blob: ca83f3854e4fcf2272474d9200cebd30c75eba82 [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 Mortonbb4cc1a82013-09-24 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 Weiner71cd3112017-05-03 14:55:13 -0700103static const char *const mem_cgroup_lru_names[] = {
Sha Zhengju58cf1882013-02-22 16:32:05 -0800104 "inactive_anon",
105 "active_anon",
106 "inactive_file",
107 "active_file",
108 "unevictable",
109};
110
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700111#define THRESHOLDS_EVENTS_TARGET 128
112#define SOFTLIMIT_EVENTS_TARGET 1024
113#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700114
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700115/*
116 * Cgroups above their limits are maintained in a RB-Tree, independent of
117 * their hierarchy representation
118 */
119
Mel Gormanef8f2322016-07-28 15:46:05 -0700120struct mem_cgroup_tree_per_node {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700121 struct rb_root rb_root;
122 spinlock_t lock;
123};
124
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700125struct mem_cgroup_tree {
126 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
127};
128
129static struct mem_cgroup_tree soft_limit_tree __read_mostly;
130
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700131/* for OOM */
132struct mem_cgroup_eventfd_list {
133 struct list_head list;
134 struct eventfd_ctx *eventfd;
135};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800136
Tejun Heo79bd9812013-11-22 18:20:42 -0500137/*
138 * cgroup_event represents events which userspace want to receive.
139 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500140struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500141 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500142 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500143 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500144 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500145 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500146 * eventfd to signal userspace about the event.
147 */
148 struct eventfd_ctx *eventfd;
149 /*
150 * Each of these stored in a list by the cgroup.
151 */
152 struct list_head list;
153 /*
Tejun Heofba94802013-11-22 18:20:43 -0500154 * register_event() callback will be used to add new userspace
155 * waiter for changes related to this event. Use eventfd_signal()
156 * on eventfd to send notification to userspace.
157 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500158 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500159 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500160 /*
161 * unregister_event() callback will be called when userspace closes
162 * the eventfd or on cgroup removing. This callback must be set,
163 * if you want provide notification functionality.
164 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500165 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500166 struct eventfd_ctx *eventfd);
167 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500168 * All fields below needed to unregister event when
169 * userspace closes eventfd.
170 */
171 poll_table pt;
172 wait_queue_head_t *wqh;
Ingo Molnarac6424b2017-06-20 12:06:13 +0200173 wait_queue_entry_t wait;
Tejun Heo79bd9812013-11-22 18:20:42 -0500174 struct work_struct remove;
175};
176
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700177static void mem_cgroup_threshold(struct mem_cgroup *memcg);
178static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800179
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800180/* Stuffs for move charges at task migration. */
181/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800182 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800183 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800184#define MOVE_ANON 0x1U
185#define MOVE_FILE 0x2U
186#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800187
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800188/* "mc" and its members are protected by cgroup_mutex */
189static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800190 spinlock_t lock; /* for from, to */
Tejun Heo264a0ae2016-04-21 19:09:02 -0400191 struct mm_struct *mm;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800192 struct mem_cgroup *from;
193 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800194 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800195 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800196 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800197 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800198 struct task_struct *moving_task; /* a task moving charges */
199 wait_queue_head_t waitq; /* a waitq for other context */
200} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700201 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800202 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
203};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800204
Balbir Singh4e416952009-09-23 15:56:39 -0700205/*
206 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
207 * limit reclaim to prevent infinite loops, if they ever occur.
208 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700209#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700210#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700211
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800212enum charge_type {
213 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700214 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800215 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700216 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700217 NR_CHARGE_TYPE,
218};
219
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800220/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800221enum res_type {
222 _MEM,
223 _MEMSWAP,
224 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800225 _KMEM,
Vladimir Davydovd55f90b2016-01-20 15:02:44 -0800226 _TCP,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800227};
228
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700229#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
230#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800231#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700232/* Used for OOM nofiier */
233#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800234
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700235/* Some nice accessors for the vmpressure. */
236struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
237{
238 if (!memcg)
239 memcg = root_mem_cgroup;
240 return &memcg->vmpressure;
241}
242
243struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
244{
245 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
246}
247
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700248static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
249{
250 return (memcg == root_mem_cgroup);
251}
252
Johannes Weiner127424c2016-01-20 15:02:32 -0800253#ifndef CONFIG_SLOB
Glauber Costa55007d82012-12-18 14:22:38 -0800254/*
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800255 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800256 * The main reason for not using cgroup id for this:
257 * this works better in sparse environments, where we have a lot of memcgs,
258 * but only a few kmem-limited. Or also, if we have, for instance, 200
259 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
260 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800261 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800262 * The current size of the caches array is stored in memcg_nr_cache_ids. It
263 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800264 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800265static DEFINE_IDA(memcg_cache_ida);
266int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800267
Vladimir Davydov05257a12015-02-12 14:59:01 -0800268/* Protects memcg_nr_cache_ids */
269static DECLARE_RWSEM(memcg_cache_ids_sem);
270
271void memcg_get_cache_ids(void)
272{
273 down_read(&memcg_cache_ids_sem);
274}
275
276void memcg_put_cache_ids(void)
277{
278 up_read(&memcg_cache_ids_sem);
279}
280
Glauber Costa55007d82012-12-18 14:22:38 -0800281/*
282 * MIN_SIZE is different than 1, because we would like to avoid going through
283 * the alloc/free process all the time. In a small machine, 4 kmem-limited
284 * cgroups is a reasonable guess. In the future, it could be a parameter or
285 * tunable, but that is strictly not necessary.
286 *
Li Zefanb8627832013-09-23 16:56:47 +0800287 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800288 * this constant directly from cgroup, but it is understandable that this is
289 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800290 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800291 * increase ours as well if it increases.
292 */
293#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800294#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800295
Glauber Costad7f25f82012-12-18 14:22:40 -0800296/*
297 * A lot of the calls to the cache allocation functions are expected to be
298 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
299 * conditional to this static branch, we'll have to allow modules that does
300 * kmem_cache_alloc and the such to see this symbol as well
301 */
Johannes Weineref129472016-01-14 15:21:34 -0800302DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
Glauber Costad7f25f82012-12-18 14:22:40 -0800303EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800304
Tejun Heo17cc4df2017-02-22 15:41:36 -0800305struct workqueue_struct *memcg_kmem_cache_wq;
306
Johannes Weiner127424c2016-01-20 15:02:32 -0800307#endif /* !CONFIG_SLOB */
Glauber Costaa8964b92012-12-18 14:22:09 -0800308
Tejun Heoad7fa852015-05-27 20:00:02 -0400309/**
310 * mem_cgroup_css_from_page - css of the memcg associated with a page
311 * @page: page of interest
312 *
313 * If memcg is bound to the default hierarchy, css of the memcg associated
314 * with @page is returned. The returned css remains associated with @page
315 * until it is released.
316 *
317 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
318 * is returned.
Tejun Heoad7fa852015-05-27 20:00:02 -0400319 */
320struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
321{
322 struct mem_cgroup *memcg;
323
Tejun Heoad7fa852015-05-27 20:00:02 -0400324 memcg = page->mem_cgroup;
325
Tejun Heo9e10a132015-09-18 11:56:28 -0400326 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400327 memcg = root_mem_cgroup;
328
Tejun Heoad7fa852015-05-27 20:00:02 -0400329 return &memcg->css;
330}
331
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700332/**
333 * page_cgroup_ino - return inode number of the memcg a page is charged to
334 * @page: the page
335 *
336 * Look up the closest online ancestor of the memory cgroup @page is charged to
337 * and return its inode number or 0 if @page is not charged to any cgroup. It
338 * is safe to call this function without holding a reference to @page.
339 *
340 * Note, this function is inherently racy, because there is nothing to prevent
341 * the cgroup inode from getting torn down and potentially reallocated a moment
342 * after page_cgroup_ino() returns, so it only should be used by callers that
343 * do not care (such as procfs interfaces).
344 */
345ino_t page_cgroup_ino(struct page *page)
346{
347 struct mem_cgroup *memcg;
348 unsigned long ino = 0;
349
350 rcu_read_lock();
351 memcg = READ_ONCE(page->mem_cgroup);
352 while (memcg && !(memcg->css.flags & CSS_ONLINE))
353 memcg = parent_mem_cgroup(memcg);
354 if (memcg)
355 ino = cgroup_ino(memcg->css.cgroup);
356 rcu_read_unlock();
357 return ino;
358}
359
Mel Gormanef8f2322016-07-28 15:46:05 -0700360static struct mem_cgroup_per_node *
361mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700362{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700363 int nid = page_to_nid(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700364
Mel Gormanef8f2322016-07-28 15:46:05 -0700365 return memcg->nodeinfo[nid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700366}
367
Mel Gormanef8f2322016-07-28 15:46:05 -0700368static struct mem_cgroup_tree_per_node *
369soft_limit_tree_node(int nid)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700370{
Mel Gormanef8f2322016-07-28 15:46:05 -0700371 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700372}
373
Mel Gormanef8f2322016-07-28 15:46:05 -0700374static struct mem_cgroup_tree_per_node *
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700375soft_limit_tree_from_page(struct page *page)
376{
377 int nid = page_to_nid(page);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700378
Mel Gormanef8f2322016-07-28 15:46:05 -0700379 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700380}
381
Mel Gormanef8f2322016-07-28 15:46:05 -0700382static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz,
383 struct mem_cgroup_tree_per_node *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800384 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700385{
386 struct rb_node **p = &mctz->rb_root.rb_node;
387 struct rb_node *parent = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700388 struct mem_cgroup_per_node *mz_node;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700389
390 if (mz->on_tree)
391 return;
392
393 mz->usage_in_excess = new_usage_in_excess;
394 if (!mz->usage_in_excess)
395 return;
396 while (*p) {
397 parent = *p;
Mel Gormanef8f2322016-07-28 15:46:05 -0700398 mz_node = rb_entry(parent, struct mem_cgroup_per_node,
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700399 tree_node);
400 if (mz->usage_in_excess < mz_node->usage_in_excess)
401 p = &(*p)->rb_left;
402 /*
403 * We can't avoid mem cgroups that are over their soft
404 * limit by the same amount
405 */
406 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
407 p = &(*p)->rb_right;
408 }
409 rb_link_node(&mz->tree_node, parent, p);
410 rb_insert_color(&mz->tree_node, &mctz->rb_root);
411 mz->on_tree = true;
412}
413
Mel Gormanef8f2322016-07-28 15:46:05 -0700414static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
415 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700416{
417 if (!mz->on_tree)
418 return;
419 rb_erase(&mz->tree_node, &mctz->rb_root);
420 mz->on_tree = false;
421}
422
Mel Gormanef8f2322016-07-28 15:46:05 -0700423static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
424 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700425{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700426 unsigned long flags;
427
428 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700429 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700430 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700431}
432
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800433static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
434{
435 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700436 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800437 unsigned long excess = 0;
438
439 if (nr_pages > soft_limit)
440 excess = nr_pages - soft_limit;
441
442 return excess;
443}
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700444
445static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
446{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800447 unsigned long excess;
Mel Gormanef8f2322016-07-28 15:46:05 -0700448 struct mem_cgroup_per_node *mz;
449 struct mem_cgroup_tree_per_node *mctz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700450
Jianyu Zhane2318752014-06-06 14:38:20 -0700451 mctz = soft_limit_tree_from_page(page);
Laurent Dufourbfc72282017-03-09 16:17:06 -0800452 if (!mctz)
453 return;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700454 /*
455 * Necessary to update all ancestors when hierarchy is used.
456 * because their event counter is not touched.
457 */
458 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700459 mz = mem_cgroup_page_nodeinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800460 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700461 /*
462 * We have to update the tree if mz is on RB-tree or
463 * mem is over its softlimit.
464 */
465 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700466 unsigned long flags;
467
468 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700469 /* if on-tree, remove it */
470 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700471 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700472 /*
473 * Insert again. mz->usage_in_excess will be updated.
474 * If excess is 0, no tree ops.
475 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700476 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700477 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700478 }
479 }
480}
481
482static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
483{
Mel Gormanef8f2322016-07-28 15:46:05 -0700484 struct mem_cgroup_tree_per_node *mctz;
485 struct mem_cgroup_per_node *mz;
486 int nid;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700487
Jianyu Zhane2318752014-06-06 14:38:20 -0700488 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700489 mz = mem_cgroup_nodeinfo(memcg, nid);
490 mctz = soft_limit_tree_node(nid);
Laurent Dufourbfc72282017-03-09 16:17:06 -0800491 if (mctz)
492 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700493 }
494}
495
Mel Gormanef8f2322016-07-28 15:46:05 -0700496static struct mem_cgroup_per_node *
497__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700498{
499 struct rb_node *rightmost = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700500 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700501
502retry:
503 mz = NULL;
504 rightmost = rb_last(&mctz->rb_root);
505 if (!rightmost)
506 goto done; /* Nothing to reclaim from */
507
Mel Gormanef8f2322016-07-28 15:46:05 -0700508 mz = rb_entry(rightmost, struct mem_cgroup_per_node, tree_node);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700509 /*
510 * Remove the node now but someone else can add it back,
511 * we will to add it back at the end of reclaim to its correct
512 * position in the tree.
513 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700514 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800515 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400516 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700517 goto retry;
518done:
519 return mz;
520}
521
Mel Gormanef8f2322016-07-28 15:46:05 -0700522static struct mem_cgroup_per_node *
523mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700524{
Mel Gormanef8f2322016-07-28 15:46:05 -0700525 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700526
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700527 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700528 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700529 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700530 return mz;
531}
532
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700533/*
Greg Thelen484ebb32015-10-01 15:37:05 -0700534 * Return page count for single (non recursive) @memcg.
535 *
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700536 * Implementation Note: reading percpu statistics for memcg.
537 *
538 * Both of vmstat[] and percpu_counter has threshold and do periodic
539 * synchronization to implement "quick" read. There are trade-off between
540 * reading cost and precision of value. Then, we may have a chance to implement
Greg Thelen484ebb32015-10-01 15:37:05 -0700541 * a periodic synchronization of counter in memcg's counter.
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700542 *
543 * But this _read() function is used for user interface now. The user accounts
544 * memory usage by memory cgroup and he _always_ requires exact value because
545 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
546 * have to visit all online cpus and make sum. So, for now, unnecessary
547 * synchronization is not implemented. (just implemented for cpu hotplug)
548 *
549 * If there are kernel internal actions which can make use of some not-exact
550 * value, and reading all cpu value can be performance bottleneck in some
Greg Thelen484ebb32015-10-01 15:37:05 -0700551 * common workload, threshold and synchronization as vmstat[] should be
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700552 * implemented.
Matthias Kaehlcke04fecbf2017-09-06 16:22:09 -0700553 *
554 * The parameter idx can be of type enum memcg_event_item or vm_event_item.
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700555 */
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800556
Johannes Weinerccda7f42017-05-03 14:55:16 -0700557static unsigned long memcg_sum_events(struct mem_cgroup *memcg,
Matthias Kaehlcke04fecbf2017-09-06 16:22:09 -0700558 int event)
Johannes Weinere9f89742011-03-23 16:42:37 -0700559{
560 unsigned long val = 0;
561 int cpu;
562
Tejun Heo733a5722015-05-22 18:23:18 -0400563 for_each_possible_cpu(cpu)
Johannes Weinerdf0e53d2017-05-03 14:55:10 -0700564 val += per_cpu(memcg->stat->events[event], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700565 return val;
566}
567
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700568static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700569 struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800570 bool compound, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800571{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700572 /*
573 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
574 * counted as CACHE even if it's on ANON LRU.
575 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700576 if (PageAnon(page))
Johannes Weiner71cd3112017-05-03 14:55:13 -0700577 __this_cpu_add(memcg->stat->count[MEMCG_RSS], nr_pages);
Johannes Weiner9a4caf12017-05-03 14:52:45 -0700578 else {
Johannes Weiner71cd3112017-05-03 14:55:13 -0700579 __this_cpu_add(memcg->stat->count[MEMCG_CACHE], nr_pages);
Johannes Weiner9a4caf12017-05-03 14:52:45 -0700580 if (PageSwapBacked(page))
Johannes Weiner71cd3112017-05-03 14:55:13 -0700581 __this_cpu_add(memcg->stat->count[NR_SHMEM], nr_pages);
Johannes Weiner9a4caf12017-05-03 14:52:45 -0700582 }
Balaji Rao55e462b2008-05-01 04:35:12 -0700583
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800584 if (compound) {
585 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
Johannes Weiner71cd3112017-05-03 14:55:13 -0700586 __this_cpu_add(memcg->stat->count[MEMCG_RSS_HUGE], nr_pages);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800587 }
David Rientjesb070e652013-05-07 16:18:09 -0700588
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800589 /* pagein of a big page is an event. So, ignore page size */
590 if (nr_pages > 0)
Johannes Weinerdf0e53d2017-05-03 14:55:10 -0700591 __this_cpu_inc(memcg->stat->events[PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800592 else {
Johannes Weinerdf0e53d2017-05-03 14:55:10 -0700593 __this_cpu_inc(memcg->stat->events[PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800594 nr_pages = -nr_pages; /* for event */
595 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800596
Johannes Weiner13114712012-05-29 15:07:07 -0700597 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800598}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800599
Vladimir Davydov0a6b76d2016-03-17 14:18:42 -0700600unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
601 int nid, unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700602{
Michal Hockob4536f0c82017-01-10 16:58:04 -0800603 struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg);
Jianyu Zhane2318752014-06-06 14:38:20 -0700604 unsigned long nr = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -0700605 enum lru_list lru;
Ying Han889976d2011-05-26 16:25:33 -0700606
Jianyu Zhane2318752014-06-06 14:38:20 -0700607 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700608
Mel Gormanef8f2322016-07-28 15:46:05 -0700609 for_each_lru(lru) {
610 if (!(BIT(lru) & lru_mask))
611 continue;
Michal Hockob4536f0c82017-01-10 16:58:04 -0800612 nr += mem_cgroup_get_lru_size(lruvec, lru);
Jianyu Zhane2318752014-06-06 14:38:20 -0700613 }
614 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700615}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700616
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700617static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700618 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800619{
Jianyu Zhane2318752014-06-06 14:38:20 -0700620 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700621 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800622
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800623 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700624 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
625 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800626}
627
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800628static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
629 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800630{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700631 unsigned long val, next;
632
Johannes Weiner13114712012-05-29 15:07:07 -0700633 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700634 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700635 /* from time_after() in jiffies.h */
Michal Hocko6a1a8b82017-07-10 15:48:53 -0700636 if ((long)(next - val) < 0) {
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800637 switch (target) {
638 case MEM_CGROUP_TARGET_THRESH:
639 next = val + THRESHOLDS_EVENTS_TARGET;
640 break;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700641 case MEM_CGROUP_TARGET_SOFTLIMIT:
642 next = val + SOFTLIMIT_EVENTS_TARGET;
643 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800644 case MEM_CGROUP_TARGET_NUMAINFO:
645 next = val + NUMAINFO_EVENTS_TARGET;
646 break;
647 default:
648 break;
649 }
650 __this_cpu_write(memcg->stat->targets[target], next);
651 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700652 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800653 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800654}
655
656/*
657 * Check events in order.
658 *
659 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700660static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800661{
662 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800663 if (unlikely(mem_cgroup_event_ratelimit(memcg,
664 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700665 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800666 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800667
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700668 do_softlimit = mem_cgroup_event_ratelimit(memcg,
669 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -0700670#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800671 do_numainfo = mem_cgroup_event_ratelimit(memcg,
672 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -0700673#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800674 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700675 if (unlikely(do_softlimit))
676 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800677#if MAX_NUMNODES > 1
678 if (unlikely(do_numainfo))
679 atomic_inc(&memcg->numainfo_events);
680#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700681 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800682}
683
Balbir Singhcf475ad2008-04-29 01:00:16 -0700684struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800685{
Balbir Singh31a78f22008-09-28 23:09:31 +0100686 /*
687 * mm_update_next_owner() may clear mm->owner to NULL
688 * if it races with swapoff, page migration, etc.
689 * So this can be called with p == NULL.
690 */
691 if (unlikely(!p))
692 return NULL;
693
Tejun Heo073219e2014-02-08 10:36:58 -0500694 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800695}
Michal Hocko33398cf2015-09-08 15:01:02 -0700696EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800697
Johannes Weinerdf381972014-04-07 15:37:43 -0700698static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800699{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700700 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700701
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800702 rcu_read_lock();
703 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700704 /*
705 * Page cache insertions can happen withou an
706 * actual mm context, e.g. during disk probing
707 * on boot, loopback IO, acct() writes etc.
708 */
709 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700710 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700711 else {
712 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
713 if (unlikely(!memcg))
714 memcg = root_mem_cgroup;
715 }
Tejun Heoec903c02014-05-13 12:11:01 -0400716 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800717 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700718 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800719}
720
Johannes Weiner56600482012-01-12 17:17:59 -0800721/**
722 * mem_cgroup_iter - iterate over memory cgroup hierarchy
723 * @root: hierarchy root
724 * @prev: previously returned memcg, NULL on first invocation
725 * @reclaim: cookie for shared reclaim walks, NULL for full walks
726 *
727 * Returns references to children of the hierarchy below @root, or
728 * @root itself, or %NULL after a full round-trip.
729 *
730 * Caller must pass the return value in @prev on subsequent
731 * invocations for reference counting, or use mem_cgroup_iter_break()
732 * to cancel a hierarchy walk before the round-trip is complete.
733 *
734 * Reclaimers can specify a zone and a priority level in @reclaim to
735 * divide up the memcgs in the hierarchy among all concurrent
736 * reclaimers operating on the same zone and priority.
737 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700738struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800739 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700740 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700741{
Michal Hocko33398cf2015-09-08 15:01:02 -0700742 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800743 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800744 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800745 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700746
Andrew Morton694fbc02013-09-24 15:27:37 -0700747 if (mem_cgroup_disabled())
748 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800749
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700750 if (!root)
751 root = root_mem_cgroup;
752
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800753 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800754 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800755
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800756 if (!root->use_hierarchy && root != root_mem_cgroup) {
757 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800758 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700759 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800760 }
761
Michal Hocko542f85f2013-04-29 15:07:15 -0700762 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800763
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800764 if (reclaim) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700765 struct mem_cgroup_per_node *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800766
Mel Gormanef8f2322016-07-28 15:46:05 -0700767 mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800768 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700769
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800770 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700771 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800772
Vladimir Davydov6df38682015-12-29 14:54:10 -0800773 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -0700774 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800775 if (!pos || css_tryget(&pos->css))
776 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800777 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800778 * css reference reached zero, so iter->position will
779 * be cleared by ->css_released. However, we should not
780 * rely on this happening soon, because ->css_released
781 * is called from a work queue, and by busy-waiting we
782 * might block it. So we clear iter->position right
783 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800784 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800785 (void)cmpxchg(&iter->position, pos, NULL);
786 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800787 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800788
789 if (pos)
790 css = &pos->css;
791
792 for (;;) {
793 css = css_next_descendant_pre(css, &root->css);
794 if (!css) {
795 /*
796 * Reclaimers share the hierarchy walk, and a
797 * new one might jump in right at the end of
798 * the hierarchy - make sure they see at least
799 * one group and restart from the beginning.
800 */
801 if (!prev)
802 continue;
803 break;
804 }
805
806 /*
807 * Verify the css and acquire a reference. The root
808 * is provided by the caller, so we know it's alive
809 * and kicking, and don't take an extra reference.
810 */
811 memcg = mem_cgroup_from_css(css);
812
813 if (css == &root->css)
814 break;
815
Johannes Weiner0b8f73e2016-01-20 15:02:53 -0800816 if (css_tryget(css))
817 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800818
819 memcg = NULL;
820 }
821
822 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800823 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800824 * The position could have already been updated by a competing
825 * thread, so check that the value hasn't changed since we read
826 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800827 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800828 (void)cmpxchg(&iter->position, pos, memcg);
829
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800830 if (pos)
831 css_put(&pos->css);
832
833 if (!memcg)
834 iter->generation++;
835 else if (!prev)
836 reclaim->generation = iter->generation;
837 }
838
Michal Hocko542f85f2013-04-29 15:07:15 -0700839out_unlock:
840 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800841out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700842 if (prev && prev != root)
843 css_put(&prev->css);
844
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800845 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700846}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800847
Johannes Weiner56600482012-01-12 17:17:59 -0800848/**
849 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
850 * @root: hierarchy root
851 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
852 */
853void mem_cgroup_iter_break(struct mem_cgroup *root,
854 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800855{
856 if (!root)
857 root = root_mem_cgroup;
858 if (prev && prev != root)
859 css_put(&prev->css);
860}
861
Vladimir Davydov6df38682015-12-29 14:54:10 -0800862static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
863{
864 struct mem_cgroup *memcg = dead_memcg;
865 struct mem_cgroup_reclaim_iter *iter;
Mel Gormanef8f2322016-07-28 15:46:05 -0700866 struct mem_cgroup_per_node *mz;
867 int nid;
Vladimir Davydov6df38682015-12-29 14:54:10 -0800868 int i;
869
870 while ((memcg = parent_mem_cgroup(memcg))) {
871 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700872 mz = mem_cgroup_nodeinfo(memcg, nid);
873 for (i = 0; i <= DEF_PRIORITY; i++) {
874 iter = &mz->iter[i];
875 cmpxchg(&iter->position,
876 dead_memcg, NULL);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800877 }
878 }
879 }
880}
881
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700882/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800883 * Iteration constructs for visiting all cgroups (under a tree). If
884 * loops are exited prematurely (break), mem_cgroup_iter_break() must
885 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700886 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800887#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800888 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800889 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800890 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700891
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800892#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800893 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800894 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800895 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700896
Johannes Weiner925b7672012-01-12 17:18:15 -0800897/**
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700898 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
899 * @memcg: hierarchy root
900 * @fn: function to call for each task
901 * @arg: argument passed to @fn
902 *
903 * This function iterates over tasks attached to @memcg or to any of its
904 * descendants and calls @fn for each task. If @fn returns a non-zero
905 * value, the function breaks the iteration loop and returns the value.
906 * Otherwise, it will iterate over all tasks and return 0.
907 *
908 * This function must not be called for the root memory cgroup.
909 */
910int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
911 int (*fn)(struct task_struct *, void *), void *arg)
912{
913 struct mem_cgroup *iter;
914 int ret = 0;
915
916 BUG_ON(memcg == root_mem_cgroup);
917
918 for_each_mem_cgroup_tree(iter, memcg) {
919 struct css_task_iter it;
920 struct task_struct *task;
921
Tejun Heobc2fb7e2017-05-15 09:34:01 -0400922 css_task_iter_start(&iter->css, 0, &it);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700923 while (!ret && (task = css_task_iter_next(&it)))
924 ret = fn(task, arg);
925 css_task_iter_end(&it);
926 if (ret) {
927 mem_cgroup_iter_break(memcg, iter);
928 break;
929 }
930 }
931 return ret;
932}
933
934/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800935 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -0800936 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700937 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800938 *
939 * This function is only safe when following the LRU page isolation
940 * and putback protocol: the LRU lock must be held, and the page must
941 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -0700942 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700943struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat)
Minchan Kim3f58a822011-03-22 16:32:53 -0700944{
Mel Gormanef8f2322016-07-28 15:46:05 -0700945 struct mem_cgroup_per_node *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -0800946 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800947 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800948
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800949 if (mem_cgroup_disabled()) {
Mel Gorman599d0c92016-07-28 15:45:31 -0700950 lruvec = &pgdat->lruvec;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800951 goto out;
952 }
Christoph Lameterb69408e2008-10-18 20:26:14 -0700953
Johannes Weiner1306a852014-12-10 15:44:52 -0800954 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800955 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800956 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -0800957 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -0800958 */
Johannes Weiner29833312014-12-10 15:44:02 -0800959 if (!memcg)
960 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800961
Mel Gormanef8f2322016-07-28 15:46:05 -0700962 mz = mem_cgroup_page_nodeinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800963 lruvec = &mz->lruvec;
964out:
965 /*
966 * Since a node can be onlined after the mem_cgroup was created,
967 * we have to be prepared to initialize lruvec->zone here;
968 * and if offlined then reonlined, we need to reinitialize it.
969 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700970 if (unlikely(lruvec->pgdat != pgdat))
971 lruvec->pgdat = pgdat;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800972 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -0800973}
974
975/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700976 * mem_cgroup_update_lru_size - account for adding or removing an lru page
977 * @lruvec: mem_cgroup per zone lru vector
978 * @lru: index of lru list the page is sitting on
Michal Hockob4536f0c82017-01-10 16:58:04 -0800979 * @zid: zone id of the accounted pages
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700980 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -0800981 *
Hugh Dickinsca707232016-05-19 17:12:35 -0700982 * This function must be called under lru_lock, just before a page is added
983 * to or just after a page is removed from an lru list (that ordering being
984 * so as to allow it to check that lru_size 0 is consistent with list_empty).
Johannes Weiner925b7672012-01-12 17:18:15 -0800985 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700986void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
Michal Hockob4536f0c82017-01-10 16:58:04 -0800987 int zid, int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -0800988{
Mel Gormanef8f2322016-07-28 15:46:05 -0700989 struct mem_cgroup_per_node *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700990 unsigned long *lru_size;
Hugh Dickinsca707232016-05-19 17:12:35 -0700991 long size;
Johannes Weiner925b7672012-01-12 17:18:15 -0800992
993 if (mem_cgroup_disabled())
994 return;
995
Mel Gormanef8f2322016-07-28 15:46:05 -0700996 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
Michal Hockob4536f0c82017-01-10 16:58:04 -0800997 lru_size = &mz->lru_zone_size[zid][lru];
Hugh Dickinsca707232016-05-19 17:12:35 -0700998
999 if (nr_pages < 0)
1000 *lru_size += nr_pages;
1001
1002 size = *lru_size;
Michal Hockob4536f0c82017-01-10 16:58:04 -08001003 if (WARN_ONCE(size < 0,
1004 "%s(%p, %d, %d): lru_size %ld\n",
1005 __func__, lruvec, lru, nr_pages, size)) {
Hugh Dickinsca707232016-05-19 17:12:35 -07001006 VM_BUG_ON(1);
1007 *lru_size = 0;
1008 }
1009
1010 if (nr_pages > 0)
1011 *lru_size += nr_pages;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001012}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001013
Johannes Weiner2314b422014-12-10 15:44:33 -08001014bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001015{
Johannes Weiner2314b422014-12-10 15:44:33 -08001016 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001017 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001018 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001019
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001020 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001021 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001022 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001023 task_unlock(p);
1024 } else {
1025 /*
1026 * All threads may have already detached their mm's, but the oom
1027 * killer still needs to detect if they have already been oom
1028 * killed to prevent needlessly killing additional tasks.
1029 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001030 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001031 task_memcg = mem_cgroup_from_task(task);
1032 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001033 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001034 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001035 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1036 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001037 return ret;
1038}
1039
Johannes Weiner19942822011-02-01 15:52:43 -08001040/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001041 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001042 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001043 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001044 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001045 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001046 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001047static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001048{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001049 unsigned long margin = 0;
1050 unsigned long count;
1051 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001052
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001053 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001054 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001055 if (count < limit)
1056 margin = limit - count;
1057
Johannes Weiner7941d212016-01-14 15:21:23 -08001058 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001059 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001060 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001061 if (count <= limit)
1062 margin = min(margin, limit - count);
Li RongQingcbedbac2016-05-27 14:27:43 -07001063 else
1064 margin = 0;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001065 }
1066
1067 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001068}
1069
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001070/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001071 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001072 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001073 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1074 * moving cgroups. This is for waiting at high-memory pressure
1075 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001076 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001077static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001078{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001079 struct mem_cgroup *from;
1080 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001081 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001082 /*
1083 * Unlike task_move routines, we access mc.to, mc.from not under
1084 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1085 */
1086 spin_lock(&mc.lock);
1087 from = mc.from;
1088 to = mc.to;
1089 if (!from)
1090 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001091
Johannes Weiner2314b422014-12-10 15:44:33 -08001092 ret = mem_cgroup_is_descendant(from, memcg) ||
1093 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001094unlock:
1095 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001096 return ret;
1097}
1098
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001099static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001100{
1101 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001102 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001103 DEFINE_WAIT(wait);
1104 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1105 /* moving charge context might have finished. */
1106 if (mc.moving_task)
1107 schedule();
1108 finish_wait(&mc.waitq, &wait);
1109 return true;
1110 }
1111 }
1112 return false;
1113}
1114
Johannes Weiner71cd3112017-05-03 14:55:13 -07001115unsigned int memcg1_stats[] = {
1116 MEMCG_CACHE,
1117 MEMCG_RSS,
1118 MEMCG_RSS_HUGE,
1119 NR_SHMEM,
1120 NR_FILE_MAPPED,
1121 NR_FILE_DIRTY,
1122 NR_WRITEBACK,
1123 MEMCG_SWAP,
1124};
1125
1126static const char *const memcg1_stat_names[] = {
1127 "cache",
1128 "rss",
1129 "rss_huge",
1130 "shmem",
1131 "mapped_file",
1132 "dirty",
1133 "writeback",
1134 "swap",
1135};
1136
Sha Zhengju58cf1882013-02-22 16:32:05 -08001137#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001138/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001139 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001140 * @memcg: The memory cgroup that went over limit
1141 * @p: Task that is going to be killed
1142 *
1143 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1144 * enabled
1145 */
1146void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1147{
Sha Zhengju58cf1882013-02-22 16:32:05 -08001148 struct mem_cgroup *iter;
1149 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001150
Balbir Singhe2224322009-04-02 16:57:39 -07001151 rcu_read_lock();
1152
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001153 if (p) {
1154 pr_info("Task in ");
1155 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1156 pr_cont(" killed as a result of limit of ");
1157 } else {
1158 pr_info("Memory limit reached of cgroup ");
1159 }
1160
Tejun Heoe61734c2014-02-12 09:29:50 -05001161 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001162 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001163
Balbir Singhe2224322009-04-02 16:57:39 -07001164 rcu_read_unlock();
1165
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001166 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1167 K((u64)page_counter_read(&memcg->memory)),
1168 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1169 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1170 K((u64)page_counter_read(&memcg->memsw)),
1171 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1172 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1173 K((u64)page_counter_read(&memcg->kmem)),
1174 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001175
1176 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001177 pr_info("Memory cgroup stats for ");
1178 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001179 pr_cont(":");
1180
Johannes Weiner71cd3112017-05-03 14:55:13 -07001181 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
1182 if (memcg1_stats[i] == MEMCG_SWAP && !do_swap_account)
Sha Zhengju58cf1882013-02-22 16:32:05 -08001183 continue;
Johannes Weiner71cd3112017-05-03 14:55:13 -07001184 pr_cont(" %s:%luKB", memcg1_stat_names[i],
Johannes Weinerccda7f42017-05-03 14:55:16 -07001185 K(memcg_page_state(iter, memcg1_stats[i])));
Sha Zhengju58cf1882013-02-22 16:32:05 -08001186 }
1187
1188 for (i = 0; i < NR_LRU_LISTS; i++)
1189 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1190 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1191
1192 pr_cont("\n");
1193 }
Balbir Singhe2224322009-04-02 16:57:39 -07001194}
1195
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001196/*
1197 * This function returns the number of memcg under hierarchy tree. Returns
1198 * 1(self count) if no children.
1199 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001200static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001201{
1202 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001203 struct mem_cgroup *iter;
1204
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001205 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001206 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001207 return num;
1208}
1209
Balbir Singh6d61ef42009-01-07 18:08:06 -08001210/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001211 * Return the memory (and swap, if configured) limit for a memcg.
1212 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001213unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001214{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001215 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001216
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001217 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001218 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001219 unsigned long memsw_limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001220 unsigned long swap_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001221
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001222 memsw_limit = memcg->memsw.limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001223 swap_limit = memcg->swap.limit;
1224 swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1225 limit = min(limit + swap_limit, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001226 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001227 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001228}
1229
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001230static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001231 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001232{
David Rientjes6e0fc462015-09-08 15:00:36 -07001233 struct oom_control oc = {
1234 .zonelist = NULL,
1235 .nodemask = NULL,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001236 .memcg = memcg,
David Rientjes6e0fc462015-09-08 15:00:36 -07001237 .gfp_mask = gfp_mask,
1238 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001239 };
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001240 bool ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001241
Johannes Weinerdc564012015-06-24 16:57:19 -07001242 mutex_lock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001243 ret = out_of_memory(&oc);
Johannes Weinerdc564012015-06-24 16:57:19 -07001244 mutex_unlock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001245 return ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001246}
1247
Michele Curtiae6e71d2014-12-12 16:56:35 -08001248#if MAX_NUMNODES > 1
1249
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001250/**
1251 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001252 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001253 * @nid: the node ID to be checked.
1254 * @noswap : specify true here if the user wants flle only information.
1255 *
1256 * This function returns whether the specified memcg contains any
1257 * reclaimable pages on a node. Returns true if there are any reclaimable
1258 * pages in the node.
1259 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001260static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001261 int nid, bool noswap)
1262{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001263 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001264 return true;
1265 if (noswap || !total_swap_pages)
1266 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001267 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001268 return true;
1269 return false;
1270
1271}
Ying Han889976d2011-05-26 16:25:33 -07001272
1273/*
1274 * Always updating the nodemask is not very good - even if we have an empty
1275 * list or the wrong list here, we can start from some node and traverse all
1276 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1277 *
1278 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001279static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001280{
1281 int nid;
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001282 /*
1283 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1284 * pagein/pageout changes since the last update.
1285 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001286 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001287 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001288 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001289 return;
1290
Ying Han889976d2011-05-26 16:25:33 -07001291 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001292 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001293
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001294 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001295
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001296 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1297 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001298 }
KAMEZAWA Hiroyuki453a9bf32011-07-08 15:39:43 -07001299
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001300 atomic_set(&memcg->numainfo_events, 0);
1301 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001302}
1303
1304/*
1305 * Selecting a node where we start reclaim from. Because what we need is just
1306 * reducing usage counter, start from anywhere is O,K. Considering
1307 * memory reclaim from current node, there are pros. and cons.
1308 *
1309 * Freeing memory from current node means freeing memory from a node which
1310 * we'll use or we've used. So, it may make LRU bad. And if several threads
1311 * hit limits, it will see a contention on a node. But freeing from remote
1312 * node means more costs for memory reclaim because of memory latency.
1313 *
1314 * Now, we use round-robin. Better algorithm is welcomed.
1315 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001316int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001317{
1318 int node;
1319
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001320 mem_cgroup_may_update_nodemask(memcg);
1321 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001322
Andrew Morton0edaf862016-05-19 17:10:58 -07001323 node = next_node_in(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001324 /*
Michal Hockofda3d692016-05-19 17:11:34 -07001325 * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1326 * last time it really checked all the LRUs due to rate limiting.
1327 * Fallback to the current node in that case for simplicity.
Ying Han889976d2011-05-26 16:25:33 -07001328 */
1329 if (unlikely(node == MAX_NUMNODES))
1330 node = numa_node_id();
1331
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001332 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001333 return node;
1334}
Ying Han889976d2011-05-26 16:25:33 -07001335#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001336int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001337{
1338 return 0;
1339}
1340#endif
1341
Andrew Morton0608f432013-09-24 15:27:41 -07001342static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
Mel Gormanef8f2322016-07-28 15:46:05 -07001343 pg_data_t *pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001344 gfp_t gfp_mask,
1345 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001346{
Andrew Morton0608f432013-09-24 15:27:41 -07001347 struct mem_cgroup *victim = NULL;
1348 int total = 0;
1349 int loop = 0;
1350 unsigned long excess;
1351 unsigned long nr_scanned;
1352 struct mem_cgroup_reclaim_cookie reclaim = {
Mel Gormanef8f2322016-07-28 15:46:05 -07001353 .pgdat = pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001354 .priority = 0,
1355 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001356
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001357 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001358
Andrew Morton0608f432013-09-24 15:27:41 -07001359 while (1) {
1360 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1361 if (!victim) {
1362 loop++;
1363 if (loop >= 2) {
1364 /*
1365 * If we have not been able to reclaim
1366 * anything, it might because there are
1367 * no reclaimable pages under this hierarchy
1368 */
1369 if (!total)
1370 break;
1371 /*
1372 * We want to do more targeted reclaim.
1373 * excess >> 2 is not to excessive so as to
1374 * reclaim too much, nor too less that we keep
1375 * coming back to reclaim from this cgroup
1376 */
1377 if (total >= (excess >> 2) ||
1378 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1379 break;
1380 }
1381 continue;
1382 }
Mel Gormana9dd0a82016-07-28 15:46:02 -07001383 total += mem_cgroup_shrink_node(victim, gfp_mask, false,
Mel Gormanef8f2322016-07-28 15:46:05 -07001384 pgdat, &nr_scanned);
Andrew Morton0608f432013-09-24 15:27:41 -07001385 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001386 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001387 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001388 }
Andrew Morton0608f432013-09-24 15:27:41 -07001389 mem_cgroup_iter_break(root_memcg, victim);
1390 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001391}
1392
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001393#ifdef CONFIG_LOCKDEP
1394static struct lockdep_map memcg_oom_lock_dep_map = {
1395 .name = "memcg_oom_lock",
1396};
1397#endif
1398
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001399static DEFINE_SPINLOCK(memcg_oom_lock);
1400
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001401/*
1402 * Check OOM-Killer is already running under our hierarchy.
1403 * If someone is running, return false.
1404 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001405static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001406{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001407 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001408
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001409 spin_lock(&memcg_oom_lock);
1410
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001411 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001412 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001413 /*
1414 * this subtree of our hierarchy is already locked
1415 * so we cannot give a lock.
1416 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001417 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001418 mem_cgroup_iter_break(memcg, iter);
1419 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001420 } else
1421 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001422 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001423
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001424 if (failed) {
1425 /*
1426 * OK, we failed to lock the whole subtree so we have
1427 * to clean up what we set up to the failing subtree
1428 */
1429 for_each_mem_cgroup_tree(iter, memcg) {
1430 if (iter == failed) {
1431 mem_cgroup_iter_break(memcg, iter);
1432 break;
1433 }
1434 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001435 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001436 } else
1437 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001438
1439 spin_unlock(&memcg_oom_lock);
1440
1441 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001442}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001443
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001444static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001445{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001446 struct mem_cgroup *iter;
1447
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001448 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001449 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001450 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001451 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001452 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001453}
1454
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001455static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001456{
1457 struct mem_cgroup *iter;
1458
Tejun Heoc2b42d32015-06-24 16:58:23 -07001459 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001460 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001461 iter->under_oom++;
1462 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_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001466{
1467 struct mem_cgroup *iter;
1468
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001469 /*
1470 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001471 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001472 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001473 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001474 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001475 if (iter->under_oom > 0)
1476 iter->under_oom--;
1477 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001478}
1479
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001480static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1481
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001482struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001483 struct mem_cgroup *memcg;
Ingo Molnarac6424b2017-06-20 12:06:13 +02001484 wait_queue_entry_t wait;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001485};
1486
Ingo Molnarac6424b2017-06-20 12:06:13 +02001487static int memcg_oom_wake_function(wait_queue_entry_t *wait,
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001488 unsigned mode, int sync, void *arg)
1489{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001490 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1491 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001492 struct oom_wait_info *oom_wait_info;
1493
1494 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001495 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001496
Johannes Weiner2314b422014-12-10 15:44:33 -08001497 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1498 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001499 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001500 return autoremove_wake_function(wait, mode, sync, arg);
1501}
1502
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001503static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001504{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001505 /*
1506 * For the following lockless ->under_oom test, the only required
1507 * guarantee is that it must see the state asserted by an OOM when
1508 * this function is called as a result of userland actions
1509 * triggered by the notification of the OOM. This is trivially
1510 * achieved by invoking mem_cgroup_mark_under_oom() before
1511 * triggering notification.
1512 */
1513 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001514 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001515}
1516
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001517static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001518{
Andrew Mortond0db7af2016-06-08 15:33:47 -07001519 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001520 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001521 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001522 * We are in the middle of the charge context here, so we
1523 * don't want to block when potentially sitting on a callstack
1524 * that holds all kinds of filesystem and mm locks.
1525 *
1526 * Also, the caller may handle a failed allocation gracefully
1527 * (like optional page cache readahead) and so an OOM killer
1528 * invocation might not even be necessary.
1529 *
1530 * That's why we don't do anything here except remember the
1531 * OOM context and then deal with it at the end of the page
1532 * fault when the stack is unwound, the locks are released,
1533 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001534 */
Johannes Weiner49426422013-10-16 13:46:59 -07001535 css_get(&memcg->css);
Tejun Heo626ebc42015-11-05 18:46:09 -08001536 current->memcg_in_oom = memcg;
1537 current->memcg_oom_gfp_mask = mask;
1538 current->memcg_oom_order = order;
Johannes Weiner49426422013-10-16 13:46:59 -07001539}
1540
1541/**
1542 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1543 * @handle: actually kill/wait or just clean up the OOM state
1544 *
1545 * This has to be called at the end of a page fault if the memcg OOM
1546 * handler was enabled.
1547 *
1548 * Memcg supports userspace OOM handling where failed allocations must
1549 * sleep on a waitqueue until the userspace task resolves the
1550 * situation. Sleeping directly in the charge context with all kinds
1551 * of locks held is not a good idea, instead we remember an OOM state
1552 * in the task and mem_cgroup_oom_synchronize() has to be called at
1553 * the end of the page fault to complete the OOM handling.
1554 *
1555 * Returns %true if an ongoing memcg OOM situation was detected and
1556 * completed, %false otherwise.
1557 */
1558bool mem_cgroup_oom_synchronize(bool handle)
1559{
Tejun Heo626ebc42015-11-05 18:46:09 -08001560 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001561 struct oom_wait_info owait;
1562 bool locked;
1563
1564 /* OOM is global, do not handle */
1565 if (!memcg)
1566 return false;
1567
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001568 if (!handle)
Johannes Weiner49426422013-10-16 13:46:59 -07001569 goto cleanup;
1570
1571 owait.memcg = memcg;
1572 owait.wait.flags = 0;
1573 owait.wait.func = memcg_oom_wake_function;
1574 owait.wait.private = current;
Ingo Molnar2055da92017-06-20 12:06:46 +02001575 INIT_LIST_HEAD(&owait.wait.entry);
Johannes Weiner49426422013-10-16 13:46:59 -07001576
1577 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001578 mem_cgroup_mark_under_oom(memcg);
1579
1580 locked = mem_cgroup_oom_trylock(memcg);
1581
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001582 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001583 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001584
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001585 if (locked && !memcg->oom_kill_disable) {
1586 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001587 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001588 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1589 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001590 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001591 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001592 mem_cgroup_unmark_under_oom(memcg);
1593 finish_wait(&memcg_oom_waitq, &owait.wait);
1594 }
1595
1596 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001597 mem_cgroup_oom_unlock(memcg);
1598 /*
1599 * There is no guarantee that an OOM-lock contender
1600 * sees the wakeups triggered by the OOM kill
1601 * uncharges. Wake any sleepers explicitely.
1602 */
1603 memcg_oom_recover(memcg);
1604 }
Johannes Weiner49426422013-10-16 13:46:59 -07001605cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001606 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001607 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001608 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001609}
1610
Johannes Weinerd7365e72014-10-29 14:50:48 -07001611/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001612 * lock_page_memcg - lock a page->mem_cgroup binding
1613 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001614 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001615 * This function protects unlocked LRU pages from being moved to
Johannes Weiner739f79f2017-08-18 15:15:48 -07001616 * another cgroup.
1617 *
1618 * It ensures lifetime of the returned memcg. Caller is responsible
1619 * for the lifetime of the page; __unlock_page_memcg() is available
1620 * when @page might get freed inside the locked section.
Balbir Singhd69b0422009-06-17 16:26:34 -07001621 */
Johannes Weiner739f79f2017-08-18 15:15:48 -07001622struct mem_cgroup *lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001623{
1624 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001625 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001626
Johannes Weiner6de22612015-02-11 15:25:01 -08001627 /*
1628 * The RCU lock is held throughout the transaction. The fast
1629 * path can get away without acquiring the memcg->move_lock
1630 * because page moving starts with an RCU grace period.
Johannes Weiner739f79f2017-08-18 15:15:48 -07001631 *
1632 * The RCU lock also protects the memcg from being freed when
1633 * the page state that is going to change is the only thing
1634 * preventing the page itself from being freed. E.g. writeback
1635 * doesn't hold a page reference and relies on PG_writeback to
1636 * keep off truncation, migration and so forth.
1637 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001638 rcu_read_lock();
1639
1640 if (mem_cgroup_disabled())
Johannes Weiner739f79f2017-08-18 15:15:48 -07001641 return NULL;
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 Weiner739f79f2017-08-18 15:15:48 -07001645 return NULL;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001646
Qiang Huangbdcbb652014-06-04 16:08:21 -07001647 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner739f79f2017-08-18 15:15:48 -07001648 return memcg;
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 Weiner739f79f2017-08-18 15:15:48 -07001664 return memcg;
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 Weiner739f79f2017-08-18 15:15:48 -07001669 * __unlock_page_memcg - unlock and unpin a memcg
1670 * @memcg: the memcg
1671 *
1672 * Unlock and unpin a memcg returned by lock_page_memcg().
Johannes Weinerd7365e72014-10-29 14:50:48 -07001673 */
Johannes Weiner739f79f2017-08-18 15:15:48 -07001674void __unlock_page_memcg(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001675{
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 Weiner739f79f2017-08-18 15:15:48 -07001687
1688/**
1689 * unlock_page_memcg - unlock a page->mem_cgroup binding
1690 * @page: the page
1691 */
1692void unlock_page_memcg(struct page *page)
1693{
1694 __unlock_page_memcg(page->mem_cgroup);
1695}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001696EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001697
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001698/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001699 * size of first charge trial. "32" comes from vmscan.c's magic value.
1700 * TODO: maybe necessary to use big numbers in big irons.
1701 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001702#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001703struct memcg_stock_pcp {
1704 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001705 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001706 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001707 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001708#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001709};
1710static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001711static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001712
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001713/**
1714 * consume_stock: Try to consume stocked charge on this cpu.
1715 * @memcg: memcg to consume from.
1716 * @nr_pages: how many pages to charge.
1717 *
1718 * The charges will only happen if @memcg matches the current cpu's memcg
1719 * stock, and at least @nr_pages are available in that stock. Failure to
1720 * service an allocation will refill the stock.
1721 *
1722 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001723 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001724static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001725{
1726 struct memcg_stock_pcp *stock;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001727 unsigned long flags;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001728 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001729
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001730 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001731 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001732
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001733 local_irq_save(flags);
1734
1735 stock = this_cpu_ptr(&memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001736 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001737 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001738 ret = true;
1739 }
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001740
1741 local_irq_restore(flags);
1742
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001743 return ret;
1744}
1745
1746/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001747 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001748 */
1749static void drain_stock(struct memcg_stock_pcp *stock)
1750{
1751 struct mem_cgroup *old = stock->cached;
1752
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001753 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001754 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001755 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001756 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001757 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001758 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001759 }
1760 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001761}
1762
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001763static void drain_local_stock(struct work_struct *dummy)
1764{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001765 struct memcg_stock_pcp *stock;
1766 unsigned long flags;
1767
1768 local_irq_save(flags);
1769
1770 stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001771 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001772 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001773
1774 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001775}
1776
1777/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001778 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001779 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001780 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001781static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001782{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001783 struct memcg_stock_pcp *stock;
1784 unsigned long flags;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001785
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001786 local_irq_save(flags);
1787
1788 stock = this_cpu_ptr(&memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001789 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001790 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001791 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001792 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001793 stock->nr_pages += nr_pages;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001794
Roman Gushchin475d0482017-09-08 16:13:09 -07001795 if (stock->nr_pages > CHARGE_BATCH)
1796 drain_stock(stock);
1797
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07001798 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001799}
1800
1801/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001802 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001803 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001804 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001805static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001806{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001807 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001808
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001809 /* If someone's already draining, avoid adding running more workers. */
1810 if (!mutex_trylock(&percpu_charge_mutex))
1811 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001812 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001813 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001814 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001815 for_each_online_cpu(cpu) {
1816 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001817 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001818
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001819 memcg = stock->cached;
1820 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001821 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001822 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001823 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001824 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1825 if (cpu == curcpu)
1826 drain_local_stock(&stock->work);
1827 else
1828 schedule_work_on(cpu, &stock->work);
1829 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001830 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001831 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001832 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001833 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001834}
1835
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01001836static int memcg_hotplug_cpu_dead(unsigned int cpu)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001837{
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001838 struct memcg_stock_pcp *stock;
1839
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001840 stock = &per_cpu(memcg_stock, cpu);
1841 drain_stock(stock);
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01001842 return 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001843}
1844
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001845static void reclaim_high(struct mem_cgroup *memcg,
1846 unsigned int nr_pages,
1847 gfp_t gfp_mask)
1848{
1849 do {
1850 if (page_counter_read(&memcg->memory) <= memcg->high)
1851 continue;
Johannes Weiner31176c72017-05-03 14:55:07 -07001852 mem_cgroup_event(memcg, MEMCG_HIGH);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001853 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1854 } while ((memcg = parent_mem_cgroup(memcg)));
1855}
1856
1857static void high_work_func(struct work_struct *work)
1858{
1859 struct mem_cgroup *memcg;
1860
1861 memcg = container_of(work, struct mem_cgroup, high_work);
1862 reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1863}
1864
Tejun Heob23afb92015-11-05 18:46:11 -08001865/*
1866 * Scheduled by try_charge() to be executed from the userland return path
1867 * and reclaims memory over the high limit.
1868 */
1869void mem_cgroup_handle_over_high(void)
1870{
1871 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001872 struct mem_cgroup *memcg;
Tejun Heob23afb92015-11-05 18:46:11 -08001873
1874 if (likely(!nr_pages))
1875 return;
1876
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001877 memcg = get_mem_cgroup_from_mm(current->mm);
1878 reclaim_high(memcg, nr_pages, GFP_KERNEL);
Tejun Heob23afb92015-11-05 18:46:11 -08001879 css_put(&memcg->css);
1880 current->memcg_nr_pages_over_high = 0;
1881}
1882
Johannes Weiner00501b52014-08-08 14:19:20 -07001883static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1884 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001885{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001886 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001887 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001888 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001889 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001890 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001891 bool may_swap = true;
1892 bool drained = false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001893
Johannes Weinerce00a962014-09-05 08:43:57 -04001894 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08001895 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001896retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001897 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08001898 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001899
Johannes Weiner7941d212016-01-14 15:21:23 -08001900 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08001901 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1902 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001903 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08001904 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001905 page_counter_uncharge(&memcg->memsw, batch);
1906 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001907 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001908 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001909 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001910 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001911
Johannes Weiner6539cc02014-08-06 16:05:42 -07001912 if (batch > nr_pages) {
1913 batch = nr_pages;
1914 goto retry;
1915 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001916
Johannes Weiner06b078f2014-08-06 16:05:44 -07001917 /*
1918 * Unlike in global OOM situations, memcg is not in a physical
1919 * memory shortage. Allow dying and OOM-killed tasks to
1920 * bypass the last charges so that they can exit quickly and
1921 * free their memory.
1922 */
Michal Hockoda99ecf2017-09-06 16:24:53 -07001923 if (unlikely(tsk_is_oom_victim(current) ||
Johannes Weiner06b078f2014-08-06 16:05:44 -07001924 fatal_signal_pending(current) ||
1925 current->flags & PF_EXITING))
Tejun Heo10d53c72015-11-05 18:46:17 -08001926 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001927
Johannes Weiner89a28482016-10-27 17:46:56 -07001928 /*
1929 * Prevent unbounded recursion when reclaim operations need to
1930 * allocate memory. This might exceed the limits temporarily,
1931 * but we prefer facilitating memory reclaim and getting back
1932 * under the limit over triggering OOM kills in these cases.
1933 */
1934 if (unlikely(current->flags & PF_MEMALLOC))
1935 goto force;
1936
Johannes Weiner06b078f2014-08-06 16:05:44 -07001937 if (unlikely(task_in_memcg_oom(current)))
1938 goto nomem;
1939
Mel Gormand0164ad2015-11-06 16:28:21 -08001940 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001941 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001942
Johannes Weiner31176c72017-05-03 14:55:07 -07001943 mem_cgroup_event(mem_over_limit, MEMCG_MAX);
Johannes Weiner241994ed2015-02-11 15:26:06 -08001944
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001945 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1946 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07001947
Johannes Weiner61e02c72014-08-06 16:08:16 -07001948 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07001949 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07001950
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001951 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001952 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001953 drained = true;
1954 goto retry;
1955 }
1956
Johannes Weiner28c34c22014-08-06 16:05:47 -07001957 if (gfp_mask & __GFP_NORETRY)
1958 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001959 /*
1960 * Even though the limit is exceeded at this point, reclaim
1961 * may have been able to free some pages. Retry the charge
1962 * before killing the task.
1963 *
1964 * Only for regular pages, though: huge pages are rather
1965 * unlikely to succeed so close to the limit, and we fall back
1966 * to regular pages anyway in case of failure.
1967 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07001968 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001969 goto retry;
1970 /*
1971 * At task move, charge accounts can be doubly counted. So, it's
1972 * better to wait until the end of task_move if something is going on.
1973 */
1974 if (mem_cgroup_wait_acct_move(mem_over_limit))
1975 goto retry;
1976
Johannes Weiner9b130612014-08-06 16:05:51 -07001977 if (nr_retries--)
1978 goto retry;
1979
Johannes Weiner06b078f2014-08-06 16:05:44 -07001980 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08001981 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001982
Johannes Weiner6539cc02014-08-06 16:05:42 -07001983 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08001984 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001985
Johannes Weiner31176c72017-05-03 14:55:07 -07001986 mem_cgroup_event(mem_over_limit, MEMCG_OOM);
Johannes Weiner241994ed2015-02-11 15:26:06 -08001987
Jerome Marchand3608de02015-11-05 18:47:29 -08001988 mem_cgroup_oom(mem_over_limit, gfp_mask,
1989 get_order(nr_pages * PAGE_SIZE));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001990nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07001991 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07001992 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08001993force:
1994 /*
1995 * The allocation either can't fail or will lead to more memory
1996 * being freed very soon. Allow memory usage go over the limit
1997 * temporarily by force charging it.
1998 */
1999 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002000 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08002001 page_counter_charge(&memcg->memsw, nr_pages);
2002 css_get_many(&memcg->css, nr_pages);
2003
2004 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002005
2006done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002007 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002008 if (batch > nr_pages)
2009 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002010
Johannes Weiner241994ed2015-02-11 15:26:06 -08002011 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002012 * If the hierarchy is above the normal consumption range, schedule
2013 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002014 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002015 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2016 * not recorded as it most likely matches current's and won't
2017 * change in the meantime. As high limit is checked again before
2018 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994ed2015-02-11 15:26:06 -08002019 */
2020 do {
Tejun Heob23afb92015-11-05 18:46:11 -08002021 if (page_counter_read(&memcg->memory) > memcg->high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002022 /* Don't bother a random interrupted task */
2023 if (in_interrupt()) {
2024 schedule_work(&memcg->high_work);
2025 break;
2026 }
Vladimir Davydov9516a182015-12-11 13:40:24 -08002027 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002028 set_notify_resume(current);
2029 break;
2030 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08002031 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002032
2033 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002034}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002035
Johannes Weiner00501b52014-08-08 14:19:20 -07002036static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002037{
Johannes Weinerce00a962014-09-05 08:43:57 -04002038 if (mem_cgroup_is_root(memcg))
2039 return;
2040
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002041 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002042 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002043 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002044
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002045 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002046}
2047
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002048static void lock_page_lru(struct page *page, int *isolated)
2049{
2050 struct zone *zone = page_zone(page);
2051
Mel Gormana52633d2016-07-28 15:45:28 -07002052 spin_lock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002053 if (PageLRU(page)) {
2054 struct lruvec *lruvec;
2055
Mel Gorman599d0c92016-07-28 15:45:31 -07002056 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002057 ClearPageLRU(page);
2058 del_page_from_lru_list(page, lruvec, page_lru(page));
2059 *isolated = 1;
2060 } else
2061 *isolated = 0;
2062}
2063
2064static void unlock_page_lru(struct page *page, int isolated)
2065{
2066 struct zone *zone = page_zone(page);
2067
2068 if (isolated) {
2069 struct lruvec *lruvec;
2070
Mel Gorman599d0c92016-07-28 15:45:31 -07002071 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002072 VM_BUG_ON_PAGE(PageLRU(page), page);
2073 SetPageLRU(page);
2074 add_page_to_lru_list(page, lruvec, page_lru(page));
2075 }
Mel Gormana52633d2016-07-28 15:45:28 -07002076 spin_unlock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002077}
2078
Johannes Weiner00501b52014-08-08 14:19:20 -07002079static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002080 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002081{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002082 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002083
Johannes Weiner1306a852014-12-10 15:44:52 -08002084 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002085
2086 /*
2087 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2088 * may already be on some other mem_cgroup's LRU. Take care of it.
2089 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002090 if (lrucare)
2091 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002092
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002093 /*
2094 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002095 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002096 *
2097 * - the page is uncharged
2098 *
2099 * - the page is off-LRU
2100 *
2101 * - an anonymous fault has exclusive page access, except for
2102 * a locked page table
2103 *
2104 * - a page cache insertion, a swapin fault, or a migration
2105 * have the page locked
2106 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002107 page->mem_cgroup = memcg;
Hugh Dickins3be912772008-02-07 00:14:19 -08002108
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002109 if (lrucare)
2110 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002111}
2112
Johannes Weiner127424c2016-01-20 15:02:32 -08002113#ifndef CONFIG_SLOB
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002114static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002115{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002116 int id, size;
2117 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002118
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002119 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002120 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2121 if (id < 0)
2122 return id;
2123
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002124 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002125 return id;
2126
2127 /*
2128 * There's no space for the new id in memcg_caches arrays,
2129 * so we have to grow them.
2130 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002131 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002132
2133 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002134 if (size < MEMCG_CACHES_MIN_SIZE)
2135 size = MEMCG_CACHES_MIN_SIZE;
2136 else if (size > MEMCG_CACHES_MAX_SIZE)
2137 size = MEMCG_CACHES_MAX_SIZE;
2138
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002139 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002140 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002141 err = memcg_update_all_list_lrus(size);
2142 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002143 memcg_nr_cache_ids = size;
2144
2145 up_write(&memcg_cache_ids_sem);
2146
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002147 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002148 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002149 return err;
2150 }
2151 return id;
2152}
2153
2154static void memcg_free_cache_id(int id)
2155{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002156 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002157}
2158
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002159struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002160 struct mem_cgroup *memcg;
2161 struct kmem_cache *cachep;
2162 struct work_struct work;
2163};
2164
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002165static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002166{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002167 struct memcg_kmem_cache_create_work *cw =
2168 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002169 struct mem_cgroup *memcg = cw->memcg;
2170 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002171
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002172 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002173
Vladimir Davydov5722d092014-04-07 15:39:24 -07002174 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002175 kfree(cw);
2176}
2177
2178/*
2179 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002180 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002181static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2182 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002183{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002184 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002185
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002186 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002187 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002188 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002189
2190 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002191
2192 cw->memcg = memcg;
2193 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002194 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002195
Tejun Heo17cc4df2017-02-22 15:41:36 -08002196 queue_work(memcg_kmem_cache_wq, &cw->work);
Glauber Costad7f25f82012-12-18 14:22:40 -08002197}
2198
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002199static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2200 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002201{
2202 /*
2203 * We need to stop accounting when we kmalloc, because if the
2204 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002205 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002206 *
2207 * However, it is better to enclose the whole function. Depending on
2208 * the debugging options enabled, INIT_WORK(), for instance, can
2209 * trigger an allocation. This too, will make us recurse. Because at
2210 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2211 * the safest choice is to do it like this, wrapping the whole function.
2212 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002213 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002214 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002215 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002216}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002217
Vladimir Davydov45264772016-07-26 15:24:21 -07002218static inline bool memcg_kmem_bypass(void)
2219{
2220 if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
2221 return true;
2222 return false;
2223}
2224
2225/**
2226 * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2227 * @cachep: the original global kmem cache
2228 *
Glauber Costad7f25f82012-12-18 14:22:40 -08002229 * Return the kmem_cache we're supposed to use for a slab allocation.
2230 * We try to use the current memcg's version of the cache.
2231 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002232 * If the cache does not exist yet, if we are the first user of it, we
2233 * create it asynchronously in a workqueue and let the current allocation
2234 * go through with the original cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002235 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002236 * This function takes a reference to the cache it returns to assure it
2237 * won't get destroyed while we are working with it. Once the caller is
2238 * done with it, memcg_kmem_put_cache() must be called to release the
2239 * reference.
Glauber Costad7f25f82012-12-18 14:22:40 -08002240 */
Vladimir Davydov45264772016-07-26 15:24:21 -07002241struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002242{
2243 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002244 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002245 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002246
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002247 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002248
Vladimir Davydov45264772016-07-26 15:24:21 -07002249 if (memcg_kmem_bypass())
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002250 return cachep;
2251
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002252 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002253 return cachep;
2254
Vladimir Davydov8135be52014-12-12 16:56:38 -08002255 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002256 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002257 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002258 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002259
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002260 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002261 if (likely(memcg_cachep))
2262 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002263
2264 /*
2265 * If we are in a safe context (can wait, and not in interrupt
2266 * context), we could be be predictable and return right away.
2267 * This would guarantee that the allocation being performed
2268 * already belongs in the new cache.
2269 *
2270 * However, there are some clashes that can arrive from locking.
2271 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002272 * memcg_create_kmem_cache, this means no further allocation
2273 * could happen with the slab_mutex held. So it's better to
2274 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002275 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002276 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002277out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002278 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002279 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002280}
Glauber Costad7f25f82012-12-18 14:22:40 -08002281
Vladimir Davydov45264772016-07-26 15:24:21 -07002282/**
2283 * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2284 * @cachep: the cache returned by memcg_kmem_get_cache
2285 */
2286void memcg_kmem_put_cache(struct kmem_cache *cachep)
Vladimir Davydov8135be52014-12-12 16:56:38 -08002287{
2288 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002289 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002290}
2291
Vladimir Davydov45264772016-07-26 15:24:21 -07002292/**
2293 * memcg_kmem_charge: charge a kmem page
2294 * @page: page to charge
2295 * @gfp: reclaim mode
2296 * @order: allocation order
2297 * @memcg: memory cgroup to charge
2298 *
2299 * Returns 0 on success, an error code on failure.
2300 */
2301int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2302 struct mem_cgroup *memcg)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002303{
2304 unsigned int nr_pages = 1 << order;
2305 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002306 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002307
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002308 ret = try_charge(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08002309 if (ret)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002310 return ret;
Johannes Weiner52c29b02016-01-20 15:02:35 -08002311
2312 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2313 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2314 cancel_charge(memcg, nr_pages);
2315 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002316 }
2317
2318 page->mem_cgroup = memcg;
2319
2320 return 0;
2321}
2322
Vladimir Davydov45264772016-07-26 15:24:21 -07002323/**
2324 * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2325 * @page: page to charge
2326 * @gfp: reclaim mode
2327 * @order: allocation order
2328 *
2329 * Returns 0 on success, an error code on failure.
2330 */
2331int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002332{
2333 struct mem_cgroup *memcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002334 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002335
Vladimir Davydov45264772016-07-26 15:24:21 -07002336 if (memcg_kmem_bypass())
2337 return 0;
2338
Johannes Weinerdf381972014-04-07 15:37:43 -07002339 memcg = get_mem_cgroup_from_mm(current->mm);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002340 if (!mem_cgroup_is_root(memcg)) {
Vladimir Davydov45264772016-07-26 15:24:21 -07002341 ret = memcg_kmem_charge_memcg(page, gfp, order, memcg);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002342 if (!ret)
2343 __SetPageKmemcg(page);
2344 }
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002345 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002346 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002347}
Vladimir Davydov45264772016-07-26 15:24:21 -07002348/**
2349 * memcg_kmem_uncharge: uncharge a kmem page
2350 * @page: page to uncharge
2351 * @order: allocation order
2352 */
2353void memcg_kmem_uncharge(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002354{
Johannes Weiner1306a852014-12-10 15:44:52 -08002355 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002356 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002357
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002358 if (!memcg)
2359 return;
2360
Sasha Levin309381fea2014-01-23 15:52:54 -08002361 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002362
Johannes Weiner52c29b02016-01-20 15:02:35 -08002363 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2364 page_counter_uncharge(&memcg->kmem, nr_pages);
2365
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002366 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002367 if (do_memsw_account())
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002368 page_counter_uncharge(&memcg->memsw, nr_pages);
2369
Johannes Weiner1306a852014-12-10 15:44:52 -08002370 page->mem_cgroup = NULL;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002371
2372 /* slab pages do not have PageKmemcg flag set */
2373 if (PageKmemcg(page))
2374 __ClearPageKmemcg(page);
2375
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002376 css_put_many(&memcg->css, nr_pages);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002377}
Johannes Weiner127424c2016-01-20 15:02:32 -08002378#endif /* !CONFIG_SLOB */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002379
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002380#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2381
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002382/*
2383 * Because tail pages are not marked as "used", set it. We're under
Mel Gormana52633d2016-07-28 15:45:28 -07002384 * zone_lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002385 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002386void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002387{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002388 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002389
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002390 if (mem_cgroup_disabled())
2391 return;
David Rientjesb070e652013-05-07 16:18:09 -07002392
Johannes Weiner29833312014-12-10 15:44:02 -08002393 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002394 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002395
Johannes Weiner71cd3112017-05-03 14:55:13 -07002396 __this_cpu_sub(head->mem_cgroup->stat->count[MEMCG_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002397 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002398}
Hugh Dickins12d27102012-01-12 17:19:52 -08002399#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002400
Andrew Mortonc255a452012-07-31 16:43:02 -07002401#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002402static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
Huang Ying38d8b4e2017-07-06 15:37:18 -07002403 int nr_entries)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002404{
Huang Ying38d8b4e2017-07-06 15:37:18 -07002405 this_cpu_add(memcg->stat->count[MEMCG_SWAP], nr_entries);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002406}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002407
2408/**
2409 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2410 * @entry: swap entry to be moved
2411 * @from: mem_cgroup which the entry is moved from
2412 * @to: mem_cgroup which the entry is moved to
2413 *
2414 * It succeeds only when the swap_cgroup's record for this entry is the same
2415 * as the mem_cgroup's id of @from.
2416 *
2417 * Returns 0 on success, -EINVAL on failure.
2418 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002419 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002420 * both res and memsw, and called css_get().
2421 */
2422static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002423 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002424{
2425 unsigned short old_id, new_id;
2426
Li Zefan34c00c32013-09-23 16:56:01 +08002427 old_id = mem_cgroup_id(from);
2428 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002429
2430 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Huang Ying38d8b4e2017-07-06 15:37:18 -07002431 mem_cgroup_swap_statistics(from, -1);
2432 mem_cgroup_swap_statistics(to, 1);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002433 return 0;
2434 }
2435 return -EINVAL;
2436}
2437#else
2438static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002439 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002440{
2441 return -EINVAL;
2442}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002443#endif
2444
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002445static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002446
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002447static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002448 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002449{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002450 unsigned long curusage;
2451 unsigned long oldusage;
2452 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002453 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002454 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002455
2456 /*
2457 * For keeping hierarchical_reclaim simple, how long we should retry
2458 * is depends on callers. We set our retry-count to be function
2459 * of # of children which we should visit in this loop.
2460 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002461 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2462 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002463
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002464 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002465
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002466 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002467 if (signal_pending(current)) {
2468 ret = -EINTR;
2469 break;
2470 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002471
2472 mutex_lock(&memcg_limit_mutex);
2473 if (limit > memcg->memsw.limit) {
2474 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002475 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002476 break;
2477 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002478 if (limit > memcg->memory.limit)
2479 enlarge = true;
2480 ret = page_counter_limit(&memcg->memory, limit);
2481 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002482
2483 if (!ret)
2484 break;
2485
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002486 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2487
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002488 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002489 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002490 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002491 retry_count--;
2492 else
2493 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002494 } while (retry_count);
2495
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002496 if (!ret && enlarge)
2497 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002498
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002499 return ret;
2500}
2501
Li Zefan338c8432009-06-17 16:27:15 -07002502static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002503 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002504{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002505 unsigned long curusage;
2506 unsigned long oldusage;
2507 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002508 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002509 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002510
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002511 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002512 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2513 mem_cgroup_count_children(memcg);
2514
2515 oldusage = page_counter_read(&memcg->memsw);
2516
2517 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002518 if (signal_pending(current)) {
2519 ret = -EINTR;
2520 break;
2521 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002522
2523 mutex_lock(&memcg_limit_mutex);
2524 if (limit < memcg->memory.limit) {
2525 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002526 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002527 break;
2528 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002529 if (limit > memcg->memsw.limit)
2530 enlarge = true;
2531 ret = page_counter_limit(&memcg->memsw, limit);
2532 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002533
2534 if (!ret)
2535 break;
2536
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002537 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2538
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002539 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002540 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002541 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002542 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002543 else
2544 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002545 } while (retry_count);
2546
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002547 if (!ret && enlarge)
2548 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002549
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002550 return ret;
2551}
2552
Mel Gormanef8f2322016-07-28 15:46:05 -07002553unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
Andrew Morton0608f432013-09-24 15:27:41 -07002554 gfp_t gfp_mask,
2555 unsigned long *total_scanned)
2556{
2557 unsigned long nr_reclaimed = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002558 struct mem_cgroup_per_node *mz, *next_mz = NULL;
Andrew Morton0608f432013-09-24 15:27:41 -07002559 unsigned long reclaimed;
2560 int loop = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002561 struct mem_cgroup_tree_per_node *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002562 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002563 unsigned long nr_scanned;
2564
2565 if (order > 0)
2566 return 0;
2567
Mel Gormanef8f2322016-07-28 15:46:05 -07002568 mctz = soft_limit_tree_node(pgdat->node_id);
Michal Hockod6507ff2016-08-02 14:02:37 -07002569
2570 /*
2571 * Do not even bother to check the largest node if the root
2572 * is empty. Do it lockless to prevent lock bouncing. Races
2573 * are acceptable as soft limit is best effort anyway.
2574 */
Laurent Dufourbfc72282017-03-09 16:17:06 -08002575 if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root))
Michal Hockod6507ff2016-08-02 14:02:37 -07002576 return 0;
2577
Andrew Morton0608f432013-09-24 15:27:41 -07002578 /*
2579 * This loop can run a while, specially if mem_cgroup's continuously
2580 * keep exceeding their soft limit and putting the system under
2581 * pressure
2582 */
2583 do {
2584 if (next_mz)
2585 mz = next_mz;
2586 else
2587 mz = mem_cgroup_largest_soft_limit_node(mctz);
2588 if (!mz)
2589 break;
2590
2591 nr_scanned = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002592 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07002593 gfp_mask, &nr_scanned);
2594 nr_reclaimed += reclaimed;
2595 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002596 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002597 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002598
2599 /*
2600 * If we failed to reclaim anything from this memory cgroup
2601 * it is time to move on to the next cgroup
2602 */
2603 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002604 if (!reclaimed)
2605 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2606
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002607 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002608 /*
2609 * One school of thought says that we should not add
2610 * back the node to the tree if reclaim returns 0.
2611 * But our reclaim could return 0, simply because due
2612 * to priority we are exposing a smaller subset of
2613 * memory to reclaim from. Consider this as a longer
2614 * term TODO.
2615 */
2616 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002617 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002618 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002619 css_put(&mz->memcg->css);
2620 loop++;
2621 /*
2622 * Could not reclaim anything and there are no more
2623 * mem cgroups to try or we seem to be looping without
2624 * reclaiming anything.
2625 */
2626 if (!nr_reclaimed &&
2627 (next_mz == NULL ||
2628 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2629 break;
2630 } while (!nr_reclaimed);
2631 if (next_mz)
2632 css_put(&next_mz->memcg->css);
2633 return nr_reclaimed;
2634}
2635
Tejun Heoea280e72014-05-16 13:22:48 -04002636/*
2637 * Test whether @memcg has children, dead or alive. Note that this
2638 * function doesn't care whether @memcg has use_hierarchy enabled and
2639 * returns %true if there are child csses according to the cgroup
2640 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2641 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002642static inline bool memcg_has_children(struct mem_cgroup *memcg)
2643{
Tejun Heoea280e72014-05-16 13:22:48 -04002644 bool ret;
2645
Tejun Heoea280e72014-05-16 13:22:48 -04002646 rcu_read_lock();
2647 ret = css_next_child(NULL, &memcg->css);
2648 rcu_read_unlock();
2649 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002650}
2651
2652/*
Greg Thelen51038172016-05-20 16:58:18 -07002653 * Reclaims as many pages from the given memcg as possible.
Michal Hockoc26251f2012-10-26 13:37:28 +02002654 *
2655 * Caller is responsible for holding css reference for memcg.
2656 */
2657static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2658{
2659 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002660
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002661 /* we call try-to-free pages for make this cgroup empty */
2662 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002663 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002664 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002665 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002666
Michal Hockoc26251f2012-10-26 13:37:28 +02002667 if (signal_pending(current))
2668 return -EINTR;
2669
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002670 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2671 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002672 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002673 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002674 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002675 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002676 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002677
2678 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002679
2680 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002681}
2682
Tejun Heo6770c642014-05-13 12:16:21 -04002683static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2684 char *buf, size_t nbytes,
2685 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002686{
Tejun Heo6770c642014-05-13 12:16:21 -04002687 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002688
Michal Hockod8423012012-10-26 13:37:29 +02002689 if (mem_cgroup_is_root(memcg))
2690 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002691 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002692}
2693
Tejun Heo182446d2013-08-08 20:11:24 -04002694static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2695 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002696{
Tejun Heo182446d2013-08-08 20:11:24 -04002697 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002698}
2699
Tejun Heo182446d2013-08-08 20:11:24 -04002700static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2701 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002702{
2703 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002704 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002705 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002706
Glauber Costa567fb432012-07-31 16:43:07 -07002707 if (memcg->use_hierarchy == val)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002708 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07002709
Balbir Singh18f59ea2009-01-07 18:08:07 -08002710 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002711 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002712 * in the child subtrees. If it is unset, then the change can
2713 * occur, provided the current cgroup has no children.
2714 *
2715 * For the root cgroup, parent_mem is NULL, we allow value to be
2716 * set if there are no children.
2717 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002718 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002719 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002720 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002721 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002722 else
2723 retval = -EBUSY;
2724 } else
2725 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002726
Balbir Singh18f59ea2009-01-07 18:08:07 -08002727 return retval;
2728}
2729
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002730static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
Johannes Weinerce00a962014-09-05 08:43:57 -04002731{
2732 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002733 int i;
Johannes Weinerce00a962014-09-05 08:43:57 -04002734
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002735 memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
Johannes Weinerce00a962014-09-05 08:43:57 -04002736
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002737 for_each_mem_cgroup_tree(iter, memcg) {
2738 for (i = 0; i < MEMCG_NR_STAT; i++)
Johannes Weinerccda7f42017-05-03 14:55:16 -07002739 stat[i] += memcg_page_state(iter, i);
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002740 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002741}
2742
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002743static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
Johannes Weiner587d9f72016-01-20 15:03:19 -08002744{
2745 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002746 int i;
Johannes Weiner587d9f72016-01-20 15:03:19 -08002747
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002748 memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
Johannes Weiner587d9f72016-01-20 15:03:19 -08002749
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002750 for_each_mem_cgroup_tree(iter, memcg) {
2751 for (i = 0; i < MEMCG_NR_EVENTS; i++)
Johannes Weinerccda7f42017-05-03 14:55:16 -07002752 events[i] += memcg_sum_events(iter, i);
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002753 }
Johannes Weiner587d9f72016-01-20 15:03:19 -08002754}
2755
Andrew Morton6f646152015-11-06 16:28:58 -08002756static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04002757{
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002758 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002759
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002760 if (mem_cgroup_is_root(memcg)) {
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002761 struct mem_cgroup *iter;
2762
2763 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weinerccda7f42017-05-03 14:55:16 -07002764 val += memcg_page_state(iter, MEMCG_CACHE);
2765 val += memcg_page_state(iter, MEMCG_RSS);
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002766 if (swap)
Johannes Weinerccda7f42017-05-03 14:55:16 -07002767 val += memcg_page_state(iter, MEMCG_SWAP);
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002768 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002769 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002770 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002771 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002772 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002773 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002774 }
Michal Hockoc12176d2015-11-05 18:50:29 -08002775 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002776}
2777
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002778enum {
2779 RES_USAGE,
2780 RES_LIMIT,
2781 RES_MAX_USAGE,
2782 RES_FAILCNT,
2783 RES_SOFT_LIMIT,
2784};
Johannes Weinerce00a962014-09-05 08:43:57 -04002785
Tejun Heo791badb2013-12-05 12:28:02 -05002786static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002787 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002788{
Tejun Heo182446d2013-08-08 20:11:24 -04002789 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002790 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002791
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002792 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002793 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002794 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002795 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002796 case _MEMSWAP:
2797 counter = &memcg->memsw;
2798 break;
2799 case _KMEM:
2800 counter = &memcg->kmem;
2801 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002802 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08002803 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002804 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002805 default:
2806 BUG();
2807 }
2808
2809 switch (MEMFILE_ATTR(cft->private)) {
2810 case RES_USAGE:
2811 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08002812 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002813 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08002814 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002815 return (u64)page_counter_read(counter) * PAGE_SIZE;
2816 case RES_LIMIT:
2817 return (u64)counter->limit * PAGE_SIZE;
2818 case RES_MAX_USAGE:
2819 return (u64)counter->watermark * PAGE_SIZE;
2820 case RES_FAILCNT:
2821 return counter->failcnt;
2822 case RES_SOFT_LIMIT:
2823 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002824 default:
2825 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002826 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002827}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002828
Johannes Weiner127424c2016-01-20 15:02:32 -08002829#ifndef CONFIG_SLOB
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002830static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002831{
Vladimir Davydovd6441632014-01-23 15:53:09 -08002832 int memcg_id;
2833
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002834 if (cgroup_memory_nokmem)
2835 return 0;
2836
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002837 BUG_ON(memcg->kmemcg_id >= 0);
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002838 BUG_ON(memcg->kmem_state);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002839
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002840 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002841 if (memcg_id < 0)
2842 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002843
Johannes Weineref129472016-01-14 15:21:34 -08002844 static_branch_inc(&memcg_kmem_enabled_key);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002845 /*
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002846 * A memory cgroup is considered kmem-online as soon as it gets
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002847 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002848 * guarantee no one starts accounting before all call sites are
2849 * patched.
2850 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002851 memcg->kmemcg_id = memcg_id;
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002852 memcg->kmem_state = KMEM_ONLINE;
Tejun Heobc2791f2017-02-22 15:41:21 -08002853 INIT_LIST_HEAD(&memcg->kmem_caches);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002854
2855 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002856}
2857
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002858static void memcg_offline_kmem(struct mem_cgroup *memcg)
2859{
2860 struct cgroup_subsys_state *css;
2861 struct mem_cgroup *parent, *child;
2862 int kmemcg_id;
2863
2864 if (memcg->kmem_state != KMEM_ONLINE)
2865 return;
2866 /*
2867 * Clear the online state before clearing memcg_caches array
2868 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2869 * guarantees that no cache will be created for this cgroup
2870 * after we are done (see memcg_create_kmem_cache()).
2871 */
2872 memcg->kmem_state = KMEM_ALLOCATED;
2873
2874 memcg_deactivate_kmem_caches(memcg);
2875
2876 kmemcg_id = memcg->kmemcg_id;
2877 BUG_ON(kmemcg_id < 0);
2878
2879 parent = parent_mem_cgroup(memcg);
2880 if (!parent)
2881 parent = root_mem_cgroup;
2882
2883 /*
2884 * Change kmemcg_id of this cgroup and all its descendants to the
2885 * parent's id, and then move all entries from this cgroup's list_lrus
2886 * to ones of the parent. After we have finished, all list_lrus
2887 * corresponding to this cgroup are guaranteed to remain empty. The
2888 * ordering is imposed by list_lru_node->lock taken by
2889 * memcg_drain_all_list_lrus().
2890 */
Tejun Heo3a06bb72016-06-03 14:55:44 -07002891 rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002892 css_for_each_descendant_pre(css, &memcg->css) {
2893 child = mem_cgroup_from_css(css);
2894 BUG_ON(child->kmemcg_id != kmemcg_id);
2895 child->kmemcg_id = parent->kmemcg_id;
2896 if (!memcg->use_hierarchy)
2897 break;
2898 }
Tejun Heo3a06bb72016-06-03 14:55:44 -07002899 rcu_read_unlock();
2900
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002901 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2902
2903 memcg_free_cache_id(kmemcg_id);
2904}
2905
2906static void memcg_free_kmem(struct mem_cgroup *memcg)
2907{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002908 /* css_alloc() failed, offlining didn't happen */
2909 if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2910 memcg_offline_kmem(memcg);
2911
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002912 if (memcg->kmem_state == KMEM_ALLOCATED) {
2913 memcg_destroy_kmem_caches(memcg);
2914 static_branch_dec(&memcg_kmem_enabled_key);
2915 WARN_ON(page_counter_read(&memcg->kmem));
2916 }
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002917}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002918#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002919static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08002920{
2921 return 0;
2922}
2923static void memcg_offline_kmem(struct mem_cgroup *memcg)
2924{
2925}
2926static void memcg_free_kmem(struct mem_cgroup *memcg)
2927{
2928}
2929#endif /* !CONFIG_SLOB */
2930
Johannes Weiner127424c2016-01-20 15:02:32 -08002931static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2932 unsigned long limit)
2933{
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002934 int ret;
Johannes Weiner127424c2016-01-20 15:02:32 -08002935
2936 mutex_lock(&memcg_limit_mutex);
Johannes Weiner127424c2016-01-20 15:02:32 -08002937 ret = page_counter_limit(&memcg->kmem, limit);
Johannes Weiner127424c2016-01-20 15:02:32 -08002938 mutex_unlock(&memcg_limit_mutex);
2939 return ret;
2940}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002941
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002942static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2943{
2944 int ret;
2945
2946 mutex_lock(&memcg_limit_mutex);
2947
Johannes Weiner0db15292016-01-20 15:02:50 -08002948 ret = page_counter_limit(&memcg->tcpmem, limit);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002949 if (ret)
2950 goto out;
2951
Johannes Weiner0db15292016-01-20 15:02:50 -08002952 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002953 /*
2954 * The active flag needs to be written after the static_key
2955 * update. This is what guarantees that the socket activation
Johannes Weiner2d758072016-10-07 17:00:58 -07002956 * function is the last one to run. See mem_cgroup_sk_alloc()
2957 * for details, and note that we don't mark any socket as
2958 * belonging to this memcg until that flag is up.
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002959 *
2960 * We need to do this, because static_keys will span multiple
2961 * sites, but we can't control their order. If we mark a socket
2962 * as accounted, but the accounting functions are not patched in
2963 * yet, we'll lose accounting.
2964 *
Johannes Weiner2d758072016-10-07 17:00:58 -07002965 * We never race with the readers in mem_cgroup_sk_alloc(),
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002966 * because when this value change, the code to process it is not
2967 * patched in yet.
2968 */
2969 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08002970 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002971 }
2972out:
2973 mutex_unlock(&memcg_limit_mutex);
2974 return ret;
2975}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002976
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002977/*
2978 * The user of this function is...
2979 * RES_LIMIT.
2980 */
Tejun Heo451af502014-05-13 12:16:21 -04002981static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2982 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002983{
Tejun Heo451af502014-05-13 12:16:21 -04002984 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002985 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002986 int ret;
2987
Tejun Heo451af502014-05-13 12:16:21 -04002988 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08002989 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002990 if (ret)
2991 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07002992
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002993 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002994 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07002995 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2996 ret = -EINVAL;
2997 break;
2998 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002999 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3000 case _MEM:
3001 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003002 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003003 case _MEMSWAP:
3004 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3005 break;
3006 case _KMEM:
3007 ret = memcg_update_kmem_limit(memcg, nr_pages);
3008 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003009 case _TCP:
3010 ret = memcg_update_tcp_limit(memcg, nr_pages);
3011 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003012 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003013 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003014 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003015 memcg->soft_limit = nr_pages;
3016 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003017 break;
3018 }
Tejun Heo451af502014-05-13 12:16:21 -04003019 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003020}
3021
Tejun Heo6770c642014-05-13 12:16:21 -04003022static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3023 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003024{
Tejun Heo6770c642014-05-13 12:16:21 -04003025 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003026 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003027
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003028 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3029 case _MEM:
3030 counter = &memcg->memory;
3031 break;
3032 case _MEMSWAP:
3033 counter = &memcg->memsw;
3034 break;
3035 case _KMEM:
3036 counter = &memcg->kmem;
3037 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003038 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003039 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003040 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003041 default:
3042 BUG();
3043 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003044
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003045 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003046 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003047 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003048 break;
3049 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003050 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003051 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003052 default:
3053 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003054 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003055
Tejun Heo6770c642014-05-13 12:16:21 -04003056 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003057}
3058
Tejun Heo182446d2013-08-08 20:11:24 -04003059static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003060 struct cftype *cft)
3061{
Tejun Heo182446d2013-08-08 20:11:24 -04003062 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003063}
3064
Daisuke Nishimura02491442010-03-10 15:22:17 -08003065#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003066static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003067 struct cftype *cft, u64 val)
3068{
Tejun Heo182446d2013-08-08 20:11:24 -04003069 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003070
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003071 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003072 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003073
Glauber Costaee5e8472013-02-22 16:34:50 -08003074 /*
3075 * No kind of locking is needed in here, because ->can_attach() will
3076 * check this value once in the beginning of the process, and then carry
3077 * on with stale data. This means that changes to this value will only
3078 * affect task migrations starting after the change.
3079 */
3080 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003081 return 0;
3082}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003083#else
Tejun Heo182446d2013-08-08 20:11:24 -04003084static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003085 struct cftype *cft, u64 val)
3086{
3087 return -ENOSYS;
3088}
3089#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003090
Ying Han406eb0c2011-05-26 16:25:37 -07003091#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003092static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003093{
Greg Thelen25485de2013-11-12 15:07:40 -08003094 struct numa_stat {
3095 const char *name;
3096 unsigned int lru_mask;
3097 };
3098
3099 static const struct numa_stat stats[] = {
3100 { "total", LRU_ALL },
3101 { "file", LRU_ALL_FILE },
3102 { "anon", LRU_ALL_ANON },
3103 { "unevictable", BIT(LRU_UNEVICTABLE) },
3104 };
3105 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003106 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003107 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003108 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003109
Greg Thelen25485de2013-11-12 15:07:40 -08003110 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3111 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3112 seq_printf(m, "%s=%lu", stat->name, nr);
3113 for_each_node_state(nid, N_MEMORY) {
3114 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3115 stat->lru_mask);
3116 seq_printf(m, " N%d=%lu", nid, nr);
3117 }
3118 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003119 }
Ying Han406eb0c2011-05-26 16:25:37 -07003120
Ying Han071aee12013-11-12 15:07:41 -08003121 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3122 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003123
Ying Han071aee12013-11-12 15:07:41 -08003124 nr = 0;
3125 for_each_mem_cgroup_tree(iter, memcg)
3126 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3127 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3128 for_each_node_state(nid, N_MEMORY) {
3129 nr = 0;
3130 for_each_mem_cgroup_tree(iter, memcg)
3131 nr += mem_cgroup_node_nr_lru_pages(
3132 iter, nid, stat->lru_mask);
3133 seq_printf(m, " N%d=%lu", nid, nr);
3134 }
3135 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003136 }
Ying Han406eb0c2011-05-26 16:25:37 -07003137
Ying Han406eb0c2011-05-26 16:25:37 -07003138 return 0;
3139}
3140#endif /* CONFIG_NUMA */
3141
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07003142/* Universal VM events cgroup1 shows, original sort order */
3143unsigned int memcg1_events[] = {
3144 PGPGIN,
3145 PGPGOUT,
3146 PGFAULT,
3147 PGMAJFAULT,
3148};
3149
3150static const char *const memcg1_event_names[] = {
3151 "pgpgin",
3152 "pgpgout",
3153 "pgfault",
3154 "pgmajfault",
3155};
3156
Tejun Heo2da8ca82013-12-05 12:28:04 -05003157static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003158{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003159 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003160 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003161 struct mem_cgroup *mi;
3162 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003163
Johannes Weiner71cd3112017-05-03 14:55:13 -07003164 BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats));
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003165 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3166
Johannes Weiner71cd3112017-05-03 14:55:13 -07003167 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
3168 if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003169 continue;
Johannes Weiner71cd3112017-05-03 14:55:13 -07003170 seq_printf(m, "%s %lu\n", memcg1_stat_names[i],
Johannes Weinerccda7f42017-05-03 14:55:16 -07003171 memcg_page_state(memcg, memcg1_stats[i]) *
Johannes Weiner71cd3112017-05-03 14:55:13 -07003172 PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003173 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003174
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07003175 for (i = 0; i < ARRAY_SIZE(memcg1_events); i++)
3176 seq_printf(m, "%s %lu\n", memcg1_event_names[i],
Johannes Weinerccda7f42017-05-03 14:55:16 -07003177 memcg_sum_events(memcg, memcg1_events[i]));
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003178
3179 for (i = 0; i < NR_LRU_LISTS; i++)
3180 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3181 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3182
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003183 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003184 memory = memsw = PAGE_COUNTER_MAX;
3185 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3186 memory = min(memory, mi->memory.limit);
3187 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003188 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003189 seq_printf(m, "hierarchical_memory_limit %llu\n",
3190 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08003191 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003192 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3193 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003194
Johannes Weiner71cd3112017-05-03 14:55:13 -07003195 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003196 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003197
Johannes Weiner71cd3112017-05-03 14:55:13 -07003198 if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003199 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003200 for_each_mem_cgroup_tree(mi, memcg)
Johannes Weinerccda7f42017-05-03 14:55:16 -07003201 val += memcg_page_state(mi, memcg1_stats[i]) *
Johannes Weiner71cd3112017-05-03 14:55:13 -07003202 PAGE_SIZE;
3203 seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003204 }
3205
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07003206 for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) {
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003207 unsigned long long val = 0;
3208
3209 for_each_mem_cgroup_tree(mi, memcg)
Johannes Weinerccda7f42017-05-03 14:55:16 -07003210 val += memcg_sum_events(mi, memcg1_events[i]);
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07003211 seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003212 }
3213
3214 for (i = 0; i < NR_LRU_LISTS; i++) {
3215 unsigned long long val = 0;
3216
3217 for_each_mem_cgroup_tree(mi, memcg)
3218 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3219 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003220 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003221
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003222#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003223 {
Mel Gormanef8f2322016-07-28 15:46:05 -07003224 pg_data_t *pgdat;
3225 struct mem_cgroup_per_node *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003226 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003227 unsigned long recent_rotated[2] = {0, 0};
3228 unsigned long recent_scanned[2] = {0, 0};
3229
Mel Gormanef8f2322016-07-28 15:46:05 -07003230 for_each_online_pgdat(pgdat) {
3231 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
3232 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003233
Mel Gormanef8f2322016-07-28 15:46:05 -07003234 recent_rotated[0] += rstat->recent_rotated[0];
3235 recent_rotated[1] += rstat->recent_rotated[1];
3236 recent_scanned[0] += rstat->recent_scanned[0];
3237 recent_scanned[1] += rstat->recent_scanned[1];
3238 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003239 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3240 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3241 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3242 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003243 }
3244#endif
3245
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003246 return 0;
3247}
3248
Tejun Heo182446d2013-08-08 20:11:24 -04003249static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3250 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003251{
Tejun Heo182446d2013-08-08 20:11:24 -04003252 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003253
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003254 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003255}
3256
Tejun Heo182446d2013-08-08 20:11:24 -04003257static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3258 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003259{
Tejun Heo182446d2013-08-08 20:11:24 -04003260 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003261
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003262 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003263 return -EINVAL;
3264
Linus Torvalds14208b02014-06-09 15:03:33 -07003265 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003266 memcg->swappiness = val;
3267 else
3268 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003269
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003270 return 0;
3271}
3272
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003273static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3274{
3275 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003276 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003277 int i;
3278
3279 rcu_read_lock();
3280 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003281 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003282 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003283 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003284
3285 if (!t)
3286 goto unlock;
3287
Johannes Weinerce00a962014-09-05 08:43:57 -04003288 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003289
3290 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003291 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003292 * If it's not true, a threshold was crossed after last
3293 * call of __mem_cgroup_threshold().
3294 */
Phil Carmody5407a562010-05-26 14:42:42 -07003295 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003296
3297 /*
3298 * Iterate backward over array of thresholds starting from
3299 * current_threshold and check if a threshold is crossed.
3300 * If none of thresholds below usage is crossed, we read
3301 * only one element of the array here.
3302 */
3303 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3304 eventfd_signal(t->entries[i].eventfd, 1);
3305
3306 /* i = current_threshold + 1 */
3307 i++;
3308
3309 /*
3310 * Iterate forward over array of thresholds starting from
3311 * current_threshold+1 and check if a threshold is crossed.
3312 * If none of thresholds above usage is crossed, we read
3313 * only one element of the array here.
3314 */
3315 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3316 eventfd_signal(t->entries[i].eventfd, 1);
3317
3318 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003319 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003320unlock:
3321 rcu_read_unlock();
3322}
3323
3324static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3325{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003326 while (memcg) {
3327 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08003328 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003329 __mem_cgroup_threshold(memcg, true);
3330
3331 memcg = parent_mem_cgroup(memcg);
3332 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003333}
3334
3335static int compare_thresholds(const void *a, const void *b)
3336{
3337 const struct mem_cgroup_threshold *_a = a;
3338 const struct mem_cgroup_threshold *_b = b;
3339
Greg Thelen2bff24a2013-09-11 14:23:08 -07003340 if (_a->threshold > _b->threshold)
3341 return 1;
3342
3343 if (_a->threshold < _b->threshold)
3344 return -1;
3345
3346 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003347}
3348
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003349static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003350{
3351 struct mem_cgroup_eventfd_list *ev;
3352
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003353 spin_lock(&memcg_oom_lock);
3354
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003355 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003356 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003357
3358 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003359 return 0;
3360}
3361
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003362static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003363{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003364 struct mem_cgroup *iter;
3365
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003366 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003367 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003368}
3369
Tejun Heo59b6f872013-11-22 18:20:43 -05003370static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003371 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003372{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003373 struct mem_cgroup_thresholds *thresholds;
3374 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003375 unsigned long threshold;
3376 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003377 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003378
Johannes Weiner650c5e52015-02-11 15:26:03 -08003379 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003380 if (ret)
3381 return ret;
3382
3383 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003384
Johannes Weiner05b84302014-08-06 16:05:59 -07003385 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003386 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003387 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003388 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003389 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003390 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003391 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003392 BUG();
3393
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003394 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003395 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003396 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3397
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003398 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003399
3400 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003401 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003402 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003403 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003404 ret = -ENOMEM;
3405 goto unlock;
3406 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003407 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003408
3409 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003410 if (thresholds->primary) {
3411 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003412 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003413 }
3414
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003415 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003416 new->entries[size - 1].eventfd = eventfd;
3417 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003418
3419 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003420 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003421 compare_thresholds, NULL);
3422
3423 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003424 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003425 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003426 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003427 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003428 * new->current_threshold will not be used until
3429 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003430 * it here.
3431 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003432 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003433 } else
3434 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003435 }
3436
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003437 /* Free old spare buffer and save old primary buffer as spare */
3438 kfree(thresholds->spare);
3439 thresholds->spare = thresholds->primary;
3440
3441 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003442
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003443 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003444 synchronize_rcu();
3445
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003446unlock:
3447 mutex_unlock(&memcg->thresholds_lock);
3448
3449 return ret;
3450}
3451
Tejun Heo59b6f872013-11-22 18:20:43 -05003452static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003453 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003454{
Tejun Heo59b6f872013-11-22 18:20:43 -05003455 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003456}
3457
Tejun Heo59b6f872013-11-22 18:20:43 -05003458static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003459 struct eventfd_ctx *eventfd, const char *args)
3460{
Tejun Heo59b6f872013-11-22 18:20:43 -05003461 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003462}
3463
Tejun Heo59b6f872013-11-22 18:20:43 -05003464static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003465 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003466{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003467 struct mem_cgroup_thresholds *thresholds;
3468 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003469 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003470 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003471
3472 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003473
3474 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003475 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003476 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003477 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003478 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003479 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003480 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003481 BUG();
3482
Anton Vorontsov371528c2012-02-24 05:14:46 +04003483 if (!thresholds->primary)
3484 goto unlock;
3485
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003486 /* Check if a threshold crossed before removing */
3487 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3488
3489 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003490 size = 0;
3491 for (i = 0; i < thresholds->primary->size; i++) {
3492 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003493 size++;
3494 }
3495
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003496 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003497
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003498 /* Set thresholds array to NULL if we don't have thresholds */
3499 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003500 kfree(new);
3501 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003502 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003503 }
3504
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003505 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003506
3507 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003508 new->current_threshold = -1;
3509 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3510 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003511 continue;
3512
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003513 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003514 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003515 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003516 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003517 * until rcu_assign_pointer(), so it's safe to increment
3518 * it here.
3519 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003520 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003521 }
3522 j++;
3523 }
3524
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003525swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003526 /* Swap primary and spare array */
3527 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003528
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003529 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003530
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003531 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003532 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08003533
3534 /* If all events are unregistered, free the spare array */
3535 if (!new) {
3536 kfree(thresholds->spare);
3537 thresholds->spare = NULL;
3538 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04003539unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003540 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003541}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003542
Tejun Heo59b6f872013-11-22 18:20:43 -05003543static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003544 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003545{
Tejun Heo59b6f872013-11-22 18:20:43 -05003546 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003547}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003548
Tejun Heo59b6f872013-11-22 18:20:43 -05003549static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003550 struct eventfd_ctx *eventfd)
3551{
Tejun Heo59b6f872013-11-22 18:20:43 -05003552 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003553}
3554
Tejun Heo59b6f872013-11-22 18:20:43 -05003555static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003556 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003557{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003558 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003559
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003560 event = kmalloc(sizeof(*event), GFP_KERNEL);
3561 if (!event)
3562 return -ENOMEM;
3563
Michal Hocko1af8efe2011-07-26 16:08:24 -07003564 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003565
3566 event->eventfd = eventfd;
3567 list_add(&event->list, &memcg->oom_notify);
3568
3569 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07003570 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003571 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003572 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003573
3574 return 0;
3575}
3576
Tejun Heo59b6f872013-11-22 18:20:43 -05003577static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003578 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003579{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003580 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003581
Michal Hocko1af8efe2011-07-26 16:08:24 -07003582 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003583
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003584 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003585 if (ev->eventfd == eventfd) {
3586 list_del(&ev->list);
3587 kfree(ev);
3588 }
3589 }
3590
Michal Hocko1af8efe2011-07-26 16:08:24 -07003591 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003592}
3593
Tejun Heo2da8ca82013-12-05 12:28:04 -05003594static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003595{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003596 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003597
Tejun Heo791badb2013-12-05 12:28:02 -05003598 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07003599 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
Konstantin Khlebnikov8e675f72017-07-06 15:40:28 -07003600 seq_printf(sf, "oom_kill %lu\n", memcg_sum_events(memcg, OOM_KILL));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003601 return 0;
3602}
3603
Tejun Heo182446d2013-08-08 20:11:24 -04003604static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003605 struct cftype *cft, u64 val)
3606{
Tejun Heo182446d2013-08-08 20:11:24 -04003607 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003608
3609 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003610 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003611 return -EINVAL;
3612
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003613 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003614 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003615 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003616
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003617 return 0;
3618}
3619
Tejun Heo52ebea72015-05-22 17:13:37 -04003620#ifdef CONFIG_CGROUP_WRITEBACK
3621
3622struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3623{
3624 return &memcg->cgwb_list;
3625}
3626
Tejun Heo841710a2015-05-22 18:23:33 -04003627static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3628{
3629 return wb_domain_init(&memcg->cgwb_domain, gfp);
3630}
3631
3632static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3633{
3634 wb_domain_exit(&memcg->cgwb_domain);
3635}
3636
Tejun Heo2529bb32015-05-22 18:23:34 -04003637static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3638{
3639 wb_domain_size_changed(&memcg->cgwb_domain);
3640}
3641
Tejun Heo841710a2015-05-22 18:23:33 -04003642struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3643{
3644 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3645
3646 if (!memcg->css.parent)
3647 return NULL;
3648
3649 return &memcg->cgwb_domain;
3650}
3651
Tejun Heoc2aa7232015-05-22 18:23:35 -04003652/**
3653 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3654 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003655 * @pfilepages: out parameter for number of file pages
3656 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04003657 * @pdirty: out parameter for number of dirty pages
3658 * @pwriteback: out parameter for number of pages under writeback
3659 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003660 * Determine the numbers of file, headroom, dirty, and writeback pages in
3661 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3662 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003663 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003664 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3665 * headroom is calculated as the lowest headroom of itself and the
3666 * ancestors. Note that this doesn't consider the actual amount of
3667 * available memory in the system. The caller should further cap
3668 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003669 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003670void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3671 unsigned long *pheadroom, unsigned long *pdirty,
3672 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04003673{
3674 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3675 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003676
Johannes Weinerccda7f42017-05-03 14:55:16 -07003677 *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY);
Tejun Heoc2aa7232015-05-22 18:23:35 -04003678
3679 /* this should eventually include NR_UNSTABLE_NFS */
Johannes Weinerccda7f42017-05-03 14:55:16 -07003680 *pwriteback = memcg_page_state(memcg, NR_WRITEBACK);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003681 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3682 (1 << LRU_ACTIVE_FILE));
3683 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003684
Tejun Heoc2aa7232015-05-22 18:23:35 -04003685 while ((parent = parent_mem_cgroup(memcg))) {
3686 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3687 unsigned long used = page_counter_read(&memcg->memory);
3688
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003689 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04003690 memcg = parent;
3691 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04003692}
3693
Tejun Heo841710a2015-05-22 18:23:33 -04003694#else /* CONFIG_CGROUP_WRITEBACK */
3695
3696static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3697{
3698 return 0;
3699}
3700
3701static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3702{
3703}
3704
Tejun Heo2529bb32015-05-22 18:23:34 -04003705static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3706{
3707}
3708
Tejun Heo52ebea72015-05-22 17:13:37 -04003709#endif /* CONFIG_CGROUP_WRITEBACK */
3710
Tejun Heo79bd9812013-11-22 18:20:42 -05003711/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003712 * DO NOT USE IN NEW FILES.
3713 *
3714 * "cgroup.event_control" implementation.
3715 *
3716 * This is way over-engineered. It tries to support fully configurable
3717 * events for each user. Such level of flexibility is completely
3718 * unnecessary especially in the light of the planned unified hierarchy.
3719 *
3720 * Please deprecate this and replace with something simpler if at all
3721 * possible.
3722 */
3723
3724/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003725 * Unregister event and free resources.
3726 *
3727 * Gets called from workqueue.
3728 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003729static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003730{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003731 struct mem_cgroup_event *event =
3732 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003733 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003734
3735 remove_wait_queue(event->wqh, &event->wait);
3736
Tejun Heo59b6f872013-11-22 18:20:43 -05003737 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003738
3739 /* Notify userspace the event is going away. */
3740 eventfd_signal(event->eventfd, 1);
3741
3742 eventfd_ctx_put(event->eventfd);
3743 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003744 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003745}
3746
3747/*
3748 * Gets called on POLLHUP on eventfd when user closes it.
3749 *
3750 * Called with wqh->lock held and interrupts disabled.
3751 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02003752static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode,
Tejun Heo3bc942f2013-11-22 18:20:44 -05003753 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003754{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003755 struct mem_cgroup_event *event =
3756 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003757 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003758 unsigned long flags = (unsigned long)key;
3759
3760 if (flags & POLLHUP) {
3761 /*
3762 * If the event has been detached at cgroup removal, we
3763 * can simply return knowing the other side will cleanup
3764 * for us.
3765 *
3766 * We can't race against event freeing since the other
3767 * side will require wqh->lock via remove_wait_queue(),
3768 * which we hold.
3769 */
Tejun Heofba94802013-11-22 18:20:43 -05003770 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003771 if (!list_empty(&event->list)) {
3772 list_del_init(&event->list);
3773 /*
3774 * We are in atomic context, but cgroup_event_remove()
3775 * may sleep, so we have to call it in workqueue.
3776 */
3777 schedule_work(&event->remove);
3778 }
Tejun Heofba94802013-11-22 18:20:43 -05003779 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003780 }
3781
3782 return 0;
3783}
3784
Tejun Heo3bc942f2013-11-22 18:20:44 -05003785static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003786 wait_queue_head_t *wqh, poll_table *pt)
3787{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003788 struct mem_cgroup_event *event =
3789 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003790
3791 event->wqh = wqh;
3792 add_wait_queue(wqh, &event->wait);
3793}
3794
3795/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003796 * DO NOT USE IN NEW FILES.
3797 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003798 * Parse input and register new cgroup event handler.
3799 *
3800 * Input must be in format '<event_fd> <control_fd> <args>'.
3801 * Interpretation of args is defined by control file implementation.
3802 */
Tejun Heo451af502014-05-13 12:16:21 -04003803static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3804 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003805{
Tejun Heo451af502014-05-13 12:16:21 -04003806 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003807 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003808 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003809 struct cgroup_subsys_state *cfile_css;
3810 unsigned int efd, cfd;
3811 struct fd efile;
3812 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003813 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003814 char *endp;
3815 int ret;
3816
Tejun Heo451af502014-05-13 12:16:21 -04003817 buf = strstrip(buf);
3818
3819 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003820 if (*endp != ' ')
3821 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003822 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003823
Tejun Heo451af502014-05-13 12:16:21 -04003824 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003825 if ((*endp != ' ') && (*endp != '\0'))
3826 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003827 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003828
3829 event = kzalloc(sizeof(*event), GFP_KERNEL);
3830 if (!event)
3831 return -ENOMEM;
3832
Tejun Heo59b6f872013-11-22 18:20:43 -05003833 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003834 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003835 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3836 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3837 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003838
3839 efile = fdget(efd);
3840 if (!efile.file) {
3841 ret = -EBADF;
3842 goto out_kfree;
3843 }
3844
3845 event->eventfd = eventfd_ctx_fileget(efile.file);
3846 if (IS_ERR(event->eventfd)) {
3847 ret = PTR_ERR(event->eventfd);
3848 goto out_put_efile;
3849 }
3850
3851 cfile = fdget(cfd);
3852 if (!cfile.file) {
3853 ret = -EBADF;
3854 goto out_put_eventfd;
3855 }
3856
3857 /* the process need read permission on control file */
3858 /* AV: shouldn't we check that it's been opened for read instead? */
3859 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3860 if (ret < 0)
3861 goto out_put_cfile;
3862
Tejun Heo79bd9812013-11-22 18:20:42 -05003863 /*
Tejun Heofba94802013-11-22 18:20:43 -05003864 * Determine the event callbacks and set them in @event. This used
3865 * to be done via struct cftype but cgroup core no longer knows
3866 * about these events. The following is crude but the whole thing
3867 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003868 *
3869 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003870 */
Al Virob5830432014-10-31 01:22:04 -04003871 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003872
3873 if (!strcmp(name, "memory.usage_in_bytes")) {
3874 event->register_event = mem_cgroup_usage_register_event;
3875 event->unregister_event = mem_cgroup_usage_unregister_event;
3876 } else if (!strcmp(name, "memory.oom_control")) {
3877 event->register_event = mem_cgroup_oom_register_event;
3878 event->unregister_event = mem_cgroup_oom_unregister_event;
3879 } else if (!strcmp(name, "memory.pressure_level")) {
3880 event->register_event = vmpressure_register_event;
3881 event->unregister_event = vmpressure_unregister_event;
3882 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003883 event->register_event = memsw_cgroup_usage_register_event;
3884 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003885 } else {
3886 ret = -EINVAL;
3887 goto out_put_cfile;
3888 }
3889
3890 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003891 * Verify @cfile should belong to @css. Also, remaining events are
3892 * automatically removed on cgroup destruction but the removal is
3893 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003894 */
Al Virob5830432014-10-31 01:22:04 -04003895 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003896 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003897 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003898 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003899 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003900 if (cfile_css != css) {
3901 css_put(cfile_css);
3902 goto out_put_cfile;
3903 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003904
Tejun Heo451af502014-05-13 12:16:21 -04003905 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003906 if (ret)
3907 goto out_put_css;
3908
3909 efile.file->f_op->poll(efile.file, &event->pt);
3910
Tejun Heofba94802013-11-22 18:20:43 -05003911 spin_lock(&memcg->event_list_lock);
3912 list_add(&event->list, &memcg->event_list);
3913 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003914
3915 fdput(cfile);
3916 fdput(efile);
3917
Tejun Heo451af502014-05-13 12:16:21 -04003918 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05003919
3920out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05003921 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003922out_put_cfile:
3923 fdput(cfile);
3924out_put_eventfd:
3925 eventfd_ctx_put(event->eventfd);
3926out_put_efile:
3927 fdput(efile);
3928out_kfree:
3929 kfree(event);
3930
3931 return ret;
3932}
3933
Johannes Weiner241994ed2015-02-11 15:26:06 -08003934static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003935 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003936 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003937 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003938 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003939 },
3940 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003941 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003942 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04003943 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003944 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003945 },
3946 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003947 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003948 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003949 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003950 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003951 },
3952 {
Balbir Singh296c81d2009-09-23 15:56:36 -07003953 .name = "soft_limit_in_bytes",
3954 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003955 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003956 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07003957 },
3958 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003959 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003960 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04003961 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003962 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003963 },
Balbir Singh8697d332008-02-07 00:13:59 -08003964 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003965 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003966 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003967 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003968 {
3969 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04003970 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003971 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08003972 {
3973 .name = "use_hierarchy",
3974 .write_u64 = mem_cgroup_hierarchy_write,
3975 .read_u64 = mem_cgroup_hierarchy_read,
3976 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003977 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05003978 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04003979 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04003980 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05003981 },
3982 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003983 .name = "swappiness",
3984 .read_u64 = mem_cgroup_swappiness_read,
3985 .write_u64 = mem_cgroup_swappiness_write,
3986 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003987 {
3988 .name = "move_charge_at_immigrate",
3989 .read_u64 = mem_cgroup_move_charge_read,
3990 .write_u64 = mem_cgroup_move_charge_write,
3991 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003992 {
3993 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003994 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003995 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003996 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3997 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003998 {
3999 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004000 },
Ying Han406eb0c2011-05-26 16:25:37 -07004001#ifdef CONFIG_NUMA
4002 {
4003 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004004 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07004005 },
4006#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004007 {
4008 .name = "kmem.limit_in_bytes",
4009 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004010 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004011 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004012 },
4013 {
4014 .name = "kmem.usage_in_bytes",
4015 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004016 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004017 },
4018 {
4019 .name = "kmem.failcnt",
4020 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004021 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004022 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004023 },
4024 {
4025 .name = "kmem.max_usage_in_bytes",
4026 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004027 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004028 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004029 },
Glauber Costa749c5412012-12-18 14:23:01 -08004030#ifdef CONFIG_SLABINFO
4031 {
4032 .name = "kmem.slabinfo",
Tejun Heobc2791f2017-02-22 15:41:21 -08004033 .seq_start = memcg_slab_start,
4034 .seq_next = memcg_slab_next,
4035 .seq_stop = memcg_slab_stop,
Vladimir Davydovb0475012014-12-10 15:44:19 -08004036 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004037 },
4038#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004039 {
4040 .name = "kmem.tcp.limit_in_bytes",
4041 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4042 .write = mem_cgroup_write,
4043 .read_u64 = mem_cgroup_read_u64,
4044 },
4045 {
4046 .name = "kmem.tcp.usage_in_bytes",
4047 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4048 .read_u64 = mem_cgroup_read_u64,
4049 },
4050 {
4051 .name = "kmem.tcp.failcnt",
4052 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4053 .write = mem_cgroup_reset,
4054 .read_u64 = mem_cgroup_read_u64,
4055 },
4056 {
4057 .name = "kmem.tcp.max_usage_in_bytes",
4058 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4059 .write = mem_cgroup_reset,
4060 .read_u64 = mem_cgroup_read_u64,
4061 },
Tejun Heo6bc10342012-04-01 12:09:55 -07004062 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004063};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004064
Johannes Weiner73f576c2016-07-20 15:44:57 -07004065/*
4066 * Private memory cgroup IDR
4067 *
4068 * Swap-out records and page cache shadow entries need to store memcg
4069 * references in constrained space, so we maintain an ID space that is
4070 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4071 * memory-controlled cgroups to 64k.
4072 *
4073 * However, there usually are many references to the oflline CSS after
4074 * the cgroup has been destroyed, such as page cache or reclaimable
4075 * slab objects, that don't need to hang on to the ID. We want to keep
4076 * those dead CSS from occupying IDs, or we might quickly exhaust the
4077 * relatively small ID space and prevent the creation of new cgroups
4078 * even when there are much fewer than 64k cgroups - possibly none.
4079 *
4080 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4081 * be freed and recycled when it's no longer needed, which is usually
4082 * when the CSS is offlined.
4083 *
4084 * The only exception to that are records of swapped out tmpfs/shmem
4085 * pages that need to be attributed to live ancestors on swapin. But
4086 * those references are manageable from userspace.
4087 */
4088
4089static DEFINE_IDR(mem_cgroup_idr);
4090
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004091static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004092{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004093 VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0);
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004094 atomic_add(n, &memcg->id.ref);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004095}
4096
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004097static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004098{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004099 VM_BUG_ON(atomic_read(&memcg->id.ref) < n);
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004100 if (atomic_sub_and_test(n, &memcg->id.ref)) {
Johannes Weiner73f576c2016-07-20 15:44:57 -07004101 idr_remove(&mem_cgroup_idr, memcg->id.id);
4102 memcg->id.id = 0;
4103
4104 /* Memcg ID pins CSS */
4105 css_put(&memcg->css);
4106 }
4107}
4108
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004109static inline void mem_cgroup_id_get(struct mem_cgroup *memcg)
4110{
4111 mem_cgroup_id_get_many(memcg, 1);
4112}
4113
4114static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
4115{
4116 mem_cgroup_id_put_many(memcg, 1);
4117}
4118
Johannes Weiner73f576c2016-07-20 15:44:57 -07004119/**
4120 * mem_cgroup_from_id - look up a memcg from a memcg id
4121 * @id: the memcg id to look up
4122 *
4123 * Caller must hold rcu_read_lock().
4124 */
4125struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
4126{
4127 WARN_ON_ONCE(!rcu_read_lock_held());
4128 return idr_find(&mem_cgroup_idr, id);
4129}
4130
Mel Gormanef8f2322016-07-28 15:46:05 -07004131static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004132{
4133 struct mem_cgroup_per_node *pn;
Mel Gormanef8f2322016-07-28 15:46:05 -07004134 int tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004135 /*
4136 * This routine is called against possible nodes.
4137 * But it's BUG to call kmalloc() against offline node.
4138 *
4139 * TODO: this routine can waste much memory for nodes which will
4140 * never be onlined. It's better to use memory hotplug callback
4141 * function.
4142 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004143 if (!node_state(node, N_NORMAL_MEMORY))
4144 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004145 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004146 if (!pn)
4147 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004148
Johannes Weiner00f3ca22017-07-06 15:40:52 -07004149 pn->lruvec_stat = alloc_percpu(struct lruvec_stat);
4150 if (!pn->lruvec_stat) {
4151 kfree(pn);
4152 return 1;
4153 }
4154
Mel Gormanef8f2322016-07-28 15:46:05 -07004155 lruvec_init(&pn->lruvec);
4156 pn->usage_in_excess = 0;
4157 pn->on_tree = false;
4158 pn->memcg = memcg;
4159
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004160 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004161 return 0;
4162}
4163
Mel Gormanef8f2322016-07-28 15:46:05 -07004164static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004165{
Johannes Weiner00f3ca22017-07-06 15:40:52 -07004166 struct mem_cgroup_per_node *pn = memcg->nodeinfo[node];
4167
4168 free_percpu(pn->lruvec_stat);
4169 kfree(pn);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004170}
4171
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08004172static void __mem_cgroup_free(struct mem_cgroup *memcg)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004173{
4174 int node;
4175
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004176 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004177 free_mem_cgroup_per_node_info(memcg, node);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004178 free_percpu(memcg->stat);
4179 kfree(memcg);
4180}
4181
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08004182static void mem_cgroup_free(struct mem_cgroup *memcg)
4183{
4184 memcg_wb_domain_exit(memcg);
4185 __mem_cgroup_free(memcg);
4186}
4187
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004188static struct mem_cgroup *mem_cgroup_alloc(void)
4189{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004190 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004191 size_t size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004192 int node;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004193
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004194 size = sizeof(struct mem_cgroup);
4195 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004196
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004197 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004198 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004199 return NULL;
4200
Johannes Weiner73f576c2016-07-20 15:44:57 -07004201 memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
4202 1, MEM_CGROUP_ID_MAX,
4203 GFP_KERNEL);
4204 if (memcg->id.id < 0)
4205 goto fail;
4206
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004207 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4208 if (!memcg->stat)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004209 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004210
Bob Liu3ed28fa2012-01-12 17:19:04 -08004211 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004212 if (alloc_mem_cgroup_per_node_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004213 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004214
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004215 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4216 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004217
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004218 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08004219 memcg->last_scanned_node = MAX_NUMNODES;
4220 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004221 mutex_init(&memcg->thresholds_lock);
4222 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004223 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004224 INIT_LIST_HEAD(&memcg->event_list);
4225 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004226 memcg->socket_pressure = jiffies;
Johannes Weiner127424c2016-01-20 15:02:32 -08004227#ifndef CONFIG_SLOB
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004228 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004229#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004230#ifdef CONFIG_CGROUP_WRITEBACK
4231 INIT_LIST_HEAD(&memcg->cgwb_list);
4232#endif
Johannes Weiner73f576c2016-07-20 15:44:57 -07004233 idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004234 return memcg;
4235fail:
Johannes Weiner73f576c2016-07-20 15:44:57 -07004236 if (memcg->id.id > 0)
4237 idr_remove(&mem_cgroup_idr, memcg->id.id);
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08004238 __mem_cgroup_free(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004239 return NULL;
Glauber Costad142e3e2013-02-22 16:34:52 -08004240}
4241
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004242static struct cgroup_subsys_state * __ref
4243mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004244{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004245 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4246 struct mem_cgroup *memcg;
4247 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08004248
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004249 memcg = mem_cgroup_alloc();
4250 if (!memcg)
4251 return ERR_PTR(error);
Li Zefan4219b2d2013-09-23 16:56:29 +08004252
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004253 memcg->high = PAGE_COUNTER_MAX;
4254 memcg->soft_limit = PAGE_COUNTER_MAX;
4255 if (parent) {
4256 memcg->swappiness = mem_cgroup_swappiness(parent);
4257 memcg->oom_kill_disable = parent->oom_kill_disable;
4258 }
4259 if (parent && parent->use_hierarchy) {
4260 memcg->use_hierarchy = true;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004261 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004262 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004263 page_counter_init(&memcg->memsw, &parent->memsw);
4264 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08004265 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004266 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004267 page_counter_init(&memcg->memory, NULL);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004268 page_counter_init(&memcg->swap, NULL);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004269 page_counter_init(&memcg->memsw, NULL);
4270 page_counter_init(&memcg->kmem, NULL);
Johannes Weiner0db15292016-01-20 15:02:50 -08004271 page_counter_init(&memcg->tcpmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004272 /*
4273 * Deeper hierachy with use_hierarchy == false doesn't make
4274 * much sense so let cgroup subsystem know about this
4275 * unfortunate state in our controller.
4276 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004277 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004278 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004279 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08004280
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004281 /* The following stuff does not apply to the root */
4282 if (!parent) {
4283 root_mem_cgroup = memcg;
4284 return &memcg->css;
4285 }
4286
Vladimir Davydovb313aee2016-03-17 14:18:27 -07004287 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004288 if (error)
4289 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08004290
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004291 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004292 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004293
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004294 return &memcg->css;
4295fail:
4296 mem_cgroup_free(memcg);
Tejun Heoea3a9642016-06-24 14:49:58 -07004297 return ERR_PTR(-ENOMEM);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004298}
4299
Johannes Weiner73f576c2016-07-20 15:44:57 -07004300static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004301{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004302 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4303
Johannes Weiner73f576c2016-07-20 15:44:57 -07004304 /* Online state pins memcg ID, memcg ID pins CSS */
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004305 atomic_set(&memcg->id.ref, 1);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004306 css_get(css);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004307 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004308}
4309
Tejun Heoeb954192013-08-08 20:11:23 -04004310static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004311{
Tejun Heoeb954192013-08-08 20:11:23 -04004312 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004313 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004314
4315 /*
4316 * Unregister events and notify userspace.
4317 * Notify userspace about cgroup removing only after rmdir of cgroup
4318 * directory to avoid race between userspace and kernelspace.
4319 */
Tejun Heofba94802013-11-22 18:20:43 -05004320 spin_lock(&memcg->event_list_lock);
4321 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004322 list_del_init(&event->list);
4323 schedule_work(&event->remove);
4324 }
Tejun Heofba94802013-11-22 18:20:43 -05004325 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004326
Roman Gushchin63677c742017-09-06 16:21:47 -07004327 memcg->low = 0;
4328
Johannes Weiner567e9ab2016-01-20 15:02:24 -08004329 memcg_offline_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004330 wb_memcg_offline(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004331
4332 mem_cgroup_id_put(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004333}
4334
Vladimir Davydov6df38682015-12-29 14:54:10 -08004335static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4336{
4337 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4338
4339 invalidate_reclaim_iterators(memcg);
4340}
4341
Tejun Heoeb954192013-08-08 20:11:23 -04004342static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004343{
Tejun Heoeb954192013-08-08 20:11:23 -04004344 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004345
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004346 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004347 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004348
Johannes Weiner0db15292016-01-20 15:02:50 -08004349 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004350 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004351
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004352 vmpressure_cleanup(&memcg->vmpressure);
4353 cancel_work_sync(&memcg->high_work);
4354 mem_cgroup_remove_from_trees(memcg);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004355 memcg_free_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004356 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004357}
4358
Tejun Heo1ced9532014-07-08 18:02:57 -04004359/**
4360 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4361 * @css: the target css
4362 *
4363 * Reset the states of the mem_cgroup associated with @css. This is
4364 * invoked when the userland requests disabling on the default hierarchy
4365 * but the memcg is pinned through dependency. The memcg should stop
4366 * applying policies and should revert to the vanilla state as it may be
4367 * made visible again.
4368 *
4369 * The current implementation only resets the essential configurations.
4370 * This needs to be expanded to cover all the visible parts.
4371 */
4372static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4373{
4374 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4375
Vladimir Davydovd334c9bc2016-03-17 14:19:38 -07004376 page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4377 page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4378 page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4379 page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4380 page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
Johannes Weiner241994ed2015-02-11 15:26:06 -08004381 memcg->low = 0;
4382 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004383 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004384 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004385}
4386
Daisuke Nishimura02491442010-03-10 15:22:17 -08004387#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004388/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004389static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004390{
Johannes Weiner05b84302014-08-06 16:05:59 -07004391 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004392
Mel Gormand0164ad2015-11-06 16:28:21 -08004393 /* Try a single bulk charge without reclaim first, kswapd may wake */
4394 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004395 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004396 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004397 return ret;
4398 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004399
David Rientjes36745342017-01-24 15:18:10 -08004400 /* Try charges one by one with reclaim, but do not retry */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004401 while (count--) {
David Rientjes36745342017-01-24 15:18:10 -08004402 ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004403 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004404 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004405 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004406 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004407 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004408 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004409}
4410
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004411union mc_target {
4412 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004413 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004414};
4415
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004416enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004417 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004418 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004419 MC_TARGET_SWAP,
Jérôme Glissec733a822017-09-08 16:11:54 -07004420 MC_TARGET_DEVICE,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004421};
4422
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004423static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4424 unsigned long addr, pte_t ptent)
4425{
Jérôme Glissec733a822017-09-08 16:11:54 -07004426 struct page *page = _vm_normal_page(vma, addr, ptent, true);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004427
4428 if (!page || !page_mapped(page))
4429 return NULL;
4430 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004431 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004432 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004433 } else {
4434 if (!(mc.flags & MOVE_FILE))
4435 return NULL;
4436 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004437 if (!get_page_unless_zero(page))
4438 return NULL;
4439
4440 return page;
4441}
4442
Jérôme Glissec733a822017-09-08 16:11:54 -07004443#if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004444static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004445 pte_t ptent, swp_entry_t *entry)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004446{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004447 struct page *page = NULL;
4448 swp_entry_t ent = pte_to_swp_entry(ptent);
4449
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004450 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004451 return NULL;
Jérôme Glissec733a822017-09-08 16:11:54 -07004452
4453 /*
4454 * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to
4455 * a device and because they are not accessible by CPU they are store
4456 * as special swap entry in the CPU page table.
4457 */
4458 if (is_device_private_entry(ent)) {
4459 page = device_private_entry_to_page(ent);
4460 /*
4461 * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have
4462 * a refcount of 1 when free (unlike normal page)
4463 */
4464 if (!page_ref_add_unless(page, 1, 1))
4465 return NULL;
4466 return page;
4467 }
4468
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004469 /*
4470 * Because lookup_swap_cache() updates some statistics counter,
4471 * we call find_get_page() with swapper_space directly.
4472 */
Huang Yingf6ab1f72016-10-07 17:00:21 -07004473 page = find_get_page(swap_address_space(ent), swp_offset(ent));
Johannes Weiner7941d212016-01-14 15:21:23 -08004474 if (do_memsw_account())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004475 entry->val = ent.val;
4476
4477 return page;
4478}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004479#else
4480static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004481 pte_t ptent, swp_entry_t *entry)
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004482{
4483 return NULL;
4484}
4485#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004486
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004487static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4488 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4489{
4490 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004491 struct address_space *mapping;
4492 pgoff_t pgoff;
4493
4494 if (!vma->vm_file) /* anonymous vma */
4495 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004496 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004497 return NULL;
4498
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004499 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004500 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004501
4502 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004503#ifdef CONFIG_SWAP
4504 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004505 if (shmem_mapping(mapping)) {
4506 page = find_get_entry(mapping, pgoff);
4507 if (radix_tree_exceptional_entry(page)) {
4508 swp_entry_t swp = radix_to_swp_entry(page);
Johannes Weiner7941d212016-01-14 15:21:23 -08004509 if (do_memsw_account())
Johannes Weiner139b6a62014-05-06 12:50:05 -07004510 *entry = swp;
Huang Yingf6ab1f72016-10-07 17:00:21 -07004511 page = find_get_page(swap_address_space(swp),
4512 swp_offset(swp));
Johannes Weiner139b6a62014-05-06 12:50:05 -07004513 }
4514 } else
4515 page = find_get_page(mapping, pgoff);
4516#else
4517 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004518#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004519 return page;
4520}
4521
Chen Gangb1b0dea2015-04-14 15:47:35 -07004522/**
4523 * mem_cgroup_move_account - move account of the page
4524 * @page: the page
Li RongQing25843c22016-07-26 15:26:56 -07004525 * @compound: charge the page as compound or small page
Chen Gangb1b0dea2015-04-14 15:47:35 -07004526 * @from: mem_cgroup which the page is moved from.
4527 * @to: mem_cgroup which the page is moved to. @from != @to.
4528 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08004529 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07004530 *
4531 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4532 * from old cgroup.
4533 */
4534static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004535 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07004536 struct mem_cgroup *from,
4537 struct mem_cgroup *to)
4538{
4539 unsigned long flags;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004540 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004541 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004542 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004543
4544 VM_BUG_ON(from == to);
4545 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004546 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07004547
4548 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07004549 * Prevent mem_cgroup_migrate() from looking at
Hugh Dickins45637ba2015-11-05 18:49:40 -08004550 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07004551 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004552 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004553 if (!trylock_page(page))
4554 goto out;
4555
4556 ret = -EINVAL;
4557 if (page->mem_cgroup != from)
4558 goto out_unlock;
4559
Greg Thelenc4843a72015-05-22 17:13:16 -04004560 anon = PageAnon(page);
4561
Chen Gangb1b0dea2015-04-14 15:47:35 -07004562 spin_lock_irqsave(&from->move_lock, flags);
4563
Greg Thelenc4843a72015-05-22 17:13:16 -04004564 if (!anon && page_mapped(page)) {
Johannes Weiner71cd3112017-05-03 14:55:13 -07004565 __this_cpu_sub(from->stat->count[NR_FILE_MAPPED], nr_pages);
4566 __this_cpu_add(to->stat->count[NR_FILE_MAPPED], nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004567 }
4568
Greg Thelenc4843a72015-05-22 17:13:16 -04004569 /*
4570 * move_lock grabbed above and caller set from->moving_account, so
Johannes Weinerccda7f42017-05-03 14:55:16 -07004571 * mod_memcg_page_state will serialize updates to PageDirty.
Greg Thelenc4843a72015-05-22 17:13:16 -04004572 * So mapping should be stable for dirty pages.
4573 */
4574 if (!anon && PageDirty(page)) {
4575 struct address_space *mapping = page_mapping(page);
4576
4577 if (mapping_cap_account_dirty(mapping)) {
Johannes Weiner71cd3112017-05-03 14:55:13 -07004578 __this_cpu_sub(from->stat->count[NR_FILE_DIRTY],
Greg Thelenc4843a72015-05-22 17:13:16 -04004579 nr_pages);
Johannes Weiner71cd3112017-05-03 14:55:13 -07004580 __this_cpu_add(to->stat->count[NR_FILE_DIRTY],
Greg Thelenc4843a72015-05-22 17:13:16 -04004581 nr_pages);
4582 }
4583 }
4584
Chen Gangb1b0dea2015-04-14 15:47:35 -07004585 if (PageWriteback(page)) {
Johannes Weiner71cd3112017-05-03 14:55:13 -07004586 __this_cpu_sub(from->stat->count[NR_WRITEBACK], nr_pages);
4587 __this_cpu_add(to->stat->count[NR_WRITEBACK], nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004588 }
4589
4590 /*
4591 * It is safe to change page->mem_cgroup here because the page
4592 * is referenced, charged, and isolated - we can't race with
4593 * uncharging, charging, migration, or LRU putback.
4594 */
4595
4596 /* caller should have done css_get */
4597 page->mem_cgroup = to;
4598 spin_unlock_irqrestore(&from->move_lock, flags);
4599
4600 ret = 0;
4601
4602 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004603 mem_cgroup_charge_statistics(to, page, compound, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004604 memcg_check_events(to, page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004605 mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004606 memcg_check_events(from, page);
4607 local_irq_enable();
4608out_unlock:
4609 unlock_page(page);
4610out:
4611 return ret;
4612}
4613
Li RongQing7cf78062016-05-27 14:27:46 -07004614/**
4615 * get_mctgt_type - get target type of moving charge
4616 * @vma: the vma the pte to be checked belongs
4617 * @addr: the address corresponding to the pte to be checked
4618 * @ptent: the pte to be checked
4619 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4620 *
4621 * Returns
4622 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4623 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4624 * move charge. if @target is not NULL, the page is stored in target->page
4625 * with extra refcnt got(Callers should handle it).
4626 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4627 * target for charge migration. if @target is not NULL, the entry is stored
4628 * in target->ent.
Jérôme Glissedf6ad692017-09-08 16:12:24 -07004629 * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PUBLIC
4630 * or MEMORY_DEVICE_PRIVATE (so ZONE_DEVICE page and thus not on the lru).
4631 * For now we such page is charge like a regular page would be as for all
4632 * intent and purposes it is just special memory taking the place of a
4633 * regular page.
Jérôme Glissec733a822017-09-08 16:11:54 -07004634 *
4635 * See Documentations/vm/hmm.txt and include/linux/hmm.h
Li RongQing7cf78062016-05-27 14:27:46 -07004636 *
4637 * Called with pte lock held.
4638 */
4639
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004640static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004641 unsigned long addr, pte_t ptent, union mc_target *target)
4642{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004643 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004644 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004645 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004646
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004647 if (pte_present(ptent))
4648 page = mc_handle_present_pte(vma, addr, ptent);
4649 else if (is_swap_pte(ptent))
Li RongQing48406ef2016-07-26 15:22:14 -07004650 page = mc_handle_swap_pte(vma, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004651 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004652 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004653
4654 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004655 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004656 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004657 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004658 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004659 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004660 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004661 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004662 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004663 ret = MC_TARGET_PAGE;
Jérôme Glissedf6ad692017-09-08 16:12:24 -07004664 if (is_device_private_page(page) ||
4665 is_device_public_page(page))
Jérôme Glissec733a822017-09-08 16:11:54 -07004666 ret = MC_TARGET_DEVICE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004667 if (target)
4668 target->page = page;
4669 }
4670 if (!ret || !target)
4671 put_page(page);
4672 }
Huang Ying3e14a572017-09-06 16:22:37 -07004673 /*
4674 * There is a swap entry and a page doesn't exist or isn't charged.
4675 * But we cannot move a tail-page in a THP.
4676 */
4677 if (ent.val && !ret && (!page || !PageTransCompound(page)) &&
Li Zefan34c00c32013-09-23 16:56:01 +08004678 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004679 ret = MC_TARGET_SWAP;
4680 if (target)
4681 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004682 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004683 return ret;
4684}
4685
Naoya Horiguchi12724852012-03-21 16:34:28 -07004686#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4687/*
Huang Yingd6810d72017-09-06 16:22:45 -07004688 * We don't consider PMD mapped swapping or file mapped pages because THP does
4689 * not support them for now.
Naoya Horiguchi12724852012-03-21 16:34:28 -07004690 * Caller should make sure that pmd_trans_huge(pmd) is true.
4691 */
4692static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4693 unsigned long addr, pmd_t pmd, union mc_target *target)
4694{
4695 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004696 enum mc_target_type ret = MC_TARGET_NONE;
4697
Zi Yan84c3fc42017-09-08 16:11:01 -07004698 if (unlikely(is_swap_pmd(pmd))) {
4699 VM_BUG_ON(thp_migration_supported() &&
4700 !is_pmd_migration_entry(pmd));
4701 return ret;
4702 }
Naoya Horiguchi12724852012-03-21 16:34:28 -07004703 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004704 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004705 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004706 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004707 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004708 ret = MC_TARGET_PAGE;
4709 if (target) {
4710 get_page(page);
4711 target->page = page;
4712 }
4713 }
4714 return ret;
4715}
4716#else
4717static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4718 unsigned long addr, pmd_t pmd, union mc_target *target)
4719{
4720 return MC_TARGET_NONE;
4721}
4722#endif
4723
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004724static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4725 unsigned long addr, unsigned long end,
4726 struct mm_walk *walk)
4727{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004728 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004729 pte_t *pte;
4730 spinlock_t *ptl;
4731
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004732 ptl = pmd_trans_huge_lock(pmd, vma);
4733 if (ptl) {
Jérôme Glissec733a822017-09-08 16:11:54 -07004734 /*
4735 * Note their can not be MC_TARGET_DEVICE for now as we do not
4736 * support transparent huge page with MEMORY_DEVICE_PUBLIC or
4737 * MEMORY_DEVICE_PRIVATE but this might change.
4738 */
Naoya Horiguchi12724852012-03-21 16:34:28 -07004739 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4740 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004741 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004742 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004743 }
Dave Hansen03319322011-03-22 16:32:56 -07004744
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004745 if (pmd_trans_unstable(pmd))
4746 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004747 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4748 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004749 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004750 mc.precharge++; /* increment precharge temporarily */
4751 pte_unmap_unlock(pte - 1, ptl);
4752 cond_resched();
4753
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004754 return 0;
4755}
4756
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004757static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4758{
4759 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004760
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004761 struct mm_walk mem_cgroup_count_precharge_walk = {
4762 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4763 .mm = mm,
4764 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004765 down_read(&mm->mmap_sem);
James Morse0247f3f2016-10-07 17:00:12 -07004766 walk_page_range(0, mm->highest_vm_end,
4767 &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004768 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004769
4770 precharge = mc.precharge;
4771 mc.precharge = 0;
4772
4773 return precharge;
4774}
4775
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004776static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4777{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004778 unsigned long precharge = mem_cgroup_count_precharge(mm);
4779
4780 VM_BUG_ON(mc.moving_task);
4781 mc.moving_task = current;
4782 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004783}
4784
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004785/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4786static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004787{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004788 struct mem_cgroup *from = mc.from;
4789 struct mem_cgroup *to = mc.to;
4790
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004791 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004792 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004793 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004794 mc.precharge = 0;
4795 }
4796 /*
4797 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4798 * we must uncharge here.
4799 */
4800 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004801 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004802 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004803 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004804 /* we must fixup refcnts and charges */
4805 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004806 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004807 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004808 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004809
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004810 mem_cgroup_id_put_many(mc.from, mc.moved_swap);
4811
Johannes Weiner05b84302014-08-06 16:05:59 -07004812 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004813 * we charged both to->memory and to->memsw, so we
4814 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004815 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004816 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004817 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004818
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004819 mem_cgroup_id_get_many(mc.to, mc.moved_swap);
4820 css_put_many(&mc.to->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004821
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004822 mc.moved_swap = 0;
4823 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004824 memcg_oom_recover(from);
4825 memcg_oom_recover(to);
4826 wake_up_all(&mc.waitq);
4827}
4828
4829static void mem_cgroup_clear_mc(void)
4830{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004831 struct mm_struct *mm = mc.mm;
4832
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004833 /*
4834 * we must clear moving_task before waking up waiters at the end of
4835 * task migration.
4836 */
4837 mc.moving_task = NULL;
4838 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004839 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004840 mc.from = NULL;
4841 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04004842 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004843 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004844
4845 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004846}
4847
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004848static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004849{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004850 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07004851 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07004852 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04004853 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004854 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004855 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004856 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004857
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004858 /* charge immigration isn't supported on the default hierarchy */
4859 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07004860 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004861
Tejun Heo4530edd2015-09-11 15:00:19 -04004862 /*
4863 * Multi-process migrations only happen on the default hierarchy
4864 * where charge immigration is not used. Perform charge
4865 * immigration if @tset contains a leader and whine if there are
4866 * multiple.
4867 */
4868 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004869 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04004870 WARN_ON_ONCE(p);
4871 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004872 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04004873 }
4874 if (!p)
4875 return 0;
4876
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004877 /*
4878 * We are now commited to this value whatever it is. Changes in this
4879 * tunable will only affect upcoming migrations, not the current one.
4880 * So we need to save it, and keep it going.
4881 */
4882 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4883 if (!move_flags)
4884 return 0;
4885
Tejun Heo9f2115f2015-09-08 15:01:10 -07004886 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004887
Tejun Heo9f2115f2015-09-08 15:01:10 -07004888 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004889
Tejun Heo9f2115f2015-09-08 15:01:10 -07004890 mm = get_task_mm(p);
4891 if (!mm)
4892 return 0;
4893 /* We move charges only when we move a owner of the mm */
4894 if (mm->owner == p) {
4895 VM_BUG_ON(mc.from);
4896 VM_BUG_ON(mc.to);
4897 VM_BUG_ON(mc.precharge);
4898 VM_BUG_ON(mc.moved_charge);
4899 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004900
Tejun Heo9f2115f2015-09-08 15:01:10 -07004901 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004902 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004903 mc.from = from;
4904 mc.to = memcg;
4905 mc.flags = move_flags;
4906 spin_unlock(&mc.lock);
4907 /* We set mc.moving_task later */
4908
4909 ret = mem_cgroup_precharge_mc(mm);
4910 if (ret)
4911 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004912 } else {
4913 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004914 }
4915 return ret;
4916}
4917
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004918static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004919{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004920 if (mc.to)
4921 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004922}
4923
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004924static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4925 unsigned long addr, unsigned long end,
4926 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004927{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004928 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004929 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004930 pte_t *pte;
4931 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004932 enum mc_target_type target_type;
4933 union mc_target target;
4934 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004935
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004936 ptl = pmd_trans_huge_lock(pmd, vma);
4937 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004938 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004939 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004940 return 0;
4941 }
4942 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4943 if (target_type == MC_TARGET_PAGE) {
4944 page = target.page;
4945 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004946 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08004947 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004948 mc.precharge -= HPAGE_PMD_NR;
4949 mc.moved_charge += HPAGE_PMD_NR;
4950 }
4951 putback_lru_page(page);
4952 }
4953 put_page(page);
Jérôme Glissec733a822017-09-08 16:11:54 -07004954 } else if (target_type == MC_TARGET_DEVICE) {
4955 page = target.page;
4956 if (!mem_cgroup_move_account(page, true,
4957 mc.from, mc.to)) {
4958 mc.precharge -= HPAGE_PMD_NR;
4959 mc.moved_charge += HPAGE_PMD_NR;
4960 }
4961 put_page(page);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004962 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004963 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004964 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004965 }
4966
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004967 if (pmd_trans_unstable(pmd))
4968 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004969retry:
4970 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4971 for (; addr != end; addr += PAGE_SIZE) {
4972 pte_t ptent = *(pte++);
Jérôme Glissec733a822017-09-08 16:11:54 -07004973 bool device = false;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004974 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004975
4976 if (!mc.precharge)
4977 break;
4978
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004979 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Jérôme Glissec733a822017-09-08 16:11:54 -07004980 case MC_TARGET_DEVICE:
4981 device = true;
4982 /* fall through */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004983 case MC_TARGET_PAGE:
4984 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08004985 /*
4986 * We can have a part of the split pmd here. Moving it
4987 * can be done but it would be too convoluted so simply
4988 * ignore such a partial THP and keep it in original
4989 * memcg. There should be somebody mapping the head.
4990 */
4991 if (PageTransCompound(page))
4992 goto put;
Jérôme Glissec733a822017-09-08 16:11:54 -07004993 if (!device && isolate_lru_page(page))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004994 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004995 if (!mem_cgroup_move_account(page, false,
4996 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004997 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004998 /* we uncharge from mc.from later. */
4999 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005000 }
Jérôme Glissec733a822017-09-08 16:11:54 -07005001 if (!device)
5002 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005003put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005004 put_page(page);
5005 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005006 case MC_TARGET_SWAP:
5007 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07005008 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005009 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005010 /* we fixup refcnts and charges later. */
5011 mc.moved_swap++;
5012 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08005013 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005014 default:
5015 break;
5016 }
5017 }
5018 pte_unmap_unlock(pte - 1, ptl);
5019 cond_resched();
5020
5021 if (addr != end) {
5022 /*
5023 * We have consumed all precharges we got in can_attach().
5024 * We try charge one by one, but don't do any additional
5025 * charges to mc.to if we have failed in charge once in attach()
5026 * phase.
5027 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005028 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005029 if (!ret)
5030 goto retry;
5031 }
5032
5033 return ret;
5034}
5035
Tejun Heo264a0ae2016-04-21 19:09:02 -04005036static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005037{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005038 struct mm_walk mem_cgroup_move_charge_walk = {
5039 .pmd_entry = mem_cgroup_move_charge_pte_range,
Tejun Heo264a0ae2016-04-21 19:09:02 -04005040 .mm = mc.mm,
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005041 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005042
5043 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08005044 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07005045 * Signal lock_page_memcg() to take the memcg's move_lock
5046 * while we're moving its pages to another memcg. Then wait
5047 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08005048 */
5049 atomic_inc(&mc.from->moving_account);
5050 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005051retry:
Tejun Heo264a0ae2016-04-21 19:09:02 -04005052 if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005053 /*
5054 * Someone who are holding the mmap_sem might be waiting in
5055 * waitq. So we cancel all extra charges, wake up all waiters,
5056 * and retry. Because we cancel precharges, we might not be able
5057 * to move enough charges, but moving charge is a best-effort
5058 * feature anyway, so it wouldn't be a big problem.
5059 */
5060 __mem_cgroup_clear_mc();
5061 cond_resched();
5062 goto retry;
5063 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005064 /*
5065 * When we have consumed all precharges and failed in doing
5066 * additional charge, the page walk just aborts.
5067 */
James Morse0247f3f2016-10-07 17:00:12 -07005068 walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk);
5069
Tejun Heo264a0ae2016-04-21 19:09:02 -04005070 up_read(&mc.mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08005071 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005072}
5073
Tejun Heo264a0ae2016-04-21 19:09:02 -04005074static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08005075{
Tejun Heo264a0ae2016-04-21 19:09:02 -04005076 if (mc.to) {
5077 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005078 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04005079 }
Balbir Singh67e465a2008-02-07 00:13:54 -08005080}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005081#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005082static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005083{
5084 return 0;
5085}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005086static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005087{
5088}
Tejun Heo264a0ae2016-04-21 19:09:02 -04005089static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005090{
5091}
5092#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005093
Tejun Heof00baae2013-04-15 13:41:15 -07005094/*
5095 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005096 * to verify whether we're attached to the default hierarchy on each mount
5097 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005098 */
Tejun Heoeb954192013-08-08 20:11:23 -04005099static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005100{
5101 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005102 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005103 * guarantees that @root doesn't have any children, so turning it
5104 * on for the root memcg is enough.
5105 */
Tejun Heo9e10a132015-09-18 11:56:28 -04005106 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee5902015-03-12 16:26:19 -07005107 root_mem_cgroup->use_hierarchy = true;
5108 else
5109 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005110}
5111
Johannes Weiner241994ed2015-02-11 15:26:06 -08005112static u64 memory_current_read(struct cgroup_subsys_state *css,
5113 struct cftype *cft)
5114{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005115 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5116
5117 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005118}
5119
5120static int memory_low_show(struct seq_file *m, void *v)
5121{
5122 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005123 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005124
5125 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005126 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005127 else
5128 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5129
5130 return 0;
5131}
5132
5133static ssize_t memory_low_write(struct kernfs_open_file *of,
5134 char *buf, size_t nbytes, loff_t off)
5135{
5136 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5137 unsigned long low;
5138 int err;
5139
5140 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005141 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005142 if (err)
5143 return err;
5144
5145 memcg->low = low;
5146
5147 return nbytes;
5148}
5149
5150static int memory_high_show(struct seq_file *m, void *v)
5151{
5152 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005153 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005154
5155 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005156 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005157 else
5158 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5159
5160 return 0;
5161}
5162
5163static ssize_t memory_high_write(struct kernfs_open_file *of,
5164 char *buf, size_t nbytes, loff_t off)
5165{
5166 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner588083b2016-03-17 14:20:25 -07005167 unsigned long nr_pages;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005168 unsigned long high;
5169 int err;
5170
5171 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005172 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005173 if (err)
5174 return err;
5175
5176 memcg->high = high;
5177
Johannes Weiner588083b2016-03-17 14:20:25 -07005178 nr_pages = page_counter_read(&memcg->memory);
5179 if (nr_pages > high)
5180 try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5181 GFP_KERNEL, true);
5182
Tejun Heo2529bb32015-05-22 18:23:34 -04005183 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005184 return nbytes;
5185}
5186
5187static int memory_max_show(struct seq_file *m, void *v)
5188{
5189 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005190 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005191
5192 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005193 seq_puts(m, "max\n");
Johannes Weiner241994ed2015-02-11 15:26:06 -08005194 else
5195 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5196
5197 return 0;
5198}
5199
5200static ssize_t memory_max_write(struct kernfs_open_file *of,
5201 char *buf, size_t nbytes, loff_t off)
5202{
5203 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005204 unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5205 bool drained = false;
Johannes Weiner241994ed2015-02-11 15:26:06 -08005206 unsigned long max;
5207 int err;
5208
5209 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005210 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005211 if (err)
5212 return err;
5213
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005214 xchg(&memcg->memory.limit, max);
5215
5216 for (;;) {
5217 unsigned long nr_pages = page_counter_read(&memcg->memory);
5218
5219 if (nr_pages <= max)
5220 break;
5221
5222 if (signal_pending(current)) {
5223 err = -EINTR;
5224 break;
5225 }
5226
5227 if (!drained) {
5228 drain_all_stock(memcg);
5229 drained = true;
5230 continue;
5231 }
5232
5233 if (nr_reclaims) {
5234 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5235 GFP_KERNEL, true))
5236 nr_reclaims--;
5237 continue;
5238 }
5239
Johannes Weiner31176c72017-05-03 14:55:07 -07005240 mem_cgroup_event(memcg, MEMCG_OOM);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005241 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5242 break;
5243 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08005244
Tejun Heo2529bb32015-05-22 18:23:34 -04005245 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08005246 return nbytes;
5247}
5248
5249static int memory_events_show(struct seq_file *m, void *v)
5250{
5251 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5252
Johannes Weinerccda7f42017-05-03 14:55:16 -07005253 seq_printf(m, "low %lu\n", memcg_sum_events(memcg, MEMCG_LOW));
5254 seq_printf(m, "high %lu\n", memcg_sum_events(memcg, MEMCG_HIGH));
5255 seq_printf(m, "max %lu\n", memcg_sum_events(memcg, MEMCG_MAX));
5256 seq_printf(m, "oom %lu\n", memcg_sum_events(memcg, MEMCG_OOM));
Konstantin Khlebnikov8e675f72017-07-06 15:40:28 -07005257 seq_printf(m, "oom_kill %lu\n", memcg_sum_events(memcg, OOM_KILL));
Johannes Weiner241994ed2015-02-11 15:26:06 -08005258
5259 return 0;
5260}
5261
Johannes Weiner587d9f72016-01-20 15:03:19 -08005262static int memory_stat_show(struct seq_file *m, void *v)
5263{
5264 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005265 unsigned long stat[MEMCG_NR_STAT];
5266 unsigned long events[MEMCG_NR_EVENTS];
Johannes Weiner587d9f72016-01-20 15:03:19 -08005267 int i;
5268
5269 /*
5270 * Provide statistics on the state of the memory subsystem as
5271 * well as cumulative event counters that show past behavior.
5272 *
5273 * This list is ordered following a combination of these gradients:
5274 * 1) generic big picture -> specifics and details
5275 * 2) reflecting userspace activity -> reflecting kernel heuristics
5276 *
5277 * Current memory state:
5278 */
5279
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005280 tree_stat(memcg, stat);
5281 tree_events(memcg, events);
5282
Johannes Weiner587d9f72016-01-20 15:03:19 -08005283 seq_printf(m, "anon %llu\n",
Johannes Weiner71cd3112017-05-03 14:55:13 -07005284 (u64)stat[MEMCG_RSS] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005285 seq_printf(m, "file %llu\n",
Johannes Weiner71cd3112017-05-03 14:55:13 -07005286 (u64)stat[MEMCG_CACHE] * PAGE_SIZE);
Vladimir Davydov12580e42016-03-17 14:17:38 -07005287 seq_printf(m, "kernel_stack %llu\n",
Andy Lutomirskiefdc9492016-07-28 15:48:17 -07005288 (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005289 seq_printf(m, "slab %llu\n",
Johannes Weiner32049292017-07-06 15:40:46 -07005290 (u64)(stat[NR_SLAB_RECLAIMABLE] +
5291 stat[NR_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
Johannes Weinerb2807f02016-01-20 15:03:22 -08005292 seq_printf(m, "sock %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005293 (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005294
Johannes Weiner9a4caf12017-05-03 14:52:45 -07005295 seq_printf(m, "shmem %llu\n",
Johannes Weiner71cd3112017-05-03 14:55:13 -07005296 (u64)stat[NR_SHMEM] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005297 seq_printf(m, "file_mapped %llu\n",
Johannes Weiner71cd3112017-05-03 14:55:13 -07005298 (u64)stat[NR_FILE_MAPPED] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005299 seq_printf(m, "file_dirty %llu\n",
Johannes Weiner71cd3112017-05-03 14:55:13 -07005300 (u64)stat[NR_FILE_DIRTY] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005301 seq_printf(m, "file_writeback %llu\n",
Johannes Weiner71cd3112017-05-03 14:55:13 -07005302 (u64)stat[NR_WRITEBACK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005303
5304 for (i = 0; i < NR_LRU_LISTS; i++) {
5305 struct mem_cgroup *mi;
5306 unsigned long val = 0;
5307
5308 for_each_mem_cgroup_tree(mi, memcg)
5309 val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5310 seq_printf(m, "%s %llu\n",
5311 mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5312 }
5313
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005314 seq_printf(m, "slab_reclaimable %llu\n",
Johannes Weiner32049292017-07-06 15:40:46 -07005315 (u64)stat[NR_SLAB_RECLAIMABLE] * PAGE_SIZE);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005316 seq_printf(m, "slab_unreclaimable %llu\n",
Johannes Weiner32049292017-07-06 15:40:46 -07005317 (u64)stat[NR_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005318
Johannes Weiner587d9f72016-01-20 15:03:19 -08005319 /* Accumulated memory events */
5320
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07005321 seq_printf(m, "pgfault %lu\n", events[PGFAULT]);
5322 seq_printf(m, "pgmajfault %lu\n", events[PGMAJFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005323
Roman Gushchin22621852017-07-06 15:40:25 -07005324 seq_printf(m, "pgrefill %lu\n", events[PGREFILL]);
5325 seq_printf(m, "pgscan %lu\n", events[PGSCAN_KSWAPD] +
5326 events[PGSCAN_DIRECT]);
5327 seq_printf(m, "pgsteal %lu\n", events[PGSTEAL_KSWAPD] +
5328 events[PGSTEAL_DIRECT]);
5329 seq_printf(m, "pgactivate %lu\n", events[PGACTIVATE]);
5330 seq_printf(m, "pgdeactivate %lu\n", events[PGDEACTIVATE]);
5331 seq_printf(m, "pglazyfree %lu\n", events[PGLAZYFREE]);
5332 seq_printf(m, "pglazyfreed %lu\n", events[PGLAZYFREED]);
5333
Johannes Weiner2a2e4882017-05-03 14:55:03 -07005334 seq_printf(m, "workingset_refault %lu\n",
Johannes Weiner71cd3112017-05-03 14:55:13 -07005335 stat[WORKINGSET_REFAULT]);
Johannes Weiner2a2e4882017-05-03 14:55:03 -07005336 seq_printf(m, "workingset_activate %lu\n",
Johannes Weiner71cd3112017-05-03 14:55:13 -07005337 stat[WORKINGSET_ACTIVATE]);
Johannes Weiner2a2e4882017-05-03 14:55:03 -07005338 seq_printf(m, "workingset_nodereclaim %lu\n",
Johannes Weiner71cd3112017-05-03 14:55:13 -07005339 stat[WORKINGSET_NODERECLAIM]);
Johannes Weiner2a2e4882017-05-03 14:55:03 -07005340
Johannes Weiner587d9f72016-01-20 15:03:19 -08005341 return 0;
5342}
5343
Johannes Weiner241994ed2015-02-11 15:26:06 -08005344static struct cftype memory_files[] = {
5345 {
5346 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005347 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005348 .read_u64 = memory_current_read,
5349 },
5350 {
5351 .name = "low",
5352 .flags = CFTYPE_NOT_ON_ROOT,
5353 .seq_show = memory_low_show,
5354 .write = memory_low_write,
5355 },
5356 {
5357 .name = "high",
5358 .flags = CFTYPE_NOT_ON_ROOT,
5359 .seq_show = memory_high_show,
5360 .write = memory_high_write,
5361 },
5362 {
5363 .name = "max",
5364 .flags = CFTYPE_NOT_ON_ROOT,
5365 .seq_show = memory_max_show,
5366 .write = memory_max_write,
5367 },
5368 {
5369 .name = "events",
5370 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04005371 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994ed2015-02-11 15:26:06 -08005372 .seq_show = memory_events_show,
5373 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08005374 {
5375 .name = "stat",
5376 .flags = CFTYPE_NOT_ON_ROOT,
5377 .seq_show = memory_stat_show,
5378 },
Johannes Weiner241994ed2015-02-11 15:26:06 -08005379 { } /* terminate */
5380};
5381
Tejun Heo073219e2014-02-08 10:36:58 -05005382struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005383 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005384 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005385 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08005386 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08005387 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005388 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005389 .can_attach = mem_cgroup_can_attach,
5390 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04005391 .post_attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07005392 .bind = mem_cgroup_bind,
Johannes Weiner241994ed2015-02-11 15:26:06 -08005393 .dfl_cftypes = memory_files,
5394 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005395 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005396};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005397
Johannes Weiner241994ed2015-02-11 15:26:06 -08005398/**
Johannes Weiner241994ed2015-02-11 15:26:06 -08005399 * mem_cgroup_low - check if memory consumption is below the normal range
Sean Christopherson34c81052017-07-10 15:48:05 -07005400 * @root: the top ancestor of the sub-tree being checked
Johannes Weiner241994ed2015-02-11 15:26:06 -08005401 * @memcg: the memory cgroup to check
5402 *
5403 * Returns %true if memory consumption of @memcg, and that of all
Sean Christopherson34c81052017-07-10 15:48:05 -07005404 * ancestors up to (but not including) @root, is below the normal range.
5405 *
5406 * @root is exclusive; it is never low when looked at directly and isn't
5407 * checked when traversing the hierarchy.
5408 *
5409 * Excluding @root enables using memory.low to prioritize memory usage
5410 * between cgroups within a subtree of the hierarchy that is limited by
5411 * memory.high or memory.max.
5412 *
5413 * For example, given cgroup A with children B and C:
5414 *
5415 * A
5416 * / \
5417 * B C
5418 *
5419 * and
5420 *
5421 * 1. A/memory.current > A/memory.high
5422 * 2. A/B/memory.current < A/B/memory.low
5423 * 3. A/C/memory.current >= A/C/memory.low
5424 *
5425 * As 'A' is high, i.e. triggers reclaim from 'A', and 'B' is low, we
5426 * should reclaim from 'C' until 'A' is no longer high or until we can
5427 * no longer reclaim from 'C'. If 'A', i.e. @root, isn't excluded by
5428 * mem_cgroup_low when reclaming from 'A', then 'B' won't be considered
5429 * low and we will reclaim indiscriminately from both 'B' and 'C'.
Johannes Weiner241994ed2015-02-11 15:26:06 -08005430 */
5431bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5432{
5433 if (mem_cgroup_disabled())
5434 return false;
5435
Sean Christopherson34c81052017-07-10 15:48:05 -07005436 if (!root)
5437 root = root_mem_cgroup;
5438 if (memcg == root)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005439 return false;
5440
Sean Christopherson34c81052017-07-10 15:48:05 -07005441 for (; memcg != root; memcg = parent_mem_cgroup(memcg)) {
Michal Hocko4e54ded2015-02-27 15:51:46 -08005442 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994ed2015-02-11 15:26:06 -08005443 return false;
5444 }
Sean Christopherson34c81052017-07-10 15:48:05 -07005445
Johannes Weiner241994ed2015-02-11 15:26:06 -08005446 return true;
5447}
5448
Johannes Weiner00501b52014-08-08 14:19:20 -07005449/**
5450 * mem_cgroup_try_charge - try charging a page
5451 * @page: page to charge
5452 * @mm: mm context of the victim
5453 * @gfp_mask: reclaim mode
5454 * @memcgp: charged memcg return
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 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5458 * pages according to @gfp_mask if necessary.
5459 *
5460 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5461 * Otherwise, an error code is returned.
5462 *
5463 * After page->mapping has been set up, the caller must finalize the
5464 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5465 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5466 */
5467int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005468 gfp_t gfp_mask, struct mem_cgroup **memcgp,
5469 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005470{
5471 struct mem_cgroup *memcg = NULL;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005472 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005473 int ret = 0;
5474
5475 if (mem_cgroup_disabled())
5476 goto out;
5477
5478 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005479 /*
5480 * Every swap fault against a single page tries to charge the
5481 * page, bail as early as possible. shmem_unuse() encounters
5482 * already charged pages, too. The USED bit is protected by
5483 * the page lock, which serializes swap cache removal, which
5484 * in turn serializes uncharging.
5485 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005486 VM_BUG_ON_PAGE(!PageLocked(page), page);
Huang Yingabe28952017-09-06 16:22:41 -07005487 if (compound_head(page)->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005488 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005489
Vladimir Davydov37e84352016-01-20 15:02:56 -08005490 if (do_swap_account) {
Vladimir Davydove993d902015-09-09 15:35:35 -07005491 swp_entry_t ent = { .val = page_private(page), };
5492 unsigned short id = lookup_swap_cgroup_id(ent);
5493
5494 rcu_read_lock();
5495 memcg = mem_cgroup_from_id(id);
5496 if (memcg && !css_tryget_online(&memcg->css))
5497 memcg = NULL;
5498 rcu_read_unlock();
5499 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005500 }
5501
Johannes Weiner00501b52014-08-08 14:19:20 -07005502 if (!memcg)
5503 memcg = get_mem_cgroup_from_mm(mm);
5504
5505 ret = try_charge(memcg, gfp_mask, nr_pages);
5506
5507 css_put(&memcg->css);
Johannes Weiner00501b52014-08-08 14:19:20 -07005508out:
5509 *memcgp = memcg;
5510 return ret;
5511}
5512
5513/**
5514 * mem_cgroup_commit_charge - commit a page charge
5515 * @page: page to charge
5516 * @memcg: memcg to charge the page to
5517 * @lrucare: page might be on LRU already
Li RongQing25843c22016-07-26 15:26:56 -07005518 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005519 *
5520 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5521 * after page->mapping has been set up. This must happen atomically
5522 * as part of the page instantiation, i.e. under the page table lock
5523 * for anonymous pages, under the page lock for page and swap cache.
5524 *
5525 * In addition, the page must not be on the LRU during the commit, to
5526 * prevent racing with task migration. If it might be, use @lrucare.
5527 *
5528 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5529 */
5530void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005531 bool lrucare, bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005532{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005533 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005534
5535 VM_BUG_ON_PAGE(!page->mapping, page);
5536 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5537
5538 if (mem_cgroup_disabled())
5539 return;
5540 /*
5541 * Swap faults will attempt to charge the same page multiple
5542 * times. But reuse_swap_page() might have removed the page
5543 * from swapcache already, so we can't check PageSwapCache().
5544 */
5545 if (!memcg)
5546 return;
5547
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005548 commit_charge(page, memcg, lrucare);
5549
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005550 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005551 mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005552 memcg_check_events(memcg, page);
5553 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005554
Johannes Weiner7941d212016-01-14 15:21:23 -08005555 if (do_memsw_account() && PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005556 swp_entry_t entry = { .val = page_private(page) };
5557 /*
5558 * The swap entry might not get freed for a long time,
5559 * let's not wait for it. The page already received a
5560 * memory+swap charge, drop the swap entry duplicate.
5561 */
Huang Ying38d8b4e2017-07-06 15:37:18 -07005562 mem_cgroup_uncharge_swap(entry, nr_pages);
Johannes Weiner00501b52014-08-08 14:19:20 -07005563 }
5564}
5565
5566/**
5567 * mem_cgroup_cancel_charge - cancel a page charge
5568 * @page: page to charge
5569 * @memcg: memcg to charge the page to
Li RongQing25843c22016-07-26 15:26:56 -07005570 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005571 *
5572 * Cancel a charge transaction started by mem_cgroup_try_charge().
5573 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005574void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5575 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005576{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005577 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005578
5579 if (mem_cgroup_disabled())
5580 return;
5581 /*
5582 * Swap faults will attempt to charge the same page multiple
5583 * times. But reuse_swap_page() might have removed the page
5584 * from swapcache already, so we can't check PageSwapCache().
5585 */
5586 if (!memcg)
5587 return;
5588
Johannes Weiner00501b52014-08-08 14:19:20 -07005589 cancel_charge(memcg, nr_pages);
5590}
5591
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07005592struct uncharge_gather {
5593 struct mem_cgroup *memcg;
5594 unsigned long pgpgout;
5595 unsigned long nr_anon;
5596 unsigned long nr_file;
5597 unsigned long nr_kmem;
5598 unsigned long nr_huge;
5599 unsigned long nr_shmem;
5600 struct page *dummy_page;
5601};
5602
5603static inline void uncharge_gather_clear(struct uncharge_gather *ug)
Johannes Weiner747db952014-08-08 14:19:24 -07005604{
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07005605 memset(ug, 0, sizeof(*ug));
5606}
5607
5608static void uncharge_batch(const struct uncharge_gather *ug)
5609{
5610 unsigned long nr_pages = ug->nr_anon + ug->nr_file + ug->nr_kmem;
Johannes Weiner747db952014-08-08 14:19:24 -07005611 unsigned long flags;
5612
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07005613 if (!mem_cgroup_is_root(ug->memcg)) {
5614 page_counter_uncharge(&ug->memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08005615 if (do_memsw_account())
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07005616 page_counter_uncharge(&ug->memcg->memsw, nr_pages);
5617 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem)
5618 page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem);
5619 memcg_oom_recover(ug->memcg);
Johannes Weinerce00a962014-09-05 08:43:57 -04005620 }
Johannes Weiner747db952014-08-08 14:19:24 -07005621
5622 local_irq_save(flags);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07005623 __this_cpu_sub(ug->memcg->stat->count[MEMCG_RSS], ug->nr_anon);
5624 __this_cpu_sub(ug->memcg->stat->count[MEMCG_CACHE], ug->nr_file);
5625 __this_cpu_sub(ug->memcg->stat->count[MEMCG_RSS_HUGE], ug->nr_huge);
5626 __this_cpu_sub(ug->memcg->stat->count[NR_SHMEM], ug->nr_shmem);
5627 __this_cpu_add(ug->memcg->stat->events[PGPGOUT], ug->pgpgout);
5628 __this_cpu_add(ug->memcg->stat->nr_page_events, nr_pages);
5629 memcg_check_events(ug->memcg, ug->dummy_page);
Johannes Weiner747db952014-08-08 14:19:24 -07005630 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005631
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07005632 if (!mem_cgroup_is_root(ug->memcg))
5633 css_put_many(&ug->memcg->css, nr_pages);
5634}
5635
5636static void uncharge_page(struct page *page, struct uncharge_gather *ug)
5637{
5638 VM_BUG_ON_PAGE(PageLRU(page), page);
5639 VM_BUG_ON_PAGE(!PageHWPoison(page) && page_count(page), page);
5640
5641 if (!page->mem_cgroup)
5642 return;
5643
5644 /*
5645 * Nobody should be changing or seriously looking at
5646 * page->mem_cgroup at this point, we have fully
5647 * exclusive access to the page.
5648 */
5649
5650 if (ug->memcg != page->mem_cgroup) {
5651 if (ug->memcg) {
5652 uncharge_batch(ug);
5653 uncharge_gather_clear(ug);
5654 }
5655 ug->memcg = page->mem_cgroup;
5656 }
5657
5658 if (!PageKmemcg(page)) {
5659 unsigned int nr_pages = 1;
5660
5661 if (PageTransHuge(page)) {
5662 nr_pages <<= compound_order(page);
5663 ug->nr_huge += nr_pages;
5664 }
5665 if (PageAnon(page))
5666 ug->nr_anon += nr_pages;
5667 else {
5668 ug->nr_file += nr_pages;
5669 if (PageSwapBacked(page))
5670 ug->nr_shmem += nr_pages;
5671 }
5672 ug->pgpgout++;
5673 } else {
5674 ug->nr_kmem += 1 << compound_order(page);
5675 __ClearPageKmemcg(page);
5676 }
5677
5678 ug->dummy_page = page;
5679 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005680}
5681
5682static void uncharge_list(struct list_head *page_list)
5683{
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07005684 struct uncharge_gather ug;
Johannes Weiner747db952014-08-08 14:19:24 -07005685 struct list_head *next;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07005686
5687 uncharge_gather_clear(&ug);
Johannes Weiner747db952014-08-08 14:19:24 -07005688
Johannes Weiner8b592652016-03-17 14:20:31 -07005689 /*
5690 * Note that the list can be a single page->lru; hence the
5691 * do-while loop instead of a simple list_for_each_entry().
5692 */
Johannes Weiner747db952014-08-08 14:19:24 -07005693 next = page_list->next;
5694 do {
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07005695 struct page *page;
5696
Johannes Weiner747db952014-08-08 14:19:24 -07005697 page = list_entry(next, struct page, lru);
5698 next = page->lru.next;
5699
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07005700 uncharge_page(page, &ug);
Johannes Weiner747db952014-08-08 14:19:24 -07005701 } while (next != page_list);
5702
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07005703 if (ug.memcg)
5704 uncharge_batch(&ug);
Johannes Weiner747db952014-08-08 14:19:24 -07005705}
5706
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005707/**
5708 * mem_cgroup_uncharge - uncharge a page
5709 * @page: page to uncharge
5710 *
5711 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5712 * mem_cgroup_commit_charge().
5713 */
5714void mem_cgroup_uncharge(struct page *page)
5715{
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07005716 struct uncharge_gather ug;
5717
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005718 if (mem_cgroup_disabled())
5719 return;
5720
Johannes Weiner747db952014-08-08 14:19:24 -07005721 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005722 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005723 return;
5724
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07005725 uncharge_gather_clear(&ug);
5726 uncharge_page(page, &ug);
5727 uncharge_batch(&ug);
Johannes Weiner747db952014-08-08 14:19:24 -07005728}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005729
Johannes Weiner747db952014-08-08 14:19:24 -07005730/**
5731 * mem_cgroup_uncharge_list - uncharge a list of page
5732 * @page_list: list of pages to uncharge
5733 *
5734 * Uncharge a list of pages previously charged with
5735 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5736 */
5737void mem_cgroup_uncharge_list(struct list_head *page_list)
5738{
5739 if (mem_cgroup_disabled())
5740 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005741
Johannes Weiner747db952014-08-08 14:19:24 -07005742 if (!list_empty(page_list))
5743 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005744}
5745
5746/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005747 * mem_cgroup_migrate - charge a page's replacement
5748 * @oldpage: currently circulating page
5749 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005750 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005751 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5752 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005753 *
5754 * Both pages must be locked, @newpage->mapping must be set up.
5755 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005756void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005757{
Johannes Weiner29833312014-12-10 15:44:02 -08005758 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005759 unsigned int nr_pages;
5760 bool compound;
Tejun Heod93c4132016-06-24 14:49:54 -07005761 unsigned long flags;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005762
5763 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5764 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005765 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005766 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5767 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005768
5769 if (mem_cgroup_disabled())
5770 return;
5771
5772 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005773 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005774 return;
5775
Hugh Dickins45637ba2015-11-05 18:49:40 -08005776 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08005777 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005778 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005779 return;
5780
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005781 /* Force-charge the new page. The old one will be freed soon */
5782 compound = PageTransHuge(newpage);
5783 nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5784
5785 page_counter_charge(&memcg->memory, nr_pages);
5786 if (do_memsw_account())
5787 page_counter_charge(&memcg->memsw, nr_pages);
5788 css_get_many(&memcg->css, nr_pages);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005789
Johannes Weiner9cf76662016-03-15 14:57:58 -07005790 commit_charge(newpage, memcg, false);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005791
Tejun Heod93c4132016-06-24 14:49:54 -07005792 local_irq_save(flags);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005793 mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5794 memcg_check_events(memcg, newpage);
Tejun Heod93c4132016-06-24 14:49:54 -07005795 local_irq_restore(flags);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005796}
5797
Johannes Weineref129472016-01-14 15:21:34 -08005798DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08005799EXPORT_SYMBOL(memcg_sockets_enabled_key);
5800
Johannes Weiner2d758072016-10-07 17:00:58 -07005801void mem_cgroup_sk_alloc(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08005802{
5803 struct mem_cgroup *memcg;
5804
Johannes Weiner2d758072016-10-07 17:00:58 -07005805 if (!mem_cgroup_sockets_enabled)
5806 return;
5807
5808 /*
5809 * Socket cloning can throw us here with sk_memcg already
Johannes Weiner11092082016-01-14 15:21:26 -08005810 * filled. It won't however, necessarily happen from
5811 * process context. So the test for root memcg given
5812 * the current task's memcg won't help us in this case.
5813 *
5814 * Respecting the original socket's memcg is a better
5815 * decision in this case.
5816 */
5817 if (sk->sk_memcg) {
5818 BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5819 css_get(&sk->sk_memcg->css);
5820 return;
5821 }
5822
5823 rcu_read_lock();
5824 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005825 if (memcg == root_mem_cgroup)
5826 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08005827 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005828 goto out;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005829 if (css_tryget_online(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08005830 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005831out:
Johannes Weiner11092082016-01-14 15:21:26 -08005832 rcu_read_unlock();
5833}
Johannes Weiner11092082016-01-14 15:21:26 -08005834
Johannes Weiner2d758072016-10-07 17:00:58 -07005835void mem_cgroup_sk_free(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08005836{
Johannes Weiner2d758072016-10-07 17:00:58 -07005837 if (sk->sk_memcg)
5838 css_put(&sk->sk_memcg->css);
Johannes Weiner11092082016-01-14 15:21:26 -08005839}
5840
5841/**
5842 * mem_cgroup_charge_skmem - charge socket memory
5843 * @memcg: memcg to charge
5844 * @nr_pages: number of pages to charge
5845 *
5846 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5847 * @memcg's configured limit, %false if the charge had to be forced.
5848 */
5849bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5850{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005851 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08005852
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005853 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005854 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005855
Johannes Weiner0db15292016-01-20 15:02:50 -08005856 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5857 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005858 return true;
5859 }
Johannes Weiner0db15292016-01-20 15:02:50 -08005860 page_counter_charge(&memcg->tcpmem, nr_pages);
5861 memcg->tcpmem_pressure = 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005862 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08005863 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005864
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005865 /* Don't block in the packet receive path */
5866 if (in_softirq())
5867 gfp_mask = GFP_NOWAIT;
5868
Johannes Weinerb2807f02016-01-20 15:03:22 -08005869 this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5870
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005871 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5872 return true;
5873
5874 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005875 return false;
5876}
5877
5878/**
5879 * mem_cgroup_uncharge_skmem - uncharge socket memory
5880 * @memcg - memcg to uncharge
5881 * @nr_pages - number of pages to uncharge
5882 */
5883void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5884{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005885 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005886 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005887 return;
5888 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005889
Johannes Weinerb2807f02016-01-20 15:03:22 -08005890 this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5891
Roman Gushchin475d0482017-09-08 16:13:09 -07005892 refill_stock(memcg, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005893}
5894
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005895static int __init cgroup_memory(char *s)
5896{
5897 char *token;
5898
5899 while ((token = strsep(&s, ",")) != NULL) {
5900 if (!*token)
5901 continue;
5902 if (!strcmp(token, "nosocket"))
5903 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08005904 if (!strcmp(token, "nokmem"))
5905 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005906 }
5907 return 0;
5908}
5909__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08005910
Michal Hocko2d110852013-02-22 16:34:43 -08005911/*
Michal Hocko10813122013-02-22 16:35:41 -08005912 * subsys_initcall() for memory controller.
5913 *
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01005914 * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this
5915 * context because of lock dependencies (cgroup_lock -> cpu hotplug) but
5916 * basically everything that doesn't depend on a specific mem_cgroup structure
5917 * should be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005918 */
5919static int __init mem_cgroup_init(void)
5920{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005921 int cpu, node;
5922
Vladimir Davydov13583c32016-12-12 16:41:29 -08005923#ifndef CONFIG_SLOB
5924 /*
5925 * Kmem cache creation is mostly done with the slab_mutex held,
Tejun Heo17cc4df2017-02-22 15:41:36 -08005926 * so use a workqueue with limited concurrency to avoid stalling
5927 * all worker threads in case lots of cgroups are created and
5928 * destroyed simultaneously.
Vladimir Davydov13583c32016-12-12 16:41:29 -08005929 */
Tejun Heo17cc4df2017-02-22 15:41:36 -08005930 memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1);
5931 BUG_ON(!memcg_kmem_cache_wq);
Vladimir Davydov13583c32016-12-12 16:41:29 -08005932#endif
5933
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01005934 cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL,
5935 memcg_hotplug_cpu_dead);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005936
5937 for_each_possible_cpu(cpu)
5938 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5939 drain_local_stock);
5940
5941 for_each_node(node) {
5942 struct mem_cgroup_tree_per_node *rtpn;
Johannes Weiner95a045f2015-02-11 15:26:33 -08005943
5944 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5945 node_online(node) ? node : NUMA_NO_NODE);
5946
Mel Gormanef8f2322016-07-28 15:46:05 -07005947 rtpn->rb_root = RB_ROOT;
5948 spin_lock_init(&rtpn->lock);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005949 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5950 }
5951
Michal Hocko2d110852013-02-22 16:34:43 -08005952 return 0;
5953}
5954subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005955
5956#ifdef CONFIG_MEMCG_SWAP
Arnd Bergmann358c07f2016-08-25 15:17:08 -07005957static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
5958{
5959 while (!atomic_inc_not_zero(&memcg->id.ref)) {
5960 /*
5961 * The root cgroup cannot be destroyed, so it's refcount must
5962 * always be >= 1.
5963 */
5964 if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
5965 VM_BUG_ON(1);
5966 break;
5967 }
5968 memcg = parent_mem_cgroup(memcg);
5969 if (!memcg)
5970 memcg = root_mem_cgroup;
5971 }
5972 return memcg;
5973}
5974
Johannes Weiner21afa382015-02-11 15:26:36 -08005975/**
5976 * mem_cgroup_swapout - transfer a memsw charge to swap
5977 * @page: page whose memsw charge to transfer
5978 * @entry: swap entry to move the charge to
5979 *
5980 * Transfer the memsw charge of @page to @entry.
5981 */
5982void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5983{
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005984 struct mem_cgroup *memcg, *swap_memcg;
Huang Yingd6810d72017-09-06 16:22:45 -07005985 unsigned int nr_entries;
Johannes Weiner21afa382015-02-11 15:26:36 -08005986 unsigned short oldid;
5987
5988 VM_BUG_ON_PAGE(PageLRU(page), page);
5989 VM_BUG_ON_PAGE(page_count(page), page);
5990
Johannes Weiner7941d212016-01-14 15:21:23 -08005991 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08005992 return;
5993
5994 memcg = page->mem_cgroup;
5995
5996 /* Readahead page, never charged */
5997 if (!memcg)
5998 return;
5999
Vladimir Davydov1f47b612016-08-11 15:33:00 -07006000 /*
6001 * In case the memcg owning these pages has been offlined and doesn't
6002 * have an ID allocated to it anymore, charge the closest online
6003 * ancestor for the swap instead and transfer the memory+swap charge.
6004 */
6005 swap_memcg = mem_cgroup_id_get_online(memcg);
Huang Yingd6810d72017-09-06 16:22:45 -07006006 nr_entries = hpage_nr_pages(page);
6007 /* Get references for the tail pages, too */
6008 if (nr_entries > 1)
6009 mem_cgroup_id_get_many(swap_memcg, nr_entries - 1);
6010 oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg),
6011 nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08006012 VM_BUG_ON_PAGE(oldid, page);
Huang Yingd6810d72017-09-06 16:22:45 -07006013 mem_cgroup_swap_statistics(swap_memcg, nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08006014
6015 page->mem_cgroup = NULL;
6016
6017 if (!mem_cgroup_is_root(memcg))
Huang Yingd6810d72017-09-06 16:22:45 -07006018 page_counter_uncharge(&memcg->memory, nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08006019
Vladimir Davydov1f47b612016-08-11 15:33:00 -07006020 if (memcg != swap_memcg) {
6021 if (!mem_cgroup_is_root(swap_memcg))
Huang Yingd6810d72017-09-06 16:22:45 -07006022 page_counter_charge(&swap_memcg->memsw, nr_entries);
6023 page_counter_uncharge(&memcg->memsw, nr_entries);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07006024 }
6025
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07006026 /*
6027 * Interrupts should be disabled here because the caller holds the
6028 * mapping->tree_lock lock which is taken with interrupts-off. It is
6029 * important here to have the interrupts disabled because it is the
6030 * only synchronisation we have for udpating the per-CPU variables.
6031 */
6032 VM_BUG_ON(!irqs_disabled());
Huang Yingd6810d72017-09-06 16:22:45 -07006033 mem_cgroup_charge_statistics(memcg, page, PageTransHuge(page),
6034 -nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08006035 memcg_check_events(memcg, page);
Johannes Weiner73f576c2016-07-20 15:44:57 -07006036
6037 if (!mem_cgroup_is_root(memcg))
6038 css_put(&memcg->css);
Johannes Weiner21afa382015-02-11 15:26:36 -08006039}
6040
Huang Ying38d8b4e2017-07-06 15:37:18 -07006041/**
6042 * mem_cgroup_try_charge_swap - try charging swap space for a page
Vladimir Davydov37e84352016-01-20 15:02:56 -08006043 * @page: page being added to swap
6044 * @entry: swap entry to charge
6045 *
Huang Ying38d8b4e2017-07-06 15:37:18 -07006046 * Try to charge @page's memcg for the swap space at @entry.
Vladimir Davydov37e84352016-01-20 15:02:56 -08006047 *
6048 * Returns 0 on success, -ENOMEM on failure.
6049 */
6050int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
6051{
Huang Ying38d8b4e2017-07-06 15:37:18 -07006052 unsigned int nr_pages = hpage_nr_pages(page);
Vladimir Davydov37e84352016-01-20 15:02:56 -08006053 struct page_counter *counter;
Huang Ying38d8b4e2017-07-06 15:37:18 -07006054 struct mem_cgroup *memcg;
Vladimir Davydov37e84352016-01-20 15:02:56 -08006055 unsigned short oldid;
6056
6057 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
6058 return 0;
6059
6060 memcg = page->mem_cgroup;
6061
6062 /* Readahead page, never charged */
6063 if (!memcg)
6064 return 0;
6065
Vladimir Davydov1f47b612016-08-11 15:33:00 -07006066 memcg = mem_cgroup_id_get_online(memcg);
Vladimir Davydov37e84352016-01-20 15:02:56 -08006067
Vladimir Davydov1f47b612016-08-11 15:33:00 -07006068 if (!mem_cgroup_is_root(memcg) &&
Huang Ying38d8b4e2017-07-06 15:37:18 -07006069 !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) {
Vladimir Davydov1f47b612016-08-11 15:33:00 -07006070 mem_cgroup_id_put(memcg);
6071 return -ENOMEM;
6072 }
6073
Huang Ying38d8b4e2017-07-06 15:37:18 -07006074 /* Get references for the tail pages, too */
6075 if (nr_pages > 1)
6076 mem_cgroup_id_get_many(memcg, nr_pages - 1);
6077 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08006078 VM_BUG_ON_PAGE(oldid, page);
Huang Ying38d8b4e2017-07-06 15:37:18 -07006079 mem_cgroup_swap_statistics(memcg, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08006080
Vladimir Davydov37e84352016-01-20 15:02:56 -08006081 return 0;
6082}
6083
Johannes Weiner21afa382015-02-11 15:26:36 -08006084/**
Huang Ying38d8b4e2017-07-06 15:37:18 -07006085 * mem_cgroup_uncharge_swap - uncharge swap space
Johannes Weiner21afa382015-02-11 15:26:36 -08006086 * @entry: swap entry to uncharge
Huang Ying38d8b4e2017-07-06 15:37:18 -07006087 * @nr_pages: the amount of swap space to uncharge
Johannes Weiner21afa382015-02-11 15:26:36 -08006088 */
Huang Ying38d8b4e2017-07-06 15:37:18 -07006089void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages)
Johannes Weiner21afa382015-02-11 15:26:36 -08006090{
6091 struct mem_cgroup *memcg;
6092 unsigned short id;
6093
Vladimir Davydov37e84352016-01-20 15:02:56 -08006094 if (!do_swap_account)
Johannes Weiner21afa382015-02-11 15:26:36 -08006095 return;
6096
Huang Ying38d8b4e2017-07-06 15:37:18 -07006097 id = swap_cgroup_record(entry, 0, nr_pages);
Johannes Weiner21afa382015-02-11 15:26:36 -08006098 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07006099 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08006100 if (memcg) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08006101 if (!mem_cgroup_is_root(memcg)) {
6102 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Huang Ying38d8b4e2017-07-06 15:37:18 -07006103 page_counter_uncharge(&memcg->swap, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08006104 else
Huang Ying38d8b4e2017-07-06 15:37:18 -07006105 page_counter_uncharge(&memcg->memsw, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08006106 }
Huang Ying38d8b4e2017-07-06 15:37:18 -07006107 mem_cgroup_swap_statistics(memcg, -nr_pages);
6108 mem_cgroup_id_put_many(memcg, nr_pages);
Johannes Weiner21afa382015-02-11 15:26:36 -08006109 }
6110 rcu_read_unlock();
6111}
6112
Vladimir Davydovd8b38432016-01-20 15:03:07 -08006113long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
6114{
6115 long nr_swap_pages = get_nr_swap_pages();
6116
6117 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
6118 return nr_swap_pages;
6119 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
6120 nr_swap_pages = min_t(long, nr_swap_pages,
6121 READ_ONCE(memcg->swap.limit) -
6122 page_counter_read(&memcg->swap));
6123 return nr_swap_pages;
6124}
6125
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08006126bool mem_cgroup_swap_full(struct page *page)
6127{
6128 struct mem_cgroup *memcg;
6129
6130 VM_BUG_ON_PAGE(!PageLocked(page), page);
6131
6132 if (vm_swap_full())
6133 return true;
6134 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
6135 return false;
6136
6137 memcg = page->mem_cgroup;
6138 if (!memcg)
6139 return false;
6140
6141 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
6142 if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
6143 return true;
6144
6145 return false;
6146}
6147
Johannes Weiner21afa382015-02-11 15:26:36 -08006148/* for remember boot option*/
6149#ifdef CONFIG_MEMCG_SWAP_ENABLED
6150static int really_do_swap_account __initdata = 1;
6151#else
6152static int really_do_swap_account __initdata;
6153#endif
6154
6155static int __init enable_swap_account(char *s)
6156{
6157 if (!strcmp(s, "1"))
6158 really_do_swap_account = 1;
6159 else if (!strcmp(s, "0"))
6160 really_do_swap_account = 0;
6161 return 1;
6162}
6163__setup("swapaccount=", enable_swap_account);
6164
Vladimir Davydov37e84352016-01-20 15:02:56 -08006165static u64 swap_current_read(struct cgroup_subsys_state *css,
6166 struct cftype *cft)
6167{
6168 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6169
6170 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
6171}
6172
6173static int swap_max_show(struct seq_file *m, void *v)
6174{
6175 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
6176 unsigned long max = READ_ONCE(memcg->swap.limit);
6177
6178 if (max == PAGE_COUNTER_MAX)
6179 seq_puts(m, "max\n");
6180 else
6181 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
6182
6183 return 0;
6184}
6185
6186static ssize_t swap_max_write(struct kernfs_open_file *of,
6187 char *buf, size_t nbytes, loff_t off)
6188{
6189 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6190 unsigned long max;
6191 int err;
6192
6193 buf = strstrip(buf);
6194 err = page_counter_memparse(buf, "max", &max);
6195 if (err)
6196 return err;
6197
6198 mutex_lock(&memcg_limit_mutex);
6199 err = page_counter_limit(&memcg->swap, max);
6200 mutex_unlock(&memcg_limit_mutex);
6201 if (err)
6202 return err;
6203
6204 return nbytes;
6205}
6206
6207static struct cftype swap_files[] = {
6208 {
6209 .name = "swap.current",
6210 .flags = CFTYPE_NOT_ON_ROOT,
6211 .read_u64 = swap_current_read,
6212 },
6213 {
6214 .name = "swap.max",
6215 .flags = CFTYPE_NOT_ON_ROOT,
6216 .seq_show = swap_max_show,
6217 .write = swap_max_write,
6218 },
6219 { } /* terminate */
6220};
6221
Johannes Weiner21afa382015-02-11 15:26:36 -08006222static struct cftype memsw_cgroup_files[] = {
6223 {
6224 .name = "memsw.usage_in_bytes",
6225 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6226 .read_u64 = mem_cgroup_read_u64,
6227 },
6228 {
6229 .name = "memsw.max_usage_in_bytes",
6230 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6231 .write = mem_cgroup_reset,
6232 .read_u64 = mem_cgroup_read_u64,
6233 },
6234 {
6235 .name = "memsw.limit_in_bytes",
6236 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6237 .write = mem_cgroup_write,
6238 .read_u64 = mem_cgroup_read_u64,
6239 },
6240 {
6241 .name = "memsw.failcnt",
6242 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6243 .write = mem_cgroup_reset,
6244 .read_u64 = mem_cgroup_read_u64,
6245 },
6246 { }, /* terminate */
6247};
6248
6249static int __init mem_cgroup_swap_init(void)
6250{
6251 if (!mem_cgroup_disabled() && really_do_swap_account) {
6252 do_swap_account = 1;
Vladimir Davydov37e84352016-01-20 15:02:56 -08006253 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
6254 swap_files));
Johannes Weiner21afa382015-02-11 15:26:36 -08006255 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
6256 memsw_cgroup_files));
6257 }
6258 return 0;
6259}
6260subsys_initcall(mem_cgroup_swap_init);
6261
6262#endif /* CONFIG_MEMCG_SWAP */