blob: 856db3ba3093806de05d96199ff8ff41ab206b6b [file] [log] [blame]
Thomas Gleixnerc942fdd2019-05-27 08:55:06 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002/* memcontrol.c - Memory Controller
3 *
4 * Copyright IBM Corporation, 2007
5 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08007 * Copyright 2007 OpenVZ SWsoft Inc
8 * Author: Pavel Emelianov <xemul@openvz.org>
9 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080010 * Memory thresholds
11 * Copyright (C) 2009 Nokia Corporation
12 * Author: Kirill A. Shutemov
13 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080014 * Kernel Memory Controller
15 * Copyright (C) 2012 Parallels Inc. and Google Inc.
16 * Authors: Glauber Costa and Suleiman Souhlal
17 *
Johannes Weiner1575e682015-04-14 15:44:51 -070018 * Native page reclaim
19 * Charge lifetime sanitation
20 * Lockless page tracking & accounting
21 * Unified hierarchy configuration model
22 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
Balbir Singh8cdea7c2008-02-07 00:13:50 -080023 */
24
Johannes Weiner3e32cb22014-12-10 15:42:31 -080025#include <linux/page_counter.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080026#include <linux/memcontrol.h>
27#include <linux/cgroup.h>
Christoph Hellwiga5201102019-08-28 16:19:53 +020028#include <linux/pagewalk.h>
Ingo Molnar6e84f312017-02-08 18:51:29 +010029#include <linux/sched/mm.h>
Hugh Dickins3a4f8a02017-02-24 14:59:36 -080030#include <linux/shmem_fs.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080031#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080032#include <linux/pagemap.h>
Chris Down1ff9e6e2019-03-05 15:48:09 -080033#include <linux/vm_event_item.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080034#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080036#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080037#include <linux/bit_spinlock.h>
38#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070039#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040040#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080041#include <linux/mutex.h>
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -070042#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070043#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080044#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080045#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080046#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080047#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050048#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080049#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080051#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070052#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070053#include <linux/mm_inline.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080054#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080055#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070056#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070057#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050058#include <linux/file.h>
Tejun Heob23afb92015-11-05 18:46:11 -080059#include <linux/tracehook.h>
Chris Down0e4b01d2019-09-23 15:34:55 -070060#include <linux/psi.h>
Johannes Weinerc8713d02019-07-11 20:55:59 -070061#include <linux/seq_buf.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080062#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000063#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070064#include <net/ip.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080065#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080066
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080067#include <linux/uaccess.h>
Balbir Singh8697d332008-02-07 00:13:59 -080068
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070069#include <trace/events/vmscan.h>
70
Tejun Heo073219e2014-02-08 10:36:58 -050071struct cgroup_subsys memory_cgrp_subsys __read_mostly;
72EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080073
Johannes Weiner7d828602016-01-14 15:20:56 -080074struct mem_cgroup *root_mem_cgroup __read_mostly;
75
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080076/* Socket memory accounting disabled? */
77static bool cgroup_memory_nosocket;
78
Vladimir Davydov04823c82016-01-20 15:02:38 -080079/* Kernel memory accounting disabled? */
80static bool cgroup_memory_nokmem;
81
Johannes Weiner21afa382015-02-11 15:26:36 -080082/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070083#ifdef CONFIG_MEMCG_SWAP
Johannes Weinereccb52e2020-06-03 16:02:11 -070084bool cgroup_memory_noswap __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080085#else
Johannes Weinereccb52e2020-06-03 16:02:11 -070086#define cgroup_memory_noswap 1
Johannes Weiner2d1c4982020-06-03 16:02:14 -070087#endif
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080088
Tejun Heo97b27822019-08-26 09:06:56 -070089#ifdef CONFIG_CGROUP_WRITEBACK
90static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq);
91#endif
92
Johannes Weiner7941d212016-01-14 15:21:23 -080093/* Whether legacy memory+swap accounting is active */
94static bool do_memsw_account(void)
95{
Johannes Weinereccb52e2020-06-03 16:02:11 -070096 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_noswap;
Johannes Weiner7941d212016-01-14 15:21:23 -080097}
98
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -070099#define THRESHOLDS_EVENTS_TARGET 128
100#define SOFTLIMIT_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700101
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700102/*
103 * Cgroups above their limits are maintained in a RB-Tree, independent of
104 * their hierarchy representation
105 */
106
Mel Gormanef8f2322016-07-28 15:46:05 -0700107struct mem_cgroup_tree_per_node {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700108 struct rb_root rb_root;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700109 struct rb_node *rb_rightmost;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700110 spinlock_t lock;
111};
112
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700113struct mem_cgroup_tree {
114 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
115};
116
117static struct mem_cgroup_tree soft_limit_tree __read_mostly;
118
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700119/* for OOM */
120struct mem_cgroup_eventfd_list {
121 struct list_head list;
122 struct eventfd_ctx *eventfd;
123};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800124
Tejun Heo79bd9812013-11-22 18:20:42 -0500125/*
126 * cgroup_event represents events which userspace want to receive.
127 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500128struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500129 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500130 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500131 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500132 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500133 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500134 * eventfd to signal userspace about the event.
135 */
136 struct eventfd_ctx *eventfd;
137 /*
138 * Each of these stored in a list by the cgroup.
139 */
140 struct list_head list;
141 /*
Tejun Heofba94802013-11-22 18:20:43 -0500142 * register_event() callback will be used to add new userspace
143 * waiter for changes related to this event. Use eventfd_signal()
144 * on eventfd to send notification to userspace.
145 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500146 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500147 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500148 /*
149 * unregister_event() callback will be called when userspace closes
150 * the eventfd or on cgroup removing. This callback must be set,
151 * if you want provide notification functionality.
152 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500153 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500154 struct eventfd_ctx *eventfd);
155 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500156 * All fields below needed to unregister event when
157 * userspace closes eventfd.
158 */
159 poll_table pt;
160 wait_queue_head_t *wqh;
Ingo Molnarac6424b2017-06-20 12:06:13 +0200161 wait_queue_entry_t wait;
Tejun Heo79bd9812013-11-22 18:20:42 -0500162 struct work_struct remove;
163};
164
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700165static void mem_cgroup_threshold(struct mem_cgroup *memcg);
166static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800167
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800168/* Stuffs for move charges at task migration. */
169/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800170 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800171 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800172#define MOVE_ANON 0x1U
173#define MOVE_FILE 0x2U
174#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800175
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800176/* "mc" and its members are protected by cgroup_mutex */
177static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800178 spinlock_t lock; /* for from, to */
Tejun Heo264a0ae2016-04-21 19:09:02 -0400179 struct mm_struct *mm;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800180 struct mem_cgroup *from;
181 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800182 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800183 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800184 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800185 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800186 struct task_struct *moving_task; /* a task moving charges */
187 wait_queue_head_t waitq; /* a waitq for other context */
188} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700189 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800190 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
191};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800192
Balbir Singh4e416952009-09-23 15:56:39 -0700193/*
194 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
195 * limit reclaim to prevent infinite loops, if they ever occur.
196 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700197#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700198#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700199
KAMEZAWA Hiroyuki217bc312008-02-07 00:14:17 -0800200enum charge_type {
201 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
Kamezawa Hiroyuki41326c12012-07-31 16:41:40 -0700202 MEM_CGROUP_CHARGE_TYPE_ANON,
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800203 MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
KAMEZAWA Hiroyuki8a9478c2009-06-17 16:27:17 -0700204 MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
KAMEZAWA Hiroyukic05555b2008-10-18 20:28:11 -0700205 NR_CHARGE_TYPE,
206};
207
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800208/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800209enum res_type {
210 _MEM,
211 _MEMSWAP,
212 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800213 _KMEM,
Vladimir Davydovd55f90b2016-01-20 15:02:44 -0800214 _TCP,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800215};
216
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700217#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
218#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800219#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700220/* Used for OOM nofiier */
221#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800222
Kirill Tkhaib05706f2018-08-17 15:47:33 -0700223/*
224 * Iteration constructs for visiting all cgroups (under a tree). If
225 * loops are exited prematurely (break), mem_cgroup_iter_break() must
226 * be used for reference counting.
227 */
228#define for_each_mem_cgroup_tree(iter, root) \
229 for (iter = mem_cgroup_iter(root, NULL, NULL); \
230 iter != NULL; \
231 iter = mem_cgroup_iter(root, iter, NULL))
232
233#define for_each_mem_cgroup(iter) \
234 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
235 iter != NULL; \
236 iter = mem_cgroup_iter(NULL, iter, NULL))
237
Tetsuo Handa7775fac2019-03-05 15:46:47 -0800238static inline bool should_force_charge(void)
239{
240 return tsk_is_oom_victim(current) || fatal_signal_pending(current) ||
241 (current->flags & PF_EXITING);
242}
243
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700244/* Some nice accessors for the vmpressure. */
245struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
246{
247 if (!memcg)
248 memcg = root_mem_cgroup;
249 return &memcg->vmpressure;
250}
251
252struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
253{
254 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
255}
256
Kirill Tkhai84c07d12018-08-17 15:47:25 -0700257#ifdef CONFIG_MEMCG_KMEM
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700258extern spinlock_t css_set_lock;
259
260static void obj_cgroup_release(struct percpu_ref *ref)
261{
262 struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt);
263 struct mem_cgroup *memcg;
264 unsigned int nr_bytes;
265 unsigned int nr_pages;
266 unsigned long flags;
267
268 /*
269 * At this point all allocated objects are freed, and
270 * objcg->nr_charged_bytes can't have an arbitrary byte value.
271 * However, it can be PAGE_SIZE or (x * PAGE_SIZE).
272 *
273 * The following sequence can lead to it:
274 * 1) CPU0: objcg == stock->cached_objcg
275 * 2) CPU1: we do a small allocation (e.g. 92 bytes),
276 * PAGE_SIZE bytes are charged
277 * 3) CPU1: a process from another memcg is allocating something,
278 * the stock if flushed,
279 * objcg->nr_charged_bytes = PAGE_SIZE - 92
280 * 5) CPU0: we do release this object,
281 * 92 bytes are added to stock->nr_bytes
282 * 6) CPU0: stock is flushed,
283 * 92 bytes are added to objcg->nr_charged_bytes
284 *
285 * In the result, nr_charged_bytes == PAGE_SIZE.
286 * This page will be uncharged in obj_cgroup_release().
287 */
288 nr_bytes = atomic_read(&objcg->nr_charged_bytes);
289 WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1));
290 nr_pages = nr_bytes >> PAGE_SHIFT;
291
292 spin_lock_irqsave(&css_set_lock, flags);
293 memcg = obj_cgroup_memcg(objcg);
294 if (nr_pages)
295 __memcg_kmem_uncharge(memcg, nr_pages);
296 list_del(&objcg->list);
297 mem_cgroup_put(memcg);
298 spin_unlock_irqrestore(&css_set_lock, flags);
299
300 percpu_ref_exit(ref);
301 kfree_rcu(objcg, rcu);
302}
303
304static struct obj_cgroup *obj_cgroup_alloc(void)
305{
306 struct obj_cgroup *objcg;
307 int ret;
308
309 objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL);
310 if (!objcg)
311 return NULL;
312
313 ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0,
314 GFP_KERNEL);
315 if (ret) {
316 kfree(objcg);
317 return NULL;
318 }
319 INIT_LIST_HEAD(&objcg->list);
320 return objcg;
321}
322
323static void memcg_reparent_objcgs(struct mem_cgroup *memcg,
324 struct mem_cgroup *parent)
325{
326 struct obj_cgroup *objcg, *iter;
327
328 objcg = rcu_replace_pointer(memcg->objcg, NULL, true);
329
330 spin_lock_irq(&css_set_lock);
331
332 /* Move active objcg to the parent's list */
333 xchg(&objcg->memcg, parent);
334 css_get(&parent->css);
335 list_add(&objcg->list, &parent->objcg_list);
336
337 /* Move already reparented objcgs to the parent's list */
338 list_for_each_entry(iter, &memcg->objcg_list, list) {
339 css_get(&parent->css);
340 xchg(&iter->memcg, parent);
341 css_put(&memcg->css);
342 }
343 list_splice(&memcg->objcg_list, &parent->objcg_list);
344
345 spin_unlock_irq(&css_set_lock);
346
347 percpu_ref_kill(&objcg->refcnt);
348}
349
Glauber Costa55007d82012-12-18 14:22:38 -0800350/*
Roman Gushchin98556092020-08-06 23:21:10 -0700351 * This will be used as a shrinker list's index.
Li Zefanb8627832013-09-23 16:56:47 +0800352 * The main reason for not using cgroup id for this:
353 * this works better in sparse environments, where we have a lot of memcgs,
354 * but only a few kmem-limited. Or also, if we have, for instance, 200
355 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
356 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800357 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800358 * The current size of the caches array is stored in memcg_nr_cache_ids. It
359 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800360 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800361static DEFINE_IDA(memcg_cache_ida);
362int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800363
Vladimir Davydov05257a12015-02-12 14:59:01 -0800364/* Protects memcg_nr_cache_ids */
365static DECLARE_RWSEM(memcg_cache_ids_sem);
366
367void memcg_get_cache_ids(void)
368{
369 down_read(&memcg_cache_ids_sem);
370}
371
372void memcg_put_cache_ids(void)
373{
374 up_read(&memcg_cache_ids_sem);
375}
376
Glauber Costa55007d82012-12-18 14:22:38 -0800377/*
378 * MIN_SIZE is different than 1, because we would like to avoid going through
379 * the alloc/free process all the time. In a small machine, 4 kmem-limited
380 * cgroups is a reasonable guess. In the future, it could be a parameter or
381 * tunable, but that is strictly not necessary.
382 *
Li Zefanb8627832013-09-23 16:56:47 +0800383 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800384 * this constant directly from cgroup, but it is understandable that this is
385 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800386 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800387 * increase ours as well if it increases.
388 */
389#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800390#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800391
Glauber Costad7f25f82012-12-18 14:22:40 -0800392/*
393 * A lot of the calls to the cache allocation functions are expected to be
Roman Gushchin272911a2020-08-06 23:21:17 -0700394 * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are
Glauber Costad7f25f82012-12-18 14:22:40 -0800395 * conditional to this static branch, we'll have to allow modules that does
396 * kmem_cache_alloc and the such to see this symbol as well
397 */
Johannes Weineref129472016-01-14 15:21:34 -0800398DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
Glauber Costad7f25f82012-12-18 14:22:40 -0800399EXPORT_SYMBOL(memcg_kmem_enabled_key);
Yang Shi0a432dc2019-09-23 15:38:12 -0700400#endif
Tejun Heo17cc4df2017-02-22 15:41:36 -0800401
Kirill Tkhai0a4465d2018-08-17 15:47:37 -0700402static int memcg_shrinker_map_size;
403static DEFINE_MUTEX(memcg_shrinker_map_mutex);
404
405static void memcg_free_shrinker_map_rcu(struct rcu_head *head)
406{
407 kvfree(container_of(head, struct memcg_shrinker_map, rcu));
408}
409
410static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg,
411 int size, int old_size)
412{
413 struct memcg_shrinker_map *new, *old;
414 int nid;
415
416 lockdep_assert_held(&memcg_shrinker_map_mutex);
417
418 for_each_node(nid) {
419 old = rcu_dereference_protected(
420 mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true);
421 /* Not yet online memcg */
422 if (!old)
423 return 0;
424
Kirill Tkhai86daf942020-04-01 21:06:33 -0700425 new = kvmalloc_node(sizeof(*new) + size, GFP_KERNEL, nid);
Kirill Tkhai0a4465d2018-08-17 15:47:37 -0700426 if (!new)
427 return -ENOMEM;
428
429 /* Set all old bits, clear all new bits */
430 memset(new->map, (int)0xff, old_size);
431 memset((void *)new->map + old_size, 0, size - old_size);
432
433 rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new);
434 call_rcu(&old->rcu, memcg_free_shrinker_map_rcu);
435 }
436
437 return 0;
438}
439
440static void memcg_free_shrinker_maps(struct mem_cgroup *memcg)
441{
442 struct mem_cgroup_per_node *pn;
443 struct memcg_shrinker_map *map;
444 int nid;
445
446 if (mem_cgroup_is_root(memcg))
447 return;
448
449 for_each_node(nid) {
450 pn = mem_cgroup_nodeinfo(memcg, nid);
451 map = rcu_dereference_protected(pn->shrinker_map, true);
452 if (map)
453 kvfree(map);
454 rcu_assign_pointer(pn->shrinker_map, NULL);
455 }
456}
457
458static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg)
459{
460 struct memcg_shrinker_map *map;
461 int nid, size, ret = 0;
462
463 if (mem_cgroup_is_root(memcg))
464 return 0;
465
466 mutex_lock(&memcg_shrinker_map_mutex);
467 size = memcg_shrinker_map_size;
468 for_each_node(nid) {
Kirill Tkhai86daf942020-04-01 21:06:33 -0700469 map = kvzalloc_node(sizeof(*map) + size, GFP_KERNEL, nid);
Kirill Tkhai0a4465d2018-08-17 15:47:37 -0700470 if (!map) {
471 memcg_free_shrinker_maps(memcg);
472 ret = -ENOMEM;
473 break;
474 }
475 rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map);
476 }
477 mutex_unlock(&memcg_shrinker_map_mutex);
478
479 return ret;
480}
481
482int memcg_expand_shrinker_maps(int new_id)
483{
484 int size, old_size, ret = 0;
485 struct mem_cgroup *memcg;
486
487 size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long);
488 old_size = memcg_shrinker_map_size;
489 if (size <= old_size)
490 return 0;
491
492 mutex_lock(&memcg_shrinker_map_mutex);
493 if (!root_mem_cgroup)
494 goto unlock;
495
496 for_each_mem_cgroup(memcg) {
497 if (mem_cgroup_is_root(memcg))
498 continue;
499 ret = memcg_expand_one_shrinker_map(memcg, size, old_size);
Vasily Averin75866af2020-02-20 20:04:18 -0800500 if (ret) {
501 mem_cgroup_iter_break(NULL, memcg);
Kirill Tkhai0a4465d2018-08-17 15:47:37 -0700502 goto unlock;
Vasily Averin75866af2020-02-20 20:04:18 -0800503 }
Kirill Tkhai0a4465d2018-08-17 15:47:37 -0700504 }
505unlock:
506 if (!ret)
507 memcg_shrinker_map_size = size;
508 mutex_unlock(&memcg_shrinker_map_mutex);
509 return ret;
510}
Kirill Tkhaifae91d62018-08-17 15:48:10 -0700511
512void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id)
513{
514 if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) {
515 struct memcg_shrinker_map *map;
516
517 rcu_read_lock();
518 map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map);
Kirill Tkhaif90280d2018-08-17 15:48:25 -0700519 /* Pairs with smp mb in shrink_slab() */
520 smp_mb__before_atomic();
Kirill Tkhaifae91d62018-08-17 15:48:10 -0700521 set_bit(shrinker_id, map->map);
522 rcu_read_unlock();
523 }
524}
525
Tejun Heoad7fa852015-05-27 20:00:02 -0400526/**
527 * mem_cgroup_css_from_page - css of the memcg associated with a page
528 * @page: page of interest
529 *
530 * If memcg is bound to the default hierarchy, css of the memcg associated
531 * with @page is returned. The returned css remains associated with @page
532 * until it is released.
533 *
534 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
535 * is returned.
Tejun Heoad7fa852015-05-27 20:00:02 -0400536 */
537struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
538{
539 struct mem_cgroup *memcg;
540
Tejun Heoad7fa852015-05-27 20:00:02 -0400541 memcg = page->mem_cgroup;
542
Tejun Heo9e10a132015-09-18 11:56:28 -0400543 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400544 memcg = root_mem_cgroup;
545
Tejun Heoad7fa852015-05-27 20:00:02 -0400546 return &memcg->css;
547}
548
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700549/**
550 * page_cgroup_ino - return inode number of the memcg a page is charged to
551 * @page: the page
552 *
553 * Look up the closest online ancestor of the memory cgroup @page is charged to
554 * and return its inode number or 0 if @page is not charged to any cgroup. It
555 * is safe to call this function without holding a reference to @page.
556 *
557 * Note, this function is inherently racy, because there is nothing to prevent
558 * the cgroup inode from getting torn down and potentially reallocated a moment
559 * after page_cgroup_ino() returns, so it only should be used by callers that
560 * do not care (such as procfs interfaces).
561 */
562ino_t page_cgroup_ino(struct page *page)
563{
564 struct mem_cgroup *memcg;
565 unsigned long ino = 0;
566
567 rcu_read_lock();
Roman Gushchin98556092020-08-06 23:21:10 -0700568 memcg = page->mem_cgroup;
Roman Gushchin286e04b2020-08-06 23:20:52 -0700569
Roman Gushchin98556092020-08-06 23:21:10 -0700570 /*
571 * The lowest bit set means that memcg isn't a valid
572 * memcg pointer, but a obj_cgroups pointer.
573 * In this case the page is shared and doesn't belong
574 * to any specific memory cgroup.
575 */
576 if ((unsigned long) memcg & 0x1UL)
577 memcg = NULL;
Roman Gushchin286e04b2020-08-06 23:20:52 -0700578
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700579 while (memcg && !(memcg->css.flags & CSS_ONLINE))
580 memcg = parent_mem_cgroup(memcg);
581 if (memcg)
582 ino = cgroup_ino(memcg->css.cgroup);
583 rcu_read_unlock();
584 return ino;
585}
586
Mel Gormanef8f2322016-07-28 15:46:05 -0700587static struct mem_cgroup_per_node *
588mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700589{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700590 int nid = page_to_nid(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700591
Mel Gormanef8f2322016-07-28 15:46:05 -0700592 return memcg->nodeinfo[nid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700593}
594
Mel Gormanef8f2322016-07-28 15:46:05 -0700595static struct mem_cgroup_tree_per_node *
596soft_limit_tree_node(int nid)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700597{
Mel Gormanef8f2322016-07-28 15:46:05 -0700598 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700599}
600
Mel Gormanef8f2322016-07-28 15:46:05 -0700601static struct mem_cgroup_tree_per_node *
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700602soft_limit_tree_from_page(struct page *page)
603{
604 int nid = page_to_nid(page);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700605
Mel Gormanef8f2322016-07-28 15:46:05 -0700606 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700607}
608
Mel Gormanef8f2322016-07-28 15:46:05 -0700609static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz,
610 struct mem_cgroup_tree_per_node *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800611 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700612{
613 struct rb_node **p = &mctz->rb_root.rb_node;
614 struct rb_node *parent = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700615 struct mem_cgroup_per_node *mz_node;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700616 bool rightmost = true;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700617
618 if (mz->on_tree)
619 return;
620
621 mz->usage_in_excess = new_usage_in_excess;
622 if (!mz->usage_in_excess)
623 return;
624 while (*p) {
625 parent = *p;
Mel Gormanef8f2322016-07-28 15:46:05 -0700626 mz_node = rb_entry(parent, struct mem_cgroup_per_node,
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700627 tree_node);
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700628 if (mz->usage_in_excess < mz_node->usage_in_excess) {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700629 p = &(*p)->rb_left;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700630 rightmost = false;
631 }
632
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700633 /*
634 * We can't avoid mem cgroups that are over their soft
635 * limit by the same amount
636 */
637 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
638 p = &(*p)->rb_right;
639 }
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700640
641 if (rightmost)
642 mctz->rb_rightmost = &mz->tree_node;
643
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700644 rb_link_node(&mz->tree_node, parent, p);
645 rb_insert_color(&mz->tree_node, &mctz->rb_root);
646 mz->on_tree = true;
647}
648
Mel Gormanef8f2322016-07-28 15:46:05 -0700649static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
650 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700651{
652 if (!mz->on_tree)
653 return;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700654
655 if (&mz->tree_node == mctz->rb_rightmost)
656 mctz->rb_rightmost = rb_prev(&mz->tree_node);
657
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700658 rb_erase(&mz->tree_node, &mctz->rb_root);
659 mz->on_tree = false;
660}
661
Mel Gormanef8f2322016-07-28 15:46:05 -0700662static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
663 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700664{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700665 unsigned long flags;
666
667 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700668 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700669 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700670}
671
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800672static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
673{
674 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700675 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800676 unsigned long excess = 0;
677
678 if (nr_pages > soft_limit)
679 excess = nr_pages - soft_limit;
680
681 return excess;
682}
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700683
684static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
685{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800686 unsigned long excess;
Mel Gormanef8f2322016-07-28 15:46:05 -0700687 struct mem_cgroup_per_node *mz;
688 struct mem_cgroup_tree_per_node *mctz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700689
Jianyu Zhane2318752014-06-06 14:38:20 -0700690 mctz = soft_limit_tree_from_page(page);
Laurent Dufourbfc72282017-03-09 16:17:06 -0800691 if (!mctz)
692 return;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700693 /*
694 * Necessary to update all ancestors when hierarchy is used.
695 * because their event counter is not touched.
696 */
697 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700698 mz = mem_cgroup_page_nodeinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800699 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700700 /*
701 * We have to update the tree if mz is on RB-tree or
702 * mem is over its softlimit.
703 */
704 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700705 unsigned long flags;
706
707 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700708 /* if on-tree, remove it */
709 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700710 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700711 /*
712 * Insert again. mz->usage_in_excess will be updated.
713 * If excess is 0, no tree ops.
714 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700715 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700716 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700717 }
718 }
719}
720
721static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
722{
Mel Gormanef8f2322016-07-28 15:46:05 -0700723 struct mem_cgroup_tree_per_node *mctz;
724 struct mem_cgroup_per_node *mz;
725 int nid;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700726
Jianyu Zhane2318752014-06-06 14:38:20 -0700727 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700728 mz = mem_cgroup_nodeinfo(memcg, nid);
729 mctz = soft_limit_tree_node(nid);
Laurent Dufourbfc72282017-03-09 16:17:06 -0800730 if (mctz)
731 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700732 }
733}
734
Mel Gormanef8f2322016-07-28 15:46:05 -0700735static struct mem_cgroup_per_node *
736__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700737{
Mel Gormanef8f2322016-07-28 15:46:05 -0700738 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700739
740retry:
741 mz = NULL;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700742 if (!mctz->rb_rightmost)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700743 goto done; /* Nothing to reclaim from */
744
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700745 mz = rb_entry(mctz->rb_rightmost,
746 struct mem_cgroup_per_node, tree_node);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700747 /*
748 * Remove the node now but someone else can add it back,
749 * we will to add it back at the end of reclaim to its correct
750 * position in the tree.
751 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700752 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800753 if (!soft_limit_excess(mz->memcg) ||
Shakeel Butt8965aa22020-04-01 21:07:10 -0700754 !css_tryget(&mz->memcg->css))
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700755 goto retry;
756done:
757 return mz;
758}
759
Mel Gormanef8f2322016-07-28 15:46:05 -0700760static struct mem_cgroup_per_node *
761mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700762{
Mel Gormanef8f2322016-07-28 15:46:05 -0700763 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700764
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700765 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700766 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700767 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700768 return mz;
769}
770
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700771/**
772 * __mod_memcg_state - update cgroup memory statistics
773 * @memcg: the memory cgroup
774 * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item
775 * @val: delta to add to the counter, can be negative
776 */
777void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val)
778{
Roman Gushchinea426c22020-08-06 23:20:35 -0700779 long x, threshold = MEMCG_CHARGE_BATCH;
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700780
781 if (mem_cgroup_disabled())
782 return;
783
Roman Gushchinea426c22020-08-06 23:20:35 -0700784 if (vmstat_item_in_bytes(idx))
785 threshold <<= PAGE_SHIFT;
786
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700787 x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]);
Roman Gushchinea426c22020-08-06 23:20:35 -0700788 if (unlikely(abs(x) > threshold)) {
Johannes Weiner42a30032019-05-14 15:47:12 -0700789 struct mem_cgroup *mi;
790
Yafang Shao766a4c12019-07-16 16:26:06 -0700791 /*
792 * Batch local counters to keep them in sync with
793 * the hierarchical ones.
794 */
795 __this_cpu_add(memcg->vmstats_local->stat[idx], x);
Johannes Weiner42a30032019-05-14 15:47:12 -0700796 for (mi = memcg; mi; mi = parent_mem_cgroup(mi))
797 atomic_long_add(x, &mi->vmstats[idx]);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700798 x = 0;
799 }
800 __this_cpu_write(memcg->vmstats_percpu->stat[idx], x);
801}
802
Johannes Weiner42a30032019-05-14 15:47:12 -0700803static struct mem_cgroup_per_node *
804parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid)
805{
806 struct mem_cgroup *parent;
807
808 parent = parent_mem_cgroup(pn->memcg);
809 if (!parent)
810 return NULL;
811 return mem_cgroup_nodeinfo(parent, nid);
812}
813
Roman Gushchineedc4e52020-08-06 23:20:32 -0700814void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
815 int val)
816{
817 struct mem_cgroup_per_node *pn;
818 struct mem_cgroup *memcg;
Roman Gushchinea426c22020-08-06 23:20:35 -0700819 long x, threshold = MEMCG_CHARGE_BATCH;
Roman Gushchineedc4e52020-08-06 23:20:32 -0700820
821 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
822 memcg = pn->memcg;
823
824 /* Update memcg */
825 __mod_memcg_state(memcg, idx, val);
826
827 /* Update lruvec */
828 __this_cpu_add(pn->lruvec_stat_local->count[idx], val);
829
Roman Gushchinea426c22020-08-06 23:20:35 -0700830 if (vmstat_item_in_bytes(idx))
831 threshold <<= PAGE_SHIFT;
832
Roman Gushchineedc4e52020-08-06 23:20:32 -0700833 x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]);
Roman Gushchinea426c22020-08-06 23:20:35 -0700834 if (unlikely(abs(x) > threshold)) {
Roman Gushchineedc4e52020-08-06 23:20:32 -0700835 pg_data_t *pgdat = lruvec_pgdat(lruvec);
836 struct mem_cgroup_per_node *pi;
837
838 for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id))
839 atomic_long_add(x, &pi->lruvec_stat[idx]);
840 x = 0;
841 }
842 __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x);
843}
844
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700845/**
846 * __mod_lruvec_state - update lruvec memory statistics
847 * @lruvec: the lruvec
848 * @idx: the stat item
849 * @val: delta to add to the counter, can be negative
850 *
851 * The lruvec is the intersection of the NUMA node and a cgroup. This
852 * function updates the all three counters that are affected by a
853 * change of state at this level: per-node, per-cgroup, per-lruvec.
854 */
855void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
856 int val)
857{
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700858 /* Update node */
Roman Gushchineedc4e52020-08-06 23:20:32 -0700859 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700860
Roman Gushchineedc4e52020-08-06 23:20:32 -0700861 /* Update memcg and lruvec */
862 if (!mem_cgroup_disabled())
863 __mod_memcg_lruvec_state(lruvec, idx, val);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700864}
865
Roman Gushchinec9f0232019-08-13 15:37:41 -0700866void __mod_lruvec_slab_state(void *p, enum node_stat_item idx, int val)
867{
Roman Gushchin4f103c62020-04-01 21:06:36 -0700868 pg_data_t *pgdat = page_pgdat(virt_to_page(p));
Roman Gushchinec9f0232019-08-13 15:37:41 -0700869 struct mem_cgroup *memcg;
870 struct lruvec *lruvec;
871
872 rcu_read_lock();
Roman Gushchin4f103c62020-04-01 21:06:36 -0700873 memcg = mem_cgroup_from_obj(p);
Roman Gushchinec9f0232019-08-13 15:37:41 -0700874
875 /* Untracked pages have no memcg, no lruvec. Update only the node */
876 if (!memcg || memcg == root_mem_cgroup) {
877 __mod_node_page_state(pgdat, idx, val);
878 } else {
Johannes Weiner867e5e12019-11-30 17:55:34 -0800879 lruvec = mem_cgroup_lruvec(memcg, pgdat);
Roman Gushchinec9f0232019-08-13 15:37:41 -0700880 __mod_lruvec_state(lruvec, idx, val);
881 }
882 rcu_read_unlock();
883}
884
Roman Gushchin8380ce42020-03-28 19:17:25 -0700885void mod_memcg_obj_state(void *p, int idx, int val)
886{
887 struct mem_cgroup *memcg;
888
889 rcu_read_lock();
890 memcg = mem_cgroup_from_obj(p);
891 if (memcg)
892 mod_memcg_state(memcg, idx, val);
893 rcu_read_unlock();
894}
895
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700896/**
897 * __count_memcg_events - account VM events in a cgroup
898 * @memcg: the memory cgroup
899 * @idx: the event item
900 * @count: the number of events that occured
901 */
902void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx,
903 unsigned long count)
904{
905 unsigned long x;
906
907 if (mem_cgroup_disabled())
908 return;
909
910 x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]);
911 if (unlikely(x > MEMCG_CHARGE_BATCH)) {
Johannes Weiner42a30032019-05-14 15:47:12 -0700912 struct mem_cgroup *mi;
913
Yafang Shao766a4c12019-07-16 16:26:06 -0700914 /*
915 * Batch local counters to keep them in sync with
916 * the hierarchical ones.
917 */
918 __this_cpu_add(memcg->vmstats_local->events[idx], x);
Johannes Weiner42a30032019-05-14 15:47:12 -0700919 for (mi = memcg; mi; mi = parent_mem_cgroup(mi))
920 atomic_long_add(x, &mi->vmevents[idx]);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700921 x = 0;
922 }
923 __this_cpu_write(memcg->vmstats_percpu->events[idx], x);
924}
925
Johannes Weiner42a30032019-05-14 15:47:12 -0700926static unsigned long memcg_events(struct mem_cgroup *memcg, int event)
Johannes Weinere9f89742011-03-23 16:42:37 -0700927{
Chris Down871789d2019-05-14 15:46:57 -0700928 return atomic_long_read(&memcg->vmevents[event]);
Johannes Weinere9f89742011-03-23 16:42:37 -0700929}
930
Johannes Weiner42a30032019-05-14 15:47:12 -0700931static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event)
932{
Johannes Weiner815744d2019-06-13 15:55:46 -0700933 long x = 0;
934 int cpu;
935
936 for_each_possible_cpu(cpu)
937 x += per_cpu(memcg->vmstats_local->events[event], cpu);
938 return x;
Johannes Weiner42a30032019-05-14 15:47:12 -0700939}
940
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700941static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700942 struct page *page,
Johannes Weiner3fba69a2020-06-03 16:01:31 -0700943 int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800944{
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800945 /* pagein of a big page is an event. So, ignore page size */
946 if (nr_pages > 0)
Johannes Weinerc9019e92018-01-31 16:16:37 -0800947 __count_memcg_events(memcg, PGPGIN, 1);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800948 else {
Johannes Weinerc9019e92018-01-31 16:16:37 -0800949 __count_memcg_events(memcg, PGPGOUT, 1);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800950 nr_pages = -nr_pages; /* for event */
951 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800952
Chris Down871789d2019-05-14 15:46:57 -0700953 __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800954}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800955
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800956static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
957 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800958{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700959 unsigned long val, next;
960
Chris Down871789d2019-05-14 15:46:57 -0700961 val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events);
962 next = __this_cpu_read(memcg->vmstats_percpu->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700963 /* from time_after() in jiffies.h */
Michal Hocko6a1a8b82017-07-10 15:48:53 -0700964 if ((long)(next - val) < 0) {
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800965 switch (target) {
966 case MEM_CGROUP_TARGET_THRESH:
967 next = val + THRESHOLDS_EVENTS_TARGET;
968 break;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700969 case MEM_CGROUP_TARGET_SOFTLIMIT:
970 next = val + SOFTLIMIT_EVENTS_TARGET;
971 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800972 default:
973 break;
974 }
Chris Down871789d2019-05-14 15:46:57 -0700975 __this_cpu_write(memcg->vmstats_percpu->targets[target], next);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800976 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700977 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800978 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800979}
980
981/*
982 * Check events in order.
983 *
984 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700985static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800986{
987 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800988 if (unlikely(mem_cgroup_event_ratelimit(memcg,
989 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700990 bool do_softlimit;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800991
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700992 do_softlimit = mem_cgroup_event_ratelimit(memcg,
993 MEM_CGROUP_TARGET_SOFTLIMIT);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800994 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700995 if (unlikely(do_softlimit))
996 mem_cgroup_update_tree(memcg, page);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700997 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800998}
999
Balbir Singhcf475ad2008-04-29 01:00:16 -07001000struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001001{
Balbir Singh31a78f22008-09-28 23:09:31 +01001002 /*
1003 * mm_update_next_owner() may clear mm->owner to NULL
1004 * if it races with swapoff, page migration, etc.
1005 * So this can be called with p == NULL.
1006 */
1007 if (unlikely(!p))
1008 return NULL;
1009
Tejun Heo073219e2014-02-08 10:36:58 -05001010 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001011}
Michal Hocko33398cf2015-09-08 15:01:02 -07001012EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001013
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001014/**
1015 * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg.
1016 * @mm: mm from which memcg should be extracted. It can be NULL.
1017 *
1018 * Obtain a reference on mm->memcg and returns it if successful. Otherwise
1019 * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is
1020 * returned.
1021 */
1022struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001023{
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001024 struct mem_cgroup *memcg;
1025
1026 if (mem_cgroup_disabled())
1027 return NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001028
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001029 rcu_read_lock();
1030 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -07001031 /*
1032 * Page cache insertions can happen withou an
1033 * actual mm context, e.g. during disk probing
1034 * on boot, loopback IO, acct() writes etc.
1035 */
1036 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -07001037 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -07001038 else {
1039 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1040 if (unlikely(!memcg))
1041 memcg = root_mem_cgroup;
1042 }
Roman Gushchin00d484f2019-11-15 17:34:43 -08001043 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001044 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001045 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001046}
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001047EXPORT_SYMBOL(get_mem_cgroup_from_mm);
1048
1049/**
Shakeel Buttf745c6f2018-08-17 15:46:44 -07001050 * get_mem_cgroup_from_page: Obtain a reference on given page's memcg.
1051 * @page: page from which memcg should be extracted.
1052 *
1053 * Obtain a reference on page->memcg and returns it if successful. Otherwise
1054 * root_mem_cgroup is returned.
1055 */
1056struct mem_cgroup *get_mem_cgroup_from_page(struct page *page)
1057{
1058 struct mem_cgroup *memcg = page->mem_cgroup;
1059
1060 if (mem_cgroup_disabled())
1061 return NULL;
1062
1063 rcu_read_lock();
Shakeel Butt8965aa22020-04-01 21:07:10 -07001064 /* Page should not get uncharged and freed memcg under us. */
1065 if (!memcg || WARN_ON_ONCE(!css_tryget(&memcg->css)))
Shakeel Buttf745c6f2018-08-17 15:46:44 -07001066 memcg = root_mem_cgroup;
1067 rcu_read_unlock();
1068 return memcg;
1069}
1070EXPORT_SYMBOL(get_mem_cgroup_from_page);
1071
1072/**
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001073 * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg.
1074 */
1075static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void)
1076{
1077 if (unlikely(current->active_memcg)) {
Shakeel Butt8965aa22020-04-01 21:07:10 -07001078 struct mem_cgroup *memcg;
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001079
1080 rcu_read_lock();
Shakeel Butt8965aa22020-04-01 21:07:10 -07001081 /* current->active_memcg must hold a ref. */
1082 if (WARN_ON_ONCE(!css_tryget(&current->active_memcg->css)))
1083 memcg = root_mem_cgroup;
1084 else
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001085 memcg = current->active_memcg;
1086 rcu_read_unlock();
1087 return memcg;
1088 }
1089 return get_mem_cgroup_from_mm(current->mm);
1090}
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001091
Johannes Weiner56600482012-01-12 17:17:59 -08001092/**
1093 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1094 * @root: hierarchy root
1095 * @prev: previously returned memcg, NULL on first invocation
1096 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1097 *
1098 * Returns references to children of the hierarchy below @root, or
1099 * @root itself, or %NULL after a full round-trip.
1100 *
1101 * Caller must pass the return value in @prev on subsequent
1102 * invocations for reference counting, or use mem_cgroup_iter_break()
1103 * to cancel a hierarchy walk before the round-trip is complete.
1104 *
Honglei Wangb213b542018-03-28 16:01:12 -07001105 * Reclaimers can specify a node and a priority level in @reclaim to
Johannes Weiner56600482012-01-12 17:17:59 -08001106 * divide up the memcgs in the hierarchy among all concurrent
Honglei Wangb213b542018-03-28 16:01:12 -07001107 * reclaimers operating on the same node and priority.
Johannes Weiner56600482012-01-12 17:17:59 -08001108 */
Andrew Morton694fbc02013-09-24 15:27:37 -07001109struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -08001110 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -07001111 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001112{
Kees Cook3f649ab2020-06-03 13:09:38 -07001113 struct mem_cgroup_reclaim_iter *iter;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001114 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001115 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001116 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001117
Andrew Morton694fbc02013-09-24 15:27:37 -07001118 if (mem_cgroup_disabled())
1119 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001120
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001121 if (!root)
1122 root = root_mem_cgroup;
1123
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001124 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001125 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001126
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001127 if (!root->use_hierarchy && root != root_mem_cgroup) {
1128 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001129 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -07001130 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001131 }
1132
Michal Hocko542f85f2013-04-29 15:07:15 -07001133 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001134
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001135 if (reclaim) {
Mel Gormanef8f2322016-07-28 15:46:05 -07001136 struct mem_cgroup_per_node *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001137
Mel Gormanef8f2322016-07-28 15:46:05 -07001138 mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id);
Yafang Shao9da83f32019-11-30 17:50:03 -08001139 iter = &mz->iter;
Michal Hocko5f578162013-04-29 15:07:17 -07001140
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001141 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -07001142 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001143
Vladimir Davydov6df38682015-12-29 14:54:10 -08001144 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001145 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -08001146 if (!pos || css_tryget(&pos->css))
1147 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001148 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -08001149 * css reference reached zero, so iter->position will
1150 * be cleared by ->css_released. However, we should not
1151 * rely on this happening soon, because ->css_released
1152 * is called from a work queue, and by busy-waiting we
1153 * might block it. So we clear iter->position right
1154 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001155 */
Vladimir Davydov6df38682015-12-29 14:54:10 -08001156 (void)cmpxchg(&iter->position, pos, NULL);
1157 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001158 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001159
1160 if (pos)
1161 css = &pos->css;
1162
1163 for (;;) {
1164 css = css_next_descendant_pre(css, &root->css);
1165 if (!css) {
1166 /*
1167 * Reclaimers share the hierarchy walk, and a
1168 * new one might jump in right at the end of
1169 * the hierarchy - make sure they see at least
1170 * one group and restart from the beginning.
1171 */
1172 if (!prev)
1173 continue;
1174 break;
1175 }
1176
1177 /*
1178 * Verify the css and acquire a reference. The root
1179 * is provided by the caller, so we know it's alive
1180 * and kicking, and don't take an extra reference.
1181 */
1182 memcg = mem_cgroup_from_css(css);
1183
1184 if (css == &root->css)
1185 break;
1186
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08001187 if (css_tryget(css))
1188 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001189
1190 memcg = NULL;
1191 }
1192
1193 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001194 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -08001195 * The position could have already been updated by a competing
1196 * thread, so check that the value hasn't changed since we read
1197 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001198 */
Vladimir Davydov6df38682015-12-29 14:54:10 -08001199 (void)cmpxchg(&iter->position, pos, memcg);
1200
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001201 if (pos)
1202 css_put(&pos->css);
1203
1204 if (!memcg)
1205 iter->generation++;
1206 else if (!prev)
1207 reclaim->generation = iter->generation;
1208 }
1209
Michal Hocko542f85f2013-04-29 15:07:15 -07001210out_unlock:
1211 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001212out:
Michal Hockoc40046f2013-04-29 15:07:14 -07001213 if (prev && prev != root)
1214 css_put(&prev->css);
1215
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001216 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001217}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001218
Johannes Weiner56600482012-01-12 17:17:59 -08001219/**
1220 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1221 * @root: hierarchy root
1222 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1223 */
1224void mem_cgroup_iter_break(struct mem_cgroup *root,
1225 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001226{
1227 if (!root)
1228 root = root_mem_cgroup;
1229 if (prev && prev != root)
1230 css_put(&prev->css);
1231}
1232
Miles Chen54a83d62019-08-13 15:37:28 -07001233static void __invalidate_reclaim_iterators(struct mem_cgroup *from,
1234 struct mem_cgroup *dead_memcg)
Vladimir Davydov6df38682015-12-29 14:54:10 -08001235{
Vladimir Davydov6df38682015-12-29 14:54:10 -08001236 struct mem_cgroup_reclaim_iter *iter;
Mel Gormanef8f2322016-07-28 15:46:05 -07001237 struct mem_cgroup_per_node *mz;
1238 int nid;
Vladimir Davydov6df38682015-12-29 14:54:10 -08001239
Miles Chen54a83d62019-08-13 15:37:28 -07001240 for_each_node(nid) {
1241 mz = mem_cgroup_nodeinfo(from, nid);
Yafang Shao9da83f32019-11-30 17:50:03 -08001242 iter = &mz->iter;
1243 cmpxchg(&iter->position, dead_memcg, NULL);
Vladimir Davydov6df38682015-12-29 14:54:10 -08001244 }
1245}
1246
Miles Chen54a83d62019-08-13 15:37:28 -07001247static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
1248{
1249 struct mem_cgroup *memcg = dead_memcg;
1250 struct mem_cgroup *last;
1251
1252 do {
1253 __invalidate_reclaim_iterators(memcg, dead_memcg);
1254 last = memcg;
1255 } while ((memcg = parent_mem_cgroup(memcg)));
1256
1257 /*
1258 * When cgruop1 non-hierarchy mode is used,
1259 * parent_mem_cgroup() does not walk all the way up to the
1260 * cgroup root (root_mem_cgroup). So we have to handle
1261 * dead_memcg from cgroup root separately.
1262 */
1263 if (last != root_mem_cgroup)
1264 __invalidate_reclaim_iterators(root_mem_cgroup,
1265 dead_memcg);
1266}
1267
Johannes Weiner925b7672012-01-12 17:18:15 -08001268/**
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001269 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
1270 * @memcg: hierarchy root
1271 * @fn: function to call for each task
1272 * @arg: argument passed to @fn
1273 *
1274 * This function iterates over tasks attached to @memcg or to any of its
1275 * descendants and calls @fn for each task. If @fn returns a non-zero
1276 * value, the function breaks the iteration loop and returns the value.
1277 * Otherwise, it will iterate over all tasks and return 0.
1278 *
1279 * This function must not be called for the root memory cgroup.
1280 */
1281int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
1282 int (*fn)(struct task_struct *, void *), void *arg)
1283{
1284 struct mem_cgroup *iter;
1285 int ret = 0;
1286
1287 BUG_ON(memcg == root_mem_cgroup);
1288
1289 for_each_mem_cgroup_tree(iter, memcg) {
1290 struct css_task_iter it;
1291 struct task_struct *task;
1292
Tetsuo Handaf168a9a2019-07-11 21:00:20 -07001293 css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001294 while (!ret && (task = css_task_iter_next(&it)))
1295 ret = fn(task, arg);
1296 css_task_iter_end(&it);
1297 if (ret) {
1298 mem_cgroup_iter_break(memcg, iter);
1299 break;
1300 }
1301 }
1302 return ret;
1303}
1304
1305/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001306 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -08001307 * @page: the page
Mike Rapoportf144c392018-02-06 15:42:16 -08001308 * @pgdat: pgdat of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001309 *
Johannes Weinera0b5b412020-06-03 16:02:27 -07001310 * This function relies on page->mem_cgroup being stable - see the
1311 * access rules in commit_charge().
Minchan Kim3f58a822011-03-22 16:32:53 -07001312 */
Mel Gorman599d0c92016-07-28 15:45:31 -07001313struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat)
Minchan Kim3f58a822011-03-22 16:32:53 -07001314{
Mel Gormanef8f2322016-07-28 15:46:05 -07001315 struct mem_cgroup_per_node *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001316 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001317 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001318
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001319 if (mem_cgroup_disabled()) {
Johannes Weiner867e5e12019-11-30 17:55:34 -08001320 lruvec = &pgdat->__lruvec;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001321 goto out;
1322 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001323
Johannes Weiner1306a852014-12-10 15:44:52 -08001324 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001325 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001326 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -08001327 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -08001328 */
Johannes Weiner29833312014-12-10 15:44:02 -08001329 if (!memcg)
1330 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001331
Mel Gormanef8f2322016-07-28 15:46:05 -07001332 mz = mem_cgroup_page_nodeinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001333 lruvec = &mz->lruvec;
1334out:
1335 /*
1336 * Since a node can be onlined after the mem_cgroup was created,
1337 * we have to be prepared to initialize lruvec->zone here;
1338 * and if offlined then reonlined, we need to reinitialize it.
1339 */
Mel Gorman599d0c92016-07-28 15:45:31 -07001340 if (unlikely(lruvec->pgdat != pgdat))
1341 lruvec->pgdat = pgdat;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001342 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001343}
1344
1345/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001346 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1347 * @lruvec: mem_cgroup per zone lru vector
1348 * @lru: index of lru list the page is sitting on
Michal Hockob4536f0c82017-01-10 16:58:04 -08001349 * @zid: zone id of the accounted pages
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001350 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001351 *
Hugh Dickinsca707232016-05-19 17:12:35 -07001352 * This function must be called under lru_lock, just before a page is added
1353 * to or just after a page is removed from an lru list (that ordering being
1354 * so as to allow it to check that lru_size 0 is consistent with list_empty).
Johannes Weiner925b7672012-01-12 17:18:15 -08001355 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001356void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
Michal Hockob4536f0c82017-01-10 16:58:04 -08001357 int zid, int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001358{
Mel Gormanef8f2322016-07-28 15:46:05 -07001359 struct mem_cgroup_per_node *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001360 unsigned long *lru_size;
Hugh Dickinsca707232016-05-19 17:12:35 -07001361 long size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001362
1363 if (mem_cgroup_disabled())
1364 return;
1365
Mel Gormanef8f2322016-07-28 15:46:05 -07001366 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
Michal Hockob4536f0c82017-01-10 16:58:04 -08001367 lru_size = &mz->lru_zone_size[zid][lru];
Hugh Dickinsca707232016-05-19 17:12:35 -07001368
1369 if (nr_pages < 0)
1370 *lru_size += nr_pages;
1371
1372 size = *lru_size;
Michal Hockob4536f0c82017-01-10 16:58:04 -08001373 if (WARN_ONCE(size < 0,
1374 "%s(%p, %d, %d): lru_size %ld\n",
1375 __func__, lruvec, lru, nr_pages, size)) {
Hugh Dickinsca707232016-05-19 17:12:35 -07001376 VM_BUG_ON(1);
1377 *lru_size = 0;
1378 }
1379
1380 if (nr_pages > 0)
1381 *lru_size += nr_pages;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001382}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001383
Johannes Weiner19942822011-02-01 15:52:43 -08001384/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001385 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001386 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001387 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001388 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001389 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001390 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001391static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001392{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001393 unsigned long margin = 0;
1394 unsigned long count;
1395 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001396
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001397 count = page_counter_read(&memcg->memory);
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001398 limit = READ_ONCE(memcg->memory.max);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001399 if (count < limit)
1400 margin = limit - count;
1401
Johannes Weiner7941d212016-01-14 15:21:23 -08001402 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001403 count = page_counter_read(&memcg->memsw);
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001404 limit = READ_ONCE(memcg->memsw.max);
Kaixu Xia1c4448e2020-06-01 21:49:36 -07001405 if (count < limit)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001406 margin = min(margin, limit - count);
Li RongQingcbedbac2016-05-27 14:27:43 -07001407 else
1408 margin = 0;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001409 }
1410
1411 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001412}
1413
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001414/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001415 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001416 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001417 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1418 * moving cgroups. This is for waiting at high-memory pressure
1419 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001420 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001421static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001422{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001423 struct mem_cgroup *from;
1424 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001425 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001426 /*
1427 * Unlike task_move routines, we access mc.to, mc.from not under
1428 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1429 */
1430 spin_lock(&mc.lock);
1431 from = mc.from;
1432 to = mc.to;
1433 if (!from)
1434 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001435
Johannes Weiner2314b422014-12-10 15:44:33 -08001436 ret = mem_cgroup_is_descendant(from, memcg) ||
1437 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001438unlock:
1439 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001440 return ret;
1441}
1442
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001443static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001444{
1445 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001446 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001447 DEFINE_WAIT(wait);
1448 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1449 /* moving charge context might have finished. */
1450 if (mc.moving_task)
1451 schedule();
1452 finish_wait(&mc.waitq, &wait);
1453 return true;
1454 }
1455 }
1456 return false;
1457}
1458
Johannes Weinerc8713d02019-07-11 20:55:59 -07001459static char *memory_stat_format(struct mem_cgroup *memcg)
1460{
1461 struct seq_buf s;
1462 int i;
Johannes Weiner71cd3112017-05-03 14:55:13 -07001463
Johannes Weinerc8713d02019-07-11 20:55:59 -07001464 seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE);
1465 if (!s.buffer)
1466 return NULL;
1467
1468 /*
1469 * Provide statistics on the state of the memory subsystem as
1470 * well as cumulative event counters that show past behavior.
1471 *
1472 * This list is ordered following a combination of these gradients:
1473 * 1) generic big picture -> specifics and details
1474 * 2) reflecting userspace activity -> reflecting kernel heuristics
1475 *
1476 * Current memory state:
1477 */
1478
1479 seq_buf_printf(&s, "anon %llu\n",
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07001480 (u64)memcg_page_state(memcg, NR_ANON_MAPPED) *
Johannes Weinerc8713d02019-07-11 20:55:59 -07001481 PAGE_SIZE);
1482 seq_buf_printf(&s, "file %llu\n",
Johannes Weiner0d1c2072020-06-03 16:01:54 -07001483 (u64)memcg_page_state(memcg, NR_FILE_PAGES) *
Johannes Weinerc8713d02019-07-11 20:55:59 -07001484 PAGE_SIZE);
1485 seq_buf_printf(&s, "kernel_stack %llu\n",
Shakeel Butt991e7672020-08-06 23:21:37 -07001486 (u64)memcg_page_state(memcg, NR_KERNEL_STACK_KB) *
Johannes Weinerc8713d02019-07-11 20:55:59 -07001487 1024);
1488 seq_buf_printf(&s, "slab %llu\n",
Roman Gushchind42f3242020-08-06 23:20:39 -07001489 (u64)(memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B) +
1490 memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B)));
Johannes Weinerc8713d02019-07-11 20:55:59 -07001491 seq_buf_printf(&s, "sock %llu\n",
1492 (u64)memcg_page_state(memcg, MEMCG_SOCK) *
1493 PAGE_SIZE);
1494
1495 seq_buf_printf(&s, "shmem %llu\n",
1496 (u64)memcg_page_state(memcg, NR_SHMEM) *
1497 PAGE_SIZE);
1498 seq_buf_printf(&s, "file_mapped %llu\n",
1499 (u64)memcg_page_state(memcg, NR_FILE_MAPPED) *
1500 PAGE_SIZE);
1501 seq_buf_printf(&s, "file_dirty %llu\n",
1502 (u64)memcg_page_state(memcg, NR_FILE_DIRTY) *
1503 PAGE_SIZE);
1504 seq_buf_printf(&s, "file_writeback %llu\n",
1505 (u64)memcg_page_state(memcg, NR_WRITEBACK) *
1506 PAGE_SIZE);
1507
Johannes Weiner468c3982020-06-03 16:02:01 -07001508#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Johannes Weinerc8713d02019-07-11 20:55:59 -07001509 seq_buf_printf(&s, "anon_thp %llu\n",
Johannes Weiner468c3982020-06-03 16:02:01 -07001510 (u64)memcg_page_state(memcg, NR_ANON_THPS) *
1511 HPAGE_PMD_SIZE);
1512#endif
Johannes Weinerc8713d02019-07-11 20:55:59 -07001513
1514 for (i = 0; i < NR_LRU_LISTS; i++)
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001515 seq_buf_printf(&s, "%s %llu\n", lru_list_name(i),
Johannes Weinerc8713d02019-07-11 20:55:59 -07001516 (u64)memcg_page_state(memcg, NR_LRU_BASE + i) *
1517 PAGE_SIZE);
1518
1519 seq_buf_printf(&s, "slab_reclaimable %llu\n",
Roman Gushchind42f3242020-08-06 23:20:39 -07001520 (u64)memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B));
Johannes Weinerc8713d02019-07-11 20:55:59 -07001521 seq_buf_printf(&s, "slab_unreclaimable %llu\n",
Roman Gushchind42f3242020-08-06 23:20:39 -07001522 (u64)memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B));
Johannes Weinerc8713d02019-07-11 20:55:59 -07001523
1524 /* Accumulated memory events */
1525
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001526 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT),
1527 memcg_events(memcg, PGFAULT));
1528 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT),
1529 memcg_events(memcg, PGMAJFAULT));
Johannes Weinerc8713d02019-07-11 20:55:59 -07001530
1531 seq_buf_printf(&s, "workingset_refault %lu\n",
1532 memcg_page_state(memcg, WORKINGSET_REFAULT));
1533 seq_buf_printf(&s, "workingset_activate %lu\n",
1534 memcg_page_state(memcg, WORKINGSET_ACTIVATE));
Yafang Shaoa6f55762020-06-01 21:49:32 -07001535 seq_buf_printf(&s, "workingset_restore %lu\n",
1536 memcg_page_state(memcg, WORKINGSET_RESTORE));
Johannes Weinerc8713d02019-07-11 20:55:59 -07001537 seq_buf_printf(&s, "workingset_nodereclaim %lu\n",
1538 memcg_page_state(memcg, WORKINGSET_NODERECLAIM));
1539
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001540 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL),
1541 memcg_events(memcg, PGREFILL));
Johannes Weinerc8713d02019-07-11 20:55:59 -07001542 seq_buf_printf(&s, "pgscan %lu\n",
1543 memcg_events(memcg, PGSCAN_KSWAPD) +
1544 memcg_events(memcg, PGSCAN_DIRECT));
1545 seq_buf_printf(&s, "pgsteal %lu\n",
1546 memcg_events(memcg, PGSTEAL_KSWAPD) +
1547 memcg_events(memcg, PGSTEAL_DIRECT));
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001548 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE),
1549 memcg_events(memcg, PGACTIVATE));
1550 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE),
1551 memcg_events(memcg, PGDEACTIVATE));
1552 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE),
1553 memcg_events(memcg, PGLAZYFREE));
1554 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED),
1555 memcg_events(memcg, PGLAZYFREED));
Johannes Weinerc8713d02019-07-11 20:55:59 -07001556
1557#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001558 seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC),
Johannes Weinerc8713d02019-07-11 20:55:59 -07001559 memcg_events(memcg, THP_FAULT_ALLOC));
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001560 seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC),
Johannes Weinerc8713d02019-07-11 20:55:59 -07001561 memcg_events(memcg, THP_COLLAPSE_ALLOC));
1562#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1563
1564 /* The above should easily fit into one page */
1565 WARN_ON_ONCE(seq_buf_has_overflowed(&s));
1566
1567 return s.buffer;
1568}
Johannes Weiner71cd3112017-05-03 14:55:13 -07001569
Sha Zhengju58cf1882013-02-22 16:32:05 -08001570#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001571/**
yuzhoujianf0c867d2018-12-28 00:36:10 -08001572 * mem_cgroup_print_oom_context: Print OOM information relevant to
1573 * memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001574 * @memcg: The memory cgroup that went over limit
1575 * @p: Task that is going to be killed
1576 *
1577 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1578 * enabled
1579 */
yuzhoujianf0c867d2018-12-28 00:36:10 -08001580void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p)
1581{
1582 rcu_read_lock();
1583
1584 if (memcg) {
1585 pr_cont(",oom_memcg=");
1586 pr_cont_cgroup_path(memcg->css.cgroup);
1587 } else
1588 pr_cont(",global_oom");
1589 if (p) {
1590 pr_cont(",task_memcg=");
1591 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1592 }
1593 rcu_read_unlock();
1594}
1595
1596/**
1597 * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to
1598 * memory controller.
1599 * @memcg: The memory cgroup that went over limit
1600 */
1601void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
Balbir Singhe2224322009-04-02 16:57:39 -07001602{
Johannes Weinerc8713d02019-07-11 20:55:59 -07001603 char *buf;
Balbir Singhe2224322009-04-02 16:57:39 -07001604
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001605 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1606 K((u64)page_counter_read(&memcg->memory)),
Chris Down15b42562020-04-01 21:07:20 -07001607 K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt);
Johannes Weinerc8713d02019-07-11 20:55:59 -07001608 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
1609 pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n",
1610 K((u64)page_counter_read(&memcg->swap)),
Chris Down32d087c2020-04-01 21:07:30 -07001611 K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt);
Johannes Weinerc8713d02019-07-11 20:55:59 -07001612 else {
1613 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1614 K((u64)page_counter_read(&memcg->memsw)),
1615 K((u64)memcg->memsw.max), memcg->memsw.failcnt);
1616 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1617 K((u64)page_counter_read(&memcg->kmem)),
1618 K((u64)memcg->kmem.max), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001619 }
Johannes Weinerc8713d02019-07-11 20:55:59 -07001620
1621 pr_info("Memory cgroup stats for ");
1622 pr_cont_cgroup_path(memcg->css.cgroup);
1623 pr_cont(":");
1624 buf = memory_stat_format(memcg);
1625 if (!buf)
1626 return;
1627 pr_info("%s", buf);
1628 kfree(buf);
Balbir Singhe2224322009-04-02 16:57:39 -07001629}
1630
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001631/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001632 * Return the memory (and swap, if configured) limit for a memcg.
1633 */
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001634unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001635{
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001636 unsigned long max;
David Rientjesa63d83f2010-08-09 17:19:46 -07001637
Chris Down15b42562020-04-01 21:07:20 -07001638 max = READ_ONCE(memcg->memory.max);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001639 if (mem_cgroup_swappiness(memcg)) {
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001640 unsigned long memsw_max;
1641 unsigned long swap_max;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001642
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001643 memsw_max = memcg->memsw.max;
Chris Down32d087c2020-04-01 21:07:30 -07001644 swap_max = READ_ONCE(memcg->swap.max);
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001645 swap_max = min(swap_max, (unsigned long)total_swap_pages);
1646 max = min(max + swap_max, memsw_max);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001647 }
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001648 return max;
David Rientjesa63d83f2010-08-09 17:19:46 -07001649}
1650
Chris Down9783aa92019-10-06 17:58:32 -07001651unsigned long mem_cgroup_size(struct mem_cgroup *memcg)
1652{
1653 return page_counter_read(&memcg->memory);
1654}
1655
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001656static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001657 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001658{
David Rientjes6e0fc462015-09-08 15:00:36 -07001659 struct oom_control oc = {
1660 .zonelist = NULL,
1661 .nodemask = NULL,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001662 .memcg = memcg,
David Rientjes6e0fc462015-09-08 15:00:36 -07001663 .gfp_mask = gfp_mask,
1664 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001665 };
Yafang Shao1378b372020-08-06 23:22:08 -07001666 bool ret = true;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001667
Tetsuo Handa7775fac2019-03-05 15:46:47 -08001668 if (mutex_lock_killable(&oom_lock))
1669 return true;
Yafang Shao1378b372020-08-06 23:22:08 -07001670
1671 if (mem_cgroup_margin(memcg) >= (1 << order))
1672 goto unlock;
1673
Tetsuo Handa7775fac2019-03-05 15:46:47 -08001674 /*
1675 * A few threads which were not waiting at mutex_lock_killable() can
1676 * fail to bail out. Therefore, check again after holding oom_lock.
1677 */
1678 ret = should_force_charge() || out_of_memory(&oc);
Yafang Shao1378b372020-08-06 23:22:08 -07001679
1680unlock:
Johannes Weinerdc564012015-06-24 16:57:19 -07001681 mutex_unlock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001682 return ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001683}
1684
Andrew Morton0608f432013-09-24 15:27:41 -07001685static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
Mel Gormanef8f2322016-07-28 15:46:05 -07001686 pg_data_t *pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001687 gfp_t gfp_mask,
1688 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001689{
Andrew Morton0608f432013-09-24 15:27:41 -07001690 struct mem_cgroup *victim = NULL;
1691 int total = 0;
1692 int loop = 0;
1693 unsigned long excess;
1694 unsigned long nr_scanned;
1695 struct mem_cgroup_reclaim_cookie reclaim = {
Mel Gormanef8f2322016-07-28 15:46:05 -07001696 .pgdat = pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001697 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001698
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001699 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001700
Andrew Morton0608f432013-09-24 15:27:41 -07001701 while (1) {
1702 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1703 if (!victim) {
1704 loop++;
1705 if (loop >= 2) {
1706 /*
1707 * If we have not been able to reclaim
1708 * anything, it might because there are
1709 * no reclaimable pages under this hierarchy
1710 */
1711 if (!total)
1712 break;
1713 /*
1714 * We want to do more targeted reclaim.
1715 * excess >> 2 is not to excessive so as to
1716 * reclaim too much, nor too less that we keep
1717 * coming back to reclaim from this cgroup
1718 */
1719 if (total >= (excess >> 2) ||
1720 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1721 break;
1722 }
1723 continue;
1724 }
Mel Gormana9dd0a82016-07-28 15:46:02 -07001725 total += mem_cgroup_shrink_node(victim, gfp_mask, false,
Mel Gormanef8f2322016-07-28 15:46:05 -07001726 pgdat, &nr_scanned);
Andrew Morton0608f432013-09-24 15:27:41 -07001727 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001728 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001729 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001730 }
Andrew Morton0608f432013-09-24 15:27:41 -07001731 mem_cgroup_iter_break(root_memcg, victim);
1732 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001733}
1734
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001735#ifdef CONFIG_LOCKDEP
1736static struct lockdep_map memcg_oom_lock_dep_map = {
1737 .name = "memcg_oom_lock",
1738};
1739#endif
1740
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001741static DEFINE_SPINLOCK(memcg_oom_lock);
1742
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001743/*
1744 * Check OOM-Killer is already running under our hierarchy.
1745 * If someone is running, return false.
1746 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001747static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001748{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001749 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001750
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001751 spin_lock(&memcg_oom_lock);
1752
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001753 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001754 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001755 /*
1756 * this subtree of our hierarchy is already locked
1757 * so we cannot give a lock.
1758 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001759 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001760 mem_cgroup_iter_break(memcg, iter);
1761 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001762 } else
1763 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001764 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001765
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001766 if (failed) {
1767 /*
1768 * OK, we failed to lock the whole subtree so we have
1769 * to clean up what we set up to the failing subtree
1770 */
1771 for_each_mem_cgroup_tree(iter, memcg) {
1772 if (iter == failed) {
1773 mem_cgroup_iter_break(memcg, iter);
1774 break;
1775 }
1776 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001777 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001778 } else
1779 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001780
1781 spin_unlock(&memcg_oom_lock);
1782
1783 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001784}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001785
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001786static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001787{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001788 struct mem_cgroup *iter;
1789
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001790 spin_lock(&memcg_oom_lock);
Qian Cai5facae42019-09-19 12:09:40 -04001791 mutex_release(&memcg_oom_lock_dep_map, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001792 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001793 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001794 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001795}
1796
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001797static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001798{
1799 struct mem_cgroup *iter;
1800
Tejun Heoc2b42d32015-06-24 16:58:23 -07001801 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001802 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001803 iter->under_oom++;
1804 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001805}
1806
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001807static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001808{
1809 struct mem_cgroup *iter;
1810
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001811 /*
1812 * When a new child is created while the hierarchy is under oom,
Tejun Heoc2b42d32015-06-24 16:58:23 -07001813 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001814 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001815 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001816 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001817 if (iter->under_oom > 0)
1818 iter->under_oom--;
1819 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001820}
1821
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001822static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1823
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001824struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001825 struct mem_cgroup *memcg;
Ingo Molnarac6424b2017-06-20 12:06:13 +02001826 wait_queue_entry_t wait;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001827};
1828
Ingo Molnarac6424b2017-06-20 12:06:13 +02001829static int memcg_oom_wake_function(wait_queue_entry_t *wait,
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001830 unsigned mode, int sync, void *arg)
1831{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001832 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1833 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001834 struct oom_wait_info *oom_wait_info;
1835
1836 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001837 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001838
Johannes Weiner2314b422014-12-10 15:44:33 -08001839 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1840 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001841 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001842 return autoremove_wake_function(wait, mode, sync, arg);
1843}
1844
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001845static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001846{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001847 /*
1848 * For the following lockless ->under_oom test, the only required
1849 * guarantee is that it must see the state asserted by an OOM when
1850 * this function is called as a result of userland actions
1851 * triggered by the notification of the OOM. This is trivially
1852 * achieved by invoking mem_cgroup_mark_under_oom() before
1853 * triggering notification.
1854 */
1855 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001856 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001857}
1858
Michal Hocko29ef6802018-08-17 15:47:11 -07001859enum oom_status {
1860 OOM_SUCCESS,
1861 OOM_FAILED,
1862 OOM_ASYNC,
1863 OOM_SKIPPED
1864};
1865
1866static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001867{
Michal Hocko7056d3a2018-12-28 00:39:57 -08001868 enum oom_status ret;
1869 bool locked;
1870
Michal Hocko29ef6802018-08-17 15:47:11 -07001871 if (order > PAGE_ALLOC_COSTLY_ORDER)
1872 return OOM_SKIPPED;
1873
Roman Gushchin7a1adfd2018-10-26 15:09:48 -07001874 memcg_memory_event(memcg, MEMCG_OOM);
1875
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001876 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001877 * We are in the middle of the charge context here, so we
1878 * don't want to block when potentially sitting on a callstack
1879 * that holds all kinds of filesystem and mm locks.
1880 *
Michal Hocko29ef6802018-08-17 15:47:11 -07001881 * cgroup1 allows disabling the OOM killer and waiting for outside
1882 * handling until the charge can succeed; remember the context and put
1883 * the task to sleep at the end of the page fault when all locks are
1884 * released.
Johannes Weiner49426422013-10-16 13:46:59 -07001885 *
Michal Hocko29ef6802018-08-17 15:47:11 -07001886 * On the other hand, in-kernel OOM killer allows for an async victim
1887 * memory reclaim (oom_reaper) and that means that we are not solely
1888 * relying on the oom victim to make a forward progress and we can
1889 * invoke the oom killer here.
1890 *
1891 * Please note that mem_cgroup_out_of_memory might fail to find a
1892 * victim and then we have to bail out from the charge path.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001893 */
Michal Hocko29ef6802018-08-17 15:47:11 -07001894 if (memcg->oom_kill_disable) {
1895 if (!current->in_user_fault)
1896 return OOM_SKIPPED;
1897 css_get(&memcg->css);
1898 current->memcg_in_oom = memcg;
1899 current->memcg_oom_gfp_mask = mask;
1900 current->memcg_oom_order = order;
1901
1902 return OOM_ASYNC;
1903 }
1904
Michal Hocko7056d3a2018-12-28 00:39:57 -08001905 mem_cgroup_mark_under_oom(memcg);
Michal Hocko29ef6802018-08-17 15:47:11 -07001906
Michal Hocko7056d3a2018-12-28 00:39:57 -08001907 locked = mem_cgroup_oom_trylock(memcg);
1908
1909 if (locked)
1910 mem_cgroup_oom_notify(memcg);
1911
1912 mem_cgroup_unmark_under_oom(memcg);
1913 if (mem_cgroup_out_of_memory(memcg, mask, order))
1914 ret = OOM_SUCCESS;
1915 else
1916 ret = OOM_FAILED;
1917
1918 if (locked)
1919 mem_cgroup_oom_unlock(memcg);
1920
1921 return ret;
Johannes Weiner49426422013-10-16 13:46:59 -07001922}
1923
1924/**
1925 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1926 * @handle: actually kill/wait or just clean up the OOM state
1927 *
1928 * This has to be called at the end of a page fault if the memcg OOM
1929 * handler was enabled.
1930 *
1931 * Memcg supports userspace OOM handling where failed allocations must
1932 * sleep on a waitqueue until the userspace task resolves the
1933 * situation. Sleeping directly in the charge context with all kinds
1934 * of locks held is not a good idea, instead we remember an OOM state
1935 * in the task and mem_cgroup_oom_synchronize() has to be called at
1936 * the end of the page fault to complete the OOM handling.
1937 *
1938 * Returns %true if an ongoing memcg OOM situation was detected and
1939 * completed, %false otherwise.
1940 */
1941bool mem_cgroup_oom_synchronize(bool handle)
1942{
Tejun Heo626ebc42015-11-05 18:46:09 -08001943 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001944 struct oom_wait_info owait;
1945 bool locked;
1946
1947 /* OOM is global, do not handle */
1948 if (!memcg)
1949 return false;
1950
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001951 if (!handle)
Johannes Weiner49426422013-10-16 13:46:59 -07001952 goto cleanup;
1953
1954 owait.memcg = memcg;
1955 owait.wait.flags = 0;
1956 owait.wait.func = memcg_oom_wake_function;
1957 owait.wait.private = current;
Ingo Molnar2055da92017-06-20 12:06:46 +02001958 INIT_LIST_HEAD(&owait.wait.entry);
Johannes Weiner49426422013-10-16 13:46:59 -07001959
1960 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001961 mem_cgroup_mark_under_oom(memcg);
1962
1963 locked = mem_cgroup_oom_trylock(memcg);
1964
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001965 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001966 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001967
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001968 if (locked && !memcg->oom_kill_disable) {
1969 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001970 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001971 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1972 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001973 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001974 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001975 mem_cgroup_unmark_under_oom(memcg);
1976 finish_wait(&memcg_oom_waitq, &owait.wait);
1977 }
1978
1979 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001980 mem_cgroup_oom_unlock(memcg);
1981 /*
1982 * There is no guarantee that an OOM-lock contender
1983 * sees the wakeups triggered by the OOM kill
1984 * uncharges. Wake any sleepers explicitely.
1985 */
1986 memcg_oom_recover(memcg);
1987 }
Johannes Weiner49426422013-10-16 13:46:59 -07001988cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001989 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001990 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001991 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001992}
1993
Johannes Weinerd7365e72014-10-29 14:50:48 -07001994/**
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07001995 * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM
1996 * @victim: task to be killed by the OOM killer
1997 * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM
1998 *
1999 * Returns a pointer to a memory cgroup, which has to be cleaned up
2000 * by killing all belonging OOM-killable tasks.
2001 *
2002 * Caller has to call mem_cgroup_put() on the returned non-NULL memcg.
2003 */
2004struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
2005 struct mem_cgroup *oom_domain)
2006{
2007 struct mem_cgroup *oom_group = NULL;
2008 struct mem_cgroup *memcg;
2009
2010 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2011 return NULL;
2012
2013 if (!oom_domain)
2014 oom_domain = root_mem_cgroup;
2015
2016 rcu_read_lock();
2017
2018 memcg = mem_cgroup_from_task(victim);
2019 if (memcg == root_mem_cgroup)
2020 goto out;
2021
2022 /*
Roman Gushchin48fe2672020-04-01 21:07:39 -07002023 * If the victim task has been asynchronously moved to a different
2024 * memory cgroup, we might end up killing tasks outside oom_domain.
2025 * In this case it's better to ignore memory.group.oom.
2026 */
2027 if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain)))
2028 goto out;
2029
2030 /*
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07002031 * Traverse the memory cgroup hierarchy from the victim task's
2032 * cgroup up to the OOMing cgroup (or root) to find the
2033 * highest-level memory cgroup with oom.group set.
2034 */
2035 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
2036 if (memcg->oom_group)
2037 oom_group = memcg;
2038
2039 if (memcg == oom_domain)
2040 break;
2041 }
2042
2043 if (oom_group)
2044 css_get(&oom_group->css);
2045out:
2046 rcu_read_unlock();
2047
2048 return oom_group;
2049}
2050
2051void mem_cgroup_print_oom_group(struct mem_cgroup *memcg)
2052{
2053 pr_info("Tasks in ");
2054 pr_cont_cgroup_path(memcg->css.cgroup);
2055 pr_cont(" are going to be killed due to memory.oom.group set\n");
2056}
2057
2058/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002059 * lock_page_memcg - lock a page->mem_cgroup binding
2060 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002061 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002062 * This function protects unlocked LRU pages from being moved to
Johannes Weiner739f79f2017-08-18 15:15:48 -07002063 * another cgroup.
2064 *
2065 * It ensures lifetime of the returned memcg. Caller is responsible
2066 * for the lifetime of the page; __unlock_page_memcg() is available
2067 * when @page might get freed inside the locked section.
Balbir Singhd69b0422009-06-17 16:26:34 -07002068 */
Johannes Weiner739f79f2017-08-18 15:15:48 -07002069struct mem_cgroup *lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002070{
Johannes Weiner9da7b522020-06-03 16:01:51 -07002071 struct page *head = compound_head(page); /* rmap on tail pages */
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002072 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08002073 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002074
Johannes Weiner6de22612015-02-11 15:25:01 -08002075 /*
2076 * The RCU lock is held throughout the transaction. The fast
2077 * path can get away without acquiring the memcg->move_lock
2078 * because page moving starts with an RCU grace period.
Johannes Weiner739f79f2017-08-18 15:15:48 -07002079 *
2080 * The RCU lock also protects the memcg from being freed when
2081 * the page state that is going to change is the only thing
2082 * preventing the page itself from being freed. E.g. writeback
2083 * doesn't hold a page reference and relies on PG_writeback to
2084 * keep off truncation, migration and so forth.
2085 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07002086 rcu_read_lock();
2087
2088 if (mem_cgroup_disabled())
Johannes Weiner739f79f2017-08-18 15:15:48 -07002089 return NULL;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002090again:
Johannes Weiner9da7b522020-06-03 16:01:51 -07002091 memcg = head->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08002092 if (unlikely(!memcg))
Johannes Weiner739f79f2017-08-18 15:15:48 -07002093 return NULL;
Johannes Weinerd7365e72014-10-29 14:50:48 -07002094
Qiang Huangbdcbb652014-06-04 16:08:21 -07002095 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner739f79f2017-08-18 15:15:48 -07002096 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002097
Johannes Weiner6de22612015-02-11 15:25:01 -08002098 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner9da7b522020-06-03 16:01:51 -07002099 if (memcg != head->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08002100 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002101 goto again;
2102 }
Johannes Weiner6de22612015-02-11 15:25:01 -08002103
2104 /*
2105 * When charge migration first begins, we can have locked and
2106 * unlocked page stat updates happening concurrently. Track
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002107 * the task who has the lock for unlock_page_memcg().
Johannes Weiner6de22612015-02-11 15:25:01 -08002108 */
2109 memcg->move_lock_task = current;
2110 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07002111
Johannes Weiner739f79f2017-08-18 15:15:48 -07002112 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002113}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002114EXPORT_SYMBOL(lock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002115
Johannes Weinerd7365e72014-10-29 14:50:48 -07002116/**
Johannes Weiner739f79f2017-08-18 15:15:48 -07002117 * __unlock_page_memcg - unlock and unpin a memcg
2118 * @memcg: the memcg
2119 *
2120 * Unlock and unpin a memcg returned by lock_page_memcg().
Johannes Weinerd7365e72014-10-29 14:50:48 -07002121 */
Johannes Weiner739f79f2017-08-18 15:15:48 -07002122void __unlock_page_memcg(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002123{
Johannes Weiner6de22612015-02-11 15:25:01 -08002124 if (memcg && memcg->move_lock_task == current) {
2125 unsigned long flags = memcg->move_lock_flags;
2126
2127 memcg->move_lock_task = NULL;
2128 memcg->move_lock_flags = 0;
2129
2130 spin_unlock_irqrestore(&memcg->move_lock, flags);
2131 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002132
Johannes Weinerd7365e72014-10-29 14:50:48 -07002133 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002134}
Johannes Weiner739f79f2017-08-18 15:15:48 -07002135
2136/**
2137 * unlock_page_memcg - unlock a page->mem_cgroup binding
2138 * @page: the page
2139 */
2140void unlock_page_memcg(struct page *page)
2141{
Johannes Weiner9da7b522020-06-03 16:01:51 -07002142 struct page *head = compound_head(page);
2143
2144 __unlock_page_memcg(head->mem_cgroup);
Johannes Weiner739f79f2017-08-18 15:15:48 -07002145}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002146EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002147
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002148struct memcg_stock_pcp {
2149 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002150 unsigned int nr_pages;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002151
2152#ifdef CONFIG_MEMCG_KMEM
2153 struct obj_cgroup *cached_objcg;
2154 unsigned int nr_bytes;
2155#endif
2156
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002157 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002158 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002159#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002160};
2161static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002162static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002163
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002164#ifdef CONFIG_MEMCG_KMEM
2165static void drain_obj_stock(struct memcg_stock_pcp *stock);
2166static bool obj_stock_flush_required(struct memcg_stock_pcp *stock,
2167 struct mem_cgroup *root_memcg);
2168
2169#else
2170static inline void drain_obj_stock(struct memcg_stock_pcp *stock)
2171{
2172}
2173static bool obj_stock_flush_required(struct memcg_stock_pcp *stock,
2174 struct mem_cgroup *root_memcg)
2175{
2176 return false;
2177}
2178#endif
2179
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002180/**
2181 * consume_stock: Try to consume stocked charge on this cpu.
2182 * @memcg: memcg to consume from.
2183 * @nr_pages: how many pages to charge.
2184 *
2185 * The charges will only happen if @memcg matches the current cpu's memcg
2186 * stock, and at least @nr_pages are available in that stock. Failure to
2187 * service an allocation will refill the stock.
2188 *
2189 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002190 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002191static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002192{
2193 struct memcg_stock_pcp *stock;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002194 unsigned long flags;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002195 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002196
Johannes Weinera983b5e2018-01-31 16:16:45 -08002197 if (nr_pages > MEMCG_CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002198 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002199
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002200 local_irq_save(flags);
2201
2202 stock = this_cpu_ptr(&memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002203 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002204 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002205 ret = true;
2206 }
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002207
2208 local_irq_restore(flags);
2209
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002210 return ret;
2211}
2212
2213/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002214 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002215 */
2216static void drain_stock(struct memcg_stock_pcp *stock)
2217{
2218 struct mem_cgroup *old = stock->cached;
2219
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07002220 if (!old)
2221 return;
2222
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002223 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002224 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002225 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002226 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002227 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002228 }
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07002229
2230 css_put(&old->css);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002231 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002232}
2233
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002234static void drain_local_stock(struct work_struct *dummy)
2235{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002236 struct memcg_stock_pcp *stock;
2237 unsigned long flags;
2238
Michal Hocko72f01842017-10-03 16:14:53 -07002239 /*
2240 * The only protection from memory hotplug vs. drain_stock races is
2241 * that we always operate on local CPU stock here with IRQ disabled
2242 */
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002243 local_irq_save(flags);
2244
2245 stock = this_cpu_ptr(&memcg_stock);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002246 drain_obj_stock(stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002247 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002248 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002249
2250 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002251}
2252
2253/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002254 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002255 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002256 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002257static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002258{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002259 struct memcg_stock_pcp *stock;
2260 unsigned long flags;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002261
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002262 local_irq_save(flags);
2263
2264 stock = this_cpu_ptr(&memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002265 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002266 drain_stock(stock);
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07002267 css_get(&memcg->css);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002268 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002269 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002270 stock->nr_pages += nr_pages;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002271
Johannes Weinera983b5e2018-01-31 16:16:45 -08002272 if (stock->nr_pages > MEMCG_CHARGE_BATCH)
Roman Gushchin475d0482017-09-08 16:13:09 -07002273 drain_stock(stock);
2274
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002275 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002276}
2277
2278/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002279 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002280 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002281 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002282static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002283{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002284 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002285
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002286 /* If someone's already draining, avoid adding running more workers. */
2287 if (!mutex_trylock(&percpu_charge_mutex))
2288 return;
Michal Hocko72f01842017-10-03 16:14:53 -07002289 /*
2290 * Notify other cpus that system-wide "drain" is running
2291 * We do not care about races with the cpu hotplug because cpu down
2292 * as well as workers from this path always operate on the local
2293 * per-cpu data. CPU up doesn't touch memcg_stock at all.
2294 */
Johannes Weiner5af12d02011-08-25 15:59:07 -07002295 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002296 for_each_online_cpu(cpu) {
2297 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002298 struct mem_cgroup *memcg;
Roman Gushchine1a366b2019-09-23 15:34:58 -07002299 bool flush = false;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002300
Roman Gushchine1a366b2019-09-23 15:34:58 -07002301 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002302 memcg = stock->cached;
Roman Gushchine1a366b2019-09-23 15:34:58 -07002303 if (memcg && stock->nr_pages &&
2304 mem_cgroup_is_descendant(memcg, root_memcg))
2305 flush = true;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002306 if (obj_stock_flush_required(stock, root_memcg))
2307 flush = true;
Roman Gushchine1a366b2019-09-23 15:34:58 -07002308 rcu_read_unlock();
2309
2310 if (flush &&
2311 !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
Michal Hockod1a05b62011-07-26 16:08:27 -07002312 if (cpu == curcpu)
2313 drain_local_stock(&stock->work);
2314 else
2315 schedule_work_on(cpu, &stock->work);
2316 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002317 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002318 put_cpu();
Michal Hocko9f50fad2011-08-09 11:56:26 +02002319 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002320}
2321
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01002322static int memcg_hotplug_cpu_dead(unsigned int cpu)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002323{
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002324 struct memcg_stock_pcp *stock;
Johannes Weiner42a30032019-05-14 15:47:12 -07002325 struct mem_cgroup *memcg, *mi;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002326
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002327 stock = &per_cpu(memcg_stock, cpu);
2328 drain_stock(stock);
Johannes Weinera983b5e2018-01-31 16:16:45 -08002329
2330 for_each_mem_cgroup(memcg) {
2331 int i;
2332
2333 for (i = 0; i < MEMCG_NR_STAT; i++) {
2334 int nid;
2335 long x;
2336
Chris Down871789d2019-05-14 15:46:57 -07002337 x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0);
Johannes Weiner815744d2019-06-13 15:55:46 -07002338 if (x)
Johannes Weiner42a30032019-05-14 15:47:12 -07002339 for (mi = memcg; mi; mi = parent_mem_cgroup(mi))
2340 atomic_long_add(x, &memcg->vmstats[i]);
Johannes Weinera983b5e2018-01-31 16:16:45 -08002341
2342 if (i >= NR_VM_NODE_STAT_ITEMS)
2343 continue;
2344
2345 for_each_node(nid) {
2346 struct mem_cgroup_per_node *pn;
2347
2348 pn = mem_cgroup_nodeinfo(memcg, nid);
2349 x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0);
Johannes Weiner815744d2019-06-13 15:55:46 -07002350 if (x)
Johannes Weiner42a30032019-05-14 15:47:12 -07002351 do {
2352 atomic_long_add(x, &pn->lruvec_stat[i]);
2353 } while ((pn = parent_nodeinfo(pn, nid)));
Johannes Weinera983b5e2018-01-31 16:16:45 -08002354 }
2355 }
2356
Johannes Weinere27be242018-04-10 16:29:45 -07002357 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
Johannes Weinera983b5e2018-01-31 16:16:45 -08002358 long x;
2359
Chris Down871789d2019-05-14 15:46:57 -07002360 x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0);
Johannes Weiner815744d2019-06-13 15:55:46 -07002361 if (x)
Johannes Weiner42a30032019-05-14 15:47:12 -07002362 for (mi = memcg; mi; mi = parent_mem_cgroup(mi))
2363 atomic_long_add(x, &memcg->vmevents[i]);
Johannes Weinera983b5e2018-01-31 16:16:45 -08002364 }
2365 }
2366
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01002367 return 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002368}
2369
Chris Downb3ff9292020-08-06 23:21:54 -07002370static unsigned long reclaim_high(struct mem_cgroup *memcg,
2371 unsigned int nr_pages,
2372 gfp_t gfp_mask)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002373{
Chris Downb3ff9292020-08-06 23:21:54 -07002374 unsigned long nr_reclaimed = 0;
2375
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002376 do {
Jakub Kicinskid1663a92020-06-01 21:49:49 -07002377 if (page_counter_read(&memcg->memory) <=
2378 READ_ONCE(memcg->memory.high))
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002379 continue;
Johannes Weinere27be242018-04-10 16:29:45 -07002380 memcg_memory_event(memcg, MEMCG_HIGH);
Chris Downb3ff9292020-08-06 23:21:54 -07002381 nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages,
2382 gfp_mask, true);
Chris Down4bf17302020-04-06 20:03:30 -07002383 } while ((memcg = parent_mem_cgroup(memcg)) &&
2384 !mem_cgroup_is_root(memcg));
Chris Downb3ff9292020-08-06 23:21:54 -07002385
2386 return nr_reclaimed;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002387}
2388
2389static void high_work_func(struct work_struct *work)
2390{
2391 struct mem_cgroup *memcg;
2392
2393 memcg = container_of(work, struct mem_cgroup, high_work);
Johannes Weinera983b5e2018-01-31 16:16:45 -08002394 reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002395}
2396
Tejun Heob23afb92015-11-05 18:46:11 -08002397/*
Chris Down0e4b01d2019-09-23 15:34:55 -07002398 * Clamp the maximum sleep time per allocation batch to 2 seconds. This is
2399 * enough to still cause a significant slowdown in most cases, while still
2400 * allowing diagnostics and tracing to proceed without becoming stuck.
2401 */
2402#define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ)
2403
2404/*
2405 * When calculating the delay, we use these either side of the exponentiation to
2406 * maintain precision and scale to a reasonable number of jiffies (see the table
2407 * below.
2408 *
2409 * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the
2410 * overage ratio to a delay.
2411 * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down down the
2412 * proposed penalty in order to reduce to a reasonable number of jiffies, and
2413 * to produce a reasonable delay curve.
2414 *
2415 * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a
2416 * reasonable delay curve compared to precision-adjusted overage, not
2417 * penalising heavily at first, but still making sure that growth beyond the
2418 * limit penalises misbehaviour cgroups by slowing them down exponentially. For
2419 * example, with a high of 100 megabytes:
2420 *
2421 * +-------+------------------------+
2422 * | usage | time to allocate in ms |
2423 * +-------+------------------------+
2424 * | 100M | 0 |
2425 * | 101M | 6 |
2426 * | 102M | 25 |
2427 * | 103M | 57 |
2428 * | 104M | 102 |
2429 * | 105M | 159 |
2430 * | 106M | 230 |
2431 * | 107M | 313 |
2432 * | 108M | 409 |
2433 * | 109M | 518 |
2434 * | 110M | 639 |
2435 * | 111M | 774 |
2436 * | 112M | 921 |
2437 * | 113M | 1081 |
2438 * | 114M | 1254 |
2439 * | 115M | 1439 |
2440 * | 116M | 1638 |
2441 * | 117M | 1849 |
2442 * | 118M | 2000 |
2443 * | 119M | 2000 |
2444 * | 120M | 2000 |
2445 * +-------+------------------------+
2446 */
2447 #define MEMCG_DELAY_PRECISION_SHIFT 20
2448 #define MEMCG_DELAY_SCALING_SHIFT 14
2449
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002450static u64 calculate_overage(unsigned long usage, unsigned long high)
2451{
2452 u64 overage;
2453
2454 if (usage <= high)
2455 return 0;
2456
2457 /*
2458 * Prevent division by 0 in overage calculation by acting as if
2459 * it was a threshold of 1 page
2460 */
2461 high = max(high, 1UL);
2462
2463 overage = usage - high;
2464 overage <<= MEMCG_DELAY_PRECISION_SHIFT;
2465 return div64_u64(overage, high);
2466}
2467
2468static u64 mem_find_max_overage(struct mem_cgroup *memcg)
2469{
2470 u64 overage, max_overage = 0;
2471
2472 do {
2473 overage = calculate_overage(page_counter_read(&memcg->memory),
Jakub Kicinskid1663a92020-06-01 21:49:49 -07002474 READ_ONCE(memcg->memory.high));
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002475 max_overage = max(overage, max_overage);
2476 } while ((memcg = parent_mem_cgroup(memcg)) &&
2477 !mem_cgroup_is_root(memcg));
2478
2479 return max_overage;
2480}
2481
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002482static u64 swap_find_max_overage(struct mem_cgroup *memcg)
2483{
2484 u64 overage, max_overage = 0;
2485
2486 do {
2487 overage = calculate_overage(page_counter_read(&memcg->swap),
2488 READ_ONCE(memcg->swap.high));
2489 if (overage)
2490 memcg_memory_event(memcg, MEMCG_SWAP_HIGH);
2491 max_overage = max(overage, max_overage);
2492 } while ((memcg = parent_mem_cgroup(memcg)) &&
2493 !mem_cgroup_is_root(memcg));
2494
2495 return max_overage;
2496}
2497
Chris Down0e4b01d2019-09-23 15:34:55 -07002498/*
Chris Downe26733e2020-03-21 18:22:23 -07002499 * Get the number of jiffies that we should penalise a mischievous cgroup which
2500 * is exceeding its memory.high by checking both it and its ancestors.
Tejun Heob23afb92015-11-05 18:46:11 -08002501 */
Chris Downe26733e2020-03-21 18:22:23 -07002502static unsigned long calculate_high_delay(struct mem_cgroup *memcg,
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002503 unsigned int nr_pages,
2504 u64 max_overage)
Tejun Heob23afb92015-11-05 18:46:11 -08002505{
Chris Downe26733e2020-03-21 18:22:23 -07002506 unsigned long penalty_jiffies;
Chris Downe26733e2020-03-21 18:22:23 -07002507
2508 if (!max_overage)
2509 return 0;
Chris Down0e4b01d2019-09-23 15:34:55 -07002510
2511 /*
Chris Down0e4b01d2019-09-23 15:34:55 -07002512 * We use overage compared to memory.high to calculate the number of
2513 * jiffies to sleep (penalty_jiffies). Ideally this value should be
2514 * fairly lenient on small overages, and increasingly harsh when the
2515 * memcg in question makes it clear that it has no intention of stopping
2516 * its crazy behaviour, so we exponentially increase the delay based on
2517 * overage amount.
2518 */
Chris Downe26733e2020-03-21 18:22:23 -07002519 penalty_jiffies = max_overage * max_overage * HZ;
2520 penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT;
2521 penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT;
Chris Down0e4b01d2019-09-23 15:34:55 -07002522
2523 /*
2524 * Factor in the task's own contribution to the overage, such that four
2525 * N-sized allocations are throttled approximately the same as one
2526 * 4N-sized allocation.
2527 *
2528 * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or
2529 * larger the current charge patch is than that.
2530 */
Jakub Kicinskiff144e62020-06-01 21:49:45 -07002531 return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH;
Chris Downe26733e2020-03-21 18:22:23 -07002532}
2533
2534/*
2535 * Scheduled by try_charge() to be executed from the userland return path
2536 * and reclaims memory over the high limit.
2537 */
2538void mem_cgroup_handle_over_high(void)
2539{
2540 unsigned long penalty_jiffies;
2541 unsigned long pflags;
Chris Downb3ff9292020-08-06 23:21:54 -07002542 unsigned long nr_reclaimed;
Chris Downe26733e2020-03-21 18:22:23 -07002543 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Chris Downd977aa92020-08-06 23:21:58 -07002544 int nr_retries = MAX_RECLAIM_RETRIES;
Chris Downe26733e2020-03-21 18:22:23 -07002545 struct mem_cgroup *memcg;
Chris Downb3ff9292020-08-06 23:21:54 -07002546 bool in_retry = false;
Chris Downe26733e2020-03-21 18:22:23 -07002547
2548 if (likely(!nr_pages))
2549 return;
2550
2551 memcg = get_mem_cgroup_from_mm(current->mm);
Chris Downe26733e2020-03-21 18:22:23 -07002552 current->memcg_nr_pages_over_high = 0;
2553
Chris Downb3ff9292020-08-06 23:21:54 -07002554retry_reclaim:
2555 /*
2556 * The allocating task should reclaim at least the batch size, but for
2557 * subsequent retries we only want to do what's necessary to prevent oom
2558 * or breaching resource isolation.
2559 *
2560 * This is distinct from memory.max or page allocator behaviour because
2561 * memory.high is currently batched, whereas memory.max and the page
2562 * allocator run every time an allocation is made.
2563 */
2564 nr_reclaimed = reclaim_high(memcg,
2565 in_retry ? SWAP_CLUSTER_MAX : nr_pages,
2566 GFP_KERNEL);
2567
Chris Downe26733e2020-03-21 18:22:23 -07002568 /*
2569 * memory.high is breached and reclaim is unable to keep up. Throttle
2570 * allocators proactively to slow down excessive growth.
2571 */
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002572 penalty_jiffies = calculate_high_delay(memcg, nr_pages,
2573 mem_find_max_overage(memcg));
Chris Down0e4b01d2019-09-23 15:34:55 -07002574
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002575 penalty_jiffies += calculate_high_delay(memcg, nr_pages,
2576 swap_find_max_overage(memcg));
2577
Chris Down0e4b01d2019-09-23 15:34:55 -07002578 /*
Jakub Kicinskiff144e62020-06-01 21:49:45 -07002579 * Clamp the max delay per usermode return so as to still keep the
2580 * application moving forwards and also permit diagnostics, albeit
2581 * extremely slowly.
2582 */
2583 penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES);
2584
2585 /*
Chris Down0e4b01d2019-09-23 15:34:55 -07002586 * Don't sleep if the amount of jiffies this memcg owes us is so low
2587 * that it's not even worth doing, in an attempt to be nice to those who
2588 * go only a small amount over their memory.high value and maybe haven't
2589 * been aggressively reclaimed enough yet.
2590 */
2591 if (penalty_jiffies <= HZ / 100)
2592 goto out;
2593
2594 /*
Chris Downb3ff9292020-08-06 23:21:54 -07002595 * If reclaim is making forward progress but we're still over
2596 * memory.high, we want to encourage that rather than doing allocator
2597 * throttling.
2598 */
2599 if (nr_reclaimed || nr_retries--) {
2600 in_retry = true;
2601 goto retry_reclaim;
2602 }
2603
2604 /*
Chris Down0e4b01d2019-09-23 15:34:55 -07002605 * If we exit early, we're guaranteed to die (since
2606 * schedule_timeout_killable sets TASK_KILLABLE). This means we don't
2607 * need to account for any ill-begotten jiffies to pay them off later.
2608 */
2609 psi_memstall_enter(&pflags);
2610 schedule_timeout_killable(penalty_jiffies);
2611 psi_memstall_leave(&pflags);
2612
2613out:
2614 css_put(&memcg->css);
Tejun Heob23afb92015-11-05 18:46:11 -08002615}
2616
Johannes Weiner00501b52014-08-08 14:19:20 -07002617static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
2618 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002619{
Johannes Weinera983b5e2018-01-31 16:16:45 -08002620 unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages);
Chris Downd977aa92020-08-06 23:21:58 -07002621 int nr_retries = MAX_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002622 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002623 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002624 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002625 bool may_swap = true;
2626 bool drained = false;
Michal Hocko29ef6802018-08-17 15:47:11 -07002627 enum oom_status oom_status;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002628
Johannes Weinerce00a962014-09-05 08:43:57 -04002629 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08002630 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002631retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07002632 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08002633 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002634
Johannes Weiner7941d212016-01-14 15:21:23 -08002635 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08002636 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
2637 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002638 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08002639 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002640 page_counter_uncharge(&memcg->memsw, batch);
2641 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002642 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002643 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002644 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002645 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002646
Johannes Weiner6539cc02014-08-06 16:05:42 -07002647 if (batch > nr_pages) {
2648 batch = nr_pages;
2649 goto retry;
2650 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002651
Johannes Weiner06b078f2014-08-06 16:05:44 -07002652 /*
Johannes Weiner869712f2019-11-05 21:17:13 -08002653 * Memcg doesn't have a dedicated reserve for atomic
2654 * allocations. But like the global atomic pool, we need to
2655 * put the burden of reclaim on regular allocation requests
2656 * and let these go through as privileged allocations.
2657 */
2658 if (gfp_mask & __GFP_ATOMIC)
2659 goto force;
2660
2661 /*
Johannes Weiner06b078f2014-08-06 16:05:44 -07002662 * Unlike in global OOM situations, memcg is not in a physical
2663 * memory shortage. Allow dying and OOM-killed tasks to
2664 * bypass the last charges so that they can exit quickly and
2665 * free their memory.
2666 */
Tetsuo Handa7775fac2019-03-05 15:46:47 -08002667 if (unlikely(should_force_charge()))
Tejun Heo10d53c72015-11-05 18:46:17 -08002668 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07002669
Johannes Weiner89a28482016-10-27 17:46:56 -07002670 /*
2671 * Prevent unbounded recursion when reclaim operations need to
2672 * allocate memory. This might exceed the limits temporarily,
2673 * but we prefer facilitating memory reclaim and getting back
2674 * under the limit over triggering OOM kills in these cases.
2675 */
2676 if (unlikely(current->flags & PF_MEMALLOC))
2677 goto force;
2678
Johannes Weiner06b078f2014-08-06 16:05:44 -07002679 if (unlikely(task_in_memcg_oom(current)))
2680 goto nomem;
2681
Mel Gormand0164ad2015-11-06 16:28:21 -08002682 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002683 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002684
Johannes Weinere27be242018-04-10 16:29:45 -07002685 memcg_memory_event(mem_over_limit, MEMCG_MAX);
Johannes Weiner241994ed2015-02-11 15:26:06 -08002686
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002687 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
2688 gfp_mask, may_swap);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002689
Johannes Weiner61e02c72014-08-06 16:08:16 -07002690 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07002691 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07002692
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002693 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002694 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002695 drained = true;
2696 goto retry;
2697 }
2698
Johannes Weiner28c34c22014-08-06 16:05:47 -07002699 if (gfp_mask & __GFP_NORETRY)
2700 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002701 /*
2702 * Even though the limit is exceeded at this point, reclaim
2703 * may have been able to free some pages. Retry the charge
2704 * before killing the task.
2705 *
2706 * Only for regular pages, though: huge pages are rather
2707 * unlikely to succeed so close to the limit, and we fall back
2708 * to regular pages anyway in case of failure.
2709 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002710 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002711 goto retry;
2712 /*
2713 * At task move, charge accounts can be doubly counted. So, it's
2714 * better to wait until the end of task_move if something is going on.
2715 */
2716 if (mem_cgroup_wait_acct_move(mem_over_limit))
2717 goto retry;
2718
Johannes Weiner9b130612014-08-06 16:05:51 -07002719 if (nr_retries--)
2720 goto retry;
2721
Shakeel Butt38d38492019-07-11 20:55:48 -07002722 if (gfp_mask & __GFP_RETRY_MAYFAIL)
Michal Hocko29ef6802018-08-17 15:47:11 -07002723 goto nomem;
2724
Johannes Weiner06b078f2014-08-06 16:05:44 -07002725 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08002726 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07002727
Johannes Weiner6539cc02014-08-06 16:05:42 -07002728 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08002729 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002730
Michal Hocko29ef6802018-08-17 15:47:11 -07002731 /*
2732 * keep retrying as long as the memcg oom killer is able to make
2733 * a forward progress or bypass the charge if the oom killer
2734 * couldn't make any progress.
2735 */
2736 oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask,
Jerome Marchand3608de02015-11-05 18:47:29 -08002737 get_order(nr_pages * PAGE_SIZE));
Michal Hocko29ef6802018-08-17 15:47:11 -07002738 switch (oom_status) {
2739 case OOM_SUCCESS:
Chris Downd977aa92020-08-06 23:21:58 -07002740 nr_retries = MAX_RECLAIM_RETRIES;
Michal Hocko29ef6802018-08-17 15:47:11 -07002741 goto retry;
2742 case OOM_FAILED:
2743 goto force;
2744 default:
2745 goto nomem;
2746 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002747nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002748 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002749 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08002750force:
2751 /*
2752 * The allocation either can't fail or will lead to more memory
2753 * being freed very soon. Allow memory usage go over the limit
2754 * temporarily by force charging it.
2755 */
2756 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002757 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08002758 page_counter_charge(&memcg->memsw, nr_pages);
Tejun Heo10d53c72015-11-05 18:46:17 -08002759
2760 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002761
2762done_restock:
2763 if (batch > nr_pages)
2764 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002765
Johannes Weiner241994ed2015-02-11 15:26:06 -08002766 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002767 * If the hierarchy is above the normal consumption range, schedule
2768 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002769 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002770 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2771 * not recorded as it most likely matches current's and won't
2772 * change in the meantime. As high limit is checked again before
2773 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994ed2015-02-11 15:26:06 -08002774 */
2775 do {
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002776 bool mem_high, swap_high;
2777
2778 mem_high = page_counter_read(&memcg->memory) >
2779 READ_ONCE(memcg->memory.high);
2780 swap_high = page_counter_read(&memcg->swap) >
2781 READ_ONCE(memcg->swap.high);
2782
2783 /* Don't bother a random interrupted task */
2784 if (in_interrupt()) {
2785 if (mem_high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002786 schedule_work(&memcg->high_work);
2787 break;
2788 }
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002789 continue;
2790 }
2791
2792 if (mem_high || swap_high) {
2793 /*
2794 * The allocating tasks in this cgroup will need to do
2795 * reclaim or be throttled to prevent further growth
2796 * of the memory or swap footprints.
2797 *
2798 * Target some best-effort fairness between the tasks,
2799 * and distribute reclaim work and delay penalties
2800 * based on how much each task is actually allocating.
2801 */
Vladimir Davydov9516a182015-12-11 13:40:24 -08002802 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002803 set_notify_resume(current);
2804 break;
2805 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08002806 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002807
2808 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002809}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002810
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07002811#if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU)
Johannes Weiner00501b52014-08-08 14:19:20 -07002812static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002813{
Johannes Weinerce00a962014-09-05 08:43:57 -04002814 if (mem_cgroup_is_root(memcg))
2815 return;
2816
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002817 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002818 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002819 page_counter_uncharge(&memcg->memsw, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002820}
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07002821#endif
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002822
Johannes Weinerd9eb1ea2020-06-03 16:02:24 -07002823static void commit_charge(struct page *page, struct mem_cgroup *memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002824{
Johannes Weiner1306a852014-12-10 15:44:52 -08002825 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002826 /*
Johannes Weinera0b5b412020-06-03 16:02:27 -07002827 * Any of the following ensures page->mem_cgroup stability:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002828 *
Johannes Weinera0b5b412020-06-03 16:02:27 -07002829 * - the page lock
2830 * - LRU isolation
2831 * - lock_page_memcg()
2832 * - exclusive reference
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002833 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002834 page->mem_cgroup = memcg;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002835}
2836
Kirill Tkhai84c07d12018-08-17 15:47:25 -07002837#ifdef CONFIG_MEMCG_KMEM
Roman Gushchin10befea2020-08-06 23:21:27 -07002838int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
2839 gfp_t gfp)
2840{
2841 unsigned int objects = objs_per_slab_page(s, page);
2842 void *vec;
2843
2844 vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp,
2845 page_to_nid(page));
2846 if (!vec)
2847 return -ENOMEM;
2848
2849 if (cmpxchg(&page->obj_cgroups, NULL,
2850 (struct obj_cgroup **) ((unsigned long)vec | 0x1UL)))
2851 kfree(vec);
2852 else
2853 kmemleak_not_leak(vec);
2854
2855 return 0;
2856}
2857
Roman Gushchin8380ce42020-03-28 19:17:25 -07002858/*
2859 * Returns a pointer to the memory cgroup to which the kernel object is charged.
2860 *
2861 * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(),
2862 * cgroup_mutex, etc.
2863 */
2864struct mem_cgroup *mem_cgroup_from_obj(void *p)
2865{
2866 struct page *page;
2867
2868 if (mem_cgroup_disabled())
2869 return NULL;
2870
2871 page = virt_to_head_page(p);
2872
2873 /*
Roman Gushchin98556092020-08-06 23:21:10 -07002874 * Slab objects are accounted individually, not per-page.
2875 * Memcg membership data for each individual object is saved in
2876 * the page->obj_cgroups.
Roman Gushchin8380ce42020-03-28 19:17:25 -07002877 */
Roman Gushchin98556092020-08-06 23:21:10 -07002878 if (page_has_obj_cgroups(page)) {
2879 struct obj_cgroup *objcg;
2880 unsigned int off;
2881
2882 off = obj_to_index(page->slab_cache, page, p);
2883 objcg = page_obj_cgroups(page)[off];
Roman Gushchin10befea2020-08-06 23:21:27 -07002884 if (objcg)
2885 return obj_cgroup_memcg(objcg);
2886
2887 return NULL;
Roman Gushchin98556092020-08-06 23:21:10 -07002888 }
Roman Gushchin8380ce42020-03-28 19:17:25 -07002889
2890 /* All other pages use page->mem_cgroup */
2891 return page->mem_cgroup;
2892}
2893
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002894__always_inline struct obj_cgroup *get_obj_cgroup_from_current(void)
2895{
2896 struct obj_cgroup *objcg = NULL;
2897 struct mem_cgroup *memcg;
2898
2899 if (unlikely(!current->mm && !current->active_memcg))
2900 return NULL;
2901
2902 rcu_read_lock();
2903 if (unlikely(current->active_memcg))
2904 memcg = rcu_dereference(current->active_memcg);
2905 else
2906 memcg = mem_cgroup_from_task(current);
2907
2908 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) {
2909 objcg = rcu_dereference(memcg->objcg);
2910 if (objcg && obj_cgroup_tryget(objcg))
2911 break;
2912 }
2913 rcu_read_unlock();
2914
2915 return objcg;
2916}
2917
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002918static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002919{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002920 int id, size;
2921 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002922
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002923 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002924 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2925 if (id < 0)
2926 return id;
2927
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002928 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002929 return id;
2930
2931 /*
2932 * There's no space for the new id in memcg_caches arrays,
2933 * so we have to grow them.
2934 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002935 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002936
2937 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002938 if (size < MEMCG_CACHES_MIN_SIZE)
2939 size = MEMCG_CACHES_MIN_SIZE;
2940 else if (size > MEMCG_CACHES_MAX_SIZE)
2941 size = MEMCG_CACHES_MAX_SIZE;
2942
Roman Gushchin98556092020-08-06 23:21:10 -07002943 err = memcg_update_all_list_lrus(size);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002944 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002945 memcg_nr_cache_ids = size;
2946
2947 up_write(&memcg_cache_ids_sem);
2948
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002949 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002950 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002951 return err;
2952 }
2953 return id;
2954}
2955
2956static void memcg_free_cache_id(int id)
2957{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002958 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002959}
2960
Vladimir Davydov45264772016-07-26 15:24:21 -07002961/**
Roman Gushchin4b13f642020-04-01 21:06:56 -07002962 * __memcg_kmem_charge: charge a number of kernel pages to a memcg
Roman Gushchin10eaec22020-04-01 21:06:39 -07002963 * @memcg: memory cgroup to charge
Vladimir Davydov45264772016-07-26 15:24:21 -07002964 * @gfp: reclaim mode
Roman Gushchin92d0510c2020-04-01 21:06:49 -07002965 * @nr_pages: number of pages to charge
Vladimir Davydov45264772016-07-26 15:24:21 -07002966 *
2967 * Returns 0 on success, an error code on failure.
2968 */
Roman Gushchin4b13f642020-04-01 21:06:56 -07002969int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp,
2970 unsigned int nr_pages)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002971{
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002972 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002973 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002974
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002975 ret = try_charge(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08002976 if (ret)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002977 return ret;
Johannes Weiner52c29b02016-01-20 15:02:35 -08002978
2979 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2980 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
Michal Hockoe55d9d92019-09-25 16:45:53 -07002981
2982 /*
2983 * Enforce __GFP_NOFAIL allocation because callers are not
2984 * prepared to see failures and likely do not have any failure
2985 * handling code.
2986 */
2987 if (gfp & __GFP_NOFAIL) {
2988 page_counter_charge(&memcg->kmem, nr_pages);
2989 return 0;
2990 }
Johannes Weiner52c29b02016-01-20 15:02:35 -08002991 cancel_charge(memcg, nr_pages);
2992 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002993 }
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002994 return 0;
2995}
2996
Vladimir Davydov45264772016-07-26 15:24:21 -07002997/**
Roman Gushchin4b13f642020-04-01 21:06:56 -07002998 * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg
2999 * @memcg: memcg to uncharge
3000 * @nr_pages: number of pages to uncharge
3001 */
3002void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages)
3003{
3004 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
3005 page_counter_uncharge(&memcg->kmem, nr_pages);
3006
3007 page_counter_uncharge(&memcg->memory, nr_pages);
3008 if (do_memsw_account())
3009 page_counter_uncharge(&memcg->memsw, nr_pages);
3010}
3011
3012/**
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003013 * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup
Vladimir Davydov45264772016-07-26 15:24:21 -07003014 * @page: page to charge
3015 * @gfp: reclaim mode
3016 * @order: allocation order
3017 *
3018 * Returns 0 on success, an error code on failure.
3019 */
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003020int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003021{
3022 struct mem_cgroup *memcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07003023 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003024
Shakeel Butt60cd4bc2019-03-05 15:43:13 -08003025 if (memcg_kmem_bypass())
Vladimir Davydov45264772016-07-26 15:24:21 -07003026 return 0;
3027
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07003028 memcg = get_mem_cgroup_from_current();
Vladimir Davydovc4159a72016-08-08 23:03:12 +03003029 if (!mem_cgroup_is_root(memcg)) {
Roman Gushchin4b13f642020-04-01 21:06:56 -07003030 ret = __memcg_kmem_charge(memcg, gfp, 1 << order);
Roman Gushchin4d96ba32019-07-11 20:56:31 -07003031 if (!ret) {
3032 page->mem_cgroup = memcg;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03003033 __SetPageKmemcg(page);
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07003034 return 0;
Roman Gushchin4d96ba32019-07-11 20:56:31 -07003035 }
Vladimir Davydovc4159a72016-08-08 23:03:12 +03003036 }
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003037 css_put(&memcg->css);
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08003038 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003039}
Roman Gushchin49a18ea2019-07-11 20:56:13 -07003040
3041/**
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003042 * __memcg_kmem_uncharge_page: uncharge a kmem page
Vladimir Davydov45264772016-07-26 15:24:21 -07003043 * @page: page to uncharge
3044 * @order: allocation order
3045 */
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003046void __memcg_kmem_uncharge_page(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003047{
Johannes Weiner1306a852014-12-10 15:44:52 -08003048 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003049 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003050
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003051 if (!memcg)
3052 return;
3053
Sasha Levin309381fea2014-01-23 15:52:54 -08003054 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Roman Gushchin4b13f642020-04-01 21:06:56 -07003055 __memcg_kmem_uncharge(memcg, nr_pages);
Johannes Weiner1306a852014-12-10 15:44:52 -08003056 page->mem_cgroup = NULL;
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07003057 css_put(&memcg->css);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03003058
3059 /* slab pages do not have PageKmemcg flag set */
3060 if (PageKmemcg(page))
3061 __ClearPageKmemcg(page);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08003062}
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003063
3064static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes)
3065{
3066 struct memcg_stock_pcp *stock;
3067 unsigned long flags;
3068 bool ret = false;
3069
3070 local_irq_save(flags);
3071
3072 stock = this_cpu_ptr(&memcg_stock);
3073 if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) {
3074 stock->nr_bytes -= nr_bytes;
3075 ret = true;
3076 }
3077
3078 local_irq_restore(flags);
3079
3080 return ret;
3081}
3082
3083static void drain_obj_stock(struct memcg_stock_pcp *stock)
3084{
3085 struct obj_cgroup *old = stock->cached_objcg;
3086
3087 if (!old)
3088 return;
3089
3090 if (stock->nr_bytes) {
3091 unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT;
3092 unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1);
3093
3094 if (nr_pages) {
3095 rcu_read_lock();
3096 __memcg_kmem_uncharge(obj_cgroup_memcg(old), nr_pages);
3097 rcu_read_unlock();
3098 }
3099
3100 /*
3101 * The leftover is flushed to the centralized per-memcg value.
3102 * On the next attempt to refill obj stock it will be moved
3103 * to a per-cpu stock (probably, on an other CPU), see
3104 * refill_obj_stock().
3105 *
3106 * How often it's flushed is a trade-off between the memory
3107 * limit enforcement accuracy and potential CPU contention,
3108 * so it might be changed in the future.
3109 */
3110 atomic_add(nr_bytes, &old->nr_charged_bytes);
3111 stock->nr_bytes = 0;
3112 }
3113
3114 obj_cgroup_put(old);
3115 stock->cached_objcg = NULL;
3116}
3117
3118static bool obj_stock_flush_required(struct memcg_stock_pcp *stock,
3119 struct mem_cgroup *root_memcg)
3120{
3121 struct mem_cgroup *memcg;
3122
3123 if (stock->cached_objcg) {
3124 memcg = obj_cgroup_memcg(stock->cached_objcg);
3125 if (memcg && mem_cgroup_is_descendant(memcg, root_memcg))
3126 return true;
3127 }
3128
3129 return false;
3130}
3131
3132static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes)
3133{
3134 struct memcg_stock_pcp *stock;
3135 unsigned long flags;
3136
3137 local_irq_save(flags);
3138
3139 stock = this_cpu_ptr(&memcg_stock);
3140 if (stock->cached_objcg != objcg) { /* reset if necessary */
3141 drain_obj_stock(stock);
3142 obj_cgroup_get(objcg);
3143 stock->cached_objcg = objcg;
3144 stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0);
3145 }
3146 stock->nr_bytes += nr_bytes;
3147
3148 if (stock->nr_bytes > PAGE_SIZE)
3149 drain_obj_stock(stock);
3150
3151 local_irq_restore(flags);
3152}
3153
3154int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size)
3155{
3156 struct mem_cgroup *memcg;
3157 unsigned int nr_pages, nr_bytes;
3158 int ret;
3159
3160 if (consume_obj_stock(objcg, size))
3161 return 0;
3162
3163 /*
3164 * In theory, memcg->nr_charged_bytes can have enough
3165 * pre-charged bytes to satisfy the allocation. However,
3166 * flushing memcg->nr_charged_bytes requires two atomic
3167 * operations, and memcg->nr_charged_bytes can't be big,
3168 * so it's better to ignore it and try grab some new pages.
3169 * memcg->nr_charged_bytes will be flushed in
3170 * refill_obj_stock(), called from this function or
3171 * independently later.
3172 */
3173 rcu_read_lock();
3174 memcg = obj_cgroup_memcg(objcg);
3175 css_get(&memcg->css);
3176 rcu_read_unlock();
3177
3178 nr_pages = size >> PAGE_SHIFT;
3179 nr_bytes = size & (PAGE_SIZE - 1);
3180
3181 if (nr_bytes)
3182 nr_pages += 1;
3183
3184 ret = __memcg_kmem_charge(memcg, gfp, nr_pages);
3185 if (!ret && nr_bytes)
3186 refill_obj_stock(objcg, PAGE_SIZE - nr_bytes);
3187
3188 css_put(&memcg->css);
3189 return ret;
3190}
3191
3192void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size)
3193{
3194 refill_obj_stock(objcg, size);
3195}
3196
Kirill Tkhai84c07d12018-08-17 15:47:25 -07003197#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003198
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003199#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3200
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003201/*
3202 * Because tail pages are not marked as "used", set it. We're under
Andrey Ryabininf4b7e272019-03-05 15:49:39 -08003203 * pgdat->lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003204 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003205void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003206{
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07003207 struct mem_cgroup *memcg = head->mem_cgroup;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003208 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003209
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08003210 if (mem_cgroup_disabled())
3211 return;
David Rientjesb070e652013-05-07 16:18:09 -07003212
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07003213 for (i = 1; i < HPAGE_PMD_NR; i++) {
3214 css_get(&memcg->css);
3215 head[i].mem_cgroup = memcg;
3216 }
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003217}
Hugh Dickins12d27102012-01-12 17:19:52 -08003218#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003219
Andrew Mortonc255a452012-07-31 16:43:02 -07003220#ifdef CONFIG_MEMCG_SWAP
Daisuke Nishimura02491442010-03-10 15:22:17 -08003221/**
3222 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3223 * @entry: swap entry to be moved
3224 * @from: mem_cgroup which the entry is moved from
3225 * @to: mem_cgroup which the entry is moved to
3226 *
3227 * It succeeds only when the swap_cgroup's record for this entry is the same
3228 * as the mem_cgroup's id of @from.
3229 *
3230 * Returns 0 on success, -EINVAL on failure.
3231 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003232 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08003233 * both res and memsw, and called css_get().
3234 */
3235static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003236 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003237{
3238 unsigned short old_id, new_id;
3239
Li Zefan34c00c32013-09-23 16:56:01 +08003240 old_id = mem_cgroup_id(from);
3241 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003242
3243 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Johannes Weinerc9019e92018-01-31 16:16:37 -08003244 mod_memcg_state(from, MEMCG_SWAP, -1);
3245 mod_memcg_state(to, MEMCG_SWAP, 1);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003246 return 0;
3247 }
3248 return -EINVAL;
3249}
3250#else
3251static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003252 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003253{
3254 return -EINVAL;
3255}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003256#endif
3257
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003258static DEFINE_MUTEX(memcg_max_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003259
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003260static int mem_cgroup_resize_max(struct mem_cgroup *memcg,
3261 unsigned long max, bool memsw)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003262{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003263 bool enlarge = false;
Shakeel Buttbb4a7ea2018-06-07 17:07:27 -07003264 bool drained = false;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003265 int ret;
Yu Zhaoc054a782018-01-31 16:20:02 -08003266 bool limits_invariant;
3267 struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003268
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003269 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003270 if (signal_pending(current)) {
3271 ret = -EINTR;
3272 break;
3273 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003274
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003275 mutex_lock(&memcg_max_mutex);
Yu Zhaoc054a782018-01-31 16:20:02 -08003276 /*
3277 * Make sure that the new limit (memsw or memory limit) doesn't
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003278 * break our basic invariant rule memory.max <= memsw.max.
Yu Zhaoc054a782018-01-31 16:20:02 -08003279 */
Chris Down15b42562020-04-01 21:07:20 -07003280 limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) :
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003281 max <= memcg->memsw.max;
Yu Zhaoc054a782018-01-31 16:20:02 -08003282 if (!limits_invariant) {
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003283 mutex_unlock(&memcg_max_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003284 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003285 break;
3286 }
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003287 if (max > counter->max)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003288 enlarge = true;
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003289 ret = page_counter_set_max(counter, max);
3290 mutex_unlock(&memcg_max_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003291
3292 if (!ret)
3293 break;
3294
Shakeel Buttbb4a7ea2018-06-07 17:07:27 -07003295 if (!drained) {
3296 drain_all_stock(memcg);
3297 drained = true;
3298 continue;
3299 }
3300
Andrey Ryabinin1ab5c052018-01-31 16:20:37 -08003301 if (!try_to_free_mem_cgroup_pages(memcg, 1,
3302 GFP_KERNEL, !memsw)) {
3303 ret = -EBUSY;
3304 break;
3305 }
3306 } while (true);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003307
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003308 if (!ret && enlarge)
3309 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003310
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003311 return ret;
3312}
3313
Mel Gormanef8f2322016-07-28 15:46:05 -07003314unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
Andrew Morton0608f432013-09-24 15:27:41 -07003315 gfp_t gfp_mask,
3316 unsigned long *total_scanned)
3317{
3318 unsigned long nr_reclaimed = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07003319 struct mem_cgroup_per_node *mz, *next_mz = NULL;
Andrew Morton0608f432013-09-24 15:27:41 -07003320 unsigned long reclaimed;
3321 int loop = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07003322 struct mem_cgroup_tree_per_node *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003323 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07003324 unsigned long nr_scanned;
3325
3326 if (order > 0)
3327 return 0;
3328
Mel Gormanef8f2322016-07-28 15:46:05 -07003329 mctz = soft_limit_tree_node(pgdat->node_id);
Michal Hockod6507ff2016-08-02 14:02:37 -07003330
3331 /*
3332 * Do not even bother to check the largest node if the root
3333 * is empty. Do it lockless to prevent lock bouncing. Races
3334 * are acceptable as soft limit is best effort anyway.
3335 */
Laurent Dufourbfc72282017-03-09 16:17:06 -08003336 if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root))
Michal Hockod6507ff2016-08-02 14:02:37 -07003337 return 0;
3338
Andrew Morton0608f432013-09-24 15:27:41 -07003339 /*
3340 * This loop can run a while, specially if mem_cgroup's continuously
3341 * keep exceeding their soft limit and putting the system under
3342 * pressure
3343 */
3344 do {
3345 if (next_mz)
3346 mz = next_mz;
3347 else
3348 mz = mem_cgroup_largest_soft_limit_node(mctz);
3349 if (!mz)
3350 break;
3351
3352 nr_scanned = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07003353 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07003354 gfp_mask, &nr_scanned);
3355 nr_reclaimed += reclaimed;
3356 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003357 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08003358 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07003359
3360 /*
3361 * If we failed to reclaim anything from this memory cgroup
3362 * it is time to move on to the next cgroup
3363 */
3364 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08003365 if (!reclaimed)
3366 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
3367
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003368 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07003369 /*
3370 * One school of thought says that we should not add
3371 * back the node to the tree if reclaim returns 0.
3372 * But our reclaim could return 0, simply because due
3373 * to priority we are exposing a smaller subset of
3374 * memory to reclaim from. Consider this as a longer
3375 * term TODO.
3376 */
3377 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07003378 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003379 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07003380 css_put(&mz->memcg->css);
3381 loop++;
3382 /*
3383 * Could not reclaim anything and there are no more
3384 * mem cgroups to try or we seem to be looping without
3385 * reclaiming anything.
3386 */
3387 if (!nr_reclaimed &&
3388 (next_mz == NULL ||
3389 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3390 break;
3391 } while (!nr_reclaimed);
3392 if (next_mz)
3393 css_put(&next_mz->memcg->css);
3394 return nr_reclaimed;
3395}
3396
Tejun Heoea280e72014-05-16 13:22:48 -04003397/*
3398 * Test whether @memcg has children, dead or alive. Note that this
3399 * function doesn't care whether @memcg has use_hierarchy enabled and
3400 * returns %true if there are child csses according to the cgroup
Ethon Paulb8f29352020-06-04 16:49:28 -07003401 * hierarchy. Testing use_hierarchy is the caller's responsibility.
Tejun Heoea280e72014-05-16 13:22:48 -04003402 */
Glauber Costab5f99b52013-02-22 16:34:53 -08003403static inline bool memcg_has_children(struct mem_cgroup *memcg)
3404{
Tejun Heoea280e72014-05-16 13:22:48 -04003405 bool ret;
3406
Tejun Heoea280e72014-05-16 13:22:48 -04003407 rcu_read_lock();
3408 ret = css_next_child(NULL, &memcg->css);
3409 rcu_read_unlock();
3410 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08003411}
3412
3413/*
Greg Thelen51038172016-05-20 16:58:18 -07003414 * Reclaims as many pages from the given memcg as possible.
Michal Hockoc26251f2012-10-26 13:37:28 +02003415 *
3416 * Caller is responsible for holding css reference for memcg.
3417 */
3418static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
3419{
Chris Downd977aa92020-08-06 23:21:58 -07003420 int nr_retries = MAX_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02003421
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003422 /* we call try-to-free pages for make this cgroup empty */
3423 lru_add_drain_all();
Junaid Shahidd12c60f2018-06-07 17:07:31 -07003424
3425 drain_all_stock(memcg);
3426
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003427 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003428 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003429 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003430
Michal Hockoc26251f2012-10-26 13:37:28 +02003431 if (signal_pending(current))
3432 return -EINTR;
3433
Johannes Weinerb70a2a22014-10-09 15:28:56 -07003434 progress = try_to_free_mem_cgroup_pages(memcg, 1,
3435 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003436 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003437 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003438 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003439 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003440 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003441
3442 }
Michal Hockoab5196c2012-10-26 13:37:32 +02003443
3444 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003445}
3446
Tejun Heo6770c642014-05-13 12:16:21 -04003447static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
3448 char *buf, size_t nbytes,
3449 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003450{
Tejun Heo6770c642014-05-13 12:16:21 -04003451 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02003452
Michal Hockod8423012012-10-26 13:37:29 +02003453 if (mem_cgroup_is_root(memcg))
3454 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04003455 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003456}
3457
Tejun Heo182446d2013-08-08 20:11:24 -04003458static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
3459 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08003460{
Tejun Heo182446d2013-08-08 20:11:24 -04003461 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003462}
3463
Tejun Heo182446d2013-08-08 20:11:24 -04003464static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
3465 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08003466{
3467 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04003468 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04003469 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003470
Glauber Costa567fb432012-07-31 16:43:07 -07003471 if (memcg->use_hierarchy == val)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003472 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07003473
Balbir Singh18f59ea2009-01-07 18:08:07 -08003474 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003475 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08003476 * in the child subtrees. If it is unset, then the change can
3477 * occur, provided the current cgroup has no children.
3478 *
3479 * For the root cgroup, parent_mem is NULL, we allow value to be
3480 * set if there are no children.
3481 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003482 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08003483 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04003484 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003485 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003486 else
3487 retval = -EBUSY;
3488 } else
3489 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07003490
Balbir Singh18f59ea2009-01-07 18:08:07 -08003491 return retval;
3492}
3493
Andrew Morton6f646152015-11-06 16:28:58 -08003494static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04003495{
Johannes Weiner42a30032019-05-14 15:47:12 -07003496 unsigned long val;
Johannes Weinerce00a962014-09-05 08:43:57 -04003497
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003498 if (mem_cgroup_is_root(memcg)) {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07003499 val = memcg_page_state(memcg, NR_FILE_PAGES) +
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07003500 memcg_page_state(memcg, NR_ANON_MAPPED);
Johannes Weiner42a30032019-05-14 15:47:12 -07003501 if (swap)
3502 val += memcg_page_state(memcg, MEMCG_SWAP);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003503 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04003504 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003505 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04003506 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003507 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04003508 }
Michal Hockoc12176d2015-11-05 18:50:29 -08003509 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04003510}
3511
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003512enum {
3513 RES_USAGE,
3514 RES_LIMIT,
3515 RES_MAX_USAGE,
3516 RES_FAILCNT,
3517 RES_SOFT_LIMIT,
3518};
Johannes Weinerce00a962014-09-05 08:43:57 -04003519
Tejun Heo791badb2013-12-05 12:28:02 -05003520static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07003521 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003522{
Tejun Heo182446d2013-08-08 20:11:24 -04003523 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003524 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07003525
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003526 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003527 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003528 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08003529 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003530 case _MEMSWAP:
3531 counter = &memcg->memsw;
3532 break;
3533 case _KMEM:
3534 counter = &memcg->kmem;
3535 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003536 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003537 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003538 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003539 default:
3540 BUG();
3541 }
3542
3543 switch (MEMFILE_ATTR(cft->private)) {
3544 case RES_USAGE:
3545 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08003546 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003547 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08003548 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003549 return (u64)page_counter_read(counter) * PAGE_SIZE;
3550 case RES_LIMIT:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003551 return (u64)counter->max * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003552 case RES_MAX_USAGE:
3553 return (u64)counter->watermark * PAGE_SIZE;
3554 case RES_FAILCNT:
3555 return counter->failcnt;
3556 case RES_SOFT_LIMIT:
3557 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003558 default:
3559 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003560 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003561}
Glauber Costa510fc4e2012-12-18 14:21:47 -08003562
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003563static void memcg_flush_percpu_vmstats(struct mem_cgroup *memcg)
Roman Gushchinc350a992019-08-24 17:54:47 -07003564{
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003565 unsigned long stat[MEMCG_NR_STAT] = {0};
Roman Gushchinc350a992019-08-24 17:54:47 -07003566 struct mem_cgroup *mi;
3567 int node, cpu, i;
Roman Gushchinc350a992019-08-24 17:54:47 -07003568
3569 for_each_online_cpu(cpu)
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003570 for (i = 0; i < MEMCG_NR_STAT; i++)
Shakeel Butt6c1c2802019-08-30 16:04:53 -07003571 stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu);
Roman Gushchinc350a992019-08-24 17:54:47 -07003572
3573 for (mi = memcg; mi; mi = parent_mem_cgroup(mi))
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003574 for (i = 0; i < MEMCG_NR_STAT; i++)
Roman Gushchinc350a992019-08-24 17:54:47 -07003575 atomic_long_add(stat[i], &mi->vmstats[i]);
3576
3577 for_each_node(node) {
3578 struct mem_cgroup_per_node *pn = memcg->nodeinfo[node];
3579 struct mem_cgroup_per_node *pi;
3580
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003581 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
Roman Gushchinc350a992019-08-24 17:54:47 -07003582 stat[i] = 0;
3583
3584 for_each_online_cpu(cpu)
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003585 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
Shakeel Butt6c1c2802019-08-30 16:04:53 -07003586 stat[i] += per_cpu(
3587 pn->lruvec_stat_cpu->count[i], cpu);
Roman Gushchinc350a992019-08-24 17:54:47 -07003588
3589 for (pi = pn; pi; pi = parent_nodeinfo(pi, node))
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003590 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
Roman Gushchinc350a992019-08-24 17:54:47 -07003591 atomic_long_add(stat[i], &pi->lruvec_stat[i]);
3592 }
3593}
3594
Roman Gushchinbb65f892019-08-24 17:54:50 -07003595static void memcg_flush_percpu_vmevents(struct mem_cgroup *memcg)
3596{
3597 unsigned long events[NR_VM_EVENT_ITEMS];
3598 struct mem_cgroup *mi;
3599 int cpu, i;
3600
3601 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
3602 events[i] = 0;
3603
3604 for_each_online_cpu(cpu)
3605 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
Shakeel Butt6c1c2802019-08-30 16:04:53 -07003606 events[i] += per_cpu(memcg->vmstats_percpu->events[i],
3607 cpu);
Roman Gushchinbb65f892019-08-24 17:54:50 -07003608
3609 for (mi = memcg; mi; mi = parent_mem_cgroup(mi))
3610 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
3611 atomic_long_add(events[i], &mi->vmevents[i]);
3612}
3613
Kirill Tkhai84c07d12018-08-17 15:47:25 -07003614#ifdef CONFIG_MEMCG_KMEM
Johannes Weiner567e9ab2016-01-20 15:02:24 -08003615static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08003616{
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003617 struct obj_cgroup *objcg;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003618 int memcg_id;
3619
Vladimir Davydovb313aee2016-03-17 14:18:27 -07003620 if (cgroup_memory_nokmem)
3621 return 0;
3622
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003623 BUG_ON(memcg->kmemcg_id >= 0);
Johannes Weiner567e9ab2016-01-20 15:02:24 -08003624 BUG_ON(memcg->kmem_state);
Vladimir Davydovd6441632014-01-23 15:53:09 -08003625
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07003626 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003627 if (memcg_id < 0)
3628 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003629
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003630 objcg = obj_cgroup_alloc();
3631 if (!objcg) {
3632 memcg_free_cache_id(memcg_id);
3633 return -ENOMEM;
3634 }
3635 objcg->memcg = memcg;
3636 rcu_assign_pointer(memcg->objcg, objcg);
3637
Roman Gushchind648bcc2020-08-06 23:20:28 -07003638 static_branch_enable(&memcg_kmem_enabled_key);
3639
Vladimir Davydovd6441632014-01-23 15:53:09 -08003640 /*
Johannes Weiner567e9ab2016-01-20 15:02:24 -08003641 * A memory cgroup is considered kmem-online as soon as it gets
Vladimir Davydov900a38f2014-12-12 16:55:10 -08003642 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08003643 * guarantee no one starts accounting before all call sites are
3644 * patched.
3645 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08003646 memcg->kmemcg_id = memcg_id;
Johannes Weiner567e9ab2016-01-20 15:02:24 -08003647 memcg->kmem_state = KMEM_ONLINE;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003648
3649 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003650}
3651
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003652static void memcg_offline_kmem(struct mem_cgroup *memcg)
3653{
3654 struct cgroup_subsys_state *css;
3655 struct mem_cgroup *parent, *child;
3656 int kmemcg_id;
3657
3658 if (memcg->kmem_state != KMEM_ONLINE)
3659 return;
Roman Gushchin98556092020-08-06 23:21:10 -07003660
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003661 memcg->kmem_state = KMEM_ALLOCATED;
3662
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003663 parent = parent_mem_cgroup(memcg);
3664 if (!parent)
3665 parent = root_mem_cgroup;
3666
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003667 memcg_reparent_objcgs(memcg, parent);
Roman Gushchinfb2f2b02019-07-11 20:56:34 -07003668
3669 kmemcg_id = memcg->kmemcg_id;
3670 BUG_ON(kmemcg_id < 0);
3671
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003672 /*
3673 * Change kmemcg_id of this cgroup and all its descendants to the
3674 * parent's id, and then move all entries from this cgroup's list_lrus
3675 * to ones of the parent. After we have finished, all list_lrus
3676 * corresponding to this cgroup are guaranteed to remain empty. The
3677 * ordering is imposed by list_lru_node->lock taken by
3678 * memcg_drain_all_list_lrus().
3679 */
Tejun Heo3a06bb72016-06-03 14:55:44 -07003680 rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003681 css_for_each_descendant_pre(css, &memcg->css) {
3682 child = mem_cgroup_from_css(css);
3683 BUG_ON(child->kmemcg_id != kmemcg_id);
3684 child->kmemcg_id = parent->kmemcg_id;
3685 if (!memcg->use_hierarchy)
3686 break;
3687 }
Tejun Heo3a06bb72016-06-03 14:55:44 -07003688 rcu_read_unlock();
3689
Kirill Tkhai9bec5c32018-08-17 15:47:58 -07003690 memcg_drain_all_list_lrus(kmemcg_id, parent);
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003691
3692 memcg_free_cache_id(kmemcg_id);
3693}
3694
3695static void memcg_free_kmem(struct mem_cgroup *memcg)
3696{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003697 /* css_alloc() failed, offlining didn't happen */
3698 if (unlikely(memcg->kmem_state == KMEM_ONLINE))
3699 memcg_offline_kmem(memcg);
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003700}
Vladimir Davydovd6441632014-01-23 15:53:09 -08003701#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003702static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08003703{
3704 return 0;
3705}
3706static void memcg_offline_kmem(struct mem_cgroup *memcg)
3707{
3708}
3709static void memcg_free_kmem(struct mem_cgroup *memcg)
3710{
3711}
Kirill Tkhai84c07d12018-08-17 15:47:25 -07003712#endif /* CONFIG_MEMCG_KMEM */
Johannes Weiner127424c2016-01-20 15:02:32 -08003713
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003714static int memcg_update_kmem_max(struct mem_cgroup *memcg,
3715 unsigned long max)
Johannes Weiner127424c2016-01-20 15:02:32 -08003716{
Vladimir Davydovb313aee2016-03-17 14:18:27 -07003717 int ret;
Johannes Weiner127424c2016-01-20 15:02:32 -08003718
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003719 mutex_lock(&memcg_max_mutex);
3720 ret = page_counter_set_max(&memcg->kmem, max);
3721 mutex_unlock(&memcg_max_mutex);
Johannes Weiner127424c2016-01-20 15:02:32 -08003722 return ret;
3723}
Glauber Costa510fc4e2012-12-18 14:21:47 -08003724
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003725static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003726{
3727 int ret;
3728
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003729 mutex_lock(&memcg_max_mutex);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003730
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003731 ret = page_counter_set_max(&memcg->tcpmem, max);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003732 if (ret)
3733 goto out;
3734
Johannes Weiner0db15292016-01-20 15:02:50 -08003735 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003736 /*
3737 * The active flag needs to be written after the static_key
3738 * update. This is what guarantees that the socket activation
Johannes Weiner2d758072016-10-07 17:00:58 -07003739 * function is the last one to run. See mem_cgroup_sk_alloc()
3740 * for details, and note that we don't mark any socket as
3741 * belonging to this memcg until that flag is up.
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003742 *
3743 * We need to do this, because static_keys will span multiple
3744 * sites, but we can't control their order. If we mark a socket
3745 * as accounted, but the accounting functions are not patched in
3746 * yet, we'll lose accounting.
3747 *
Johannes Weiner2d758072016-10-07 17:00:58 -07003748 * We never race with the readers in mem_cgroup_sk_alloc(),
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003749 * because when this value change, the code to process it is not
3750 * patched in yet.
3751 */
3752 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08003753 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003754 }
3755out:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003756 mutex_unlock(&memcg_max_mutex);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003757 return ret;
3758}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003759
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003760/*
3761 * The user of this function is...
3762 * RES_LIMIT.
3763 */
Tejun Heo451af502014-05-13 12:16:21 -04003764static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
3765 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003766{
Tejun Heo451af502014-05-13 12:16:21 -04003767 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003768 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003769 int ret;
3770
Tejun Heo451af502014-05-13 12:16:21 -04003771 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08003772 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003773 if (ret)
3774 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07003775
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003776 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003777 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003778 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3779 ret = -EINVAL;
3780 break;
3781 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003782 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3783 case _MEM:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003784 ret = mem_cgroup_resize_max(memcg, nr_pages, false);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003785 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003786 case _MEMSWAP:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003787 ret = mem_cgroup_resize_max(memcg, nr_pages, true);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003788 break;
3789 case _KMEM:
Michal Hocko01581152019-09-23 15:37:22 -07003790 pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. "
3791 "Please report your usecase to linux-mm@kvack.org if you "
3792 "depend on this functionality.\n");
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003793 ret = memcg_update_kmem_max(memcg, nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003794 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003795 case _TCP:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003796 ret = memcg_update_tcp_max(memcg, nr_pages);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003797 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003798 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003799 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003800 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003801 memcg->soft_limit = nr_pages;
3802 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003803 break;
3804 }
Tejun Heo451af502014-05-13 12:16:21 -04003805 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003806}
3807
Tejun Heo6770c642014-05-13 12:16:21 -04003808static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3809 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003810{
Tejun Heo6770c642014-05-13 12:16:21 -04003811 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003812 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003813
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003814 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3815 case _MEM:
3816 counter = &memcg->memory;
3817 break;
3818 case _MEMSWAP:
3819 counter = &memcg->memsw;
3820 break;
3821 case _KMEM:
3822 counter = &memcg->kmem;
3823 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003824 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003825 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003826 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003827 default:
3828 BUG();
3829 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003830
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003831 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003832 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003833 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003834 break;
3835 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003836 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003837 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003838 default:
3839 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003840 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003841
Tejun Heo6770c642014-05-13 12:16:21 -04003842 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003843}
3844
Tejun Heo182446d2013-08-08 20:11:24 -04003845static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003846 struct cftype *cft)
3847{
Tejun Heo182446d2013-08-08 20:11:24 -04003848 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003849}
3850
Daisuke Nishimura02491442010-03-10 15:22:17 -08003851#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003852static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003853 struct cftype *cft, u64 val)
3854{
Tejun Heo182446d2013-08-08 20:11:24 -04003855 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003856
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003857 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003858 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003859
Glauber Costaee5e8472013-02-22 16:34:50 -08003860 /*
3861 * No kind of locking is needed in here, because ->can_attach() will
3862 * check this value once in the beginning of the process, and then carry
3863 * on with stale data. This means that changes to this value will only
3864 * affect task migrations starting after the change.
3865 */
3866 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003867 return 0;
3868}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003869#else
Tejun Heo182446d2013-08-08 20:11:24 -04003870static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003871 struct cftype *cft, u64 val)
3872{
3873 return -ENOSYS;
3874}
3875#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003876
Ying Han406eb0c2011-05-26 16:25:37 -07003877#ifdef CONFIG_NUMA
Johannes Weiner113b7df2019-05-13 17:18:11 -07003878
3879#define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
3880#define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
3881#define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
3882
3883static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003884 int nid, unsigned int lru_mask, bool tree)
Johannes Weiner113b7df2019-05-13 17:18:11 -07003885{
Johannes Weiner867e5e12019-11-30 17:55:34 -08003886 struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid));
Johannes Weiner113b7df2019-05-13 17:18:11 -07003887 unsigned long nr = 0;
3888 enum lru_list lru;
3889
3890 VM_BUG_ON((unsigned)nid >= nr_node_ids);
3891
3892 for_each_lru(lru) {
3893 if (!(BIT(lru) & lru_mask))
3894 continue;
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003895 if (tree)
3896 nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru);
3897 else
3898 nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru);
Johannes Weiner113b7df2019-05-13 17:18:11 -07003899 }
3900 return nr;
3901}
3902
3903static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003904 unsigned int lru_mask,
3905 bool tree)
Johannes Weiner113b7df2019-05-13 17:18:11 -07003906{
3907 unsigned long nr = 0;
3908 enum lru_list lru;
3909
3910 for_each_lru(lru) {
3911 if (!(BIT(lru) & lru_mask))
3912 continue;
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003913 if (tree)
3914 nr += memcg_page_state(memcg, NR_LRU_BASE + lru);
3915 else
3916 nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru);
Johannes Weiner113b7df2019-05-13 17:18:11 -07003917 }
3918 return nr;
3919}
3920
Tejun Heo2da8ca82013-12-05 12:28:04 -05003921static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003922{
Greg Thelen25485de2013-11-12 15:07:40 -08003923 struct numa_stat {
3924 const char *name;
3925 unsigned int lru_mask;
3926 };
3927
3928 static const struct numa_stat stats[] = {
3929 { "total", LRU_ALL },
3930 { "file", LRU_ALL_FILE },
3931 { "anon", LRU_ALL_ANON },
3932 { "unevictable", BIT(LRU_UNEVICTABLE) },
3933 };
3934 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003935 int nid;
Chris Downaa9694b2019-03-05 15:45:52 -08003936 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Ying Han406eb0c2011-05-26 16:25:37 -07003937
Greg Thelen25485de2013-11-12 15:07:40 -08003938 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003939 seq_printf(m, "%s=%lu", stat->name,
3940 mem_cgroup_nr_lru_pages(memcg, stat->lru_mask,
3941 false));
3942 for_each_node_state(nid, N_MEMORY)
3943 seq_printf(m, " N%d=%lu", nid,
3944 mem_cgroup_node_nr_lru_pages(memcg, nid,
3945 stat->lru_mask, false));
Greg Thelen25485de2013-11-12 15:07:40 -08003946 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003947 }
Ying Han406eb0c2011-05-26 16:25:37 -07003948
Ying Han071aee12013-11-12 15:07:41 -08003949 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
Ying Han406eb0c2011-05-26 16:25:37 -07003950
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003951 seq_printf(m, "hierarchical_%s=%lu", stat->name,
3952 mem_cgroup_nr_lru_pages(memcg, stat->lru_mask,
3953 true));
3954 for_each_node_state(nid, N_MEMORY)
3955 seq_printf(m, " N%d=%lu", nid,
3956 mem_cgroup_node_nr_lru_pages(memcg, nid,
3957 stat->lru_mask, true));
Ying Han071aee12013-11-12 15:07:41 -08003958 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003959 }
Ying Han406eb0c2011-05-26 16:25:37 -07003960
Ying Han406eb0c2011-05-26 16:25:37 -07003961 return 0;
3962}
3963#endif /* CONFIG_NUMA */
3964
Johannes Weinerc8713d02019-07-11 20:55:59 -07003965static const unsigned int memcg1_stats[] = {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07003966 NR_FILE_PAGES,
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07003967 NR_ANON_MAPPED,
Johannes Weiner468c3982020-06-03 16:02:01 -07003968#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3969 NR_ANON_THPS,
3970#endif
Johannes Weinerc8713d02019-07-11 20:55:59 -07003971 NR_SHMEM,
3972 NR_FILE_MAPPED,
3973 NR_FILE_DIRTY,
3974 NR_WRITEBACK,
3975 MEMCG_SWAP,
3976};
3977
3978static const char *const memcg1_stat_names[] = {
3979 "cache",
3980 "rss",
Johannes Weiner468c3982020-06-03 16:02:01 -07003981#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Johannes Weinerc8713d02019-07-11 20:55:59 -07003982 "rss_huge",
Johannes Weiner468c3982020-06-03 16:02:01 -07003983#endif
Johannes Weinerc8713d02019-07-11 20:55:59 -07003984 "shmem",
3985 "mapped_file",
3986 "dirty",
3987 "writeback",
3988 "swap",
3989};
3990
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07003991/* Universal VM events cgroup1 shows, original sort order */
Greg Thelen8dd53fd2018-06-07 17:07:23 -07003992static const unsigned int memcg1_events[] = {
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07003993 PGPGIN,
3994 PGPGOUT,
3995 PGFAULT,
3996 PGMAJFAULT,
3997};
3998
Tejun Heo2da8ca82013-12-05 12:28:04 -05003999static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004000{
Chris Downaa9694b2019-03-05 15:45:52 -08004001 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004002 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004003 struct mem_cgroup *mi;
4004 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004005
Johannes Weiner71cd3112017-05-03 14:55:13 -07004006 BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats));
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08004007
Johannes Weiner71cd3112017-05-03 14:55:13 -07004008 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
Johannes Weiner468c3982020-06-03 16:02:01 -07004009 unsigned long nr;
4010
Johannes Weiner71cd3112017-05-03 14:55:13 -07004011 if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004012 continue;
Johannes Weiner468c3982020-06-03 16:02:01 -07004013 nr = memcg_page_state_local(memcg, memcg1_stats[i]);
4014#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4015 if (memcg1_stats[i] == NR_ANON_THPS)
4016 nr *= HPAGE_PMD_NR;
4017#endif
4018 seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004019 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004020
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07004021 for (i = 0; i < ARRAY_SIZE(memcg1_events); i++)
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08004022 seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]),
Johannes Weiner205b20c2019-05-14 15:47:06 -07004023 memcg_events_local(memcg, memcg1_events[i]));
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004024
4025 for (i = 0; i < NR_LRU_LISTS; i++)
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08004026 seq_printf(m, "%s %lu\n", lru_list_name(i),
Johannes Weiner205b20c2019-05-14 15:47:06 -07004027 memcg_page_state_local(memcg, NR_LRU_BASE + i) *
Johannes Weiner21d89d12019-05-13 17:18:08 -07004028 PAGE_SIZE);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004029
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004030 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004031 memory = memsw = PAGE_COUNTER_MAX;
4032 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
Chris Down15b42562020-04-01 21:07:20 -07004033 memory = min(memory, READ_ONCE(mi->memory.max));
4034 memsw = min(memsw, READ_ONCE(mi->memsw.max));
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004035 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004036 seq_printf(m, "hierarchical_memory_limit %llu\n",
4037 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08004038 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004039 seq_printf(m, "hierarchical_memsw_limit %llu\n",
4040 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004041
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07004042 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
Johannes Weiner71cd3112017-05-03 14:55:13 -07004043 if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004044 continue;
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07004045 seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i],
Yafang Shaodd923992019-07-11 20:52:11 -07004046 (u64)memcg_page_state(memcg, memcg1_stats[i]) *
4047 PAGE_SIZE);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004048 }
4049
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07004050 for (i = 0; i < ARRAY_SIZE(memcg1_events); i++)
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08004051 seq_printf(m, "total_%s %llu\n",
4052 vm_event_name(memcg1_events[i]),
Yafang Shaodd923992019-07-11 20:52:11 -07004053 (u64)memcg_events(memcg, memcg1_events[i]));
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004054
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07004055 for (i = 0; i < NR_LRU_LISTS; i++)
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08004056 seq_printf(m, "total_%s %llu\n", lru_list_name(i),
Johannes Weiner42a30032019-05-14 15:47:12 -07004057 (u64)memcg_page_state(memcg, NR_LRU_BASE + i) *
4058 PAGE_SIZE);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004059
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004060#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004061 {
Mel Gormanef8f2322016-07-28 15:46:05 -07004062 pg_data_t *pgdat;
4063 struct mem_cgroup_per_node *mz;
Johannes Weiner1431d4d2020-06-03 16:02:53 -07004064 unsigned long anon_cost = 0;
4065 unsigned long file_cost = 0;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004066
Mel Gormanef8f2322016-07-28 15:46:05 -07004067 for_each_online_pgdat(pgdat) {
4068 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004069
Johannes Weiner1431d4d2020-06-03 16:02:53 -07004070 anon_cost += mz->lruvec.anon_cost;
4071 file_cost += mz->lruvec.file_cost;
Mel Gormanef8f2322016-07-28 15:46:05 -07004072 }
Johannes Weiner1431d4d2020-06-03 16:02:53 -07004073 seq_printf(m, "anon_cost %lu\n", anon_cost);
4074 seq_printf(m, "file_cost %lu\n", file_cost);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004075 }
4076#endif
4077
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004078 return 0;
4079}
4080
Tejun Heo182446d2013-08-08 20:11:24 -04004081static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
4082 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004083{
Tejun Heo182446d2013-08-08 20:11:24 -04004084 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004085
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004086 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004087}
4088
Tejun Heo182446d2013-08-08 20:11:24 -04004089static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
4090 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004091{
Tejun Heo182446d2013-08-08 20:11:24 -04004092 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08004093
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07004094 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004095 return -EINVAL;
4096
Linus Torvalds14208b02014-06-09 15:03:33 -07004097 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07004098 memcg->swappiness = val;
4099 else
4100 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08004101
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004102 return 0;
4103}
4104
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004105static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4106{
4107 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004108 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004109 int i;
4110
4111 rcu_read_lock();
4112 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004113 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004114 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004115 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004116
4117 if (!t)
4118 goto unlock;
4119
Johannes Weinerce00a962014-09-05 08:43:57 -04004120 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004121
4122 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07004123 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004124 * If it's not true, a threshold was crossed after last
4125 * call of __mem_cgroup_threshold().
4126 */
Phil Carmody5407a562010-05-26 14:42:42 -07004127 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004128
4129 /*
4130 * Iterate backward over array of thresholds starting from
4131 * current_threshold and check if a threshold is crossed.
4132 * If none of thresholds below usage is crossed, we read
4133 * only one element of the array here.
4134 */
4135 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4136 eventfd_signal(t->entries[i].eventfd, 1);
4137
4138 /* i = current_threshold + 1 */
4139 i++;
4140
4141 /*
4142 * Iterate forward over array of thresholds starting from
4143 * current_threshold+1 and check if a threshold is crossed.
4144 * If none of thresholds above usage is crossed, we read
4145 * only one element of the array here.
4146 */
4147 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4148 eventfd_signal(t->entries[i].eventfd, 1);
4149
4150 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004151 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004152unlock:
4153 rcu_read_unlock();
4154}
4155
4156static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4157{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004158 while (memcg) {
4159 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08004160 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004161 __mem_cgroup_threshold(memcg, true);
4162
4163 memcg = parent_mem_cgroup(memcg);
4164 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004165}
4166
4167static int compare_thresholds(const void *a, const void *b)
4168{
4169 const struct mem_cgroup_threshold *_a = a;
4170 const struct mem_cgroup_threshold *_b = b;
4171
Greg Thelen2bff24a2013-09-11 14:23:08 -07004172 if (_a->threshold > _b->threshold)
4173 return 1;
4174
4175 if (_a->threshold < _b->threshold)
4176 return -1;
4177
4178 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004179}
4180
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004181static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004182{
4183 struct mem_cgroup_eventfd_list *ev;
4184
Michal Hocko2bcf2e92014-07-30 16:08:33 -07004185 spin_lock(&memcg_oom_lock);
4186
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004187 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004188 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07004189
4190 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004191 return 0;
4192}
4193
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004194static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004195{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004196 struct mem_cgroup *iter;
4197
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004198 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004199 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004200}
4201
Tejun Heo59b6f872013-11-22 18:20:43 -05004202static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004203 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004204{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004205 struct mem_cgroup_thresholds *thresholds;
4206 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004207 unsigned long threshold;
4208 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004209 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004210
Johannes Weiner650c5e52015-02-11 15:26:03 -08004211 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004212 if (ret)
4213 return ret;
4214
4215 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004216
Johannes Weiner05b84302014-08-06 16:05:59 -07004217 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004218 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004219 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07004220 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004221 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004222 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07004223 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004224 BUG();
4225
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004226 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004227 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004228 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4229
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004230 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004231
4232 /* Allocate memory for new array of thresholds */
Gustavo A. R. Silva67b80462019-03-05 15:44:05 -08004233 new = kmalloc(struct_size(new, entries, size), GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004234 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004235 ret = -ENOMEM;
4236 goto unlock;
4237 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004238 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004239
4240 /* Copy thresholds (if any) to new array */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004241 if (thresholds->primary) {
4242 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004243 sizeof(struct mem_cgroup_threshold));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004244 }
4245
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004246 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004247 new->entries[size - 1].eventfd = eventfd;
4248 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004249
4250 /* Sort thresholds. Registering of new threshold isn't time-critical */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004251 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004252 compare_thresholds, NULL);
4253
4254 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004255 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004256 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07004257 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004258 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004259 * new->current_threshold will not be used until
4260 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004261 * it here.
4262 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004263 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07004264 } else
4265 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004266 }
4267
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004268 /* Free old spare buffer and save old primary buffer as spare */
4269 kfree(thresholds->spare);
4270 thresholds->spare = thresholds->primary;
4271
4272 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004273
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004274 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004275 synchronize_rcu();
4276
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004277unlock:
4278 mutex_unlock(&memcg->thresholds_lock);
4279
4280 return ret;
4281}
4282
Tejun Heo59b6f872013-11-22 18:20:43 -05004283static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004284 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004285{
Tejun Heo59b6f872013-11-22 18:20:43 -05004286 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05004287}
4288
Tejun Heo59b6f872013-11-22 18:20:43 -05004289static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004290 struct eventfd_ctx *eventfd, const char *args)
4291{
Tejun Heo59b6f872013-11-22 18:20:43 -05004292 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05004293}
4294
Tejun Heo59b6f872013-11-22 18:20:43 -05004295static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004296 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004297{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004298 struct mem_cgroup_thresholds *thresholds;
4299 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004300 unsigned long usage;
Chunguang Xu7d366652020-03-21 18:22:10 -07004301 int i, j, size, entries;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004302
4303 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07004304
4305 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004306 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004307 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07004308 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004309 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004310 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07004311 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004312 BUG();
4313
Anton Vorontsov371528c2012-02-24 05:14:46 +04004314 if (!thresholds->primary)
4315 goto unlock;
4316
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004317 /* Check if a threshold crossed before removing */
4318 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4319
4320 /* Calculate new number of threshold */
Chunguang Xu7d366652020-03-21 18:22:10 -07004321 size = entries = 0;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004322 for (i = 0; i < thresholds->primary->size; i++) {
4323 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004324 size++;
Chunguang Xu7d366652020-03-21 18:22:10 -07004325 else
4326 entries++;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004327 }
4328
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004329 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004330
Chunguang Xu7d366652020-03-21 18:22:10 -07004331 /* If no items related to eventfd have been cleared, nothing to do */
4332 if (!entries)
4333 goto unlock;
4334
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004335 /* Set thresholds array to NULL if we don't have thresholds */
4336 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004337 kfree(new);
4338 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004339 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004340 }
4341
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004342 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004343
4344 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004345 new->current_threshold = -1;
4346 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4347 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004348 continue;
4349
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004350 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07004351 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004352 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004353 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004354 * until rcu_assign_pointer(), so it's safe to increment
4355 * it here.
4356 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004357 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004358 }
4359 j++;
4360 }
4361
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004362swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004363 /* Swap primary and spare array */
4364 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07004365
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004366 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004367
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004368 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004369 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08004370
4371 /* If all events are unregistered, free the spare array */
4372 if (!new) {
4373 kfree(thresholds->spare);
4374 thresholds->spare = NULL;
4375 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04004376unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004377 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004378}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004379
Tejun Heo59b6f872013-11-22 18:20:43 -05004380static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004381 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004382{
Tejun Heo59b6f872013-11-22 18:20:43 -05004383 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05004384}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004385
Tejun Heo59b6f872013-11-22 18:20:43 -05004386static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004387 struct eventfd_ctx *eventfd)
4388{
Tejun Heo59b6f872013-11-22 18:20:43 -05004389 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05004390}
4391
Tejun Heo59b6f872013-11-22 18:20:43 -05004392static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004393 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004394{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004395 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004396
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004397 event = kmalloc(sizeof(*event), GFP_KERNEL);
4398 if (!event)
4399 return -ENOMEM;
4400
Michal Hocko1af8efe2011-07-26 16:08:24 -07004401 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004402
4403 event->eventfd = eventfd;
4404 list_add(&event->list, &memcg->oom_notify);
4405
4406 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07004407 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004408 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004409 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004410
4411 return 0;
4412}
4413
Tejun Heo59b6f872013-11-22 18:20:43 -05004414static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004415 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004416{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004417 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004418
Michal Hocko1af8efe2011-07-26 16:08:24 -07004419 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004420
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004421 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004422 if (ev->eventfd == eventfd) {
4423 list_del(&ev->list);
4424 kfree(ev);
4425 }
4426 }
4427
Michal Hocko1af8efe2011-07-26 16:08:24 -07004428 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004429}
4430
Tejun Heo2da8ca82013-12-05 12:28:04 -05004431static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004432{
Chris Downaa9694b2019-03-05 15:45:52 -08004433 struct mem_cgroup *memcg = mem_cgroup_from_seq(sf);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004434
Tejun Heo791badb2013-12-05 12:28:02 -05004435 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07004436 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
Roman Gushchinfe6bdfc2018-06-14 15:28:05 -07004437 seq_printf(sf, "oom_kill %lu\n",
4438 atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL]));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004439 return 0;
4440}
4441
Tejun Heo182446d2013-08-08 20:11:24 -04004442static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004443 struct cftype *cft, u64 val)
4444{
Tejun Heo182446d2013-08-08 20:11:24 -04004445 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004446
4447 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07004448 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004449 return -EINVAL;
4450
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004451 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004452 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004453 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07004454
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004455 return 0;
4456}
4457
Tejun Heo52ebea72015-05-22 17:13:37 -04004458#ifdef CONFIG_CGROUP_WRITEBACK
4459
Tejun Heo3a8e9ac2019-08-29 15:47:19 -07004460#include <trace/events/writeback.h>
4461
Tejun Heo841710a2015-05-22 18:23:33 -04004462static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
4463{
4464 return wb_domain_init(&memcg->cgwb_domain, gfp);
4465}
4466
4467static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
4468{
4469 wb_domain_exit(&memcg->cgwb_domain);
4470}
4471
Tejun Heo2529bb32015-05-22 18:23:34 -04004472static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
4473{
4474 wb_domain_size_changed(&memcg->cgwb_domain);
4475}
4476
Tejun Heo841710a2015-05-22 18:23:33 -04004477struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
4478{
4479 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
4480
4481 if (!memcg->css.parent)
4482 return NULL;
4483
4484 return &memcg->cgwb_domain;
4485}
4486
Greg Thelen0b3d6e62019-04-05 18:39:18 -07004487/*
4488 * idx can be of type enum memcg_stat_item or node_stat_item.
4489 * Keep in sync with memcg_exact_page().
4490 */
4491static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx)
4492{
Chris Down871789d2019-05-14 15:46:57 -07004493 long x = atomic_long_read(&memcg->vmstats[idx]);
Greg Thelen0b3d6e62019-04-05 18:39:18 -07004494 int cpu;
4495
4496 for_each_online_cpu(cpu)
Chris Down871789d2019-05-14 15:46:57 -07004497 x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx];
Greg Thelen0b3d6e62019-04-05 18:39:18 -07004498 if (x < 0)
4499 x = 0;
4500 return x;
4501}
4502
Tejun Heoc2aa7232015-05-22 18:23:35 -04004503/**
4504 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
4505 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004506 * @pfilepages: out parameter for number of file pages
4507 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04004508 * @pdirty: out parameter for number of dirty pages
4509 * @pwriteback: out parameter for number of pages under writeback
4510 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004511 * Determine the numbers of file, headroom, dirty, and writeback pages in
4512 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
4513 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04004514 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004515 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
4516 * headroom is calculated as the lowest headroom of itself and the
4517 * ancestors. Note that this doesn't consider the actual amount of
4518 * available memory in the system. The caller should further cap
4519 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04004520 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004521void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
4522 unsigned long *pheadroom, unsigned long *pdirty,
4523 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04004524{
4525 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
4526 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04004527
Greg Thelen0b3d6e62019-04-05 18:39:18 -07004528 *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY);
Tejun Heoc2aa7232015-05-22 18:23:35 -04004529
Greg Thelen0b3d6e62019-04-05 18:39:18 -07004530 *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK);
Johannes Weiner21d89d12019-05-13 17:18:08 -07004531 *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) +
4532 memcg_exact_page_state(memcg, NR_ACTIVE_FILE);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004533 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04004534
Tejun Heoc2aa7232015-05-22 18:23:35 -04004535 while ((parent = parent_mem_cgroup(memcg))) {
Chris Down15b42562020-04-01 21:07:20 -07004536 unsigned long ceiling = min(READ_ONCE(memcg->memory.max),
Jakub Kicinskid1663a92020-06-01 21:49:49 -07004537 READ_ONCE(memcg->memory.high));
Tejun Heoc2aa7232015-05-22 18:23:35 -04004538 unsigned long used = page_counter_read(&memcg->memory);
4539
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004540 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04004541 memcg = parent;
4542 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04004543}
4544
Tejun Heo97b27822019-08-26 09:06:56 -07004545/*
4546 * Foreign dirty flushing
4547 *
4548 * There's an inherent mismatch between memcg and writeback. The former
4549 * trackes ownership per-page while the latter per-inode. This was a
4550 * deliberate design decision because honoring per-page ownership in the
4551 * writeback path is complicated, may lead to higher CPU and IO overheads
4552 * and deemed unnecessary given that write-sharing an inode across
4553 * different cgroups isn't a common use-case.
4554 *
4555 * Combined with inode majority-writer ownership switching, this works well
4556 * enough in most cases but there are some pathological cases. For
4557 * example, let's say there are two cgroups A and B which keep writing to
4558 * different but confined parts of the same inode. B owns the inode and
4559 * A's memory is limited far below B's. A's dirty ratio can rise enough to
4560 * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid
4561 * triggering background writeback. A will be slowed down without a way to
4562 * make writeback of the dirty pages happen.
4563 *
4564 * Conditions like the above can lead to a cgroup getting repatedly and
4565 * severely throttled after making some progress after each
4566 * dirty_expire_interval while the underyling IO device is almost
4567 * completely idle.
4568 *
4569 * Solving this problem completely requires matching the ownership tracking
4570 * granularities between memcg and writeback in either direction. However,
4571 * the more egregious behaviors can be avoided by simply remembering the
4572 * most recent foreign dirtying events and initiating remote flushes on
4573 * them when local writeback isn't enough to keep the memory clean enough.
4574 *
4575 * The following two functions implement such mechanism. When a foreign
4576 * page - a page whose memcg and writeback ownerships don't match - is
4577 * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning
4578 * bdi_writeback on the page owning memcg. When balance_dirty_pages()
4579 * decides that the memcg needs to sleep due to high dirty ratio, it calls
4580 * mem_cgroup_flush_foreign() which queues writeback on the recorded
4581 * foreign bdi_writebacks which haven't expired. Both the numbers of
4582 * recorded bdi_writebacks and concurrent in-flight foreign writebacks are
4583 * limited to MEMCG_CGWB_FRN_CNT.
4584 *
4585 * The mechanism only remembers IDs and doesn't hold any object references.
4586 * As being wrong occasionally doesn't matter, updates and accesses to the
4587 * records are lockless and racy.
4588 */
4589void mem_cgroup_track_foreign_dirty_slowpath(struct page *page,
4590 struct bdi_writeback *wb)
4591{
4592 struct mem_cgroup *memcg = page->mem_cgroup;
4593 struct memcg_cgwb_frn *frn;
4594 u64 now = get_jiffies_64();
4595 u64 oldest_at = now;
4596 int oldest = -1;
4597 int i;
4598
Tejun Heo3a8e9ac2019-08-29 15:47:19 -07004599 trace_track_foreign_dirty(page, wb);
4600
Tejun Heo97b27822019-08-26 09:06:56 -07004601 /*
4602 * Pick the slot to use. If there is already a slot for @wb, keep
4603 * using it. If not replace the oldest one which isn't being
4604 * written out.
4605 */
4606 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) {
4607 frn = &memcg->cgwb_frn[i];
4608 if (frn->bdi_id == wb->bdi->id &&
4609 frn->memcg_id == wb->memcg_css->id)
4610 break;
4611 if (time_before64(frn->at, oldest_at) &&
4612 atomic_read(&frn->done.cnt) == 1) {
4613 oldest = i;
4614 oldest_at = frn->at;
4615 }
4616 }
4617
4618 if (i < MEMCG_CGWB_FRN_CNT) {
4619 /*
4620 * Re-using an existing one. Update timestamp lazily to
4621 * avoid making the cacheline hot. We want them to be
4622 * reasonably up-to-date and significantly shorter than
4623 * dirty_expire_interval as that's what expires the record.
4624 * Use the shorter of 1s and dirty_expire_interval / 8.
4625 */
4626 unsigned long update_intv =
4627 min_t(unsigned long, HZ,
4628 msecs_to_jiffies(dirty_expire_interval * 10) / 8);
4629
4630 if (time_before64(frn->at, now - update_intv))
4631 frn->at = now;
4632 } else if (oldest >= 0) {
4633 /* replace the oldest free one */
4634 frn = &memcg->cgwb_frn[oldest];
4635 frn->bdi_id = wb->bdi->id;
4636 frn->memcg_id = wb->memcg_css->id;
4637 frn->at = now;
4638 }
4639}
4640
4641/* issue foreign writeback flushes for recorded foreign dirtying events */
4642void mem_cgroup_flush_foreign(struct bdi_writeback *wb)
4643{
4644 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
4645 unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10);
4646 u64 now = jiffies_64;
4647 int i;
4648
4649 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) {
4650 struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i];
4651
4652 /*
4653 * If the record is older than dirty_expire_interval,
4654 * writeback on it has already started. No need to kick it
4655 * off again. Also, don't start a new one if there's
4656 * already one in flight.
4657 */
4658 if (time_after64(frn->at, now - intv) &&
4659 atomic_read(&frn->done.cnt) == 1) {
4660 frn->at = 0;
Tejun Heo3a8e9ac2019-08-29 15:47:19 -07004661 trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id);
Tejun Heo97b27822019-08-26 09:06:56 -07004662 cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0,
4663 WB_REASON_FOREIGN_FLUSH,
4664 &frn->done);
4665 }
4666 }
4667}
4668
Tejun Heo841710a2015-05-22 18:23:33 -04004669#else /* CONFIG_CGROUP_WRITEBACK */
4670
4671static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
4672{
4673 return 0;
4674}
4675
4676static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
4677{
4678}
4679
Tejun Heo2529bb32015-05-22 18:23:34 -04004680static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
4681{
4682}
4683
Tejun Heo52ebea72015-05-22 17:13:37 -04004684#endif /* CONFIG_CGROUP_WRITEBACK */
4685
Tejun Heo79bd9812013-11-22 18:20:42 -05004686/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05004687 * DO NOT USE IN NEW FILES.
4688 *
4689 * "cgroup.event_control" implementation.
4690 *
4691 * This is way over-engineered. It tries to support fully configurable
4692 * events for each user. Such level of flexibility is completely
4693 * unnecessary especially in the light of the planned unified hierarchy.
4694 *
4695 * Please deprecate this and replace with something simpler if at all
4696 * possible.
4697 */
4698
4699/*
Tejun Heo79bd9812013-11-22 18:20:42 -05004700 * Unregister event and free resources.
4701 *
4702 * Gets called from workqueue.
4703 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05004704static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05004705{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004706 struct mem_cgroup_event *event =
4707 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05004708 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05004709
4710 remove_wait_queue(event->wqh, &event->wait);
4711
Tejun Heo59b6f872013-11-22 18:20:43 -05004712 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05004713
4714 /* Notify userspace the event is going away. */
4715 eventfd_signal(event->eventfd, 1);
4716
4717 eventfd_ctx_put(event->eventfd);
4718 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05004719 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05004720}
4721
4722/*
Linus Torvaldsa9a08842018-02-11 14:34:03 -08004723 * Gets called on EPOLLHUP on eventfd when user closes it.
Tejun Heo79bd9812013-11-22 18:20:42 -05004724 *
4725 * Called with wqh->lock held and interrupts disabled.
4726 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02004727static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode,
Tejun Heo3bc942f2013-11-22 18:20:44 -05004728 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05004729{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004730 struct mem_cgroup_event *event =
4731 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05004732 struct mem_cgroup *memcg = event->memcg;
Al Viro3ad6f932017-07-03 20:14:56 -04004733 __poll_t flags = key_to_poll(key);
Tejun Heo79bd9812013-11-22 18:20:42 -05004734
Linus Torvaldsa9a08842018-02-11 14:34:03 -08004735 if (flags & EPOLLHUP) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004736 /*
4737 * If the event has been detached at cgroup removal, we
4738 * can simply return knowing the other side will cleanup
4739 * for us.
4740 *
4741 * We can't race against event freeing since the other
4742 * side will require wqh->lock via remove_wait_queue(),
4743 * which we hold.
4744 */
Tejun Heofba94802013-11-22 18:20:43 -05004745 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004746 if (!list_empty(&event->list)) {
4747 list_del_init(&event->list);
4748 /*
4749 * We are in atomic context, but cgroup_event_remove()
4750 * may sleep, so we have to call it in workqueue.
4751 */
4752 schedule_work(&event->remove);
4753 }
Tejun Heofba94802013-11-22 18:20:43 -05004754 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004755 }
4756
4757 return 0;
4758}
4759
Tejun Heo3bc942f2013-11-22 18:20:44 -05004760static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05004761 wait_queue_head_t *wqh, poll_table *pt)
4762{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004763 struct mem_cgroup_event *event =
4764 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05004765
4766 event->wqh = wqh;
4767 add_wait_queue(wqh, &event->wait);
4768}
4769
4770/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05004771 * DO NOT USE IN NEW FILES.
4772 *
Tejun Heo79bd9812013-11-22 18:20:42 -05004773 * Parse input and register new cgroup event handler.
4774 *
4775 * Input must be in format '<event_fd> <control_fd> <args>'.
4776 * Interpretation of args is defined by control file implementation.
4777 */
Tejun Heo451af502014-05-13 12:16:21 -04004778static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
4779 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05004780{
Tejun Heo451af502014-05-13 12:16:21 -04004781 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05004782 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004783 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05004784 struct cgroup_subsys_state *cfile_css;
4785 unsigned int efd, cfd;
4786 struct fd efile;
4787 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05004788 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05004789 char *endp;
4790 int ret;
4791
Tejun Heo451af502014-05-13 12:16:21 -04004792 buf = strstrip(buf);
4793
4794 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05004795 if (*endp != ' ')
4796 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04004797 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05004798
Tejun Heo451af502014-05-13 12:16:21 -04004799 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05004800 if ((*endp != ' ') && (*endp != '\0'))
4801 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04004802 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05004803
4804 event = kzalloc(sizeof(*event), GFP_KERNEL);
4805 if (!event)
4806 return -ENOMEM;
4807
Tejun Heo59b6f872013-11-22 18:20:43 -05004808 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05004809 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004810 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
4811 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
4812 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05004813
4814 efile = fdget(efd);
4815 if (!efile.file) {
4816 ret = -EBADF;
4817 goto out_kfree;
4818 }
4819
4820 event->eventfd = eventfd_ctx_fileget(efile.file);
4821 if (IS_ERR(event->eventfd)) {
4822 ret = PTR_ERR(event->eventfd);
4823 goto out_put_efile;
4824 }
4825
4826 cfile = fdget(cfd);
4827 if (!cfile.file) {
4828 ret = -EBADF;
4829 goto out_put_eventfd;
4830 }
4831
4832 /* the process need read permission on control file */
4833 /* AV: shouldn't we check that it's been opened for read instead? */
4834 ret = inode_permission(file_inode(cfile.file), MAY_READ);
4835 if (ret < 0)
4836 goto out_put_cfile;
4837
Tejun Heo79bd9812013-11-22 18:20:42 -05004838 /*
Tejun Heofba94802013-11-22 18:20:43 -05004839 * Determine the event callbacks and set them in @event. This used
4840 * to be done via struct cftype but cgroup core no longer knows
4841 * about these events. The following is crude but the whole thing
4842 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05004843 *
4844 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05004845 */
Al Virob5830432014-10-31 01:22:04 -04004846 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05004847
4848 if (!strcmp(name, "memory.usage_in_bytes")) {
4849 event->register_event = mem_cgroup_usage_register_event;
4850 event->unregister_event = mem_cgroup_usage_unregister_event;
4851 } else if (!strcmp(name, "memory.oom_control")) {
4852 event->register_event = mem_cgroup_oom_register_event;
4853 event->unregister_event = mem_cgroup_oom_unregister_event;
4854 } else if (!strcmp(name, "memory.pressure_level")) {
4855 event->register_event = vmpressure_register_event;
4856 event->unregister_event = vmpressure_unregister_event;
4857 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05004858 event->register_event = memsw_cgroup_usage_register_event;
4859 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05004860 } else {
4861 ret = -EINVAL;
4862 goto out_put_cfile;
4863 }
4864
4865 /*
Tejun Heob5557c42013-11-22 18:20:42 -05004866 * Verify @cfile should belong to @css. Also, remaining events are
4867 * automatically removed on cgroup destruction but the removal is
4868 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05004869 */
Al Virob5830432014-10-31 01:22:04 -04004870 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04004871 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05004872 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05004873 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05004874 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05004875 if (cfile_css != css) {
4876 css_put(cfile_css);
4877 goto out_put_cfile;
4878 }
Tejun Heo79bd9812013-11-22 18:20:42 -05004879
Tejun Heo451af502014-05-13 12:16:21 -04004880 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05004881 if (ret)
4882 goto out_put_css;
4883
Christoph Hellwig9965ed172018-03-05 07:26:05 -08004884 vfs_poll(efile.file, &event->pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05004885
Tejun Heofba94802013-11-22 18:20:43 -05004886 spin_lock(&memcg->event_list_lock);
4887 list_add(&event->list, &memcg->event_list);
4888 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004889
4890 fdput(cfile);
4891 fdput(efile);
4892
Tejun Heo451af502014-05-13 12:16:21 -04004893 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05004894
4895out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05004896 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05004897out_put_cfile:
4898 fdput(cfile);
4899out_put_eventfd:
4900 eventfd_ctx_put(event->eventfd);
4901out_put_efile:
4902 fdput(efile);
4903out_kfree:
4904 kfree(event);
4905
4906 return ret;
4907}
4908
Johannes Weiner241994ed2015-02-11 15:26:06 -08004909static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004910 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004911 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004912 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004913 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004914 },
4915 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004916 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004917 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004918 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004919 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004920 },
4921 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004922 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004923 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004924 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004925 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004926 },
4927 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004928 .name = "soft_limit_in_bytes",
4929 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004930 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004931 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07004932 },
4933 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004934 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004935 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004936 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004937 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004938 },
Balbir Singh8697d332008-02-07 00:13:59 -08004939 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004940 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004941 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004942 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004943 {
4944 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04004945 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004946 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004947 {
4948 .name = "use_hierarchy",
4949 .write_u64 = mem_cgroup_hierarchy_write,
4950 .read_u64 = mem_cgroup_hierarchy_read,
4951 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004952 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05004953 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04004954 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04004955 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05004956 },
4957 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004958 .name = "swappiness",
4959 .read_u64 = mem_cgroup_swappiness_read,
4960 .write_u64 = mem_cgroup_swappiness_write,
4961 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004962 {
4963 .name = "move_charge_at_immigrate",
4964 .read_u64 = mem_cgroup_move_charge_read,
4965 .write_u64 = mem_cgroup_move_charge_write,
4966 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004967 {
4968 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004969 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004970 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004971 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4972 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004973 {
4974 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004975 },
Ying Han406eb0c2011-05-26 16:25:37 -07004976#ifdef CONFIG_NUMA
4977 {
4978 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004979 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07004980 },
4981#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004982 {
4983 .name = "kmem.limit_in_bytes",
4984 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004985 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004986 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004987 },
4988 {
4989 .name = "kmem.usage_in_bytes",
4990 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004991 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004992 },
4993 {
4994 .name = "kmem.failcnt",
4995 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004996 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004997 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004998 },
4999 {
5000 .name = "kmem.max_usage_in_bytes",
5001 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04005002 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005003 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08005004 },
Yafang Shaoa87425a2020-04-01 21:06:30 -07005005#if defined(CONFIG_MEMCG_KMEM) && \
5006 (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG))
Glauber Costa749c5412012-12-18 14:23:01 -08005007 {
5008 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08005009 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08005010 },
5011#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08005012 {
5013 .name = "kmem.tcp.limit_in_bytes",
5014 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
5015 .write = mem_cgroup_write,
5016 .read_u64 = mem_cgroup_read_u64,
5017 },
5018 {
5019 .name = "kmem.tcp.usage_in_bytes",
5020 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
5021 .read_u64 = mem_cgroup_read_u64,
5022 },
5023 {
5024 .name = "kmem.tcp.failcnt",
5025 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
5026 .write = mem_cgroup_reset,
5027 .read_u64 = mem_cgroup_read_u64,
5028 },
5029 {
5030 .name = "kmem.tcp.max_usage_in_bytes",
5031 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
5032 .write = mem_cgroup_reset,
5033 .read_u64 = mem_cgroup_read_u64,
5034 },
Tejun Heo6bc10342012-04-01 12:09:55 -07005035 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07005036};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005037
Johannes Weiner73f576c2016-07-20 15:44:57 -07005038/*
5039 * Private memory cgroup IDR
5040 *
5041 * Swap-out records and page cache shadow entries need to store memcg
5042 * references in constrained space, so we maintain an ID space that is
5043 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
5044 * memory-controlled cgroups to 64k.
5045 *
Ethon Paulb8f29352020-06-04 16:49:28 -07005046 * However, there usually are many references to the offline CSS after
Johannes Weiner73f576c2016-07-20 15:44:57 -07005047 * the cgroup has been destroyed, such as page cache or reclaimable
5048 * slab objects, that don't need to hang on to the ID. We want to keep
5049 * those dead CSS from occupying IDs, or we might quickly exhaust the
5050 * relatively small ID space and prevent the creation of new cgroups
5051 * even when there are much fewer than 64k cgroups - possibly none.
5052 *
5053 * Maintain a private 16-bit ID space for memcg, and allow the ID to
5054 * be freed and recycled when it's no longer needed, which is usually
5055 * when the CSS is offlined.
5056 *
5057 * The only exception to that are records of swapped out tmpfs/shmem
5058 * pages that need to be attributed to live ancestors on swapin. But
5059 * those references are manageable from userspace.
5060 */
5061
5062static DEFINE_IDR(mem_cgroup_idr);
5063
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005064static void mem_cgroup_id_remove(struct mem_cgroup *memcg)
5065{
5066 if (memcg->id.id > 0) {
5067 idr_remove(&mem_cgroup_idr, memcg->id.id);
5068 memcg->id.id = 0;
5069 }
5070}
5071
Vincenzo Frascinoc1514c02020-04-01 21:07:13 -07005072static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg,
5073 unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07005074{
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07005075 refcount_add(n, &memcg->id.ref);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005076}
5077
Vladimir Davydov615d66c2016-08-11 15:33:03 -07005078static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07005079{
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07005080 if (refcount_sub_and_test(n, &memcg->id.ref)) {
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005081 mem_cgroup_id_remove(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005082
5083 /* Memcg ID pins CSS */
5084 css_put(&memcg->css);
5085 }
5086}
5087
Vladimir Davydov615d66c2016-08-11 15:33:03 -07005088static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
5089{
5090 mem_cgroup_id_put_many(memcg, 1);
5091}
5092
Johannes Weiner73f576c2016-07-20 15:44:57 -07005093/**
5094 * mem_cgroup_from_id - look up a memcg from a memcg id
5095 * @id: the memcg id to look up
5096 *
5097 * Caller must hold rcu_read_lock().
5098 */
5099struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
5100{
5101 WARN_ON_ONCE(!rcu_read_lock_held());
5102 return idr_find(&mem_cgroup_idr, id);
5103}
5104
Mel Gormanef8f2322016-07-28 15:46:05 -07005105static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005106{
5107 struct mem_cgroup_per_node *pn;
Mel Gormanef8f2322016-07-28 15:46:05 -07005108 int tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005109 /*
5110 * This routine is called against possible nodes.
5111 * But it's BUG to call kmalloc() against offline node.
5112 *
5113 * TODO: this routine can waste much memory for nodes which will
5114 * never be onlined. It's better to use memory hotplug callback
5115 * function.
5116 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07005117 if (!node_state(node, N_NORMAL_MEMORY))
5118 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08005119 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005120 if (!pn)
5121 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005122
Johannes Weiner815744d2019-06-13 15:55:46 -07005123 pn->lruvec_stat_local = alloc_percpu(struct lruvec_stat);
5124 if (!pn->lruvec_stat_local) {
5125 kfree(pn);
5126 return 1;
5127 }
5128
Johannes Weinera983b5e2018-01-31 16:16:45 -08005129 pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat);
5130 if (!pn->lruvec_stat_cpu) {
Johannes Weiner815744d2019-06-13 15:55:46 -07005131 free_percpu(pn->lruvec_stat_local);
Johannes Weiner00f3ca22017-07-06 15:40:52 -07005132 kfree(pn);
5133 return 1;
5134 }
5135
Mel Gormanef8f2322016-07-28 15:46:05 -07005136 lruvec_init(&pn->lruvec);
5137 pn->usage_in_excess = 0;
5138 pn->on_tree = false;
5139 pn->memcg = memcg;
5140
Johannes Weiner54f72fe2013-07-08 15:59:49 -07005141 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005142 return 0;
5143}
5144
Mel Gormanef8f2322016-07-28 15:46:05 -07005145static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005146{
Johannes Weiner00f3ca22017-07-06 15:40:52 -07005147 struct mem_cgroup_per_node *pn = memcg->nodeinfo[node];
5148
Michal Hocko4eaf4312018-04-10 16:29:52 -07005149 if (!pn)
5150 return;
5151
Johannes Weinera983b5e2018-01-31 16:16:45 -08005152 free_percpu(pn->lruvec_stat_cpu);
Johannes Weiner815744d2019-06-13 15:55:46 -07005153 free_percpu(pn->lruvec_stat_local);
Johannes Weiner00f3ca22017-07-06 15:40:52 -07005154 kfree(pn);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005155}
5156
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005157static void __mem_cgroup_free(struct mem_cgroup *memcg)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005158{
5159 int node;
5160
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005161 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07005162 free_mem_cgroup_per_node_info(memcg, node);
Chris Down871789d2019-05-14 15:46:57 -07005163 free_percpu(memcg->vmstats_percpu);
Johannes Weiner815744d2019-06-13 15:55:46 -07005164 free_percpu(memcg->vmstats_local);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005165 kfree(memcg);
5166}
5167
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005168static void mem_cgroup_free(struct mem_cgroup *memcg)
5169{
5170 memcg_wb_domain_exit(memcg);
Shakeel Butt7961eee2019-11-05 21:16:21 -08005171 /*
5172 * Flush percpu vmstats and vmevents to guarantee the value correctness
5173 * on parent's and all ancestor levels.
5174 */
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08005175 memcg_flush_percpu_vmstats(memcg);
Shakeel Butt7961eee2019-11-05 21:16:21 -08005176 memcg_flush_percpu_vmevents(memcg);
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005177 __mem_cgroup_free(memcg);
5178}
5179
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005180static struct mem_cgroup *mem_cgroup_alloc(void)
5181{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005182 struct mem_cgroup *memcg;
Alexey Dobriyanb9726c22019-03-05 15:48:26 -08005183 unsigned int size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005184 int node;
Tejun Heo97b27822019-08-26 09:06:56 -07005185 int __maybe_unused i;
Yafang Shao11d67612020-05-07 18:35:43 -07005186 long error = -ENOMEM;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005187
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08005188 size = sizeof(struct mem_cgroup);
5189 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005190
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08005191 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005192 if (!memcg)
Yafang Shao11d67612020-05-07 18:35:43 -07005193 return ERR_PTR(error);
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07005194
Johannes Weiner73f576c2016-07-20 15:44:57 -07005195 memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
5196 1, MEM_CGROUP_ID_MAX,
5197 GFP_KERNEL);
Yafang Shao11d67612020-05-07 18:35:43 -07005198 if (memcg->id.id < 0) {
5199 error = memcg->id.id;
Johannes Weiner73f576c2016-07-20 15:44:57 -07005200 goto fail;
Yafang Shao11d67612020-05-07 18:35:43 -07005201 }
Johannes Weiner73f576c2016-07-20 15:44:57 -07005202
Johannes Weiner815744d2019-06-13 15:55:46 -07005203 memcg->vmstats_local = alloc_percpu(struct memcg_vmstats_percpu);
5204 if (!memcg->vmstats_local)
5205 goto fail;
5206
Chris Down871789d2019-05-14 15:46:57 -07005207 memcg->vmstats_percpu = alloc_percpu(struct memcg_vmstats_percpu);
5208 if (!memcg->vmstats_percpu)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005209 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08005210
Bob Liu3ed28fa2012-01-12 17:19:04 -08005211 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07005212 if (alloc_mem_cgroup_per_node_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005213 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07005214
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005215 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
5216 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08005217
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005218 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08005219 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08005220 mutex_init(&memcg->thresholds_lock);
5221 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07005222 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05005223 INIT_LIST_HEAD(&memcg->event_list);
5224 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005225 memcg->socket_pressure = jiffies;
Kirill Tkhai84c07d12018-08-17 15:47:25 -07005226#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydov900a38f2014-12-12 16:55:10 -08005227 memcg->kmemcg_id = -1;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07005228 INIT_LIST_HEAD(&memcg->objcg_list);
Vladimir Davydov900a38f2014-12-12 16:55:10 -08005229#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04005230#ifdef CONFIG_CGROUP_WRITEBACK
5231 INIT_LIST_HEAD(&memcg->cgwb_list);
Tejun Heo97b27822019-08-26 09:06:56 -07005232 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++)
5233 memcg->cgwb_frn[i].done =
5234 __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq);
Tejun Heo52ebea72015-05-22 17:13:37 -04005235#endif
Yang Shi87eaceb2019-09-23 15:38:15 -07005236#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5237 spin_lock_init(&memcg->deferred_split_queue.split_queue_lock);
5238 INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue);
5239 memcg->deferred_split_queue.split_queue_len = 0;
5240#endif
Johannes Weiner73f576c2016-07-20 15:44:57 -07005241 idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005242 return memcg;
5243fail:
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005244 mem_cgroup_id_remove(memcg);
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005245 __mem_cgroup_free(memcg);
Yafang Shao11d67612020-05-07 18:35:43 -07005246 return ERR_PTR(error);
Glauber Costad142e3e2013-02-22 16:34:52 -08005247}
5248
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005249static struct cgroup_subsys_state * __ref
5250mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08005251{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005252 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
5253 struct mem_cgroup *memcg;
5254 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08005255
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005256 memcg = mem_cgroup_alloc();
Yafang Shao11d67612020-05-07 18:35:43 -07005257 if (IS_ERR(memcg))
5258 return ERR_CAST(memcg);
Li Zefan4219b2d2013-09-23 16:56:29 +08005259
Jakub Kicinskid1663a92020-06-01 21:49:49 -07005260 page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005261 memcg->soft_limit = PAGE_COUNTER_MAX;
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07005262 page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005263 if (parent) {
5264 memcg->swappiness = mem_cgroup_swappiness(parent);
5265 memcg->oom_kill_disable = parent->oom_kill_disable;
5266 }
5267 if (parent && parent->use_hierarchy) {
5268 memcg->use_hierarchy = true;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005269 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08005270 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005271 page_counter_init(&memcg->memsw, &parent->memsw);
5272 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08005273 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005274 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005275 page_counter_init(&memcg->memory, NULL);
Vladimir Davydov37e84352016-01-20 15:02:56 -08005276 page_counter_init(&memcg->swap, NULL);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005277 page_counter_init(&memcg->memsw, NULL);
5278 page_counter_init(&memcg->kmem, NULL);
Johannes Weiner0db15292016-01-20 15:02:50 -08005279 page_counter_init(&memcg->tcpmem, NULL);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07005280 /*
5281 * Deeper hierachy with use_hierarchy == false doesn't make
5282 * much sense so let cgroup subsystem know about this
5283 * unfortunate state in our controller.
5284 */
Glauber Costad142e3e2013-02-22 16:34:52 -08005285 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05005286 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08005287 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08005288
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005289 /* The following stuff does not apply to the root */
5290 if (!parent) {
5291 root_mem_cgroup = memcg;
5292 return &memcg->css;
5293 }
5294
Vladimir Davydovb313aee2016-03-17 14:18:27 -07005295 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005296 if (error)
5297 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08005298
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005299 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08005300 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005301
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005302 return &memcg->css;
5303fail:
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005304 mem_cgroup_id_remove(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005305 mem_cgroup_free(memcg);
Yafang Shao11d67612020-05-07 18:35:43 -07005306 return ERR_PTR(error);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005307}
5308
Johannes Weiner73f576c2016-07-20 15:44:57 -07005309static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005310{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07005311 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5312
Kirill Tkhai0a4465d2018-08-17 15:47:37 -07005313 /*
5314 * A memcg must be visible for memcg_expand_shrinker_maps()
5315 * by the time the maps are allocated. So, we allocate maps
5316 * here, when for_each_mem_cgroup() can't skip it.
5317 */
5318 if (memcg_alloc_shrinker_maps(memcg)) {
5319 mem_cgroup_id_remove(memcg);
5320 return -ENOMEM;
5321 }
5322
Johannes Weiner73f576c2016-07-20 15:44:57 -07005323 /* Online state pins memcg ID, memcg ID pins CSS */
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07005324 refcount_set(&memcg->id.ref, 1);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005325 css_get(css);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07005326 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005327}
5328
Tejun Heoeb954192013-08-08 20:11:23 -04005329static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005330{
Tejun Heoeb954192013-08-08 20:11:23 -04005331 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05005332 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05005333
5334 /*
5335 * Unregister events and notify userspace.
5336 * Notify userspace about cgroup removing only after rmdir of cgroup
5337 * directory to avoid race between userspace and kernelspace.
5338 */
Tejun Heofba94802013-11-22 18:20:43 -05005339 spin_lock(&memcg->event_list_lock);
5340 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05005341 list_del_init(&event->list);
5342 schedule_work(&event->remove);
5343 }
Tejun Heofba94802013-11-22 18:20:43 -05005344 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07005345
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07005346 page_counter_set_min(&memcg->memory, 0);
Roman Gushchin23067152018-06-07 17:06:22 -07005347 page_counter_set_low(&memcg->memory, 0);
Roman Gushchin63677c742017-09-06 16:21:47 -07005348
Johannes Weiner567e9ab2016-01-20 15:02:24 -08005349 memcg_offline_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04005350 wb_memcg_offline(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005351
Roman Gushchin591edfb2018-10-26 15:03:23 -07005352 drain_all_stock(memcg);
5353
Johannes Weiner73f576c2016-07-20 15:44:57 -07005354 mem_cgroup_id_put(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005355}
5356
Vladimir Davydov6df38682015-12-29 14:54:10 -08005357static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
5358{
5359 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5360
5361 invalidate_reclaim_iterators(memcg);
5362}
5363
Tejun Heoeb954192013-08-08 20:11:23 -04005364static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005365{
Tejun Heoeb954192013-08-08 20:11:23 -04005366 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo97b27822019-08-26 09:06:56 -07005367 int __maybe_unused i;
Daisuke Nishimurac268e992009-01-15 13:51:13 -08005368
Tejun Heo97b27822019-08-26 09:06:56 -07005369#ifdef CONFIG_CGROUP_WRITEBACK
5370 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++)
5371 wb_wait_for_completion(&memcg->cgwb_frn[i].done);
5372#endif
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005373 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08005374 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08005375
Johannes Weiner0db15292016-01-20 15:02:50 -08005376 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08005377 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08005378
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005379 vmpressure_cleanup(&memcg->vmpressure);
5380 cancel_work_sync(&memcg->high_work);
5381 mem_cgroup_remove_from_trees(memcg);
Kirill Tkhai0a4465d2018-08-17 15:47:37 -07005382 memcg_free_shrinker_maps(memcg);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005383 memcg_free_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005384 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005385}
5386
Tejun Heo1ced9532014-07-08 18:02:57 -04005387/**
5388 * mem_cgroup_css_reset - reset the states of a mem_cgroup
5389 * @css: the target css
5390 *
5391 * Reset the states of the mem_cgroup associated with @css. This is
5392 * invoked when the userland requests disabling on the default hierarchy
5393 * but the memcg is pinned through dependency. The memcg should stop
5394 * applying policies and should revert to the vanilla state as it may be
5395 * made visible again.
5396 *
5397 * The current implementation only resets the essential configurations.
5398 * This needs to be expanded to cover all the visible parts.
5399 */
5400static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
5401{
5402 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5403
Roman Gushchinbbec2e12018-06-07 17:06:18 -07005404 page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX);
5405 page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX);
5406 page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX);
5407 page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX);
5408 page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX);
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07005409 page_counter_set_min(&memcg->memory, 0);
Roman Gushchin23067152018-06-07 17:06:22 -07005410 page_counter_set_low(&memcg->memory, 0);
Jakub Kicinskid1663a92020-06-01 21:49:49 -07005411 page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX);
Johannes Weiner24d404d2015-01-08 14:32:35 -08005412 memcg->soft_limit = PAGE_COUNTER_MAX;
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07005413 page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX);
Tejun Heo2529bb32015-05-22 18:23:34 -04005414 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04005415}
5416
Daisuke Nishimura02491442010-03-10 15:22:17 -08005417#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005418/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005419static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005420{
Johannes Weiner05b84302014-08-06 16:05:59 -07005421 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07005422
Mel Gormand0164ad2015-11-06 16:28:21 -08005423 /* Try a single bulk charge without reclaim first, kswapd may wake */
5424 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07005425 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005426 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005427 return ret;
5428 }
Johannes Weiner9476db92014-08-06 16:05:55 -07005429
David Rientjes36745342017-01-24 15:18:10 -08005430 /* Try charges one by one with reclaim, but do not retry */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005431 while (count--) {
David Rientjes36745342017-01-24 15:18:10 -08005432 ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005433 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005434 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005435 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07005436 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005437 }
Johannes Weiner9476db92014-08-06 16:05:55 -07005438 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005439}
5440
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005441union mc_target {
5442 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005443 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005444};
5445
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005446enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005447 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005448 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005449 MC_TARGET_SWAP,
Jérôme Glissec733a822017-09-08 16:11:54 -07005450 MC_TARGET_DEVICE,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005451};
5452
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005453static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5454 unsigned long addr, pte_t ptent)
5455{
Christoph Hellwig25b29952019-06-13 22:50:49 +02005456 struct page *page = vm_normal_page(vma, addr, ptent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005457
5458 if (!page || !page_mapped(page))
5459 return NULL;
5460 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005461 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005462 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005463 } else {
5464 if (!(mc.flags & MOVE_FILE))
5465 return NULL;
5466 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005467 if (!get_page_unless_zero(page))
5468 return NULL;
5469
5470 return page;
5471}
5472
Jérôme Glissec733a822017-09-08 16:11:54 -07005473#if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005474static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07005475 pte_t ptent, swp_entry_t *entry)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005476{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005477 struct page *page = NULL;
5478 swp_entry_t ent = pte_to_swp_entry(ptent);
5479
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005480 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005481 return NULL;
Jérôme Glissec733a822017-09-08 16:11:54 -07005482
5483 /*
5484 * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to
5485 * a device and because they are not accessible by CPU they are store
5486 * as special swap entry in the CPU page table.
5487 */
5488 if (is_device_private_entry(ent)) {
5489 page = device_private_entry_to_page(ent);
5490 /*
5491 * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have
5492 * a refcount of 1 when free (unlike normal page)
5493 */
5494 if (!page_ref_add_unless(page, 1, 1))
5495 return NULL;
5496 return page;
5497 }
5498
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005499 /*
5500 * Because lookup_swap_cache() updates some statistics counter,
5501 * we call find_get_page() with swapper_space directly.
5502 */
Huang Yingf6ab1f72016-10-07 17:00:21 -07005503 page = find_get_page(swap_address_space(ent), swp_offset(ent));
Johannes Weiner2d1c4982020-06-03 16:02:14 -07005504 entry->val = ent.val;
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005505
5506 return page;
5507}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005508#else
5509static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07005510 pte_t ptent, swp_entry_t *entry)
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005511{
5512 return NULL;
5513}
5514#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005515
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005516static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5517 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5518{
5519 struct page *page = NULL;
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005520 struct address_space *mapping;
5521 pgoff_t pgoff;
5522
5523 if (!vma->vm_file) /* anonymous vma */
5524 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005525 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005526 return NULL;
5527
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005528 mapping = vma->vm_file->f_mapping;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08005529 pgoff = linear_page_index(vma, addr);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005530
5531 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005532#ifdef CONFIG_SWAP
5533 /* shmem/tmpfs may report page out on swap: account for that too. */
Johannes Weiner139b6a62014-05-06 12:50:05 -07005534 if (shmem_mapping(mapping)) {
5535 page = find_get_entry(mapping, pgoff);
Matthew Wilcox3159f942017-11-03 13:30:42 -04005536 if (xa_is_value(page)) {
Johannes Weiner139b6a62014-05-06 12:50:05 -07005537 swp_entry_t swp = radix_to_swp_entry(page);
Johannes Weiner2d1c4982020-06-03 16:02:14 -07005538 *entry = swp;
Huang Yingf6ab1f72016-10-07 17:00:21 -07005539 page = find_get_page(swap_address_space(swp),
5540 swp_offset(swp));
Johannes Weiner139b6a62014-05-06 12:50:05 -07005541 }
5542 } else
5543 page = find_get_page(mapping, pgoff);
5544#else
5545 page = find_get_page(mapping, pgoff);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005546#endif
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005547 return page;
5548}
5549
Chen Gangb1b0dea2015-04-14 15:47:35 -07005550/**
5551 * mem_cgroup_move_account - move account of the page
5552 * @page: the page
Li RongQing25843c22016-07-26 15:26:56 -07005553 * @compound: charge the page as compound or small page
Chen Gangb1b0dea2015-04-14 15:47:35 -07005554 * @from: mem_cgroup which the page is moved from.
5555 * @to: mem_cgroup which the page is moved to. @from != @to.
5556 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08005557 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07005558 *
5559 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
5560 * from old cgroup.
5561 */
5562static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005563 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07005564 struct mem_cgroup *from,
5565 struct mem_cgroup *to)
5566{
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005567 struct lruvec *from_vec, *to_vec;
5568 struct pglist_data *pgdat;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005569 unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07005570 int ret;
5571
5572 VM_BUG_ON(from == to);
5573 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005574 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07005575
5576 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005577 * Prevent mem_cgroup_migrate() from looking at
Hugh Dickins45637ba2015-11-05 18:49:40 -08005578 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07005579 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005580 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07005581 if (!trylock_page(page))
5582 goto out;
5583
5584 ret = -EINVAL;
5585 if (page->mem_cgroup != from)
5586 goto out_unlock;
5587
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005588 pgdat = page_pgdat(page);
Johannes Weiner867e5e12019-11-30 17:55:34 -08005589 from_vec = mem_cgroup_lruvec(from, pgdat);
5590 to_vec = mem_cgroup_lruvec(to, pgdat);
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005591
Johannes Weinerabb242f2020-06-03 16:01:28 -07005592 lock_page_memcg(page);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005593
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07005594 if (PageAnon(page)) {
5595 if (page_mapped(page)) {
5596 __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages);
5597 __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages);
Johannes Weiner468c3982020-06-03 16:02:01 -07005598 if (PageTransHuge(page)) {
5599 __mod_lruvec_state(from_vec, NR_ANON_THPS,
5600 -nr_pages);
5601 __mod_lruvec_state(to_vec, NR_ANON_THPS,
5602 nr_pages);
5603 }
5604
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07005605 }
5606 } else {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07005607 __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages);
5608 __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages);
5609
5610 if (PageSwapBacked(page)) {
5611 __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages);
5612 __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages);
5613 }
5614
Johannes Weiner49e50d22020-06-03 16:01:47 -07005615 if (page_mapped(page)) {
5616 __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages);
5617 __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages);
5618 }
Chen Gangb1b0dea2015-04-14 15:47:35 -07005619
Johannes Weiner49e50d22020-06-03 16:01:47 -07005620 if (PageDirty(page)) {
5621 struct address_space *mapping = page_mapping(page);
Greg Thelenc4843a72015-05-22 17:13:16 -04005622
Johannes Weiner49e50d22020-06-03 16:01:47 -07005623 if (mapping_cap_account_dirty(mapping)) {
5624 __mod_lruvec_state(from_vec, NR_FILE_DIRTY,
5625 -nr_pages);
5626 __mod_lruvec_state(to_vec, NR_FILE_DIRTY,
5627 nr_pages);
5628 }
Greg Thelenc4843a72015-05-22 17:13:16 -04005629 }
5630 }
5631
Chen Gangb1b0dea2015-04-14 15:47:35 -07005632 if (PageWriteback(page)) {
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005633 __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages);
5634 __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005635 }
5636
5637 /*
Johannes Weinerabb242f2020-06-03 16:01:28 -07005638 * All state has been migrated, let's switch to the new memcg.
5639 *
Chen Gangb1b0dea2015-04-14 15:47:35 -07005640 * It is safe to change page->mem_cgroup here because the page
Johannes Weinerabb242f2020-06-03 16:01:28 -07005641 * is referenced, charged, isolated, and locked: we can't race
5642 * with (un)charging, migration, LRU putback, or anything else
5643 * that would rely on a stable page->mem_cgroup.
5644 *
5645 * Note that lock_page_memcg is a memcg lock, not a page lock,
5646 * to save space. As soon as we switch page->mem_cgroup to a
5647 * new memcg that isn't locked, the above state can change
5648 * concurrently again. Make sure we're truly done with it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07005649 */
Johannes Weinerabb242f2020-06-03 16:01:28 -07005650 smp_mb();
Chen Gangb1b0dea2015-04-14 15:47:35 -07005651
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07005652 css_get(&to->css);
5653 css_put(&from->css);
5654
5655 page->mem_cgroup = to;
Yang Shi87eaceb2019-09-23 15:38:15 -07005656
Johannes Weinerabb242f2020-06-03 16:01:28 -07005657 __unlock_page_memcg(from);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005658
5659 ret = 0;
5660
5661 local_irq_disable();
Johannes Weiner3fba69a2020-06-03 16:01:31 -07005662 mem_cgroup_charge_statistics(to, page, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005663 memcg_check_events(to, page);
Johannes Weiner3fba69a2020-06-03 16:01:31 -07005664 mem_cgroup_charge_statistics(from, page, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005665 memcg_check_events(from, page);
5666 local_irq_enable();
5667out_unlock:
5668 unlock_page(page);
5669out:
5670 return ret;
5671}
5672
Li RongQing7cf78062016-05-27 14:27:46 -07005673/**
5674 * get_mctgt_type - get target type of moving charge
5675 * @vma: the vma the pte to be checked belongs
5676 * @addr: the address corresponding to the pte to be checked
5677 * @ptent: the pte to be checked
5678 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5679 *
5680 * Returns
5681 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5682 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5683 * move charge. if @target is not NULL, the page is stored in target->page
5684 * with extra refcnt got(Callers should handle it).
5685 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5686 * target for charge migration. if @target is not NULL, the entry is stored
5687 * in target->ent.
Christoph Hellwig25b29952019-06-13 22:50:49 +02005688 * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE
5689 * (so ZONE_DEVICE page and thus not on the lru).
Jérôme Glissedf6ad692017-09-08 16:12:24 -07005690 * For now we such page is charge like a regular page would be as for all
5691 * intent and purposes it is just special memory taking the place of a
5692 * regular page.
Jérôme Glissec733a822017-09-08 16:11:54 -07005693 *
5694 * See Documentations/vm/hmm.txt and include/linux/hmm.h
Li RongQing7cf78062016-05-27 14:27:46 -07005695 *
5696 * Called with pte lock held.
5697 */
5698
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005699static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005700 unsigned long addr, pte_t ptent, union mc_target *target)
5701{
Daisuke Nishimura02491442010-03-10 15:22:17 -08005702 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005703 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005704 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005705
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005706 if (pte_present(ptent))
5707 page = mc_handle_present_pte(vma, addr, ptent);
5708 else if (is_swap_pte(ptent))
Li RongQing48406ef2016-07-26 15:22:14 -07005709 page = mc_handle_swap_pte(vma, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08005710 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005711 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005712
5713 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005714 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005715 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005716 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005717 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08005718 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005719 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08005720 */
Johannes Weiner1306a852014-12-10 15:44:52 -08005721 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005722 ret = MC_TARGET_PAGE;
Christoph Hellwig25b29952019-06-13 22:50:49 +02005723 if (is_device_private_page(page))
Jérôme Glissec733a822017-09-08 16:11:54 -07005724 ret = MC_TARGET_DEVICE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005725 if (target)
5726 target->page = page;
5727 }
5728 if (!ret || !target)
5729 put_page(page);
5730 }
Huang Ying3e14a572017-09-06 16:22:37 -07005731 /*
5732 * There is a swap entry and a page doesn't exist or isn't charged.
5733 * But we cannot move a tail-page in a THP.
5734 */
5735 if (ent.val && !ret && (!page || !PageTransCompound(page)) &&
Li Zefan34c00c32013-09-23 16:56:01 +08005736 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07005737 ret = MC_TARGET_SWAP;
5738 if (target)
5739 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005740 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005741 return ret;
5742}
5743
Naoya Horiguchi12724852012-03-21 16:34:28 -07005744#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5745/*
Huang Yingd6810d72017-09-06 16:22:45 -07005746 * We don't consider PMD mapped swapping or file mapped pages because THP does
5747 * not support them for now.
Naoya Horiguchi12724852012-03-21 16:34:28 -07005748 * Caller should make sure that pmd_trans_huge(pmd) is true.
5749 */
5750static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5751 unsigned long addr, pmd_t pmd, union mc_target *target)
5752{
5753 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005754 enum mc_target_type ret = MC_TARGET_NONE;
5755
Zi Yan84c3fc42017-09-08 16:11:01 -07005756 if (unlikely(is_swap_pmd(pmd))) {
5757 VM_BUG_ON(thp_migration_supported() &&
5758 !is_pmd_migration_entry(pmd));
5759 return ret;
5760 }
Naoya Horiguchi12724852012-03-21 16:34:28 -07005761 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08005762 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005763 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07005764 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08005765 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005766 ret = MC_TARGET_PAGE;
5767 if (target) {
5768 get_page(page);
5769 target->page = page;
5770 }
5771 }
5772 return ret;
5773}
5774#else
5775static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5776 unsigned long addr, pmd_t pmd, union mc_target *target)
5777{
5778 return MC_TARGET_NONE;
5779}
5780#endif
5781
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005782static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5783 unsigned long addr, unsigned long end,
5784 struct mm_walk *walk)
5785{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005786 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005787 pte_t *pte;
5788 spinlock_t *ptl;
5789
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08005790 ptl = pmd_trans_huge_lock(pmd, vma);
5791 if (ptl) {
Jérôme Glissec733a822017-09-08 16:11:54 -07005792 /*
5793 * Note their can not be MC_TARGET_DEVICE for now as we do not
Christoph Hellwig25b29952019-06-13 22:50:49 +02005794 * support transparent huge page with MEMORY_DEVICE_PRIVATE but
5795 * this might change.
Jérôme Glissec733a822017-09-08 16:11:54 -07005796 */
Naoya Horiguchi12724852012-03-21 16:34:28 -07005797 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
5798 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08005799 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005800 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005801 }
Dave Hansen03319322011-03-22 16:32:56 -07005802
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005803 if (pmd_trans_unstable(pmd))
5804 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005805 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5806 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005807 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005808 mc.precharge++; /* increment precharge temporarily */
5809 pte_unmap_unlock(pte - 1, ptl);
5810 cond_resched();
5811
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005812 return 0;
5813}
5814
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02005815static const struct mm_walk_ops precharge_walk_ops = {
5816 .pmd_entry = mem_cgroup_count_precharge_pte_range,
5817};
5818
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005819static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5820{
5821 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005822
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07005823 mmap_read_lock(mm);
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02005824 walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL);
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07005825 mmap_read_unlock(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005826
5827 precharge = mc.precharge;
5828 mc.precharge = 0;
5829
5830 return precharge;
5831}
5832
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005833static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5834{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005835 unsigned long precharge = mem_cgroup_count_precharge(mm);
5836
5837 VM_BUG_ON(mc.moving_task);
5838 mc.moving_task = current;
5839 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005840}
5841
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005842/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5843static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005844{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005845 struct mem_cgroup *from = mc.from;
5846 struct mem_cgroup *to = mc.to;
5847
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005848 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005849 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005850 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005851 mc.precharge = 0;
5852 }
5853 /*
5854 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5855 * we must uncharge here.
5856 */
5857 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005858 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005859 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005860 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005861 /* we must fixup refcnts and charges */
5862 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005863 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04005864 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005865 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005866
Vladimir Davydov615d66c2016-08-11 15:33:03 -07005867 mem_cgroup_id_put_many(mc.from, mc.moved_swap);
5868
Johannes Weiner05b84302014-08-06 16:05:59 -07005869 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005870 * we charged both to->memory and to->memsw, so we
5871 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07005872 */
Johannes Weinerce00a962014-09-05 08:43:57 -04005873 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005874 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005875
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005876 mc.moved_swap = 0;
5877 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005878 memcg_oom_recover(from);
5879 memcg_oom_recover(to);
5880 wake_up_all(&mc.waitq);
5881}
5882
5883static void mem_cgroup_clear_mc(void)
5884{
Tejun Heo264a0ae2016-04-21 19:09:02 -04005885 struct mm_struct *mm = mc.mm;
5886
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005887 /*
5888 * we must clear moving_task before waking up waiters at the end of
5889 * task migration.
5890 */
5891 mc.moving_task = NULL;
5892 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005893 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005894 mc.from = NULL;
5895 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04005896 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005897 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04005898
5899 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005900}
5901
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005902static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005903{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005904 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07005905 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07005906 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04005907 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07005908 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005909 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07005910 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005911
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005912 /* charge immigration isn't supported on the default hierarchy */
5913 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07005914 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005915
Tejun Heo4530edd2015-09-11 15:00:19 -04005916 /*
5917 * Multi-process migrations only happen on the default hierarchy
5918 * where charge immigration is not used. Perform charge
5919 * immigration if @tset contains a leader and whine if there are
5920 * multiple.
5921 */
5922 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005923 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04005924 WARN_ON_ONCE(p);
5925 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005926 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04005927 }
5928 if (!p)
5929 return 0;
5930
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005931 /*
5932 * We are now commited to this value whatever it is. Changes in this
5933 * tunable will only affect upcoming migrations, not the current one.
5934 * So we need to save it, and keep it going.
5935 */
5936 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
5937 if (!move_flags)
5938 return 0;
5939
Tejun Heo9f2115f2015-09-08 15:01:10 -07005940 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005941
Tejun Heo9f2115f2015-09-08 15:01:10 -07005942 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08005943
Tejun Heo9f2115f2015-09-08 15:01:10 -07005944 mm = get_task_mm(p);
5945 if (!mm)
5946 return 0;
5947 /* We move charges only when we move a owner of the mm */
5948 if (mm->owner == p) {
5949 VM_BUG_ON(mc.from);
5950 VM_BUG_ON(mc.to);
5951 VM_BUG_ON(mc.precharge);
5952 VM_BUG_ON(mc.moved_charge);
5953 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005954
Tejun Heo9f2115f2015-09-08 15:01:10 -07005955 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04005956 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07005957 mc.from = from;
5958 mc.to = memcg;
5959 mc.flags = move_flags;
5960 spin_unlock(&mc.lock);
5961 /* We set mc.moving_task later */
5962
5963 ret = mem_cgroup_precharge_mc(mm);
5964 if (ret)
5965 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04005966 } else {
5967 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005968 }
5969 return ret;
5970}
5971
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005972static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005973{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08005974 if (mc.to)
5975 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005976}
5977
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005978static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5979 unsigned long addr, unsigned long end,
5980 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005981{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005982 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005983 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005984 pte_t *pte;
5985 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005986 enum mc_target_type target_type;
5987 union mc_target target;
5988 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005989
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08005990 ptl = pmd_trans_huge_lock(pmd, vma);
5991 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07005992 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08005993 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07005994 return 0;
5995 }
5996 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
5997 if (target_type == MC_TARGET_PAGE) {
5998 page = target.page;
5999 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08006000 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08006001 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006002 mc.precharge -= HPAGE_PMD_NR;
6003 mc.moved_charge += HPAGE_PMD_NR;
6004 }
6005 putback_lru_page(page);
6006 }
6007 put_page(page);
Jérôme Glissec733a822017-09-08 16:11:54 -07006008 } else if (target_type == MC_TARGET_DEVICE) {
6009 page = target.page;
6010 if (!mem_cgroup_move_account(page, true,
6011 mc.from, mc.to)) {
6012 mc.precharge -= HPAGE_PMD_NR;
6013 mc.moved_charge += HPAGE_PMD_NR;
6014 }
6015 put_page(page);
Naoya Horiguchi12724852012-03-21 16:34:28 -07006016 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006017 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006018 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006019 }
6020
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006021 if (pmd_trans_unstable(pmd))
6022 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006023retry:
6024 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6025 for (; addr != end; addr += PAGE_SIZE) {
6026 pte_t ptent = *(pte++);
Jérôme Glissec733a822017-09-08 16:11:54 -07006027 bool device = false;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006028 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006029
6030 if (!mc.precharge)
6031 break;
6032
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006033 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Jérôme Glissec733a822017-09-08 16:11:54 -07006034 case MC_TARGET_DEVICE:
6035 device = true;
Joe Perchese4a9bc52020-04-06 20:08:39 -07006036 fallthrough;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006037 case MC_TARGET_PAGE:
6038 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08006039 /*
6040 * We can have a part of the split pmd here. Moving it
6041 * can be done but it would be too convoluted so simply
6042 * ignore such a partial THP and keep it in original
6043 * memcg. There should be somebody mapping the head.
6044 */
6045 if (PageTransCompound(page))
6046 goto put;
Jérôme Glissec733a822017-09-08 16:11:54 -07006047 if (!device && isolate_lru_page(page))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006048 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08006049 if (!mem_cgroup_move_account(page, false,
6050 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006051 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006052 /* we uncharge from mc.from later. */
6053 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006054 }
Jérôme Glissec733a822017-09-08 16:11:54 -07006055 if (!device)
6056 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006057put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006058 put_page(page);
6059 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006060 case MC_TARGET_SWAP:
6061 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07006062 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08006063 mc.precharge--;
Hugh Dickins8d22a932020-07-23 21:15:24 -07006064 mem_cgroup_id_get_many(mc.to, 1);
6065 /* we fixup other refcnts and charges later. */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006066 mc.moved_swap++;
6067 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08006068 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006069 default:
6070 break;
6071 }
6072 }
6073 pte_unmap_unlock(pte - 1, ptl);
6074 cond_resched();
6075
6076 if (addr != end) {
6077 /*
6078 * We have consumed all precharges we got in can_attach().
6079 * We try charge one by one, but don't do any additional
6080 * charges to mc.to if we have failed in charge once in attach()
6081 * phase.
6082 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006083 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006084 if (!ret)
6085 goto retry;
6086 }
6087
6088 return ret;
6089}
6090
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02006091static const struct mm_walk_ops charge_walk_ops = {
6092 .pmd_entry = mem_cgroup_move_charge_pte_range,
6093};
6094
Tejun Heo264a0ae2016-04-21 19:09:02 -04006095static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006096{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006097 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08006098 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07006099 * Signal lock_page_memcg() to take the memcg's move_lock
6100 * while we're moving its pages to another memcg. Then wait
6101 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08006102 */
6103 atomic_inc(&mc.from->moving_account);
6104 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006105retry:
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07006106 if (unlikely(!mmap_read_trylock(mc.mm))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006107 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07006108 * Someone who are holding the mmap_lock might be waiting in
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006109 * waitq. So we cancel all extra charges, wake up all waiters,
6110 * and retry. Because we cancel precharges, we might not be able
6111 * to move enough charges, but moving charge is a best-effort
6112 * feature anyway, so it wouldn't be a big problem.
6113 */
6114 __mem_cgroup_clear_mc();
6115 cond_resched();
6116 goto retry;
6117 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08006118 /*
6119 * When we have consumed all precharges and failed in doing
6120 * additional charge, the page walk just aborts.
6121 */
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02006122 walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops,
6123 NULL);
James Morse0247f3f2016-10-07 17:00:12 -07006124
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07006125 mmap_read_unlock(mc.mm);
Johannes Weiner312722c2014-12-10 15:44:25 -08006126 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006127}
6128
Tejun Heo264a0ae2016-04-21 19:09:02 -04006129static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08006130{
Tejun Heo264a0ae2016-04-21 19:09:02 -04006131 if (mc.to) {
6132 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006133 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04006134 }
Balbir Singh67e465a2008-02-07 00:13:54 -08006135}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006136#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006137static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006138{
6139 return 0;
6140}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006141static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006142{
6143}
Tejun Heo264a0ae2016-04-21 19:09:02 -04006144static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006145{
6146}
6147#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08006148
Tejun Heof00baae2013-04-15 13:41:15 -07006149/*
6150 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04006151 * to verify whether we're attached to the default hierarchy on each mount
6152 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07006153 */
Tejun Heoeb954192013-08-08 20:11:23 -04006154static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07006155{
6156 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04006157 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07006158 * guarantees that @root doesn't have any children, so turning it
6159 * on for the root memcg is enough.
6160 */
Tejun Heo9e10a132015-09-18 11:56:28 -04006161 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee5902015-03-12 16:26:19 -07006162 root_mem_cgroup->use_hierarchy = true;
6163 else
6164 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07006165}
6166
Chris Down677dc972019-03-05 15:45:55 -08006167static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value)
6168{
6169 if (value == PAGE_COUNTER_MAX)
6170 seq_puts(m, "max\n");
6171 else
6172 seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE);
6173
6174 return 0;
6175}
6176
Johannes Weiner241994ed2015-02-11 15:26:06 -08006177static u64 memory_current_read(struct cgroup_subsys_state *css,
6178 struct cftype *cft)
6179{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08006180 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6181
6182 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006183}
6184
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006185static int memory_min_show(struct seq_file *m, void *v)
6186{
Chris Down677dc972019-03-05 15:45:55 -08006187 return seq_puts_memcg_tunable(m,
6188 READ_ONCE(mem_cgroup_from_seq(m)->memory.min));
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006189}
6190
6191static ssize_t memory_min_write(struct kernfs_open_file *of,
6192 char *buf, size_t nbytes, loff_t off)
6193{
6194 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6195 unsigned long min;
6196 int err;
6197
6198 buf = strstrip(buf);
6199 err = page_counter_memparse(buf, "max", &min);
6200 if (err)
6201 return err;
6202
6203 page_counter_set_min(&memcg->memory, min);
6204
6205 return nbytes;
6206}
6207
Johannes Weiner241994ed2015-02-11 15:26:06 -08006208static int memory_low_show(struct seq_file *m, void *v)
6209{
Chris Down677dc972019-03-05 15:45:55 -08006210 return seq_puts_memcg_tunable(m,
6211 READ_ONCE(mem_cgroup_from_seq(m)->memory.low));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006212}
6213
6214static ssize_t memory_low_write(struct kernfs_open_file *of,
6215 char *buf, size_t nbytes, loff_t off)
6216{
6217 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6218 unsigned long low;
6219 int err;
6220
6221 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08006222 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006223 if (err)
6224 return err;
6225
Roman Gushchin23067152018-06-07 17:06:22 -07006226 page_counter_set_low(&memcg->memory, low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006227
6228 return nbytes;
6229}
6230
6231static int memory_high_show(struct seq_file *m, void *v)
6232{
Jakub Kicinskid1663a92020-06-01 21:49:49 -07006233 return seq_puts_memcg_tunable(m,
6234 READ_ONCE(mem_cgroup_from_seq(m)->memory.high));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006235}
6236
6237static ssize_t memory_high_write(struct kernfs_open_file *of,
6238 char *buf, size_t nbytes, loff_t off)
6239{
6240 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Chris Downd977aa92020-08-06 23:21:58 -07006241 unsigned int nr_retries = MAX_RECLAIM_RETRIES;
Johannes Weiner8c8c3832019-11-30 17:50:09 -08006242 bool drained = false;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006243 unsigned long high;
6244 int err;
6245
6246 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08006247 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006248 if (err)
6249 return err;
6250
Johannes Weiner8c8c3832019-11-30 17:50:09 -08006251 for (;;) {
6252 unsigned long nr_pages = page_counter_read(&memcg->memory);
6253 unsigned long reclaimed;
Johannes Weiner588083b2016-03-17 14:20:25 -07006254
Johannes Weiner8c8c3832019-11-30 17:50:09 -08006255 if (nr_pages <= high)
6256 break;
6257
6258 if (signal_pending(current))
6259 break;
6260
6261 if (!drained) {
6262 drain_all_stock(memcg);
6263 drained = true;
6264 continue;
6265 }
6266
6267 reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
6268 GFP_KERNEL, true);
6269
6270 if (!reclaimed && !nr_retries--)
6271 break;
6272 }
6273
Roman Gushchin536d3bf22020-08-06 23:21:51 -07006274 page_counter_set_high(&memcg->memory, high);
6275
Johannes Weiner19ce33a2020-08-06 23:22:12 -07006276 memcg_wb_domain_size_changed(memcg);
6277
Johannes Weiner241994ed2015-02-11 15:26:06 -08006278 return nbytes;
6279}
6280
6281static int memory_max_show(struct seq_file *m, void *v)
6282{
Chris Down677dc972019-03-05 15:45:55 -08006283 return seq_puts_memcg_tunable(m,
6284 READ_ONCE(mem_cgroup_from_seq(m)->memory.max));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006285}
6286
6287static ssize_t memory_max_write(struct kernfs_open_file *of,
6288 char *buf, size_t nbytes, loff_t off)
6289{
6290 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Chris Downd977aa92020-08-06 23:21:58 -07006291 unsigned int nr_reclaims = MAX_RECLAIM_RETRIES;
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006292 bool drained = false;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006293 unsigned long max;
6294 int err;
6295
6296 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08006297 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006298 if (err)
6299 return err;
6300
Roman Gushchinbbec2e12018-06-07 17:06:18 -07006301 xchg(&memcg->memory.max, max);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006302
6303 for (;;) {
6304 unsigned long nr_pages = page_counter_read(&memcg->memory);
6305
6306 if (nr_pages <= max)
6307 break;
6308
Johannes Weiner7249c9f2019-11-30 17:50:06 -08006309 if (signal_pending(current))
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006310 break;
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006311
6312 if (!drained) {
6313 drain_all_stock(memcg);
6314 drained = true;
6315 continue;
6316 }
6317
6318 if (nr_reclaims) {
6319 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
6320 GFP_KERNEL, true))
6321 nr_reclaims--;
6322 continue;
6323 }
6324
Johannes Weinere27be242018-04-10 16:29:45 -07006325 memcg_memory_event(memcg, MEMCG_OOM);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006326 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
6327 break;
6328 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08006329
Tejun Heo2529bb32015-05-22 18:23:34 -04006330 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006331 return nbytes;
6332}
6333
Shakeel Butt1e577f92019-07-11 20:55:55 -07006334static void __memory_events_show(struct seq_file *m, atomic_long_t *events)
6335{
6336 seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW]));
6337 seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH]));
6338 seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX]));
6339 seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM]));
6340 seq_printf(m, "oom_kill %lu\n",
6341 atomic_long_read(&events[MEMCG_OOM_KILL]));
6342}
6343
Johannes Weiner241994ed2015-02-11 15:26:06 -08006344static int memory_events_show(struct seq_file *m, void *v)
6345{
Chris Downaa9694b2019-03-05 15:45:52 -08006346 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006347
Shakeel Butt1e577f92019-07-11 20:55:55 -07006348 __memory_events_show(m, memcg->memory_events);
6349 return 0;
6350}
Johannes Weiner241994ed2015-02-11 15:26:06 -08006351
Shakeel Butt1e577f92019-07-11 20:55:55 -07006352static int memory_events_local_show(struct seq_file *m, void *v)
6353{
6354 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
6355
6356 __memory_events_show(m, memcg->memory_events_local);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006357 return 0;
6358}
6359
Johannes Weiner587d9f72016-01-20 15:03:19 -08006360static int memory_stat_show(struct seq_file *m, void *v)
6361{
Chris Downaa9694b2019-03-05 15:45:52 -08006362 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Johannes Weinerc8713d02019-07-11 20:55:59 -07006363 char *buf;
Johannes Weiner587d9f72016-01-20 15:03:19 -08006364
Johannes Weinerc8713d02019-07-11 20:55:59 -07006365 buf = memory_stat_format(memcg);
6366 if (!buf)
6367 return -ENOMEM;
6368 seq_puts(m, buf);
6369 kfree(buf);
Johannes Weiner587d9f72016-01-20 15:03:19 -08006370 return 0;
6371}
6372
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006373static int memory_oom_group_show(struct seq_file *m, void *v)
6374{
Chris Downaa9694b2019-03-05 15:45:52 -08006375 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006376
6377 seq_printf(m, "%d\n", memcg->oom_group);
6378
6379 return 0;
6380}
6381
6382static ssize_t memory_oom_group_write(struct kernfs_open_file *of,
6383 char *buf, size_t nbytes, loff_t off)
6384{
6385 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6386 int ret, oom_group;
6387
6388 buf = strstrip(buf);
6389 if (!buf)
6390 return -EINVAL;
6391
6392 ret = kstrtoint(buf, 0, &oom_group);
6393 if (ret)
6394 return ret;
6395
6396 if (oom_group != 0 && oom_group != 1)
6397 return -EINVAL;
6398
6399 memcg->oom_group = oom_group;
6400
6401 return nbytes;
6402}
6403
Johannes Weiner241994ed2015-02-11 15:26:06 -08006404static struct cftype memory_files[] = {
6405 {
6406 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08006407 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994ed2015-02-11 15:26:06 -08006408 .read_u64 = memory_current_read,
6409 },
6410 {
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006411 .name = "min",
6412 .flags = CFTYPE_NOT_ON_ROOT,
6413 .seq_show = memory_min_show,
6414 .write = memory_min_write,
6415 },
6416 {
Johannes Weiner241994ed2015-02-11 15:26:06 -08006417 .name = "low",
6418 .flags = CFTYPE_NOT_ON_ROOT,
6419 .seq_show = memory_low_show,
6420 .write = memory_low_write,
6421 },
6422 {
6423 .name = "high",
6424 .flags = CFTYPE_NOT_ON_ROOT,
6425 .seq_show = memory_high_show,
6426 .write = memory_high_write,
6427 },
6428 {
6429 .name = "max",
6430 .flags = CFTYPE_NOT_ON_ROOT,
6431 .seq_show = memory_max_show,
6432 .write = memory_max_write,
6433 },
6434 {
6435 .name = "events",
6436 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04006437 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994ed2015-02-11 15:26:06 -08006438 .seq_show = memory_events_show,
6439 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08006440 {
Shakeel Butt1e577f92019-07-11 20:55:55 -07006441 .name = "events.local",
6442 .flags = CFTYPE_NOT_ON_ROOT,
6443 .file_offset = offsetof(struct mem_cgroup, events_local_file),
6444 .seq_show = memory_events_local_show,
6445 },
6446 {
Johannes Weiner587d9f72016-01-20 15:03:19 -08006447 .name = "stat",
Johannes Weiner587d9f72016-01-20 15:03:19 -08006448 .seq_show = memory_stat_show,
6449 },
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006450 {
6451 .name = "oom.group",
6452 .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE,
6453 .seq_show = memory_oom_group_show,
6454 .write = memory_oom_group_write,
6455 },
Johannes Weiner241994ed2015-02-11 15:26:06 -08006456 { } /* terminate */
6457};
6458
Tejun Heo073219e2014-02-08 10:36:58 -05006459struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08006460 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08006461 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08006462 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08006463 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08006464 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04006465 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006466 .can_attach = mem_cgroup_can_attach,
6467 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04006468 .post_attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07006469 .bind = mem_cgroup_bind,
Johannes Weiner241994ed2015-02-11 15:26:06 -08006470 .dfl_cftypes = memory_files,
6471 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006472 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006473};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006474
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006475/*
6476 * This function calculates an individual cgroup's effective
6477 * protection which is derived from its own memory.min/low, its
6478 * parent's and siblings' settings, as well as the actual memory
6479 * distribution in the tree.
6480 *
6481 * The following rules apply to the effective protection values:
6482 *
6483 * 1. At the first level of reclaim, effective protection is equal to
6484 * the declared protection in memory.min and memory.low.
6485 *
6486 * 2. To enable safe delegation of the protection configuration, at
6487 * subsequent levels the effective protection is capped to the
6488 * parent's effective protection.
6489 *
6490 * 3. To make complex and dynamic subtrees easier to configure, the
6491 * user is allowed to overcommit the declared protection at a given
6492 * level. If that is the case, the parent's effective protection is
6493 * distributed to the children in proportion to how much protection
6494 * they have declared and how much of it they are utilizing.
6495 *
6496 * This makes distribution proportional, but also work-conserving:
6497 * if one cgroup claims much more protection than it uses memory,
6498 * the unused remainder is available to its siblings.
6499 *
6500 * 4. Conversely, when the declared protection is undercommitted at a
6501 * given level, the distribution of the larger parental protection
6502 * budget is NOT proportional. A cgroup's protection from a sibling
6503 * is capped to its own memory.min/low setting.
6504 *
Johannes Weiner8a931f82020-04-01 21:07:07 -07006505 * 5. However, to allow protecting recursive subtrees from each other
6506 * without having to declare each individual cgroup's fixed share
6507 * of the ancestor's claim to protection, any unutilized -
6508 * "floating" - protection from up the tree is distributed in
6509 * proportion to each cgroup's *usage*. This makes the protection
6510 * neutral wrt sibling cgroups and lets them compete freely over
6511 * the shared parental protection budget, but it protects the
6512 * subtree as a whole from neighboring subtrees.
6513 *
6514 * Note that 4. and 5. are not in conflict: 4. is about protecting
6515 * against immediate siblings whereas 5. is about protecting against
6516 * neighboring subtrees.
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006517 */
6518static unsigned long effective_protection(unsigned long usage,
Johannes Weiner8a931f82020-04-01 21:07:07 -07006519 unsigned long parent_usage,
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006520 unsigned long setting,
6521 unsigned long parent_effective,
6522 unsigned long siblings_protected)
6523{
6524 unsigned long protected;
Johannes Weiner8a931f82020-04-01 21:07:07 -07006525 unsigned long ep;
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006526
6527 protected = min(usage, setting);
6528 /*
6529 * If all cgroups at this level combined claim and use more
6530 * protection then what the parent affords them, distribute
6531 * shares in proportion to utilization.
6532 *
6533 * We are using actual utilization rather than the statically
6534 * claimed protection in order to be work-conserving: claimed
6535 * but unused protection is available to siblings that would
6536 * otherwise get a smaller chunk than what they claimed.
6537 */
6538 if (siblings_protected > parent_effective)
6539 return protected * parent_effective / siblings_protected;
6540
6541 /*
6542 * Ok, utilized protection of all children is within what the
6543 * parent affords them, so we know whatever this child claims
6544 * and utilizes is effectively protected.
6545 *
6546 * If there is unprotected usage beyond this value, reclaim
6547 * will apply pressure in proportion to that amount.
6548 *
6549 * If there is unutilized protection, the cgroup will be fully
6550 * shielded from reclaim, but we do return a smaller value for
6551 * protection than what the group could enjoy in theory. This
6552 * is okay. With the overcommit distribution above, effective
6553 * protection is always dependent on how memory is actually
6554 * consumed among the siblings anyway.
6555 */
Johannes Weiner8a931f82020-04-01 21:07:07 -07006556 ep = protected;
6557
6558 /*
6559 * If the children aren't claiming (all of) the protection
6560 * afforded to them by the parent, distribute the remainder in
6561 * proportion to the (unprotected) memory of each cgroup. That
6562 * way, cgroups that aren't explicitly prioritized wrt each
6563 * other compete freely over the allowance, but they are
6564 * collectively protected from neighboring trees.
6565 *
6566 * We're using unprotected memory for the weight so that if
6567 * some cgroups DO claim explicit protection, we don't protect
6568 * the same bytes twice.
Johannes Weinercd324ed2020-06-25 20:30:16 -07006569 *
6570 * Check both usage and parent_usage against the respective
6571 * protected values. One should imply the other, but they
6572 * aren't read atomically - make sure the division is sane.
Johannes Weiner8a931f82020-04-01 21:07:07 -07006573 */
6574 if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT))
6575 return ep;
Johannes Weinercd324ed2020-06-25 20:30:16 -07006576 if (parent_effective > siblings_protected &&
6577 parent_usage > siblings_protected &&
6578 usage > protected) {
Johannes Weiner8a931f82020-04-01 21:07:07 -07006579 unsigned long unclaimed;
6580
6581 unclaimed = parent_effective - siblings_protected;
6582 unclaimed *= usage - protected;
6583 unclaimed /= parent_usage - siblings_protected;
6584
6585 ep += unclaimed;
6586 }
6587
6588 return ep;
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006589}
6590
Johannes Weiner241994ed2015-02-11 15:26:06 -08006591/**
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006592 * mem_cgroup_protected - check if memory consumption is in the normal range
Sean Christopherson34c81052017-07-10 15:48:05 -07006593 * @root: the top ancestor of the sub-tree being checked
Johannes Weiner241994ed2015-02-11 15:26:06 -08006594 * @memcg: the memory cgroup to check
6595 *
Roman Gushchin23067152018-06-07 17:06:22 -07006596 * WARNING: This function is not stateless! It can only be used as part
6597 * of a top-down tree iteration, not for isolated queries.
Johannes Weiner241994ed2015-02-11 15:26:06 -08006598 */
Chris Down45c7f7e2020-08-06 23:22:05 -07006599void mem_cgroup_calculate_protection(struct mem_cgroup *root,
6600 struct mem_cgroup *memcg)
Johannes Weiner241994ed2015-02-11 15:26:06 -08006601{
Johannes Weiner8a931f82020-04-01 21:07:07 -07006602 unsigned long usage, parent_usage;
Roman Gushchin23067152018-06-07 17:06:22 -07006603 struct mem_cgroup *parent;
6604
Johannes Weiner241994ed2015-02-11 15:26:06 -08006605 if (mem_cgroup_disabled())
Chris Down45c7f7e2020-08-06 23:22:05 -07006606 return;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006607
Sean Christopherson34c81052017-07-10 15:48:05 -07006608 if (!root)
6609 root = root_mem_cgroup;
Yafang Shao22f74962020-08-06 23:22:01 -07006610
6611 /*
6612 * Effective values of the reclaim targets are ignored so they
6613 * can be stale. Have a look at mem_cgroup_protection for more
6614 * details.
6615 * TODO: calculation should be more robust so that we do not need
6616 * that special casing.
6617 */
Sean Christopherson34c81052017-07-10 15:48:05 -07006618 if (memcg == root)
Chris Down45c7f7e2020-08-06 23:22:05 -07006619 return;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006620
Roman Gushchin23067152018-06-07 17:06:22 -07006621 usage = page_counter_read(&memcg->memory);
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006622 if (!usage)
Chris Down45c7f7e2020-08-06 23:22:05 -07006623 return;
Sean Christopherson34c81052017-07-10 15:48:05 -07006624
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006625 parent = parent_mem_cgroup(memcg);
Roman Gushchindf2a4192018-06-14 15:26:17 -07006626 /* No parent means a non-hierarchical mode on v1 memcg */
6627 if (!parent)
Chris Down45c7f7e2020-08-06 23:22:05 -07006628 return;
Roman Gushchindf2a4192018-06-14 15:26:17 -07006629
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006630 if (parent == root) {
Chris Downc3d53202020-04-01 21:07:27 -07006631 memcg->memory.emin = READ_ONCE(memcg->memory.min);
Chris Down03960e32020-06-25 20:30:22 -07006632 memcg->memory.elow = READ_ONCE(memcg->memory.low);
Chris Down45c7f7e2020-08-06 23:22:05 -07006633 return;
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006634 }
6635
Johannes Weiner8a931f82020-04-01 21:07:07 -07006636 parent_usage = page_counter_read(&parent->memory);
6637
Chris Downb3a78222020-04-01 21:07:33 -07006638 WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage,
Chris Downc3d53202020-04-01 21:07:27 -07006639 READ_ONCE(memcg->memory.min),
6640 READ_ONCE(parent->memory.emin),
Chris Downb3a78222020-04-01 21:07:33 -07006641 atomic_long_read(&parent->memory.children_min_usage)));
Roman Gushchin23067152018-06-07 17:06:22 -07006642
Chris Downb3a78222020-04-01 21:07:33 -07006643 WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage,
Chris Down03960e32020-06-25 20:30:22 -07006644 READ_ONCE(memcg->memory.low),
6645 READ_ONCE(parent->memory.elow),
Chris Downb3a78222020-04-01 21:07:33 -07006646 atomic_long_read(&parent->memory.children_low_usage)));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006647}
6648
Johannes Weiner00501b52014-08-08 14:19:20 -07006649/**
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07006650 * mem_cgroup_charge - charge a newly allocated page to a cgroup
Johannes Weiner00501b52014-08-08 14:19:20 -07006651 * @page: page to charge
6652 * @mm: mm context of the victim
6653 * @gfp_mask: reclaim mode
Johannes Weiner00501b52014-08-08 14:19:20 -07006654 *
6655 * Try to charge @page to the memcg that @mm belongs to, reclaiming
6656 * pages according to @gfp_mask if necessary.
6657 *
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07006658 * Returns 0 on success. Otherwise, an error code is returned.
Johannes Weiner00501b52014-08-08 14:19:20 -07006659 */
Johannes Weinerd9eb1ea2020-06-03 16:02:24 -07006660int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask)
Johannes Weiner00501b52014-08-08 14:19:20 -07006661{
Johannes Weiner3fba69a2020-06-03 16:01:31 -07006662 unsigned int nr_pages = hpage_nr_pages(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07006663 struct mem_cgroup *memcg = NULL;
Johannes Weiner00501b52014-08-08 14:19:20 -07006664 int ret = 0;
6665
6666 if (mem_cgroup_disabled())
6667 goto out;
6668
6669 if (PageSwapCache(page)) {
Johannes Weiner2d1c4982020-06-03 16:02:14 -07006670 swp_entry_t ent = { .val = page_private(page), };
6671 unsigned short id;
6672
Johannes Weiner00501b52014-08-08 14:19:20 -07006673 /*
6674 * Every swap fault against a single page tries to charge the
6675 * page, bail as early as possible. shmem_unuse() encounters
Johannes Weinereccb52e2020-06-03 16:02:11 -07006676 * already charged pages, too. page->mem_cgroup is protected
6677 * by the page lock, which serializes swap cache removal, which
Johannes Weiner00501b52014-08-08 14:19:20 -07006678 * in turn serializes uncharging.
6679 */
Vladimir Davydove993d902015-09-09 15:35:35 -07006680 VM_BUG_ON_PAGE(!PageLocked(page), page);
Huang Yingabe28952017-09-06 16:22:41 -07006681 if (compound_head(page)->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07006682 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07006683
Johannes Weiner2d1c4982020-06-03 16:02:14 -07006684 id = lookup_swap_cgroup_id(ent);
6685 rcu_read_lock();
6686 memcg = mem_cgroup_from_id(id);
6687 if (memcg && !css_tryget_online(&memcg->css))
6688 memcg = NULL;
6689 rcu_read_unlock();
Johannes Weiner00501b52014-08-08 14:19:20 -07006690 }
6691
Johannes Weiner00501b52014-08-08 14:19:20 -07006692 if (!memcg)
6693 memcg = get_mem_cgroup_from_mm(mm);
6694
6695 ret = try_charge(memcg, gfp_mask, nr_pages);
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07006696 if (ret)
6697 goto out_put;
Johannes Weiner00501b52014-08-08 14:19:20 -07006698
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07006699 css_get(&memcg->css);
Johannes Weinerd9eb1ea2020-06-03 16:02:24 -07006700 commit_charge(page, memcg);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006701
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006702 local_irq_disable();
Johannes Weiner3fba69a2020-06-03 16:01:31 -07006703 mem_cgroup_charge_statistics(memcg, page, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006704 memcg_check_events(memcg, page);
6705 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07006706
Johannes Weiner2d1c4982020-06-03 16:02:14 -07006707 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07006708 swp_entry_t entry = { .val = page_private(page) };
6709 /*
6710 * The swap entry might not get freed for a long time,
6711 * let's not wait for it. The page already received a
6712 * memory+swap charge, drop the swap entry duplicate.
6713 */
Huang Ying38d8b4e2017-07-06 15:37:18 -07006714 mem_cgroup_uncharge_swap(entry, nr_pages);
Johannes Weiner00501b52014-08-08 14:19:20 -07006715 }
Johannes Weiner00501b52014-08-08 14:19:20 -07006716
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07006717out_put:
6718 css_put(&memcg->css);
6719out:
6720 return ret;
Johannes Weiner3fea5a42020-06-03 16:01:41 -07006721}
6722
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006723struct uncharge_gather {
6724 struct mem_cgroup *memcg;
Johannes Weiner9f762db2020-06-03 16:01:44 -07006725 unsigned long nr_pages;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006726 unsigned long pgpgout;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006727 unsigned long nr_kmem;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006728 struct page *dummy_page;
6729};
6730
6731static inline void uncharge_gather_clear(struct uncharge_gather *ug)
Johannes Weiner747db952014-08-08 14:19:24 -07006732{
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006733 memset(ug, 0, sizeof(*ug));
6734}
6735
6736static void uncharge_batch(const struct uncharge_gather *ug)
6737{
Johannes Weiner747db952014-08-08 14:19:24 -07006738 unsigned long flags;
6739
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006740 if (!mem_cgroup_is_root(ug->memcg)) {
Johannes Weiner9f762db2020-06-03 16:01:44 -07006741 page_counter_uncharge(&ug->memcg->memory, ug->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08006742 if (do_memsw_account())
Johannes Weiner9f762db2020-06-03 16:01:44 -07006743 page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006744 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem)
6745 page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem);
6746 memcg_oom_recover(ug->memcg);
Johannes Weinerce00a962014-09-05 08:43:57 -04006747 }
Johannes Weiner747db952014-08-08 14:19:24 -07006748
6749 local_irq_save(flags);
Johannes Weinerc9019e92018-01-31 16:16:37 -08006750 __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout);
Johannes Weiner9f762db2020-06-03 16:01:44 -07006751 __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006752 memcg_check_events(ug->memcg, ug->dummy_page);
Johannes Weiner747db952014-08-08 14:19:24 -07006753 local_irq_restore(flags);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006754}
6755
6756static void uncharge_page(struct page *page, struct uncharge_gather *ug)
6757{
Johannes Weiner9f762db2020-06-03 16:01:44 -07006758 unsigned long nr_pages;
6759
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006760 VM_BUG_ON_PAGE(PageLRU(page), page);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006761
6762 if (!page->mem_cgroup)
6763 return;
6764
6765 /*
6766 * Nobody should be changing or seriously looking at
6767 * page->mem_cgroup at this point, we have fully
6768 * exclusive access to the page.
6769 */
6770
6771 if (ug->memcg != page->mem_cgroup) {
6772 if (ug->memcg) {
6773 uncharge_batch(ug);
6774 uncharge_gather_clear(ug);
6775 }
6776 ug->memcg = page->mem_cgroup;
6777 }
6778
Johannes Weiner9f762db2020-06-03 16:01:44 -07006779 nr_pages = compound_nr(page);
6780 ug->nr_pages += nr_pages;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006781
Johannes Weiner9f762db2020-06-03 16:01:44 -07006782 if (!PageKmemcg(page)) {
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006783 ug->pgpgout++;
6784 } else {
Johannes Weiner9f762db2020-06-03 16:01:44 -07006785 ug->nr_kmem += nr_pages;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006786 __ClearPageKmemcg(page);
6787 }
6788
6789 ug->dummy_page = page;
6790 page->mem_cgroup = NULL;
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07006791 css_put(&ug->memcg->css);
Johannes Weiner747db952014-08-08 14:19:24 -07006792}
6793
6794static void uncharge_list(struct list_head *page_list)
6795{
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006796 struct uncharge_gather ug;
Johannes Weiner747db952014-08-08 14:19:24 -07006797 struct list_head *next;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006798
6799 uncharge_gather_clear(&ug);
Johannes Weiner747db952014-08-08 14:19:24 -07006800
Johannes Weiner8b592652016-03-17 14:20:31 -07006801 /*
6802 * Note that the list can be a single page->lru; hence the
6803 * do-while loop instead of a simple list_for_each_entry().
6804 */
Johannes Weiner747db952014-08-08 14:19:24 -07006805 next = page_list->next;
6806 do {
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006807 struct page *page;
6808
Johannes Weiner747db952014-08-08 14:19:24 -07006809 page = list_entry(next, struct page, lru);
6810 next = page->lru.next;
6811
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006812 uncharge_page(page, &ug);
Johannes Weiner747db952014-08-08 14:19:24 -07006813 } while (next != page_list);
6814
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006815 if (ug.memcg)
6816 uncharge_batch(&ug);
Johannes Weiner747db952014-08-08 14:19:24 -07006817}
6818
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006819/**
6820 * mem_cgroup_uncharge - uncharge a page
6821 * @page: page to uncharge
6822 *
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07006823 * Uncharge a page previously charged with mem_cgroup_charge().
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006824 */
6825void mem_cgroup_uncharge(struct page *page)
6826{
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006827 struct uncharge_gather ug;
6828
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006829 if (mem_cgroup_disabled())
6830 return;
6831
Johannes Weiner747db952014-08-08 14:19:24 -07006832 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08006833 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006834 return;
6835
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006836 uncharge_gather_clear(&ug);
6837 uncharge_page(page, &ug);
6838 uncharge_batch(&ug);
Johannes Weiner747db952014-08-08 14:19:24 -07006839}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006840
Johannes Weiner747db952014-08-08 14:19:24 -07006841/**
6842 * mem_cgroup_uncharge_list - uncharge a list of page
6843 * @page_list: list of pages to uncharge
6844 *
6845 * Uncharge a list of pages previously charged with
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07006846 * mem_cgroup_charge().
Johannes Weiner747db952014-08-08 14:19:24 -07006847 */
6848void mem_cgroup_uncharge_list(struct list_head *page_list)
6849{
6850 if (mem_cgroup_disabled())
6851 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006852
Johannes Weiner747db952014-08-08 14:19:24 -07006853 if (!list_empty(page_list))
6854 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006855}
6856
6857/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07006858 * mem_cgroup_migrate - charge a page's replacement
6859 * @oldpage: currently circulating page
6860 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006861 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07006862 * Charge @newpage as a replacement page for @oldpage. @oldpage will
6863 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006864 *
6865 * Both pages must be locked, @newpage->mapping must be set up.
6866 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07006867void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006868{
Johannes Weiner29833312014-12-10 15:44:02 -08006869 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08006870 unsigned int nr_pages;
Tejun Heod93c4132016-06-24 14:49:54 -07006871 unsigned long flags;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006872
6873 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
6874 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006875 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006876 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
6877 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006878
6879 if (mem_cgroup_disabled())
6880 return;
6881
6882 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08006883 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006884 return;
6885
Hugh Dickins45637ba2015-11-05 18:49:40 -08006886 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08006887 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08006888 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006889 return;
6890
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08006891 /* Force-charge the new page. The old one will be freed soon */
Kaitao Cheng928552702020-01-30 22:13:42 -08006892 nr_pages = hpage_nr_pages(newpage);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08006893
6894 page_counter_charge(&memcg->memory, nr_pages);
6895 if (do_memsw_account())
6896 page_counter_charge(&memcg->memsw, nr_pages);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006897
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07006898 css_get(&memcg->css);
Johannes Weinerd9eb1ea2020-06-03 16:02:24 -07006899 commit_charge(newpage, memcg);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08006900
Tejun Heod93c4132016-06-24 14:49:54 -07006901 local_irq_save(flags);
Johannes Weiner3fba69a2020-06-03 16:01:31 -07006902 mem_cgroup_charge_statistics(memcg, newpage, nr_pages);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08006903 memcg_check_events(memcg, newpage);
Tejun Heod93c4132016-06-24 14:49:54 -07006904 local_irq_restore(flags);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006905}
6906
Johannes Weineref129472016-01-14 15:21:34 -08006907DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08006908EXPORT_SYMBOL(memcg_sockets_enabled_key);
6909
Johannes Weiner2d758072016-10-07 17:00:58 -07006910void mem_cgroup_sk_alloc(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08006911{
6912 struct mem_cgroup *memcg;
6913
Johannes Weiner2d758072016-10-07 17:00:58 -07006914 if (!mem_cgroup_sockets_enabled)
6915 return;
6916
Shakeel Butte876ecc2020-03-09 22:16:05 -07006917 /* Do not associate the sock with unrelated interrupted task's memcg. */
6918 if (in_interrupt())
6919 return;
6920
Johannes Weiner11092082016-01-14 15:21:26 -08006921 rcu_read_lock();
6922 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006923 if (memcg == root_mem_cgroup)
6924 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08006925 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006926 goto out;
Shakeel Butt8965aa22020-04-01 21:07:10 -07006927 if (css_tryget(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08006928 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006929out:
Johannes Weiner11092082016-01-14 15:21:26 -08006930 rcu_read_unlock();
6931}
Johannes Weiner11092082016-01-14 15:21:26 -08006932
Johannes Weiner2d758072016-10-07 17:00:58 -07006933void mem_cgroup_sk_free(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08006934{
Johannes Weiner2d758072016-10-07 17:00:58 -07006935 if (sk->sk_memcg)
6936 css_put(&sk->sk_memcg->css);
Johannes Weiner11092082016-01-14 15:21:26 -08006937}
6938
6939/**
6940 * mem_cgroup_charge_skmem - charge socket memory
6941 * @memcg: memcg to charge
6942 * @nr_pages: number of pages to charge
6943 *
6944 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
6945 * @memcg's configured limit, %false if the charge had to be forced.
6946 */
6947bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
6948{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006949 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08006950
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006951 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08006952 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006953
Johannes Weiner0db15292016-01-20 15:02:50 -08006954 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
6955 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006956 return true;
6957 }
Johannes Weiner0db15292016-01-20 15:02:50 -08006958 page_counter_charge(&memcg->tcpmem, nr_pages);
6959 memcg->tcpmem_pressure = 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006960 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08006961 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08006962
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006963 /* Don't block in the packet receive path */
6964 if (in_softirq())
6965 gfp_mask = GFP_NOWAIT;
6966
Johannes Weinerc9019e92018-01-31 16:16:37 -08006967 mod_memcg_state(memcg, MEMCG_SOCK, nr_pages);
Johannes Weinerb2807f02016-01-20 15:03:22 -08006968
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006969 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
6970 return true;
6971
6972 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08006973 return false;
6974}
6975
6976/**
6977 * mem_cgroup_uncharge_skmem - uncharge socket memory
Mike Rapoportb7701a52018-02-06 15:42:13 -08006978 * @memcg: memcg to uncharge
6979 * @nr_pages: number of pages to uncharge
Johannes Weiner11092082016-01-14 15:21:26 -08006980 */
6981void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
6982{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006983 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08006984 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006985 return;
6986 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08006987
Johannes Weinerc9019e92018-01-31 16:16:37 -08006988 mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages);
Johannes Weinerb2807f02016-01-20 15:03:22 -08006989
Roman Gushchin475d0482017-09-08 16:13:09 -07006990 refill_stock(memcg, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08006991}
6992
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006993static int __init cgroup_memory(char *s)
6994{
6995 char *token;
6996
6997 while ((token = strsep(&s, ",")) != NULL) {
6998 if (!*token)
6999 continue;
7000 if (!strcmp(token, "nosocket"))
7001 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08007002 if (!strcmp(token, "nokmem"))
7003 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007004 }
7005 return 0;
7006}
7007__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08007008
Michal Hocko2d110852013-02-22 16:34:43 -08007009/*
Michal Hocko10813122013-02-22 16:35:41 -08007010 * subsys_initcall() for memory controller.
7011 *
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01007012 * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this
7013 * context because of lock dependencies (cgroup_lock -> cpu hotplug) but
7014 * basically everything that doesn't depend on a specific mem_cgroup structure
7015 * should be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08007016 */
7017static int __init mem_cgroup_init(void)
7018{
Johannes Weiner95a045f2015-02-11 15:26:33 -08007019 int cpu, node;
7020
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01007021 cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL,
7022 memcg_hotplug_cpu_dead);
Johannes Weiner95a045f2015-02-11 15:26:33 -08007023
7024 for_each_possible_cpu(cpu)
7025 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
7026 drain_local_stock);
7027
7028 for_each_node(node) {
7029 struct mem_cgroup_tree_per_node *rtpn;
Johannes Weiner95a045f2015-02-11 15:26:33 -08007030
7031 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
7032 node_online(node) ? node : NUMA_NO_NODE);
7033
Mel Gormanef8f2322016-07-28 15:46:05 -07007034 rtpn->rb_root = RB_ROOT;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -07007035 rtpn->rb_rightmost = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -07007036 spin_lock_init(&rtpn->lock);
Johannes Weiner95a045f2015-02-11 15:26:33 -08007037 soft_limit_tree.rb_tree_per_node[node] = rtpn;
7038 }
7039
Michal Hocko2d110852013-02-22 16:34:43 -08007040 return 0;
7041}
7042subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08007043
7044#ifdef CONFIG_MEMCG_SWAP
Arnd Bergmann358c07f2016-08-25 15:17:08 -07007045static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
7046{
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07007047 while (!refcount_inc_not_zero(&memcg->id.ref)) {
Arnd Bergmann358c07f2016-08-25 15:17:08 -07007048 /*
7049 * The root cgroup cannot be destroyed, so it's refcount must
7050 * always be >= 1.
7051 */
7052 if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
7053 VM_BUG_ON(1);
7054 break;
7055 }
7056 memcg = parent_mem_cgroup(memcg);
7057 if (!memcg)
7058 memcg = root_mem_cgroup;
7059 }
7060 return memcg;
7061}
7062
Johannes Weiner21afa382015-02-11 15:26:36 -08007063/**
7064 * mem_cgroup_swapout - transfer a memsw charge to swap
7065 * @page: page whose memsw charge to transfer
7066 * @entry: swap entry to move the charge to
7067 *
7068 * Transfer the memsw charge of @page to @entry.
7069 */
7070void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
7071{
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007072 struct mem_cgroup *memcg, *swap_memcg;
Huang Yingd6810d72017-09-06 16:22:45 -07007073 unsigned int nr_entries;
Johannes Weiner21afa382015-02-11 15:26:36 -08007074 unsigned short oldid;
7075
7076 VM_BUG_ON_PAGE(PageLRU(page), page);
7077 VM_BUG_ON_PAGE(page_count(page), page);
7078
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007079 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Johannes Weiner21afa382015-02-11 15:26:36 -08007080 return;
7081
7082 memcg = page->mem_cgroup;
7083
7084 /* Readahead page, never charged */
7085 if (!memcg)
7086 return;
7087
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007088 /*
7089 * In case the memcg owning these pages has been offlined and doesn't
7090 * have an ID allocated to it anymore, charge the closest online
7091 * ancestor for the swap instead and transfer the memory+swap charge.
7092 */
7093 swap_memcg = mem_cgroup_id_get_online(memcg);
Huang Yingd6810d72017-09-06 16:22:45 -07007094 nr_entries = hpage_nr_pages(page);
7095 /* Get references for the tail pages, too */
7096 if (nr_entries > 1)
7097 mem_cgroup_id_get_many(swap_memcg, nr_entries - 1);
7098 oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg),
7099 nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007100 VM_BUG_ON_PAGE(oldid, page);
Johannes Weinerc9019e92018-01-31 16:16:37 -08007101 mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007102
7103 page->mem_cgroup = NULL;
7104
7105 if (!mem_cgroup_is_root(memcg))
Huang Yingd6810d72017-09-06 16:22:45 -07007106 page_counter_uncharge(&memcg->memory, nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007107
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007108 if (!cgroup_memory_noswap && memcg != swap_memcg) {
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007109 if (!mem_cgroup_is_root(swap_memcg))
Huang Yingd6810d72017-09-06 16:22:45 -07007110 page_counter_charge(&swap_memcg->memsw, nr_entries);
7111 page_counter_uncharge(&memcg->memsw, nr_entries);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007112 }
7113
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07007114 /*
7115 * Interrupts should be disabled here because the caller holds the
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07007116 * i_pages lock which is taken with interrupts-off. It is
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07007117 * important here to have the interrupts disabled because it is the
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07007118 * only synchronisation we have for updating the per-CPU variables.
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07007119 */
7120 VM_BUG_ON(!irqs_disabled());
Johannes Weiner3fba69a2020-06-03 16:01:31 -07007121 mem_cgroup_charge_statistics(memcg, page, -nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007122 memcg_check_events(memcg, page);
Johannes Weiner73f576c2016-07-20 15:44:57 -07007123
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07007124 css_put(&memcg->css);
Johannes Weiner21afa382015-02-11 15:26:36 -08007125}
7126
Huang Ying38d8b4e2017-07-06 15:37:18 -07007127/**
7128 * mem_cgroup_try_charge_swap - try charging swap space for a page
Vladimir Davydov37e84352016-01-20 15:02:56 -08007129 * @page: page being added to swap
7130 * @entry: swap entry to charge
7131 *
Huang Ying38d8b4e2017-07-06 15:37:18 -07007132 * Try to charge @page's memcg for the swap space at @entry.
Vladimir Davydov37e84352016-01-20 15:02:56 -08007133 *
7134 * Returns 0 on success, -ENOMEM on failure.
7135 */
7136int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
7137{
Huang Ying38d8b4e2017-07-06 15:37:18 -07007138 unsigned int nr_pages = hpage_nr_pages(page);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007139 struct page_counter *counter;
Huang Ying38d8b4e2017-07-06 15:37:18 -07007140 struct mem_cgroup *memcg;
Vladimir Davydov37e84352016-01-20 15:02:56 -08007141 unsigned short oldid;
7142
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007143 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov37e84352016-01-20 15:02:56 -08007144 return 0;
7145
7146 memcg = page->mem_cgroup;
7147
7148 /* Readahead page, never charged */
7149 if (!memcg)
7150 return 0;
7151
Tejun Heof3a53a32018-06-07 17:05:35 -07007152 if (!entry.val) {
7153 memcg_memory_event(memcg, MEMCG_SWAP_FAIL);
Tejun Heobb98f2c2018-06-07 17:05:31 -07007154 return 0;
Tejun Heof3a53a32018-06-07 17:05:35 -07007155 }
Tejun Heobb98f2c2018-06-07 17:05:31 -07007156
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007157 memcg = mem_cgroup_id_get_online(memcg);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007158
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007159 if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) &&
Huang Ying38d8b4e2017-07-06 15:37:18 -07007160 !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) {
Tejun Heof3a53a32018-06-07 17:05:35 -07007161 memcg_memory_event(memcg, MEMCG_SWAP_MAX);
7162 memcg_memory_event(memcg, MEMCG_SWAP_FAIL);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007163 mem_cgroup_id_put(memcg);
7164 return -ENOMEM;
7165 }
7166
Huang Ying38d8b4e2017-07-06 15:37:18 -07007167 /* Get references for the tail pages, too */
7168 if (nr_pages > 1)
7169 mem_cgroup_id_get_many(memcg, nr_pages - 1);
7170 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007171 VM_BUG_ON_PAGE(oldid, page);
Johannes Weinerc9019e92018-01-31 16:16:37 -08007172 mod_memcg_state(memcg, MEMCG_SWAP, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007173
Vladimir Davydov37e84352016-01-20 15:02:56 -08007174 return 0;
7175}
7176
Johannes Weiner21afa382015-02-11 15:26:36 -08007177/**
Huang Ying38d8b4e2017-07-06 15:37:18 -07007178 * mem_cgroup_uncharge_swap - uncharge swap space
Johannes Weiner21afa382015-02-11 15:26:36 -08007179 * @entry: swap entry to uncharge
Huang Ying38d8b4e2017-07-06 15:37:18 -07007180 * @nr_pages: the amount of swap space to uncharge
Johannes Weiner21afa382015-02-11 15:26:36 -08007181 */
Huang Ying38d8b4e2017-07-06 15:37:18 -07007182void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages)
Johannes Weiner21afa382015-02-11 15:26:36 -08007183{
7184 struct mem_cgroup *memcg;
7185 unsigned short id;
7186
Huang Ying38d8b4e2017-07-06 15:37:18 -07007187 id = swap_cgroup_record(entry, 0, nr_pages);
Johannes Weiner21afa382015-02-11 15:26:36 -08007188 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07007189 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08007190 if (memcg) {
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007191 if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08007192 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Huang Ying38d8b4e2017-07-06 15:37:18 -07007193 page_counter_uncharge(&memcg->swap, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007194 else
Huang Ying38d8b4e2017-07-06 15:37:18 -07007195 page_counter_uncharge(&memcg->memsw, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007196 }
Johannes Weinerc9019e92018-01-31 16:16:37 -08007197 mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages);
Huang Ying38d8b4e2017-07-06 15:37:18 -07007198 mem_cgroup_id_put_many(memcg, nr_pages);
Johannes Weiner21afa382015-02-11 15:26:36 -08007199 }
7200 rcu_read_unlock();
7201}
7202
Vladimir Davydovd8b38432016-01-20 15:03:07 -08007203long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
7204{
7205 long nr_swap_pages = get_nr_swap_pages();
7206
Johannes Weinereccb52e2020-06-03 16:02:11 -07007207 if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydovd8b38432016-01-20 15:03:07 -08007208 return nr_swap_pages;
7209 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
7210 nr_swap_pages = min_t(long, nr_swap_pages,
Roman Gushchinbbec2e12018-06-07 17:06:18 -07007211 READ_ONCE(memcg->swap.max) -
Vladimir Davydovd8b38432016-01-20 15:03:07 -08007212 page_counter_read(&memcg->swap));
7213 return nr_swap_pages;
7214}
7215
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007216bool mem_cgroup_swap_full(struct page *page)
7217{
7218 struct mem_cgroup *memcg;
7219
7220 VM_BUG_ON_PAGE(!PageLocked(page), page);
7221
7222 if (vm_swap_full())
7223 return true;
Johannes Weinereccb52e2020-06-03 16:02:11 -07007224 if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007225 return false;
7226
7227 memcg = page->mem_cgroup;
7228 if (!memcg)
7229 return false;
7230
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007231 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) {
7232 unsigned long usage = page_counter_read(&memcg->swap);
7233
7234 if (usage * 2 >= READ_ONCE(memcg->swap.high) ||
7235 usage * 2 >= READ_ONCE(memcg->swap.max))
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007236 return true;
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007237 }
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007238
7239 return false;
7240}
7241
Johannes Weinereccb52e2020-06-03 16:02:11 -07007242static int __init setup_swap_account(char *s)
Johannes Weiner21afa382015-02-11 15:26:36 -08007243{
7244 if (!strcmp(s, "1"))
Johannes Weinereccb52e2020-06-03 16:02:11 -07007245 cgroup_memory_noswap = 0;
Johannes Weiner21afa382015-02-11 15:26:36 -08007246 else if (!strcmp(s, "0"))
Johannes Weinereccb52e2020-06-03 16:02:11 -07007247 cgroup_memory_noswap = 1;
Johannes Weiner21afa382015-02-11 15:26:36 -08007248 return 1;
7249}
Johannes Weinereccb52e2020-06-03 16:02:11 -07007250__setup("swapaccount=", setup_swap_account);
Johannes Weiner21afa382015-02-11 15:26:36 -08007251
Vladimir Davydov37e84352016-01-20 15:02:56 -08007252static u64 swap_current_read(struct cgroup_subsys_state *css,
7253 struct cftype *cft)
7254{
7255 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
7256
7257 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
7258}
7259
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007260static int swap_high_show(struct seq_file *m, void *v)
7261{
7262 return seq_puts_memcg_tunable(m,
7263 READ_ONCE(mem_cgroup_from_seq(m)->swap.high));
7264}
7265
7266static ssize_t swap_high_write(struct kernfs_open_file *of,
7267 char *buf, size_t nbytes, loff_t off)
7268{
7269 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
7270 unsigned long high;
7271 int err;
7272
7273 buf = strstrip(buf);
7274 err = page_counter_memparse(buf, "max", &high);
7275 if (err)
7276 return err;
7277
7278 page_counter_set_high(&memcg->swap, high);
7279
7280 return nbytes;
7281}
7282
Vladimir Davydov37e84352016-01-20 15:02:56 -08007283static int swap_max_show(struct seq_file *m, void *v)
7284{
Chris Down677dc972019-03-05 15:45:55 -08007285 return seq_puts_memcg_tunable(m,
7286 READ_ONCE(mem_cgroup_from_seq(m)->swap.max));
Vladimir Davydov37e84352016-01-20 15:02:56 -08007287}
7288
7289static ssize_t swap_max_write(struct kernfs_open_file *of,
7290 char *buf, size_t nbytes, loff_t off)
7291{
7292 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
7293 unsigned long max;
7294 int err;
7295
7296 buf = strstrip(buf);
7297 err = page_counter_memparse(buf, "max", &max);
7298 if (err)
7299 return err;
7300
Tejun Heobe091022018-06-07 17:09:21 -07007301 xchg(&memcg->swap.max, max);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007302
7303 return nbytes;
7304}
7305
Tejun Heof3a53a32018-06-07 17:05:35 -07007306static int swap_events_show(struct seq_file *m, void *v)
7307{
Chris Downaa9694b2019-03-05 15:45:52 -08007308 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Tejun Heof3a53a32018-06-07 17:05:35 -07007309
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007310 seq_printf(m, "high %lu\n",
7311 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH]));
Tejun Heof3a53a32018-06-07 17:05:35 -07007312 seq_printf(m, "max %lu\n",
7313 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX]));
7314 seq_printf(m, "fail %lu\n",
7315 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL]));
7316
7317 return 0;
7318}
7319
Vladimir Davydov37e84352016-01-20 15:02:56 -08007320static struct cftype swap_files[] = {
7321 {
7322 .name = "swap.current",
7323 .flags = CFTYPE_NOT_ON_ROOT,
7324 .read_u64 = swap_current_read,
7325 },
7326 {
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007327 .name = "swap.high",
7328 .flags = CFTYPE_NOT_ON_ROOT,
7329 .seq_show = swap_high_show,
7330 .write = swap_high_write,
7331 },
7332 {
Vladimir Davydov37e84352016-01-20 15:02:56 -08007333 .name = "swap.max",
7334 .flags = CFTYPE_NOT_ON_ROOT,
7335 .seq_show = swap_max_show,
7336 .write = swap_max_write,
7337 },
Tejun Heof3a53a32018-06-07 17:05:35 -07007338 {
7339 .name = "swap.events",
7340 .flags = CFTYPE_NOT_ON_ROOT,
7341 .file_offset = offsetof(struct mem_cgroup, swap_events_file),
7342 .seq_show = swap_events_show,
7343 },
Vladimir Davydov37e84352016-01-20 15:02:56 -08007344 { } /* terminate */
7345};
7346
Johannes Weinereccb52e2020-06-03 16:02:11 -07007347static struct cftype memsw_files[] = {
Johannes Weiner21afa382015-02-11 15:26:36 -08007348 {
7349 .name = "memsw.usage_in_bytes",
7350 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
7351 .read_u64 = mem_cgroup_read_u64,
7352 },
7353 {
7354 .name = "memsw.max_usage_in_bytes",
7355 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
7356 .write = mem_cgroup_reset,
7357 .read_u64 = mem_cgroup_read_u64,
7358 },
7359 {
7360 .name = "memsw.limit_in_bytes",
7361 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
7362 .write = mem_cgroup_write,
7363 .read_u64 = mem_cgroup_read_u64,
7364 },
7365 {
7366 .name = "memsw.failcnt",
7367 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
7368 .write = mem_cgroup_reset,
7369 .read_u64 = mem_cgroup_read_u64,
7370 },
7371 { }, /* terminate */
7372};
7373
Bhupesh Sharma82ff1652020-07-23 21:15:21 -07007374/*
7375 * If mem_cgroup_swap_init() is implemented as a subsys_initcall()
7376 * instead of a core_initcall(), this could mean cgroup_memory_noswap still
7377 * remains set to false even when memcg is disabled via "cgroup_disable=memory"
7378 * boot parameter. This may result in premature OOPS inside
7379 * mem_cgroup_get_nr_swap_pages() function in corner cases.
7380 */
Johannes Weiner21afa382015-02-11 15:26:36 -08007381static int __init mem_cgroup_swap_init(void)
7382{
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007383 /* No memory control -> no swap control */
7384 if (mem_cgroup_disabled())
7385 cgroup_memory_noswap = true;
7386
7387 if (cgroup_memory_noswap)
Johannes Weinereccb52e2020-06-03 16:02:11 -07007388 return 0;
7389
7390 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files));
7391 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files));
7392
Johannes Weiner21afa382015-02-11 15:26:36 -08007393 return 0;
7394}
Bhupesh Sharma82ff1652020-07-23 21:15:21 -07007395core_initcall(mem_cgroup_swap_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08007396
7397#endif /* CONFIG_MEMCG_SWAP */