blob: fdc790a55ac11aea26f2efa8a93ea93c88f48168 [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08009 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080013 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
Johannes Weiner1575e682015-04-14 15:44:51 -070017 * Native page reclaim
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
22 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -080023 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
27 *
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
32 */
33
Johannes Weiner3e32cb22014-12-10 15:42:31 -080034#include <linux/page_counter.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080035#include <linux/memcontrol.h>
36#include <linux/cgroup.h>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080037#include <linux/mm.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080038#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080039#include <linux/pagemap.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080040#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080041#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080042#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080043#include <linux/bit_spinlock.h>
44#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070045#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040046#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080047#include <linux/mutex.h>
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -070048#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070049#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080051#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080052#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080053#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050054#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080055#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080056#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080057#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070058#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070059#include <linux/mm_inline.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080060#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080061#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070062#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070063#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050064#include <linux/file.h>
Tejun Heob23afb92015-11-05 18:46:11 -080065#include <linux/tracehook.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080066#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000067#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070068#include <net/ip.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080069#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080070
Balbir Singh8697d332008-02-07 00:13:59 -080071#include <asm/uaccess.h>
72
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070073#include <trace/events/vmscan.h>
74
Tejun Heo073219e2014-02-08 10:36:58 -050075struct cgroup_subsys memory_cgrp_subsys __read_mostly;
76EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080077
Johannes Weiner7d828602016-01-14 15:20:56 -080078struct mem_cgroup *root_mem_cgroup __read_mostly;
79
KAMEZAWA Hiroyukia181b0e2008-07-25 01:47:08 -070080#define MEM_CGROUP_RECLAIM_RETRIES 5
Balbir Singh8cdea7c2008-02-07 00:13:50 -080081
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080082/* Socket memory accounting disabled? */
83static bool cgroup_memory_nosocket;
84
Vladimir Davydov04823c82016-01-20 15:02:38 -080085/* Kernel memory accounting disabled? */
86static bool cgroup_memory_nokmem;
87
Johannes Weiner21afa382015-02-11 15:26:36 -080088/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070089#ifdef CONFIG_MEMCG_SWAP
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080090int do_swap_account __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080091#else
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070092#define do_swap_account 0
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080093#endif
94
Johannes Weiner7941d212016-01-14 15:21:23 -080095/* Whether legacy memory+swap accounting is active */
96static bool do_memsw_account(void)
97{
98 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account;
99}
100
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700101static const char * const mem_cgroup_stat_names[] = {
102 "cache",
103 "rss",
David Rientjesb070e652013-05-07 16:18:09 -0700104 "rss_huge",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700105 "mapped_file",
Greg Thelenc4843a72015-05-22 17:13:16 -0400106 "dirty",
Sha Zhengju3ea67d02013-09-12 15:13:53 -0700107 "writeback",
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700108 "swap",
109};
110
Johannes Weineraf7c4b02012-05-29 15:07:08 -0700111static const char * const mem_cgroup_events_names[] = {
112 "pgpgin",
113 "pgpgout",
114 "pgfault",
115 "pgmajfault",
116};
117
Sha Zhengju58cf1882013-02-22 16:32:05 -0800118static const char * const mem_cgroup_lru_names[] = {
119 "inactive_anon",
120 "active_anon",
121 "inactive_file",
122 "active_file",
123 "unevictable",
124};
125
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700126#define THRESHOLDS_EVENTS_TARGET 128
127#define SOFTLIMIT_EVENTS_TARGET 1024
128#define NUMAINFO_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700129
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700130/*
131 * Cgroups above their limits are maintained in a RB-Tree, independent of
132 * their hierarchy representation
133 */
134
Mel Gormanef8f2322016-07-28 15:46:05 -0700135struct mem_cgroup_tree_per_node {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700136 struct rb_root rb_root;
137 spinlock_t lock;
138};
139
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700140struct mem_cgroup_tree {
141 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
142};
143
144static struct mem_cgroup_tree soft_limit_tree __read_mostly;
145
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700146/* for OOM */
147struct mem_cgroup_eventfd_list {
148 struct list_head list;
149 struct eventfd_ctx *eventfd;
150};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800151
Tejun Heo79bd9812013-11-22 18:20:42 -0500152/*
153 * cgroup_event represents events which userspace want to receive.
154 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500155struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500156 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500157 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500158 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500159 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500160 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500161 * eventfd to signal userspace about the event.
162 */
163 struct eventfd_ctx *eventfd;
164 /*
165 * Each of these stored in a list by the cgroup.
166 */
167 struct list_head list;
168 /*
Tejun Heofba94802013-11-22 18:20:43 -0500169 * register_event() callback will be used to add new userspace
170 * waiter for changes related to this event. Use eventfd_signal()
171 * on eventfd to send notification to userspace.
172 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500173 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500174 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500175 /*
176 * unregister_event() callback will be called when userspace closes
177 * the eventfd or on cgroup removing. This callback must be set,
178 * if you want provide notification functionality.
179 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500180 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500181 struct eventfd_ctx *eventfd);
182 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500183 * All fields below needed to unregister event when
184 * userspace closes eventfd.
185 */
186 poll_table pt;
187 wait_queue_head_t *wqh;
188 wait_queue_t wait;
189 struct work_struct remove;
190};
191
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700192static void mem_cgroup_threshold(struct mem_cgroup *memcg);
193static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800194
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800195/* Stuffs for move charges at task migration. */
196/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800197 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800198 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800199#define MOVE_ANON 0x1U
200#define MOVE_FILE 0x2U
201#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800202
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800203/* "mc" and its members are protected by cgroup_mutex */
204static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800205 spinlock_t lock; /* for from, to */
Tejun Heo264a0ae2016-04-21 19:09:02 -0400206 struct mm_struct *mm;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800207 struct mem_cgroup *from;
208 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800209 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800210 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800211 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800212 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800213 struct task_struct *moving_task; /* a task moving charges */
214 wait_queue_head_t waitq; /* a waitq for other context */
215} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700216 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800217 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
218};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800219
Balbir Singh4e416952009-09-23 15:56:39 -0700220/*
221 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
222 * limit reclaim to prevent infinite loops, if they ever occur.
223 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700224#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700225#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700226
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800227enum charge_type {
228 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700229 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800230 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700231 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700232 NR_CHARGE_TYPE,
233};
234
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800235/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800236enum res_type {
237 _MEM,
238 _MEMSWAP,
239 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800240 _KMEM,
Vladimir Davydovd55f90b2016-01-20 15:02:44 -0800241 _TCP,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800242};
243
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700244#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
245#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800246#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700247/* Used for OOM nofiier */
248#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800249
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700250/* Some nice accessors for the vmpressure. */
251struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
252{
253 if (!memcg)
254 memcg = root_mem_cgroup;
255 return &memcg->vmpressure;
256}
257
258struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
259{
260 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
261}
262
Michal Hocko7ffc0ed2012-10-08 16:33:13 -0700263static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
264{
265 return (memcg == root_mem_cgroup);
266}
267
Johannes Weiner127424c2016-01-20 15:02:32 -0800268#ifndef CONFIG_SLOB
Glauber Costa55007d82012-12-18 14:22:38 -0800269/*
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800270 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
Li Zefanb8627832013-09-23 16:56:47 +0800271 * The main reason for not using cgroup id for this:
272 * this works better in sparse environments, where we have a lot of memcgs,
273 * but only a few kmem-limited. Or also, if we have, for instance, 200
274 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
275 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800276 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800277 * The current size of the caches array is stored in memcg_nr_cache_ids. It
278 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800279 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800280static DEFINE_IDA(memcg_cache_ida);
281int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800282
Vladimir Davydov05257a12015-02-12 14:59:01 -0800283/* Protects memcg_nr_cache_ids */
284static DECLARE_RWSEM(memcg_cache_ids_sem);
285
286void memcg_get_cache_ids(void)
287{
288 down_read(&memcg_cache_ids_sem);
289}
290
291void memcg_put_cache_ids(void)
292{
293 up_read(&memcg_cache_ids_sem);
294}
295
Glauber Costa55007d82012-12-18 14:22:38 -0800296/*
297 * MIN_SIZE is different than 1, because we would like to avoid going through
298 * the alloc/free process all the time. In a small machine, 4 kmem-limited
299 * cgroups is a reasonable guess. In the future, it could be a parameter or
300 * tunable, but that is strictly not necessary.
301 *
Li Zefanb8627832013-09-23 16:56:47 +0800302 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800303 * this constant directly from cgroup, but it is understandable that this is
304 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800305 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800306 * increase ours as well if it increases.
307 */
308#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800309#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800310
Glauber Costad7f25f82012-12-18 14:22:40 -0800311/*
312 * A lot of the calls to the cache allocation functions are expected to be
313 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
314 * conditional to this static branch, we'll have to allow modules that does
315 * kmem_cache_alloc and the such to see this symbol as well
316 */
Johannes Weineref129472016-01-14 15:21:34 -0800317DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
Glauber Costad7f25f82012-12-18 14:22:40 -0800318EXPORT_SYMBOL(memcg_kmem_enabled_key);
Glauber Costaa8964b92012-12-18 14:22:09 -0800319
Johannes Weiner127424c2016-01-20 15:02:32 -0800320#endif /* !CONFIG_SLOB */
Glauber Costaa8964b92012-12-18 14:22:09 -0800321
Tejun Heoad7fa852015-05-27 20:00:02 -0400322/**
323 * mem_cgroup_css_from_page - css of the memcg associated with a page
324 * @page: page of interest
325 *
326 * If memcg is bound to the default hierarchy, css of the memcg associated
327 * with @page is returned. The returned css remains associated with @page
328 * until it is released.
329 *
330 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
331 * is returned.
Tejun Heoad7fa852015-05-27 20:00:02 -0400332 */
333struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
334{
335 struct mem_cgroup *memcg;
336
Tejun Heoad7fa852015-05-27 20:00:02 -0400337 memcg = page->mem_cgroup;
338
Tejun Heo9e10a132015-09-18 11:56:28 -0400339 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400340 memcg = root_mem_cgroup;
341
Tejun Heoad7fa852015-05-27 20:00:02 -0400342 return &memcg->css;
343}
344
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700345/**
346 * page_cgroup_ino - return inode number of the memcg a page is charged to
347 * @page: the page
348 *
349 * Look up the closest online ancestor of the memory cgroup @page is charged to
350 * and return its inode number or 0 if @page is not charged to any cgroup. It
351 * is safe to call this function without holding a reference to @page.
352 *
353 * Note, this function is inherently racy, because there is nothing to prevent
354 * the cgroup inode from getting torn down and potentially reallocated a moment
355 * after page_cgroup_ino() returns, so it only should be used by callers that
356 * do not care (such as procfs interfaces).
357 */
358ino_t page_cgroup_ino(struct page *page)
359{
360 struct mem_cgroup *memcg;
361 unsigned long ino = 0;
362
363 rcu_read_lock();
364 memcg = READ_ONCE(page->mem_cgroup);
365 while (memcg && !(memcg->css.flags & CSS_ONLINE))
366 memcg = parent_mem_cgroup(memcg);
367 if (memcg)
368 ino = cgroup_ino(memcg->css.cgroup);
369 rcu_read_unlock();
370 return ino;
371}
372
Mel Gormanef8f2322016-07-28 15:46:05 -0700373static struct mem_cgroup_per_node *
374mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700375{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700376 int nid = page_to_nid(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700377
Mel Gormanef8f2322016-07-28 15:46:05 -0700378 return memcg->nodeinfo[nid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700379}
380
Mel Gormanef8f2322016-07-28 15:46:05 -0700381static struct mem_cgroup_tree_per_node *
382soft_limit_tree_node(int nid)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700383{
Mel Gormanef8f2322016-07-28 15:46:05 -0700384 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700385}
386
Mel Gormanef8f2322016-07-28 15:46:05 -0700387static struct mem_cgroup_tree_per_node *
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700388soft_limit_tree_from_page(struct page *page)
389{
390 int nid = page_to_nid(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700391
Mel Gormanef8f2322016-07-28 15:46:05 -0700392 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700393}
394
Mel Gormanef8f2322016-07-28 15:46:05 -0700395static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz,
396 struct mem_cgroup_tree_per_node *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800397 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700398{
399 struct rb_node **p = &mctz->rb_root.rb_node;
400 struct rb_node *parent = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700401 struct mem_cgroup_per_node *mz_node;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700402
403 if (mz->on_tree)
404 return;
405
406 mz->usage_in_excess = new_usage_in_excess;
407 if (!mz->usage_in_excess)
408 return;
409 while (*p) {
410 parent = *p;
Mel Gormanef8f2322016-07-28 15:46:05 -0700411 mz_node = rb_entry(parent, struct mem_cgroup_per_node,
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700412 tree_node);
413 if (mz->usage_in_excess < mz_node->usage_in_excess)
414 p = &(*p)->rb_left;
415 /*
416 * We can't avoid mem cgroups that are over their soft
417 * limit by the same amount
418 */
419 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
420 p = &(*p)->rb_right;
421 }
422 rb_link_node(&mz->tree_node, parent, p);
423 rb_insert_color(&mz->tree_node, &mctz->rb_root);
424 mz->on_tree = true;
425}
426
Mel Gormanef8f2322016-07-28 15:46:05 -0700427static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
428 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700429{
430 if (!mz->on_tree)
431 return;
432 rb_erase(&mz->tree_node, &mctz->rb_root);
433 mz->on_tree = false;
434}
435
Mel Gormanef8f2322016-07-28 15:46:05 -0700436static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
437 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700438{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700439 unsigned long flags;
440
441 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700442 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700443 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700444}
445
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800446static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
447{
448 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700449 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800450 unsigned long excess = 0;
451
452 if (nr_pages > soft_limit)
453 excess = nr_pages - soft_limit;
454
455 return excess;
456}
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700457
458static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
459{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800460 unsigned long excess;
Mel Gormanef8f2322016-07-28 15:46:05 -0700461 struct mem_cgroup_per_node *mz;
462 struct mem_cgroup_tree_per_node *mctz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700463
Jianyu Zhane2318752014-06-06 14:38:20 -0700464 mctz = soft_limit_tree_from_page(page);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700465 /*
466 * Necessary to update all ancestors when hierarchy is used.
467 * because their event counter is not touched.
468 */
469 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700470 mz = mem_cgroup_page_nodeinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800471 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700472 /*
473 * We have to update the tree if mz is on RB-tree or
474 * mem is over its softlimit.
475 */
476 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700477 unsigned long flags;
478
479 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700480 /* if on-tree, remove it */
481 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700482 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700483 /*
484 * Insert again. mz->usage_in_excess will be updated.
485 * If excess is 0, no tree ops.
486 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700487 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700488 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700489 }
490 }
491}
492
493static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
494{
Mel Gormanef8f2322016-07-28 15:46:05 -0700495 struct mem_cgroup_tree_per_node *mctz;
496 struct mem_cgroup_per_node *mz;
497 int nid;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700498
Jianyu Zhane2318752014-06-06 14:38:20 -0700499 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700500 mz = mem_cgroup_nodeinfo(memcg, nid);
501 mctz = soft_limit_tree_node(nid);
502 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700503 }
504}
505
Mel Gormanef8f2322016-07-28 15:46:05 -0700506static struct mem_cgroup_per_node *
507__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700508{
509 struct rb_node *rightmost = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700510 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700511
512retry:
513 mz = NULL;
514 rightmost = rb_last(&mctz->rb_root);
515 if (!rightmost)
516 goto done; /* Nothing to reclaim from */
517
Mel Gormanef8f2322016-07-28 15:46:05 -0700518 mz = rb_entry(rightmost, struct mem_cgroup_per_node, tree_node);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700519 /*
520 * Remove the node now but someone else can add it back,
521 * we will to add it back at the end of reclaim to its correct
522 * position in the tree.
523 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700524 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800525 if (!soft_limit_excess(mz->memcg) ||
Tejun Heoec903c02014-05-13 12:11:01 -0400526 !css_tryget_online(&mz->memcg->css))
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700527 goto retry;
528done:
529 return mz;
530}
531
Mel Gormanef8f2322016-07-28 15:46:05 -0700532static struct mem_cgroup_per_node *
533mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700534{
Mel Gormanef8f2322016-07-28 15:46:05 -0700535 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700536
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700537 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700538 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700539 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700540 return mz;
541}
542
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700543/*
Greg Thelen484ebb32015-10-01 15:37:05 -0700544 * Return page count for single (non recursive) @memcg.
545 *
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700546 * Implementation Note: reading percpu statistics for memcg.
547 *
548 * Both of vmstat[] and percpu_counter has threshold and do periodic
549 * synchronization to implement "quick" read. There are trade-off between
550 * reading cost and precision of value. Then, we may have a chance to implement
Greg Thelen484ebb32015-10-01 15:37:05 -0700551 * a periodic synchronization of counter in memcg's counter.
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700552 *
553 * But this _read() function is used for user interface now. The user accounts
554 * memory usage by memory cgroup and he _always_ requires exact value because
555 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
556 * have to visit all online cpus and make sum. So, for now, unnecessary
557 * synchronization is not implemented. (just implemented for cpu hotplug)
558 *
559 * If there are kernel internal actions which can make use of some not-exact
560 * value, and reading all cpu value can be performance bottleneck in some
Greg Thelen484ebb32015-10-01 15:37:05 -0700561 * common workload, threshold and synchronization as vmstat[] should be
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700562 * implemented.
563 */
Greg Thelen484ebb32015-10-01 15:37:05 -0700564static unsigned long
565mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800566{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700567 long val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800568 int cpu;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800569
Greg Thelen484ebb32015-10-01 15:37:05 -0700570 /* Per-cpu values can be negative, use a signed accumulator */
Tejun Heo733a5722015-05-22 18:23:18 -0400571 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700572 val += per_cpu(memcg->stat->count[idx], cpu);
Greg Thelen484ebb32015-10-01 15:37:05 -0700573 /*
574 * Summing races with updates, so val may be negative. Avoid exposing
575 * transient negative values.
576 */
577 if (val < 0)
578 val = 0;
KAMEZAWA Hiroyukic62b1a32010-03-10 15:22:29 -0800579 return val;
580}
581
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700582static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
Johannes Weinere9f89742011-03-23 16:42:37 -0700583 enum mem_cgroup_events_index idx)
584{
585 unsigned long val = 0;
586 int cpu;
587
Tejun Heo733a5722015-05-22 18:23:18 -0400588 for_each_possible_cpu(cpu)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700589 val += per_cpu(memcg->stat->events[idx], cpu);
Johannes Weinere9f89742011-03-23 16:42:37 -0700590 return val;
591}
592
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700593static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700594 struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800595 bool compound, int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800596{
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700597 /*
598 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
599 * counted as CACHE even if it's on ANON LRU.
600 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700601 if (PageAnon(page))
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700602 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700603 nr_pages);
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800604 else
KAMEZAWA Hiroyukib24028572012-03-21 16:34:22 -0700605 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700606 nr_pages);
Balaji Rao55e462b2008-05-01 04:35:12 -0700607
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800608 if (compound) {
609 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
David Rientjesb070e652013-05-07 16:18:09 -0700610 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
611 nr_pages);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800612 }
David Rientjesb070e652013-05-07 16:18:09 -0700613
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800614 /* pagein of a big page is an event. So, ignore page size */
615 if (nr_pages > 0)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700616 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800617 else {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700618 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800619 nr_pages = -nr_pages; /* for event */
620 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800621
Johannes Weiner13114712012-05-29 15:07:07 -0700622 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800623}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800624
Vladimir Davydov0a6b76d2016-03-17 14:18:42 -0700625unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
626 int nid, unsigned int lru_mask)
Ying Han889976d2011-05-26 16:25:33 -0700627{
Michal Hocko07fc9572017-01-10 16:58:04 -0800628 struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg);
Jianyu Zhane2318752014-06-06 14:38:20 -0700629 unsigned long nr = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -0700630 enum lru_list lru;
Ying Han889976d2011-05-26 16:25:33 -0700631
Jianyu Zhane2318752014-06-06 14:38:20 -0700632 VM_BUG_ON((unsigned)nid >= nr_node_ids);
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700633
Mel Gormanef8f2322016-07-28 15:46:05 -0700634 for_each_lru(lru) {
635 if (!(BIT(lru) & lru_mask))
636 continue;
Michal Hocko07fc9572017-01-10 16:58:04 -0800637 nr += mem_cgroup_get_lru_size(lruvec, lru);
Jianyu Zhane2318752014-06-06 14:38:20 -0700638 }
639 return nr;
Ying Han889976d2011-05-26 16:25:33 -0700640}
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700641
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700642static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
KAMEZAWA Hiroyukibb2a0de2011-07-26 16:08:22 -0700643 unsigned int lru_mask)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800644{
Jianyu Zhane2318752014-06-06 14:38:20 -0700645 unsigned long nr = 0;
Ying Han889976d2011-05-26 16:25:33 -0700646 int nid;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800647
Lai Jiangshan31aaea42012-12-12 13:51:27 -0800648 for_each_node_state(nid, N_MEMORY)
Jianyu Zhane2318752014-06-06 14:38:20 -0700649 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
650 return nr;
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800651}
652
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800653static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
654 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800655{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700656 unsigned long val, next;
657
Johannes Weiner13114712012-05-29 15:07:07 -0700658 val = __this_cpu_read(memcg->stat->nr_page_events);
Steven Rostedt47994012011-11-02 13:38:33 -0700659 next = __this_cpu_read(memcg->stat->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700660 /* from time_after() in jiffies.h */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800661 if ((long)next - (long)val < 0) {
662 switch (target) {
663 case MEM_CGROUP_TARGET_THRESH:
664 next = val + THRESHOLDS_EVENTS_TARGET;
665 break;
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700666 case MEM_CGROUP_TARGET_SOFTLIMIT:
667 next = val + SOFTLIMIT_EVENTS_TARGET;
668 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800669 case MEM_CGROUP_TARGET_NUMAINFO:
670 next = val + NUMAINFO_EVENTS_TARGET;
671 break;
672 default:
673 break;
674 }
675 __this_cpu_write(memcg->stat->targets[target], next);
676 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700677 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800678 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800679}
680
681/*
682 * Check events in order.
683 *
684 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700685static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800686{
687 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800688 if (unlikely(mem_cgroup_event_ratelimit(memcg,
689 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700690 bool do_softlimit;
Andrew Morton82b3f2a2012-02-03 15:37:14 -0800691 bool do_numainfo __maybe_unused;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800692
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700693 do_softlimit = mem_cgroup_event_ratelimit(memcg,
694 MEM_CGROUP_TARGET_SOFTLIMIT);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700695#if MAX_NUMNODES > 1
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800696 do_numainfo = mem_cgroup_event_ratelimit(memcg,
697 MEM_CGROUP_TARGET_NUMAINFO);
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -0700698#endif
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800699 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a2013-09-24 15:27:40 -0700700 if (unlikely(do_softlimit))
701 mem_cgroup_update_tree(memcg, page);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800702#if MAX_NUMNODES > 1
703 if (unlikely(do_numainfo))
704 atomic_inc(&memcg->numainfo_events);
705#endif
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700706 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800707}
708
Balbir Singhcf475ad2008-04-29 01:00:16 -0700709struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800710{
Balbir Singh31a78f22008-09-28 23:09:31 +0100711 /*
712 * mm_update_next_owner() may clear mm->owner to NULL
713 * if it races with swapoff, page migration, etc.
714 * So this can be called with p == NULL.
715 */
716 if (unlikely(!p))
717 return NULL;
718
Tejun Heo073219e2014-02-08 10:36:58 -0500719 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800720}
Michal Hocko33398cf2015-09-08 15:01:02 -0700721EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800722
Johannes Weinerdf381972014-04-07 15:37:43 -0700723static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800724{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700725 struct mem_cgroup *memcg = NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700726
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800727 rcu_read_lock();
728 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -0700729 /*
730 * Page cache insertions can happen withou an
731 * actual mm context, e.g. during disk probing
732 * on boot, loopback IO, acct() writes etc.
733 */
734 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -0700735 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -0700736 else {
737 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
738 if (unlikely(!memcg))
739 memcg = root_mem_cgroup;
740 }
Tejun Heoec903c02014-05-13 12:11:01 -0400741 } while (!css_tryget_online(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800742 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700743 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800744}
745
Johannes Weiner56600482012-01-12 17:17:59 -0800746/**
747 * mem_cgroup_iter - iterate over memory cgroup hierarchy
748 * @root: hierarchy root
749 * @prev: previously returned memcg, NULL on first invocation
750 * @reclaim: cookie for shared reclaim walks, NULL for full walks
751 *
752 * Returns references to children of the hierarchy below @root, or
753 * @root itself, or %NULL after a full round-trip.
754 *
755 * Caller must pass the return value in @prev on subsequent
756 * invocations for reference counting, or use mem_cgroup_iter_break()
757 * to cancel a hierarchy walk before the round-trip is complete.
758 *
759 * Reclaimers can specify a zone and a priority level in @reclaim to
760 * divide up the memcgs in the hierarchy among all concurrent
761 * reclaimers operating on the same zone and priority.
762 */
Andrew Morton694fbc02013-09-24 15:27:37 -0700763struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -0800764 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -0700765 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -0700766{
Michal Hocko33398cf2015-09-08 15:01:02 -0700767 struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800768 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800769 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800770 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700771
Andrew Morton694fbc02013-09-24 15:27:37 -0700772 if (mem_cgroup_disabled())
773 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -0800774
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -0700775 if (!root)
776 root = root_mem_cgroup;
777
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800778 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800779 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800780
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800781 if (!root->use_hierarchy && root != root_mem_cgroup) {
782 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800783 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -0700784 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800785 }
786
Michal Hocko542f85f2013-04-29 15:07:15 -0700787 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800788
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800789 if (reclaim) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700790 struct mem_cgroup_per_node *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800791
Mel Gormanef8f2322016-07-28 15:46:05 -0700792 mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800793 iter = &mz->iter[reclaim->priority];
Michal Hocko5f578162013-04-29 15:07:17 -0700794
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800795 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -0700796 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800797
Vladimir Davydov6df38682015-12-29 14:54:10 -0800798 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -0700799 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800800 if (!pos || css_tryget(&pos->css))
801 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800802 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800803 * css reference reached zero, so iter->position will
804 * be cleared by ->css_released. However, we should not
805 * rely on this happening soon, because ->css_released
806 * is called from a work queue, and by busy-waiting we
807 * might block it. So we clear iter->position right
808 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800809 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800810 (void)cmpxchg(&iter->position, pos, NULL);
811 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800812 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800813
814 if (pos)
815 css = &pos->css;
816
817 for (;;) {
818 css = css_next_descendant_pre(css, &root->css);
819 if (!css) {
820 /*
821 * Reclaimers share the hierarchy walk, and a
822 * new one might jump in right at the end of
823 * the hierarchy - make sure they see at least
824 * one group and restart from the beginning.
825 */
826 if (!prev)
827 continue;
828 break;
829 }
830
831 /*
832 * Verify the css and acquire a reference. The root
833 * is provided by the caller, so we know it's alive
834 * and kicking, and don't take an extra reference.
835 */
836 memcg = mem_cgroup_from_css(css);
837
838 if (css == &root->css)
839 break;
840
Johannes Weiner0b8f73e2016-01-20 15:02:53 -0800841 if (css_tryget(css))
842 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800843
844 memcg = NULL;
845 }
846
847 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800848 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -0800849 * The position could have already been updated by a competing
850 * thread, so check that the value hasn't changed since we read
851 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800852 */
Vladimir Davydov6df38682015-12-29 14:54:10 -0800853 (void)cmpxchg(&iter->position, pos, memcg);
854
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800855 if (pos)
856 css_put(&pos->css);
857
858 if (!memcg)
859 iter->generation++;
860 else if (!prev)
861 reclaim->generation = iter->generation;
862 }
863
Michal Hocko542f85f2013-04-29 15:07:15 -0700864out_unlock:
865 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -0800866out:
Michal Hockoc40046f2013-04-29 15:07:14 -0700867 if (prev && prev != root)
868 css_put(&prev->css);
869
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800870 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700871}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800872
Johannes Weiner56600482012-01-12 17:17:59 -0800873/**
874 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
875 * @root: hierarchy root
876 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
877 */
878void mem_cgroup_iter_break(struct mem_cgroup *root,
879 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800880{
881 if (!root)
882 root = root_mem_cgroup;
883 if (prev && prev != root)
884 css_put(&prev->css);
885}
886
Vladimir Davydov6df38682015-12-29 14:54:10 -0800887static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
888{
889 struct mem_cgroup *memcg = dead_memcg;
890 struct mem_cgroup_reclaim_iter *iter;
Mel Gormanef8f2322016-07-28 15:46:05 -0700891 struct mem_cgroup_per_node *mz;
892 int nid;
Vladimir Davydov6df38682015-12-29 14:54:10 -0800893 int i;
894
895 while ((memcg = parent_mem_cgroup(memcg))) {
896 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700897 mz = mem_cgroup_nodeinfo(memcg, nid);
898 for (i = 0; i <= DEF_PRIORITY; i++) {
899 iter = &mz->iter[i];
900 cmpxchg(&iter->position,
901 dead_memcg, NULL);
Vladimir Davydov6df38682015-12-29 14:54:10 -0800902 }
903 }
904 }
905}
906
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700907/*
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800908 * Iteration constructs for visiting all cgroups (under a tree). If
909 * loops are exited prematurely (break), mem_cgroup_iter_break() must
910 * be used for reference counting.
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700911 */
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800912#define for_each_mem_cgroup_tree(iter, root) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800913 for (iter = mem_cgroup_iter(root, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800914 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800915 iter = mem_cgroup_iter(root, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700916
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800917#define for_each_mem_cgroup(iter) \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800918 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
Johannes Weiner9f3a0d02012-01-12 17:17:48 -0800919 iter != NULL; \
Johannes Weiner527a5ec2012-01-12 17:17:55 -0800920 iter = mem_cgroup_iter(NULL, iter, NULL))
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -0700921
Johannes Weiner925b7672012-01-12 17:18:15 -0800922/**
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700923 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
924 * @memcg: hierarchy root
925 * @fn: function to call for each task
926 * @arg: argument passed to @fn
927 *
928 * This function iterates over tasks attached to @memcg or to any of its
929 * descendants and calls @fn for each task. If @fn returns a non-zero
930 * value, the function breaks the iteration loop and returns the value.
931 * Otherwise, it will iterate over all tasks and return 0.
932 *
933 * This function must not be called for the root memory cgroup.
934 */
935int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
936 int (*fn)(struct task_struct *, void *), void *arg)
937{
938 struct mem_cgroup *iter;
939 int ret = 0;
940
941 BUG_ON(memcg == root_mem_cgroup);
942
943 for_each_mem_cgroup_tree(iter, memcg) {
944 struct css_task_iter it;
945 struct task_struct *task;
946
947 css_task_iter_start(&iter->css, &it);
948 while (!ret && (task = css_task_iter_next(&it)))
949 ret = fn(task, arg);
950 css_task_iter_end(&it);
951 if (ret) {
952 mem_cgroup_iter_break(memcg, iter);
953 break;
954 }
955 }
956 return ret;
957}
958
959/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800960 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -0800961 * @page: the page
Hugh Dickinsfa9add62012-05-29 15:07:09 -0700962 * @zone: zone of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800963 *
964 * This function is only safe when following the LRU page isolation
965 * and putback protocol: the LRU lock must be held, and the page must
966 * either be PageLRU() or the caller must have isolated/allocated it.
Minchan Kim3f58a822011-03-22 16:32:53 -0700967 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700968struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat)
Minchan Kim3f58a822011-03-22 16:32:53 -0700969{
Mel Gormanef8f2322016-07-28 15:46:05 -0700970 struct mem_cgroup_per_node *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -0800971 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800972 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800973
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800974 if (mem_cgroup_disabled()) {
Mel Gorman599d0c92016-07-28 15:45:31 -0700975 lruvec = &pgdat->lruvec;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800976 goto out;
977 }
Christoph Lameterb69408e2008-10-18 20:26:14 -0700978
Johannes Weiner1306a852014-12-10 15:44:52 -0800979 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800980 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -0800981 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -0800982 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -0800983 */
Johannes Weiner29833312014-12-10 15:44:02 -0800984 if (!memcg)
985 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -0800986
Mel Gormanef8f2322016-07-28 15:46:05 -0700987 mz = mem_cgroup_page_nodeinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800988 lruvec = &mz->lruvec;
989out:
990 /*
991 * Since a node can be onlined after the mem_cgroup was created,
992 * we have to be prepared to initialize lruvec->zone here;
993 * and if offlined then reonlined, we need to reinitialize it.
994 */
Mel Gorman599d0c92016-07-28 15:45:31 -0700995 if (unlikely(lruvec->pgdat != pgdat))
996 lruvec->pgdat = pgdat;
Hugh Dickinsbea8c152012-11-16 14:14:54 -0800997 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -0800998}
999
1000/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001001 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1002 * @lruvec: mem_cgroup per zone lru vector
1003 * @lru: index of lru list the page is sitting on
Michal Hocko07fc9572017-01-10 16:58:04 -08001004 * @zid: zone id of the accounted pages
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001005 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001006 *
Hugh Dickinsca707232016-05-19 17:12:35 -07001007 * This function must be called under lru_lock, just before a page is added
1008 * to or just after a page is removed from an lru list (that ordering being
1009 * so as to allow it to check that lru_size 0 is consistent with list_empty).
Johannes Weiner925b7672012-01-12 17:18:15 -08001010 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001011void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
Michal Hocko07fc9572017-01-10 16:58:04 -08001012 int zid, int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001013{
Mel Gormanef8f2322016-07-28 15:46:05 -07001014 struct mem_cgroup_per_node *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001015 unsigned long *lru_size;
Hugh Dickinsca707232016-05-19 17:12:35 -07001016 long size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001017
1018 if (mem_cgroup_disabled())
1019 return;
1020
Mel Gormanef8f2322016-07-28 15:46:05 -07001021 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
Michal Hocko07fc9572017-01-10 16:58:04 -08001022 lru_size = &mz->lru_zone_size[zid][lru];
Hugh Dickinsca707232016-05-19 17:12:35 -07001023
1024 if (nr_pages < 0)
1025 *lru_size += nr_pages;
1026
1027 size = *lru_size;
Michal Hocko07fc9572017-01-10 16:58:04 -08001028 if (WARN_ONCE(size < 0,
1029 "%s(%p, %d, %d): lru_size %ld\n",
1030 __func__, lruvec, lru, nr_pages, size)) {
Hugh Dickinsca707232016-05-19 17:12:35 -07001031 VM_BUG_ON(1);
1032 *lru_size = 0;
1033 }
1034
1035 if (nr_pages > 0)
1036 *lru_size += nr_pages;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001037}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001038
Johannes Weiner2314b422014-12-10 15:44:33 -08001039bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
Johannes Weinerc3ac9a82012-05-29 15:06:25 -07001040{
Johannes Weiner2314b422014-12-10 15:44:33 -08001041 struct mem_cgroup *task_memcg;
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001042 struct task_struct *p;
David Rientjesffbdccf2013-07-03 15:01:23 -07001043 bool ret;
David Rientjes4c4a2212008-02-07 00:14:06 -08001044
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -07001045 p = find_lock_task_mm(task);
David Rientjesde077d22012-01-12 17:18:52 -08001046 if (p) {
Johannes Weiner2314b422014-12-10 15:44:33 -08001047 task_memcg = get_mem_cgroup_from_mm(p->mm);
David Rientjesde077d22012-01-12 17:18:52 -08001048 task_unlock(p);
1049 } else {
1050 /*
1051 * All threads may have already detached their mm's, but the oom
1052 * killer still needs to detect if they have already been oom
1053 * killed to prevent needlessly killing additional tasks.
1054 */
David Rientjesffbdccf2013-07-03 15:01:23 -07001055 rcu_read_lock();
Johannes Weiner2314b422014-12-10 15:44:33 -08001056 task_memcg = mem_cgroup_from_task(task);
1057 css_get(&task_memcg->css);
David Rientjesffbdccf2013-07-03 15:01:23 -07001058 rcu_read_unlock();
David Rientjesde077d22012-01-12 17:18:52 -08001059 }
Johannes Weiner2314b422014-12-10 15:44:33 -08001060 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1061 css_put(&task_memcg->css);
David Rientjes4c4a2212008-02-07 00:14:06 -08001062 return ret;
1063}
1064
Johannes Weiner19942822011-02-01 15:52:43 -08001065/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001066 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001067 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001068 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001069 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001070 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001071 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001072static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001073{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001074 unsigned long margin = 0;
1075 unsigned long count;
1076 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001077
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001078 count = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -07001079 limit = READ_ONCE(memcg->memory.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001080 if (count < limit)
1081 margin = limit - count;
1082
Johannes Weiner7941d212016-01-14 15:21:23 -08001083 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001084 count = page_counter_read(&memcg->memsw);
Jason Low4db0c3c2015-04-15 16:14:08 -07001085 limit = READ_ONCE(memcg->memsw.limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001086 if (count <= limit)
1087 margin = min(margin, limit - count);
Li RongQingcbedbac2016-05-27 14:27:43 -07001088 else
1089 margin = 0;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001090 }
1091
1092 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001093}
1094
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001095/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001096 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001097 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001098 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1099 * moving cgroups. This is for waiting at high-memory pressure
1100 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001101 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001102static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001103{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001104 struct mem_cgroup *from;
1105 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001106 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001107 /*
1108 * Unlike task_move routines, we access mc.to, mc.from not under
1109 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1110 */
1111 spin_lock(&mc.lock);
1112 from = mc.from;
1113 to = mc.to;
1114 if (!from)
1115 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001116
Johannes Weiner2314b422014-12-10 15:44:33 -08001117 ret = mem_cgroup_is_descendant(from, memcg) ||
1118 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001119unlock:
1120 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001121 return ret;
1122}
1123
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001124static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001125{
1126 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001127 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001128 DEFINE_WAIT(wait);
1129 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1130 /* moving charge context might have finished. */
1131 if (mc.moving_task)
1132 schedule();
1133 finish_wait(&mc.waitq, &wait);
1134 return true;
1135 }
1136 }
1137 return false;
1138}
1139
Sha Zhengju58cf1882013-02-22 16:32:05 -08001140#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001141/**
Sha Zhengju58cf1882013-02-22 16:32:05 -08001142 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001143 * @memcg: The memory cgroup that went over limit
1144 * @p: Task that is going to be killed
1145 *
1146 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1147 * enabled
1148 */
1149void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1150{
Sha Zhengju58cf1882013-02-22 16:32:05 -08001151 struct mem_cgroup *iter;
1152 unsigned int i;
Balbir Singhe2224322009-04-02 16:57:39 -07001153
Balbir Singhe2224322009-04-02 16:57:39 -07001154 rcu_read_lock();
1155
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -07001156 if (p) {
1157 pr_info("Task in ");
1158 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1159 pr_cont(" killed as a result of limit of ");
1160 } else {
1161 pr_info("Memory limit reached of cgroup ");
1162 }
1163
Tejun Heoe61734c2014-02-12 09:29:50 -05001164 pr_cont_cgroup_path(memcg->css.cgroup);
Greg Thelen0346dad2015-01-26 12:58:38 -08001165 pr_cont("\n");
Balbir Singhe2224322009-04-02 16:57:39 -07001166
Balbir Singhe2224322009-04-02 16:57:39 -07001167 rcu_read_unlock();
1168
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001169 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1170 K((u64)page_counter_read(&memcg->memory)),
1171 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1172 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1173 K((u64)page_counter_read(&memcg->memsw)),
1174 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1175 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1176 K((u64)page_counter_read(&memcg->kmem)),
1177 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001178
1179 for_each_mem_cgroup_tree(iter, memcg) {
Tejun Heoe61734c2014-02-12 09:29:50 -05001180 pr_info("Memory cgroup stats for ");
1181 pr_cont_cgroup_path(iter->css.cgroup);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001182 pr_cont(":");
1183
1184 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08001185 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
Sha Zhengju58cf1882013-02-22 16:32:05 -08001186 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07001187 pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
Sha Zhengju58cf1882013-02-22 16:32:05 -08001188 K(mem_cgroup_read_stat(iter, i)));
1189 }
1190
1191 for (i = 0; i < NR_LRU_LISTS; i++)
1192 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1193 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1194
1195 pr_cont("\n");
1196 }
Balbir Singhe2224322009-04-02 16:57:39 -07001197}
1198
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001199/*
1200 * This function returns the number of memcg under hierarchy tree. Returns
1201 * 1(self count) if no children.
1202 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001203static int mem_cgroup_count_children(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001204{
1205 int num = 0;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001206 struct mem_cgroup *iter;
1207
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001208 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001209 num++;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001210 return num;
1211}
1212
Balbir Singh6d61ef42009-01-07 18:08:06 -08001213/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001214 * Return the memory (and swap, if configured) limit for a memcg.
1215 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001216unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001217{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001218 unsigned long limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001219
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001220 limit = memcg->memory.limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001221 if (mem_cgroup_swappiness(memcg)) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001222 unsigned long memsw_limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001223 unsigned long swap_limit;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001224
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001225 memsw_limit = memcg->memsw.limit;
Vladimir Davydov37e84352016-01-20 15:02:56 -08001226 swap_limit = memcg->swap.limit;
1227 swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1228 limit = min(limit + swap_limit, memsw_limit);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001229 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001230 return limit;
David Rientjesa63d83f2010-08-09 17:19:46 -07001231}
1232
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001233static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001234 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001235{
David Rientjes6e0fc462015-09-08 15:00:36 -07001236 struct oom_control oc = {
1237 .zonelist = NULL,
1238 .nodemask = NULL,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001239 .memcg = memcg,
David Rientjes6e0fc462015-09-08 15:00:36 -07001240 .gfp_mask = gfp_mask,
1241 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001242 };
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001243 bool ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001244
Johannes Weinerdc564012015-06-24 16:57:19 -07001245 mutex_lock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001246 ret = out_of_memory(&oc);
Johannes Weinerdc564012015-06-24 16:57:19 -07001247 mutex_unlock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001248 return ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001249}
1250
Michele Curtiae6e71d2014-12-12 16:56:35 -08001251#if MAX_NUMNODES > 1
1252
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001253/**
1254 * test_mem_cgroup_node_reclaimable
Wanpeng Lidad75572012-06-20 12:53:01 -07001255 * @memcg: the target memcg
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001256 * @nid: the node ID to be checked.
1257 * @noswap : specify true here if the user wants flle only information.
1258 *
1259 * This function returns whether the specified memcg contains any
1260 * reclaimable pages on a node. Returns true if there are any reclaimable
1261 * pages in the node.
1262 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001263static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001264 int nid, bool noswap)
1265{
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001266 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001267 return true;
1268 if (noswap || !total_swap_pages)
1269 return false;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001270 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
KAMEZAWA Hiroyuki4d0c0662011-07-08 15:39:42 -07001271 return true;
1272 return false;
1273
1274}
Ying Han889976d2011-05-26 16:25:33 -07001275
1276/*
1277 * Always updating the nodemask is not very good - even if we have an empty
1278 * list or the wrong list here, we can start from some node and traverse all
1279 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1280 *
1281 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001282static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001283{
1284 int nid;
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001285 /*
1286 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1287 * pagein/pageout changes since the last update.
1288 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001289 if (!atomic_read(&memcg->numainfo_events))
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001290 return;
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001291 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
Ying Han889976d2011-05-26 16:25:33 -07001292 return;
1293
Ying Han889976d2011-05-26 16:25:33 -07001294 /* make a nodemask where this memcg uses memory from */
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001295 memcg->scan_nodes = node_states[N_MEMORY];
Ying Han889976d2011-05-26 16:25:33 -07001296
Lai Jiangshan31aaea42012-12-12 13:51:27 -08001297 for_each_node_mask(nid, node_states[N_MEMORY]) {
Ying Han889976d2011-05-26 16:25:33 -07001298
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001299 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1300 node_clear(nid, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001301 }
KAMEZAWA Hiroyuki453a9bf2011-07-08 15:39:43 -07001302
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001303 atomic_set(&memcg->numainfo_events, 0);
1304 atomic_set(&memcg->numainfo_updating, 0);
Ying Han889976d2011-05-26 16:25:33 -07001305}
1306
1307/*
1308 * Selecting a node where we start reclaim from. Because what we need is just
1309 * reducing usage counter, start from anywhere is O,K. Considering
1310 * memory reclaim from current node, there are pros. and cons.
1311 *
1312 * Freeing memory from current node means freeing memory from a node which
1313 * we'll use or we've used. So, it may make LRU bad. And if several threads
1314 * hit limits, it will see a contention on a node. But freeing from remote
1315 * node means more costs for memory reclaim because of memory latency.
1316 *
1317 * Now, we use round-robin. Better algorithm is welcomed.
1318 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001319int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001320{
1321 int node;
1322
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001323 mem_cgroup_may_update_nodemask(memcg);
1324 node = memcg->last_scanned_node;
Ying Han889976d2011-05-26 16:25:33 -07001325
Andrew Morton0edaf862016-05-19 17:10:58 -07001326 node = next_node_in(node, memcg->scan_nodes);
Ying Han889976d2011-05-26 16:25:33 -07001327 /*
Michal Hockofda3d692016-05-19 17:11:34 -07001328 * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1329 * last time it really checked all the LRUs due to rate limiting.
1330 * Fallback to the current node in that case for simplicity.
Ying Han889976d2011-05-26 16:25:33 -07001331 */
1332 if (unlikely(node == MAX_NUMNODES))
1333 node = numa_node_id();
1334
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001335 memcg->last_scanned_node = node;
Ying Han889976d2011-05-26 16:25:33 -07001336 return node;
1337}
Ying Han889976d2011-05-26 16:25:33 -07001338#else
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001339int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
Ying Han889976d2011-05-26 16:25:33 -07001340{
1341 return 0;
1342}
1343#endif
1344
Andrew Morton0608f432013-09-24 15:27:41 -07001345static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
Mel Gormanef8f2322016-07-28 15:46:05 -07001346 pg_data_t *pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001347 gfp_t gfp_mask,
1348 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001349{
Andrew Morton0608f432013-09-24 15:27:41 -07001350 struct mem_cgroup *victim = NULL;
1351 int total = 0;
1352 int loop = 0;
1353 unsigned long excess;
1354 unsigned long nr_scanned;
1355 struct mem_cgroup_reclaim_cookie reclaim = {
Mel Gormanef8f2322016-07-28 15:46:05 -07001356 .pgdat = pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001357 .priority = 0,
1358 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001359
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001360 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001361
Andrew Morton0608f432013-09-24 15:27:41 -07001362 while (1) {
1363 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1364 if (!victim) {
1365 loop++;
1366 if (loop >= 2) {
1367 /*
1368 * If we have not been able to reclaim
1369 * anything, it might because there are
1370 * no reclaimable pages under this hierarchy
1371 */
1372 if (!total)
1373 break;
1374 /*
1375 * We want to do more targeted reclaim.
1376 * excess >> 2 is not to excessive so as to
1377 * reclaim too much, nor too less that we keep
1378 * coming back to reclaim from this cgroup
1379 */
1380 if (total >= (excess >> 2) ||
1381 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1382 break;
1383 }
1384 continue;
1385 }
Mel Gormana9dd0a82016-07-28 15:46:02 -07001386 total += mem_cgroup_shrink_node(victim, gfp_mask, false,
Mel Gormanef8f2322016-07-28 15:46:05 -07001387 pgdat, &nr_scanned);
Andrew Morton0608f432013-09-24 15:27:41 -07001388 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001389 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001390 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001391 }
Andrew Morton0608f432013-09-24 15:27:41 -07001392 mem_cgroup_iter_break(root_memcg, victim);
1393 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001394}
1395
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001396#ifdef CONFIG_LOCKDEP
1397static struct lockdep_map memcg_oom_lock_dep_map = {
1398 .name = "memcg_oom_lock",
1399};
1400#endif
1401
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001402static DEFINE_SPINLOCK(memcg_oom_lock);
1403
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001404/*
1405 * Check OOM-Killer is already running under our hierarchy.
1406 * If someone is running, return false.
1407 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001408static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001409{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001410 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001411
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001412 spin_lock(&memcg_oom_lock);
1413
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001414 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001415 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001416 /*
1417 * this subtree of our hierarchy is already locked
1418 * so we cannot give a lock.
1419 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001420 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001421 mem_cgroup_iter_break(memcg, iter);
1422 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001423 } else
1424 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001425 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001426
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001427 if (failed) {
1428 /*
1429 * OK, we failed to lock the whole subtree so we have
1430 * to clean up what we set up to the failing subtree
1431 */
1432 for_each_mem_cgroup_tree(iter, memcg) {
1433 if (iter == failed) {
1434 mem_cgroup_iter_break(memcg, iter);
1435 break;
1436 }
1437 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001438 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001439 } else
1440 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001441
1442 spin_unlock(&memcg_oom_lock);
1443
1444 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001445}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001446
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001447static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001448{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001449 struct mem_cgroup *iter;
1450
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001451 spin_lock(&memcg_oom_lock);
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001452 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001453 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001454 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001455 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001456}
1457
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001458static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001459{
1460 struct mem_cgroup *iter;
1461
Tejun Heoc2b42d32015-06-24 16:58:23 -07001462 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001463 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001464 iter->under_oom++;
1465 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001466}
1467
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001468static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001469{
1470 struct mem_cgroup *iter;
1471
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001472 /*
1473 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001474 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001475 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001476 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001477 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001478 if (iter->under_oom > 0)
1479 iter->under_oom--;
1480 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001481}
1482
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001483static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1484
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001485struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001486 struct mem_cgroup *memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001487 wait_queue_t wait;
1488};
1489
1490static int memcg_oom_wake_function(wait_queue_t *wait,
1491 unsigned mode, int sync, void *arg)
1492{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001493 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1494 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001495 struct oom_wait_info *oom_wait_info;
1496
1497 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001498 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001499
Johannes Weiner2314b422014-12-10 15:44:33 -08001500 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1501 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001502 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001503 return autoremove_wake_function(wait, mode, sync, arg);
1504}
1505
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001506static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001507{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001508 /*
1509 * For the following lockless ->under_oom test, the only required
1510 * guarantee is that it must see the state asserted by an OOM when
1511 * this function is called as a result of userland actions
1512 * triggered by the notification of the OOM. This is trivially
1513 * achieved by invoking mem_cgroup_mark_under_oom() before
1514 * triggering notification.
1515 */
1516 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001517 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001518}
1519
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001520static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001521{
Andrew Mortond0db7af2016-06-08 15:33:47 -07001522 if (!current->memcg_may_oom)
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001523 return;
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001524 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001525 * We are in the middle of the charge context here, so we
1526 * don't want to block when potentially sitting on a callstack
1527 * that holds all kinds of filesystem and mm locks.
1528 *
1529 * Also, the caller may handle a failed allocation gracefully
1530 * (like optional page cache readahead) and so an OOM killer
1531 * invocation might not even be necessary.
1532 *
1533 * That's why we don't do anything here except remember the
1534 * OOM context and then deal with it at the end of the page
1535 * fault when the stack is unwound, the locks are released,
1536 * and when we know whether the fault was overall successful.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001537 */
Johannes Weiner49426422013-10-16 13:46:59 -07001538 css_get(&memcg->css);
Tejun Heo626ebc42015-11-05 18:46:09 -08001539 current->memcg_in_oom = memcg;
1540 current->memcg_oom_gfp_mask = mask;
1541 current->memcg_oom_order = order;
Johannes Weiner49426422013-10-16 13:46:59 -07001542}
1543
1544/**
1545 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1546 * @handle: actually kill/wait or just clean up the OOM state
1547 *
1548 * This has to be called at the end of a page fault if the memcg OOM
1549 * handler was enabled.
1550 *
1551 * Memcg supports userspace OOM handling where failed allocations must
1552 * sleep on a waitqueue until the userspace task resolves the
1553 * situation. Sleeping directly in the charge context with all kinds
1554 * of locks held is not a good idea, instead we remember an OOM state
1555 * in the task and mem_cgroup_oom_synchronize() has to be called at
1556 * the end of the page fault to complete the OOM handling.
1557 *
1558 * Returns %true if an ongoing memcg OOM situation was detected and
1559 * completed, %false otherwise.
1560 */
1561bool mem_cgroup_oom_synchronize(bool handle)
1562{
Tejun Heo626ebc42015-11-05 18:46:09 -08001563 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001564 struct oom_wait_info owait;
1565 bool locked;
1566
1567 /* OOM is global, do not handle */
1568 if (!memcg)
1569 return false;
1570
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001571 if (!handle)
Johannes Weiner49426422013-10-16 13:46:59 -07001572 goto cleanup;
1573
1574 owait.memcg = memcg;
1575 owait.wait.flags = 0;
1576 owait.wait.func = memcg_oom_wake_function;
1577 owait.wait.private = current;
1578 INIT_LIST_HEAD(&owait.wait.task_list);
1579
1580 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001581 mem_cgroup_mark_under_oom(memcg);
1582
1583 locked = mem_cgroup_oom_trylock(memcg);
1584
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001585 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001586 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001587
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001588 if (locked && !memcg->oom_kill_disable) {
1589 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001590 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001591 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1592 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001593 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001594 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001595 mem_cgroup_unmark_under_oom(memcg);
1596 finish_wait(&memcg_oom_waitq, &owait.wait);
1597 }
1598
1599 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001600 mem_cgroup_oom_unlock(memcg);
1601 /*
1602 * There is no guarantee that an OOM-lock contender
1603 * sees the wakeups triggered by the OOM kill
1604 * uncharges. Wake any sleepers explicitely.
1605 */
1606 memcg_oom_recover(memcg);
1607 }
Johannes Weiner49426422013-10-16 13:46:59 -07001608cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001609 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001610 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001611 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001612}
1613
Johannes Weinerd7365e72014-10-29 14:50:48 -07001614/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001615 * lock_page_memcg - lock a page->mem_cgroup binding
1616 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001617 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001618 * This function protects unlocked LRU pages from being moved to
1619 * another cgroup and stabilizes their page->mem_cgroup binding.
Balbir Singhd69b0422009-06-17 16:26:34 -07001620 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001621void lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001622{
1623 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08001624 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001625
Johannes Weiner6de22612015-02-11 15:25:01 -08001626 /*
1627 * The RCU lock is held throughout the transaction. The fast
1628 * path can get away without acquiring the memcg->move_lock
1629 * because page moving starts with an RCU grace period.
Johannes Weiner6de22612015-02-11 15:25:01 -08001630 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07001631 rcu_read_lock();
1632
1633 if (mem_cgroup_disabled())
Johannes Weiner62cccb82016-03-15 14:57:22 -07001634 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001635again:
Johannes Weiner1306a852014-12-10 15:44:52 -08001636 memcg = page->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08001637 if (unlikely(!memcg))
Johannes Weiner62cccb82016-03-15 14:57:22 -07001638 return;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001639
Qiang Huangbdcbb652014-06-04 16:08:21 -07001640 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001641 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001642
Johannes Weiner6de22612015-02-11 15:25:01 -08001643 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner1306a852014-12-10 15:44:52 -08001644 if (memcg != page->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08001645 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001646 goto again;
1647 }
Johannes Weiner6de22612015-02-11 15:25:01 -08001648
1649 /*
1650 * When charge migration first begins, we can have locked and
1651 * unlocked page stat updates happening concurrently. Track
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001652 * the task who has the lock for unlock_page_memcg().
Johannes Weiner6de22612015-02-11 15:25:01 -08001653 */
1654 memcg->move_lock_task = current;
1655 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07001656
Johannes Weiner62cccb82016-03-15 14:57:22 -07001657 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001658}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001659EXPORT_SYMBOL(lock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001660
Johannes Weinerd7365e72014-10-29 14:50:48 -07001661/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001662 * unlock_page_memcg - unlock a page->mem_cgroup binding
Johannes Weiner62cccb82016-03-15 14:57:22 -07001663 * @page: the page
Johannes Weinerd7365e72014-10-29 14:50:48 -07001664 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001665void unlock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001666{
Johannes Weiner62cccb82016-03-15 14:57:22 -07001667 struct mem_cgroup *memcg = page->mem_cgroup;
1668
Johannes Weiner6de22612015-02-11 15:25:01 -08001669 if (memcg && memcg->move_lock_task == current) {
1670 unsigned long flags = memcg->move_lock_flags;
1671
1672 memcg->move_lock_task = NULL;
1673 memcg->move_lock_flags = 0;
1674
1675 spin_unlock_irqrestore(&memcg->move_lock, flags);
1676 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001677
Johannes Weinerd7365e72014-10-29 14:50:48 -07001678 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001679}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001680EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07001681
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08001682/*
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001683 * size of first charge trial. "32" comes from vmscan.c's magic value.
1684 * TODO: maybe necessary to use big numbers in big irons.
1685 */
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001686#define CHARGE_BATCH 32U
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001687struct memcg_stock_pcp {
1688 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001689 unsigned int nr_pages;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001690 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001691 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07001692#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001693};
1694static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02001695static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001696
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001697/**
1698 * consume_stock: Try to consume stocked charge on this cpu.
1699 * @memcg: memcg to consume from.
1700 * @nr_pages: how many pages to charge.
1701 *
1702 * The charges will only happen if @memcg matches the current cpu's memcg
1703 * stock, and at least @nr_pages are available in that stock. Failure to
1704 * service an allocation will refill the stock.
1705 *
1706 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001707 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001708static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001709{
1710 struct memcg_stock_pcp *stock;
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001711 unsigned long flags;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001712 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001713
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001714 if (nr_pages > CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001715 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001716
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001717 local_irq_save(flags);
1718
1719 stock = this_cpu_ptr(&memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001720 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08001721 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001722 ret = true;
1723 }
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001724
1725 local_irq_restore(flags);
1726
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001727 return ret;
1728}
1729
1730/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001731 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001732 */
1733static void drain_stock(struct memcg_stock_pcp *stock)
1734{
1735 struct mem_cgroup *old = stock->cached;
1736
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001737 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001738 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001739 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001740 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001741 css_put_many(&old->css, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001742 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001743 }
1744 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001745}
1746
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001747static void drain_local_stock(struct work_struct *dummy)
1748{
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001749 struct memcg_stock_pcp *stock;
1750 unsigned long flags;
1751
1752 local_irq_save(flags);
1753
1754 stock = this_cpu_ptr(&memcg_stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001755 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001756 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001757
1758 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001759}
1760
1761/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001762 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01001763 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001764 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001765static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001766{
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001767 struct memcg_stock_pcp *stock;
1768 unsigned long flags;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001769
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001770 local_irq_save(flags);
1771
1772 stock = this_cpu_ptr(&memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001773 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001774 drain_stock(stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001775 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001776 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07001777 stock->nr_pages += nr_pages;
Johannes Weinerdb2ba402016-09-19 14:44:36 -07001778
1779 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001780}
1781
1782/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001783 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001784 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001785 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001786static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001787{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001788 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07001789
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001790 /* If someone's already draining, avoid adding running more workers. */
1791 if (!mutex_trylock(&percpu_charge_mutex))
1792 return;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001793 /* Notify other cpus that system-wide "drain" is running */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001794 get_online_cpus();
Johannes Weiner5af12d02011-08-25 15:59:07 -07001795 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001796 for_each_online_cpu(cpu) {
1797 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001798 struct mem_cgroup *memcg;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001799
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001800 memcg = stock->cached;
1801 if (!memcg || !stock->nr_pages)
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07001802 continue;
Johannes Weiner2314b422014-12-10 15:44:33 -08001803 if (!mem_cgroup_is_descendant(memcg, root_memcg))
Michal Hocko3e920412011-07-26 16:08:29 -07001804 continue;
Michal Hockod1a05b62011-07-26 16:08:27 -07001805 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1806 if (cpu == curcpu)
1807 drain_local_stock(&stock->work);
1808 else
1809 schedule_work_on(cpu, &stock->work);
1810 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001811 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07001812 put_cpu();
Andrew Mortonf894ffa2013-09-12 15:13:35 -07001813 put_online_cpus();
Michal Hocko9f50fad2011-08-09 11:56:26 +02001814 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001815}
1816
Paul Gortmaker0db06282013-06-19 14:53:51 -04001817static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001818 unsigned long action,
1819 void *hcpu)
1820{
1821 int cpu = (unsigned long)hcpu;
1822 struct memcg_stock_pcp *stock;
1823
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001824 if (action == CPU_ONLINE)
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001825 return NOTIFY_OK;
KAMEZAWA Hiroyuki1489eba2010-10-27 15:33:42 -07001826
Kirill A. Shutemovd8330492012-04-12 12:49:11 -07001827 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001828 return NOTIFY_OK;
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001829
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001830 stock = &per_cpu(memcg_stock, cpu);
1831 drain_stock(stock);
1832 return NOTIFY_OK;
1833}
1834
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001835static void reclaim_high(struct mem_cgroup *memcg,
1836 unsigned int nr_pages,
1837 gfp_t gfp_mask)
1838{
1839 do {
1840 if (page_counter_read(&memcg->memory) <= memcg->high)
1841 continue;
1842 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
1843 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1844 } while ((memcg = parent_mem_cgroup(memcg)));
1845}
1846
1847static void high_work_func(struct work_struct *work)
1848{
1849 struct mem_cgroup *memcg;
1850
1851 memcg = container_of(work, struct mem_cgroup, high_work);
1852 reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1853}
1854
Tejun Heob23afb92015-11-05 18:46:11 -08001855/*
1856 * Scheduled by try_charge() to be executed from the userland return path
1857 * and reclaims memory over the high limit.
1858 */
1859void mem_cgroup_handle_over_high(void)
1860{
1861 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001862 struct mem_cgroup *memcg;
Tejun Heob23afb92015-11-05 18:46:11 -08001863
1864 if (likely(!nr_pages))
1865 return;
1866
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08001867 memcg = get_mem_cgroup_from_mm(current->mm);
1868 reclaim_high(memcg, nr_pages, GFP_KERNEL);
Tejun Heob23afb92015-11-05 18:46:11 -08001869 css_put(&memcg->css);
1870 current->memcg_nr_pages_over_high = 0;
1871}
1872
Johannes Weiner00501b52014-08-08 14:19:20 -07001873static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1874 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001875{
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001876 unsigned int batch = max(CHARGE_BATCH, nr_pages);
Johannes Weiner9b130612014-08-06 16:05:51 -07001877 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001878 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001879 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001880 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001881 bool may_swap = true;
1882 bool drained = false;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001883
Johannes Weinerce00a962014-09-05 08:43:57 -04001884 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08001885 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001886retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07001887 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08001888 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08001889
Johannes Weiner7941d212016-01-14 15:21:23 -08001890 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08001891 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1892 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001893 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08001894 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001895 page_counter_uncharge(&memcg->memsw, batch);
1896 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001897 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001898 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001899 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07001900 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001901
Johannes Weiner6539cc02014-08-06 16:05:42 -07001902 if (batch > nr_pages) {
1903 batch = nr_pages;
1904 goto retry;
1905 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08001906
Johannes Weiner06b078f2014-08-06 16:05:44 -07001907 /*
1908 * Unlike in global OOM situations, memcg is not in a physical
1909 * memory shortage. Allow dying and OOM-killed tasks to
1910 * bypass the last charges so that they can exit quickly and
1911 * free their memory.
1912 */
1913 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1914 fatal_signal_pending(current) ||
1915 current->flags & PF_EXITING))
Tejun Heo10d53c72015-11-05 18:46:17 -08001916 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001917
Johannes Weiner89a28482016-10-27 17:46:56 -07001918 /*
1919 * Prevent unbounded recursion when reclaim operations need to
1920 * allocate memory. This might exceed the limits temporarily,
1921 * but we prefer facilitating memory reclaim and getting back
1922 * under the limit over triggering OOM kills in these cases.
1923 */
1924 if (unlikely(current->flags & PF_MEMALLOC))
1925 goto force;
1926
Johannes Weiner06b078f2014-08-06 16:05:44 -07001927 if (unlikely(task_in_memcg_oom(current)))
1928 goto nomem;
1929
Mel Gormand0164ad2015-11-06 16:28:21 -08001930 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001931 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001932
Johannes Weiner241994e2015-02-11 15:26:06 -08001933 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
1934
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001935 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1936 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07001937
Johannes Weiner61e02c72014-08-06 16:08:16 -07001938 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07001939 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07001940
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001941 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08001942 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07001943 drained = true;
1944 goto retry;
1945 }
1946
Johannes Weiner28c34c22014-08-06 16:05:47 -07001947 if (gfp_mask & __GFP_NORETRY)
1948 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001949 /*
1950 * Even though the limit is exceeded at this point, reclaim
1951 * may have been able to free some pages. Retry the charge
1952 * before killing the task.
1953 *
1954 * Only for regular pages, though: huge pages are rather
1955 * unlikely to succeed so close to the limit, and we fall back
1956 * to regular pages anyway in case of failure.
1957 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07001958 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07001959 goto retry;
1960 /*
1961 * At task move, charge accounts can be doubly counted. So, it's
1962 * better to wait until the end of task_move if something is going on.
1963 */
1964 if (mem_cgroup_wait_acct_move(mem_over_limit))
1965 goto retry;
1966
Johannes Weiner9b130612014-08-06 16:05:51 -07001967 if (nr_retries--)
1968 goto retry;
1969
Johannes Weiner06b078f2014-08-06 16:05:44 -07001970 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08001971 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07001972
Johannes Weiner6539cc02014-08-06 16:05:42 -07001973 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08001974 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001975
Johannes Weiner241994e2015-02-11 15:26:06 -08001976 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
1977
Jerome Marchand3608de02015-11-05 18:47:29 -08001978 mem_cgroup_oom(mem_over_limit, gfp_mask,
1979 get_order(nr_pages * PAGE_SIZE));
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08001980nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07001981 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07001982 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08001983force:
1984 /*
1985 * The allocation either can't fail or will lead to more memory
1986 * being freed very soon. Allow memory usage go over the limit
1987 * temporarily by force charging it.
1988 */
1989 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08001990 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08001991 page_counter_charge(&memcg->memsw, nr_pages);
1992 css_get_many(&memcg->css, nr_pages);
1993
1994 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07001995
1996done_restock:
Johannes Weinere8ea14c2014-12-10 15:42:42 -08001997 css_get_many(&memcg->css, batch);
Johannes Weiner6539cc02014-08-06 16:05:42 -07001998 if (batch > nr_pages)
1999 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002000
Johannes Weiner241994e2015-02-11 15:26:06 -08002001 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002002 * If the hierarchy is above the normal consumption range, schedule
2003 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002004 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002005 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2006 * not recorded as it most likely matches current's and won't
2007 * change in the meantime. As high limit is checked again before
2008 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994e2015-02-11 15:26:06 -08002009 */
2010 do {
Tejun Heob23afb92015-11-05 18:46:11 -08002011 if (page_counter_read(&memcg->memory) > memcg->high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002012 /* Don't bother a random interrupted task */
2013 if (in_interrupt()) {
2014 schedule_work(&memcg->high_work);
2015 break;
2016 }
Vladimir Davydov9516a182015-12-11 13:40:24 -08002017 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002018 set_notify_resume(current);
2019 break;
2020 }
Johannes Weiner241994e2015-02-11 15:26:06 -08002021 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002022
2023 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002024}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002025
Johannes Weiner00501b52014-08-08 14:19:20 -07002026static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002027{
Johannes Weinerce00a962014-09-05 08:43:57 -04002028 if (mem_cgroup_is_root(memcg))
2029 return;
2030
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002031 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002032 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002033 page_counter_uncharge(&memcg->memsw, nr_pages);
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002034
Johannes Weinere8ea14c2014-12-10 15:42:42 -08002035 css_put_many(&memcg->css, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002036}
2037
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002038static void lock_page_lru(struct page *page, int *isolated)
2039{
2040 struct zone *zone = page_zone(page);
2041
Mel Gormana52633d2016-07-28 15:45:28 -07002042 spin_lock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002043 if (PageLRU(page)) {
2044 struct lruvec *lruvec;
2045
Mel Gorman599d0c92016-07-28 15:45:31 -07002046 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002047 ClearPageLRU(page);
2048 del_page_from_lru_list(page, lruvec, page_lru(page));
2049 *isolated = 1;
2050 } else
2051 *isolated = 0;
2052}
2053
2054static void unlock_page_lru(struct page *page, int isolated)
2055{
2056 struct zone *zone = page_zone(page);
2057
2058 if (isolated) {
2059 struct lruvec *lruvec;
2060
Mel Gorman599d0c92016-07-28 15:45:31 -07002061 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002062 VM_BUG_ON_PAGE(PageLRU(page), page);
2063 SetPageLRU(page);
2064 add_page_to_lru_list(page, lruvec, page_lru(page));
2065 }
Mel Gormana52633d2016-07-28 15:45:28 -07002066 spin_unlock_irq(zone_lru_lock(zone));
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002067}
2068
Johannes Weiner00501b52014-08-08 14:19:20 -07002069static void commit_charge(struct page *page, struct mem_cgroup *memcg,
Johannes Weiner6abb5a82014-08-08 14:19:33 -07002070 bool lrucare)
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002071{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002072 int isolated;
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002073
Johannes Weiner1306a852014-12-10 15:44:52 -08002074 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002075
2076 /*
2077 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2078 * may already be on some other mem_cgroup's LRU. Take care of it.
2079 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002080 if (lrucare)
2081 lock_page_lru(page, &isolated);
Hugh Dickins9ce70c02012-03-05 14:59:16 -08002082
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002083 /*
2084 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08002085 * page->mem_cgroup at this point:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002086 *
2087 * - the page is uncharged
2088 *
2089 * - the page is off-LRU
2090 *
2091 * - an anonymous fault has exclusive page access, except for
2092 * a locked page table
2093 *
2094 * - a page cache insertion, a swapin fault, or a migration
2095 * have the page locked
2096 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002097 page->mem_cgroup = memcg;
Hugh Dickins3be91272008-02-07 00:14:19 -08002098
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002099 if (lrucare)
2100 unlock_page_lru(page, isolated);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002101}
2102
Johannes Weiner127424c2016-01-20 15:02:32 -08002103#ifndef CONFIG_SLOB
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002104static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002105{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002106 int id, size;
2107 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002108
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002109 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002110 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2111 if (id < 0)
2112 return id;
2113
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002114 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002115 return id;
2116
2117 /*
2118 * There's no space for the new id in memcg_caches arrays,
2119 * so we have to grow them.
2120 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002121 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002122
2123 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002124 if (size < MEMCG_CACHES_MIN_SIZE)
2125 size = MEMCG_CACHES_MIN_SIZE;
2126 else if (size > MEMCG_CACHES_MAX_SIZE)
2127 size = MEMCG_CACHES_MAX_SIZE;
2128
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002129 err = memcg_update_all_caches(size);
Vladimir Davydov05257a12015-02-12 14:59:01 -08002130 if (!err)
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002131 err = memcg_update_all_list_lrus(size);
2132 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002133 memcg_nr_cache_ids = size;
2134
2135 up_write(&memcg_cache_ids_sem);
2136
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002137 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002138 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002139 return err;
2140 }
2141 return id;
2142}
2143
2144static void memcg_free_cache_id(int id)
2145{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002146 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002147}
2148
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002149struct memcg_kmem_cache_create_work {
Vladimir Davydov5722d092014-04-07 15:39:24 -07002150 struct mem_cgroup *memcg;
2151 struct kmem_cache *cachep;
2152 struct work_struct work;
2153};
2154
Vladimir Davydov754ae7e2016-12-12 16:41:29 -08002155static struct workqueue_struct *memcg_kmem_cache_create_wq;
2156
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002157static void memcg_kmem_cache_create_func(struct work_struct *w)
Glauber Costad7f25f82012-12-18 14:22:40 -08002158{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002159 struct memcg_kmem_cache_create_work *cw =
2160 container_of(w, struct memcg_kmem_cache_create_work, work);
Vladimir Davydov5722d092014-04-07 15:39:24 -07002161 struct mem_cgroup *memcg = cw->memcg;
2162 struct kmem_cache *cachep = cw->cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002163
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002164 memcg_create_kmem_cache(memcg, cachep);
Vladimir Davydovbd673142014-06-04 16:07:40 -07002165
Vladimir Davydov5722d092014-04-07 15:39:24 -07002166 css_put(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002167 kfree(cw);
2168}
2169
2170/*
2171 * Enqueue the creation of a per-memcg kmem_cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002172 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002173static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2174 struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002175{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002176 struct memcg_kmem_cache_create_work *cw;
Glauber Costad7f25f82012-12-18 14:22:40 -08002177
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002178 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002179 if (!cw)
Glauber Costad7f25f82012-12-18 14:22:40 -08002180 return;
Vladimir Davydov8135be52014-12-12 16:56:38 -08002181
2182 css_get(&memcg->css);
Glauber Costad7f25f82012-12-18 14:22:40 -08002183
2184 cw->memcg = memcg;
2185 cw->cachep = cachep;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002186 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
Glauber Costad7f25f82012-12-18 14:22:40 -08002187
Vladimir Davydov754ae7e2016-12-12 16:41:29 -08002188 queue_work(memcg_kmem_cache_create_wq, &cw->work);
Glauber Costad7f25f82012-12-18 14:22:40 -08002189}
2190
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002191static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2192 struct kmem_cache *cachep)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002193{
2194 /*
2195 * We need to stop accounting when we kmalloc, because if the
2196 * corresponding kmalloc cache is not yet created, the first allocation
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002197 * in __memcg_schedule_kmem_cache_create will recurse.
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002198 *
2199 * However, it is better to enclose the whole function. Depending on
2200 * the debugging options enabled, INIT_WORK(), for instance, can
2201 * trigger an allocation. This too, will make us recurse. Because at
2202 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2203 * the safest choice is to do it like this, wrapping the whole function.
2204 */
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002205 current->memcg_kmem_skip_account = 1;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002206 __memcg_schedule_kmem_cache_create(memcg, cachep);
Vladimir Davydov6f185c22014-12-12 16:55:15 -08002207 current->memcg_kmem_skip_account = 0;
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002208}
Vladimir Davydovc67a8a62014-06-04 16:07:39 -07002209
Vladimir Davydov45264772016-07-26 15:24:21 -07002210static inline bool memcg_kmem_bypass(void)
2211{
2212 if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
2213 return true;
2214 return false;
2215}
2216
2217/**
2218 * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2219 * @cachep: the original global kmem cache
2220 *
Glauber Costad7f25f82012-12-18 14:22:40 -08002221 * Return the kmem_cache we're supposed to use for a slab allocation.
2222 * We try to use the current memcg's version of the cache.
2223 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002224 * If the cache does not exist yet, if we are the first user of it, we
2225 * create it asynchronously in a workqueue and let the current allocation
2226 * go through with the original cache.
Glauber Costad7f25f82012-12-18 14:22:40 -08002227 *
Vladimir Davydov45264772016-07-26 15:24:21 -07002228 * This function takes a reference to the cache it returns to assure it
2229 * won't get destroyed while we are working with it. Once the caller is
2230 * done with it, memcg_kmem_put_cache() must be called to release the
2231 * reference.
Glauber Costad7f25f82012-12-18 14:22:40 -08002232 */
Vladimir Davydov45264772016-07-26 15:24:21 -07002233struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep)
Glauber Costad7f25f82012-12-18 14:22:40 -08002234{
2235 struct mem_cgroup *memcg;
Vladimir Davydov959c8962014-01-23 15:52:59 -08002236 struct kmem_cache *memcg_cachep;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002237 int kmemcg_id;
Glauber Costad7f25f82012-12-18 14:22:40 -08002238
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002239 VM_BUG_ON(!is_root_cache(cachep));
Glauber Costad7f25f82012-12-18 14:22:40 -08002240
Vladimir Davydov45264772016-07-26 15:24:21 -07002241 if (memcg_kmem_bypass())
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08002242 return cachep;
2243
Vladimir Davydov9d100c52014-12-12 16:54:53 -08002244 if (current->memcg_kmem_skip_account)
Glauber Costa0e9d92f2012-12-18 14:22:42 -08002245 return cachep;
2246
Vladimir Davydov8135be52014-12-12 16:56:38 -08002247 memcg = get_mem_cgroup_from_mm(current->mm);
Jason Low4db0c3c2015-04-15 16:14:08 -07002248 kmemcg_id = READ_ONCE(memcg->kmemcg_id);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002249 if (kmemcg_id < 0)
Li Zefanca0dde92013-04-29 15:08:57 -07002250 goto out;
Glauber Costad7f25f82012-12-18 14:22:40 -08002251
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002252 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002253 if (likely(memcg_cachep))
2254 return memcg_cachep;
Li Zefanca0dde92013-04-29 15:08:57 -07002255
2256 /*
2257 * If we are in a safe context (can wait, and not in interrupt
2258 * context), we could be be predictable and return right away.
2259 * This would guarantee that the allocation being performed
2260 * already belongs in the new cache.
2261 *
2262 * However, there are some clashes that can arrive from locking.
2263 * For instance, because we acquire the slab_mutex while doing
Vladimir Davydov776ed0f2014-06-04 16:10:02 -07002264 * memcg_create_kmem_cache, this means no further allocation
2265 * could happen with the slab_mutex held. So it's better to
2266 * defer everything.
Li Zefanca0dde92013-04-29 15:08:57 -07002267 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -08002268 memcg_schedule_kmem_cache_create(memcg, cachep);
Li Zefanca0dde92013-04-29 15:08:57 -07002269out:
Vladimir Davydov8135be52014-12-12 16:56:38 -08002270 css_put(&memcg->css);
Li Zefanca0dde92013-04-29 15:08:57 -07002271 return cachep;
Glauber Costad7f25f82012-12-18 14:22:40 -08002272}
Glauber Costad7f25f82012-12-18 14:22:40 -08002273
Vladimir Davydov45264772016-07-26 15:24:21 -07002274/**
2275 * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2276 * @cachep: the cache returned by memcg_kmem_get_cache
2277 */
2278void memcg_kmem_put_cache(struct kmem_cache *cachep)
Vladimir Davydov8135be52014-12-12 16:56:38 -08002279{
2280 if (!is_root_cache(cachep))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08002281 css_put(&cachep->memcg_params.memcg->css);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002282}
2283
Vladimir Davydov45264772016-07-26 15:24:21 -07002284/**
2285 * memcg_kmem_charge: charge a kmem page
2286 * @page: page to charge
2287 * @gfp: reclaim mode
2288 * @order: allocation order
2289 * @memcg: memory cgroup to charge
2290 *
2291 * Returns 0 on success, an error code on failure.
2292 */
2293int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2294 struct mem_cgroup *memcg)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002295{
2296 unsigned int nr_pages = 1 << order;
2297 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002298 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002299
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002300 ret = try_charge(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08002301 if (ret)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002302 return ret;
Johannes Weiner52c29b02016-01-20 15:02:35 -08002303
2304 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2305 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2306 cancel_charge(memcg, nr_pages);
2307 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002308 }
2309
2310 page->mem_cgroup = memcg;
2311
2312 return 0;
2313}
2314
Vladimir Davydov45264772016-07-26 15:24:21 -07002315/**
2316 * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2317 * @page: page to charge
2318 * @gfp: reclaim mode
2319 * @order: allocation order
2320 *
2321 * Returns 0 on success, an error code on failure.
2322 */
2323int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002324{
2325 struct mem_cgroup *memcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07002326 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002327
Vladimir Davydov45264772016-07-26 15:24:21 -07002328 if (memcg_kmem_bypass())
2329 return 0;
2330
Johannes Weinerdf381972014-04-07 15:37:43 -07002331 memcg = get_mem_cgroup_from_mm(current->mm);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002332 if (!mem_cgroup_is_root(memcg)) {
Vladimir Davydov45264772016-07-26 15:24:21 -07002333 ret = memcg_kmem_charge_memcg(page, gfp, order, memcg);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002334 if (!ret)
2335 __SetPageKmemcg(page);
2336 }
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002337 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08002338 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002339}
Vladimir Davydov45264772016-07-26 15:24:21 -07002340/**
2341 * memcg_kmem_uncharge: uncharge a kmem page
2342 * @page: page to uncharge
2343 * @order: allocation order
2344 */
2345void memcg_kmem_uncharge(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002346{
Johannes Weiner1306a852014-12-10 15:44:52 -08002347 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002348 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002349
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002350 if (!memcg)
2351 return;
2352
Sasha Levin309381fea2014-01-23 15:52:54 -08002353 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Johannes Weiner29833312014-12-10 15:44:02 -08002354
Johannes Weiner52c29b02016-01-20 15:02:35 -08002355 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2356 page_counter_uncharge(&memcg->kmem, nr_pages);
2357
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002358 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002359 if (do_memsw_account())
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002360 page_counter_uncharge(&memcg->memsw, nr_pages);
2361
Johannes Weiner1306a852014-12-10 15:44:52 -08002362 page->mem_cgroup = NULL;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03002363
2364 /* slab pages do not have PageKmemcg flag set */
2365 if (PageKmemcg(page))
2366 __ClearPageKmemcg(page);
2367
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002368 css_put_many(&memcg->css, nr_pages);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002369}
Johannes Weiner127424c2016-01-20 15:02:32 -08002370#endif /* !CONFIG_SLOB */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08002371
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002372#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2373
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002374/*
2375 * Because tail pages are not marked as "used", set it. We're under
Mel Gormana52633d2016-07-28 15:45:28 -07002376 * zone_lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002377 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002378void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002379{
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08002380 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002381
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08002382 if (mem_cgroup_disabled())
2383 return;
David Rientjesb070e652013-05-07 16:18:09 -07002384
Johannes Weiner29833312014-12-10 15:44:02 -08002385 for (i = 1; i < HPAGE_PMD_NR; i++)
Johannes Weiner1306a852014-12-10 15:44:52 -08002386 head[i].mem_cgroup = head->mem_cgroup;
Michal Hockob9982f82014-12-10 15:43:51 -08002387
Johannes Weiner1306a852014-12-10 15:44:52 -08002388 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
David Rientjesb070e652013-05-07 16:18:09 -07002389 HPAGE_PMD_NR);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002390}
Hugh Dickins12d27102012-01-12 17:19:52 -08002391#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08002392
Andrew Mortonc255a452012-07-31 16:43:02 -07002393#ifdef CONFIG_MEMCG_SWAP
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002394static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2395 bool charge)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002396{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002397 int val = (charge) ? 1 : -1;
2398 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002399}
Daisuke Nishimura02491442010-03-10 15:22:17 -08002400
2401/**
2402 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2403 * @entry: swap entry to be moved
2404 * @from: mem_cgroup which the entry is moved from
2405 * @to: mem_cgroup which the entry is moved to
2406 *
2407 * It succeeds only when the swap_cgroup's record for this entry is the same
2408 * as the mem_cgroup's id of @from.
2409 *
2410 * Returns 0 on success, -EINVAL on failure.
2411 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002412 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08002413 * both res and memsw, and called css_get().
2414 */
2415static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002416 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002417{
2418 unsigned short old_id, new_id;
2419
Li Zefan34c00c32013-09-23 16:56:01 +08002420 old_id = mem_cgroup_id(from);
2421 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002422
2423 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08002424 mem_cgroup_swap_statistics(from, false);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002425 mem_cgroup_swap_statistics(to, true);
Daisuke Nishimura02491442010-03-10 15:22:17 -08002426 return 0;
2427 }
2428 return -EINVAL;
2429}
2430#else
2431static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07002432 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08002433{
2434 return -EINVAL;
2435}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002436#endif
2437
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002438static DEFINE_MUTEX(memcg_limit_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07002439
KOSAKI Motohirod38d2a72009-01-06 14:39:44 -08002440static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002441 unsigned long limit)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002442{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002443 unsigned long curusage;
2444 unsigned long oldusage;
2445 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002446 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002447 int ret;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002448
2449 /*
2450 * For keeping hierarchical_reclaim simple, how long we should retry
2451 * is depends on callers. We set our retry-count to be function
2452 * of # of children which we should visit in this loop.
2453 */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002454 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2455 mem_cgroup_count_children(memcg);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002456
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002457 oldusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002458
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002459 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002460 if (signal_pending(current)) {
2461 ret = -EINTR;
2462 break;
2463 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002464
2465 mutex_lock(&memcg_limit_mutex);
2466 if (limit > memcg->memsw.limit) {
2467 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002468 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002469 break;
2470 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002471 if (limit > memcg->memory.limit)
2472 enlarge = true;
2473 ret = page_counter_limit(&memcg->memory, limit);
2474 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002475
2476 if (!ret)
2477 break;
2478
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002479 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2480
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002481 curusage = page_counter_read(&memcg->memory);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002482 /* Usage is reduced ? */
Andrew Mortonf894ffa2013-09-12 15:13:35 -07002483 if (curusage >= oldusage)
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002484 retry_count--;
2485 else
2486 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002487 } while (retry_count);
2488
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002489 if (!ret && enlarge)
2490 memcg_oom_recover(memcg);
KOSAKI Motohiro14797e22009-01-07 18:08:18 -08002491
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002492 return ret;
2493}
2494
Li Zefan338c8432009-06-17 16:27:15 -07002495static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002496 unsigned long limit)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002497{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002498 unsigned long curusage;
2499 unsigned long oldusage;
2500 bool enlarge = false;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002501 int retry_count;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002502 int ret;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002503
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002504 /* see mem_cgroup_resize_res_limit */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002505 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2506 mem_cgroup_count_children(memcg);
2507
2508 oldusage = page_counter_read(&memcg->memsw);
2509
2510 do {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002511 if (signal_pending(current)) {
2512 ret = -EINTR;
2513 break;
2514 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002515
2516 mutex_lock(&memcg_limit_mutex);
2517 if (limit < memcg->memory.limit) {
2518 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002519 ret = -EINVAL;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002520 break;
2521 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002522 if (limit > memcg->memsw.limit)
2523 enlarge = true;
2524 ret = page_counter_limit(&memcg->memsw, limit);
2525 mutex_unlock(&memcg_limit_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002526
2527 if (!ret)
2528 break;
2529
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002530 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2531
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002532 curusage = page_counter_read(&memcg->memsw);
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002533 /* Usage is reduced ? */
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002534 if (curusage >= oldusage)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002535 retry_count--;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07002536 else
2537 oldusage = curusage;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002538 } while (retry_count);
2539
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002540 if (!ret && enlarge)
2541 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002542
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002543 return ret;
2544}
2545
Mel Gormanef8f2322016-07-28 15:46:05 -07002546unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
Andrew Morton0608f432013-09-24 15:27:41 -07002547 gfp_t gfp_mask,
2548 unsigned long *total_scanned)
2549{
2550 unsigned long nr_reclaimed = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002551 struct mem_cgroup_per_node *mz, *next_mz = NULL;
Andrew Morton0608f432013-09-24 15:27:41 -07002552 unsigned long reclaimed;
2553 int loop = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002554 struct mem_cgroup_tree_per_node *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002555 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07002556 unsigned long nr_scanned;
2557
2558 if (order > 0)
2559 return 0;
2560
Mel Gormanef8f2322016-07-28 15:46:05 -07002561 mctz = soft_limit_tree_node(pgdat->node_id);
Michal Hockod6507ff2016-08-02 14:02:37 -07002562
2563 /*
2564 * Do not even bother to check the largest node if the root
2565 * is empty. Do it lockless to prevent lock bouncing. Races
2566 * are acceptable as soft limit is best effort anyway.
2567 */
2568 if (RB_EMPTY_ROOT(&mctz->rb_root))
2569 return 0;
2570
Andrew Morton0608f432013-09-24 15:27:41 -07002571 /*
2572 * This loop can run a while, specially if mem_cgroup's continuously
2573 * keep exceeding their soft limit and putting the system under
2574 * pressure
2575 */
2576 do {
2577 if (next_mz)
2578 mz = next_mz;
2579 else
2580 mz = mem_cgroup_largest_soft_limit_node(mctz);
2581 if (!mz)
2582 break;
2583
2584 nr_scanned = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07002585 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07002586 gfp_mask, &nr_scanned);
2587 nr_reclaimed += reclaimed;
2588 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002589 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002590 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07002591
2592 /*
2593 * If we failed to reclaim anything from this memory cgroup
2594 * it is time to move on to the next cgroup
2595 */
2596 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08002597 if (!reclaimed)
2598 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2599
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002600 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07002601 /*
2602 * One school of thought says that we should not add
2603 * back the node to the tree if reclaim returns 0.
2604 * But our reclaim could return 0, simply because due
2605 * to priority we are exposing a smaller subset of
2606 * memory to reclaim from. Consider this as a longer
2607 * term TODO.
2608 */
2609 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07002610 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002611 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07002612 css_put(&mz->memcg->css);
2613 loop++;
2614 /*
2615 * Could not reclaim anything and there are no more
2616 * mem cgroups to try or we seem to be looping without
2617 * reclaiming anything.
2618 */
2619 if (!nr_reclaimed &&
2620 (next_mz == NULL ||
2621 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2622 break;
2623 } while (!nr_reclaimed);
2624 if (next_mz)
2625 css_put(&next_mz->memcg->css);
2626 return nr_reclaimed;
2627}
2628
Tejun Heoea280e72014-05-16 13:22:48 -04002629/*
2630 * Test whether @memcg has children, dead or alive. Note that this
2631 * function doesn't care whether @memcg has use_hierarchy enabled and
2632 * returns %true if there are child csses according to the cgroup
2633 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2634 */
Glauber Costab5f99b52013-02-22 16:34:53 -08002635static inline bool memcg_has_children(struct mem_cgroup *memcg)
2636{
Tejun Heoea280e72014-05-16 13:22:48 -04002637 bool ret;
2638
Tejun Heoea280e72014-05-16 13:22:48 -04002639 rcu_read_lock();
2640 ret = css_next_child(NULL, &memcg->css);
2641 rcu_read_unlock();
2642 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08002643}
2644
2645/*
Greg Thelen51038172016-05-20 16:58:18 -07002646 * Reclaims as many pages from the given memcg as possible.
Michal Hockoc26251f2012-10-26 13:37:28 +02002647 *
2648 * Caller is responsible for holding css reference for memcg.
2649 */
2650static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2651{
2652 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02002653
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002654 /* we call try-to-free pages for make this cgroup empty */
2655 lru_add_drain_all();
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002656 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002657 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002658 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002659
Michal Hockoc26251f2012-10-26 13:37:28 +02002660 if (signal_pending(current))
2661 return -EINTR;
2662
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002663 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2664 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002665 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002666 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002667 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02002668 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002669 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08002670
2671 }
Michal Hockoab5196c2012-10-26 13:37:32 +02002672
2673 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08002674}
2675
Tejun Heo6770c642014-05-13 12:16:21 -04002676static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2677 char *buf, size_t nbytes,
2678 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002679{
Tejun Heo6770c642014-05-13 12:16:21 -04002680 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02002681
Michal Hockod8423012012-10-26 13:37:29 +02002682 if (mem_cgroup_is_root(memcg))
2683 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04002684 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08002685}
2686
Tejun Heo182446d2013-08-08 20:11:24 -04002687static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2688 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002689{
Tejun Heo182446d2013-08-08 20:11:24 -04002690 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002691}
2692
Tejun Heo182446d2013-08-08 20:11:24 -04002693static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2694 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08002695{
2696 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04002697 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04002698 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08002699
Glauber Costa567fb432012-07-31 16:43:07 -07002700 if (memcg->use_hierarchy == val)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002701 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07002702
Balbir Singh18f59ea2009-01-07 18:08:07 -08002703 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002704 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08002705 * in the child subtrees. If it is unset, then the change can
2706 * occur, provided the current cgroup has no children.
2707 *
2708 * For the root cgroup, parent_mem is NULL, we allow value to be
2709 * set if there are no children.
2710 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002711 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08002712 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04002713 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002714 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08002715 else
2716 retval = -EBUSY;
2717 } else
2718 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07002719
Balbir Singh18f59ea2009-01-07 18:08:07 -08002720 return retval;
2721}
2722
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002723static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
Johannes Weinerce00a962014-09-05 08:43:57 -04002724{
2725 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002726 int i;
Johannes Weinerce00a962014-09-05 08:43:57 -04002727
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002728 memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
Johannes Weinerce00a962014-09-05 08:43:57 -04002729
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002730 for_each_mem_cgroup_tree(iter, memcg) {
2731 for (i = 0; i < MEMCG_NR_STAT; i++)
2732 stat[i] += mem_cgroup_read_stat(iter, i);
2733 }
Johannes Weinerce00a962014-09-05 08:43:57 -04002734}
2735
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002736static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
Johannes Weiner587d9f72016-01-20 15:03:19 -08002737{
2738 struct mem_cgroup *iter;
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002739 int i;
Johannes Weiner587d9f72016-01-20 15:03:19 -08002740
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002741 memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
Johannes Weiner587d9f72016-01-20 15:03:19 -08002742
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002743 for_each_mem_cgroup_tree(iter, memcg) {
2744 for (i = 0; i < MEMCG_NR_EVENTS; i++)
2745 events[i] += mem_cgroup_read_events(iter, i);
2746 }
Johannes Weiner587d9f72016-01-20 15:03:19 -08002747}
2748
Andrew Morton6f646152015-11-06 16:28:58 -08002749static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04002750{
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002751 unsigned long val = 0;
Johannes Weinerce00a962014-09-05 08:43:57 -04002752
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002753 if (mem_cgroup_is_root(memcg)) {
Vladimir Davydov72b54e72016-03-17 14:17:32 -07002754 struct mem_cgroup *iter;
2755
2756 for_each_mem_cgroup_tree(iter, memcg) {
2757 val += mem_cgroup_read_stat(iter,
2758 MEM_CGROUP_STAT_CACHE);
2759 val += mem_cgroup_read_stat(iter,
2760 MEM_CGROUP_STAT_RSS);
2761 if (swap)
2762 val += mem_cgroup_read_stat(iter,
2763 MEM_CGROUP_STAT_SWAP);
2764 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002765 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04002766 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002767 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04002768 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002769 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04002770 }
Michal Hockoc12176d2015-11-05 18:50:29 -08002771 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04002772}
2773
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002774enum {
2775 RES_USAGE,
2776 RES_LIMIT,
2777 RES_MAX_USAGE,
2778 RES_FAILCNT,
2779 RES_SOFT_LIMIT,
2780};
Johannes Weinerce00a962014-09-05 08:43:57 -04002781
Tejun Heo791badb2013-12-05 12:28:02 -05002782static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07002783 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002784{
Tejun Heo182446d2013-08-08 20:11:24 -04002785 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002786 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07002787
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002788 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002789 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002790 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08002791 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002792 case _MEMSWAP:
2793 counter = &memcg->memsw;
2794 break;
2795 case _KMEM:
2796 counter = &memcg->kmem;
2797 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002798 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08002799 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002800 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002801 default:
2802 BUG();
2803 }
2804
2805 switch (MEMFILE_ATTR(cft->private)) {
2806 case RES_USAGE:
2807 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08002808 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002809 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08002810 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002811 return (u64)page_counter_read(counter) * PAGE_SIZE;
2812 case RES_LIMIT:
2813 return (u64)counter->limit * PAGE_SIZE;
2814 case RES_MAX_USAGE:
2815 return (u64)counter->watermark * PAGE_SIZE;
2816 case RES_FAILCNT:
2817 return counter->failcnt;
2818 case RES_SOFT_LIMIT:
2819 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002820 default:
2821 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002822 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002823}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002824
Johannes Weiner127424c2016-01-20 15:02:32 -08002825#ifndef CONFIG_SLOB
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002826static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08002827{
Vladimir Davydovd6441632014-01-23 15:53:09 -08002828 int memcg_id;
2829
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002830 if (cgroup_memory_nokmem)
2831 return 0;
2832
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08002833 BUG_ON(memcg->kmemcg_id >= 0);
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002834 BUG_ON(memcg->kmem_state);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002835
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002836 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002837 if (memcg_id < 0)
2838 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002839
Johannes Weineref129472016-01-14 15:21:34 -08002840 static_branch_inc(&memcg_kmem_enabled_key);
Vladimir Davydovd6441632014-01-23 15:53:09 -08002841 /*
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002842 * A memory cgroup is considered kmem-online as soon as it gets
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002843 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08002844 * guarantee no one starts accounting before all call sites are
2845 * patched.
2846 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08002847 memcg->kmemcg_id = memcg_id;
Johannes Weiner567e9ab2016-01-20 15:02:24 -08002848 memcg->kmem_state = KMEM_ONLINE;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002849
2850 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08002851}
2852
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002853static void memcg_offline_kmem(struct mem_cgroup *memcg)
2854{
2855 struct cgroup_subsys_state *css;
2856 struct mem_cgroup *parent, *child;
2857 int kmemcg_id;
2858
2859 if (memcg->kmem_state != KMEM_ONLINE)
2860 return;
2861 /*
2862 * Clear the online state before clearing memcg_caches array
2863 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2864 * guarantees that no cache will be created for this cgroup
2865 * after we are done (see memcg_create_kmem_cache()).
2866 */
2867 memcg->kmem_state = KMEM_ALLOCATED;
2868
2869 memcg_deactivate_kmem_caches(memcg);
2870
2871 kmemcg_id = memcg->kmemcg_id;
2872 BUG_ON(kmemcg_id < 0);
2873
2874 parent = parent_mem_cgroup(memcg);
2875 if (!parent)
2876 parent = root_mem_cgroup;
2877
2878 /*
2879 * Change kmemcg_id of this cgroup and all its descendants to the
2880 * parent's id, and then move all entries from this cgroup's list_lrus
2881 * to ones of the parent. After we have finished, all list_lrus
2882 * corresponding to this cgroup are guaranteed to remain empty. The
2883 * ordering is imposed by list_lru_node->lock taken by
2884 * memcg_drain_all_list_lrus().
2885 */
Tejun Heo3a06bb72016-06-03 14:55:44 -07002886 rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002887 css_for_each_descendant_pre(css, &memcg->css) {
2888 child = mem_cgroup_from_css(css);
2889 BUG_ON(child->kmemcg_id != kmemcg_id);
2890 child->kmemcg_id = parent->kmemcg_id;
2891 if (!memcg->use_hierarchy)
2892 break;
2893 }
Tejun Heo3a06bb72016-06-03 14:55:44 -07002894 rcu_read_unlock();
2895
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002896 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2897
2898 memcg_free_cache_id(kmemcg_id);
2899}
2900
2901static void memcg_free_kmem(struct mem_cgroup *memcg)
2902{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002903 /* css_alloc() failed, offlining didn't happen */
2904 if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2905 memcg_offline_kmem(memcg);
2906
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002907 if (memcg->kmem_state == KMEM_ALLOCATED) {
2908 memcg_destroy_kmem_caches(memcg);
2909 static_branch_dec(&memcg_kmem_enabled_key);
2910 WARN_ON(page_counter_read(&memcg->kmem));
2911 }
Johannes Weiner8e0a8912016-01-20 15:02:26 -08002912}
Vladimir Davydovd6441632014-01-23 15:53:09 -08002913#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08002914static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08002915{
2916 return 0;
2917}
2918static void memcg_offline_kmem(struct mem_cgroup *memcg)
2919{
2920}
2921static void memcg_free_kmem(struct mem_cgroup *memcg)
2922{
2923}
2924#endif /* !CONFIG_SLOB */
2925
Johannes Weiner127424c2016-01-20 15:02:32 -08002926static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2927 unsigned long limit)
2928{
Vladimir Davydovb313aee2016-03-17 14:18:27 -07002929 int ret;
Johannes Weiner127424c2016-01-20 15:02:32 -08002930
2931 mutex_lock(&memcg_limit_mutex);
Johannes Weiner127424c2016-01-20 15:02:32 -08002932 ret = page_counter_limit(&memcg->kmem, limit);
Johannes Weiner127424c2016-01-20 15:02:32 -08002933 mutex_unlock(&memcg_limit_mutex);
2934 return ret;
2935}
Glauber Costa510fc4e2012-12-18 14:21:47 -08002936
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002937static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2938{
2939 int ret;
2940
2941 mutex_lock(&memcg_limit_mutex);
2942
Johannes Weiner0db15292016-01-20 15:02:50 -08002943 ret = page_counter_limit(&memcg->tcpmem, limit);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002944 if (ret)
2945 goto out;
2946
Johannes Weiner0db15292016-01-20 15:02:50 -08002947 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002948 /*
2949 * The active flag needs to be written after the static_key
2950 * update. This is what guarantees that the socket activation
Johannes Weiner2d758072016-10-07 17:00:58 -07002951 * function is the last one to run. See mem_cgroup_sk_alloc()
2952 * for details, and note that we don't mark any socket as
2953 * belonging to this memcg until that flag is up.
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002954 *
2955 * We need to do this, because static_keys will span multiple
2956 * sites, but we can't control their order. If we mark a socket
2957 * as accounted, but the accounting functions are not patched in
2958 * yet, we'll lose accounting.
2959 *
Johannes Weiner2d758072016-10-07 17:00:58 -07002960 * We never race with the readers in mem_cgroup_sk_alloc(),
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002961 * because when this value change, the code to process it is not
2962 * patched in yet.
2963 */
2964 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08002965 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002966 }
2967out:
2968 mutex_unlock(&memcg_limit_mutex);
2969 return ret;
2970}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08002971
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002972/*
2973 * The user of this function is...
2974 * RES_LIMIT.
2975 */
Tejun Heo451af502014-05-13 12:16:21 -04002976static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2977 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002978{
Tejun Heo451af502014-05-13 12:16:21 -04002979 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002980 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002981 int ret;
2982
Tejun Heo451af502014-05-13 12:16:21 -04002983 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08002984 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002985 if (ret)
2986 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07002987
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002988 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07002989 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07002990 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2991 ret = -EINVAL;
2992 break;
2993 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002994 switch (MEMFILE_TYPE(of_cft(of)->private)) {
2995 case _MEM:
2996 ret = mem_cgroup_resize_limit(memcg, nr_pages);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08002997 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002998 case _MEMSWAP:
2999 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3000 break;
3001 case _KMEM:
3002 ret = memcg_update_kmem_limit(memcg, nr_pages);
3003 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003004 case _TCP:
3005 ret = memcg_update_tcp_limit(memcg, nr_pages);
3006 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003007 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003008 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003009 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003010 memcg->soft_limit = nr_pages;
3011 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003012 break;
3013 }
Tejun Heo451af502014-05-13 12:16:21 -04003014 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003015}
3016
Tejun Heo6770c642014-05-13 12:16:21 -04003017static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3018 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003019{
Tejun Heo6770c642014-05-13 12:16:21 -04003020 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003021 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003022
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003023 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3024 case _MEM:
3025 counter = &memcg->memory;
3026 break;
3027 case _MEMSWAP:
3028 counter = &memcg->memsw;
3029 break;
3030 case _KMEM:
3031 counter = &memcg->kmem;
3032 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003033 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003034 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003035 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003036 default:
3037 BUG();
3038 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003039
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003040 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003041 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003042 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003043 break;
3044 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003045 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003046 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003047 default:
3048 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003049 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003050
Tejun Heo6770c642014-05-13 12:16:21 -04003051 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003052}
3053
Tejun Heo182446d2013-08-08 20:11:24 -04003054static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003055 struct cftype *cft)
3056{
Tejun Heo182446d2013-08-08 20:11:24 -04003057 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003058}
3059
Daisuke Nishimura02491442010-03-10 15:22:17 -08003060#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003061static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003062 struct cftype *cft, u64 val)
3063{
Tejun Heo182446d2013-08-08 20:11:24 -04003064 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003065
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003066 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003067 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003068
Glauber Costaee5e8472013-02-22 16:34:50 -08003069 /*
3070 * No kind of locking is needed in here, because ->can_attach() will
3071 * check this value once in the beginning of the process, and then carry
3072 * on with stale data. This means that changes to this value will only
3073 * affect task migrations starting after the change.
3074 */
3075 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003076 return 0;
3077}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003078#else
Tejun Heo182446d2013-08-08 20:11:24 -04003079static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003080 struct cftype *cft, u64 val)
3081{
3082 return -ENOSYS;
3083}
3084#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003085
Ying Han406eb0c2011-05-26 16:25:37 -07003086#ifdef CONFIG_NUMA
Tejun Heo2da8ca82013-12-05 12:28:04 -05003087static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003088{
Greg Thelen25485de2013-11-12 15:07:40 -08003089 struct numa_stat {
3090 const char *name;
3091 unsigned int lru_mask;
3092 };
3093
3094 static const struct numa_stat stats[] = {
3095 { "total", LRU_ALL },
3096 { "file", LRU_ALL_FILE },
3097 { "anon", LRU_ALL_ANON },
3098 { "unevictable", BIT(LRU_UNEVICTABLE) },
3099 };
3100 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003101 int nid;
Greg Thelen25485de2013-11-12 15:07:40 -08003102 unsigned long nr;
Tejun Heo2da8ca82013-12-05 12:28:04 -05003103 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Ying Han406eb0c2011-05-26 16:25:37 -07003104
Greg Thelen25485de2013-11-12 15:07:40 -08003105 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3106 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3107 seq_printf(m, "%s=%lu", stat->name, nr);
3108 for_each_node_state(nid, N_MEMORY) {
3109 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3110 stat->lru_mask);
3111 seq_printf(m, " N%d=%lu", nid, nr);
3112 }
3113 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003114 }
Ying Han406eb0c2011-05-26 16:25:37 -07003115
Ying Han071aee12013-11-12 15:07:41 -08003116 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3117 struct mem_cgroup *iter;
Ying Han406eb0c2011-05-26 16:25:37 -07003118
Ying Han071aee12013-11-12 15:07:41 -08003119 nr = 0;
3120 for_each_mem_cgroup_tree(iter, memcg)
3121 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3122 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3123 for_each_node_state(nid, N_MEMORY) {
3124 nr = 0;
3125 for_each_mem_cgroup_tree(iter, memcg)
3126 nr += mem_cgroup_node_nr_lru_pages(
3127 iter, nid, stat->lru_mask);
3128 seq_printf(m, " N%d=%lu", nid, nr);
3129 }
3130 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003131 }
Ying Han406eb0c2011-05-26 16:25:37 -07003132
Ying Han406eb0c2011-05-26 16:25:37 -07003133 return 0;
3134}
3135#endif /* CONFIG_NUMA */
3136
Tejun Heo2da8ca82013-12-05 12:28:04 -05003137static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003138{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003139 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003140 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003141 struct mem_cgroup *mi;
3142 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003143
Greg Thelen0ca44b12015-02-11 15:25:58 -08003144 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3145 MEM_CGROUP_STAT_NSTATS);
3146 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3147 MEM_CGROUP_EVENTS_NSTATS);
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003148 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3149
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003150 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Johannes Weiner7941d212016-01-14 15:21:23 -08003151 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003152 continue;
Greg Thelen484ebb32015-10-01 15:37:05 -07003153 seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003154 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003155 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003156
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003157 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3158 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3159 mem_cgroup_read_events(memcg, i));
3160
3161 for (i = 0; i < NR_LRU_LISTS; i++)
3162 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3163 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3164
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003165 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003166 memory = memsw = PAGE_COUNTER_MAX;
3167 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3168 memory = min(memory, mi->memory.limit);
3169 memsw = min(memsw, mi->memsw.limit);
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003170 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003171 seq_printf(m, "hierarchical_memory_limit %llu\n",
3172 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08003173 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003174 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3175 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003176
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003177 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
Greg Thelen484ebb32015-10-01 15:37:05 -07003178 unsigned long long val = 0;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003179
Johannes Weiner7941d212016-01-14 15:21:23 -08003180 if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003181 continue;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003182 for_each_mem_cgroup_tree(mi, memcg)
3183 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
Greg Thelen484ebb32015-10-01 15:37:05 -07003184 seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003185 }
3186
3187 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3188 unsigned long long val = 0;
3189
3190 for_each_mem_cgroup_tree(mi, memcg)
3191 val += mem_cgroup_read_events(mi, i);
3192 seq_printf(m, "total_%s %llu\n",
3193 mem_cgroup_events_names[i], val);
3194 }
3195
3196 for (i = 0; i < NR_LRU_LISTS; i++) {
3197 unsigned long long val = 0;
3198
3199 for_each_mem_cgroup_tree(mi, memcg)
3200 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3201 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003202 }
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003203
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003204#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003205 {
Mel Gormanef8f2322016-07-28 15:46:05 -07003206 pg_data_t *pgdat;
3207 struct mem_cgroup_per_node *mz;
Hugh Dickins89abfab2012-05-29 15:06:53 -07003208 struct zone_reclaim_stat *rstat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003209 unsigned long recent_rotated[2] = {0, 0};
3210 unsigned long recent_scanned[2] = {0, 0};
3211
Mel Gormanef8f2322016-07-28 15:46:05 -07003212 for_each_online_pgdat(pgdat) {
3213 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
3214 rstat = &mz->lruvec.reclaim_stat;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003215
Mel Gormanef8f2322016-07-28 15:46:05 -07003216 recent_rotated[0] += rstat->recent_rotated[0];
3217 recent_rotated[1] += rstat->recent_rotated[1];
3218 recent_scanned[0] += rstat->recent_scanned[0];
3219 recent_scanned[1] += rstat->recent_scanned[1];
3220 }
Johannes Weiner78ccf5b2012-05-29 15:07:06 -07003221 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3222 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3223 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3224 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003225 }
3226#endif
3227
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003228 return 0;
3229}
3230
Tejun Heo182446d2013-08-08 20:11:24 -04003231static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3232 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003233{
Tejun Heo182446d2013-08-08 20:11:24 -04003234 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003235
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07003236 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003237}
3238
Tejun Heo182446d2013-08-08 20:11:24 -04003239static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3240 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003241{
Tejun Heo182446d2013-08-08 20:11:24 -04003242 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08003243
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003244 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003245 return -EINVAL;
3246
Linus Torvalds14208b02014-06-09 15:03:33 -07003247 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003248 memcg->swappiness = val;
3249 else
3250 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08003251
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003252 return 0;
3253}
3254
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003255static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3256{
3257 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003258 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003259 int i;
3260
3261 rcu_read_lock();
3262 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003263 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003264 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003265 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003266
3267 if (!t)
3268 goto unlock;
3269
Johannes Weinerce00a962014-09-05 08:43:57 -04003270 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003271
3272 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07003273 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003274 * If it's not true, a threshold was crossed after last
3275 * call of __mem_cgroup_threshold().
3276 */
Phil Carmody5407a562010-05-26 14:42:42 -07003277 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003278
3279 /*
3280 * Iterate backward over array of thresholds starting from
3281 * current_threshold and check if a threshold is crossed.
3282 * If none of thresholds below usage is crossed, we read
3283 * only one element of the array here.
3284 */
3285 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3286 eventfd_signal(t->entries[i].eventfd, 1);
3287
3288 /* i = current_threshold + 1 */
3289 i++;
3290
3291 /*
3292 * Iterate forward over array of thresholds starting from
3293 * current_threshold+1 and check if a threshold is crossed.
3294 * If none of thresholds above usage is crossed, we read
3295 * only one element of the array here.
3296 */
3297 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3298 eventfd_signal(t->entries[i].eventfd, 1);
3299
3300 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07003301 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003302unlock:
3303 rcu_read_unlock();
3304}
3305
3306static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3307{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003308 while (memcg) {
3309 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08003310 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07003311 __mem_cgroup_threshold(memcg, true);
3312
3313 memcg = parent_mem_cgroup(memcg);
3314 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003315}
3316
3317static int compare_thresholds(const void *a, const void *b)
3318{
3319 const struct mem_cgroup_threshold *_a = a;
3320 const struct mem_cgroup_threshold *_b = b;
3321
Greg Thelen2bff24a2013-09-11 14:23:08 -07003322 if (_a->threshold > _b->threshold)
3323 return 1;
3324
3325 if (_a->threshold < _b->threshold)
3326 return -1;
3327
3328 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003329}
3330
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003331static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003332{
3333 struct mem_cgroup_eventfd_list *ev;
3334
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003335 spin_lock(&memcg_oom_lock);
3336
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003337 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003338 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07003339
3340 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003341 return 0;
3342}
3343
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003344static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003345{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003346 struct mem_cgroup *iter;
3347
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003348 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07003349 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003350}
3351
Tejun Heo59b6f872013-11-22 18:20:43 -05003352static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003353 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003354{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003355 struct mem_cgroup_thresholds *thresholds;
3356 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003357 unsigned long threshold;
3358 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003359 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003360
Johannes Weiner650c5e52015-02-11 15:26:03 -08003361 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003362 if (ret)
3363 return ret;
3364
3365 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003366
Johannes Weiner05b84302014-08-06 16:05:59 -07003367 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003368 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003369 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003370 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003371 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003372 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003373 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003374 BUG();
3375
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003376 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003377 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003378 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3379
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003380 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003381
3382 /* Allocate memory for new array of thresholds */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003383 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003384 GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003385 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003386 ret = -ENOMEM;
3387 goto unlock;
3388 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003389 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003390
3391 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003392 if (thresholds->primary) {
3393 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003394 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003395 }
3396
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003397 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003398 new->entries[size - 1].eventfd = eventfd;
3399 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003400
3401 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003402 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003403 compare_thresholds, NULL);
3404
3405 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003406 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003407 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07003408 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003409 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003410 * new->current_threshold will not be used until
3411 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003412 * it here.
3413 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003414 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07003415 } else
3416 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003417 }
3418
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003419 /* Free old spare buffer and save old primary buffer as spare */
3420 kfree(thresholds->spare);
3421 thresholds->spare = thresholds->primary;
3422
3423 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003424
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003425 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003426 synchronize_rcu();
3427
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003428unlock:
3429 mutex_unlock(&memcg->thresholds_lock);
3430
3431 return ret;
3432}
3433
Tejun Heo59b6f872013-11-22 18:20:43 -05003434static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003435 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003436{
Tejun Heo59b6f872013-11-22 18:20:43 -05003437 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003438}
3439
Tejun Heo59b6f872013-11-22 18:20:43 -05003440static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003441 struct eventfd_ctx *eventfd, const char *args)
3442{
Tejun Heo59b6f872013-11-22 18:20:43 -05003443 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003444}
3445
Tejun Heo59b6f872013-11-22 18:20:43 -05003446static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003447 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003448{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003449 struct mem_cgroup_thresholds *thresholds;
3450 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003451 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003452 int i, j, size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003453
3454 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07003455
3456 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003457 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003458 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07003459 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003460 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04003461 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07003462 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003463 BUG();
3464
Anton Vorontsov371528c2012-02-24 05:14:46 +04003465 if (!thresholds->primary)
3466 goto unlock;
3467
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003468 /* Check if a threshold crossed before removing */
3469 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3470
3471 /* Calculate new number of threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003472 size = 0;
3473 for (i = 0; i < thresholds->primary->size; i++) {
3474 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003475 size++;
3476 }
3477
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003478 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003479
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003480 /* Set thresholds array to NULL if we don't have thresholds */
3481 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003482 kfree(new);
3483 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003484 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003485 }
3486
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003487 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003488
3489 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003490 new->current_threshold = -1;
3491 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3492 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003493 continue;
3494
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003495 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07003496 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003497 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003498 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003499 * until rcu_assign_pointer(), so it's safe to increment
3500 * it here.
3501 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003502 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003503 }
3504 j++;
3505 }
3506
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003507swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003508 /* Swap primary and spare array */
3509 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07003510
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07003511 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003512
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07003513 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003514 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08003515
3516 /* If all events are unregistered, free the spare array */
3517 if (!new) {
3518 kfree(thresholds->spare);
3519 thresholds->spare = NULL;
3520 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04003521unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003522 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08003523}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003524
Tejun Heo59b6f872013-11-22 18:20:43 -05003525static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003526 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003527{
Tejun Heo59b6f872013-11-22 18:20:43 -05003528 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05003529}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003530
Tejun Heo59b6f872013-11-22 18:20:43 -05003531static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003532 struct eventfd_ctx *eventfd)
3533{
Tejun Heo59b6f872013-11-22 18:20:43 -05003534 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05003535}
3536
Tejun Heo59b6f872013-11-22 18:20:43 -05003537static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003538 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003539{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003540 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003541
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003542 event = kmalloc(sizeof(*event), GFP_KERNEL);
3543 if (!event)
3544 return -ENOMEM;
3545
Michal Hocko1af8efe2011-07-26 16:08:24 -07003546 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003547
3548 event->eventfd = eventfd;
3549 list_add(&event->list, &memcg->oom_notify);
3550
3551 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07003552 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003553 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07003554 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003555
3556 return 0;
3557}
3558
Tejun Heo59b6f872013-11-22 18:20:43 -05003559static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05003560 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003561{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003562 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003563
Michal Hocko1af8efe2011-07-26 16:08:24 -07003564 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003565
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003566 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003567 if (ev->eventfd == eventfd) {
3568 list_del(&ev->list);
3569 kfree(ev);
3570 }
3571 }
3572
Michal Hocko1af8efe2011-07-26 16:08:24 -07003573 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003574}
3575
Tejun Heo2da8ca82013-12-05 12:28:04 -05003576static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003577{
Tejun Heo2da8ca82013-12-05 12:28:04 -05003578 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003579
Tejun Heo791badb2013-12-05 12:28:02 -05003580 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07003581 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003582 return 0;
3583}
3584
Tejun Heo182446d2013-08-08 20:11:24 -04003585static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003586 struct cftype *cft, u64 val)
3587{
Tejun Heo182446d2013-08-08 20:11:24 -04003588 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003589
3590 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07003591 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003592 return -EINVAL;
3593
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003594 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07003595 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003596 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07003597
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003598 return 0;
3599}
3600
Tejun Heo52ebea72015-05-22 17:13:37 -04003601#ifdef CONFIG_CGROUP_WRITEBACK
3602
3603struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3604{
3605 return &memcg->cgwb_list;
3606}
3607
Tejun Heo841710a2015-05-22 18:23:33 -04003608static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3609{
3610 return wb_domain_init(&memcg->cgwb_domain, gfp);
3611}
3612
3613static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3614{
3615 wb_domain_exit(&memcg->cgwb_domain);
3616}
3617
Tejun Heo2529bb32015-05-22 18:23:34 -04003618static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3619{
3620 wb_domain_size_changed(&memcg->cgwb_domain);
3621}
3622
Tejun Heo841710a2015-05-22 18:23:33 -04003623struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3624{
3625 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3626
3627 if (!memcg->css.parent)
3628 return NULL;
3629
3630 return &memcg->cgwb_domain;
3631}
3632
Tejun Heoc2aa7232015-05-22 18:23:35 -04003633/**
3634 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3635 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003636 * @pfilepages: out parameter for number of file pages
3637 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04003638 * @pdirty: out parameter for number of dirty pages
3639 * @pwriteback: out parameter for number of pages under writeback
3640 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003641 * Determine the numbers of file, headroom, dirty, and writeback pages in
3642 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3643 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003644 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003645 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3646 * headroom is calculated as the lowest headroom of itself and the
3647 * ancestors. Note that this doesn't consider the actual amount of
3648 * available memory in the system. The caller should further cap
3649 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04003650 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003651void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3652 unsigned long *pheadroom, unsigned long *pdirty,
3653 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04003654{
3655 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3656 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003657
3658 *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3659
3660 /* this should eventually include NR_UNSTABLE_NFS */
3661 *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003662 *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3663 (1 << LRU_ACTIVE_FILE));
3664 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04003665
Tejun Heoc2aa7232015-05-22 18:23:35 -04003666 while ((parent = parent_mem_cgroup(memcg))) {
3667 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3668 unsigned long used = page_counter_read(&memcg->memory);
3669
Tejun Heoc5edf9c2015-09-29 13:04:26 -04003670 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04003671 memcg = parent;
3672 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04003673}
3674
Tejun Heo841710a2015-05-22 18:23:33 -04003675#else /* CONFIG_CGROUP_WRITEBACK */
3676
3677static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3678{
3679 return 0;
3680}
3681
3682static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3683{
3684}
3685
Tejun Heo2529bb32015-05-22 18:23:34 -04003686static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3687{
3688}
3689
Tejun Heo52ebea72015-05-22 17:13:37 -04003690#endif /* CONFIG_CGROUP_WRITEBACK */
3691
Tejun Heo79bd9812013-11-22 18:20:42 -05003692/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003693 * DO NOT USE IN NEW FILES.
3694 *
3695 * "cgroup.event_control" implementation.
3696 *
3697 * This is way over-engineered. It tries to support fully configurable
3698 * events for each user. Such level of flexibility is completely
3699 * unnecessary especially in the light of the planned unified hierarchy.
3700 *
3701 * Please deprecate this and replace with something simpler if at all
3702 * possible.
3703 */
3704
3705/*
Tejun Heo79bd9812013-11-22 18:20:42 -05003706 * Unregister event and free resources.
3707 *
3708 * Gets called from workqueue.
3709 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003710static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05003711{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003712 struct mem_cgroup_event *event =
3713 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05003714 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003715
3716 remove_wait_queue(event->wqh, &event->wait);
3717
Tejun Heo59b6f872013-11-22 18:20:43 -05003718 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05003719
3720 /* Notify userspace the event is going away. */
3721 eventfd_signal(event->eventfd, 1);
3722
3723 eventfd_ctx_put(event->eventfd);
3724 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05003725 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003726}
3727
3728/*
3729 * Gets called on POLLHUP on eventfd when user closes it.
3730 *
3731 * Called with wqh->lock held and interrupts disabled.
3732 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05003733static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3734 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05003735{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003736 struct mem_cgroup_event *event =
3737 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05003738 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003739 unsigned long flags = (unsigned long)key;
3740
3741 if (flags & POLLHUP) {
3742 /*
3743 * If the event has been detached at cgroup removal, we
3744 * can simply return knowing the other side will cleanup
3745 * for us.
3746 *
3747 * We can't race against event freeing since the other
3748 * side will require wqh->lock via remove_wait_queue(),
3749 * which we hold.
3750 */
Tejun Heofba94802013-11-22 18:20:43 -05003751 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003752 if (!list_empty(&event->list)) {
3753 list_del_init(&event->list);
3754 /*
3755 * We are in atomic context, but cgroup_event_remove()
3756 * may sleep, so we have to call it in workqueue.
3757 */
3758 schedule_work(&event->remove);
3759 }
Tejun Heofba94802013-11-22 18:20:43 -05003760 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003761 }
3762
3763 return 0;
3764}
3765
Tejun Heo3bc942f2013-11-22 18:20:44 -05003766static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05003767 wait_queue_head_t *wqh, poll_table *pt)
3768{
Tejun Heo3bc942f2013-11-22 18:20:44 -05003769 struct mem_cgroup_event *event =
3770 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05003771
3772 event->wqh = wqh;
3773 add_wait_queue(wqh, &event->wait);
3774}
3775
3776/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05003777 * DO NOT USE IN NEW FILES.
3778 *
Tejun Heo79bd9812013-11-22 18:20:42 -05003779 * Parse input and register new cgroup event handler.
3780 *
3781 * Input must be in format '<event_fd> <control_fd> <args>'.
3782 * Interpretation of args is defined by control file implementation.
3783 */
Tejun Heo451af502014-05-13 12:16:21 -04003784static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3785 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05003786{
Tejun Heo451af502014-05-13 12:16:21 -04003787 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05003788 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003789 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05003790 struct cgroup_subsys_state *cfile_css;
3791 unsigned int efd, cfd;
3792 struct fd efile;
3793 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05003794 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05003795 char *endp;
3796 int ret;
3797
Tejun Heo451af502014-05-13 12:16:21 -04003798 buf = strstrip(buf);
3799
3800 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003801 if (*endp != ' ')
3802 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003803 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003804
Tejun Heo451af502014-05-13 12:16:21 -04003805 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05003806 if ((*endp != ' ') && (*endp != '\0'))
3807 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04003808 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05003809
3810 event = kzalloc(sizeof(*event), GFP_KERNEL);
3811 if (!event)
3812 return -ENOMEM;
3813
Tejun Heo59b6f872013-11-22 18:20:43 -05003814 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05003815 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05003816 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3817 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3818 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05003819
3820 efile = fdget(efd);
3821 if (!efile.file) {
3822 ret = -EBADF;
3823 goto out_kfree;
3824 }
3825
3826 event->eventfd = eventfd_ctx_fileget(efile.file);
3827 if (IS_ERR(event->eventfd)) {
3828 ret = PTR_ERR(event->eventfd);
3829 goto out_put_efile;
3830 }
3831
3832 cfile = fdget(cfd);
3833 if (!cfile.file) {
3834 ret = -EBADF;
3835 goto out_put_eventfd;
3836 }
3837
3838 /* the process need read permission on control file */
3839 /* AV: shouldn't we check that it's been opened for read instead? */
3840 ret = inode_permission(file_inode(cfile.file), MAY_READ);
3841 if (ret < 0)
3842 goto out_put_cfile;
3843
Tejun Heo79bd9812013-11-22 18:20:42 -05003844 /*
Tejun Heofba94802013-11-22 18:20:43 -05003845 * Determine the event callbacks and set them in @event. This used
3846 * to be done via struct cftype but cgroup core no longer knows
3847 * about these events. The following is crude but the whole thing
3848 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05003849 *
3850 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05003851 */
Al Virob5830432014-10-31 01:22:04 -04003852 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05003853
3854 if (!strcmp(name, "memory.usage_in_bytes")) {
3855 event->register_event = mem_cgroup_usage_register_event;
3856 event->unregister_event = mem_cgroup_usage_unregister_event;
3857 } else if (!strcmp(name, "memory.oom_control")) {
3858 event->register_event = mem_cgroup_oom_register_event;
3859 event->unregister_event = mem_cgroup_oom_unregister_event;
3860 } else if (!strcmp(name, "memory.pressure_level")) {
3861 event->register_event = vmpressure_register_event;
3862 event->unregister_event = vmpressure_unregister_event;
3863 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05003864 event->register_event = memsw_cgroup_usage_register_event;
3865 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05003866 } else {
3867 ret = -EINVAL;
3868 goto out_put_cfile;
3869 }
3870
3871 /*
Tejun Heob5557c42013-11-22 18:20:42 -05003872 * Verify @cfile should belong to @css. Also, remaining events are
3873 * automatically removed on cgroup destruction but the removal is
3874 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05003875 */
Al Virob5830432014-10-31 01:22:04 -04003876 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04003877 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05003878 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05003879 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05003880 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05003881 if (cfile_css != css) {
3882 css_put(cfile_css);
3883 goto out_put_cfile;
3884 }
Tejun Heo79bd9812013-11-22 18:20:42 -05003885
Tejun Heo451af502014-05-13 12:16:21 -04003886 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05003887 if (ret)
3888 goto out_put_css;
3889
3890 efile.file->f_op->poll(efile.file, &event->pt);
3891
Tejun Heofba94802013-11-22 18:20:43 -05003892 spin_lock(&memcg->event_list_lock);
3893 list_add(&event->list, &memcg->event_list);
3894 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05003895
3896 fdput(cfile);
3897 fdput(efile);
3898
Tejun Heo451af502014-05-13 12:16:21 -04003899 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05003900
3901out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05003902 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05003903out_put_cfile:
3904 fdput(cfile);
3905out_put_eventfd:
3906 eventfd_ctx_put(event->eventfd);
3907out_put_efile:
3908 fdput(efile);
3909out_kfree:
3910 kfree(event);
3911
3912 return ret;
3913}
3914
Johannes Weiner241994e2015-02-11 15:26:06 -08003915static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003916 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003917 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003918 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003919 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003920 },
3921 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003922 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003923 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04003924 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003925 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003926 },
3927 {
Balbir Singh0eea1032008-02-07 00:13:57 -08003928 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003929 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003930 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003931 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003932 },
3933 {
Balbir Singh296c81d2009-09-23 15:56:36 -07003934 .name = "soft_limit_in_bytes",
3935 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003936 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003937 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07003938 },
3939 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003940 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003941 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04003942 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05003943 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003944 },
Balbir Singh8697d332008-02-07 00:13:59 -08003945 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003946 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003947 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003948 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003949 {
3950 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04003951 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003952 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08003953 {
3954 .name = "use_hierarchy",
3955 .write_u64 = mem_cgroup_hierarchy_write,
3956 .read_u64 = mem_cgroup_hierarchy_read,
3957 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003958 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05003959 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04003960 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04003961 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05003962 },
3963 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08003964 .name = "swappiness",
3965 .read_u64 = mem_cgroup_swappiness_read,
3966 .write_u64 = mem_cgroup_swappiness_write,
3967 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003968 {
3969 .name = "move_charge_at_immigrate",
3970 .read_u64 = mem_cgroup_move_charge_read,
3971 .write_u64 = mem_cgroup_move_charge_write,
3972 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003973 {
3974 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003975 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003976 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07003977 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3978 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003979 {
3980 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07003981 },
Ying Han406eb0c2011-05-26 16:25:37 -07003982#ifdef CONFIG_NUMA
3983 {
3984 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05003985 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07003986 },
3987#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08003988 {
3989 .name = "kmem.limit_in_bytes",
3990 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04003991 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05003992 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08003993 },
3994 {
3995 .name = "kmem.usage_in_bytes",
3996 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05003997 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08003998 },
3999 {
4000 .name = "kmem.failcnt",
4001 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004002 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004003 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004004 },
4005 {
4006 .name = "kmem.max_usage_in_bytes",
4007 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004008 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004009 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004010 },
Glauber Costa749c5412012-12-18 14:23:01 -08004011#ifdef CONFIG_SLABINFO
4012 {
4013 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004014 .seq_start = slab_start,
4015 .seq_next = slab_next,
4016 .seq_stop = slab_stop,
4017 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004018 },
4019#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004020 {
4021 .name = "kmem.tcp.limit_in_bytes",
4022 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4023 .write = mem_cgroup_write,
4024 .read_u64 = mem_cgroup_read_u64,
4025 },
4026 {
4027 .name = "kmem.tcp.usage_in_bytes",
4028 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4029 .read_u64 = mem_cgroup_read_u64,
4030 },
4031 {
4032 .name = "kmem.tcp.failcnt",
4033 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4034 .write = mem_cgroup_reset,
4035 .read_u64 = mem_cgroup_read_u64,
4036 },
4037 {
4038 .name = "kmem.tcp.max_usage_in_bytes",
4039 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4040 .write = mem_cgroup_reset,
4041 .read_u64 = mem_cgroup_read_u64,
4042 },
Tejun Heo6bc10342012-04-01 12:09:55 -07004043 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004044};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004045
Johannes Weiner73f576c2016-07-20 15:44:57 -07004046/*
4047 * Private memory cgroup IDR
4048 *
4049 * Swap-out records and page cache shadow entries need to store memcg
4050 * references in constrained space, so we maintain an ID space that is
4051 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4052 * memory-controlled cgroups to 64k.
4053 *
4054 * However, there usually are many references to the oflline CSS after
4055 * the cgroup has been destroyed, such as page cache or reclaimable
4056 * slab objects, that don't need to hang on to the ID. We want to keep
4057 * those dead CSS from occupying IDs, or we might quickly exhaust the
4058 * relatively small ID space and prevent the creation of new cgroups
4059 * even when there are much fewer than 64k cgroups - possibly none.
4060 *
4061 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4062 * be freed and recycled when it's no longer needed, which is usually
4063 * when the CSS is offlined.
4064 *
4065 * The only exception to that are records of swapped out tmpfs/shmem
4066 * pages that need to be attributed to live ancestors on swapin. But
4067 * those references are manageable from userspace.
4068 */
4069
4070static DEFINE_IDR(mem_cgroup_idr);
4071
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004072static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004073{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004074 VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0);
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004075 atomic_add(n, &memcg->id.ref);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004076}
4077
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004078static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07004079{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004080 VM_BUG_ON(atomic_read(&memcg->id.ref) < n);
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004081 if (atomic_sub_and_test(n, &memcg->id.ref)) {
Johannes Weiner73f576c2016-07-20 15:44:57 -07004082 idr_remove(&mem_cgroup_idr, memcg->id.id);
4083 memcg->id.id = 0;
4084
4085 /* Memcg ID pins CSS */
4086 css_put(&memcg->css);
4087 }
4088}
4089
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004090static inline void mem_cgroup_id_get(struct mem_cgroup *memcg)
4091{
4092 mem_cgroup_id_get_many(memcg, 1);
4093}
4094
4095static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
4096{
4097 mem_cgroup_id_put_many(memcg, 1);
4098}
4099
Johannes Weiner73f576c2016-07-20 15:44:57 -07004100/**
4101 * mem_cgroup_from_id - look up a memcg from a memcg id
4102 * @id: the memcg id to look up
4103 *
4104 * Caller must hold rcu_read_lock().
4105 */
4106struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
4107{
4108 WARN_ON_ONCE(!rcu_read_lock_held());
4109 return idr_find(&mem_cgroup_idr, id);
4110}
4111
Mel Gormanef8f2322016-07-28 15:46:05 -07004112static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004113{
4114 struct mem_cgroup_per_node *pn;
Mel Gormanef8f2322016-07-28 15:46:05 -07004115 int tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004116 /*
4117 * This routine is called against possible nodes.
4118 * But it's BUG to call kmalloc() against offline node.
4119 *
4120 * TODO: this routine can waste much memory for nodes which will
4121 * never be onlined. It's better to use memory hotplug callback
4122 * function.
4123 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07004124 if (!node_state(node, N_NORMAL_MEMORY))
4125 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08004126 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004127 if (!pn)
4128 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004129
Mel Gormanef8f2322016-07-28 15:46:05 -07004130 lruvec_init(&pn->lruvec);
4131 pn->usage_in_excess = 0;
4132 pn->on_tree = false;
4133 pn->memcg = memcg;
4134
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004135 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004136 return 0;
4137}
4138
Mel Gormanef8f2322016-07-28 15:46:05 -07004139static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004140{
Johannes Weiner54f72fe2013-07-08 15:59:49 -07004141 kfree(memcg->nodeinfo[node]);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08004142}
4143
Tahsin Erdogan16ace912017-03-09 16:17:26 -08004144static void __mem_cgroup_free(struct mem_cgroup *memcg)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004145{
4146 int node;
4147
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004148 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004149 free_mem_cgroup_per_node_info(memcg, node);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004150 free_percpu(memcg->stat);
4151 kfree(memcg);
4152}
4153
Tahsin Erdogan16ace912017-03-09 16:17:26 -08004154static void mem_cgroup_free(struct mem_cgroup *memcg)
4155{
4156 memcg_wb_domain_exit(memcg);
4157 __mem_cgroup_free(memcg);
4158}
4159
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004160static struct mem_cgroup *mem_cgroup_alloc(void)
4161{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004162 struct mem_cgroup *memcg;
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004163 size_t size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004164 int node;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004165
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004166 size = sizeof(struct mem_cgroup);
4167 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07004168
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08004169 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004170 if (!memcg)
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07004171 return NULL;
4172
Johannes Weiner73f576c2016-07-20 15:44:57 -07004173 memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
4174 1, MEM_CGROUP_ID_MAX,
4175 GFP_KERNEL);
4176 if (memcg->id.id < 0)
4177 goto fail;
4178
Hugh Dickinsd79154b2012-03-21 16:34:18 -07004179 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4180 if (!memcg->stat)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004181 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08004182
Bob Liu3ed28fa2012-01-12 17:19:04 -08004183 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07004184 if (alloc_mem_cgroup_per_node_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004185 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07004186
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004187 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4188 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08004189
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004190 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08004191 memcg->last_scanned_node = MAX_NUMNODES;
4192 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08004193 mutex_init(&memcg->thresholds_lock);
4194 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004195 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05004196 INIT_LIST_HEAD(&memcg->event_list);
4197 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004198 memcg->socket_pressure = jiffies;
Johannes Weiner127424c2016-01-20 15:02:32 -08004199#ifndef CONFIG_SLOB
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004200 memcg->kmemcg_id = -1;
Vladimir Davydov900a38f2014-12-12 16:55:10 -08004201#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04004202#ifdef CONFIG_CGROUP_WRITEBACK
4203 INIT_LIST_HEAD(&memcg->cgwb_list);
4204#endif
Johannes Weiner73f576c2016-07-20 15:44:57 -07004205 idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004206 return memcg;
4207fail:
Johannes Weiner73f576c2016-07-20 15:44:57 -07004208 if (memcg->id.id > 0)
4209 idr_remove(&mem_cgroup_idr, memcg->id.id);
Tahsin Erdogan16ace912017-03-09 16:17:26 -08004210 __mem_cgroup_free(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004211 return NULL;
Glauber Costad142e3e2013-02-22 16:34:52 -08004212}
4213
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004214static struct cgroup_subsys_state * __ref
4215mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08004216{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004217 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4218 struct mem_cgroup *memcg;
4219 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08004220
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004221 memcg = mem_cgroup_alloc();
4222 if (!memcg)
4223 return ERR_PTR(error);
Li Zefan4219b2d2013-09-23 16:56:29 +08004224
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004225 memcg->high = PAGE_COUNTER_MAX;
4226 memcg->soft_limit = PAGE_COUNTER_MAX;
4227 if (parent) {
4228 memcg->swappiness = mem_cgroup_swappiness(parent);
4229 memcg->oom_kill_disable = parent->oom_kill_disable;
4230 }
4231 if (parent && parent->use_hierarchy) {
4232 memcg->use_hierarchy = true;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004233 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004234 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004235 page_counter_init(&memcg->memsw, &parent->memsw);
4236 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08004237 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08004238 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004239 page_counter_init(&memcg->memory, NULL);
Vladimir Davydov37e84352016-01-20 15:02:56 -08004240 page_counter_init(&memcg->swap, NULL);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004241 page_counter_init(&memcg->memsw, NULL);
4242 page_counter_init(&memcg->kmem, NULL);
Johannes Weiner0db15292016-01-20 15:02:50 -08004243 page_counter_init(&memcg->tcpmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07004244 /*
4245 * Deeper hierachy with use_hierarchy == false doesn't make
4246 * much sense so let cgroup subsystem know about this
4247 * unfortunate state in our controller.
4248 */
Glauber Costad142e3e2013-02-22 16:34:52 -08004249 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05004250 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08004251 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08004252
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004253 /* The following stuff does not apply to the root */
4254 if (!parent) {
4255 root_mem_cgroup = memcg;
4256 return &memcg->css;
4257 }
4258
Vladimir Davydovb313aee2016-03-17 14:18:27 -07004259 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004260 if (error)
4261 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08004262
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004263 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004264 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004265
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004266 return &memcg->css;
4267fail:
4268 mem_cgroup_free(memcg);
Tejun Heoea3a9642016-06-24 14:49:58 -07004269 return ERR_PTR(-ENOMEM);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004270}
4271
Johannes Weiner73f576c2016-07-20 15:44:57 -07004272static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004273{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004274 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4275
Johannes Weiner73f576c2016-07-20 15:44:57 -07004276 /* Online state pins memcg ID, memcg ID pins CSS */
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07004277 atomic_set(&memcg->id.ref, 1);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004278 css_get(css);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07004279 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004280}
4281
Tejun Heoeb954192013-08-08 20:11:23 -04004282static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004283{
Tejun Heoeb954192013-08-08 20:11:23 -04004284 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004285 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05004286
4287 /*
4288 * Unregister events and notify userspace.
4289 * Notify userspace about cgroup removing only after rmdir of cgroup
4290 * directory to avoid race between userspace and kernelspace.
4291 */
Tejun Heofba94802013-11-22 18:20:43 -05004292 spin_lock(&memcg->event_list_lock);
4293 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004294 list_del_init(&event->list);
4295 schedule_work(&event->remove);
4296 }
Tejun Heofba94802013-11-22 18:20:43 -05004297 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07004298
Johannes Weiner567e9ab2016-01-20 15:02:24 -08004299 memcg_offline_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04004300 wb_memcg_offline(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07004301
4302 mem_cgroup_id_put(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08004303}
4304
Vladimir Davydov6df38682015-12-29 14:54:10 -08004305static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4306{
4307 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4308
4309 invalidate_reclaim_iterators(memcg);
4310}
4311
Tejun Heoeb954192013-08-08 20:11:23 -04004312static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004313{
Tejun Heoeb954192013-08-08 20:11:23 -04004314 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimurac268e992009-01-15 13:51:13 -08004315
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08004316 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08004317 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004318
Johannes Weiner0db15292016-01-20 15:02:50 -08004319 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004320 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08004321
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004322 vmpressure_cleanup(&memcg->vmpressure);
4323 cancel_work_sync(&memcg->high_work);
4324 mem_cgroup_remove_from_trees(memcg);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08004325 memcg_free_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08004326 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004327}
4328
Tejun Heo1ced9532014-07-08 18:02:57 -04004329/**
4330 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4331 * @css: the target css
4332 *
4333 * Reset the states of the mem_cgroup associated with @css. This is
4334 * invoked when the userland requests disabling on the default hierarchy
4335 * but the memcg is pinned through dependency. The memcg should stop
4336 * applying policies and should revert to the vanilla state as it may be
4337 * made visible again.
4338 *
4339 * The current implementation only resets the essential configurations.
4340 * This needs to be expanded to cover all the visible parts.
4341 */
4342static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4343{
4344 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4345
Vladimir Davydovd334c9b2016-03-17 14:19:38 -07004346 page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4347 page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4348 page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4349 page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4350 page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
Johannes Weiner241994e2015-02-11 15:26:06 -08004351 memcg->low = 0;
4352 memcg->high = PAGE_COUNTER_MAX;
Johannes Weiner24d404d2015-01-08 14:32:35 -08004353 memcg->soft_limit = PAGE_COUNTER_MAX;
Tejun Heo2529bb32015-05-22 18:23:34 -04004354 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04004355}
4356
Daisuke Nishimura02491442010-03-10 15:22:17 -08004357#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004358/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004359static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004360{
Johannes Weiner05b84302014-08-06 16:05:59 -07004361 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07004362
Mel Gormand0164ad2015-11-06 16:28:21 -08004363 /* Try a single bulk charge without reclaim first, kswapd may wake */
4364 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07004365 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004366 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004367 return ret;
4368 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004369
David Rientjesf5f415c2017-01-24 15:18:10 -08004370 /* Try charges one by one with reclaim, but do not retry */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004371 while (count--) {
David Rientjesf5f415c2017-01-24 15:18:10 -08004372 ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004373 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08004374 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004375 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07004376 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004377 }
Johannes Weiner9476db92014-08-06 16:05:55 -07004378 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004379}
4380
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004381union mc_target {
4382 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004383 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004384};
4385
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004386enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004387 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004388 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004389 MC_TARGET_SWAP,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004390};
4391
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004392static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4393 unsigned long addr, pte_t ptent)
4394{
4395 struct page *page = vm_normal_page(vma, addr, ptent);
4396
4397 if (!page || !page_mapped(page))
4398 return NULL;
4399 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004400 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004401 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004402 } else {
4403 if (!(mc.flags & MOVE_FILE))
4404 return NULL;
4405 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004406 if (!get_page_unless_zero(page))
4407 return NULL;
4408
4409 return page;
4410}
4411
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004412#ifdef CONFIG_SWAP
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004413static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004414 pte_t ptent, swp_entry_t *entry)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004415{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004416 struct page *page = NULL;
4417 swp_entry_t ent = pte_to_swp_entry(ptent);
4418
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004419 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004420 return NULL;
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004421 /*
4422 * Because lookup_swap_cache() updates some statistics counter,
4423 * we call find_get_page() with swapper_space directly.
4424 */
Huang Yingf6ab1f72016-10-07 17:00:21 -07004425 page = find_get_page(swap_address_space(ent), swp_offset(ent));
Johannes Weiner7941d212016-01-14 15:21:23 -08004426 if (do_memsw_account())
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004427 entry->val = ent.val;
4428
4429 return page;
4430}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004431#else
4432static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07004433 pte_t ptent, swp_entry_t *entry)
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07004434{
4435 return NULL;
4436}
4437#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004438
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004439static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4440 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4441{
4442 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004443 struct address_space *mapping;
4444 pgoff_t pgoff;
4445
4446 if (!vma->vm_file) /* anonymous vma */
4447 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004448 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004449 return NULL;
4450
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004451 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004452 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004453
4454 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004455#ifdef CONFIG_SWAP
4456 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07004457 if (shmem_mapping(mapping)) {
4458 page = find_get_entry(mapping, pgoff);
4459 if (radix_tree_exceptional_entry(page)) {
4460 swp_entry_t swp = radix_to_swp_entry(page);
Johannes Weiner7941d212016-01-14 15:21:23 -08004461 if (do_memsw_account())
Johannes Weiner139b6a62014-05-06 12:50:05 -07004462 *entry = swp;
Huang Yingf6ab1f72016-10-07 17:00:21 -07004463 page = find_get_page(swap_address_space(swp),
4464 swp_offset(swp));
Johannes Weiner139b6a62014-05-06 12:50:05 -07004465 }
4466 } else
4467 page = find_get_page(mapping, pgoff);
4468#else
4469 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07004470#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004471 return page;
4472}
4473
Chen Gangb1b0dea2015-04-14 15:47:35 -07004474/**
4475 * mem_cgroup_move_account - move account of the page
4476 * @page: the page
Li RongQing25843c22016-07-26 15:26:56 -07004477 * @compound: charge the page as compound or small page
Chen Gangb1b0dea2015-04-14 15:47:35 -07004478 * @from: mem_cgroup which the page is moved from.
4479 * @to: mem_cgroup which the page is moved to. @from != @to.
4480 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08004481 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07004482 *
4483 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4484 * from old cgroup.
4485 */
4486static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004487 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07004488 struct mem_cgroup *from,
4489 struct mem_cgroup *to)
4490{
4491 unsigned long flags;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004492 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004493 int ret;
Greg Thelenc4843a72015-05-22 17:13:16 -04004494 bool anon;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004495
4496 VM_BUG_ON(from == to);
4497 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004498 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07004499
4500 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07004501 * Prevent mem_cgroup_migrate() from looking at
Hugh Dickins45637ba2015-11-05 18:49:40 -08004502 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07004503 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004504 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07004505 if (!trylock_page(page))
4506 goto out;
4507
4508 ret = -EINVAL;
4509 if (page->mem_cgroup != from)
4510 goto out_unlock;
4511
Greg Thelenc4843a72015-05-22 17:13:16 -04004512 anon = PageAnon(page);
4513
Chen Gangb1b0dea2015-04-14 15:47:35 -07004514 spin_lock_irqsave(&from->move_lock, flags);
4515
Greg Thelenc4843a72015-05-22 17:13:16 -04004516 if (!anon && page_mapped(page)) {
Chen Gangb1b0dea2015-04-14 15:47:35 -07004517 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4518 nr_pages);
4519 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4520 nr_pages);
4521 }
4522
Greg Thelenc4843a72015-05-22 17:13:16 -04004523 /*
4524 * move_lock grabbed above and caller set from->moving_account, so
4525 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4526 * So mapping should be stable for dirty pages.
4527 */
4528 if (!anon && PageDirty(page)) {
4529 struct address_space *mapping = page_mapping(page);
4530
4531 if (mapping_cap_account_dirty(mapping)) {
4532 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4533 nr_pages);
4534 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4535 nr_pages);
4536 }
4537 }
4538
Chen Gangb1b0dea2015-04-14 15:47:35 -07004539 if (PageWriteback(page)) {
4540 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4541 nr_pages);
4542 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4543 nr_pages);
4544 }
4545
4546 /*
4547 * It is safe to change page->mem_cgroup here because the page
4548 * is referenced, charged, and isolated - we can't race with
4549 * uncharging, charging, migration, or LRU putback.
4550 */
4551
4552 /* caller should have done css_get */
4553 page->mem_cgroup = to;
4554 spin_unlock_irqrestore(&from->move_lock, flags);
4555
4556 ret = 0;
4557
4558 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004559 mem_cgroup_charge_statistics(to, page, compound, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004560 memcg_check_events(to, page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004561 mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07004562 memcg_check_events(from, page);
4563 local_irq_enable();
4564out_unlock:
4565 unlock_page(page);
4566out:
4567 return ret;
4568}
4569
Li RongQing7cf78062016-05-27 14:27:46 -07004570/**
4571 * get_mctgt_type - get target type of moving charge
4572 * @vma: the vma the pte to be checked belongs
4573 * @addr: the address corresponding to the pte to be checked
4574 * @ptent: the pte to be checked
4575 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4576 *
4577 * Returns
4578 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4579 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4580 * move charge. if @target is not NULL, the page is stored in target->page
4581 * with extra refcnt got(Callers should handle it).
4582 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4583 * target for charge migration. if @target is not NULL, the entry is stored
4584 * in target->ent.
4585 *
4586 * Called with pte lock held.
4587 */
4588
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004589static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004590 unsigned long addr, pte_t ptent, union mc_target *target)
4591{
Daisuke Nishimura02491442010-03-10 15:22:17 -08004592 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004593 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004594 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004595
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004596 if (pte_present(ptent))
4597 page = mc_handle_present_pte(vma, addr, ptent);
4598 else if (is_swap_pte(ptent))
Li RongQing48406ef2016-07-26 15:22:14 -07004599 page = mc_handle_swap_pte(vma, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08004600 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07004601 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004602
4603 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004604 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004605 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004606 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004607 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08004608 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07004609 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08004610 */
Johannes Weiner1306a852014-12-10 15:44:52 -08004611 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004612 ret = MC_TARGET_PAGE;
4613 if (target)
4614 target->page = page;
4615 }
4616 if (!ret || !target)
4617 put_page(page);
4618 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07004619 /* There is a swap entry and a page doesn't exist or isn't charged */
4620 if (ent.val && !ret &&
Li Zefan34c00c32013-09-23 16:56:01 +08004621 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07004622 ret = MC_TARGET_SWAP;
4623 if (target)
4624 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004625 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004626 return ret;
4627}
4628
Naoya Horiguchi12724852012-03-21 16:34:28 -07004629#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4630/*
4631 * We don't consider swapping or file mapped pages because THP does not
4632 * support them for now.
4633 * Caller should make sure that pmd_trans_huge(pmd) is true.
4634 */
4635static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4636 unsigned long addr, pmd_t pmd, union mc_target *target)
4637{
4638 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004639 enum mc_target_type ret = MC_TARGET_NONE;
4640
4641 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08004642 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004643 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07004644 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08004645 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004646 ret = MC_TARGET_PAGE;
4647 if (target) {
4648 get_page(page);
4649 target->page = page;
4650 }
4651 }
4652 return ret;
4653}
4654#else
4655static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4656 unsigned long addr, pmd_t pmd, union mc_target *target)
4657{
4658 return MC_TARGET_NONE;
4659}
4660#endif
4661
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004662static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4663 unsigned long addr, unsigned long end,
4664 struct mm_walk *walk)
4665{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004666 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004667 pte_t *pte;
4668 spinlock_t *ptl;
4669
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004670 ptl = pmd_trans_huge_lock(pmd, vma);
4671 if (ptl) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004672 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4673 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004674 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004675 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004676 }
Dave Hansen03319322011-03-22 16:32:56 -07004677
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004678 if (pmd_trans_unstable(pmd))
4679 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004680 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4681 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004682 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004683 mc.precharge++; /* increment precharge temporarily */
4684 pte_unmap_unlock(pte - 1, ptl);
4685 cond_resched();
4686
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004687 return 0;
4688}
4689
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004690static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4691{
4692 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004693
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004694 struct mm_walk mem_cgroup_count_precharge_walk = {
4695 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4696 .mm = mm,
4697 };
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004698 down_read(&mm->mmap_sem);
James Morse0247f3f2016-10-07 17:00:12 -07004699 walk_page_range(0, mm->highest_vm_end,
4700 &mem_cgroup_count_precharge_walk);
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004701 up_read(&mm->mmap_sem);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004702
4703 precharge = mc.precharge;
4704 mc.precharge = 0;
4705
4706 return precharge;
4707}
4708
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004709static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4710{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004711 unsigned long precharge = mem_cgroup_count_precharge(mm);
4712
4713 VM_BUG_ON(mc.moving_task);
4714 mc.moving_task = current;
4715 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004716}
4717
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004718/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4719static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004720{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004721 struct mem_cgroup *from = mc.from;
4722 struct mem_cgroup *to = mc.to;
4723
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004724 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004725 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004726 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004727 mc.precharge = 0;
4728 }
4729 /*
4730 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4731 * we must uncharge here.
4732 */
4733 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07004734 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004735 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004736 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004737 /* we must fixup refcnts and charges */
4738 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004739 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04004740 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004741 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004742
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004743 mem_cgroup_id_put_many(mc.from, mc.moved_swap);
4744
Johannes Weiner05b84302014-08-06 16:05:59 -07004745 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004746 * we charged both to->memory and to->memsw, so we
4747 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07004748 */
Johannes Weinerce00a962014-09-05 08:43:57 -04004749 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004750 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004751
Vladimir Davydov615d66c2016-08-11 15:33:03 -07004752 mem_cgroup_id_get_many(mc.to, mc.moved_swap);
4753 css_put_many(&mc.to->css, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004754
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004755 mc.moved_swap = 0;
4756 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004757 memcg_oom_recover(from);
4758 memcg_oom_recover(to);
4759 wake_up_all(&mc.waitq);
4760}
4761
4762static void mem_cgroup_clear_mc(void)
4763{
Tejun Heo264a0ae2016-04-21 19:09:02 -04004764 struct mm_struct *mm = mc.mm;
4765
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004766 /*
4767 * we must clear moving_task before waking up waiters at the end of
4768 * task migration.
4769 */
4770 mc.moving_task = NULL;
4771 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004772 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004773 mc.from = NULL;
4774 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04004775 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07004776 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004777
4778 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004779}
4780
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004781static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004782{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004783 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07004784 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07004785 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04004786 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004787 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004788 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004789 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004790
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004791 /* charge immigration isn't supported on the default hierarchy */
4792 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07004793 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004794
Tejun Heo4530edd2015-09-11 15:00:19 -04004795 /*
4796 * Multi-process migrations only happen on the default hierarchy
4797 * where charge immigration is not used. Perform charge
4798 * immigration if @tset contains a leader and whine if there are
4799 * multiple.
4800 */
4801 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004802 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04004803 WARN_ON_ONCE(p);
4804 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004805 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04004806 }
4807 if (!p)
4808 return 0;
4809
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004810 /*
4811 * We are now commited to this value whatever it is. Changes in this
4812 * tunable will only affect upcoming migrations, not the current one.
4813 * So we need to save it, and keep it going.
4814 */
4815 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4816 if (!move_flags)
4817 return 0;
4818
Tejun Heo9f2115f2015-09-08 15:01:10 -07004819 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004820
Tejun Heo9f2115f2015-09-08 15:01:10 -07004821 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08004822
Tejun Heo9f2115f2015-09-08 15:01:10 -07004823 mm = get_task_mm(p);
4824 if (!mm)
4825 return 0;
4826 /* We move charges only when we move a owner of the mm */
4827 if (mm->owner == p) {
4828 VM_BUG_ON(mc.from);
4829 VM_BUG_ON(mc.to);
4830 VM_BUG_ON(mc.precharge);
4831 VM_BUG_ON(mc.moved_charge);
4832 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004833
Tejun Heo9f2115f2015-09-08 15:01:10 -07004834 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04004835 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07004836 mc.from = from;
4837 mc.to = memcg;
4838 mc.flags = move_flags;
4839 spin_unlock(&mc.lock);
4840 /* We set mc.moving_task later */
4841
4842 ret = mem_cgroup_precharge_mc(mm);
4843 if (ret)
4844 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04004845 } else {
4846 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004847 }
4848 return ret;
4849}
4850
Amit Pundir3ba5a3b2015-12-21 16:01:10 +05304851static int mem_cgroup_allow_attach(struct cgroup_taskset *tset)
Rom Lemarchand5c85c742014-11-07 09:42:40 -08004852{
Amit Pundir3ba5a3b2015-12-21 16:01:10 +05304853 return subsys_cgroup_allow_attach(tset);
Rom Lemarchand5c85c742014-11-07 09:42:40 -08004854}
4855
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05004856static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004857{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08004858 if (mc.to)
4859 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004860}
4861
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004862static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4863 unsigned long addr, unsigned long end,
4864 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004865{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004866 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004867 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004868 pte_t *pte;
4869 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004870 enum mc_target_type target_type;
4871 union mc_target target;
4872 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004873
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08004874 ptl = pmd_trans_huge_lock(pmd, vma);
4875 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07004876 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004877 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07004878 return 0;
4879 }
4880 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4881 if (target_type == MC_TARGET_PAGE) {
4882 page = target.page;
4883 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004884 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08004885 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07004886 mc.precharge -= HPAGE_PMD_NR;
4887 mc.moved_charge += HPAGE_PMD_NR;
4888 }
4889 putback_lru_page(page);
4890 }
4891 put_page(page);
4892 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08004893 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07004894 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07004895 }
4896
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07004897 if (pmd_trans_unstable(pmd))
4898 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004899retry:
4900 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4901 for (; addr != end; addr += PAGE_SIZE) {
4902 pte_t ptent = *(pte++);
Daisuke Nishimura02491442010-03-10 15:22:17 -08004903 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004904
4905 if (!mc.precharge)
4906 break;
4907
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004908 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004909 case MC_TARGET_PAGE:
4910 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08004911 /*
4912 * We can have a part of the split pmd here. Moving it
4913 * can be done but it would be too convoluted so simply
4914 * ignore such a partial THP and keep it in original
4915 * memcg. There should be somebody mapping the head.
4916 */
4917 if (PageTransCompound(page))
4918 goto put;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004919 if (isolate_lru_page(page))
4920 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08004921 if (!mem_cgroup_move_account(page, false,
4922 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004923 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004924 /* we uncharge from mc.from later. */
4925 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004926 }
4927 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07004928put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004929 put_page(page);
4930 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08004931 case MC_TARGET_SWAP:
4932 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07004933 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08004934 mc.precharge--;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08004935 /* we fixup refcnts and charges later. */
4936 mc.moved_swap++;
4937 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08004938 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004939 default:
4940 break;
4941 }
4942 }
4943 pte_unmap_unlock(pte - 1, ptl);
4944 cond_resched();
4945
4946 if (addr != end) {
4947 /*
4948 * We have consumed all precharges we got in can_attach().
4949 * We try charge one by one, but don't do any additional
4950 * charges to mc.to if we have failed in charge once in attach()
4951 * phase.
4952 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08004953 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004954 if (!ret)
4955 goto retry;
4956 }
4957
4958 return ret;
4959}
4960
Tejun Heo264a0ae2016-04-21 19:09:02 -04004961static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004962{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004963 struct mm_walk mem_cgroup_move_charge_walk = {
4964 .pmd_entry = mem_cgroup_move_charge_pte_range,
Tejun Heo264a0ae2016-04-21 19:09:02 -04004965 .mm = mc.mm,
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004966 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08004967
4968 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08004969 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07004970 * Signal lock_page_memcg() to take the memcg's move_lock
4971 * while we're moving its pages to another memcg. Then wait
4972 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08004973 */
4974 atomic_inc(&mc.from->moving_account);
4975 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004976retry:
Tejun Heo264a0ae2016-04-21 19:09:02 -04004977 if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08004978 /*
4979 * Someone who are holding the mmap_sem might be waiting in
4980 * waitq. So we cancel all extra charges, wake up all waiters,
4981 * and retry. Because we cancel precharges, we might not be able
4982 * to move enough charges, but moving charge is a best-effort
4983 * feature anyway, so it wouldn't be a big problem.
4984 */
4985 __mem_cgroup_clear_mc();
4986 cond_resched();
4987 goto retry;
4988 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08004989 /*
4990 * When we have consumed all precharges and failed in doing
4991 * additional charge, the page walk just aborts.
4992 */
James Morse0247f3f2016-10-07 17:00:12 -07004993 walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk);
4994
Tejun Heo264a0ae2016-04-21 19:09:02 -04004995 up_read(&mc.mm->mmap_sem);
Johannes Weiner312722c2014-12-10 15:44:25 -08004996 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004997}
4998
Tejun Heo264a0ae2016-04-21 19:09:02 -04004999static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08005000{
Tejun Heo264a0ae2016-04-21 19:09:02 -04005001 if (mc.to) {
5002 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07005003 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04005004 }
Balbir Singh67e465a2008-02-07 00:13:54 -08005005}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005006#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005007static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005008{
5009 return 0;
5010}
Amit Pundir3ba5a3b2015-12-21 16:01:10 +05305011static int mem_cgroup_allow_attach(struct cgroup_taskset *tset)
Rom Lemarchand5c85c742014-11-07 09:42:40 -08005012{
5013 return 0;
5014}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005015static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005016{
5017}
Tejun Heo264a0ae2016-04-21 19:09:02 -04005018static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07005019{
5020}
5021#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08005022
Tejun Heof00baae2013-04-15 13:41:15 -07005023/*
5024 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04005025 * to verify whether we're attached to the default hierarchy on each mount
5026 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07005027 */
Tejun Heoeb954192013-08-08 20:11:23 -04005028static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07005029{
5030 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04005031 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07005032 * guarantees that @root doesn't have any children, so turning it
5033 * on for the root memcg is enough.
5034 */
Tejun Heo9e10a132015-09-18 11:56:28 -04005035 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee592015-03-12 16:26:19 -07005036 root_mem_cgroup->use_hierarchy = true;
5037 else
5038 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07005039}
5040
Johannes Weiner241994e2015-02-11 15:26:06 -08005041static u64 memory_current_read(struct cgroup_subsys_state *css,
5042 struct cftype *cft)
5043{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005044 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5045
5046 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994e2015-02-11 15:26:06 -08005047}
5048
5049static int memory_low_show(struct seq_file *m, void *v)
5050{
5051 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005052 unsigned long low = READ_ONCE(memcg->low);
Johannes Weiner241994e2015-02-11 15:26:06 -08005053
5054 if (low == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005055 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005056 else
5057 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5058
5059 return 0;
5060}
5061
5062static ssize_t memory_low_write(struct kernfs_open_file *of,
5063 char *buf, size_t nbytes, loff_t off)
5064{
5065 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5066 unsigned long low;
5067 int err;
5068
5069 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005070 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994e2015-02-11 15:26:06 -08005071 if (err)
5072 return err;
5073
5074 memcg->low = low;
5075
5076 return nbytes;
5077}
5078
5079static int memory_high_show(struct seq_file *m, void *v)
5080{
5081 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005082 unsigned long high = READ_ONCE(memcg->high);
Johannes Weiner241994e2015-02-11 15:26:06 -08005083
5084 if (high == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005085 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005086 else
5087 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5088
5089 return 0;
5090}
5091
5092static ssize_t memory_high_write(struct kernfs_open_file *of,
5093 char *buf, size_t nbytes, loff_t off)
5094{
5095 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner588083b2016-03-17 14:20:25 -07005096 unsigned long nr_pages;
Johannes Weiner241994e2015-02-11 15:26:06 -08005097 unsigned long high;
5098 int err;
5099
5100 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005101 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994e2015-02-11 15:26:06 -08005102 if (err)
5103 return err;
5104
5105 memcg->high = high;
5106
Johannes Weiner588083b2016-03-17 14:20:25 -07005107 nr_pages = page_counter_read(&memcg->memory);
5108 if (nr_pages > high)
5109 try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5110 GFP_KERNEL, true);
5111
Tejun Heo2529bb32015-05-22 18:23:34 -04005112 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005113 return nbytes;
5114}
5115
5116static int memory_max_show(struct seq_file *m, void *v)
5117{
5118 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Jason Low4db0c3c2015-04-15 16:14:08 -07005119 unsigned long max = READ_ONCE(memcg->memory.limit);
Johannes Weiner241994e2015-02-11 15:26:06 -08005120
5121 if (max == PAGE_COUNTER_MAX)
Johannes Weinerd2973692015-02-27 15:52:04 -08005122 seq_puts(m, "max\n");
Johannes Weiner241994e2015-02-11 15:26:06 -08005123 else
5124 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5125
5126 return 0;
5127}
5128
5129static ssize_t memory_max_write(struct kernfs_open_file *of,
5130 char *buf, size_t nbytes, loff_t off)
5131{
5132 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005133 unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5134 bool drained = false;
Johannes Weiner241994e2015-02-11 15:26:06 -08005135 unsigned long max;
5136 int err;
5137
5138 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08005139 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994e2015-02-11 15:26:06 -08005140 if (err)
5141 return err;
5142
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07005143 xchg(&memcg->memory.limit, max);
5144
5145 for (;;) {
5146 unsigned long nr_pages = page_counter_read(&memcg->memory);
5147
5148 if (nr_pages <= max)
5149 break;
5150
5151 if (signal_pending(current)) {
5152 err = -EINTR;
5153 break;
5154 }
5155
5156 if (!drained) {
5157 drain_all_stock(memcg);
5158 drained = true;
5159 continue;
5160 }
5161
5162 if (nr_reclaims) {
5163 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5164 GFP_KERNEL, true))
5165 nr_reclaims--;
5166 continue;
5167 }
5168
5169 mem_cgroup_events(memcg, MEMCG_OOM, 1);
5170 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5171 break;
5172 }
Johannes Weiner241994e2015-02-11 15:26:06 -08005173
Tejun Heo2529bb32015-05-22 18:23:34 -04005174 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994e2015-02-11 15:26:06 -08005175 return nbytes;
5176}
5177
5178static int memory_events_show(struct seq_file *m, void *v)
5179{
5180 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5181
5182 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5183 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5184 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5185 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5186
5187 return 0;
5188}
5189
Johannes Weiner587d9f72016-01-20 15:03:19 -08005190static int memory_stat_show(struct seq_file *m, void *v)
5191{
5192 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005193 unsigned long stat[MEMCG_NR_STAT];
5194 unsigned long events[MEMCG_NR_EVENTS];
Johannes Weiner587d9f72016-01-20 15:03:19 -08005195 int i;
5196
5197 /*
5198 * Provide statistics on the state of the memory subsystem as
5199 * well as cumulative event counters that show past behavior.
5200 *
5201 * This list is ordered following a combination of these gradients:
5202 * 1) generic big picture -> specifics and details
5203 * 2) reflecting userspace activity -> reflecting kernel heuristics
5204 *
5205 * Current memory state:
5206 */
5207
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005208 tree_stat(memcg, stat);
5209 tree_events(memcg, events);
5210
Johannes Weiner587d9f72016-01-20 15:03:19 -08005211 seq_printf(m, "anon %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005212 (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005213 seq_printf(m, "file %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005214 (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE);
Vladimir Davydov12580e42016-03-17 14:17:38 -07005215 seq_printf(m, "kernel_stack %llu\n",
Andy Lutomirskiefdc9492016-07-28 15:48:17 -07005216 (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005217 seq_printf(m, "slab %llu\n",
5218 (u64)(stat[MEMCG_SLAB_RECLAIMABLE] +
5219 stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
Johannes Weinerb2807f02016-01-20 15:03:22 -08005220 seq_printf(m, "sock %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005221 (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005222
5223 seq_printf(m, "file_mapped %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005224 (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005225 seq_printf(m, "file_dirty %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005226 (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005227 seq_printf(m, "file_writeback %llu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005228 (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005229
5230 for (i = 0; i < NR_LRU_LISTS; i++) {
5231 struct mem_cgroup *mi;
5232 unsigned long val = 0;
5233
5234 for_each_mem_cgroup_tree(mi, memcg)
5235 val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5236 seq_printf(m, "%s %llu\n",
5237 mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5238 }
5239
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07005240 seq_printf(m, "slab_reclaimable %llu\n",
5241 (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE);
5242 seq_printf(m, "slab_unreclaimable %llu\n",
5243 (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5244
Johannes Weiner587d9f72016-01-20 15:03:19 -08005245 /* Accumulated memory events */
5246
5247 seq_printf(m, "pgfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005248 events[MEM_CGROUP_EVENTS_PGFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005249 seq_printf(m, "pgmajfault %lu\n",
Vladimir Davydov72b54e72016-03-17 14:17:32 -07005250 events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
Johannes Weiner587d9f72016-01-20 15:03:19 -08005251
5252 return 0;
5253}
5254
Johannes Weiner241994e2015-02-11 15:26:06 -08005255static struct cftype memory_files[] = {
5256 {
5257 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08005258 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994e2015-02-11 15:26:06 -08005259 .read_u64 = memory_current_read,
5260 },
5261 {
5262 .name = "low",
5263 .flags = CFTYPE_NOT_ON_ROOT,
5264 .seq_show = memory_low_show,
5265 .write = memory_low_write,
5266 },
5267 {
5268 .name = "high",
5269 .flags = CFTYPE_NOT_ON_ROOT,
5270 .seq_show = memory_high_show,
5271 .write = memory_high_write,
5272 },
5273 {
5274 .name = "max",
5275 .flags = CFTYPE_NOT_ON_ROOT,
5276 .seq_show = memory_max_show,
5277 .write = memory_max_write,
5278 },
5279 {
5280 .name = "events",
5281 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04005282 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994e2015-02-11 15:26:06 -08005283 .seq_show = memory_events_show,
5284 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08005285 {
5286 .name = "stat",
5287 .flags = CFTYPE_NOT_ON_ROOT,
5288 .seq_show = memory_stat_show,
5289 },
Johannes Weiner241994e2015-02-11 15:26:06 -08005290 { } /* terminate */
5291};
5292
Tejun Heo073219e2014-02-08 10:36:58 -05005293struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08005294 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08005295 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08005296 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08005297 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08005298 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04005299 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005300 .can_attach = mem_cgroup_can_attach,
5301 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04005302 .post_attach = mem_cgroup_move_task,
Rom Lemarchand5c85c742014-11-07 09:42:40 -08005303 .allow_attach = mem_cgroup_allow_attach,
Tejun Heof00baae2013-04-15 13:41:15 -07005304 .bind = mem_cgroup_bind,
Johannes Weiner241994e2015-02-11 15:26:06 -08005305 .dfl_cftypes = memory_files,
5306 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005307 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005308};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08005309
Johannes Weiner241994e2015-02-11 15:26:06 -08005310/**
Johannes Weiner241994e2015-02-11 15:26:06 -08005311 * mem_cgroup_low - check if memory consumption is below the normal range
5312 * @root: the highest ancestor to consider
5313 * @memcg: the memory cgroup to check
5314 *
5315 * Returns %true if memory consumption of @memcg, and that of all
5316 * configurable ancestors up to @root, is below the normal range.
5317 */
5318bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5319{
5320 if (mem_cgroup_disabled())
5321 return false;
5322
5323 /*
5324 * The toplevel group doesn't have a configurable range, so
5325 * it's never low when looked at directly, and it is not
5326 * considered an ancestor when assessing the hierarchy.
5327 */
5328
5329 if (memcg == root_mem_cgroup)
5330 return false;
5331
Michal Hocko4e54ded2015-02-27 15:51:46 -08005332 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005333 return false;
5334
5335 while (memcg != root) {
5336 memcg = parent_mem_cgroup(memcg);
5337
5338 if (memcg == root_mem_cgroup)
5339 break;
5340
Michal Hocko4e54ded2015-02-27 15:51:46 -08005341 if (page_counter_read(&memcg->memory) >= memcg->low)
Johannes Weiner241994e2015-02-11 15:26:06 -08005342 return false;
5343 }
5344 return true;
5345}
5346
Johannes Weiner00501b52014-08-08 14:19:20 -07005347/**
5348 * mem_cgroup_try_charge - try charging a page
5349 * @page: page to charge
5350 * @mm: mm context of the victim
5351 * @gfp_mask: reclaim mode
5352 * @memcgp: charged memcg return
Li RongQing25843c22016-07-26 15:26:56 -07005353 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005354 *
5355 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5356 * pages according to @gfp_mask if necessary.
5357 *
5358 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5359 * Otherwise, an error code is returned.
5360 *
5361 * After page->mapping has been set up, the caller must finalize the
5362 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5363 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5364 */
5365int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005366 gfp_t gfp_mask, struct mem_cgroup **memcgp,
5367 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005368{
5369 struct mem_cgroup *memcg = NULL;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005370 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005371 int ret = 0;
5372
5373 if (mem_cgroup_disabled())
5374 goto out;
5375
5376 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005377 /*
5378 * Every swap fault against a single page tries to charge the
5379 * page, bail as early as possible. shmem_unuse() encounters
5380 * already charged pages, too. The USED bit is protected by
5381 * the page lock, which serializes swap cache removal, which
5382 * in turn serializes uncharging.
5383 */
Vladimir Davydove993d902015-09-09 15:35:35 -07005384 VM_BUG_ON_PAGE(!PageLocked(page), page);
Johannes Weiner1306a852014-12-10 15:44:52 -08005385 if (page->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07005386 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07005387
Vladimir Davydov37e84352016-01-20 15:02:56 -08005388 if (do_swap_account) {
Vladimir Davydove993d902015-09-09 15:35:35 -07005389 swp_entry_t ent = { .val = page_private(page), };
5390 unsigned short id = lookup_swap_cgroup_id(ent);
5391
5392 rcu_read_lock();
5393 memcg = mem_cgroup_from_id(id);
5394 if (memcg && !css_tryget_online(&memcg->css))
5395 memcg = NULL;
5396 rcu_read_unlock();
5397 }
Johannes Weiner00501b52014-08-08 14:19:20 -07005398 }
5399
Johannes Weiner00501b52014-08-08 14:19:20 -07005400 if (!memcg)
5401 memcg = get_mem_cgroup_from_mm(mm);
5402
5403 ret = try_charge(memcg, gfp_mask, nr_pages);
5404
5405 css_put(&memcg->css);
Johannes Weiner00501b52014-08-08 14:19:20 -07005406out:
5407 *memcgp = memcg;
5408 return ret;
5409}
5410
5411/**
5412 * mem_cgroup_commit_charge - commit a page charge
5413 * @page: page to charge
5414 * @memcg: memcg to charge the page to
5415 * @lrucare: page might be on LRU already
Li RongQing25843c22016-07-26 15:26:56 -07005416 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005417 *
5418 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5419 * after page->mapping has been set up. This must happen atomically
5420 * as part of the page instantiation, i.e. under the page table lock
5421 * for anonymous pages, under the page lock for page and swap cache.
5422 *
5423 * In addition, the page must not be on the LRU during the commit, to
5424 * prevent racing with task migration. If it might be, use @lrucare.
5425 *
5426 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5427 */
5428void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005429 bool lrucare, bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005430{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005431 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005432
5433 VM_BUG_ON_PAGE(!page->mapping, page);
5434 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5435
5436 if (mem_cgroup_disabled())
5437 return;
5438 /*
5439 * Swap faults will attempt to charge the same page multiple
5440 * times. But reuse_swap_page() might have removed the page
5441 * from swapcache already, so we can't check PageSwapCache().
5442 */
5443 if (!memcg)
5444 return;
5445
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005446 commit_charge(page, memcg, lrucare);
5447
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005448 local_irq_disable();
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005449 mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005450 memcg_check_events(memcg, page);
5451 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07005452
Johannes Weiner7941d212016-01-14 15:21:23 -08005453 if (do_memsw_account() && PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005454 swp_entry_t entry = { .val = page_private(page) };
5455 /*
5456 * The swap entry might not get freed for a long time,
5457 * let's not wait for it. The page already received a
5458 * memory+swap charge, drop the swap entry duplicate.
5459 */
5460 mem_cgroup_uncharge_swap(entry);
5461 }
5462}
5463
5464/**
5465 * mem_cgroup_cancel_charge - cancel a page charge
5466 * @page: page to charge
5467 * @memcg: memcg to charge the page to
Li RongQing25843c22016-07-26 15:26:56 -07005468 * @compound: charge the page as compound or small page
Johannes Weiner00501b52014-08-08 14:19:20 -07005469 *
5470 * Cancel a charge transaction started by mem_cgroup_try_charge().
5471 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005472void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5473 bool compound)
Johannes Weiner00501b52014-08-08 14:19:20 -07005474{
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005475 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Johannes Weiner00501b52014-08-08 14:19:20 -07005476
5477 if (mem_cgroup_disabled())
5478 return;
5479 /*
5480 * Swap faults will attempt to charge the same page multiple
5481 * times. But reuse_swap_page() might have removed the page
5482 * from swapcache already, so we can't check PageSwapCache().
5483 */
5484 if (!memcg)
5485 return;
5486
Johannes Weiner00501b52014-08-08 14:19:20 -07005487 cancel_charge(memcg, nr_pages);
5488}
5489
Johannes Weiner747db952014-08-08 14:19:24 -07005490static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
Johannes Weiner747db952014-08-08 14:19:24 -07005491 unsigned long nr_anon, unsigned long nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005492 unsigned long nr_huge, unsigned long nr_kmem,
5493 struct page *dummy_page)
Johannes Weiner747db952014-08-08 14:19:24 -07005494{
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005495 unsigned long nr_pages = nr_anon + nr_file + nr_kmem;
Johannes Weiner747db952014-08-08 14:19:24 -07005496 unsigned long flags;
5497
Johannes Weinerce00a962014-09-05 08:43:57 -04005498 if (!mem_cgroup_is_root(memcg)) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005499 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08005500 if (do_memsw_account())
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005501 page_counter_uncharge(&memcg->memsw, nr_pages);
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005502 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem)
5503 page_counter_uncharge(&memcg->kmem, nr_kmem);
Johannes Weinerce00a962014-09-05 08:43:57 -04005504 memcg_oom_recover(memcg);
5505 }
Johannes Weiner747db952014-08-08 14:19:24 -07005506
5507 local_irq_save(flags);
5508 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5509 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5510 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5511 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005512 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005513 memcg_check_events(memcg, dummy_page);
5514 local_irq_restore(flags);
Johannes Weinere8ea14c2014-12-10 15:42:42 -08005515
5516 if (!mem_cgroup_is_root(memcg))
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005517 css_put_many(&memcg->css, nr_pages);
Johannes Weiner747db952014-08-08 14:19:24 -07005518}
5519
5520static void uncharge_list(struct list_head *page_list)
5521{
5522 struct mem_cgroup *memcg = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005523 unsigned long nr_anon = 0;
5524 unsigned long nr_file = 0;
5525 unsigned long nr_huge = 0;
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005526 unsigned long nr_kmem = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005527 unsigned long pgpgout = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005528 struct list_head *next;
5529 struct page *page;
5530
Johannes Weiner8b592652016-03-17 14:20:31 -07005531 /*
5532 * Note that the list can be a single page->lru; hence the
5533 * do-while loop instead of a simple list_for_each_entry().
5534 */
Johannes Weiner747db952014-08-08 14:19:24 -07005535 next = page_list->next;
5536 do {
Johannes Weiner747db952014-08-08 14:19:24 -07005537 page = list_entry(next, struct page, lru);
5538 next = page->lru.next;
5539
5540 VM_BUG_ON_PAGE(PageLRU(page), page);
5541 VM_BUG_ON_PAGE(page_count(page), page);
5542
Johannes Weiner1306a852014-12-10 15:44:52 -08005543 if (!page->mem_cgroup)
Johannes Weiner747db952014-08-08 14:19:24 -07005544 continue;
5545
5546 /*
5547 * Nobody should be changing or seriously looking at
Johannes Weiner1306a852014-12-10 15:44:52 -08005548 * page->mem_cgroup at this point, we have fully
Johannes Weiner29833312014-12-10 15:44:02 -08005549 * exclusive access to the page.
Johannes Weiner747db952014-08-08 14:19:24 -07005550 */
5551
Johannes Weiner1306a852014-12-10 15:44:52 -08005552 if (memcg != page->mem_cgroup) {
Johannes Weiner747db952014-08-08 14:19:24 -07005553 if (memcg) {
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005554 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005555 nr_huge, nr_kmem, page);
5556 pgpgout = nr_anon = nr_file =
5557 nr_huge = nr_kmem = 0;
Johannes Weiner747db952014-08-08 14:19:24 -07005558 }
Johannes Weiner1306a852014-12-10 15:44:52 -08005559 memcg = page->mem_cgroup;
Johannes Weiner747db952014-08-08 14:19:24 -07005560 }
5561
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005562 if (!PageKmemcg(page)) {
5563 unsigned int nr_pages = 1;
Johannes Weiner747db952014-08-08 14:19:24 -07005564
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005565 if (PageTransHuge(page)) {
5566 nr_pages <<= compound_order(page);
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005567 nr_huge += nr_pages;
5568 }
5569 if (PageAnon(page))
5570 nr_anon += nr_pages;
5571 else
5572 nr_file += nr_pages;
5573 pgpgout++;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03005574 } else {
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005575 nr_kmem += 1 << compound_order(page);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03005576 __ClearPageKmemcg(page);
5577 }
Johannes Weiner747db952014-08-08 14:19:24 -07005578
Johannes Weiner1306a852014-12-10 15:44:52 -08005579 page->mem_cgroup = NULL;
Johannes Weiner747db952014-08-08 14:19:24 -07005580 } while (next != page_list);
5581
5582 if (memcg)
Johannes Weiner18eca2e2014-12-10 15:43:57 -08005583 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
Vladimir Davydov5e8d35f2016-07-26 15:24:27 -07005584 nr_huge, nr_kmem, page);
Johannes Weiner747db952014-08-08 14:19:24 -07005585}
5586
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005587/**
5588 * mem_cgroup_uncharge - uncharge a page
5589 * @page: page to uncharge
5590 *
5591 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5592 * mem_cgroup_commit_charge().
5593 */
5594void mem_cgroup_uncharge(struct page *page)
5595{
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005596 if (mem_cgroup_disabled())
5597 return;
5598
Johannes Weiner747db952014-08-08 14:19:24 -07005599 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08005600 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005601 return;
5602
Johannes Weiner747db952014-08-08 14:19:24 -07005603 INIT_LIST_HEAD(&page->lru);
5604 uncharge_list(&page->lru);
5605}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005606
Johannes Weiner747db952014-08-08 14:19:24 -07005607/**
5608 * mem_cgroup_uncharge_list - uncharge a list of page
5609 * @page_list: list of pages to uncharge
5610 *
5611 * Uncharge a list of pages previously charged with
5612 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5613 */
5614void mem_cgroup_uncharge_list(struct list_head *page_list)
5615{
5616 if (mem_cgroup_disabled())
5617 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005618
Johannes Weiner747db952014-08-08 14:19:24 -07005619 if (!list_empty(page_list))
5620 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005621}
5622
5623/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005624 * mem_cgroup_migrate - charge a page's replacement
5625 * @oldpage: currently circulating page
5626 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005627 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005628 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5629 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005630 *
5631 * Both pages must be locked, @newpage->mapping must be set up.
5632 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005633void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005634{
Johannes Weiner29833312014-12-10 15:44:02 -08005635 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005636 unsigned int nr_pages;
5637 bool compound;
Tejun Heod93c4132016-06-24 14:49:54 -07005638 unsigned long flags;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005639
5640 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5641 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005642 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07005643 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5644 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005645
5646 if (mem_cgroup_disabled())
5647 return;
5648
5649 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08005650 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005651 return;
5652
Hugh Dickins45637ba2015-11-05 18:49:40 -08005653 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08005654 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08005655 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005656 return;
5657
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005658 /* Force-charge the new page. The old one will be freed soon */
5659 compound = PageTransHuge(newpage);
5660 nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5661
5662 page_counter_charge(&memcg->memory, nr_pages);
5663 if (do_memsw_account())
5664 page_counter_charge(&memcg->memsw, nr_pages);
5665 css_get_many(&memcg->css, nr_pages);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005666
Johannes Weiner9cf76662016-03-15 14:57:58 -07005667 commit_charge(newpage, memcg, false);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005668
Tejun Heod93c4132016-06-24 14:49:54 -07005669 local_irq_save(flags);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08005670 mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5671 memcg_check_events(memcg, newpage);
Tejun Heod93c4132016-06-24 14:49:54 -07005672 local_irq_restore(flags);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005673}
5674
Johannes Weineref129472016-01-14 15:21:34 -08005675DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08005676EXPORT_SYMBOL(memcg_sockets_enabled_key);
5677
Johannes Weiner2d758072016-10-07 17:00:58 -07005678void mem_cgroup_sk_alloc(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08005679{
5680 struct mem_cgroup *memcg;
5681
Johannes Weiner2d758072016-10-07 17:00:58 -07005682 if (!mem_cgroup_sockets_enabled)
5683 return;
5684
5685 /*
5686 * Socket cloning can throw us here with sk_memcg already
Johannes Weiner11092082016-01-14 15:21:26 -08005687 * filled. It won't however, necessarily happen from
5688 * process context. So the test for root memcg given
5689 * the current task's memcg won't help us in this case.
5690 *
5691 * Respecting the original socket's memcg is a better
5692 * decision in this case.
5693 */
5694 if (sk->sk_memcg) {
5695 BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5696 css_get(&sk->sk_memcg->css);
5697 return;
5698 }
5699
5700 rcu_read_lock();
5701 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005702 if (memcg == root_mem_cgroup)
5703 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08005704 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005705 goto out;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005706 if (css_tryget_online(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08005707 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005708out:
Johannes Weiner11092082016-01-14 15:21:26 -08005709 rcu_read_unlock();
5710}
Johannes Weiner11092082016-01-14 15:21:26 -08005711
Johannes Weiner2d758072016-10-07 17:00:58 -07005712void mem_cgroup_sk_free(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08005713{
Johannes Weiner2d758072016-10-07 17:00:58 -07005714 if (sk->sk_memcg)
5715 css_put(&sk->sk_memcg->css);
Johannes Weiner11092082016-01-14 15:21:26 -08005716}
5717
5718/**
5719 * mem_cgroup_charge_skmem - charge socket memory
5720 * @memcg: memcg to charge
5721 * @nr_pages: number of pages to charge
5722 *
5723 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5724 * @memcg's configured limit, %false if the charge had to be forced.
5725 */
5726bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5727{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005728 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08005729
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005730 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005731 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005732
Johannes Weiner0db15292016-01-20 15:02:50 -08005733 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5734 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005735 return true;
5736 }
Johannes Weiner0db15292016-01-20 15:02:50 -08005737 page_counter_charge(&memcg->tcpmem, nr_pages);
5738 memcg->tcpmem_pressure = 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005739 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08005740 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005741
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005742 /* Don't block in the packet receive path */
5743 if (in_softirq())
5744 gfp_mask = GFP_NOWAIT;
5745
Johannes Weinerb2807f02016-01-20 15:03:22 -08005746 this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5747
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005748 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5749 return true;
5750
5751 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005752 return false;
5753}
5754
5755/**
5756 * mem_cgroup_uncharge_skmem - uncharge socket memory
5757 * @memcg - memcg to uncharge
5758 * @nr_pages - number of pages to uncharge
5759 */
5760void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5761{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005762 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08005763 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005764 return;
5765 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005766
Johannes Weinerb2807f02016-01-20 15:03:22 -08005767 this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5768
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005769 page_counter_uncharge(&memcg->memory, nr_pages);
5770 css_put_many(&memcg->css, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08005771}
5772
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005773static int __init cgroup_memory(char *s)
5774{
5775 char *token;
5776
5777 while ((token = strsep(&s, ",")) != NULL) {
5778 if (!*token)
5779 continue;
5780 if (!strcmp(token, "nosocket"))
5781 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08005782 if (!strcmp(token, "nokmem"))
5783 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005784 }
5785 return 0;
5786}
5787__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08005788
Michal Hocko2d110852013-02-22 16:34:43 -08005789/*
Michal Hocko10813122013-02-22 16:35:41 -08005790 * subsys_initcall() for memory controller.
5791 *
5792 * Some parts like hotcpu_notifier() have to be initialized from this context
5793 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5794 * everything that doesn't depend on a specific mem_cgroup structure should
5795 * be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08005796 */
5797static int __init mem_cgroup_init(void)
5798{
Johannes Weiner95a045f2015-02-11 15:26:33 -08005799 int cpu, node;
5800
Vladimir Davydov754ae7e2016-12-12 16:41:29 -08005801#ifndef CONFIG_SLOB
5802 /*
5803 * Kmem cache creation is mostly done with the slab_mutex held,
5804 * so use a special workqueue to avoid stalling all worker
5805 * threads in case lots of cgroups are created simultaneously.
5806 */
5807 memcg_kmem_cache_create_wq =
5808 alloc_ordered_workqueue("memcg_kmem_cache_create", 0);
5809 BUG_ON(!memcg_kmem_cache_create_wq);
5810#endif
5811
Michal Hocko2d110852013-02-22 16:34:43 -08005812 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005813
5814 for_each_possible_cpu(cpu)
5815 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5816 drain_local_stock);
5817
5818 for_each_node(node) {
5819 struct mem_cgroup_tree_per_node *rtpn;
Johannes Weiner95a045f2015-02-11 15:26:33 -08005820
5821 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5822 node_online(node) ? node : NUMA_NO_NODE);
5823
Mel Gormanef8f2322016-07-28 15:46:05 -07005824 rtpn->rb_root = RB_ROOT;
5825 spin_lock_init(&rtpn->lock);
Johannes Weiner95a045f2015-02-11 15:26:33 -08005826 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5827 }
5828
Michal Hocko2d110852013-02-22 16:34:43 -08005829 return 0;
5830}
5831subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08005832
5833#ifdef CONFIG_MEMCG_SWAP
Arnd Bergmann358c07f2016-08-25 15:17:08 -07005834static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
5835{
5836 while (!atomic_inc_not_zero(&memcg->id.ref)) {
5837 /*
5838 * The root cgroup cannot be destroyed, so it's refcount must
5839 * always be >= 1.
5840 */
5841 if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
5842 VM_BUG_ON(1);
5843 break;
5844 }
5845 memcg = parent_mem_cgroup(memcg);
5846 if (!memcg)
5847 memcg = root_mem_cgroup;
5848 }
5849 return memcg;
5850}
5851
Johannes Weiner21afa382015-02-11 15:26:36 -08005852/**
5853 * mem_cgroup_swapout - transfer a memsw charge to swap
5854 * @page: page whose memsw charge to transfer
5855 * @entry: swap entry to move the charge to
5856 *
5857 * Transfer the memsw charge of @page to @entry.
5858 */
5859void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5860{
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005861 struct mem_cgroup *memcg, *swap_memcg;
Johannes Weiner21afa382015-02-11 15:26:36 -08005862 unsigned short oldid;
5863
5864 VM_BUG_ON_PAGE(PageLRU(page), page);
5865 VM_BUG_ON_PAGE(page_count(page), page);
5866
Johannes Weiner7941d212016-01-14 15:21:23 -08005867 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08005868 return;
5869
5870 memcg = page->mem_cgroup;
5871
5872 /* Readahead page, never charged */
5873 if (!memcg)
5874 return;
5875
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005876 /*
5877 * In case the memcg owning these pages has been offlined and doesn't
5878 * have an ID allocated to it anymore, charge the closest online
5879 * ancestor for the swap instead and transfer the memory+swap charge.
5880 */
5881 swap_memcg = mem_cgroup_id_get_online(memcg);
5882 oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg));
Johannes Weiner21afa382015-02-11 15:26:36 -08005883 VM_BUG_ON_PAGE(oldid, page);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005884 mem_cgroup_swap_statistics(swap_memcg, true);
Johannes Weiner21afa382015-02-11 15:26:36 -08005885
5886 page->mem_cgroup = NULL;
5887
5888 if (!mem_cgroup_is_root(memcg))
5889 page_counter_uncharge(&memcg->memory, 1);
5890
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005891 if (memcg != swap_memcg) {
5892 if (!mem_cgroup_is_root(swap_memcg))
5893 page_counter_charge(&swap_memcg->memsw, 1);
5894 page_counter_uncharge(&memcg->memsw, 1);
5895 }
5896
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07005897 /*
5898 * Interrupts should be disabled here because the caller holds the
5899 * mapping->tree_lock lock which is taken with interrupts-off. It is
5900 * important here to have the interrupts disabled because it is the
5901 * only synchronisation we have for udpating the per-CPU variables.
5902 */
5903 VM_BUG_ON(!irqs_disabled());
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005904 mem_cgroup_charge_statistics(memcg, page, false, -1);
Johannes Weiner21afa382015-02-11 15:26:36 -08005905 memcg_check_events(memcg, page);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005906
5907 if (!mem_cgroup_is_root(memcg))
5908 css_put(&memcg->css);
Johannes Weiner21afa382015-02-11 15:26:36 -08005909}
5910
Vladimir Davydov37e84352016-01-20 15:02:56 -08005911/*
5912 * mem_cgroup_try_charge_swap - try charging a swap entry
5913 * @page: page being added to swap
5914 * @entry: swap entry to charge
5915 *
5916 * Try to charge @entry to the memcg that @page belongs to.
5917 *
5918 * Returns 0 on success, -ENOMEM on failure.
5919 */
5920int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5921{
5922 struct mem_cgroup *memcg;
5923 struct page_counter *counter;
5924 unsigned short oldid;
5925
5926 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5927 return 0;
5928
5929 memcg = page->mem_cgroup;
5930
5931 /* Readahead page, never charged */
5932 if (!memcg)
5933 return 0;
5934
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005935 memcg = mem_cgroup_id_get_online(memcg);
Vladimir Davydov37e84352016-01-20 15:02:56 -08005936
Vladimir Davydov1f47b612016-08-11 15:33:00 -07005937 if (!mem_cgroup_is_root(memcg) &&
5938 !page_counter_try_charge(&memcg->swap, 1, &counter)) {
5939 mem_cgroup_id_put(memcg);
5940 return -ENOMEM;
5941 }
5942
Vladimir Davydov37e84352016-01-20 15:02:56 -08005943 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5944 VM_BUG_ON_PAGE(oldid, page);
5945 mem_cgroup_swap_statistics(memcg, true);
5946
Vladimir Davydov37e84352016-01-20 15:02:56 -08005947 return 0;
5948}
5949
Johannes Weiner21afa382015-02-11 15:26:36 -08005950/**
5951 * mem_cgroup_uncharge_swap - uncharge a swap entry
5952 * @entry: swap entry to uncharge
5953 *
Vladimir Davydov37e84352016-01-20 15:02:56 -08005954 * Drop the swap charge associated with @entry.
Johannes Weiner21afa382015-02-11 15:26:36 -08005955 */
5956void mem_cgroup_uncharge_swap(swp_entry_t entry)
5957{
5958 struct mem_cgroup *memcg;
5959 unsigned short id;
5960
Vladimir Davydov37e84352016-01-20 15:02:56 -08005961 if (!do_swap_account)
Johannes Weiner21afa382015-02-11 15:26:36 -08005962 return;
5963
5964 id = swap_cgroup_record(entry, 0);
5965 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07005966 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08005967 if (memcg) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08005968 if (!mem_cgroup_is_root(memcg)) {
5969 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5970 page_counter_uncharge(&memcg->swap, 1);
5971 else
5972 page_counter_uncharge(&memcg->memsw, 1);
5973 }
Johannes Weiner21afa382015-02-11 15:26:36 -08005974 mem_cgroup_swap_statistics(memcg, false);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005975 mem_cgroup_id_put(memcg);
Johannes Weiner21afa382015-02-11 15:26:36 -08005976 }
5977 rcu_read_unlock();
5978}
5979
Vladimir Davydovd8b38432016-01-20 15:03:07 -08005980long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
5981{
5982 long nr_swap_pages = get_nr_swap_pages();
5983
5984 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5985 return nr_swap_pages;
5986 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5987 nr_swap_pages = min_t(long, nr_swap_pages,
5988 READ_ONCE(memcg->swap.limit) -
5989 page_counter_read(&memcg->swap));
5990 return nr_swap_pages;
5991}
5992
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08005993bool mem_cgroup_swap_full(struct page *page)
5994{
5995 struct mem_cgroup *memcg;
5996
5997 VM_BUG_ON_PAGE(!PageLocked(page), page);
5998
5999 if (vm_swap_full())
6000 return true;
6001 if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
6002 return false;
6003
6004 memcg = page->mem_cgroup;
6005 if (!memcg)
6006 return false;
6007
6008 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
6009 if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
6010 return true;
6011
6012 return false;
6013}
6014
Johannes Weiner21afa382015-02-11 15:26:36 -08006015/* for remember boot option*/
6016#ifdef CONFIG_MEMCG_SWAP_ENABLED
6017static int really_do_swap_account __initdata = 1;
6018#else
6019static int really_do_swap_account __initdata;
6020#endif
6021
6022static int __init enable_swap_account(char *s)
6023{
6024 if (!strcmp(s, "1"))
6025 really_do_swap_account = 1;
6026 else if (!strcmp(s, "0"))
6027 really_do_swap_account = 0;
6028 return 1;
6029}
6030__setup("swapaccount=", enable_swap_account);
6031
Vladimir Davydov37e84352016-01-20 15:02:56 -08006032static u64 swap_current_read(struct cgroup_subsys_state *css,
6033 struct cftype *cft)
6034{
6035 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6036
6037 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
6038}
6039
6040static int swap_max_show(struct seq_file *m, void *v)
6041{
6042 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
6043 unsigned long max = READ_ONCE(memcg->swap.limit);
6044
6045 if (max == PAGE_COUNTER_MAX)
6046 seq_puts(m, "max\n");
6047 else
6048 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
6049
6050 return 0;
6051}
6052
6053static ssize_t swap_max_write(struct kernfs_open_file *of,
6054 char *buf, size_t nbytes, loff_t off)
6055{
6056 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6057 unsigned long max;
6058 int err;
6059
6060 buf = strstrip(buf);
6061 err = page_counter_memparse(buf, "max", &max);
6062 if (err)
6063 return err;
6064
6065 mutex_lock(&memcg_limit_mutex);
6066 err = page_counter_limit(&memcg->swap, max);
6067 mutex_unlock(&memcg_limit_mutex);
6068 if (err)
6069 return err;
6070
6071 return nbytes;
6072}
6073
6074static struct cftype swap_files[] = {
6075 {
6076 .name = "swap.current",
6077 .flags = CFTYPE_NOT_ON_ROOT,
6078 .read_u64 = swap_current_read,
6079 },
6080 {
6081 .name = "swap.max",
6082 .flags = CFTYPE_NOT_ON_ROOT,
6083 .seq_show = swap_max_show,
6084 .write = swap_max_write,
6085 },
6086 { } /* terminate */
6087};
6088
Johannes Weiner21afa382015-02-11 15:26:36 -08006089static struct cftype memsw_cgroup_files[] = {
6090 {
6091 .name = "memsw.usage_in_bytes",
6092 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6093 .read_u64 = mem_cgroup_read_u64,
6094 },
6095 {
6096 .name = "memsw.max_usage_in_bytes",
6097 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6098 .write = mem_cgroup_reset,
6099 .read_u64 = mem_cgroup_read_u64,
6100 },
6101 {
6102 .name = "memsw.limit_in_bytes",
6103 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6104 .write = mem_cgroup_write,
6105 .read_u64 = mem_cgroup_read_u64,
6106 },
6107 {
6108 .name = "memsw.failcnt",
6109 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6110 .write = mem_cgroup_reset,
6111 .read_u64 = mem_cgroup_read_u64,
6112 },
6113 { }, /* terminate */
6114};
6115
6116static int __init mem_cgroup_swap_init(void)
6117{
6118 if (!mem_cgroup_disabled() && really_do_swap_account) {
6119 do_swap_account = 1;
Vladimir Davydov37e84352016-01-20 15:02:56 -08006120 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
6121 swap_files));
Johannes Weiner21afa382015-02-11 15:26:36 -08006122 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
6123 memsw_cgroup_files));
6124 }
6125 return 0;
6126}
6127subsys_initcall(mem_cgroup_swap_init);
6128
6129#endif /* CONFIG_MEMCG_SWAP */