comes from vmscan.c's magic value.
1691 * TODO: maybe necessary to use big numbers in big irons.
1692 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001693#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001694struct memcg_stock_pcp {
1695 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001696 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001697 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001698 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001699#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001700};
1701static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001702static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001703
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001704/**
1705 * consume_stock: Try to consume stocked charge on this cpu.
1706 * @memcg: memcg to consume from.
1707 * @nr_pages: how many pages to charge.
1708 *
1709 * The charges will only happen if @memcg matches the current cpu's memcg
1710 * stock, and at least @nr_pages are available in that stock. Failure to
1711 * service an allocation will refill the stock.
1712 *
1713 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001714 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001715static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001716{
1717 struct memcg_stock_pcp *stock;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001718 unsigned long flags;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001719 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001720
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001721 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001722 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001723
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001724 local_irq_save(flags);
1725
1726 stock = this_cpu_ptr(&memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001727 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001728 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001729 ret = true;
1730 }
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001731
1732 local_irq_restore(flags);
1733
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001734 return ret;
1735}
1736
1737/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001738 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001739 */
1740static void drain_stock(struct memcg_stock_pcp *stock)
1741{
1742 struct mem_cgroup *old = stock->cached;
1743
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001744 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001745 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001746 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001747 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001748 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001749 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001750 }
1751 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001752}
1753
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001754static void drain_local_stock(struct work_struct *dummy)
1755{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001756 struct memcg_stock_pcp *stock;
1757 unsigned long flags;
1758
1759 local_irq_save(flags);
1760
1761 stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001762 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001763 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001764
1765 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001766}
1767
1768/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001769 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001770 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001771 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001772static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001773{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001774 struct memcg_stock_pcp *stock;
1775 unsigned long flags;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001776
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001777 local_irq_save(flags);
1778
1779 stock = this_cpu_ptr(&memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001780 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001781 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001782 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001783 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001784 stock->nr_pages += nr_pages;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001785
1786 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001787}
1788
1789/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001790 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001791 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001792 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001793static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001794{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001795 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001796
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001797 /* If someone's already draining, avoid adding running more workers. */
1798 if (!mutex_trylock(&percpu_charge_mutex))
1799 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001800 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001801 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001802 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001803 for_each_online_cpu(cpu) {
1804 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001805 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001806
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001807 memcg = stock->cached;
1808 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001809 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001810 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001811 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001812 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1813 if (cpu == curcpu)
1814 drain_local_stock(&stock->work);
1815 else
1816 schedule_work_on(cpu, &stock->work);
1817 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001818 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001819 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001820 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001821 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001822}
1823
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01001824static int memcg_hotplug_cpu_dead(unsigned int cpu)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001825{
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001826 struct memcg_stock_pcp *stock;
1827
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001828 stock = &per_cpu(memcg_stock, cpu);
1829 drain_stock(stock);
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01001830 return 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001831}
1832
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001833static void reclaim_high(struct mem_cgroup *memcg,
1834 unsigned int nr_pages,
1835 gfp_t gfp_mask)
1836{
1837 do {
1838 if (page_counter_read(&memcg->memory) <= memcg->high)
1839 continue;
1840 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
1841 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1842 } while ((memcg = parent_mem_cgroup(memcg)));
1843}
1844
1845static void high_work_func(struct work_struct *work)
1846{
1847 struct mem_cgroup *memcg;
1848
1849 memcg = container_of(work, struct mem_cgroup, high_work);
1850 reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1851}
1852
Tejun Heob23afb92015-11-05 18:46:11 -08001853/*
1854 * Scheduled by try_charge() to be executed from the userland return path
1855 * and reclaims memory over the high limit.
1856 */
1857void mem_cgroup_handle_over_high(void)
1858{
1859 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001860 struct mem_cgroup *memcg;
Tejun Heob23afb92015-11-05 18:46:11 -08001861
1862 if (likely(!nr_pages))
1863 return;
1864
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001865 memcg = get_mem_cgroup_from_mm(current->mm);
1866 reclaim_high(memcg, nr_pages, GFP_KERNEL);
Tejun Heob23afb92015-11-05 18:46:11 -08001867 css_put(&memcg->css);
1868 current->memcg_nr_pages_over_high = 0;
1869}
1870
Johannes Weiner00501b52014-08-08 14:19:20 -07001871static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1872 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001873{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001874 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001875 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001876 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001877 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001878 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001879 bool may_swap = true;
1880 bool drained = false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001881
Johannes Weinerce00a962014-09-05 08:43:57 -04001882 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08001883 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001884retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001885 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08001886 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001887
Johannes Weiner7941d212016-01-14 15:21:23 -08001888 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08001889 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1890 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001891 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08001892 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001893 page_counter_uncharge(&memcg->memsw, batch);
1894 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001895 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001896 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001897 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001898 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001899
Johannes Weiner6539cc02014-08-06 16:05:42 -07001900 if (batch > nr_pages) {
1901 batch = nr_pages;
1902 goto retry;
1903 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001904
Johannes Weiner06b078f2014-08-06 16:05:44 -07001905 /*
1906 * Unlike in global OOM situations, memcg is not in a physical
1907 * memory shortage. Allow dying and OOM-killed tasks to
1908 * bypass the last charges so that they can exit quickly and
1909 * free their memory.
1910 */
1911 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1912 fatal_signal_pending(current) ||
1913 current->flags & PF_EXITING))
Tejun Heo10d53c72015-11-05 18:46:17 -08001914 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001915
Johannes Weiner89a28482016-10-27 17:46:56 -07001916 /*
1917 * Prevent unbounded recursion when reclaim operations need to
1918 * allocate memory. This might exceed the limits temporarily,
1919 * but we prefer facilitating memory reclaim and getting back
1920 * under the limit over triggering OOM kills in these cases.
1921 */
1922 if (unlikely(current->flags & PF_MEMALLOC))
1923 goto force;
1924
Johannes Weiner06b078f2014-08-06 16:05:44 -07001925 if (unlikely(task_in_memcg_oom(current)))
1926 goto nomem;
1927
Mel Gormand0164ad2015-11-06 16:28:21 -08001928 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001929 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001930
Johannes Weiner241994ed2015-02-11 15:26:06 -08001931 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
1932
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001933 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1934 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07001935
Johannes Weiner61e02c72014-08-06 16:08:16 -07001936 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07001937 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07001938
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001939 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001940 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001941 drained = true;
1942 goto retry;
1943 }
1944
Johannes Weiner28c34c22014-08-06 16:05:47 -07001945 if (gfp_mask & __GFP_NORETRY)
1946 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001947 /*
1948 * Even though the limit is exceeded at this point, reclaim
1949 * may have been able to free some pages. Retry the charge
1950 * before killing the task.
1951 *
1952 * Only for regular pages, though: huge pages are rather
1953 * unlikely to succeed so close to the limit, and we fall back
1954 * to regular pages anyway in case of failure.
1955 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07001956 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001957 goto retry;
1958 /*
1959 * At task move, charge accounts can be doubly counted. So, it's
1960 * better to wait until the end of task_move if something is going on.
1961 */
1962 if (mem_cgroup_wait_acct_move(mem_over_limit))
1963 goto retry;
1964
Johannes Weiner9b130612014-08-06 16:05:51 -07001965 if (nr_retries--)
1966 goto retry;
1967
Johannes Weiner06b078f2014-08-06 16:05:44 -07001968 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08001969 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001970
Johannes Weiner6539cc02014-08-06 16:05:42 -07001971 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08001972 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001973
Johannes Weiner241994ed2015-02-11 15:26:06 -08001974 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
1975
Jerome Marchand3608de02015-11-05 18:47:29 -08001976 mem_cgroup_oom(mem_over_limit, gfp_mask,
1977 get_order(nr_pages * PAGE_SIZE));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001978nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07001979 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07001980 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08001981force:
1982 /*
1983 * The allocation either can't fail or will lead to more memory
1984 * being freed very soon. Allow memory usage go over the limit
1985 * temporarily by force charging it.
1986 */
1987 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001988 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08001989 page_counter_charge(&memcg->memsw, nr_pages);
1990 css_get_many(&memcg->css, nr_pages);
1991
1992 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001993
1994done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001995 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07001996 if (batch > nr_pages)
1997 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08001998
Johannes Weiner241994ed2015-02-11 15:26:06 -08001999 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002000 * If the hierarchy is above the normal consumption range, schedule
2001 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002002 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002003 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2004 * not recorded as it most likely matches current's and won't
2005 * change in the meantime. As high limit is checked again before
2006 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994ed2015-02-11 15:26:06 -08002007 */
2008 do {
Tejun Heob23afb92015-11-05 18:46:11 -08002009 if (page_counter_read(&memcg->memory) > memcg->high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002010 /* Don't bother a random interrupted task */
2011 if (in_interrupt()) {
2012 schedule_work(&memcg->high_work);
2013 break;
2014 }
Vladimir Davydov9516a182015-12-11 13:40:24 -08002015 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002016 set_notify_resume(current);
2017 break;
2018 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08002019 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002020
2021 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002022}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002023
Johannes Weiner00501b52014-08-08 14:19:20 -07002024static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002025{
Johannes Weinerce00a962014-09-05 08:43:57 -04002026 if (mem_cgroup_is_root(memcg))
2027 return;
2028
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002029 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002030 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002031 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002032
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002033 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002034}
2035
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002036static void lock_page_lru(struct page *page, int *isolated)
2037{
2038 struct zone *zone = page_zone(page);
2039
Mel Gormana52633d2016-07-28 15:45:28 -07002040 spin_lock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002041 if (PageLRU(page)) {
2042 struct lruvec *lruvec;
2043
Mel Gorman599d0c92016-07-28 15:45:31 -07002044 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002045 ClearPageLRU(page);
2046 del_page_from_lru_list(page, lruvec, page_lru(page));
2047 *isolated = 1;
2048 } else
2049 *isolated = 0;
2050}
2051
2052static void unlock_page_lru(struct page *page, int isolated)
2053{
2054 struct zone *zone = page_zone(page);
2055
2056 if (isolated) {
2057 struct lruvec *lruvec;
2058
Mel Gorman599d0c92016-07-28 15:45:31 -07002059 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002060 VM_BUG_ON_PAGE(PageLRU(page), page);
2061 SetPageLRU(page);
2062 add_page_to_lru_list(page, lruvec, page_lru(page));
2063 }
Mel Gormana52633d2016-07-28 15:45:28 -07002064 spin_unlock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002065}
2066
Johannes Weiner00501b52014-08-08 14:19:20 -07002067static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002068 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002069{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002070 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002071
Johannes Weiner1306a852014-12-10 15:44:52 -08002072 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002073
2074 /*
2075 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2076 * may already be on some other mem_cgroup's LRU. Take care of it.
2077 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002078 if (lrucare)
2079 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002080
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002081 /*
2082 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002083 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002084 *
2085 * - the page is uncharged
2086 *
2087 * - the page is off-LRU
2088 *
2089 * - an anonymous fault has exclusive page access, except for
2090 * a locked page table
2091 *
2092 * - a page cache insertion, a swapin fault, or a migration
2093 * have the page locked
2094 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002095 page->mem_cgroup = memcg;
Hugh Dickins3be91272008-02-07 00:14:19 -08002096
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002097 if (lrucare)
2098 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002099}
2100
Johannes Weiner127424c2016-01-20 15:02:32 -08002101#ifndef CONFIG_SLOB
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002102static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002103{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002104 int id, size;
2105 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002106
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002107 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002108 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2109 if (id < 0)
2110 return id;
2111
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002112 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002113 return id;
2114
2115 /*
2116 * There's no space for the new id in memcg_caches arrays,
2117 * so we have to grow them.
2118 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002119 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002120
2121 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002122 if (size < MEMCG_CACHES_MIN_SIZE)
2123 size = MEMCG_CACHES_MIN_SIZE;
2124 else if (size > MEMCG_CACHES_MAX_SIZE)
2125 size = MEMCG_CACHES_MAX_SIZE;
2126
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002127 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002128 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002129 err = memcg_update_all_list_lrus(size);
2130 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002131 memcg_nr_cache_ids = size;
2132
2133 up_write(&memcg_cache_ids_sem);
2134
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002135 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002136 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002137 return err;
2138 }
2139 return id;
2140}
2141
2142static void memcg_free_cache_id(int id)
2143{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002144 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002145}
2146
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002147struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002148 struct mem_cgroup *memcg;
2149 struct kmem_cache *cachep;
2150 struct work_struct work;
2151};
2152
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002153static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002154{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002155 struct memcg_kmem_cache_create_work *cw =
2156 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002157 struct mem_cgroup *memcg = cw->memcg;
2158 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002159
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002160 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002161
Vladimir Davydov5722d092014-04-07 15:39:24 -07002162 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002163 kfree(cw);
2164}
2165
2166/*
2167 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002168 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002169static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2170 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002171{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002172 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002173
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002174 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002175 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002176 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002177
2178 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002179
2180 cw->memcg = memcg;
2181 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002182 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002183
Tejun Heo17cc4df2017-02-22 15:41:36 -08002184 queue_work(memcg_kmem_cache_wq, &cw->work);
Glauber Costad7f25f82012-12-18 14:22:40 -08002185}
2186
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002187static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2188 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002189{
2190 /*
2191 * We need to stop accounting when we kmalloc, because if the
2192 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002193 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002194 *
2195 * However, it is better to enclose the whole function. Depending on
2196 * the debugging options enabled, INIT_WORK(), for instance, can
2197 * trigger an allocation. This too, will make us recurse. Because at
2198 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2199 * the safest choice is to do it like this, wrapping the whole function.
2200 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002201 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002202 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002203 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002204}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002205
Vladimir Davydov45264772016-07-26 15:24:21 -07002206static inline bool memcg_kmem_bypass(void)
2207{
2208 if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
2209 return true;
2210 return false;
2211}
2212
2213/**
2214 * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2215 * @cachep: the original global kmem cache
2216 *
Glauber Costad7f25f82012-12-18 14:22:40 -08002217 * Return the kmem_cache we're supposed to use for a slab allocation.
2218 * We try to use the current memcg's version of the cache.
2219 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002220 * If the cache does not exist yet, if we are the first user of it, we
2221 * create it asynchronously in a workqueue and let the current allocation
2222 * go through with the original cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002223 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002224 * This function takes a reference to the cache it returns to assure it
2225 * won't get destroyed while we are working with it. Once the caller is
2226 * done with it, memcg_kmem_put_cache() must be called to release the
2227 * reference.
Glauber Costad7f25f82012-12-18 14:22:40 -08002228 */
Vladimir Davydov45264772016-07-26 15:24:21 -07002229struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002230{
2231 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002232 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002233 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002234
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002235 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002236
Vladimir Davydov45264772016-07-26 15:24:21 -07002237 if (memcg_kmem_bypass())
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002238 return cachep;
2239
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002240 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002241 return cachep;
2242
Vladimir Davydov8135be52014-12-12 16:56:38 -08002243 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002244 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002245 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002246 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002247
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002248 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002249 if (likely(memcg_cachep))
2250 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002251
2252 /*
2253 * If we are in a safe context (can wait, and not in interrupt
2254 * context), we could be be predictable and return right away.
2255 * This would guarantee that the allocation being performed
2256 * already belongs in the new cache.
2257 *
2258 * However, there are some clashes that can arrive from locking.
2259 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002260 * memcg_create_kmem_cache, this means no further allocation
2261 * could happen with the slab_mutex held. So it's better to
2262 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002263 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002264 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002265out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002266 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002267 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002268}
Glauber Costad7f25f82012-12-18 14:22:40 -08002269
Vladimir Davydov45264772016-07-26 15:24:21 -07002270/**
2271 * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2272 * @cachep: the cache returned by memcg_kmem_get_cache
2273 */
2274void memcg_kmem_put_cache(struct kmem_cache *cachep)
Vladimir Davydov8135be52014-12-12 16:56:38 -08002275{
2276 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002277 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002278}
2279
Vladimir Davydov45264772016-07-26 15:24:21 -07002280/**
2281 * memcg_kmem_charge: charge a kmem page
2282 * @page: page to charge
2283 * @gfp: reclaim mode
2284 * @order: allocation order
2285 * @memcg: memory cgroup to charge
2286 *
2287 * Returns 0 on success, an error code on failure.
2288 */
2289int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2290 struct mem_cgroup *memcg)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002291{
2292 unsigned int nr_pages = 1 << order;
2293 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002294 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002295
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002296 ret = try_charge(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08002297 if (ret)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002298 return ret;
Johannes Weiner52c29b02016-01-20 15:02:35 -08002299
2300 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2301 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2302 cancel_charge(memcg, nr_pages);
2303 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002304 }
2305
2306 page->mem_cgroup = memcg;
2307
2308 return 0;
2309}
2310
Vladimir Davydov45264772016-07-26 15:24:21 -07002311/**
2312 * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2313 * @page: page to charge
2314 * @gfp: reclaim mode
2315 * @order: allocation order
2316 *
2317 * Returns 0 on success, an error code on failure.
2318 */
2319int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002320{
2321 struct mem_cgroup *memcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002322 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002323
Vladimir Davydov45264772016-07-26 15:24:21 -07002324 if (memcg_kmem_bypass())
2325 return 0;
2326
Johannes Weinerdf381972014-04-07 15:37:43 -07002327 memcg = get_mem_cgroup_from_mm(current->mm);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002328 if (!mem_cgroup_is_root(memcg)) {
Vladimir Davydov45264772016-07-26 15:24:21 -07002329 ret = memcg_kmem_charge_memcg(page, gfp, order, memcg);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002330 if (!ret)
2331 __SetPageKmemcg(page);
2332 }
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002333 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002334 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002335}
Vladimir Davydov45264772016-07-26 15:24:21 -07002336/**
2337 * memcg_kmem_uncharge: uncharge a kmem page
2338 * @page: page to uncharge
2339 * @order: allocation order
2340 */
2341void memcg_kmem_uncharge(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002342{
Johannes Weiner1306a852014-12-10 15:44:52 -08002343 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002344 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002345
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002346 if (!memcg)
2347 return;
2348
Sasha Levin309381fea2014-01-23 15:52:54 -08002349 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002350
Johannes Weiner52c29b02016-01-20 15:02:35 -08002351 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2352 page_counter_uncharge(&memcg->kmem, nr_pages);
2353
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002354 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002355 if (do_memsw_account())
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002356 page_counter_uncharge(&memcg->memsw, nr_pages);
2357
Johannes Weiner1306a852014-12-10 15:44:52 -08002358 page->mem_cgroup = NULL;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002359
2360 /* slab pages do not have PageKmemcg flag set */
2361 if (PageKmemcg(page))
2362 __ClearPageKmemcg(page);
2363
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002364 css_put_many(&memcg->css, nr_pages);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002365}
Johannes Weiner127424c2016-01-20 15:02:32 -08002366#endif /* !CONFIG_SLOB */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002367
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002368#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2369
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002370/*
2371 * Because tail pages are not marked as "used", set it. We're under
Mel Gormana52633d2016-07-28 15:45:28 -07002372 * zone_lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002373 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002374void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002375{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002376 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002377
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002378 if (mem_cgroup_disabled())
2379 return;
David Rientjesb070e652013-05-07 16:18:09 -07002380
Johannes Weiner29833312014-12-10 15:44:02 -08002381 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002382 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002383
Johannes Weiner1306a852014-12-10 15:44:52 -08002384 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002385 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002386}
Hugh Dickins12d27102012-01-12 17:19:52 -08002387#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002388
Andrew Mortonc255a452012-07-31 16:43:02 -07002389#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002390static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2391 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002392{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002393 int val = (charge) ? 1 : -1;
2394 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002395}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002396
2397/**
2398 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2399 * @entry: swap entry to be moved
2400 * @from: mem_cgroup which the entry is moved from
2401 * @to: mem_cgroup which the entry is moved to
2402 *
2403 * It succeeds only when the swap_cgroup's record for this entry is the same
2404 * as the mem_cgroup's id of @from.
2405 *
2406 * Returns 0 on success, -EINVAL on failure.
2407 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002408 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002409 * both res and memsw, and called css_get().
2410 */
2411static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002412 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002413{
2414 unsigned short old_id, new_id;
2415
Li Zefan34c00c32013-09-23 16:56:01 +08002416 old_id = mem_cgroup_id(from);
2417 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002418
2419 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002420 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002421 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002422 return 0;
2423 }
2424 return -EINVAL;
2425}
2426#else
2427static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002428 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002429{
2430 return -EINVAL;
2431}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002432#endif
2433
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002434static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002435
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002436static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002437 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002438{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002439 unsigned long curusage;
2440 unsigned long oldusage;
2441 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002442 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002443 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002444
2445 /*
2446 * For keeping hierarchical_reclaim simple, how long we should retry
2447 * is depends on callers. We set our retry-count to be function
2448 * of # of children which we should visit in this loop.
2449 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002450 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2451 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002452
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002453 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002454
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002455 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002456 if (signal_pending(current)) {
2457 ret = -EINTR;
2458 break;
2459 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002460
2461 mutex_lock(&memcg_limit_mutex);
2462 if (limit > memcg->memsw.limit) {
2463 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002464 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002465 break;
2466 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002467 if (limit > memcg->memory.limit)
2468 enlarge = true;
2469 ret = page_counter_limit(&memcg->memory, limit);
2470 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002471
2472 if (!ret)
2473 break;
2474
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002475 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2476
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002477 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002478 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002479 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002480 retry_count--;
2481 else
2482 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002483 } while (retry_count);
2484
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002485 if (!ret && enlarge)
2486 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002487
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002488 return ret;
2489}
2490
Li Zefan338c8432009-06-17 16:27:15 -07002491static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002492 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002493{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002494 unsigned long curusage;
2495 unsigned long oldusage;
2496 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002497 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002498 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002499
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002500 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002501 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2502 mem_cgroup_count_children(memcg);
2503
2504 oldusage = page_counter_read(&memcg->memsw);
2505
2506 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002507 if (signal_pending(current)) {
2508 ret = -EINTR;
2509 break;
2510 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002511
2512 mutex_lock(&memcg_limit_mutex);
2513 if (limit < memcg->memory.limit) {
2514 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002515 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002516 break;
2517 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002518 if (limit > memcg->memsw.limit)
2519 enlarge = true;
2520 ret = page_counter_limit(&memcg->memsw, limit);
2521 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002522
2523 if (!ret)
2524 break;
2525
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002526 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2527
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002528 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002529 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002530 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002531 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002532 else
2533 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002534 } while (retry_count);
2535
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002536 if (!ret && enlarge)
2537 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002538
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002539 return ret;
2540}
2541
Mel Gormanef8f2322016-07-28 15:46:05 -07002542unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
Andrew Morton0608f432013-09-24 15:27:41 -07002543 gfp_t gfp_mask,
2544 unsigned long *total_scanned)
2545{
2546 unsigned long nr_reclaimed = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002547 struct mem_cgroup_per_node *mz, *next_mz = NULL;
Andrew Morton0608f432013-09-24 15:27:41 -07002548 unsigned long reclaimed;
2549 int loop = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002550 struct mem_cgroup_tree_per_node *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002551 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002552 unsigned long nr_scanned;
2553
2554 if (order > 0)
2555 return 0;
2556
Mel Gormanef8f2322016-07-28 15:46:05 -07002557 mctz = soft_limit_tree_node(pgdat->node_id);
Michal Hockod6507ff2016-08-02 14:02:37 -07002558
2559 /*
2560 * Do not even bother to check the largest node if the root
2561 * is empty. Do it lockless to prevent lock bouncing. Races
2562 * are acceptable as soft limit is best effort anyway.
2563 */
Laurent Dufourbfc72282017-03-09 16:17:06 -08002564 if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root))
Michal Hockod6507ff2016-08-02 14:02:37 -07002565 return 0;
2566
Andrew Morton0608f432013-09-24 15:27:41 -07002567 /*
2568 * This loop can run a while, specially if mem_cgroup's continuously
2569 * keep exceeding their soft limit and putting the system under
2570 * pressure
2571 */
2572 do {
2573 if (next_mz)
2574 mz = next_mz;
2575 else
2576 mz = mem_cgroup_largest_soft_limit_node(mctz);
2577 if (!mz)
2578 break;
2579
2580 nr_scanned = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002581 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07002582 gfp_mask, &nr_scanned);
2583 nr_reclaimed += reclaimed;
2584 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002585 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002586 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002587
2588 /*
2589 * If we failed to reclaim anything from this memory cgroup
2590 * it is time to move on to the next cgroup
2591 */
2592 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002593 if (!reclaimed)
2594 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2595
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002596 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002597 /*
2598 * One school of thought says that we should not add
2599 * back the node to the tree if reclaim returns 0.
2600 * But our reclaim could return 0, simply because due
2601 * to priority we are exposing a smaller subset of
2602 * memory to reclaim from. Consider this as a longer
2603 * term TODO.
2604 */
2605 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002606 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002607 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002608 css_put(&mz->memcg->css);
2609 loop++;
2610 /*
2611 * Could not reclaim anything and there are no more
2612 * mem cgroups to try or we seem to be looping without
2613 * reclaiming anything.
2614 */
2615 if (!nr_reclaimed &&
2616 (next_mz == NULL ||
2617 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2618 break;
2619 } while (!nr_reclaimed);
2620 if (next_mz)
2621 css_put(&next_mz->memcg->css);
2622 return nr_reclaimed;
2623}
2624
Tejun Heoea280e72014-05-16 13:22:48 -04002625/*
2626 * Test whether @memcg has children, dead or alive. Note that this
2627 * function doesn't care whether @memcg has use_hierarchy enabled and
2628 * returns %true if there are child csses according to the cgroup
2629 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2630 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002631static inline bool memcg_has_children(struct mem_cgroup *memcg)
2632{
Tejun Heoea280e72014-05-16 13:22:48 -04002633 bool ret;
2634
Tejun Heoea280e72014-05-16 13:22:48 -04002635 rcu_read_lock();
2636 ret = css_next_child(NULL, &memcg->css);
2637 rcu_read_unlock();
2638 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002639}
2640
2641/*
Greg Thelen51038172016-05-20 16:58:18 -07002642 * Reclaims as many pages from the given memcg as possible.
Michal Hockoc26251f2012-10-26 13:37:28 +02002643 *
2644 * Caller is responsible for holding css reference for memcg.
2645 */
2646static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2647{
2648 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002649
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002650 /* we call try-to-free pages for make this cgroup empty */
2651 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002652 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002653 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002654 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002655
Michal Hockoc26251f2012-10-26 13:37:28 +02002656 if (signal_pending(current))
2657 return -EINTR;
2658
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002659 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2660 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002661 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002662 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002663 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002664 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002665 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002666
2667 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002668
2669 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002670}
2671
Tejun Heo6770c642014-05-13 12:16:21 -04002672static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2673 char *buf, size_t nbytes,
2674 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002675{
Tejun Heo6770c642014-05-13 12:16:21 -04002676 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002677
Michal Hockod8423012012-10-26 13:37:29 +02002678 if (mem_cgroup_is_root(memcg))
2679 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002680 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002681}
2682
Tejun Heo182446d2013-08-08 20:11:24 -04002683static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2684 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002685{
Tejun Heo182446d2013-08-08 20:11:24 -04002686 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002687}
2688
Tejun Heo182446d2013-08-08 20:11:24 -04002689static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2690 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002691{
2692 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002693 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002694 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002695
Glauber Costa567fb432012-07-31 16:43:07 -07002696 if (memcg->use_hierarchy == val)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002697 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07002698
Balbir Singh18f59ea2009-01-07 18:08:07 -08002699 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002700 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002701 * in the child subtrees. If it is unset, then the change can
2702 * occur, provided the current cgroup has no children.
2703 *
2704 * For the root cgroup, parent_mem is NULL, we allow value to be
2705 * set if there are no children.
2706 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002707 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002708 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002709 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002710 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002711 else
2712 retval = -EBUSY;
2713 } else
2714 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002715
Balbir Singh18f59ea2009-01-07 18:08:07 -08002716 return retval;
2717}
2718
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002719static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
Johannes Weinerce00a962014-09-05 08:43:57 -04002720{
2721 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002722 int i;
Johannes Weinerce00a962014-09-05 08:43:57 -04002723
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002724 memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
Johannes Weinerce00a962014-09-05 08:43:57 -04002725
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002726 for_each_mem_cgroup_tree(iter, memcg) {
2727 for (i = 0; i < MEMCG_NR_STAT; i++)
2728 stat[i] += mem_cgroup_read_stat(iter, i);
2729 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002730}
2731
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002732static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
Johannes Weiner587d9f72016-01-20 15:03:19 -08002733{
2734 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002735 int i;
Johannes Weiner587d9f72016-01-20 15:03:19 -08002736
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002737 memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
Johannes Weiner587d9f72016-01-20 15:03:19 -08002738
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002739 for_each_mem_cgroup_tree(iter, memcg) {
2740 for (i = 0; i < MEMCG_NR_EVENTS; i++)
2741 events[i] += mem_cgroup_read_events(iter, i);
2742 }
Johannes Weiner587d9f72016-01-20 15:03:19 -08002743}
2744
Andrew Morton6f646152015-11-06 16:28:58 -08002745static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04002746{
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002747 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002748
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002749 if (mem_cgroup_is_root(memcg)) {
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002750 struct mem_cgroup *iter;
2751
2752 for_each_mem_cgroup_tree(iter, memcg) {
2753 val += mem_cgroup_read_stat(iter,
2754 MEM_CGROUP_STAT_CACHE);
2755 val += mem_cgroup_read_stat(iter,
2756 MEM_CGROUP_STAT_RSS);
2757 if (swap)
2758 val += mem_cgroup_read_stat(iter,
2759 MEM_CGROUP_STAT_SWAP);
2760 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002761 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002762 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002763 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002764 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002765 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002766 }
Michal Hockoc12176d2015-11-05 18:50:29 -08002767 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002768}
2769
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002770enum {
2771 RES_USAGE,
2772 RES_LIMIT,
2773 RES_MAX_USAGE,
2774 RES_FAILCNT,
2775 RES_SOFT_LIMIT,
2776};
Johannes Weinerce00a962014-09-05 08:43:57 -04002777
Tejun Heo791badb2013-12-05 12:28:02 -05002778static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002779 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002780{
Tejun Heo182446d2013-08-08 20:11:24 -04002781 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002782 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002783
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002784 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002785 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002786 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002787 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002788 case _MEMSWAP:
2789 counter = &memcg->memsw;
2790 break;
2791 case _KMEM:
2792 counter = &memcg->kmem;
2793 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002794 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08002795 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002796 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002797 default:
2798 BUG();
2799 }
2800
2801 switch (MEMFILE_ATTR(cft->private)) {
2802 case RES_USAGE:
2803 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08002804 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002805 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08002806 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002807 return (u64)page_counter_read(counter) * PAGE_SIZE;
2808 case RES_LIMIT:
2809 return (u64)counter->limit * PAGE_SIZE;
2810 case RES_MAX_USAGE:
2811 return (u64)counter->watermark * PAGE_SIZE;
2812 case RES_FAILCNT:
2813 return counter->failcnt;
2814 case RES_SOFT_LIMIT:
2815 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002816 default:
2817 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002818 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002819}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002820
Johannes Weiner127424c2016-01-20 15:02:32 -08002821#ifndef CONFIG_SLOB
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002822static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002823{
Vladimir Davydovd6441632014-01-23 15:53:09 -08002824 int memcg_id;
2825
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002826 if (cgroup_memory_nokmem)
2827 return 0;
2828
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002829 BUG_ON(memcg->kmemcg_id >= 0);
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002830 BUG_ON(memcg->kmem_state);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002831
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002832 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002833 if (memcg_id < 0)
2834 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002835
Johannes Weineref129472016-01-14 15:21:34 -08002836 static_branch_inc(&memcg_kmem_enabled_key);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002837 /*
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002838 * A memory cgroup is considered kmem-online as soon as it gets
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002839 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002840 * guarantee no one starts accounting before all call sites are
2841 * patched.
2842 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002843 memcg->kmemcg_id = memcg_id;
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002844 memcg->kmem_state = KMEM_ONLINE;
Tejun Heobc2791f2017-02-22 15:41:21 -08002845 INIT_LIST_HEAD(&memcg->kmem_caches);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002846
2847 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002848}
2849
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002850static void memcg_offline_kmem(struct mem_cgroup *memcg)
2851{
2852 struct cgroup_subsys_state *css;
2853 struct mem_cgroup *parent, *child;
2854 int kmemcg_id;
2855
2856 if (memcg->kmem_state != KMEM_ONLINE)
2857 return;
2858 /*
2859 * Clear the online state before clearing memcg_caches array
2860 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2861 * guarantees that no cache will be created for this cgroup
2862 * after we are done (see memcg_create_kmem_cache()).
2863 */
2864 memcg->kmem_state = KMEM_ALLOCATED;
2865
2866 memcg_deactivate_kmem_caches(memcg);
2867
2868 kmemcg_id = memcg->kmemcg_id;
2869 BUG_ON(kmemcg_id < 0);
2870
2871 parent = parent_mem_cgroup(memcg);
2872 if (!parent)
2873 parent = root_mem_cgroup;
2874
2875 /*
2876 * Change kmemcg_id of this cgroup and all its descendants to the
2877 * parent's id, and then move all entries from this cgroup's list_lrus
2878 * to ones of the parent. After we have finished, all list_lrus
2879 * corresponding to this cgroup are guaranteed to remain empty. The
2880 * ordering is imposed by list_lru_node->lock taken by
2881 * memcg_drain_all_list_lrus().
2882 */
Tejun Heo3a06bb72016-06-03 14:55:44 -07002883 rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002884 css_for_each_descendant_pre(css, &memcg->css) {
2885 child = mem_cgroup_from_css(css);
2886 BUG_ON(child->kmemcg_id != kmemcg_id);
2887 child->kmemcg_id = parent->kmemcg_id;
2888 if (!memcg->use_hierarchy)
2889 break;
2890 }
Tejun Heo3a06bb72016-06-03 14:55:44 -07002891 rcu_read_unlock();
2892
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002893 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2894
2895 memcg_free_cache_id(kmemcg_id);
2896}
2897
2898static void memcg_free_kmem(struct mem_cgroup *memcg)
2899{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002900 /* css_alloc() failed, offlining didn't happen */
2901 if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2902 memcg_offline_kmem(memcg);
2903
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002904 if (memcg->kmem_state == KMEM_ALLOCATED) {
2905 memcg_destroy_kmem_caches(memcg);
2906 static_branch_dec(&memcg_kmem_enabled_key);
2907 WARN_ON(page_counter_read(&memcg->kmem));
2908 }
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002909}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002910#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002911static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08002912{
2913 return 0;
2914}
2915static void memcg_offline_kmem(struct mem_cgroup *memcg)
2916{
2917}
2918static void memcg_free_kmem(struct mem_cgroup *memcg)
2919{
2920}
2921#endif /* !CONFIG_SLOB */
2922
Johannes Weiner127424c2016-01-20 15:02:32 -08002923static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2924 unsigned long limit)
2925{
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002926 int ret;
Johannes Weiner127424c2016-01-20 15:02:32 -08002927
2928 mutex_lock(&memcg_limit_mutex);
Johannes Weiner127424c2016-01-20 15:02:32 -08002929 ret = page_counter_limit(&memcg->kmem, limit);
Johannes Weiner127424c2016-01-20 15:02:32 -08002930 mutex_unlock(&memcg_limit_mutex);
2931 return ret;
2932}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002933
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002934static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2935{
2936 int ret;
2937
2938 mutex_lock(&memcg_limit_mutex);
2939
Johannes Weiner0db15292016-01-20 15:02:50 -08002940 ret = page_counter_limit(&memcg->tcpmem, limit);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002941 if (ret)
2942 goto out;
2943
Johannes Weiner0db15292016-01-20 15:02:50 -08002944 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002945 /*
2946 * The active flag needs to be written after the static_key
2947 * update. This is what guarantees that the socket activation
Johannes Weiner2d758072016-10-07 17:00:58 -07002948 * function is the last one to run. See mem_cgroup_sk_alloc()
2949 * for details, and note that we don't mark any socket as
2950 * belonging to this memcg until that flag is up.
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002951 *
2952 * We need to do this, because static_keys will span multiple
2953 * sites, but we can't control their order. If we mark a socket
2954 * as accounted, but the accounting functions are not patched in
2955 * yet, we'll lose accounting.
2956 *
Johannes Weiner2d758072016-10-07 17:00:58 -07002957 * We never race with the readers in mem_cgroup_sk_alloc(),
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002958 * because when this value change, the code to process it is not
2959 * patched in yet.
2960 */
2961 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08002962 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002963 }
2964out:
2965 mutex_unlock(&memcg_limit_mutex);
2966 return ret;
2967}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002968
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002969/*
2970 * The user of this function is...
2971 * RES_LIMIT.
2972 */
Tejun Heo451af502014-05-13 12:16:21 -04002973static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2974 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002975{
Tejun Heo451af502014-05-13 12:16:21 -04002976 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002977 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002978 int ret;
2979
Tejun Heo451af502014-05-13 12:16:21 -04002980 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08002981 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002982 if (ret)
2983 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07002984
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002985 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002986 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07002987 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2988 ret = -EINVAL;
2989 break;
2990 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002991 switch (MEMFILE_TYPE(of_cft(of)->private)) {
2992 case _MEM:
2993 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002994 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002995 case _MEMSWAP:
2996 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
2997 break;
2998 case _KMEM:
2999 ret = memcg_update_kmem_limit(memcg, nr_pages);
3000 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003001 case _TCP:
3002 ret = memcg_update_tcp_limit(memcg, nr_pages);
3003 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003004 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003005 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003006 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003007 memcg->soft_limit = nr_pages;
3008 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003009 break;
3010 }
Tejun Heo451af502014-05-13 12:16:21 -04003011 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003012}
3013
Tejun Heo6770c642014-05-13 12:16:21 -04003014static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3015 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003016{
Tejun Heo6770c642014-05-13 12:16:21 -04003017 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003018 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003019
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003020 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3021 case _MEM:
3022 counter = &memcg->memory;
3023 break;
3024 case _MEMSWAP:
3025 counter = &memcg->memsw;
3026 break;
3027 case _KMEM:
3028 counter = &memcg->kmem;
3029 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003030 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003031 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003032 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003033 default:
3034 BUG();
3035 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003036
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003037 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003038 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003039 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003040 break;
3041 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003042 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003043 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003044 default:
3045 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003046 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003047
Tejun Heo6770c642014-05-13 12:16:21 -04003048 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003049}
3050
Tejun Heo182446d2013-08-08 20:11:24 -04003051static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003052 struct cftype *cft)
3053{
Tejun Heo182446d2013-08-08 20:11:24 -04003054 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003055}
3056
Daisuke Nishimura02491442010-03-10 15:22:17 -08003057#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003058static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003059 struct cftype *cft, u64 val)
3060{
Tejun Heo182446d2013-08-08 20:11:24 -04003061 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003062
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003063 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003064 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003065
Glauber Costaee5e8472013-02-22 16:34:50 -08003066 /*
3067 * No kind of locking is needed in here, because ->can_attach() will
3068 * check this value once in the beginning of the process, and then carry
3069 * on with stale data. This means that changes to this value will only
3070 * affect task migrations starting after the change.
3071 */
3072 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003073 return 0;
3074}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003075#else
Tejun Heo182446d2013-08-08 20:11:24 -04003076static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003077 struct cftype *cft, u64 val)
3078{
3079 return -ENOSYS;
3080}
3081#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003082
Ying Han406eb0c2011-05-26 16:25:37 -07003083#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003084static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003085{
Greg Thelen25485de2013-11-12 15:07:40 -08003086 struct numa_stat {
3087 const char *name;
3088 unsigned int lru_mask;
3089 };
3090
3091 static const struct numa_stat stats[] = {
3092 { "total", LRU_ALL },
3093 { "file", LRU_ALL_FILE },
3094 { "anon", LRU_ALL_ANON },
3095 { "unevictable", BIT(LRU_UNEVICTABLE) },
3096 };
3097 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003098 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003099 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003100 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003101
Greg Thelen25485de2013-11-12 15:07:40 -08003102 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3103 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3104 seq_printf(m, "%s=%lu", stat->name, nr);
3105 for_each_node_state(nid, N_MEMORY) {
3106 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3107 stat->lru_mask);
3108 seq_printf(m, " N%d=%lu", nid, nr);
3109 }
3110 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003111 }
Ying Han406eb0c2011-05-26 16:25:37 -07003112
Ying Han071aee12013-11-12 15:07:41 -08003113 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3114 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003115
Ying Han071aee12013-11-12 15:07:41 -08003116 nr = 0;
3117 for_each_mem_cgroup_tree(iter, memcg)
3118 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3119 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3120 for_each_node_state(nid, N_MEMORY) {
3121 nr = 0;
3122 for_each_mem_cgroup_tree(iter, memcg)
3123 nr += mem_cgroup_node_nr_lru_pages(
3124 iter, nid, stat->lru_mask);
3125 seq_printf(m, " N%d=%lu", nid, nr);
3126 }
3127 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003128 }
Ying Han406eb0c2011-05-26 16:25:37 -07003129
Ying Han406eb0c2011-05-26 16:25:37 -07003130 return 0;
3131}
3132#endif /* CONFIG_NUMA */
3133
Tejun Heo2da8ca82013-12-05 12:28:04 -05003134static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003135{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003136 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003137 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003138 struct mem_cgroup *mi;
3139 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003140
Greg Thelen0ca44b12015-02-11 15:25:58 -08003141 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3142 MEM_CGROUP_STAT_NSTATS);
3143 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3144 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003145 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3146
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003147 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Johannes Weiner7941d212016-01-14 15:21:23 -08003148 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003149 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07003150 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003151 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003152 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003153
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003154 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3155 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3156 mem_cgroup_read_events(memcg, i));
3157
3158 for (i = 0; i < NR_LRU_LISTS; i++)
3159 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3160 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3161
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003162 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003163 memory = memsw = PAGE_COUNTER_MAX;
3164 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3165 memory = min(memory, mi->memory.limit);
3166 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003167 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003168 seq_printf(m, "hierarchical_memory_limit %llu\n",
3169 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08003170 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003171 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3172 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003173
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003174 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003175 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003176
Johannes Weiner7941d212016-01-14 15:21:23 -08003177 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003178 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003179 for_each_mem_cgroup_tree(mi, memcg)
3180 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
Greg Thelen484ebb32015-10-01 15:37:05 -07003181 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003182 }
3183
3184 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3185 unsigned long long val = 0;
3186
3187 for_each_mem_cgroup_tree(mi, memcg)
3188 val += mem_cgroup_read_events(mi, i);
3189 seq_printf(m, "total_%s %llu\n",
3190 mem_cgroup_events_names[i], val);
3191 }
3192
3193 for (i = 0; i < NR_LRU_LISTS; i++) {
3194 unsigned long long val = 0;
3195
3196 for_each_mem_cgroup_tree(mi, memcg)
3197 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3198 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003199 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003200
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003201#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003202 {
Mel Gormanef8f2322016-07-28 15:46:05 -07003203 pg_data_t *pgdat;
3204 struct mem_cgroup_per_node *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003205 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003206 unsigned long recent_rotated[2] = {0, 0};
3207 unsigned long recent_scanned[2] = {0, 0};
3208
Mel Gormanef8f2322016-07-28 15:46:05 -07003209 for_each_online_pgdat(pgdat) {
3210 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
3211 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003212
Mel Gormanef8f2322016-07-28 15:46:05 -07003213 recent_rotated[0] += rstat->recent_rotated[0];
3214 recent_rotated[1] += rstat->recent_rotated[1];
3215 recent_scanned[0] += rstat->recent_scanned[0];
3216 recent_scanned[1] += rstat->recent_scanned[1];
3217 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003218 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3219 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3220 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3221 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003222 }
3223#endif
3224
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003225 return 0;
3226}
3227
Tejun Heo182446d2013-08-08 20:11:24 -04003228static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3229 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003230{
Tejun Heo182446d2013-08-08 20:11:24 -04003231 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003232
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003233 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003234}
3235
Tejun Heo182446d2013-08-08 20:11:24 -04003236static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3237 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003238{
Tejun Heo182446d2013-08-08 20:11:24 -04003239 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003240
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003241 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003242 return -EINVAL;
3243
Linus Torvalds14208b02014-06-09 15:03:33 -07003244 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003245 memcg->swappiness = val;
3246 else
3247 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003248
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003249 return 0;
3250}
3251
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003252static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3253{
3254 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003255 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003256 int i;
3257
3258 rcu_read_lock();
3259 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003260 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003261 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003262 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003263
3264 if (!t)
3265 goto unlock;
3266
Johannes Weinerce00a962014-09-05 08:43:57 -04003267 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003268
3269 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003270 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003271 * If it's not true, a threshold was crossed after last
3272 * call of __mem_cgroup_threshold().
3273 */
Phil Carmody5407a562010-05-26 14:42:42 -07003274 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003275
3276 /*
3277 * Iterate backward over array of thresholds starting from
3278 * current_threshold and check if a threshold is crossed.
3279 * If none of thresholds below usage is crossed, we read
3280 * only one element of the array here.
3281 */
3282 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3283 eventfd_signal(t->entries[i].eventfd, 1);
3284
3285 /* i = current_threshold + 1 */
3286 i++;
3287
3288 /*
3289 * Iterate forward over array of thresholds starting from
3290 * current_threshold+1 and check if a threshold is crossed.
3291 * If none of thresholds above usage is crossed, we read
3292 * only one element of the array here.
3293 */
3294 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3295 eventfd_signal(t->entries[i].eventfd, 1);
3296
3297 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003298 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003299unlock:
3300 rcu_read_unlock();
3301}
3302
3303static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3304{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003305 while (memcg) {
3306 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08003307 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003308 __mem_cgroup_threshold(memcg, true);
3309
3310 memcg = parent_mem_cgroup(memcg);
3311 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003312}
3313
3314static int compare_thresholds(const void *a, const void *b)
3315{
3316 const struct mem_cgroup_threshold *_a = a;
3317 const struct mem_cgroup_threshold *_b = b;
3318
Greg Thelen2bff24a2013-09-11 14:23:08 -07003319 if (_a->threshold > _b->threshold)
3320 return 1;
3321
3322 if (_a->threshold < _b->threshold)
3323 return -1;
3324
3325 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003326}
3327
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003328static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003329{
3330 struct mem_cgroup_eventfd_list *ev;
3331
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003332 spin_lock(&memcg_oom_lock);
3333
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003334 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003335 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003336
3337 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003338 return 0;
3339}
3340
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003341static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003342{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003343 struct mem_cgroup *iter;
3344
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003345 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003346 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003347}
3348
Tejun Heo59b6f872013-11-22 18:20:43 -05003349static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003350 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003351{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003352 struct mem_cgroup_thresholds *thresholds;
3353 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003354 unsigned long threshold;
3355 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003356 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003357
Johannes Weiner650c5e52015-02-11 15:26:03 -08003358 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003359 if (ret)
3360 return ret;
3361
3362 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003363
Johannes Weiner05b84302014-08-06 16:05:59 -07003364 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003365 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003366 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003367 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003368 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003369 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003370 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003371 BUG();
3372
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003373 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003374 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003375 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3376
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003377 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003378
3379 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003380 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003381 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003382 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003383 ret = -ENOMEM;
3384 goto unlock;
3385 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003386 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003387
3388 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003389 if (thresholds->primary) {
3390 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003391 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003392 }
3393
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003394 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003395 new->entries[size - 1].eventfd = eventfd;
3396 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003397
3398 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003399 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003400 compare_thresholds, NULL);
3401
3402 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003403 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003404 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003405 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003406 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003407 * new->current_threshold will not be used until
3408 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003409 * it here.
3410 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003411 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003412 } else
3413 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003414 }
3415
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003416 /* Free old spare buffer and save old primary buffer as spare */
3417 kfree(thresholds->spare);
3418 thresholds->spare = thresholds->primary;
3419
3420 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003421
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003422 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003423 synchronize_rcu();
3424
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003425unlock:
3426 mutex_unlock(&memcg->thresholds_lock);
3427
3428 return ret;
3429}
3430
Tejun Heo59b6f872013-11-22 18:20:43 -05003431static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003432 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003433{
Tejun Heo59b6f872013-11-22 18:20:43 -05003434 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003435}
3436
Tejun Heo59b6f872013-11-22 18:20:43 -05003437static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003438 struct eventfd_ctx *eventfd, const char *args)
3439{
Tejun Heo59b6f872013-11-22 18:20:43 -05003440 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003441}
3442
Tejun Heo59b6f872013-11-22 18:20:43 -05003443static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003444 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003445{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003446 struct mem_cgroup_thresholds *thresholds;
3447 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003448 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003449 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003450
3451 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003452
3453 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003454 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003455 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003456 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003457 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003458 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003459 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003460 BUG();
3461
Anton Vorontsov371528c2012-02-24 05:14:46 +04003462 if (!thresholds->primary)
3463 goto unlock;
3464
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003465 /* Check if a threshold crossed before removing */
3466 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3467
3468 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003469 size = 0;
3470 for (i = 0; i < thresholds->primary->size; i++) {
3471 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003472 size++;
3473 }
3474
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003475 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003476
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003477 /* Set thresholds array to NULL if we don't have thresholds */
3478 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003479 kfree(new);
3480 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003481 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003482 }
3483
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003484 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003485
3486 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003487 new->current_threshold = -1;
3488 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3489 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003490 continue;
3491
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003492 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003493 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003494 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003495 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003496 * until rcu_assign_pointer(), so it's safe to increment
3497 * it here.
3498 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003499 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003500 }
3501 j++;
3502 }
3503
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003504swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003505 /* Swap primary and spare array */
3506 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003507
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003508 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003509
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003510 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003511 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08003512
3513 /* If all events are unregistered, free the spare array */
3514 if (!new) {
3515 kfree(thresholds->spare);
3516 thresholds->spare = NULL;
3517 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04003518unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003519 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003520}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003521
Tejun Heo59b6f872013-11-22 18:20:43 -05003522static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003523 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003524{
Tejun Heo59b6f872013-11-22 18:20:43 -05003525 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003526}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003527
Tejun Heo59b6f872013-11-22 18:20:43 -05003528static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003529 struct eventfd_ctx *eventfd)
3530{
Tejun Heo59b6f872013-11-22 18:20:43 -05003531 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003532}
3533
Tejun Heo59b6f872013-11-22 18:20:43 -05003534static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003535 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003536{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003537 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003538
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003539 event = kmalloc(sizeof(*event), GFP_KERNEL);
3540 if (!event)
3541 return -ENOMEM;
3542
Michal Hocko1af8efe2011-07-26 16:08:24 -07003543 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003544
3545 event->eventfd = eventfd;
3546 list_add(&event->list, &memcg->oom_notify);
3547
3548 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07003549 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003550 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003551 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003552
3553 return 0;
3554}
3555
Tejun Heo59b6f872013-11-22 18:20:43 -05003556static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003557 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003558{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003559 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003560
Michal Hocko1af8efe2011-07-26 16:08:24 -07003561 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003562
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003563 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003564 if (ev->eventfd == eventfd) {
3565 list_del(&ev->list);
3566 kfree(ev);
3567 }
3568 }
3569
Michal Hocko1af8efe2011-07-26 16:08:24 -07003570 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003571}
3572
Tejun Heo2da8ca82013-12-05 12:28:04 -05003573static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003574{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003575 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003576
Tejun Heo791badb2013-12-05 12:28:02 -05003577 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07003578 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003579 return 0;
3580}
3581
Tejun Heo182446d2013-08-08 20:11:24 -04003582static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003583 struct cftype *cft, u64 val)
3584{
Tejun Heo182446d2013-08-08 20:11:24 -04003585 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003586
3587 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003588 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003589 return -EINVAL;
3590
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003591 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003592 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003593 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003594
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003595 return 0;
3596}
3597
Tejun Heo52ebea72015-05-22 17:13:37 -04003598#ifdef CONFIG_CGROUP_WRITEBACK
3599
3600struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3601{
3602 return &memcg->cgwb_list;
3603}
3604
Tejun Heo841710a2015-05-22 18:23:33 -04003605static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3606{
3607 return wb_domain_init(&memcg->cgwb_domain, gfp);
3608}
3609
3610static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3611{
3612 wb_domain_exit(&memcg->cgwb_domain);
3613}
3614
Tejun Heo2529bb32015-05-22 18:23:34 -04003615static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3616{
3617 wb_domain_size_changed(&memcg->cgwb_domain);
3618}
3619
Tejun Heo841710a2015-05-22 18:23:33 -04003620struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3621{
3622 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3623
3624 if (!memcg->css.parent)
3625 return NULL;
3626
3627 return &memcg->cgwb_domain;
3628}
3629
Tejun Heoc2aa7232015-05-22 18:23:35 -04003630/**
3631 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3632 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003633 * @pfilepages: out parameter for number of file pages
3634 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04003635 * @pdirty: out parameter for number of dirty pages
3636 * @pwriteback: out parameter for number of pages under writeback
3637 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003638 * Determine the numbers of file, headroom, dirty, and writeback pages in
3639 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3640 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003641 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003642 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3643 * headroom is calculated as the lowest headroom of itself and the
3644 * ancestors. Note that this doesn't consider the actual amount of
3645 * available memory in the system. The caller should further cap
3646 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003647 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003648void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3649 unsigned long *pheadroom, unsigned long *pdirty,
3650 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04003651{
3652 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3653 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003654
3655 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3656
3657 /* this should eventually include NR_UNSTABLE_NFS */
3658 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003659 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3660 (1 << LRU_ACTIVE_FILE));
3661 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003662
Tejun Heoc2aa7232015-05-22 18:23:35 -04003663 while ((parent = parent_mem_cgroup(memcg))) {
3664 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3665 unsigned long used = page_counter_read(&memcg->memory);
3666
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003667 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04003668 memcg = parent;
3669 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04003670}
3671
Tejun Heo841710a2015-05-22 18:23:33 -04003672#else /* CONFIG_CGROUP_WRITEBACK */
3673
3674static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3675{
3676 return 0;
3677}
3678
3679static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3680{
3681}
3682
Tejun Heo2529bb32015-05-22 18:23:34 -04003683static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3684{
3685}
3686
Tejun Heo52ebea72015-05-22 17:13:37 -04003687#endif /* CONFIG_CGROUP_WRITEBACK */
3688
Tejun Heo79bd9812013-11-22 18:20:42 -05003689/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003690 * DO NOT USE IN NEW FILES.
3691 *
3692 * "cgroup.event_control" implementation.
3693 *
3694 * This is way over-engineered. It tries to support fully configurable
3695 * events for each user. Such level of flexibility is completely
3696 * unnecessary especially in the light of the planned unified hierarchy.
3697 *
3698 * Please deprecate this and replace with something simpler if at all
3699 * possible.
3700 */
3701
3702/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003703 * Unregister event and free resources.
3704 *
3705 * Gets called from workqueue.
3706 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003707static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003708{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003709 struct mem_cgroup_event *event =
3710 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003711 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003712
3713 remove_wait_queue(event->wqh, &event->wait);
3714
Tejun Heo59b6f872013-11-22 18:20:43 -05003715 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003716
3717 /* Notify userspace the event is going away. */
3718 eventfd_signal(event->eventfd, 1);
3719
3720 eventfd_ctx_put(event->eventfd);
3721 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003722 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003723}
3724
3725/*
3726 * Gets called on POLLHUP on eventfd when user closes it.
3727 *
3728 * Called with wqh->lock held and interrupts disabled.
3729 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003730static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3731 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003732{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003733 struct mem_cgroup_event *event =
3734 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003735 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003736 unsigned long flags = (unsigned long)key;
3737
3738 if (flags & POLLHUP) {
3739 /*
3740 * If the event has been detached at cgroup removal, we
3741 * can simply return knowing the other side will cleanup
3742 * for us.
3743 *
3744 * We can't race against event freeing since the other
3745 * side will require wqh->lock via remove_wait_queue(),
3746 * which we hold.
3747 */
Tejun Heofba94802013-11-22 18:20:43 -05003748 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003749 if (!list_empty(&event->list)) {
3750 list_del_init(&event->list);
3751 /*
3752 * We are in atomic context, but cgroup_event_remove()
3753 * may sleep, so we have to call it in workqueue.
3754 */
3755 schedule_work(&event->remove);
3756 }
Tejun Heofba94802013-11-22 18:20:43 -05003757 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003758 }
3759
3760 return 0;
3761}
3762
Tejun Heo3bc942f2013-11-22 18:20:44 -05003763static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003764 wait_queue_head_t *wqh, poll_table *pt)
3765{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003766 struct mem_cgroup_event *event =
3767 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003768
3769 event->wqh = wqh;
3770 add_wait_queue(wqh, &event->wait);
3771}
3772
3773/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003774 * DO NOT USE IN NEW FILES.
3775 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003776 * Parse input and register new cgroup event handler.
3777 *
3778 * Input must be in format '<event_fd> <control_fd> <args>'.
3779 * Interpretation of args is defined by control file implementation.
3780 */
Tejun Heo451af502014-05-13 12:16:21 -04003781static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3782 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003783{
Tejun Heo451af502014-05-13 12:16:21 -04003784 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003785 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003786 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003787 struct cgroup_subsys_state *cfile_css;
3788 unsigned int efd, cfd;
3789 struct fd efile;
3790 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003791 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003792 char *endp;
3793 int ret;
3794
Tejun Heo451af502014-05-13 12:16:21 -04003795 buf = strstrip(buf);
3796
3797 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003798 if (*endp != ' ')
3799 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003800 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003801
Tejun Heo451af502014-05-13 12:16:21 -04003802 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003803 if ((*endp != ' ') && (*endp != '\0'))
3804 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003805 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003806
3807 event = kzalloc(sizeof(*event), GFP_KERNEL);
3808 if (!event)
3809 return -ENOMEM;
3810
Tejun Heo59b6f872013-11-22 18:20:43 -05003811 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003812 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003813 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3814 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3815 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003816
3817 efile = fdget(efd);
3818 if (!efile.file) {
3819 ret = -EBADF;
3820 goto out_kfree;
3821 }
3822
3823 event->eventfd = eventfd_ctx_fileget(efile.file);
3824 if (IS_ERR(event->eventfd)) {
3825 ret = PTR_ERR(event->eventfd);
3826 goto out_put_efile;
3827 }
3828
3829 cfile = fdget(cfd);
3830 if (!cfile.file) {
3831 ret = -EBADF;
3832 goto out_put_eventfd;
3833 }
3834
3835 /* the process need read permission on control file */
3836 /* AV: shouldn't we check that it's been opened for read instead? */
3837 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3838 if (ret < 0)
3839 goto out_put_cfile;
3840
Tejun Heo79bd9812013-11-22 18:20:42 -05003841 /*
Tejun Heofba94802013-11-22 18:20:43 -05003842 * Determine the event callbacks and set them in @event. This used
3843 * to be done via struct cftype but cgroup core no longer knows
3844 * about these events. The following is crude but the whole thing
3845 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003846 *
3847 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003848 */
Al Virob5830432014-10-31 01:22:04 -04003849 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003850
3851 if (!strcmp(name, "memory.usage_in_bytes")) {
3852 event->register_event = mem_cgroup_usage_register_event;
3853 event->unregister_event = mem_cgroup_usage_unregister_event;
3854 } else if (!strcmp(name, "memory.oom_control")) {
3855 event->register_event = mem_cgroup_oom_register_event;
3856 event->unregister_event = mem_cgroup_oom_unregister_event;
3857 } else if (!strcmp(name, "memory.pressure_level")) {
3858 event->register_event = vmpressure_register_event;
3859 event->unregister_event = vmpressure_unregister_event;
3860 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003861 event->register_event = memsw_cgroup_usage_register_event;
3862 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003863 } else {
3864 ret = -EINVAL;
3865 goto out_put_cfile;
3866 }
3867
3868 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003869 * Verify @cfile should belong to @css. Also, remaining events are
3870 * automatically removed on cgroup destruction but the removal is
3871 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003872 */
Al Virob5830432014-10-31 01:22:04 -04003873 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003874 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003875 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003876 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003877 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003878 if (cfile_css != css) {
3879 css_put(cfile_css);
3880 goto out_put_cfile;
3881 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003882
Tejun Heo451af502014-05-13 12:16:21 -04003883 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003884 if (ret)
3885 goto out_put_css;
3886
3887 efile.file->f_op->poll(efile.file, &event->pt);
3888
Tejun Heofba94802013-11-22 18:20:43 -05003889 spin_lock(&memcg->event_list_lock);
3890 list_add(&event->list, &memcg->event_list);
3891 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003892
3893 fdput(cfile);
3894 fdput(efile);
3895
Tejun Heo451af502014-05-13 12:16:21 -04003896 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05003897
3898out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05003899 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003900out_put_cfile:
3901 fdput(cfile);
3902out_put_eventfd:
3903 eventfd_ctx_put(event->eventfd);
3904out_put_efile:
3905 fdput(efile);
3906out_kfree:
3907 kfree(event);
3908
3909 return ret;
3910}
3911
Johannes Weiner241994ed2015-02-11 15:26:06 -08003912static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003913 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003914 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003915 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003916 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003917 },
3918 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003919 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003920 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04003921 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003922 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003923 },
3924 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003925 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003926 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003927 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003928 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003929 },
3930 {
Balbir Singh296c81d2009-09-23 15:56:36 -07003931 .name = "soft_limit_in_bytes",
3932 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003933 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003934 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07003935 },
3936 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003937 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003938 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04003939 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003940 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003941 },
Balbir Singh8697d332008-02-07 00:13:59 -08003942 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003943 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003944 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003945 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003946 {
3947 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04003948 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003949 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08003950 {
3951 .name = "use_hierarchy",
3952 .write_u64 = mem_cgroup_hierarchy_write,
3953 .read_u64 = mem_cgroup_hierarchy_read,
3954 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003955 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05003956 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04003957 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04003958 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05003959 },
3960 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003961 .name = "swappiness",
3962 .read_u64 = mem_cgroup_swappiness_read,
3963 .write_u64 = mem_cgroup_swappiness_write,
3964 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003965 {
3966 .name = "move_charge_at_immigrate",
3967 .read_u64 = mem_cgroup_move_charge_read,
3968 .write_u64 = mem_cgroup_move_charge_write,
3969 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003970 {
3971 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003972 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003973 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003974 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3975 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003976 {
3977 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003978 },
Ying Han406eb0c2011-05-26 16:25:37 -07003979#ifdef CONFIG_NUMA
3980 {
3981 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003982 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07003983 },
3984#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08003985 {
3986 .name = "kmem.limit_in_bytes",
3987 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003988 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003989 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08003990 },
3991 {
3992 .name = "kmem.usage_in_bytes",
3993 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003994 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08003995 },
3996 {
3997 .name = "kmem.failcnt",
3998 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04003999 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004000 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004001 },
4002 {
4003 .name = "kmem.max_usage_in_bytes",
4004 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004005 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004006 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004007 },
Glauber Costa749c5412012-12-18 14:23:01 -08004008#ifdef CONFIG_SLABINFO
4009 {
4010 .name = "kmem.slabinfo",
Tejun Heobc2791f2017-02-22 15:41:21 -08004011 .seq_start = memcg_slab_start,
4012 .seq_next = memcg_slab_next,
4013 .seq_stop = memcg_slab_stop,
Vladimir Davydovb0475012014-12-10 15:44:19 -08004014 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004015 },
4016#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004017 {
4018 .name = "kmem.tcp.limit_in_bytes",
4019 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4020 .write = mem_cgroup_write,
4021 .read_u64 = mem_cgroup_read_u64,
4022 },
4023 {
4024 .name = "kmem.tcp.usage_in_bytes",
4025 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4026 .read_u64 = mem_cgroup_read_u64,
4027 },
4028 {
4029 .name = "kmem.tcp.failcnt",
4030 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4031 .write = mem_cgroup_reset,
4032 .read_u64 = mem_cgroup_read_u64,
4033 },
4034 {
4035 .name = "kmem.tcp.max_usage_in_bytes",
4036 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4037 .write = mem_cgroup_reset,
4038 .read_u64 = mem_cgroup_read_u64,
4039 },
Tejun Heo6bc10342012-04-01 12:09:55 -07004040 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004041};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004042
Johannes Weiner73f576c2016-07-20 15:44:57 -07004043/*
4044 * Private memory cgroup IDR
4045 *
4046 * Swap-out records and page cache shadow entries need to store memcg
4047 * references in constrained space, so we maintain an ID space that is
4048 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4049 * memory-controlled cgroups to 64k.
4050 *
4051 * However, there usually are many references to the oflline CSS after
4052 * the cgroup has been destroyed, such as page cache or reclaimable
4053 * slab objects, that don't need to hang on to the ID. We want to keep
4054 * those dead CSS from occupying IDs, or we might quickly exhaust the
4055 * relatively small ID space and prevent the creation of new cgroups
4056 * even when there are much fewer than 64k cgroups - possibly none.
4057 *
4058 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4059 * be freed and recycled when it's no longer needed, which is usually
4060 * when the CSS is offlined.
4061 *
4062 * The only exception to that are records of swapped out tmpfs/shmem
4063 * pages that need to be attributed to live ancestors on swapin. But
4064 * those references are manageable from userspace.
4065 */
4066
4067static DEFINE_IDR(mem_cgroup_idr);
4068
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004069static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004070{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004071 VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0);
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004072 atomic_add(n, &memcg->id.ref);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004073}
4074
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004075static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004076{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004077 VM_BUG_ON(atomic_read(&memcg->id.ref) < n);
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004078 if (atomic_sub_and_test(n, &memcg->id.ref)) {
Johannes Weiner73f576c2016-07-20 15:44:57 -07004079 idr_remove(&mem_cgroup_idr, memcg->id.id);
4080 memcg->id.id = 0;
4081
4082 /* Memcg ID pins CSS */
4083 css_put(&memcg->css);
4084 }
4085}
4086
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004087static inline void mem_cgroup_id_get(struct mem_cgroup *memcg)
4088{
4089 mem_cgroup_id_get_many(memcg, 1);
4090}
4091
4092static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
4093{
4094 mem_cgroup_id_put_many(memcg, 1);
4095}
4096
Johannes Weiner73f576c2016-07-20 15:44:57 -07004097/**
4098 * mem_cgroup_from_id - look up a memcg from a memcg id
4099 * @id: the memcg id to look up
4100 *
4101 * Caller must hold rcu_read_lock().
4102 */
4103struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
4104{
4105 WARN_ON_ONCE(!rcu_read_lock_held());
4106 return idr_find(&mem_cgroup_idr, id);
4107}
4108
Mel Gormanef8f2322016-07-28 15:46:05 -07004109static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004110{
4111 struct mem_cgroup_per_node *pn;
Mel Gormanef8f2322016-07-28 15:46:05 -07004112 int tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004113 /*
4114 * This routine is called against possible nodes.
4115 * But it's BUG to call kmalloc() against offline node.
4116 *
4117 * TODO: this routine can waste much memory for nodes which will
4118 * never be onlined. It's better to use memory hotplug callback
4119 * function.
4120 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004121 if (!node_state(node, N_NORMAL_MEMORY))
4122 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004123 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004124 if (!pn)
4125 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004126
Mel Gormanef8f2322016-07-28 15:46:05 -07004127 lruvec_init(&pn->lruvec);
4128 pn->usage_in_excess = 0;
4129 pn->on_tree = false;
4130 pn->memcg = memcg;
4131
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004132 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004133 return 0;
4134}
4135
Mel Gormanef8f2322016-07-28 15:46:05 -07004136static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004137{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004138 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004139}
4140
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08004141static void __mem_cgroup_free(struct mem_cgroup *memcg)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004142{
4143 int node;
4144
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004145 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004146 free_mem_cgroup_per_node_info(memcg, node);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004147 free_percpu(memcg->stat);
4148 kfree(memcg);
4149}
4150
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08004151static void mem_cgroup_free(struct mem_cgroup *memcg)
4152{
4153 memcg_wb_domain_exit(memcg);
4154 __mem_cgroup_free(memcg);
4155}
4156
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004157static struct mem_cgroup *mem_cgroup_alloc(void)
4158{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004159 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004160 size_t size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004161 int node;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004162
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004163 size = sizeof(struct mem_cgroup);
4164 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004165
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004166 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004167 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004168 return NULL;
4169
Johannes Weiner73f576c2016-07-20 15:44:57 -07004170 memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
4171 1, MEM_CGROUP_ID_MAX,
4172 GFP_KERNEL);
4173 if (memcg->id.id < 0)
4174 goto fail;
4175
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004176 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4177 if (!memcg->stat)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004178 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004179
Bob Liu3ed28fa2012-01-12 17:19:04 -08004180 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004181 if (alloc_mem_cgroup_per_node_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004182 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004183
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004184 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4185 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004186
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004187 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08004188 memcg->last_scanned_node = MAX_NUMNODES;
4189 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004190 mutex_init(&memcg->thresholds_lock);
4191 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004192 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004193 INIT_LIST_HEAD(&memcg->event_list);
4194 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004195 memcg->socket_pressure = jiffies;
Johannes Weiner127424c2016-01-20 15:02:32 -08004196#ifndef CONFIG_SLOB
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004197 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004198#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004199#ifdef CONFIG_CGROUP_WRITEBACK
4200 INIT_LIST_HEAD(&memcg->cgwb_list);
4201#endif
Johannes Weiner73f576c2016-07-20 15:44:57 -07004202 idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004203 return memcg;
4204fail:
Johannes Weiner73f576c2016-07-20 15:44:57 -07004205 if (memcg->id.id > 0)
4206 idr_remove(&mem_cgroup_idr, memcg->id.id);
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08004207 __mem_cgroup_free(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004208 return NULL;
Glauber Costad142e3e2013-02-22 16:34:52 -08004209}
4210
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004211static struct cgroup_subsys_state * __ref
4212mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004213{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004214 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4215 struct mem_cgroup *memcg;
4216 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08004217
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004218 memcg = mem_cgroup_alloc();
4219 if (!memcg)
4220 return ERR_PTR(error);
Li Zefan4219b2d2013-09-23 16:56:29 +08004221
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004222 memcg->high = PAGE_COUNTER_MAX;
4223 memcg->soft_limit = PAGE_COUNTER_MAX;
4224 if (parent) {
4225 memcg->swappiness = mem_cgroup_swappiness(parent);
4226 memcg->oom_kill_disable = parent->oom_kill_disable;
4227 }
4228 if (parent && parent->use_hierarchy) {
4229 memcg->use_hierarchy = true;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004230 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004231 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004232 page_counter_init(&memcg->memsw, &parent->memsw);
4233 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08004234 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004235 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004236 page_counter_init(&memcg->memory, NULL);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004237 page_counter_init(&memcg->swap, NULL);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004238 page_counter_init(&memcg->memsw, NULL);
4239 page_counter_init(&memcg->kmem, NULL);
Johannes Weiner0db15292016-01-20 15:02:50 -08004240 page_counter_init(&memcg->tcpmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004241 /*
4242 * Deeper hierachy with use_hierarchy == false doesn't make
4243 * much sense so let cgroup subsystem know about this
4244 * unfortunate state in our controller.
4245 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004246 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004247 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004248 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08004249
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004250 /* The following stuff does not apply to the root */
4251 if (!parent) {
4252 root_mem_cgroup = memcg;
4253 return &memcg->css;
4254 }
4255
Vladimir Davydovb313aee2016-03-17 14:18:27 -07004256 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004257 if (error)
4258 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08004259
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004260 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004261 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004262
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004263 return &memcg->css;
4264fail:
4265 mem_cgroup_free(memcg);
Tejun Heoea3a9642016-06-24 14:49:58 -07004266 return ERR_PTR(-ENOMEM);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004267}
4268
Johannes Weiner73f576c2016-07-20 15:44:57 -07004269static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004270{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004271 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4272
Johannes Weiner73f576c2016-07-20 15:44:57 -07004273 /* Online state pins memcg ID, memcg ID pins CSS */
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004274 atomic_set(&memcg->id.ref, 1);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004275 css_get(css);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004276 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004277}
4278
Tejun Heoeb954192013-08-08 20:11:23 -04004279static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004280{
Tejun Heoeb954192013-08-08 20:11:23 -04004281 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004282 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004283
4284 /*
4285 * Unregister events and notify userspace.
4286 * Notify userspace about cgroup removing only after rmdir of cgroup
4287 * directory to avoid race between userspace and kernelspace.
4288 */
Tejun Heofba94802013-11-22 18:20:43 -05004289 spin_lock(&memcg->event_list_lock);
4290 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004291 list_del_init(&event->list);
4292 schedule_work(&event->remove);
4293 }
Tejun Heofba94802013-11-22 18:20:43 -05004294 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004295
Johannes Weiner567e9ab2016-01-20 15:02:24 -08004296 memcg_offline_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004297 wb_memcg_offline(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004298
4299 mem_cgroup_id_put(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004300}
4301
Vladimir Davydov6df38682015-12-29 14:54:10 -08004302static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4303{
4304 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4305
4306 invalidate_reclaim_iterators(memcg);
4307}
4308
Tejun Heoeb954192013-08-08 20:11:23 -04004309static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004310{
Tejun Heoeb954192013-08-08 20:11:23 -04004311 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004312
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004313 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004314 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004315
Johannes Weiner0db15292016-01-20 15:02:50 -08004316 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004317 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004318
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004319 vmpressure_cleanup(&memcg->vmpressure);
4320 cancel_work_sync(&memcg->high_work);
4321 mem_cgroup_remove_from_trees(memcg);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004322 memcg_free_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004323 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004324}
4325
Tejun Heo1ced9532014-07-08 18:02:57 -04004326/**
4327 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4328 * @css: the target css
4329 *
4330 * Reset the states of the mem_cgroup associated with @css. This is
4331 * invoked when the userland requests disabling on the default hierarchy
4332 * but the memcg is pinned through dependency. The memcg should stop
4333 * applying policies and should revert to the vanilla state as it may be
4334 * made visible again.
4335 *
4336 * The current implementation only resets the essential configurations.
4337 * This needs to be expanded to cover all the visible parts.
4338 */
4339static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4340{
4341 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4342
Vladimir Davydovd334c9bc2016-03-17 14:19:38 -07004343 page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4344 page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4345 page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4346 page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4347 page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004348 memcg->low = 0;
4349 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004350 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004351 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004352}
4353
Daisuke Nishimura02491442010-03-10 15:22:17 -08004354#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004355/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004356static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004357{
Johannes Weiner05b84302014-08-06 16:05:59 -07004358 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004359
Mel Gormand0164ad2015-11-06 16:28:21 -08004360 /* Try a single bulk charge without reclaim first, kswapd may wake */
4361 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004362 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004363 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004364 return ret;
4365 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004366
David Rientjes36745342017-01-24 15:18:10 -08004367 /* Try charges one by one with reclaim, but do not retry */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004368 while (count--) {
David Rientjes36745342017-01-24 15:18:10 -08004369 ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004370 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004371 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004372 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004373 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004374 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004375 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004376}
4377
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004378union mc_target {
4379 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004380 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004381};
4382
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004383enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004384 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004385 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004386 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004387};
4388
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004389static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4390 unsigned long addr, pte_t ptent)
4391{
4392 struct page *page = vm_normal_page(vma, addr, ptent);
4393
4394 if (!page || !page_mapped(page))
4395 return NULL;
4396 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004397 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004398 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004399 } else {
4400 if (!(mc.flags & MOVE_FILE))
4401 return NULL;
4402 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004403 if (!get_page_unless_zero(page))
4404 return NULL;
4405
4406 return page;
4407}
4408
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004409#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004410static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004411 pte_t ptent, swp_entry_t *entry)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004412{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004413 struct page *page = NULL;
4414 swp_entry_t ent = pte_to_swp_entry(ptent);
4415
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004416 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004417 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004418 /*
4419 * Because lookup_swap_cache() updates some statistics counter,
4420 * we call find_get_page() with swapper_space directly.
4421 */
Huang Yingf6ab1f72016-10-07 17:00:21 -07004422 page = find_get_page(swap_address_space(ent), swp_offset(ent));
Johannes Weiner7941d212016-01-14 15:21:23 -08004423 if (do_memsw_account())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004424 entry->val = ent.val;
4425
4426 return page;
4427}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004428#else
4429static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004430 pte_t ptent, swp_entry_t *entry)
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004431{
4432 return NULL;
4433}
4434#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004435
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004436static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4437 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4438{
4439 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004440 struct address_space *mapping;
4441 pgoff_t pgoff;
4442
4443 if (!vma->vm_file) /* anonymous vma */
4444 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004445 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004446 return NULL;
4447
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004448 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004449 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004450
4451 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004452#ifdef CONFIG_SWAP
4453 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004454 if (shmem_mapping(mapping)) {
4455 page = find_get_entry(mapping, pgoff);
4456 if (radix_tree_exceptional_entry(page)) {
4457 swp_entry_t swp = radix_to_swp_entry(page);
Johannes Weiner7941d212016-01-14 15:21:23 -08004458 if (do_memsw_account())
Johannes Weiner139b6a62014-05-06 12:50:05 -07004459 *entry = swp;
Huang Yingf6ab1f72016-10-07 17:00:21 -07004460 page = find_get_page(swap_address_space(swp),
4461 swp_offset(swp));
Johannes Weiner139b6a62014-05-06 12:50:05 -07004462 }
4463 } else
4464 page = find_get_page(mapping, pgoff);
4465#else
4466 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004467#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004468 return page;
4469}
4470
Chen Gangb1b0dea2015-04-14 15:47:35 -07004471/**
4472 * mem_cgroup_move_account - move account of the page
4473 * @page: the page
Li RongQing25843c22016-07-26 15:26:56 -07004474 * @compound: charge the page as compound or small page
Chen Gangb1b0dea2015-04-14 15:47:35 -07004475 * @from: mem_cgroup which the page is moved from.
4476 * @to: mem_cgroup which the page is moved to. @from != @to.
4477 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08004478 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07004479 *
4480 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4481 * from old cgroup.
4482 */
4483static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004484 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07004485 struct mem_cgroup *from,
4486 struct mem_cgroup *to)
4487{
4488 unsigned long flags;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004489 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004490 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004491 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004492
4493 VM_BUG_ON(from == to);
4494 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004495 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07004496
4497 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07004498 * Prevent mem_cgroup_migrate() from looking at
Hugh Dickins45637ba2015-11-05 18:49:40 -08004499 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07004500 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004501 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004502 if (!trylock_page(page))
4503 goto out;
4504
4505 ret = -EINVAL;
4506 if (page->mem_cgroup != from)
4507 goto out_unlock;
4508
Greg Thelenc4843a72015-05-22 17:13:16 -04004509 anon = PageAnon(page);
4510
Chen Gangb1b0dea2015-04-14 15:47:35 -07004511 spin_lock_irqsave(&from->move_lock, flags);
4512
Greg Thelenc4843a72015-05-22 17:13:16 -04004513 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004514 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4515 nr_pages);
4516 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4517 nr_pages);
4518 }
4519
Greg Thelenc4843a72015-05-22 17:13:16 -04004520 /*
4521 * move_lock grabbed above and caller set from->moving_account, so
4522 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4523 * So mapping should be stable for dirty pages.
4524 */
4525 if (!anon && PageDirty(page)) {
4526 struct address_space *mapping = page_mapping(page);
4527
4528 if (mapping_cap_account_dirty(mapping)) {
4529 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4530 nr_pages);
4531 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4532 nr_pages);
4533 }
4534 }
4535
Chen Gangb1b0dea2015-04-14 15:47:35 -07004536 if (PageWriteback(page)) {
4537 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4538 nr_pages);
4539 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4540 nr_pages);
4541 }
4542
4543 /*
4544 * It is safe to change page->mem_cgroup here because the page
4545 * is referenced, charged, and isolated - we can't race with
4546 * uncharging, charging, migration, or LRU putback.
4547 */
4548
4549 /* caller should have done css_get */
4550 page->mem_cgroup = to;
4551 spin_unlock_irqrestore(&from->move_lock, flags);
4552
4553 ret = 0;
4554
4555 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004556 mem_cgroup_charge_statistics(to, page, compound, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004557 memcg_check_events(to, page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004558 mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004559 memcg_check_events(from, page);
4560 local_irq_enable();
4561out_unlock:
4562 unlock_page(page);
4563out:
4564 return ret;
4565}
4566
Li RongQing7cf78062016-05-27 14:27:46 -07004567/**
4568 * get_mctgt_type - get target type of moving charge
4569 * @vma: the vma the pte to be checked belongs
4570 * @addr: the address corresponding to the pte to be checked
4571 * @ptent: the pte to be checked
4572 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4573 *
4574 * Returns
4575 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4576 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4577 * move charge. if @target is not NULL, the page is stored in target->page
4578 * with extra refcnt got(Callers should handle it).
4579 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4580 * target for charge migration. if @target is not NULL, the entry is stored
4581 * in target->ent.
4582 *
4583 * Called with pte lock held.
4584 */
4585
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004586static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004587 unsigned long addr, pte_t ptent, union mc_target *target)
4588{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004589 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004590 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004591 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004592
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004593 if (pte_present(ptent))
4594 page = mc_handle_present_pte(vma, addr, ptent);
4595 else if (is_swap_pte(ptent))
Li RongQing48406ef2016-07-26 15:22:14 -07004596 page = mc_handle_swap_pte(vma, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004597 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004598 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004599
4600 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004601 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004602 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004603 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004604 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004605 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004606 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004607 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004608 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004609 ret = MC_TARGET_PAGE;
4610 if (target)
4611 target->page = page;
4612 }
4613 if (!ret || !target)
4614 put_page(page);
4615 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004616 /* There is a swap entry and a page doesn't exist or isn't charged */
4617 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004618 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004619 ret = MC_TARGET_SWAP;
4620 if (target)
4621 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004622 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004623 return ret;
4624}
4625
Naoya Horiguchi12724852012-03-21 16:34:28 -07004626#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4627/*
4628 * We don't consider swapping or file mapped pages because THP does not
4629 * support them for now.
4630 * Caller should make sure that pmd_trans_huge(pmd) is true.
4631 */
4632static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4633 unsigned long addr, pmd_t pmd, union mc_target *target)
4634{
4635 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004636 enum mc_target_type ret = MC_TARGET_NONE;
4637
4638 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004639 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004640 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004641 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004642 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004643 ret = MC_TARGET_PAGE;
4644 if (target) {
4645 get_page(page);
4646 target->page = page;
4647 }
4648 }
4649 return ret;
4650}
4651#else
4652static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4653 unsigned long addr, pmd_t pmd, union mc_target *target)
4654{
4655 return MC_TARGET_NONE;
4656}
4657#endif
4658
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004659static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4660 unsigned long addr, unsigned long end,
4661 struct mm_walk *walk)
4662{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004663 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004664 pte_t *pte;
4665 spinlock_t *ptl;
4666
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004667 ptl = pmd_trans_huge_lock(pmd, vma);
4668 if (ptl) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004669 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4670 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004671 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004672 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004673 }
Dave Hansen03319322011-03-22 16:32:56 -07004674
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004675 if (pmd_trans_unstable(pmd))
4676 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004677 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4678 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004679 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004680 mc.precharge++; /* increment precharge temporarily */
4681 pte_unmap_unlock(pte - 1, ptl);
4682 cond_resched();
4683
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004684 return 0;
4685}
4686
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004687static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4688{
4689 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004690
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004691 struct mm_walk mem_cgroup_count_precharge_walk = {
4692 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4693 .mm = mm,
4694 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004695 down_read(&mm->mmap_sem);
James Morse0247f3f2016-10-07 17:00:12 -07004696 walk_page_range(0, mm->highest_vm_end,
4697 &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004698 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004699
4700 precharge = mc.precharge;
4701 mc.precharge = 0;
4702
4703 return precharge;
4704}
4705
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004706static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4707{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004708 unsigned long precharge = mem_cgroup_count_precharge(mm);
4709
4710 VM_BUG_ON(mc.moving_task);
4711 mc.moving_task = current;
4712 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004713}
4714
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004715/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4716static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004717{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004718 struct mem_cgroup *from = mc.from;
4719 struct mem_cgroup *to = mc.to;
4720
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004721 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004722 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004723 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004724 mc.precharge = 0;
4725 }
4726 /*
4727 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4728 * we must uncharge here.
4729 */
4730 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004731 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004732 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004733 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004734 /* we must fixup refcnts and charges */
4735 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004736 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004737 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004738 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004739
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004740 mem_cgroup_id_put_many(mc.from, mc.moved_swap);
4741
Johannes Weiner05b84302014-08-06 16:05:59 -07004742 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004743 * we charged both to->memory and to->memsw, so we
4744 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004745 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004746 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004747 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004748
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004749 mem_cgroup_id_get_many(mc.to, mc.moved_swap);
4750 css_put_many(&mc.to->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004751
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004752 mc.moved_swap = 0;
4753 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004754 memcg_oom_recover(from);
4755 memcg_oom_recover(to);
4756 wake_up_all(&mc.waitq);
4757}
4758
4759static void mem_cgroup_clear_mc(void)
4760{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004761 struct mm_struct *mm = mc.mm;
4762
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004763 /*
4764 * we must clear moving_task before waking up waiters at the end of
4765 * task migration.
4766 */
4767 mc.moving_task = NULL;
4768 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004769 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004770 mc.from = NULL;
4771 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04004772 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004773 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004774
4775 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004776}
4777
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004778static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004779{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004780 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07004781 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07004782 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04004783 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004784 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004785 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004786 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004787
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004788 /* charge immigration isn't supported on the default hierarchy */
4789 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07004790 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004791
Tejun Heo4530edd2015-09-11 15:00:19 -04004792 /*
4793 * Multi-process migrations only happen on the default hierarchy
4794 * where charge immigration is not used. Perform charge
4795 * immigration if @tset contains a leader and whine if there are
4796 * multiple.
4797 */
4798 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004799 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04004800 WARN_ON_ONCE(p);
4801 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004802 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04004803 }
4804 if (!p)
4805 return 0;
4806
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004807 /*
4808 * We are now commited to this value whatever it is. Changes in this
4809 * tunable will only affect upcoming migrations, not the current one.
4810 * So we need to save it, and keep it going.
4811 */
4812 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4813 if (!move_flags)
4814 return 0;
4815
Tejun Heo9f2115f2015-09-08 15:01:10 -07004816 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004817
Tejun Heo9f2115f2015-09-08 15:01:10 -07004818 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004819
Tejun Heo9f2115f2015-09-08 15:01:10 -07004820 mm = get_task_mm(p);
4821 if (!mm)
4822 return 0;
4823 /* We move charges only when we move a owner of the mm */
4824 if (mm->owner == p) {
4825 VM_BUG_ON(mc.from);
4826 VM_BUG_ON(mc.to);
4827 VM_BUG_ON(mc.precharge);
4828 VM_BUG_ON(mc.moved_charge);
4829 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004830
Tejun Heo9f2115f2015-09-08 15:01:10 -07004831 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004832 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004833 mc.from = from;
4834 mc.to = memcg;
4835 mc.flags = move_flags;
4836 spin_unlock(&mc.lock);
4837 /* We set mc.moving_task later */
4838
4839 ret = mem_cgroup_precharge_mc(mm);
4840 if (ret)
4841 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004842 } else {
4843 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004844 }
4845 return ret;
4846}
4847
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004848static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004849{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004850 if (mc.to)
4851 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004852}
4853
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004854static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4855 unsigned long addr, unsigned long end,
4856 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004857{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004858 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004859 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004860 pte_t *pte;
4861 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004862 enum mc_target_type target_type;
4863 union mc_target target;
4864 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004865
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004866 ptl = pmd_trans_huge_lock(pmd, vma);
4867 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004868 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004869 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004870 return 0;
4871 }
4872 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4873 if (target_type == MC_TARGET_PAGE) {
4874 page = target.page;
4875 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004876 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08004877 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004878 mc.precharge -= HPAGE_PMD_NR;
4879 mc.moved_charge += HPAGE_PMD_NR;
4880 }
4881 putback_lru_page(page);
4882 }
4883 put_page(page);
4884 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004885 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004886 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004887 }
4888
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004889 if (pmd_trans_unstable(pmd))
4890 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004891retry:
4892 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4893 for (; addr != end; addr += PAGE_SIZE) {
4894 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004895 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004896
4897 if (!mc.precharge)
4898 break;
4899
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004900 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004901 case MC_TARGET_PAGE:
4902 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08004903 /*
4904 * We can have a part of the split pmd here. Moving it
4905 * can be done but it would be too convoluted so simply
4906 * ignore such a partial THP and keep it in original
4907 * memcg. There should be somebody mapping the head.
4908 */
4909 if (PageTransCompound(page))
4910 goto put;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004911 if (isolate_lru_page(page))
4912 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004913 if (!mem_cgroup_move_account(page, false,
4914 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004915 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004916 /* we uncharge from mc.from later. */
4917 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004918 }
4919 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004920put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004921 put_page(page);
4922 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004923 case MC_TARGET_SWAP:
4924 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004925 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004926 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004927 /* we fixup refcnts and charges later. */
4928 mc.moved_swap++;
4929 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004930 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004931 default:
4932 break;
4933 }
4934 }
4935 pte_unmap_unlock(pte - 1, ptl);
4936 cond_resched();
4937
4938 if (addr != end) {
4939 /*
4940 * We have consumed all precharges we got in can_attach().
4941 * We try charge one by one, but don't do any additional
4942 * charges to mc.to if we have failed in charge once in attach()
4943 * phase.
4944 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004945 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004946 if (!ret)
4947 goto retry;
4948 }
4949
4950 return ret;
4951}
4952
Tejun Heo264a0ae2016-04-21 19:09:02 -04004953static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004954{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004955 struct mm_walk mem_cgroup_move_charge_walk = {
4956 .pmd_entry = mem_cgroup_move_charge_pte_range,
Tejun Heo264a0ae2016-04-21 19:09:02 -04004957 .mm = mc.mm,
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004958 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004959
4960 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004961 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07004962 * Signal lock_page_memcg() to take the memcg's move_lock
4963 * while we're moving its pages to another memcg. Then wait
4964 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08004965 */
4966 atomic_inc(&mc.from->moving_account);
4967 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004968retry:
Tejun Heo264a0ae2016-04-21 19:09:02 -04004969 if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004970 /*
4971 * Someone who are holding the mmap_sem might be waiting in
4972 * waitq. So we cancel all extra charges, wake up all waiters,
4973 * and retry. Because we cancel precharges, we might not be able
4974 * to move enough charges, but moving charge is a best-effort
4975 * feature anyway, so it wouldn't be a big problem.
4976 */
4977 __mem_cgroup_clear_mc();
4978 cond_resched();
4979 goto retry;
4980 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004981 /*
4982 * When we have consumed all precharges and failed in doing
4983 * additional charge, the page walk just aborts.
4984 */
James Morse0247f3f2016-10-07 17:00:12 -07004985 walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk);
4986
Tejun Heo264a0ae2016-04-21 19:09:02 -04004987 up_read(&mc.mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08004988 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004989}
4990
Tejun Heo264a0ae2016-04-21 19:09:02 -04004991static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08004992{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004993 if (mc.to) {
4994 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07004995 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004996 }
Balbir Singh67e465a2008-02-07 00:13:54 -08004997}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07004998#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004999static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005000{
5001 return 0;
5002}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005003static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005004{
5005}
Tejun Heo264a0ae2016-04-21 19:09:02 -04005006static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005007{
5008}
5009#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005010
Tejun Heof00baae2013-04-15 13:41:15 -07005011/*
5012 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005013 * to verify whether we're attached to the default hierarchy on each mount
5014 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005015 */
Tejun Heoeb954192013-08-08 20:11:23 -04005016static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005017{
5018 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005019 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005020 * guarantees that @root doesn't have any children, so turning it
5021 * on for the root memcg is enough.
5022 */
Tejun Heo9e10a132015-09-18 11:56:28 -04005023 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee592015-03-12 16:26:19 -07005024 root_mem_cgroup->use_hierarchy = true;
5025 else
5026 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005027}
5028
Johannes Weiner241994ed2015-02-11 15:26:06 -08005029static u64 memory_current_read(struct cgroup_subsys_state *css,
5030 struct cftype *cft)
5031{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005032 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5033
5034 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005035}
5036
5037static int memory_low_show(struct seq_file *m, void *v)
5038{
5039 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005040 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005041
5042 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005043 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005044 else
5045 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5046
5047 return 0;
5048}
5049
5050static ssize_t memory_low_write(struct kernfs_open_file *of,
5051 char *buf, size_t nbytes, loff_t off)
5052{
5053 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5054 unsigned long low;
5055 int err;
5056
5057 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005058 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005059 if (err)
5060 return err;
5061
5062 memcg->low = low;
5063
5064 return nbytes;
5065}
5066
5067static int memory_high_show(struct seq_file *m, void *v)
5068{
5069 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005070 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005071
5072 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005073 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005074 else
5075 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5076
5077 return 0;
5078}
5079
5080static ssize_t memory_high_write(struct kernfs_open_file *of,
5081 char *buf, size_t nbytes, loff_t off)
5082{
5083 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner588083b2016-03-17 14:20:25 -07005084 unsigned long nr_pages;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005085 unsigned long high;
5086 int err;
5087
5088 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005089 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005090 if (err)
5091 return err;
5092
5093 memcg->high = high;
5094
Johannes Weiner588083b2016-03-17 14:20:25 -07005095 nr_pages = page_counter_read(&memcg->memory);
5096 if (nr_pages > high)
5097 try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5098 GFP_KERNEL, true);
5099
Tejun Heo2529bb32015-05-22 18:23:34 -04005100 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005101 return nbytes;
5102}
5103
5104static int memory_max_show(struct seq_file *m, void *v)
5105{
5106 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005107 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005108
5109 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005110 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005111 else
5112 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5113
5114 return 0;
5115}
5116
5117static ssize_t memory_max_write(struct kernfs_open_file *of,
5118 char *buf, size_t nbytes, loff_t off)
5119{
5120 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005121 unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5122 bool drained = false;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005123 unsigned long max;
5124 int err;
5125
5126 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005127 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005128 if (err)
5129 return err;
5130
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005131 xchg(&memcg->memory.limit, max);
5132
5133 for (;;) {
5134 unsigned long nr_pages = page_counter_read(&memcg->memory);
5135
5136 if (nr_pages <= max)
5137 break;
5138
5139 if (signal_pending(current)) {
5140 err = -EINTR;
5141 break;
5142 }
5143
5144 if (!drained) {
5145 drain_all_stock(memcg);
5146 drained = true;
5147 continue;
5148 }
5149
5150 if (nr_reclaims) {
5151 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5152 GFP_KERNEL, true))
5153 nr_reclaims--;
5154 continue;
5155 }
5156
5157 mem_cgroup_events(memcg, MEMCG_OOM, 1);
5158 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5159 break;
5160 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08005161
Tejun Heo2529bb32015-05-22 18:23:34 -04005162 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005163 return nbytes;
5164}
5165
5166static int memory_events_show(struct seq_file *m, void *v)
5167{
5168 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5169
5170 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5171 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5172 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5173 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5174
5175 return 0;
5176}
5177
Johannes Weiner587d9f72016-01-20 15:03:19 -08005178static int memory_stat_show(struct seq_file *m, void *v)
5179{
5180 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005181 unsigned long stat[MEMCG_NR_STAT];
5182 unsigned long events[MEMCG_NR_EVENTS];
Johannes Weiner587d9f72016-01-20 15:03:19 -08005183 int i;
5184
5185 /*
5186 * Provide statistics on the state of the memory subsystem as
5187 * well as cumulative event counters that show past behavior.
5188 *
5189 * This list is ordered following a combination of these gradients:
5190 * 1) generic big picture -> specifics and details
5191 * 2) reflecting userspace activity -> reflecting kernel heuristics
5192 *
5193 * Current memory state:
5194 */
5195
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005196 tree_stat(memcg, stat);
5197 tree_events(memcg, events);
5198
Johannes Weiner587d9f72016-01-20 15:03:19 -08005199 seq_printf(m, "anon %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005200 (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005201 seq_printf(m, "file %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005202 (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE);
Vladimir Davydov12580e42016-03-17 14:17:38 -07005203 seq_printf(m, "kernel_stack %llu\n",
Andy Lutomirskiefdc9492016-07-28 15:48:17 -07005204 (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005205 seq_printf(m, "slab %llu\n",
5206 (u64)(stat[MEMCG_SLAB_RECLAIMABLE] +
5207 stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
Johannes Weinerb2807f02016-01-20 15:03:22 -08005208 seq_printf(m, "sock %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005209 (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005210
5211 seq_printf(m, "file_mapped %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005212 (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005213 seq_printf(m, "file_dirty %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005214 (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005215 seq_printf(m, "file_writeback %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005216 (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005217
5218 for (i = 0; i < NR_LRU_LISTS; i++) {
5219 struct mem_cgroup *mi;
5220 unsigned long val = 0;
5221
5222 for_each_mem_cgroup_tree(mi, memcg)
5223 val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5224 seq_printf(m, "%s %llu\n",
5225 mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5226 }
5227
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005228 seq_printf(m, "slab_reclaimable %llu\n",
5229 (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE);
5230 seq_printf(m, "slab_unreclaimable %llu\n",
5231 (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5232
Johannes Weiner587d9f72016-01-20 15:03:19 -08005233 /* Accumulated memory events */
5234
5235 seq_printf(m, "pgfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005236 events[MEM_CGROUP_EVENTS_PGFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005237 seq_printf(m, "pgmajfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005238 events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005239
5240 return 0;
5241}
5242
Johannes Weiner241994ed2015-02-11 15:26:06 -08005243static struct cftype memory_files[] = {
5244 {
5245 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005246 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005247 .read_u64 = memory_current_read,
5248 },
5249 {
5250 .name = "low",
5251 .flags = CFTYPE_NOT_ON_ROOT,
5252 .seq_show = memory_low_show,
5253 .write = memory_low_write,
5254 },
5255 {
5256 .name = "high",
5257 .flags = CFTYPE_NOT_ON_ROOT,
5258 .seq_show = memory_high_show,
5259 .write = memory_high_write,
5260 },
5261 {
5262 .name = "max",
5263 .flags = CFTYPE_NOT_ON_ROOT,
5264 .seq_show = memory_max_show,
5265 .write = memory_max_write,
5266 },
5267 {
5268 .name = "events",
5269 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04005270 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994ed2015-02-11 15:26:06 -08005271 .seq_show = memory_events_show,
5272 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08005273 {
5274 .name = "stat",
5275 .flags = CFTYPE_NOT_ON_ROOT,
5276 .seq_show = memory_stat_show,
5277 },
Johannes Weiner241994ed2015-02-11 15:26:06 -08005278 { } /* terminate */
5279};
5280
Tejun Heo073219e2014-02-08 10:36:58 -05005281struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005282 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005283 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005284 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08005285 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08005286 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005287 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005288 .can_attach = mem_cgroup_can_attach,
5289 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04005290 .post_attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005291 .bind = mem_cgroup_bind,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005292 .dfl_cftypes = memory_files,
5293 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005294 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005295};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005296
Johannes Weiner241994ed2015-02-11 15:26:06 -08005297/**
Johannes Weiner241994ed2015-02-11 15:26:06 -08005298 * mem_cgroup_low - check if memory consumption is below the normal range
5299 * @root: the highest ancestor to consider
5300 * @memcg: the memory cgroup to check
5301 *
5302 * Returns %true if memory consumption of @memcg, and that of all
5303 * configurable ancestors up to @root, is below the normal range.
5304 */
5305bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5306{
5307 if (mem_cgroup_disabled())
5308 return false;
5309
5310 /*
5311 * The toplevel group doesn't have a configurable range, so
5312 * it's never low when looked at directly, and it is not
5313 * considered an ancestor when assessing the hierarchy.
5314 */
5315
5316 if (memcg == root_mem_cgroup)
5317 return false;
5318
Michal Hocko4e54ded2015-02-27 15:51:46 -08005319 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005320 return false;
5321
5322 while (memcg != root) {
5323 memcg = parent_mem_cgroup(memcg);
5324
5325 if (memcg == root_mem_cgroup)
5326 break;
5327
Michal Hocko4e54ded2015-02-27 15:51:46 -08005328 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005329 return false;
5330 }
5331 return true;
5332}
5333
Johannes Weiner00501b52014-08-08 14:19:20 -07005334/**
5335 * mem_cgroup_try_charge - try charging a page
5336 * @page: page to charge
5337 * @mm: mm context of the victim
5338 * @gfp_mask: reclaim mode
5339 * @memcgp: charged memcg return
Li RongQing25843c22016-07-26 15:26:56 -07005340 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005341 *
5342 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5343 * pages according to @gfp_mask if necessary.
5344 *
5345 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5346 * Otherwise, an error code is returned.
5347 *
5348 * After page->mapping has been set up, the caller must finalize the
5349 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5350 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5351 */
5352int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005353 gfp_t gfp_mask, struct mem_cgroup **memcgp,
5354 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005355{
5356 struct mem_cgroup *memcg = NULL;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005357 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005358 int ret = 0;
5359
5360 if (mem_cgroup_disabled())
5361 goto out;
5362
5363 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005364 /*
5365 * Every swap fault against a single page tries to charge the
5366 * page, bail as early as possible. shmem_unuse() encounters
5367 * already charged pages, too. The USED bit is protected by
5368 * the page lock, which serializes swap cache removal, which
5369 * in turn serializes uncharging.
5370 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005371 VM_BUG_ON_PAGE(!PageLocked(page), page);
Johannes Weiner1306a852014-12-10 15:44:52 -08005372 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005373 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005374
Vladimir Davydov37e84352016-01-20 15:02:56 -08005375 if (do_swap_account) {
Vladimir Davydove993d902015-09-09 15:35:35 -07005376 swp_entry_t ent = { .val = page_private(page), };
5377 unsigned short id = lookup_swap_cgroup_id(ent);
5378
5379 rcu_read_lock();
5380 memcg = mem_cgroup_from_id(id);
5381 if (memcg && !css_tryget_online(&memcg->css))
5382 memcg = NULL;
5383 rcu_read_unlock();
5384 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005385 }
5386
Johannes Weiner00501b52014-08-08 14:19:20 -07005387 if (!memcg)
5388 memcg = get_mem_cgroup_from_mm(mm);
5389
5390 ret = try_charge(memcg, gfp_mask, nr_pages);
5391
5392 css_put(&memcg->css);
Johannes Weiner00501b52014-08-08 14:19:20 -07005393out:
5394 *memcgp = memcg;
5395 return ret;
5396}
5397
5398/**
5399 * mem_cgroup_commit_charge - commit a page charge
5400 * @page: page to charge
5401 * @memcg: memcg to charge the page to
5402 * @lrucare: page might be on LRU already
Li RongQing25843c22016-07-26 15:26:56 -07005403 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005404 *
5405 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5406 * after page->mapping has been set up. This must happen atomically
5407 * as part of the page instantiation, i.e. under the page table lock
5408 * for anonymous pages, under the page lock for page and swap cache.
5409 *
5410 * In addition, the page must not be on the LRU during the commit, to
5411 * prevent racing with task migration. If it might be, use @lrucare.
5412 *
5413 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5414 */
5415void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005416 bool lrucare, bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005417{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005418 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005419
5420 VM_BUG_ON_PAGE(!page->mapping, page);
5421 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5422
5423 if (mem_cgroup_disabled())
5424 return;
5425 /*
5426 * Swap faults will attempt to charge the same page multiple
5427 * times. But reuse_swap_page() might have removed the page
5428 * from swapcache already, so we can't check PageSwapCache().
5429 */
5430 if (!memcg)
5431 return;
5432
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005433 commit_charge(page, memcg, lrucare);
5434
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005435 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005436 mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005437 memcg_check_events(memcg, page);
5438 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005439
Johannes Weiner7941d212016-01-14 15:21:23 -08005440 if (do_memsw_account() && PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005441 swp_entry_t entry = { .val = page_private(page) };
5442 /*
5443 * The swap entry might not get freed for a long time,
5444 * let's not wait for it. The page already received a
5445 * memory+swap charge, drop the swap entry duplicate.
5446 */
5447 mem_cgroup_uncharge_swap(entry);
5448 }
5449}
5450
5451/**
5452 * mem_cgroup_cancel_charge - cancel a page charge
5453 * @page: page to charge
5454 * @memcg: memcg to charge the page to
Li RongQing25843c22016-07-26 15:26:56 -07005455 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005456 *
5457 * Cancel a charge transaction started by mem_cgroup_try_charge().
5458 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005459void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5460 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005461{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005462 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005463
5464 if (mem_cgroup_disabled())
5465 return;
5466 /*
5467 * Swap faults will attempt to charge the same page multiple
5468 * times. But reuse_swap_page() might have removed the page
5469 * from swapcache already, so we can't check PageSwapCache().
5470 */
5471 if (!memcg)
5472 return;
5473
Johannes Weiner00501b52014-08-08 14:19:20 -07005474 cancel_charge(memcg, nr_pages);
5475}
5476
Johannes Weiner747db952014-08-08 14:19:24 -07005477static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005478 unsigned long nr_anon, unsigned long nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005479 unsigned long nr_huge, unsigned long nr_kmem,
5480 struct page *dummy_page)
Johannes Weiner747db952014-08-08 14:19:24 -07005481{
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005482 unsigned long nr_pages = nr_anon + nr_file + nr_kmem;
Johannes Weiner747db952014-08-08 14:19:24 -07005483 unsigned long flags;
5484
Johannes Weinerce00a962014-09-05 08:43:57 -04005485 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005486 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08005487 if (do_memsw_account())
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005488 page_counter_uncharge(&memcg->memsw, nr_pages);
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005489 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem)
5490 page_counter_uncharge(&memcg->kmem, nr_kmem);
Johannes Weinerce00a962014-09-05 08:43:57 -04005491 memcg_oom_recover(memcg);
5492 }
Johannes Weiner747db952014-08-08 14:19:24 -07005493
5494 local_irq_save(flags);
5495 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5496 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5497 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5498 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005499 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005500 memcg_check_events(memcg, dummy_page);
5501 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005502
5503 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005504 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005505}
5506
5507static void uncharge_list(struct list_head *page_list)
5508{
5509 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005510 unsigned long nr_anon = 0;
5511 unsigned long nr_file = 0;
5512 unsigned long nr_huge = 0;
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005513 unsigned long nr_kmem = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005514 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005515 struct list_head *next;
5516 struct page *page;
5517
Johannes Weiner8b592652016-03-17 14:20:31 -07005518 /*
5519 * Note that the list can be a single page->lru; hence the
5520 * do-while loop instead of a simple list_for_each_entry().
5521 */
Johannes Weiner747db952014-08-08 14:19:24 -07005522 next = page_list->next;
5523 do {
Johannes Weiner747db952014-08-08 14:19:24 -07005524 page = list_entry(next, struct page, lru);
5525 next = page->lru.next;
5526
5527 VM_BUG_ON_PAGE(PageLRU(page), page);
5528 VM_BUG_ON_PAGE(page_count(page), page);
5529
Johannes Weiner1306a852014-12-10 15:44:52 -08005530 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005531 continue;
5532
5533 /*
5534 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005535 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005536 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005537 */
5538
Johannes Weiner1306a852014-12-10 15:44:52 -08005539 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005540 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005541 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005542 nr_huge, nr_kmem, page);
5543 pgpgout = nr_anon = nr_file =
5544 nr_huge = nr_kmem = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005545 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005546 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005547 }
5548
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005549 if (!PageKmemcg(page)) {
5550 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005551
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005552 if (PageTransHuge(page)) {
5553 nr_pages <<= compound_order(page);
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005554 nr_huge += nr_pages;
5555 }
5556 if (PageAnon(page))
5557 nr_anon += nr_pages;
5558 else
5559 nr_file += nr_pages;
5560 pgpgout++;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03005561 } else {
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005562 nr_kmem += 1 << compound_order(page);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03005563 __ClearPageKmemcg(page);
5564 }
Johannes Weiner747db952014-08-08 14:19:24 -07005565
Johannes Weiner1306a852014-12-10 15:44:52 -08005566 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005567 } while (next != page_list);
5568
5569 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005570 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005571 nr_huge, nr_kmem, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005572}
5573
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005574/**
5575 * mem_cgroup_uncharge - uncharge a page
5576 * @page: page to uncharge
5577 *
5578 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5579 * mem_cgroup_commit_charge().
5580 */
5581void mem_cgroup_uncharge(struct page *page)
5582{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005583 if (mem_cgroup_disabled())
5584 return;
5585
Johannes Weiner747db952014-08-08 14:19:24 -07005586 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005587 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005588 return;
5589
Johannes Weiner747db952014-08-08 14:19:24 -07005590 INIT_LIST_HEAD(&page->lru);
5591 uncharge_list(&page->lru);
5592}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005593
Johannes Weiner747db952014-08-08 14:19:24 -07005594/**
5595 * mem_cgroup_uncharge_list - uncharge a list of page
5596 * @page_list: list of pages to uncharge
5597 *
5598 * Uncharge a list of pages previously charged with
5599 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5600 */
5601void mem_cgroup_uncharge_list(struct list_head *page_list)
5602{
5603 if (mem_cgroup_disabled())
5604 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005605
Johannes Weiner747db952014-08-08 14:19:24 -07005606 if (!list_empty(page_list))
5607 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005608}
5609
5610/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005611 * mem_cgroup_migrate - charge a page's replacement
5612 * @oldpage: currently circulating page
5613 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005614 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005615 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5616 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005617 *
5618 * Both pages must be locked, @newpage->mapping must be set up.
5619 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005620void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005621{
Johannes Weiner29833312014-12-10 15:44:02 -08005622 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005623 unsigned int nr_pages;
5624 bool compound;
Tejun Heod93c4132016-06-24 14:49:54 -07005625 unsigned long flags;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005626
5627 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5628 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005629 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005630 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5631 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005632
5633 if (mem_cgroup_disabled())
5634 return;
5635
5636 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005637 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005638 return;
5639
Hugh Dickins45637ba2015-11-05 18:49:40 -08005640 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08005641 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005642 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005643 return;
5644
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005645 /* Force-charge the new page. The old one will be freed soon */
5646 compound = PageTransHuge(newpage);
5647 nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5648
5649 page_counter_charge(&memcg->memory, nr_pages);
5650 if (do_memsw_account())
5651 page_counter_charge(&memcg->memsw, nr_pages);
5652 css_get_many(&memcg->css, nr_pages);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005653
Johannes Weiner9cf76662016-03-15 14:57:58 -07005654 commit_charge(newpage, memcg, false);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005655
Tejun Heod93c4132016-06-24 14:49:54 -07005656 local_irq_save(flags);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005657 mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5658 memcg_check_events(memcg, newpage);
Tejun Heod93c4132016-06-24 14:49:54 -07005659 local_irq_restore(flags);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005660}
5661
Johannes Weineref129472016-01-14 15:21:34 -08005662DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08005663EXPORT_SYMBOL(memcg_sockets_enabled_key);
5664
Johannes Weiner2d758072016-10-07 17:00:58 -07005665void mem_cgroup_sk_alloc(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08005666{
5667 struct mem_cgroup *memcg;
5668
Johannes Weiner2d758072016-10-07 17:00:58 -07005669 if (!mem_cgroup_sockets_enabled)
5670 return;
5671
5672 /*
5673 * Socket cloning can throw us here with sk_memcg already
Johannes Weiner11092082016-01-14 15:21:26 -08005674 * filled. It won't however, necessarily happen from
5675 * process context. So the test for root memcg given
5676 * the current task's memcg won't help us in this case.
5677 *
5678 * Respecting the original socket's memcg is a better
5679 * decision in this case.
5680 */
5681 if (sk->sk_memcg) {
5682 BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5683 css_get(&sk->sk_memcg->css);
5684 return;
5685 }
5686
5687 rcu_read_lock();
5688 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005689 if (memcg == root_mem_cgroup)
5690 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08005691 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005692 goto out;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005693 if (css_tryget_online(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08005694 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005695out:
Johannes Weiner11092082016-01-14 15:21:26 -08005696 rcu_read_unlock();
5697}
Johannes Weiner11092082016-01-14 15:21:26 -08005698
Johannes Weiner2d758072016-10-07 17:00:58 -07005699void mem_cgroup_sk_free(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08005700{
Johannes Weiner2d758072016-10-07 17:00:58 -07005701 if (sk->sk_memcg)
5702 css_put(&sk->sk_memcg->css);
Johannes Weiner11092082016-01-14 15:21:26 -08005703}
5704
5705/**
5706 * mem_cgroup_charge_skmem - charge socket memory
5707 * @memcg: memcg to charge
5708 * @nr_pages: number of pages to charge
5709 *
5710 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5711 * @memcg's configured limit, %false if the charge had to be forced.
5712 */
5713bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5714{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005715 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08005716
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005717 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005718 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005719
Johannes Weiner0db15292016-01-20 15:02:50 -08005720 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5721 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005722 return true;
5723 }
Johannes Weiner0db15292016-01-20 15:02:50 -08005724 page_counter_charge(&memcg->tcpmem, nr_pages);
5725 memcg->tcpmem_pressure = 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005726 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08005727 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005728
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005729 /* Don't block in the packet receive path */
5730 if (in_softirq())
5731 gfp_mask = GFP_NOWAIT;
5732
Johannes Weinerb2807f02016-01-20 15:03:22 -08005733 this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5734
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005735 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5736 return true;
5737
5738 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005739 return false;
5740}
5741
5742/**
5743 * mem_cgroup_uncharge_skmem - uncharge socket memory
5744 * @memcg - memcg to uncharge
5745 * @nr_pages - number of pages to uncharge
5746 */
5747void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5748{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005749 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005750 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005751 return;
5752 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005753
Johannes Weinerb2807f02016-01-20 15:03:22 -08005754 this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5755
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005756 page_counter_uncharge(&memcg->memory, nr_pages);
5757 css_put_many(&memcg->css, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005758}
5759
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005760static int __init cgroup_memory(char *s)
5761{
5762 char *token;
5763
5764 while ((token = strsep(&s, ",")) != NULL) {
5765 if (!*token)
5766 continue;
5767 if (!strcmp(token, "nosocket"))
5768 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08005769 if (!strcmp(token, "nokmem"))
5770 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005771 }
5772 return 0;
5773}
5774__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08005775
Michal Hocko2d110852013-02-22 16:34:43 -08005776/*
Michal Hocko10813122013-02-22 16:35:41 -08005777 * subsys_initcall() for memory controller.
5778 *
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01005779 * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this
5780 * context because of lock dependencies (cgroup_lock -> cpu hotplug) but
5781 * basically everything that doesn't depend on a specific mem_cgroup structure
5782 * should be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005783 */
5784static int __init mem_cgroup_init(void)
5785{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005786 int cpu, node;
5787
Vladimir Davydov13583c32016-12-12 16:41:29 -08005788#ifndef CONFIG_SLOB
5789 /*
5790 * Kmem cache creation is mostly done with the slab_mutex held,
Tejun Heo17cc4df2017-02-22 15:41:36 -08005791 * so use a workqueue with limited concurrency to avoid stalling
5792 * all worker threads in case lots of cgroups are created and
5793 * destroyed simultaneously.
Vladimir Davydov13583c32016-12-12 16:41:29 -08005794 */
Tejun Heo17cc4df2017-02-22 15:41:36 -08005795 memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1);
5796 BUG_ON(!memcg_kmem_cache_wq);
Vladimir Davydov13583c32016-12-12 16:41:29 -08005797#endif
5798
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01005799 cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL,
5800 memcg_hotplug_cpu_dead);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005801
5802 for_each_possible_cpu(cpu)
5803 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5804 drain_local_stock);
5805
5806 for_each_node(node) {
5807 struct mem_cgroup_tree_per_node *rtpn;
Johannes Weiner95a045f2015-02-11 15:26:33 -08005808
5809 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5810 node_online(node) ? node : NUMA_NO_NODE);
5811
Mel Gormanef8f2322016-07-28 15:46:05 -07005812 rtpn->rb_root = RB_ROOT;
5813 spin_lock_init(&rtpn->lock);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005814 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5815 }
5816
Michal Hocko2d110852013-02-22 16:34:43 -08005817 return 0;
5818}
5819subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005820
5821#ifdef CONFIG_MEMCG_SWAP
Arnd Bergmann358c07f2016-08-25 15:17:08 -07005822static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
5823{
5824 while (!atomic_inc_not_zero(&memcg->id.ref)) {
5825 /*
5826 * The root cgroup cannot be destroyed, so it's refcount must
5827 * always be >= 1.
5828 */
5829 if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
5830 VM_BUG_ON(1);
5831 break;
5832 }
5833 memcg = parent_mem_cgroup(memcg);
5834 if (!memcg)
5835 memcg = root_mem_cgroup;
5836 }
5837 return memcg;
5838}
5839
Johannes Weiner21afa382015-02-11 15:26:36 -08005840/**
5841 * mem_cgroup_swapout - transfer a memsw charge to swap
5842 * @page: page whose memsw charge to transfer
5843 * @entry: swap entry to move the charge to
5844 *
5845 * Transfer the memsw charge of @page to @entry.
5846 */
5847void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5848{
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005849 struct mem_cgroup *memcg, *swap_memcg;
Johannes Weiner21afa382015-02-11 15:26:36 -08005850 unsigned short oldid;
5851
5852 VM_BUG_ON_PAGE(PageLRU(page), page);
5853 VM_BUG_ON_PAGE(page_count(page), page);
5854
Johannes Weiner7941d212016-01-14 15:21:23 -08005855 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08005856 return;
5857
5858 memcg = page->mem_cgroup;
5859
5860 /* Readahead page, never charged */
5861 if (!memcg)
5862 return;
5863
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005864 /*
5865 * In case the memcg owning these pages has been offlined and doesn't
5866 * have an ID allocated to it anymore, charge the closest online
5867 * ancestor for the swap instead and transfer the memory+swap charge.
5868 */
5869 swap_memcg = mem_cgroup_id_get_online(memcg);
5870 oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg));
Johannes Weiner21afa382015-02-11 15:26:36 -08005871 VM_BUG_ON_PAGE(oldid, page);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005872 mem_cgroup_swap_statistics(swap_memcg, true);
Johannes Weiner21afa382015-02-11 15:26:36 -08005873
5874 page->mem_cgroup = NULL;
5875
5876 if (!mem_cgroup_is_root(memcg))
5877 page_counter_uncharge(&memcg->memory, 1);
5878
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005879 if (memcg != swap_memcg) {
5880 if (!mem_cgroup_is_root(swap_memcg))
5881 page_counter_charge(&swap_memcg->memsw, 1);
5882 page_counter_uncharge(&memcg->memsw, 1);
5883 }
5884
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07005885 /*
5886 * Interrupts should be disabled here because the caller holds the
5887 * mapping->tree_lock lock which is taken with interrupts-off. It is
5888 * important here to have the interrupts disabled because it is the
5889 * only synchronisation we have for udpating the per-CPU variables.
5890 */
5891 VM_BUG_ON(!irqs_disabled());
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005892 mem_cgroup_charge_statistics(memcg, page, false, -1);
Johannes Weiner21afa382015-02-11 15:26:36 -08005893 memcg_check_events(memcg, page);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005894
5895 if (!mem_cgroup_is_root(memcg))
5896 css_put(&memcg->css);
Johannes Weiner21afa382015-02-11 15:26:36 -08005897}
5898
Vladimir Davydov37e84352016-01-20 15:02:56 -08005899/*
5900 * mem_cgroup_try_charge_swap - try charging a swap entry
5901 * @page: page being added to swap
5902 * @entry: swap entry to charge
5903 *
5904 * Try to charge @entry to the memcg that @page belongs to.
5905 *
5906 * Returns 0 on success, -ENOMEM on failure.
5907 */
5908int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5909{
5910 struct mem_cgroup *memcg;
5911 struct page_counter *counter;
5912 unsigned short oldid;
5913
5914 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5915 return 0;
5916
5917 memcg = page->mem_cgroup;
5918
5919 /* Readahead page, never charged */
5920 if (!memcg)
5921 return 0;
5922
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005923 memcg = mem_cgroup_id_get_online(memcg);
Vladimir Davydov37e84352016-01-20 15:02:56 -08005924
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005925 if (!mem_cgroup_is_root(memcg) &&
5926 !page_counter_try_charge(&memcg->swap, 1, &counter)) {
5927 mem_cgroup_id_put(memcg);
5928 return -ENOMEM;
5929 }
5930
Vladimir Davydov37e84352016-01-20 15:02:56 -08005931 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5932 VM_BUG_ON_PAGE(oldid, page);
5933 mem_cgroup_swap_statistics(memcg, true);
5934
Vladimir Davydov37e84352016-01-20 15:02:56 -08005935 return 0;
5936}
5937
Johannes Weiner21afa382015-02-11 15:26:36 -08005938/**
5939 * mem_cgroup_uncharge_swap - uncharge a swap entry
5940 * @entry: swap entry to uncharge
5941 *
Vladimir Davydov37e84352016-01-20 15:02:56 -08005942 * Drop the swap charge associated with @entry.
Johannes Weiner21afa382015-02-11 15:26:36 -08005943 */
5944void mem_cgroup_uncharge_swap(swp_entry_t entry)
5945{
5946 struct mem_cgroup *memcg;
5947 unsigned short id;
5948
Vladimir Davydov37e84352016-01-20 15:02:56 -08005949 if (!do_swap_account)
Johannes Weiner21afa382015-02-11 15:26:36 -08005950 return;
5951
5952 id = swap_cgroup_record(entry, 0);
5953 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005954 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005955 if (memcg) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08005956 if (!mem_cgroup_is_root(memcg)) {
5957 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5958 page_counter_uncharge(&memcg->swap, 1);
5959 else
5960 page_counter_uncharge(&memcg->memsw, 1);
5961 }
Johannes Weiner21afa382015-02-11 15:26:36 -08005962 mem_cgroup_swap_statistics(memcg, false);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005963 mem_cgroup_id_put(memcg);
Johannes Weiner21afa382015-02-11 15:26:36 -08005964 }
5965 rcu_read_unlock();
5966}
5967
Vladimir Davydovd8b38432016-01-20 15:03:07 -08005968long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
5969{
5970 long nr_swap_pages = get_nr_swap_pages();
5971
5972 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5973 return nr_swap_pages;
5974 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5975 nr_swap_pages = min_t(long, nr_swap_pages,
5976 READ_ONCE(memcg->swap.limit) -
5977 page_counter_read(&memcg->swap));
5978 return nr_swap_pages;
5979}
5980
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08005981bool mem_cgroup_swap_full(struct page *page)
5982{
5983 struct mem_cgroup *memcg;
5984
5985 VM_BUG_ON_PAGE(!PageLocked(page), page);
5986
5987 if (vm_swap_full())
5988 return true;
5989 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5990 return false;
5991
5992 memcg = page->mem_cgroup;
5993 if (!memcg)
5994 return false;
5995
5996 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5997 if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
5998 return true;
5999
6000 return false;
6001}
6002
Johannes Weiner21afa382015-02-11 15:26:36 -08006003/* for remember boot option*/
6004#ifdef CONFIG_MEMCG_SWAP_ENABLED
6005static int really_do_swap_account __initdata = 1;
6006#else
6007static int really_do_swap_account __initdata;
6008#endif
6009
6010static int __init enable_swap_account(char *s)
6011{
6012 if (!strcmp(s, "1"))
6013 really_do_swap_account = 1;
6014 else if (!strcmp(s, "0"))
6015 really_do_swap_account = 0;
6016 return 1;
6017}
6018__setup("swapaccount=", enable_swap_account);
6019
Vladimir Davydov37e84352016-01-20 15:02:56 -08006020static u64 swap_current_read(struct cgroup_subsys_state *css,
6021 struct cftype *cft)
6022{
6023 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6024
6025 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
6026}
6027
6028static int swap_max_show(struct seq_file *m, void *v)
6029{
6030 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
6031 unsigned long max = READ_ONCE(memcg->swap.limit);
6032
6033 if (max == PAGE_COUNTER_MAX)
6034 seq_puts(m, "max\n");
6035 else
6036 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
6037
6038 return 0;
6039}
6040
6041static ssize_t swap_max_write(struct kernfs_open_file *of,
6042 char *buf, size_t nbytes, loff_t off)
6043{
6044 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6045 unsigned long max;
6046 int err;
6047
6048 buf = strstrip(buf);
6049 err = page_counter_memparse(buf, "max", &max);
6050 if (err)
6051 return err;
6052
6053 mutex_lock(&memcg_limit_mutex);
6054 err = page_counter_limit(&memcg->swap, max);
6055 mutex_unlock(&memcg_limit_mutex);
6056 if (err)
6057 return err;
6058
6059 return nbytes;
6060}
6061
6062static struct cftype swap_files[] = {
6063 {
6064 .name = "swap.current",
6065 .flags = CFTYPE_NOT_ON_ROOT,
6066 .read_u64 = swap_current_read,
6067 },
6068 {
6069 .name = "swap.max",
6070 .flags = CFTYPE_NOT_ON_ROOT,
6071 .seq_show = swap_max_show,
6072 .write = swap_max_write,
6073 },
6074 { } /* terminate */
6075};
6076
Johannes Weiner21afa382015-02-11 15:26:36 -08006077static struct cftype memsw_cgroup_files[] = {
6078 {
6079 .name = "memsw.usage_in_bytes",
6080 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6081 .read_u64 = mem_cgroup_read_u64,
6082 },
6083 {
6084 .name = "memsw.max_usage_in_bytes",
6085 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6086 .write = mem_cgroup_reset,
6087 .read_u64 = mem_cgroup_read_u64,
6088 },
6089 {
6090 .name = "memsw.limit_in_bytes",
6091 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6092 .write = mem_cgroup_write,
6093 .read_u64 = mem_cgroup_read_u64,
6094 },
6095 {
6096 .name = "memsw.failcnt",
6097 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6098 .write = mem_cgroup_reset,
6099 .read_u64 = mem_cgroup_read_u64,
6100 },
6101 { }, /* terminate */
6102};
6103
6104static int __init mem_cgroup_swap_init(void)
6105{
6106 if (!mem_cgroup_disabled() && really_do_swap_account) {
6107 do_swap_account = 1;
Vladimir Davydov37e84352016-01-20 15:02:56 -08006108 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
6109 swap_files));
Johannes Weiner21afa382015-02-11 15:26:36 -08006110 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
6111 memsw_cgroup_files));
6112 }
6113 return 0;
6114}
6115subsys_initcall(mem_cgroup_swap_init);
6116
6117#endif /* CONFIG_MEMCG_SWAP */