blob: cf0321d7a7842cfc1e0f49d07ec334a5b8cbd8d5 [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
Alex Shi6168d0d2020-12-15 12:34:29 -080023 *
24 * Per memcg lru locking
25 * Copyright (C) 2020 Alibaba, Inc, Alex Shi
Balbir Singh8cdea7c2008-02-07 00:13:50 -080026 */
27
Johannes Weiner3e32cb22014-12-10 15:42:31 -080028#include <linux/page_counter.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080029#include <linux/memcontrol.h>
30#include <linux/cgroup.h>
Christoph Hellwiga5201102019-08-28 16:19:53 +020031#include <linux/pagewalk.h>
Ingo Molnar6e84f312017-02-08 18:51:29 +010032#include <linux/sched/mm.h>
Hugh Dickins3a4f8a02017-02-24 14:59:36 -080033#include <linux/shmem_fs.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080034#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080035#include <linux/pagemap.h>
Chris Down1ff9e6e2019-03-05 15:48:09 -080036#include <linux/vm_event_item.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080037#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080038#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080039#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080040#include <linux/bit_spinlock.h>
41#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070042#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040043#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080044#include <linux/mutex.h>
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -070045#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070046#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080047#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080048#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080049#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080050#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050051#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080052#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080053#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080054#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070055#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070056#include <linux/mm_inline.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080057#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080058#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070059#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070060#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050061#include <linux/file.h>
Tejun Heob23afb92015-11-05 18:46:11 -080062#include <linux/tracehook.h>
Chris Down0e4b01d2019-09-23 15:34:55 -070063#include <linux/psi.h>
Johannes Weinerc8713d02019-07-11 20:55:59 -070064#include <linux/seq_buf.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080065#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000066#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070067#include <net/ip.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080068#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080069
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080070#include <linux/uaccess.h>
Balbir Singh8697d332008-02-07 00:13:59 -080071
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070072#include <trace/events/vmscan.h>
73
Tejun Heo073219e2014-02-08 10:36:58 -050074struct cgroup_subsys memory_cgrp_subsys __read_mostly;
75EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080076
Johannes Weiner7d828602016-01-14 15:20:56 -080077struct mem_cgroup *root_mem_cgroup __read_mostly;
78
Roman Gushchin37d59852020-10-17 16:13:50 -070079/* Active memory cgroup to use from an interrupt context */
80DEFINE_PER_CPU(struct mem_cgroup *, int_active_memcg);
Dan Schatzbergc74d40e2021-06-28 19:38:21 -070081EXPORT_PER_CPU_SYMBOL_GPL(int_active_memcg);
Roman Gushchin37d59852020-10-17 16:13:50 -070082
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080083/* Socket memory accounting disabled? */
Roman Gushchin0f0cace2021-06-02 18:09:29 -070084static bool cgroup_memory_nosocket __ro_after_init;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080085
Vladimir Davydov04823c82016-01-20 15:02:38 -080086/* Kernel memory accounting disabled? */
Linus Torvaldse2679922021-07-01 17:17:24 -070087bool cgroup_memory_nokmem __ro_after_init;
Vladimir Davydov04823c82016-01-20 15:02:38 -080088
Johannes Weiner21afa382015-02-11 15:26:36 -080089/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070090#ifdef CONFIG_MEMCG_SWAP
Roman Gushchin0f0cace2021-06-02 18:09:29 -070091bool cgroup_memory_noswap __ro_after_init;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080092#else
Johannes Weinereccb52e2020-06-03 16:02:11 -070093#define cgroup_memory_noswap 1
Johannes Weiner2d1c4982020-06-03 16:02:14 -070094#endif
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080095
Tejun Heo97b27822019-08-26 09:06:56 -070096#ifdef CONFIG_CGROUP_WRITEBACK
97static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq);
98#endif
99
Johannes Weiner7941d212016-01-14 15:21:23 -0800100/* Whether legacy memory+swap accounting is active */
101static bool do_memsw_account(void)
102{
Johannes Weinereccb52e2020-06-03 16:02:11 -0700103 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_noswap;
Johannes Weiner7941d212016-01-14 15:21:23 -0800104}
105
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700106#define THRESHOLDS_EVENTS_TARGET 128
107#define SOFTLIMIT_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700108
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700109/*
110 * Cgroups above their limits are maintained in a RB-Tree, independent of
111 * their hierarchy representation
112 */
113
Mel Gormanef8f2322016-07-28 15:46:05 -0700114struct mem_cgroup_tree_per_node {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700115 struct rb_root rb_root;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700116 struct rb_node *rb_rightmost;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700117 spinlock_t lock;
118};
119
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700120struct mem_cgroup_tree {
121 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
122};
123
124static struct mem_cgroup_tree soft_limit_tree __read_mostly;
125
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700126/* for OOM */
127struct mem_cgroup_eventfd_list {
128 struct list_head list;
129 struct eventfd_ctx *eventfd;
130};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800131
Tejun Heo79bd9812013-11-22 18:20:42 -0500132/*
133 * cgroup_event represents events which userspace want to receive.
134 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500135struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500136 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500137 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500138 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500139 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500140 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500141 * eventfd to signal userspace about the event.
142 */
143 struct eventfd_ctx *eventfd;
144 /*
145 * Each of these stored in a list by the cgroup.
146 */
147 struct list_head list;
148 /*
Tejun Heofba94802013-11-22 18:20:43 -0500149 * register_event() callback will be used to add new userspace
150 * waiter for changes related to this event. Use eventfd_signal()
151 * on eventfd to send notification to userspace.
152 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500153 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500154 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500155 /*
156 * unregister_event() callback will be called when userspace closes
157 * the eventfd or on cgroup removing. This callback must be set,
158 * if you want provide notification functionality.
159 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500160 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500161 struct eventfd_ctx *eventfd);
162 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500163 * All fields below needed to unregister event when
164 * userspace closes eventfd.
165 */
166 poll_table pt;
167 wait_queue_head_t *wqh;
Ingo Molnarac6424b2017-06-20 12:06:13 +0200168 wait_queue_entry_t wait;
Tejun Heo79bd9812013-11-22 18:20:42 -0500169 struct work_struct remove;
170};
171
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700172static void mem_cgroup_threshold(struct mem_cgroup *memcg);
173static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800174
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800175/* Stuffs for move charges at task migration. */
176/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800177 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800178 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800179#define MOVE_ANON 0x1U
180#define MOVE_FILE 0x2U
181#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800182
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800183/* "mc" and its members are protected by cgroup_mutex */
184static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800185 spinlock_t lock; /* for from, to */
Tejun Heo264a0ae2016-04-21 19:09:02 -0400186 struct mm_struct *mm;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800187 struct mem_cgroup *from;
188 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800189 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800190 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800191 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800192 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800193 struct task_struct *moving_task; /* a task moving charges */
194 wait_queue_head_t waitq; /* a waitq for other context */
195} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700196 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800197 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
198};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800199
Balbir Singh4e416952009-09-23 15:56:39 -0700200/*
201 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
202 * limit reclaim to prevent infinite loops, if they ever occur.
203 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700204#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700205#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700206
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800207/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800208enum res_type {
209 _MEM,
210 _MEMSWAP,
211 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800212 _KMEM,
Vladimir Davydovd55f90b2016-01-20 15:02:44 -0800213 _TCP,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800214};
215
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700216#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
217#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800218#define MEMFILE_ATTR(val) ((val) & 0xffff)
Ingo Molnarf0953a12021-05-06 18:06:47 -0700219/* Used for OOM notifier */
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700220#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800221
Kirill Tkhaib05706f2018-08-17 15:47:33 -0700222/*
223 * Iteration constructs for visiting all cgroups (under a tree). If
224 * loops are exited prematurely (break), mem_cgroup_iter_break() must
225 * be used for reference counting.
226 */
227#define for_each_mem_cgroup_tree(iter, root) \
228 for (iter = mem_cgroup_iter(root, NULL, NULL); \
229 iter != NULL; \
230 iter = mem_cgroup_iter(root, iter, NULL))
231
232#define for_each_mem_cgroup(iter) \
233 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
234 iter != NULL; \
235 iter = mem_cgroup_iter(NULL, iter, NULL))
236
Vasily Averina4ebf1b2021-11-05 13:38:09 -0700237static inline bool task_is_dying(void)
Tetsuo Handa7775fac2019-03-05 15:46:47 -0800238{
239 return tsk_is_oom_victim(current) || fatal_signal_pending(current) ||
240 (current->flags & PF_EXITING);
241}
242
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700243/* Some nice accessors for the vmpressure. */
244struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
245{
246 if (!memcg)
247 memcg = root_mem_cgroup;
248 return &memcg->vmpressure;
249}
250
Hui Su96478752021-09-02 14:59:36 -0700251struct mem_cgroup *vmpressure_to_memcg(struct vmpressure *vmpr)
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700252{
Hui Su96478752021-09-02 14:59:36 -0700253 return container_of(vmpr, struct mem_cgroup, vmpressure);
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700254}
255
Kirill Tkhai84c07d12018-08-17 15:47:25 -0700256#ifdef CONFIG_MEMCG_KMEM
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700257extern spinlock_t css_set_lock;
258
Roman Gushchin4d5c8ae2021-06-02 18:09:30 -0700259bool mem_cgroup_kmem_disabled(void)
260{
261 return cgroup_memory_nokmem;
262}
263
Muchun Songf1286fa2021-04-29 22:56:55 -0700264static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg,
265 unsigned int nr_pages);
Roman Gushchinc1a660d2021-02-24 12:03:58 -0800266
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700267static void obj_cgroup_release(struct percpu_ref *ref)
268{
269 struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt);
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700270 unsigned int nr_bytes;
271 unsigned int nr_pages;
272 unsigned long flags;
273
274 /*
275 * At this point all allocated objects are freed, and
276 * objcg->nr_charged_bytes can't have an arbitrary byte value.
277 * However, it can be PAGE_SIZE or (x * PAGE_SIZE).
278 *
279 * The following sequence can lead to it:
280 * 1) CPU0: objcg == stock->cached_objcg
281 * 2) CPU1: we do a small allocation (e.g. 92 bytes),
282 * PAGE_SIZE bytes are charged
283 * 3) CPU1: a process from another memcg is allocating something,
284 * the stock if flushed,
285 * objcg->nr_charged_bytes = PAGE_SIZE - 92
286 * 5) CPU0: we do release this object,
287 * 92 bytes are added to stock->nr_bytes
288 * 6) CPU0: stock is flushed,
289 * 92 bytes are added to objcg->nr_charged_bytes
290 *
291 * In the result, nr_charged_bytes == PAGE_SIZE.
292 * This page will be uncharged in obj_cgroup_release().
293 */
294 nr_bytes = atomic_read(&objcg->nr_charged_bytes);
295 WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1));
296 nr_pages = nr_bytes >> PAGE_SHIFT;
297
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700298 if (nr_pages)
Muchun Songf1286fa2021-04-29 22:56:55 -0700299 obj_cgroup_uncharge_pages(objcg, nr_pages);
Muchun Song271dd6b2021-06-28 19:38:06 -0700300
301 spin_lock_irqsave(&css_set_lock, flags);
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700302 list_del(&objcg->list);
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700303 spin_unlock_irqrestore(&css_set_lock, flags);
304
305 percpu_ref_exit(ref);
306 kfree_rcu(objcg, rcu);
307}
308
309static struct obj_cgroup *obj_cgroup_alloc(void)
310{
311 struct obj_cgroup *objcg;
312 int ret;
313
314 objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL);
315 if (!objcg)
316 return NULL;
317
318 ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0,
319 GFP_KERNEL);
320 if (ret) {
321 kfree(objcg);
322 return NULL;
323 }
324 INIT_LIST_HEAD(&objcg->list);
325 return objcg;
326}
327
328static void memcg_reparent_objcgs(struct mem_cgroup *memcg,
329 struct mem_cgroup *parent)
330{
331 struct obj_cgroup *objcg, *iter;
332
333 objcg = rcu_replace_pointer(memcg->objcg, NULL, true);
334
335 spin_lock_irq(&css_set_lock);
336
Muchun Song98383542021-06-28 19:38:03 -0700337 /* 1) Ready to reparent active objcg. */
338 list_add(&objcg->list, &memcg->objcg_list);
339 /* 2) Reparent active objcg and already reparented objcgs to parent. */
340 list_for_each_entry(iter, &memcg->objcg_list, list)
341 WRITE_ONCE(iter->memcg, parent);
342 /* 3) Move already reparented objcgs to the parent's list */
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700343 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
Tejun Heoad7fa852015-05-27 20:00:02 -0400402/**
403 * mem_cgroup_css_from_page - css of the memcg associated with a page
404 * @page: page of interest
405 *
406 * If memcg is bound to the default hierarchy, css of the memcg associated
407 * with @page is returned. The returned css remains associated with @page
408 * until it is released.
409 *
410 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
411 * is returned.
Tejun Heoad7fa852015-05-27 20:00:02 -0400412 */
413struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
414{
415 struct mem_cgroup *memcg;
416
Roman Gushchinbcfe06b2020-12-01 13:58:27 -0800417 memcg = page_memcg(page);
Tejun Heoad7fa852015-05-27 20:00:02 -0400418
Tejun Heo9e10a132015-09-18 11:56:28 -0400419 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400420 memcg = root_mem_cgroup;
421
Tejun Heoad7fa852015-05-27 20:00:02 -0400422 return &memcg->css;
423}
424
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700425/**
426 * page_cgroup_ino - return inode number of the memcg a page is charged to
427 * @page: the page
428 *
429 * Look up the closest online ancestor of the memory cgroup @page is charged to
430 * and return its inode number or 0 if @page is not charged to any cgroup. It
431 * is safe to call this function without holding a reference to @page.
432 *
433 * Note, this function is inherently racy, because there is nothing to prevent
434 * the cgroup inode from getting torn down and potentially reallocated a moment
435 * after page_cgroup_ino() returns, so it only should be used by callers that
436 * do not care (such as procfs interfaces).
437 */
438ino_t page_cgroup_ino(struct page *page)
439{
440 struct mem_cgroup *memcg;
441 unsigned long ino = 0;
442
443 rcu_read_lock();
Roman Gushchinbcfe06b2020-12-01 13:58:27 -0800444 memcg = page_memcg_check(page);
Roman Gushchin286e04b2020-08-06 23:20:52 -0700445
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700446 while (memcg && !(memcg->css.flags & CSS_ONLINE))
447 memcg = parent_mem_cgroup(memcg);
448 if (memcg)
449 ino = cgroup_ino(memcg->css.cgroup);
450 rcu_read_unlock();
451 return ino;
452}
453
Mel Gormanef8f2322016-07-28 15:46:05 -0700454static struct mem_cgroup_per_node *
455mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700456{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700457 int nid = page_to_nid(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700458
Mel Gormanef8f2322016-07-28 15:46:05 -0700459 return memcg->nodeinfo[nid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700460}
461
Mel Gormanef8f2322016-07-28 15:46:05 -0700462static struct mem_cgroup_tree_per_node *
463soft_limit_tree_node(int nid)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700464{
Mel Gormanef8f2322016-07-28 15:46:05 -0700465 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700466}
467
Mel Gormanef8f2322016-07-28 15:46:05 -0700468static struct mem_cgroup_tree_per_node *
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700469soft_limit_tree_from_page(struct page *page)
470{
471 int nid = page_to_nid(page);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700472
Mel Gormanef8f2322016-07-28 15:46:05 -0700473 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700474}
475
Mel Gormanef8f2322016-07-28 15:46:05 -0700476static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz,
477 struct mem_cgroup_tree_per_node *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800478 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700479{
480 struct rb_node **p = &mctz->rb_root.rb_node;
481 struct rb_node *parent = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700482 struct mem_cgroup_per_node *mz_node;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700483 bool rightmost = true;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700484
485 if (mz->on_tree)
486 return;
487
488 mz->usage_in_excess = new_usage_in_excess;
489 if (!mz->usage_in_excess)
490 return;
491 while (*p) {
492 parent = *p;
Mel Gormanef8f2322016-07-28 15:46:05 -0700493 mz_node = rb_entry(parent, struct mem_cgroup_per_node,
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700494 tree_node);
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700495 if (mz->usage_in_excess < mz_node->usage_in_excess) {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700496 p = &(*p)->rb_left;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700497 rightmost = false;
Miaohe Lin378876b2020-12-14 19:06:28 -0800498 } else {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700499 p = &(*p)->rb_right;
Miaohe Lin378876b2020-12-14 19:06:28 -0800500 }
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700501 }
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700502
503 if (rightmost)
504 mctz->rb_rightmost = &mz->tree_node;
505
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700506 rb_link_node(&mz->tree_node, parent, p);
507 rb_insert_color(&mz->tree_node, &mctz->rb_root);
508 mz->on_tree = true;
509}
510
Mel Gormanef8f2322016-07-28 15:46:05 -0700511static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
512 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700513{
514 if (!mz->on_tree)
515 return;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700516
517 if (&mz->tree_node == mctz->rb_rightmost)
518 mctz->rb_rightmost = rb_prev(&mz->tree_node);
519
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700520 rb_erase(&mz->tree_node, &mctz->rb_root);
521 mz->on_tree = false;
522}
523
Mel Gormanef8f2322016-07-28 15:46:05 -0700524static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
525 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700526{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700527 unsigned long flags;
528
529 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700530 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700531 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700532}
533
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800534static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
535{
536 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700537 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800538 unsigned long excess = 0;
539
540 if (nr_pages > soft_limit)
541 excess = nr_pages - soft_limit;
542
543 return excess;
544}
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700545
546static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
547{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800548 unsigned long excess;
Mel Gormanef8f2322016-07-28 15:46:05 -0700549 struct mem_cgroup_per_node *mz;
550 struct mem_cgroup_tree_per_node *mctz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700551
Jianyu Zhane2318752014-06-06 14:38:20 -0700552 mctz = soft_limit_tree_from_page(page);
Laurent Dufourbfc72282017-03-09 16:17:06 -0800553 if (!mctz)
554 return;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700555 /*
556 * Necessary to update all ancestors when hierarchy is used.
557 * because their event counter is not touched.
558 */
559 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700560 mz = mem_cgroup_page_nodeinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800561 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700562 /*
563 * We have to update the tree if mz is on RB-tree or
564 * mem is over its softlimit.
565 */
566 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700567 unsigned long flags;
568
569 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700570 /* if on-tree, remove it */
571 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700572 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700573 /*
574 * Insert again. mz->usage_in_excess will be updated.
575 * If excess is 0, no tree ops.
576 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700577 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700578 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700579 }
580 }
581}
582
583static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
584{
Mel Gormanef8f2322016-07-28 15:46:05 -0700585 struct mem_cgroup_tree_per_node *mctz;
586 struct mem_cgroup_per_node *mz;
587 int nid;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700588
Jianyu Zhane2318752014-06-06 14:38:20 -0700589 for_each_node(nid) {
Johannes Weinera3747b52021-04-29 22:56:14 -0700590 mz = memcg->nodeinfo[nid];
Mel Gormanef8f2322016-07-28 15:46:05 -0700591 mctz = soft_limit_tree_node(nid);
Laurent Dufourbfc72282017-03-09 16:17:06 -0800592 if (mctz)
593 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700594 }
595}
596
Mel Gormanef8f2322016-07-28 15:46:05 -0700597static struct mem_cgroup_per_node *
598__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700599{
Mel Gormanef8f2322016-07-28 15:46:05 -0700600 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700601
602retry:
603 mz = NULL;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700604 if (!mctz->rb_rightmost)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700605 goto done; /* Nothing to reclaim from */
606
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700607 mz = rb_entry(mctz->rb_rightmost,
608 struct mem_cgroup_per_node, tree_node);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700609 /*
610 * Remove the node now but someone else can add it back,
611 * we will to add it back at the end of reclaim to its correct
612 * position in the tree.
613 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700614 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800615 if (!soft_limit_excess(mz->memcg) ||
Shakeel Butt8965aa22020-04-01 21:07:10 -0700616 !css_tryget(&mz->memcg->css))
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700617 goto retry;
618done:
619 return mz;
620}
621
Mel Gormanef8f2322016-07-28 15:46:05 -0700622static struct mem_cgroup_per_node *
623mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700624{
Mel Gormanef8f2322016-07-28 15:46:05 -0700625 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700626
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700627 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700628 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700629 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700630 return mz;
631}
632
Shakeel Butt11192d92021-11-05 13:37:31 -0700633/*
634 * memcg and lruvec stats flushing
635 *
636 * Many codepaths leading to stats update or read are performance sensitive and
637 * adding stats flushing in such codepaths is not desirable. So, to optimize the
638 * flushing the kernel does:
639 *
640 * 1) Periodically and asynchronously flush the stats every 2 seconds to not let
641 * rstat update tree grow unbounded.
642 *
643 * 2) Flush the stats synchronously on reader side only when there are more than
644 * (MEMCG_CHARGE_BATCH * nr_cpus) update events. Though this optimization
645 * will let stats be out of sync by atmost (MEMCG_CHARGE_BATCH * nr_cpus) but
646 * only for 2 seconds due to (1).
647 */
648static void flush_memcg_stats_dwork(struct work_struct *w);
649static DECLARE_DEFERRABLE_WORK(stats_flush_dwork, flush_memcg_stats_dwork);
650static DEFINE_SPINLOCK(stats_flush_lock);
651static DEFINE_PER_CPU(unsigned int, stats_updates);
652static atomic_t stats_flush_threshold = ATOMIC_INIT(0);
653
654static inline void memcg_rstat_updated(struct mem_cgroup *memcg)
655{
656 cgroup_rstat_updated(memcg->css.cgroup, smp_processor_id());
657 if (!(__this_cpu_inc_return(stats_updates) % MEMCG_CHARGE_BATCH))
658 atomic_inc(&stats_flush_threshold);
659}
660
661static void __mem_cgroup_flush_stats(void)
662{
Shakeel Buttfd25a9e2021-11-05 13:37:34 -0700663 unsigned long flag;
664
665 if (!spin_trylock_irqsave(&stats_flush_lock, flag))
Shakeel Butt11192d92021-11-05 13:37:31 -0700666 return;
667
668 cgroup_rstat_flush_irqsafe(root_mem_cgroup->css.cgroup);
669 atomic_set(&stats_flush_threshold, 0);
Shakeel Buttfd25a9e2021-11-05 13:37:34 -0700670 spin_unlock_irqrestore(&stats_flush_lock, flag);
Shakeel Butt11192d92021-11-05 13:37:31 -0700671}
672
673void mem_cgroup_flush_stats(void)
674{
675 if (atomic_read(&stats_flush_threshold) > num_online_cpus())
676 __mem_cgroup_flush_stats();
677}
678
679static void flush_memcg_stats_dwork(struct work_struct *w)
680{
681 mem_cgroup_flush_stats();
682 queue_delayed_work(system_unbound_wq, &stats_flush_dwork, 2UL*HZ);
683}
684
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700685/**
686 * __mod_memcg_state - update cgroup memory statistics
687 * @memcg: the memory cgroup
688 * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item
689 * @val: delta to add to the counter, can be negative
690 */
691void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val)
692{
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700693 if (mem_cgroup_disabled())
694 return;
695
Johannes Weiner2d146aa2021-04-29 22:56:26 -0700696 __this_cpu_add(memcg->vmstats_percpu->state[idx], val);
Shakeel Butt11192d92021-11-05 13:37:31 -0700697 memcg_rstat_updated(memcg);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700698}
699
Johannes Weiner2d146aa2021-04-29 22:56:26 -0700700/* idx can be of type enum memcg_stat_item or node_stat_item. */
Johannes Weinera18e6e62021-04-29 22:56:17 -0700701static unsigned long memcg_page_state_local(struct mem_cgroup *memcg, int idx)
702{
703 long x = 0;
704 int cpu;
705
706 for_each_possible_cpu(cpu)
Johannes Weiner2d146aa2021-04-29 22:56:26 -0700707 x += per_cpu(memcg->vmstats_percpu->state[idx], cpu);
Johannes Weinera18e6e62021-04-29 22:56:17 -0700708#ifdef CONFIG_SMP
709 if (x < 0)
710 x = 0;
711#endif
712 return x;
713}
714
Roman Gushchineedc4e52020-08-06 23:20:32 -0700715void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
716 int val)
717{
718 struct mem_cgroup_per_node *pn;
719 struct mem_cgroup *memcg;
Roman Gushchineedc4e52020-08-06 23:20:32 -0700720
721 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
722 memcg = pn->memcg;
723
724 /* Update memcg */
Shakeel Butt11192d92021-11-05 13:37:31 -0700725 __this_cpu_add(memcg->vmstats_percpu->state[idx], val);
Roman Gushchineedc4e52020-08-06 23:20:32 -0700726
727 /* Update lruvec */
Shakeel Butt7e1c0d62021-09-02 14:55:00 -0700728 __this_cpu_add(pn->lruvec_stats_percpu->state[idx], val);
Shakeel Butt11192d92021-11-05 13:37:31 -0700729
730 memcg_rstat_updated(memcg);
Roman Gushchineedc4e52020-08-06 23:20:32 -0700731}
732
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700733/**
734 * __mod_lruvec_state - update lruvec memory statistics
735 * @lruvec: the lruvec
736 * @idx: the stat item
737 * @val: delta to add to the counter, can be negative
738 *
739 * The lruvec is the intersection of the NUMA node and a cgroup. This
740 * function updates the all three counters that are affected by a
741 * change of state at this level: per-node, per-cgroup, per-lruvec.
742 */
743void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
744 int val)
745{
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700746 /* Update node */
Roman Gushchineedc4e52020-08-06 23:20:32 -0700747 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700748
Roman Gushchineedc4e52020-08-06 23:20:32 -0700749 /* Update memcg and lruvec */
750 if (!mem_cgroup_disabled())
751 __mod_memcg_lruvec_state(lruvec, idx, val);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700752}
753
Shakeel Buttc47d5032020-12-14 19:07:14 -0800754void __mod_lruvec_page_state(struct page *page, enum node_stat_item idx,
755 int val)
756{
757 struct page *head = compound_head(page); /* rmap on tail pages */
Muchun Songb4e0b682021-04-29 22:56:52 -0700758 struct mem_cgroup *memcg;
Shakeel Buttc47d5032020-12-14 19:07:14 -0800759 pg_data_t *pgdat = page_pgdat(page);
760 struct lruvec *lruvec;
761
Muchun Songb4e0b682021-04-29 22:56:52 -0700762 rcu_read_lock();
763 memcg = page_memcg(head);
Shakeel Buttc47d5032020-12-14 19:07:14 -0800764 /* Untracked pages have no memcg, no lruvec. Update only the node */
Linus Torvaldsd635a692020-12-15 13:22:29 -0800765 if (!memcg) {
Muchun Songb4e0b682021-04-29 22:56:52 -0700766 rcu_read_unlock();
Shakeel Buttc47d5032020-12-14 19:07:14 -0800767 __mod_node_page_state(pgdat, idx, val);
768 return;
769 }
770
Linus Torvaldsd635a692020-12-15 13:22:29 -0800771 lruvec = mem_cgroup_lruvec(memcg, pgdat);
Shakeel Buttc47d5032020-12-14 19:07:14 -0800772 __mod_lruvec_state(lruvec, idx, val);
Muchun Songb4e0b682021-04-29 22:56:52 -0700773 rcu_read_unlock();
Shakeel Buttc47d5032020-12-14 19:07:14 -0800774}
Shakeel Buttf0c0c112020-12-14 19:07:17 -0800775EXPORT_SYMBOL(__mod_lruvec_page_state);
Shakeel Buttc47d5032020-12-14 19:07:14 -0800776
Muchun Songda3ceef2020-12-14 19:07:04 -0800777void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val)
Roman Gushchinec9f0232019-08-13 15:37:41 -0700778{
Roman Gushchin4f103c62020-04-01 21:06:36 -0700779 pg_data_t *pgdat = page_pgdat(virt_to_page(p));
Roman Gushchinec9f0232019-08-13 15:37:41 -0700780 struct mem_cgroup *memcg;
781 struct lruvec *lruvec;
782
783 rcu_read_lock();
Roman Gushchin4f103c62020-04-01 21:06:36 -0700784 memcg = mem_cgroup_from_obj(p);
Roman Gushchinec9f0232019-08-13 15:37:41 -0700785
Muchun Song8faeb1f2020-11-21 22:17:12 -0800786 /*
787 * Untracked pages have no memcg, no lruvec. Update only the
788 * node. If we reparent the slab objects to the root memcg,
789 * when we free the slab object, we need to update the per-memcg
790 * vmstats to keep it correct for the root memcg.
791 */
792 if (!memcg) {
Roman Gushchinec9f0232019-08-13 15:37:41 -0700793 __mod_node_page_state(pgdat, idx, val);
794 } else {
Johannes Weiner867e5e12019-11-30 17:55:34 -0800795 lruvec = mem_cgroup_lruvec(memcg, pgdat);
Roman Gushchinec9f0232019-08-13 15:37:41 -0700796 __mod_lruvec_state(lruvec, idx, val);
797 }
798 rcu_read_unlock();
799}
800
Waiman Long55927112021-06-28 19:37:30 -0700801/*
802 * mod_objcg_mlstate() may be called with irq enabled, so
803 * mod_memcg_lruvec_state() should be used.
804 */
Waiman Long68ac5b32021-06-28 19:37:23 -0700805static inline void mod_objcg_mlstate(struct obj_cgroup *objcg,
806 struct pglist_data *pgdat,
807 enum node_stat_item idx, int nr)
Waiman Longfdbcb2a2021-06-28 19:37:19 -0700808{
809 struct mem_cgroup *memcg;
810 struct lruvec *lruvec;
811
812 rcu_read_lock();
813 memcg = obj_cgroup_memcg(objcg);
814 lruvec = mem_cgroup_lruvec(memcg, pgdat);
Waiman Long55927112021-06-28 19:37:30 -0700815 mod_memcg_lruvec_state(lruvec, idx, nr);
Waiman Longfdbcb2a2021-06-28 19:37:19 -0700816 rcu_read_unlock();
817}
818
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700819/**
820 * __count_memcg_events - account VM events in a cgroup
821 * @memcg: the memory cgroup
822 * @idx: the event item
Ingo Molnarf0953a12021-05-06 18:06:47 -0700823 * @count: the number of events that occurred
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700824 */
825void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx,
826 unsigned long count)
827{
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700828 if (mem_cgroup_disabled())
829 return;
830
Johannes Weiner2d146aa2021-04-29 22:56:26 -0700831 __this_cpu_add(memcg->vmstats_percpu->events[idx], count);
Shakeel Butt11192d92021-11-05 13:37:31 -0700832 memcg_rstat_updated(memcg);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700833}
834
Johannes Weiner42a30032019-05-14 15:47:12 -0700835static unsigned long memcg_events(struct mem_cgroup *memcg, int event)
Johannes Weinere9f89742011-03-23 16:42:37 -0700836{
Johannes Weiner2d146aa2021-04-29 22:56:26 -0700837 return READ_ONCE(memcg->vmstats.events[event]);
Johannes Weinere9f89742011-03-23 16:42:37 -0700838}
839
Johannes Weiner42a30032019-05-14 15:47:12 -0700840static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event)
841{
Johannes Weiner815744d2019-06-13 15:55:46 -0700842 long x = 0;
843 int cpu;
844
845 for_each_possible_cpu(cpu)
Johannes Weiner2d146aa2021-04-29 22:56:26 -0700846 x += per_cpu(memcg->vmstats_percpu->events[event], cpu);
Johannes Weiner815744d2019-06-13 15:55:46 -0700847 return x;
Johannes Weiner42a30032019-05-14 15:47:12 -0700848}
849
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700850static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700851 struct page *page,
Johannes Weiner3fba69a2020-06-03 16:01:31 -0700852 int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800853{
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800854 /* pagein of a big page is an event. So, ignore page size */
855 if (nr_pages > 0)
Johannes Weinerc9019e92018-01-31 16:16:37 -0800856 __count_memcg_events(memcg, PGPGIN, 1);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800857 else {
Johannes Weinerc9019e92018-01-31 16:16:37 -0800858 __count_memcg_events(memcg, PGPGOUT, 1);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800859 nr_pages = -nr_pages; /* for event */
860 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800861
Chris Down871789d2019-05-14 15:46:57 -0700862 __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800863}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800864
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800865static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
866 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800867{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700868 unsigned long val, next;
869
Chris Down871789d2019-05-14 15:46:57 -0700870 val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events);
871 next = __this_cpu_read(memcg->vmstats_percpu->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700872 /* from time_after() in jiffies.h */
Michal Hocko6a1a8b82017-07-10 15:48:53 -0700873 if ((long)(next - val) < 0) {
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800874 switch (target) {
875 case MEM_CGROUP_TARGET_THRESH:
876 next = val + THRESHOLDS_EVENTS_TARGET;
877 break;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700878 case MEM_CGROUP_TARGET_SOFTLIMIT:
879 next = val + SOFTLIMIT_EVENTS_TARGET;
880 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800881 default:
882 break;
883 }
Chris Down871789d2019-05-14 15:46:57 -0700884 __this_cpu_write(memcg->vmstats_percpu->targets[target], next);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800885 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700886 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800887 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800888}
889
890/*
891 * Check events in order.
892 *
893 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700894static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800895{
896 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800897 if (unlikely(mem_cgroup_event_ratelimit(memcg,
898 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700899 bool do_softlimit;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800900
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700901 do_softlimit = mem_cgroup_event_ratelimit(memcg,
902 MEM_CGROUP_TARGET_SOFTLIMIT);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800903 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700904 if (unlikely(do_softlimit))
905 mem_cgroup_update_tree(memcg, page);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700906 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800907}
908
Balbir Singhcf475ad2008-04-29 01:00:16 -0700909struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800910{
Balbir Singh31a78f22008-09-28 23:09:31 +0100911 /*
912 * mm_update_next_owner() may clear mm->owner to NULL
913 * if it races with swapoff, page migration, etc.
914 * So this can be called with p == NULL.
915 */
916 if (unlikely(!p))
917 return NULL;
918
Tejun Heo073219e2014-02-08 10:36:58 -0500919 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800920}
Michal Hocko33398cf2015-09-08 15:01:02 -0700921EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800922
Dan Schatzberg04f94e32021-06-28 19:38:18 -0700923static __always_inline struct mem_cgroup *active_memcg(void)
924{
Vasily Averin55a68c82021-09-02 14:55:49 -0700925 if (!in_task())
Dan Schatzberg04f94e32021-06-28 19:38:18 -0700926 return this_cpu_read(int_active_memcg);
927 else
928 return current->active_memcg;
929}
930
Shakeel Buttd46eb14b2018-08-17 15:46:39 -0700931/**
932 * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg.
933 * @mm: mm from which memcg should be extracted. It can be NULL.
934 *
Dan Schatzberg04f94e32021-06-28 19:38:18 -0700935 * Obtain a reference on mm->memcg and returns it if successful. If mm
936 * is NULL, then the memcg is chosen as follows:
937 * 1) The active memcg, if set.
938 * 2) current->mm->memcg, if available
939 * 3) root memcg
940 * If mem_cgroup is disabled, NULL is returned.
Shakeel Buttd46eb14b2018-08-17 15:46:39 -0700941 */
942struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800943{
Shakeel Buttd46eb14b2018-08-17 15:46:39 -0700944 struct mem_cgroup *memcg;
945
946 if (mem_cgroup_disabled())
947 return NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -0700948
Muchun Song2884b6b2021-06-28 19:37:50 -0700949 /*
950 * Page cache insertions can happen without an
951 * actual mm context, e.g. during disk probing
952 * on boot, loopback IO, acct() writes etc.
953 *
954 * No need to css_get on root memcg as the reference
955 * counting is disabled on the root level in the
956 * cgroup core. See CSS_NO_REF.
957 */
Dan Schatzberg04f94e32021-06-28 19:38:18 -0700958 if (unlikely(!mm)) {
959 memcg = active_memcg();
960 if (unlikely(memcg)) {
961 /* remote memcg must hold a ref */
962 css_get(&memcg->css);
963 return memcg;
964 }
965 mm = current->mm;
966 if (unlikely(!mm))
967 return root_mem_cgroup;
968 }
Muchun Song2884b6b2021-06-28 19:37:50 -0700969
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800970 rcu_read_lock();
971 do {
Muchun Song2884b6b2021-06-28 19:37:50 -0700972 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
973 if (unlikely(!memcg))
Johannes Weinerdf381972014-04-07 15:37:43 -0700974 memcg = root_mem_cgroup;
Roman Gushchin00d484f2019-11-15 17:34:43 -0800975 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800976 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700977 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -0800978}
Shakeel Buttd46eb14b2018-08-17 15:46:39 -0700979EXPORT_SYMBOL(get_mem_cgroup_from_mm);
980
Roman Gushchin4127c652020-10-17 16:13:53 -0700981static __always_inline bool memcg_kmem_bypass(void)
982{
983 /* Allow remote memcg charging from any context. */
984 if (unlikely(active_memcg()))
985 return false;
986
987 /* Memcg to charge can't be determined. */
Vasily Averin61268912021-07-19 13:44:31 +0300988 if (!in_task() || !current->mm || (current->flags & PF_KTHREAD))
Roman Gushchin4127c652020-10-17 16:13:53 -0700989 return true;
990
991 return false;
992}
993
Shakeel Buttf745c6f2018-08-17 15:46:44 -0700994/**
Johannes Weiner56600482012-01-12 17:17:59 -0800995 * mem_cgroup_iter - iterate over memory cgroup hierarchy
996 * @root: hierarchy root
997 * @prev: previously returned memcg, NULL on first invocation
998 * @reclaim: cookie for shared reclaim walks, NULL for full walks
999 *
1000 * Returns references to children of the hierarchy below @root, or
1001 * @root itself, or %NULL after a full round-trip.
1002 *
1003 * Caller must pass the return value in @prev on subsequent
1004 * invocations for reference counting, or use mem_cgroup_iter_break()
1005 * to cancel a hierarchy walk before the round-trip is complete.
1006 *
Miaohe Lin05bdc522020-10-13 16:52:45 -07001007 * Reclaimers can specify a node in @reclaim to divide up the memcgs
1008 * in the hierarchy among all concurrent reclaimers operating on the
1009 * same node.
Johannes Weiner56600482012-01-12 17:17:59 -08001010 */
Andrew Morton694fbc02013-09-24 15:27:37 -07001011struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -08001012 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -07001013 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001014{
Kees Cook3f649ab2020-06-03 13:09:38 -07001015 struct mem_cgroup_reclaim_iter *iter;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001016 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001017 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001018 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001019
Andrew Morton694fbc02013-09-24 15:27:37 -07001020 if (mem_cgroup_disabled())
1021 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001022
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001023 if (!root)
1024 root = root_mem_cgroup;
1025
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001026 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001027 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001028
Michal Hocko542f85f2013-04-29 15:07:15 -07001029 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001030
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001031 if (reclaim) {
Mel Gormanef8f2322016-07-28 15:46:05 -07001032 struct mem_cgroup_per_node *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001033
Johannes Weinera3747b52021-04-29 22:56:14 -07001034 mz = root->nodeinfo[reclaim->pgdat->node_id];
Yafang Shao9da83f32019-11-30 17:50:03 -08001035 iter = &mz->iter;
Michal Hocko5f578162013-04-29 15:07:17 -07001036
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001037 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -07001038 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001039
Vladimir Davydov6df38682015-12-29 14:54:10 -08001040 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001041 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -08001042 if (!pos || css_tryget(&pos->css))
1043 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001044 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -08001045 * css reference reached zero, so iter->position will
1046 * be cleared by ->css_released. However, we should not
1047 * rely on this happening soon, because ->css_released
1048 * is called from a work queue, and by busy-waiting we
1049 * might block it. So we clear iter->position right
1050 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001051 */
Vladimir Davydov6df38682015-12-29 14:54:10 -08001052 (void)cmpxchg(&iter->position, pos, NULL);
1053 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001054 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001055
1056 if (pos)
1057 css = &pos->css;
1058
1059 for (;;) {
1060 css = css_next_descendant_pre(css, &root->css);
1061 if (!css) {
1062 /*
1063 * Reclaimers share the hierarchy walk, and a
1064 * new one might jump in right at the end of
1065 * the hierarchy - make sure they see at least
1066 * one group and restart from the beginning.
1067 */
1068 if (!prev)
1069 continue;
1070 break;
1071 }
1072
1073 /*
1074 * Verify the css and acquire a reference. The root
1075 * is provided by the caller, so we know it's alive
1076 * and kicking, and don't take an extra reference.
1077 */
1078 memcg = mem_cgroup_from_css(css);
1079
1080 if (css == &root->css)
1081 break;
1082
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08001083 if (css_tryget(css))
1084 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001085
1086 memcg = NULL;
1087 }
1088
1089 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001090 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -08001091 * The position could have already been updated by a competing
1092 * thread, so check that the value hasn't changed since we read
1093 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001094 */
Vladimir Davydov6df38682015-12-29 14:54:10 -08001095 (void)cmpxchg(&iter->position, pos, memcg);
1096
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001097 if (pos)
1098 css_put(&pos->css);
1099
1100 if (!memcg)
1101 iter->generation++;
1102 else if (!prev)
1103 reclaim->generation = iter->generation;
1104 }
1105
Michal Hocko542f85f2013-04-29 15:07:15 -07001106out_unlock:
1107 rcu_read_unlock();
Michal Hockoc40046f2013-04-29 15:07:14 -07001108 if (prev && prev != root)
1109 css_put(&prev->css);
1110
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001111 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001112}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001113
Johannes Weiner56600482012-01-12 17:17:59 -08001114/**
1115 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1116 * @root: hierarchy root
1117 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1118 */
1119void mem_cgroup_iter_break(struct mem_cgroup *root,
1120 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001121{
1122 if (!root)
1123 root = root_mem_cgroup;
1124 if (prev && prev != root)
1125 css_put(&prev->css);
1126}
1127
Miles Chen54a83d62019-08-13 15:37:28 -07001128static void __invalidate_reclaim_iterators(struct mem_cgroup *from,
1129 struct mem_cgroup *dead_memcg)
Vladimir Davydov6df38682015-12-29 14:54:10 -08001130{
Vladimir Davydov6df38682015-12-29 14:54:10 -08001131 struct mem_cgroup_reclaim_iter *iter;
Mel Gormanef8f2322016-07-28 15:46:05 -07001132 struct mem_cgroup_per_node *mz;
1133 int nid;
Vladimir Davydov6df38682015-12-29 14:54:10 -08001134
Miles Chen54a83d62019-08-13 15:37:28 -07001135 for_each_node(nid) {
Johannes Weinera3747b52021-04-29 22:56:14 -07001136 mz = from->nodeinfo[nid];
Yafang Shao9da83f32019-11-30 17:50:03 -08001137 iter = &mz->iter;
1138 cmpxchg(&iter->position, dead_memcg, NULL);
Vladimir Davydov6df38682015-12-29 14:54:10 -08001139 }
1140}
1141
Miles Chen54a83d62019-08-13 15:37:28 -07001142static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
1143{
1144 struct mem_cgroup *memcg = dead_memcg;
1145 struct mem_cgroup *last;
1146
1147 do {
1148 __invalidate_reclaim_iterators(memcg, dead_memcg);
1149 last = memcg;
1150 } while ((memcg = parent_mem_cgroup(memcg)));
1151
1152 /*
1153 * When cgruop1 non-hierarchy mode is used,
1154 * parent_mem_cgroup() does not walk all the way up to the
1155 * cgroup root (root_mem_cgroup). So we have to handle
1156 * dead_memcg from cgroup root separately.
1157 */
1158 if (last != root_mem_cgroup)
1159 __invalidate_reclaim_iterators(root_mem_cgroup,
1160 dead_memcg);
1161}
1162
Johannes Weiner925b7672012-01-12 17:18:15 -08001163/**
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001164 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
1165 * @memcg: hierarchy root
1166 * @fn: function to call for each task
1167 * @arg: argument passed to @fn
1168 *
1169 * This function iterates over tasks attached to @memcg or to any of its
1170 * descendants and calls @fn for each task. If @fn returns a non-zero
1171 * value, the function breaks the iteration loop and returns the value.
1172 * Otherwise, it will iterate over all tasks and return 0.
1173 *
1174 * This function must not be called for the root memory cgroup.
1175 */
1176int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
1177 int (*fn)(struct task_struct *, void *), void *arg)
1178{
1179 struct mem_cgroup *iter;
1180 int ret = 0;
1181
1182 BUG_ON(memcg == root_mem_cgroup);
1183
1184 for_each_mem_cgroup_tree(iter, memcg) {
1185 struct css_task_iter it;
1186 struct task_struct *task;
1187
Tetsuo Handaf168a9a2019-07-11 21:00:20 -07001188 css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001189 while (!ret && (task = css_task_iter_next(&it)))
1190 ret = fn(task, arg);
1191 css_task_iter_end(&it);
1192 if (ret) {
1193 mem_cgroup_iter_break(memcg, iter);
1194 break;
1195 }
1196 }
1197 return ret;
1198}
1199
Alex Shi6168d0d2020-12-15 12:34:29 -08001200#ifdef CONFIG_DEBUG_VM
1201void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page)
1202{
1203 struct mem_cgroup *memcg;
1204
1205 if (mem_cgroup_disabled())
1206 return;
1207
1208 memcg = page_memcg(page);
1209
1210 if (!memcg)
1211 VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != root_mem_cgroup, page);
1212 else
1213 VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != memcg, page);
1214}
1215#endif
1216
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001217/**
Alex Shi6168d0d2020-12-15 12:34:29 -08001218 * lock_page_lruvec - lock and return lruvec for a given page.
1219 * @page: the page
1220 *
Alex Shid7e3aba2021-02-24 12:03:47 -08001221 * These functions are safe to use under any of the following conditions:
1222 * - page locked
1223 * - PageLRU cleared
1224 * - lock_page_memcg()
1225 * - page->_refcount is zero
Alex Shi6168d0d2020-12-15 12:34:29 -08001226 */
1227struct lruvec *lock_page_lruvec(struct page *page)
1228{
1229 struct lruvec *lruvec;
Alex Shi6168d0d2020-12-15 12:34:29 -08001230
Muchun Songa9842262021-06-28 19:37:53 -07001231 lruvec = mem_cgroup_page_lruvec(page);
Alex Shi6168d0d2020-12-15 12:34:29 -08001232 spin_lock(&lruvec->lru_lock);
Alex Shi6168d0d2020-12-15 12:34:29 -08001233
1234 lruvec_memcg_debug(lruvec, page);
1235
1236 return lruvec;
1237}
1238
1239struct lruvec *lock_page_lruvec_irq(struct page *page)
1240{
1241 struct lruvec *lruvec;
Alex Shi6168d0d2020-12-15 12:34:29 -08001242
Muchun Songa9842262021-06-28 19:37:53 -07001243 lruvec = mem_cgroup_page_lruvec(page);
Alex Shi6168d0d2020-12-15 12:34:29 -08001244 spin_lock_irq(&lruvec->lru_lock);
Alex Shi6168d0d2020-12-15 12:34:29 -08001245
1246 lruvec_memcg_debug(lruvec, page);
1247
1248 return lruvec;
1249}
1250
1251struct lruvec *lock_page_lruvec_irqsave(struct page *page, unsigned long *flags)
1252{
1253 struct lruvec *lruvec;
Alex Shi6168d0d2020-12-15 12:34:29 -08001254
Muchun Songa9842262021-06-28 19:37:53 -07001255 lruvec = mem_cgroup_page_lruvec(page);
Alex Shi6168d0d2020-12-15 12:34:29 -08001256 spin_lock_irqsave(&lruvec->lru_lock, *flags);
Alex Shi6168d0d2020-12-15 12:34:29 -08001257
1258 lruvec_memcg_debug(lruvec, page);
1259
1260 return lruvec;
1261}
1262
1263/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001264 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1265 * @lruvec: mem_cgroup per zone lru vector
1266 * @lru: index of lru list the page is sitting on
Michal Hockob4536f0c82017-01-10 16:58:04 -08001267 * @zid: zone id of the accounted pages
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001268 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001269 *
Hugh Dickinsca707232016-05-19 17:12:35 -07001270 * This function must be called under lru_lock, just before a page is added
1271 * to or just after a page is removed from an lru list (that ordering being
1272 * so as to allow it to check that lru_size 0 is consistent with list_empty).
Johannes Weiner925b7672012-01-12 17:18:15 -08001273 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001274void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
Michal Hockob4536f0c82017-01-10 16:58:04 -08001275 int zid, int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001276{
Mel Gormanef8f2322016-07-28 15:46:05 -07001277 struct mem_cgroup_per_node *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001278 unsigned long *lru_size;
Hugh Dickinsca707232016-05-19 17:12:35 -07001279 long size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001280
1281 if (mem_cgroup_disabled())
1282 return;
1283
Mel Gormanef8f2322016-07-28 15:46:05 -07001284 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
Michal Hockob4536f0c82017-01-10 16:58:04 -08001285 lru_size = &mz->lru_zone_size[zid][lru];
Hugh Dickinsca707232016-05-19 17:12:35 -07001286
1287 if (nr_pages < 0)
1288 *lru_size += nr_pages;
1289
1290 size = *lru_size;
Michal Hockob4536f0c82017-01-10 16:58:04 -08001291 if (WARN_ONCE(size < 0,
1292 "%s(%p, %d, %d): lru_size %ld\n",
1293 __func__, lruvec, lru, nr_pages, size)) {
Hugh Dickinsca707232016-05-19 17:12:35 -07001294 VM_BUG_ON(1);
1295 *lru_size = 0;
1296 }
1297
1298 if (nr_pages > 0)
1299 *lru_size += nr_pages;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001300}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001301
Johannes Weiner19942822011-02-01 15:52:43 -08001302/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001303 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001304 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001305 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001306 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001307 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001308 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001309static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001310{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001311 unsigned long margin = 0;
1312 unsigned long count;
1313 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001314
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001315 count = page_counter_read(&memcg->memory);
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001316 limit = READ_ONCE(memcg->memory.max);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001317 if (count < limit)
1318 margin = limit - count;
1319
Johannes Weiner7941d212016-01-14 15:21:23 -08001320 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001321 count = page_counter_read(&memcg->memsw);
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001322 limit = READ_ONCE(memcg->memsw.max);
Kaixu Xia1c4448e2020-06-01 21:49:36 -07001323 if (count < limit)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001324 margin = min(margin, limit - count);
Li RongQingcbedbac2016-05-27 14:27:43 -07001325 else
1326 margin = 0;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001327 }
1328
1329 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001330}
1331
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001332/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001333 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001334 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001335 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1336 * moving cgroups. This is for waiting at high-memory pressure
1337 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001338 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001339static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001340{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001341 struct mem_cgroup *from;
1342 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001343 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001344 /*
1345 * Unlike task_move routines, we access mc.to, mc.from not under
1346 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1347 */
1348 spin_lock(&mc.lock);
1349 from = mc.from;
1350 to = mc.to;
1351 if (!from)
1352 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001353
Johannes Weiner2314b422014-12-10 15:44:33 -08001354 ret = mem_cgroup_is_descendant(from, memcg) ||
1355 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001356unlock:
1357 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001358 return ret;
1359}
1360
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001361static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001362{
1363 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001364 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001365 DEFINE_WAIT(wait);
1366 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1367 /* moving charge context might have finished. */
1368 if (mc.moving_task)
1369 schedule();
1370 finish_wait(&mc.waitq, &wait);
1371 return true;
1372 }
1373 }
1374 return false;
1375}
1376
Muchun Song5f9a4f42020-10-13 16:52:59 -07001377struct memory_stat {
1378 const char *name;
Muchun Song5f9a4f42020-10-13 16:52:59 -07001379 unsigned int idx;
1380};
1381
Muchun Song57b28472021-02-24 12:03:31 -08001382static const struct memory_stat memory_stats[] = {
Muchun Songfff66b72021-02-24 12:03:43 -08001383 { "anon", NR_ANON_MAPPED },
1384 { "file", NR_FILE_PAGES },
1385 { "kernel_stack", NR_KERNEL_STACK_KB },
1386 { "pagetables", NR_PAGETABLE },
1387 { "percpu", MEMCG_PERCPU_B },
1388 { "sock", MEMCG_SOCK },
1389 { "shmem", NR_SHMEM },
1390 { "file_mapped", NR_FILE_MAPPED },
1391 { "file_dirty", NR_FILE_DIRTY },
1392 { "file_writeback", NR_WRITEBACK },
Shakeel Buttb6038942021-02-24 12:03:55 -08001393#ifdef CONFIG_SWAP
1394 { "swapcached", NR_SWAPCACHE },
1395#endif
Muchun Song5f9a4f42020-10-13 16:52:59 -07001396#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Muchun Songfff66b72021-02-24 12:03:43 -08001397 { "anon_thp", NR_ANON_THPS },
1398 { "file_thp", NR_FILE_THPS },
1399 { "shmem_thp", NR_SHMEM_THPS },
Muchun Song5f9a4f42020-10-13 16:52:59 -07001400#endif
Muchun Songfff66b72021-02-24 12:03:43 -08001401 { "inactive_anon", NR_INACTIVE_ANON },
1402 { "active_anon", NR_ACTIVE_ANON },
1403 { "inactive_file", NR_INACTIVE_FILE },
1404 { "active_file", NR_ACTIVE_FILE },
1405 { "unevictable", NR_UNEVICTABLE },
1406 { "slab_reclaimable", NR_SLAB_RECLAIMABLE_B },
1407 { "slab_unreclaimable", NR_SLAB_UNRECLAIMABLE_B },
Muchun Song5f9a4f42020-10-13 16:52:59 -07001408
1409 /* The memory events */
Muchun Songfff66b72021-02-24 12:03:43 -08001410 { "workingset_refault_anon", WORKINGSET_REFAULT_ANON },
1411 { "workingset_refault_file", WORKINGSET_REFAULT_FILE },
1412 { "workingset_activate_anon", WORKINGSET_ACTIVATE_ANON },
1413 { "workingset_activate_file", WORKINGSET_ACTIVATE_FILE },
1414 { "workingset_restore_anon", WORKINGSET_RESTORE_ANON },
1415 { "workingset_restore_file", WORKINGSET_RESTORE_FILE },
1416 { "workingset_nodereclaim", WORKINGSET_NODERECLAIM },
Muchun Song5f9a4f42020-10-13 16:52:59 -07001417};
1418
Muchun Songfff66b72021-02-24 12:03:43 -08001419/* Translate stat items to the correct unit for memory.stat output */
1420static int memcg_page_state_unit(int item)
1421{
1422 switch (item) {
1423 case MEMCG_PERCPU_B:
1424 case NR_SLAB_RECLAIMABLE_B:
1425 case NR_SLAB_UNRECLAIMABLE_B:
1426 case WORKINGSET_REFAULT_ANON:
1427 case WORKINGSET_REFAULT_FILE:
1428 case WORKINGSET_ACTIVATE_ANON:
1429 case WORKINGSET_ACTIVATE_FILE:
1430 case WORKINGSET_RESTORE_ANON:
1431 case WORKINGSET_RESTORE_FILE:
1432 case WORKINGSET_NODERECLAIM:
1433 return 1;
1434 case NR_KERNEL_STACK_KB:
1435 return SZ_1K;
1436 default:
1437 return PAGE_SIZE;
1438 }
1439}
1440
1441static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg,
1442 int item)
1443{
1444 return memcg_page_state(memcg, item) * memcg_page_state_unit(item);
1445}
1446
Johannes Weinerc8713d02019-07-11 20:55:59 -07001447static char *memory_stat_format(struct mem_cgroup *memcg)
1448{
1449 struct seq_buf s;
1450 int i;
Johannes Weiner71cd3112017-05-03 14:55:13 -07001451
Johannes Weinerc8713d02019-07-11 20:55:59 -07001452 seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE);
1453 if (!s.buffer)
1454 return NULL;
1455
1456 /*
1457 * Provide statistics on the state of the memory subsystem as
1458 * well as cumulative event counters that show past behavior.
1459 *
1460 * This list is ordered following a combination of these gradients:
1461 * 1) generic big picture -> specifics and details
1462 * 2) reflecting userspace activity -> reflecting kernel heuristics
1463 *
1464 * Current memory state:
1465 */
Shakeel Buttfd25a9e2021-11-05 13:37:34 -07001466 mem_cgroup_flush_stats();
Johannes Weinerc8713d02019-07-11 20:55:59 -07001467
Muchun Song5f9a4f42020-10-13 16:52:59 -07001468 for (i = 0; i < ARRAY_SIZE(memory_stats); i++) {
1469 u64 size;
Johannes Weinerc8713d02019-07-11 20:55:59 -07001470
Muchun Songfff66b72021-02-24 12:03:43 -08001471 size = memcg_page_state_output(memcg, memory_stats[i].idx);
Muchun Song5f9a4f42020-10-13 16:52:59 -07001472 seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size);
Johannes Weinerc8713d02019-07-11 20:55:59 -07001473
Muchun Song5f9a4f42020-10-13 16:52:59 -07001474 if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) {
Muchun Songfff66b72021-02-24 12:03:43 -08001475 size += memcg_page_state_output(memcg,
1476 NR_SLAB_RECLAIMABLE_B);
Muchun Song5f9a4f42020-10-13 16:52:59 -07001477 seq_buf_printf(&s, "slab %llu\n", size);
1478 }
1479 }
Johannes Weinerc8713d02019-07-11 20:55:59 -07001480
1481 /* Accumulated memory events */
1482
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001483 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT),
1484 memcg_events(memcg, PGFAULT));
1485 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT),
1486 memcg_events(memcg, PGMAJFAULT));
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001487 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL),
1488 memcg_events(memcg, PGREFILL));
Johannes Weinerc8713d02019-07-11 20:55:59 -07001489 seq_buf_printf(&s, "pgscan %lu\n",
1490 memcg_events(memcg, PGSCAN_KSWAPD) +
1491 memcg_events(memcg, PGSCAN_DIRECT));
1492 seq_buf_printf(&s, "pgsteal %lu\n",
1493 memcg_events(memcg, PGSTEAL_KSWAPD) +
1494 memcg_events(memcg, PGSTEAL_DIRECT));
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001495 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE),
1496 memcg_events(memcg, PGACTIVATE));
1497 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE),
1498 memcg_events(memcg, PGDEACTIVATE));
1499 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE),
1500 memcg_events(memcg, PGLAZYFREE));
1501 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED),
1502 memcg_events(memcg, PGLAZYFREED));
Johannes Weinerc8713d02019-07-11 20:55:59 -07001503
1504#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001505 seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC),
Johannes Weinerc8713d02019-07-11 20:55:59 -07001506 memcg_events(memcg, THP_FAULT_ALLOC));
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001507 seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC),
Johannes Weinerc8713d02019-07-11 20:55:59 -07001508 memcg_events(memcg, THP_COLLAPSE_ALLOC));
1509#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1510
1511 /* The above should easily fit into one page */
1512 WARN_ON_ONCE(seq_buf_has_overflowed(&s));
1513
1514 return s.buffer;
1515}
Johannes Weiner71cd3112017-05-03 14:55:13 -07001516
Sha Zhengju58cf1882013-02-22 16:32:05 -08001517#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001518/**
yuzhoujianf0c867d2018-12-28 00:36:10 -08001519 * mem_cgroup_print_oom_context: Print OOM information relevant to
1520 * memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001521 * @memcg: The memory cgroup that went over limit
1522 * @p: Task that is going to be killed
1523 *
1524 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1525 * enabled
1526 */
yuzhoujianf0c867d2018-12-28 00:36:10 -08001527void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p)
1528{
1529 rcu_read_lock();
1530
1531 if (memcg) {
1532 pr_cont(",oom_memcg=");
1533 pr_cont_cgroup_path(memcg->css.cgroup);
1534 } else
1535 pr_cont(",global_oom");
1536 if (p) {
1537 pr_cont(",task_memcg=");
1538 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1539 }
1540 rcu_read_unlock();
1541}
1542
1543/**
1544 * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to
1545 * memory controller.
1546 * @memcg: The memory cgroup that went over limit
1547 */
1548void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
Balbir Singhe2224322009-04-02 16:57:39 -07001549{
Johannes Weinerc8713d02019-07-11 20:55:59 -07001550 char *buf;
Balbir Singhe2224322009-04-02 16:57:39 -07001551
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001552 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1553 K((u64)page_counter_read(&memcg->memory)),
Chris Down15b42562020-04-01 21:07:20 -07001554 K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt);
Johannes Weinerc8713d02019-07-11 20:55:59 -07001555 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
1556 pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n",
1557 K((u64)page_counter_read(&memcg->swap)),
Chris Down32d087c2020-04-01 21:07:30 -07001558 K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt);
Johannes Weinerc8713d02019-07-11 20:55:59 -07001559 else {
1560 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1561 K((u64)page_counter_read(&memcg->memsw)),
1562 K((u64)memcg->memsw.max), memcg->memsw.failcnt);
1563 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1564 K((u64)page_counter_read(&memcg->kmem)),
1565 K((u64)memcg->kmem.max), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001566 }
Johannes Weinerc8713d02019-07-11 20:55:59 -07001567
1568 pr_info("Memory cgroup stats for ");
1569 pr_cont_cgroup_path(memcg->css.cgroup);
1570 pr_cont(":");
1571 buf = memory_stat_format(memcg);
1572 if (!buf)
1573 return;
1574 pr_info("%s", buf);
1575 kfree(buf);
Balbir Singhe2224322009-04-02 16:57:39 -07001576}
1577
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001578/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001579 * Return the memory (and swap, if configured) limit for a memcg.
1580 */
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001581unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001582{
Waiman Long8d387a52020-10-13 16:52:52 -07001583 unsigned long max = READ_ONCE(memcg->memory.max);
David Rientjesa63d83f2010-08-09 17:19:46 -07001584
Waiman Long8d387a52020-10-13 16:52:52 -07001585 if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
1586 if (mem_cgroup_swappiness(memcg))
1587 max += min(READ_ONCE(memcg->swap.max),
1588 (unsigned long)total_swap_pages);
1589 } else { /* v1 */
1590 if (mem_cgroup_swappiness(memcg)) {
1591 /* Calculate swap excess capacity from memsw limit */
1592 unsigned long swap = READ_ONCE(memcg->memsw.max) - max;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001593
Waiman Long8d387a52020-10-13 16:52:52 -07001594 max += min(swap, (unsigned long)total_swap_pages);
1595 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001596 }
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001597 return max;
David Rientjesa63d83f2010-08-09 17:19:46 -07001598}
1599
Chris Down9783aa92019-10-06 17:58:32 -07001600unsigned long mem_cgroup_size(struct mem_cgroup *memcg)
1601{
1602 return page_counter_read(&memcg->memory);
1603}
1604
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001605static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001606 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001607{
David Rientjes6e0fc462015-09-08 15:00:36 -07001608 struct oom_control oc = {
1609 .zonelist = NULL,
1610 .nodemask = NULL,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001611 .memcg = memcg,
David Rientjes6e0fc462015-09-08 15:00:36 -07001612 .gfp_mask = gfp_mask,
1613 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001614 };
Yafang Shao1378b372020-08-06 23:22:08 -07001615 bool ret = true;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001616
Tetsuo Handa7775fac2019-03-05 15:46:47 -08001617 if (mutex_lock_killable(&oom_lock))
1618 return true;
Yafang Shao1378b372020-08-06 23:22:08 -07001619
1620 if (mem_cgroup_margin(memcg) >= (1 << order))
1621 goto unlock;
1622
Tetsuo Handa7775fac2019-03-05 15:46:47 -08001623 /*
1624 * A few threads which were not waiting at mutex_lock_killable() can
1625 * fail to bail out. Therefore, check again after holding oom_lock.
1626 */
Vasily Averina4ebf1b2021-11-05 13:38:09 -07001627 ret = task_is_dying() || out_of_memory(&oc);
Yafang Shao1378b372020-08-06 23:22:08 -07001628
1629unlock:
Johannes Weinerdc564012015-06-24 16:57:19 -07001630 mutex_unlock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001631 return ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001632}
1633
Andrew Morton0608f432013-09-24 15:27:41 -07001634static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
Mel Gormanef8f2322016-07-28 15:46:05 -07001635 pg_data_t *pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001636 gfp_t gfp_mask,
1637 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001638{
Andrew Morton0608f432013-09-24 15:27:41 -07001639 struct mem_cgroup *victim = NULL;
1640 int total = 0;
1641 int loop = 0;
1642 unsigned long excess;
1643 unsigned long nr_scanned;
1644 struct mem_cgroup_reclaim_cookie reclaim = {
Mel Gormanef8f2322016-07-28 15:46:05 -07001645 .pgdat = pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001646 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001647
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001648 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001649
Andrew Morton0608f432013-09-24 15:27:41 -07001650 while (1) {
1651 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1652 if (!victim) {
1653 loop++;
1654 if (loop >= 2) {
1655 /*
1656 * If we have not been able to reclaim
1657 * anything, it might because there are
1658 * no reclaimable pages under this hierarchy
1659 */
1660 if (!total)
1661 break;
1662 /*
1663 * We want to do more targeted reclaim.
1664 * excess >> 2 is not to excessive so as to
1665 * reclaim too much, nor too less that we keep
1666 * coming back to reclaim from this cgroup
1667 */
1668 if (total >= (excess >> 2) ||
1669 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1670 break;
1671 }
1672 continue;
1673 }
Mel Gormana9dd0a82016-07-28 15:46:02 -07001674 total += mem_cgroup_shrink_node(victim, gfp_mask, false,
Mel Gormanef8f2322016-07-28 15:46:05 -07001675 pgdat, &nr_scanned);
Andrew Morton0608f432013-09-24 15:27:41 -07001676 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001677 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001678 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001679 }
Andrew Morton0608f432013-09-24 15:27:41 -07001680 mem_cgroup_iter_break(root_memcg, victim);
1681 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001682}
1683
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001684#ifdef CONFIG_LOCKDEP
1685static struct lockdep_map memcg_oom_lock_dep_map = {
1686 .name = "memcg_oom_lock",
1687};
1688#endif
1689
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001690static DEFINE_SPINLOCK(memcg_oom_lock);
1691
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001692/*
1693 * Check OOM-Killer is already running under our hierarchy.
1694 * If someone is running, return false.
1695 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001696static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001697{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001698 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001699
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001700 spin_lock(&memcg_oom_lock);
1701
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001702 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001703 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001704 /*
1705 * this subtree of our hierarchy is already locked
1706 * so we cannot give a lock.
1707 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001708 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001709 mem_cgroup_iter_break(memcg, iter);
1710 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001711 } else
1712 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001713 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001714
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001715 if (failed) {
1716 /*
1717 * OK, we failed to lock the whole subtree so we have
1718 * to clean up what we set up to the failing subtree
1719 */
1720 for_each_mem_cgroup_tree(iter, memcg) {
1721 if (iter == failed) {
1722 mem_cgroup_iter_break(memcg, iter);
1723 break;
1724 }
1725 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001726 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001727 } else
1728 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001729
1730 spin_unlock(&memcg_oom_lock);
1731
1732 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001733}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001734
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001735static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001736{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001737 struct mem_cgroup *iter;
1738
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001739 spin_lock(&memcg_oom_lock);
Qian Cai5facae42019-09-19 12:09:40 -04001740 mutex_release(&memcg_oom_lock_dep_map, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001741 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001742 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001743 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001744}
1745
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001746static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001747{
1748 struct mem_cgroup *iter;
1749
Tejun Heoc2b42d32015-06-24 16:58:23 -07001750 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001751 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001752 iter->under_oom++;
1753 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001754}
1755
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001756static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001757{
1758 struct mem_cgroup *iter;
1759
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001760 /*
Ingo Molnarf0953a12021-05-06 18:06:47 -07001761 * Be careful about under_oom underflows because a child memcg
Miaohe Lin7a52d4d82020-10-13 16:53:05 -07001762 * could have been added after mem_cgroup_mark_under_oom.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001763 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001764 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001765 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001766 if (iter->under_oom > 0)
1767 iter->under_oom--;
1768 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001769}
1770
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001771static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1772
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001773struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001774 struct mem_cgroup *memcg;
Ingo Molnarac6424b2017-06-20 12:06:13 +02001775 wait_queue_entry_t wait;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001776};
1777
Ingo Molnarac6424b2017-06-20 12:06:13 +02001778static int memcg_oom_wake_function(wait_queue_entry_t *wait,
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001779 unsigned mode, int sync, void *arg)
1780{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001781 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1782 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001783 struct oom_wait_info *oom_wait_info;
1784
1785 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001786 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001787
Johannes Weiner2314b422014-12-10 15:44:33 -08001788 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1789 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001790 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001791 return autoremove_wake_function(wait, mode, sync, arg);
1792}
1793
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001794static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001795{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001796 /*
1797 * For the following lockless ->under_oom test, the only required
1798 * guarantee is that it must see the state asserted by an OOM when
1799 * this function is called as a result of userland actions
1800 * triggered by the notification of the OOM. This is trivially
1801 * achieved by invoking mem_cgroup_mark_under_oom() before
1802 * triggering notification.
1803 */
1804 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001805 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001806}
1807
Michal Hocko29ef6802018-08-17 15:47:11 -07001808enum oom_status {
1809 OOM_SUCCESS,
1810 OOM_FAILED,
1811 OOM_ASYNC,
1812 OOM_SKIPPED
1813};
1814
1815static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001816{
Michal Hocko7056d3a2018-12-28 00:39:57 -08001817 enum oom_status ret;
1818 bool locked;
1819
Michal Hocko29ef6802018-08-17 15:47:11 -07001820 if (order > PAGE_ALLOC_COSTLY_ORDER)
1821 return OOM_SKIPPED;
1822
Roman Gushchin7a1adfd2018-10-26 15:09:48 -07001823 memcg_memory_event(memcg, MEMCG_OOM);
1824
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001825 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001826 * We are in the middle of the charge context here, so we
1827 * don't want to block when potentially sitting on a callstack
1828 * that holds all kinds of filesystem and mm locks.
1829 *
Michal Hocko29ef6802018-08-17 15:47:11 -07001830 * cgroup1 allows disabling the OOM killer and waiting for outside
1831 * handling until the charge can succeed; remember the context and put
1832 * the task to sleep at the end of the page fault when all locks are
1833 * released.
Johannes Weiner49426422013-10-16 13:46:59 -07001834 *
Michal Hocko29ef6802018-08-17 15:47:11 -07001835 * On the other hand, in-kernel OOM killer allows for an async victim
1836 * memory reclaim (oom_reaper) and that means that we are not solely
1837 * relying on the oom victim to make a forward progress and we can
1838 * invoke the oom killer here.
1839 *
1840 * Please note that mem_cgroup_out_of_memory might fail to find a
1841 * victim and then we have to bail out from the charge path.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001842 */
Michal Hocko29ef6802018-08-17 15:47:11 -07001843 if (memcg->oom_kill_disable) {
1844 if (!current->in_user_fault)
1845 return OOM_SKIPPED;
1846 css_get(&memcg->css);
1847 current->memcg_in_oom = memcg;
1848 current->memcg_oom_gfp_mask = mask;
1849 current->memcg_oom_order = order;
1850
1851 return OOM_ASYNC;
1852 }
1853
Michal Hocko7056d3a2018-12-28 00:39:57 -08001854 mem_cgroup_mark_under_oom(memcg);
Michal Hocko29ef6802018-08-17 15:47:11 -07001855
Michal Hocko7056d3a2018-12-28 00:39:57 -08001856 locked = mem_cgroup_oom_trylock(memcg);
1857
1858 if (locked)
1859 mem_cgroup_oom_notify(memcg);
1860
1861 mem_cgroup_unmark_under_oom(memcg);
1862 if (mem_cgroup_out_of_memory(memcg, mask, order))
1863 ret = OOM_SUCCESS;
1864 else
1865 ret = OOM_FAILED;
1866
1867 if (locked)
1868 mem_cgroup_oom_unlock(memcg);
1869
1870 return ret;
Johannes Weiner49426422013-10-16 13:46:59 -07001871}
1872
1873/**
1874 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1875 * @handle: actually kill/wait or just clean up the OOM state
1876 *
1877 * This has to be called at the end of a page fault if the memcg OOM
1878 * handler was enabled.
1879 *
1880 * Memcg supports userspace OOM handling where failed allocations must
1881 * sleep on a waitqueue until the userspace task resolves the
1882 * situation. Sleeping directly in the charge context with all kinds
1883 * of locks held is not a good idea, instead we remember an OOM state
1884 * in the task and mem_cgroup_oom_synchronize() has to be called at
1885 * the end of the page fault to complete the OOM handling.
1886 *
1887 * Returns %true if an ongoing memcg OOM situation was detected and
1888 * completed, %false otherwise.
1889 */
1890bool mem_cgroup_oom_synchronize(bool handle)
1891{
Tejun Heo626ebc42015-11-05 18:46:09 -08001892 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001893 struct oom_wait_info owait;
1894 bool locked;
1895
1896 /* OOM is global, do not handle */
1897 if (!memcg)
1898 return false;
1899
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001900 if (!handle)
Johannes Weiner49426422013-10-16 13:46:59 -07001901 goto cleanup;
1902
1903 owait.memcg = memcg;
1904 owait.wait.flags = 0;
1905 owait.wait.func = memcg_oom_wake_function;
1906 owait.wait.private = current;
Ingo Molnar2055da92017-06-20 12:06:46 +02001907 INIT_LIST_HEAD(&owait.wait.entry);
Johannes Weiner49426422013-10-16 13:46:59 -07001908
1909 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001910 mem_cgroup_mark_under_oom(memcg);
1911
1912 locked = mem_cgroup_oom_trylock(memcg);
1913
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001914 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001915 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001916
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001917 if (locked && !memcg->oom_kill_disable) {
1918 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07001919 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08001920 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1921 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001922 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001923 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07001924 mem_cgroup_unmark_under_oom(memcg);
1925 finish_wait(&memcg_oom_waitq, &owait.wait);
1926 }
1927
1928 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001929 mem_cgroup_oom_unlock(memcg);
1930 /*
1931 * There is no guarantee that an OOM-lock contender
1932 * sees the wakeups triggered by the OOM kill
Ingo Molnarf0953a12021-05-06 18:06:47 -07001933 * uncharges. Wake any sleepers explicitly.
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001934 */
1935 memcg_oom_recover(memcg);
1936 }
Johannes Weiner49426422013-10-16 13:46:59 -07001937cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08001938 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001939 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001940 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001941}
1942
Johannes Weinerd7365e72014-10-29 14:50:48 -07001943/**
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07001944 * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM
1945 * @victim: task to be killed by the OOM killer
1946 * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM
1947 *
1948 * Returns a pointer to a memory cgroup, which has to be cleaned up
1949 * by killing all belonging OOM-killable tasks.
1950 *
1951 * Caller has to call mem_cgroup_put() on the returned non-NULL memcg.
1952 */
1953struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
1954 struct mem_cgroup *oom_domain)
1955{
1956 struct mem_cgroup *oom_group = NULL;
1957 struct mem_cgroup *memcg;
1958
1959 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
1960 return NULL;
1961
1962 if (!oom_domain)
1963 oom_domain = root_mem_cgroup;
1964
1965 rcu_read_lock();
1966
1967 memcg = mem_cgroup_from_task(victim);
1968 if (memcg == root_mem_cgroup)
1969 goto out;
1970
1971 /*
Roman Gushchin48fe2672020-04-01 21:07:39 -07001972 * If the victim task has been asynchronously moved to a different
1973 * memory cgroup, we might end up killing tasks outside oom_domain.
1974 * In this case it's better to ignore memory.group.oom.
1975 */
1976 if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain)))
1977 goto out;
1978
1979 /*
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07001980 * Traverse the memory cgroup hierarchy from the victim task's
1981 * cgroup up to the OOMing cgroup (or root) to find the
1982 * highest-level memory cgroup with oom.group set.
1983 */
1984 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
1985 if (memcg->oom_group)
1986 oom_group = memcg;
1987
1988 if (memcg == oom_domain)
1989 break;
1990 }
1991
1992 if (oom_group)
1993 css_get(&oom_group->css);
1994out:
1995 rcu_read_unlock();
1996
1997 return oom_group;
1998}
1999
2000void mem_cgroup_print_oom_group(struct mem_cgroup *memcg)
2001{
2002 pr_info("Tasks in ");
2003 pr_cont_cgroup_path(memcg->css.cgroup);
2004 pr_cont(" are going to be killed due to memory.oom.group set\n");
2005}
2006
2007/**
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08002008 * lock_page_memcg - lock a page and memcg binding
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002009 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002010 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002011 * This function protects unlocked LRU pages from being moved to
Johannes Weiner739f79f2017-08-18 15:15:48 -07002012 * another cgroup.
2013 *
Johannes Weiner1c824a62021-04-29 22:55:32 -07002014 * It ensures lifetime of the locked memcg. Caller is responsible
2015 * for the lifetime of the page.
Balbir Singhd69b0422009-06-17 16:26:34 -07002016 */
Johannes Weiner1c824a62021-04-29 22:55:32 -07002017void lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002018{
Johannes Weiner9da7b522020-06-03 16:01:51 -07002019 struct page *head = compound_head(page); /* rmap on tail pages */
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002020 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08002021 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002022
Johannes Weiner6de22612015-02-11 15:25:01 -08002023 /*
2024 * The RCU lock is held throughout the transaction. The fast
2025 * path can get away without acquiring the memcg->move_lock
2026 * because page moving starts with an RCU grace period.
Johannes Weiner739f79f2017-08-18 15:15:48 -07002027 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07002028 rcu_read_lock();
2029
2030 if (mem_cgroup_disabled())
Johannes Weiner1c824a62021-04-29 22:55:32 -07002031 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002032again:
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08002033 memcg = page_memcg(head);
Johannes Weiner29833312014-12-10 15:44:02 -08002034 if (unlikely(!memcg))
Johannes Weiner1c824a62021-04-29 22:55:32 -07002035 return;
Johannes Weinerd7365e72014-10-29 14:50:48 -07002036
Alex Shi20ad50d2020-12-15 12:33:51 -08002037#ifdef CONFIG_PROVE_LOCKING
2038 local_irq_save(flags);
2039 might_lock(&memcg->move_lock);
2040 local_irq_restore(flags);
2041#endif
2042
Qiang Huangbdcbb652014-06-04 16:08:21 -07002043 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner1c824a62021-04-29 22:55:32 -07002044 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002045
Johannes Weiner6de22612015-02-11 15:25:01 -08002046 spin_lock_irqsave(&memcg->move_lock, flags);
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08002047 if (memcg != page_memcg(head)) {
Johannes Weiner6de22612015-02-11 15:25:01 -08002048 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002049 goto again;
2050 }
Johannes Weiner6de22612015-02-11 15:25:01 -08002051
2052 /*
Johannes Weiner1c824a62021-04-29 22:55:32 -07002053 * When charge migration first begins, we can have multiple
2054 * critical sections holding the fast-path RCU lock and one
2055 * holding the slowpath move_lock. Track the task who has the
2056 * move_lock for unlock_page_memcg().
Johannes Weiner6de22612015-02-11 15:25:01 -08002057 */
2058 memcg->move_lock_task = current;
2059 memcg->move_lock_flags = flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002060}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002061EXPORT_SYMBOL(lock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002062
Johannes Weiner1c824a62021-04-29 22:55:32 -07002063static void __unlock_page_memcg(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002064{
Johannes Weiner6de22612015-02-11 15:25:01 -08002065 if (memcg && memcg->move_lock_task == current) {
2066 unsigned long flags = memcg->move_lock_flags;
2067
2068 memcg->move_lock_task = NULL;
2069 memcg->move_lock_flags = 0;
2070
2071 spin_unlock_irqrestore(&memcg->move_lock, flags);
2072 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002073
Johannes Weinerd7365e72014-10-29 14:50:48 -07002074 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002075}
Johannes Weiner739f79f2017-08-18 15:15:48 -07002076
2077/**
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08002078 * unlock_page_memcg - unlock a page and memcg binding
Johannes Weiner739f79f2017-08-18 15:15:48 -07002079 * @page: the page
2080 */
2081void unlock_page_memcg(struct page *page)
2082{
Johannes Weiner9da7b522020-06-03 16:01:51 -07002083 struct page *head = compound_head(page);
2084
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08002085 __unlock_page_memcg(page_memcg(head));
Johannes Weiner739f79f2017-08-18 15:15:48 -07002086}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002087EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002088
Waiman Long55927112021-06-28 19:37:30 -07002089struct obj_stock {
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002090#ifdef CONFIG_MEMCG_KMEM
2091 struct obj_cgroup *cached_objcg;
Waiman Long68ac5b32021-06-28 19:37:23 -07002092 struct pglist_data *cached_pgdat;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002093 unsigned int nr_bytes;
Waiman Long68ac5b32021-06-28 19:37:23 -07002094 int nr_slab_reclaimable_b;
2095 int nr_slab_unreclaimable_b;
Waiman Long55927112021-06-28 19:37:30 -07002096#else
2097 int dummy[0];
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002098#endif
Waiman Long55927112021-06-28 19:37:30 -07002099};
2100
2101struct memcg_stock_pcp {
2102 struct mem_cgroup *cached; /* this never be root cgroup */
2103 unsigned int nr_pages;
2104 struct obj_stock task_obj;
2105 struct obj_stock irq_obj;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002106
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002107 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002108 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002109#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002110};
2111static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002112static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002113
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002114#ifdef CONFIG_MEMCG_KMEM
Waiman Long55927112021-06-28 19:37:30 -07002115static void drain_obj_stock(struct obj_stock *stock);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002116static bool obj_stock_flush_required(struct memcg_stock_pcp *stock,
2117 struct mem_cgroup *root_memcg);
2118
2119#else
Waiman Long55927112021-06-28 19:37:30 -07002120static inline void drain_obj_stock(struct obj_stock *stock)
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002121{
2122}
2123static bool obj_stock_flush_required(struct memcg_stock_pcp *stock,
2124 struct mem_cgroup *root_memcg)
2125{
2126 return false;
2127}
2128#endif
2129
Waiman Long55927112021-06-28 19:37:30 -07002130/*
2131 * Most kmem_cache_alloc() calls are from user context. The irq disable/enable
2132 * sequence used in this case to access content from object stock is slow.
2133 * To optimize for user context access, there are now two object stocks for
2134 * task context and interrupt context access respectively.
2135 *
2136 * The task context object stock can be accessed by disabling preemption only
2137 * which is cheap in non-preempt kernel. The interrupt context object stock
2138 * can only be accessed after disabling interrupt. User context code can
2139 * access interrupt object stock, but not vice versa.
2140 */
2141static inline struct obj_stock *get_obj_stock(unsigned long *pflags)
2142{
2143 struct memcg_stock_pcp *stock;
2144
2145 if (likely(in_task())) {
2146 *pflags = 0UL;
2147 preempt_disable();
2148 stock = this_cpu_ptr(&memcg_stock);
2149 return &stock->task_obj;
2150 }
2151
2152 local_irq_save(*pflags);
2153 stock = this_cpu_ptr(&memcg_stock);
2154 return &stock->irq_obj;
2155}
2156
2157static inline void put_obj_stock(unsigned long flags)
2158{
2159 if (likely(in_task()))
2160 preempt_enable();
2161 else
2162 local_irq_restore(flags);
2163}
2164
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002165/**
2166 * consume_stock: Try to consume stocked charge on this cpu.
2167 * @memcg: memcg to consume from.
2168 * @nr_pages: how many pages to charge.
2169 *
2170 * The charges will only happen if @memcg matches the current cpu's memcg
2171 * stock, and at least @nr_pages are available in that stock. Failure to
2172 * service an allocation will refill the stock.
2173 *
2174 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002175 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002176static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002177{
2178 struct memcg_stock_pcp *stock;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002179 unsigned long flags;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002180 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002181
Johannes Weinera983b5e2018-01-31 16:16:45 -08002182 if (nr_pages > MEMCG_CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002183 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002184
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002185 local_irq_save(flags);
2186
2187 stock = this_cpu_ptr(&memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002188 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002189 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002190 ret = true;
2191 }
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002192
2193 local_irq_restore(flags);
2194
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002195 return ret;
2196}
2197
2198/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002199 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002200 */
2201static void drain_stock(struct memcg_stock_pcp *stock)
2202{
2203 struct mem_cgroup *old = stock->cached;
2204
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07002205 if (!old)
2206 return;
2207
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002208 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002209 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002210 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002211 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002212 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002213 }
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07002214
2215 css_put(&old->css);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002216 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002217}
2218
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002219static void drain_local_stock(struct work_struct *dummy)
2220{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002221 struct memcg_stock_pcp *stock;
2222 unsigned long flags;
2223
Michal Hocko72f01842017-10-03 16:14:53 -07002224 /*
Michal Hocko5c49cf92021-09-02 14:56:02 -07002225 * The only protection from cpu hotplug (memcg_hotplug_cpu_dead) vs.
2226 * drain_stock races is that we always operate on local CPU stock
2227 * here with IRQ disabled
Michal Hocko72f01842017-10-03 16:14:53 -07002228 */
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002229 local_irq_save(flags);
2230
2231 stock = this_cpu_ptr(&memcg_stock);
Waiman Long55927112021-06-28 19:37:30 -07002232 drain_obj_stock(&stock->irq_obj);
2233 if (in_task())
2234 drain_obj_stock(&stock->task_obj);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002235 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002236 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002237
2238 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002239}
2240
2241/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002242 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002243 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002244 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002245static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002246{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002247 struct memcg_stock_pcp *stock;
2248 unsigned long flags;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002249
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002250 local_irq_save(flags);
2251
2252 stock = this_cpu_ptr(&memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002253 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002254 drain_stock(stock);
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07002255 css_get(&memcg->css);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002256 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002257 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002258 stock->nr_pages += nr_pages;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002259
Johannes Weinera983b5e2018-01-31 16:16:45 -08002260 if (stock->nr_pages > MEMCG_CHARGE_BATCH)
Roman Gushchin475d0482017-09-08 16:13:09 -07002261 drain_stock(stock);
2262
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002263 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002264}
2265
2266/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002267 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002268 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002269 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002270static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002271{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002272 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002273
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002274 /* If someone's already draining, avoid adding running more workers. */
2275 if (!mutex_trylock(&percpu_charge_mutex))
2276 return;
Michal Hocko72f01842017-10-03 16:14:53 -07002277 /*
2278 * Notify other cpus that system-wide "drain" is running
2279 * We do not care about races with the cpu hotplug because cpu down
2280 * as well as workers from this path always operate on the local
2281 * per-cpu data. CPU up doesn't touch memcg_stock at all.
2282 */
Johannes Weiner5af12d02011-08-25 15:59:07 -07002283 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002284 for_each_online_cpu(cpu) {
2285 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002286 struct mem_cgroup *memcg;
Roman Gushchine1a366b2019-09-23 15:34:58 -07002287 bool flush = false;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002288
Roman Gushchine1a366b2019-09-23 15:34:58 -07002289 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002290 memcg = stock->cached;
Roman Gushchine1a366b2019-09-23 15:34:58 -07002291 if (memcg && stock->nr_pages &&
2292 mem_cgroup_is_descendant(memcg, root_memcg))
2293 flush = true;
Miaohe Lin27fb0952021-09-02 14:55:59 -07002294 else if (obj_stock_flush_required(stock, root_memcg))
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002295 flush = true;
Roman Gushchine1a366b2019-09-23 15:34:58 -07002296 rcu_read_unlock();
2297
2298 if (flush &&
2299 !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
Michal Hockod1a05b62011-07-26 16:08:27 -07002300 if (cpu == curcpu)
2301 drain_local_stock(&stock->work);
2302 else
2303 schedule_work_on(cpu, &stock->work);
2304 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002305 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002306 put_cpu();
Michal Hocko9f50fad2011-08-09 11:56:26 +02002307 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002308}
2309
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01002310static int memcg_hotplug_cpu_dead(unsigned int cpu)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002311{
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002312 struct memcg_stock_pcp *stock;
2313
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002314 stock = &per_cpu(memcg_stock, cpu);
2315 drain_stock(stock);
Johannes Weinera983b5e2018-01-31 16:16:45 -08002316
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01002317 return 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002318}
2319
Chris Downb3ff9292020-08-06 23:21:54 -07002320static unsigned long reclaim_high(struct mem_cgroup *memcg,
2321 unsigned int nr_pages,
2322 gfp_t gfp_mask)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002323{
Chris Downb3ff9292020-08-06 23:21:54 -07002324 unsigned long nr_reclaimed = 0;
2325
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002326 do {
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002327 unsigned long pflags;
2328
Jakub Kicinskid1663a92020-06-01 21:49:49 -07002329 if (page_counter_read(&memcg->memory) <=
2330 READ_ONCE(memcg->memory.high))
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002331 continue;
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002332
Johannes Weinere27be242018-04-10 16:29:45 -07002333 memcg_memory_event(memcg, MEMCG_HIGH);
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002334
2335 psi_memstall_enter(&pflags);
Chris Downb3ff9292020-08-06 23:21:54 -07002336 nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages,
2337 gfp_mask, true);
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002338 psi_memstall_leave(&pflags);
Chris Down4bf17302020-04-06 20:03:30 -07002339 } while ((memcg = parent_mem_cgroup(memcg)) &&
2340 !mem_cgroup_is_root(memcg));
Chris Downb3ff9292020-08-06 23:21:54 -07002341
2342 return nr_reclaimed;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002343}
2344
2345static void high_work_func(struct work_struct *work)
2346{
2347 struct mem_cgroup *memcg;
2348
2349 memcg = container_of(work, struct mem_cgroup, high_work);
Johannes Weinera983b5e2018-01-31 16:16:45 -08002350 reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002351}
2352
Tejun Heob23afb92015-11-05 18:46:11 -08002353/*
Chris Down0e4b01d2019-09-23 15:34:55 -07002354 * Clamp the maximum sleep time per allocation batch to 2 seconds. This is
2355 * enough to still cause a significant slowdown in most cases, while still
2356 * allowing diagnostics and tracing to proceed without becoming stuck.
2357 */
2358#define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ)
2359
2360/*
2361 * When calculating the delay, we use these either side of the exponentiation to
2362 * maintain precision and scale to a reasonable number of jiffies (see the table
2363 * below.
2364 *
2365 * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the
2366 * overage ratio to a delay.
Randy Dunlapac5ddd02020-08-11 18:33:02 -07002367 * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the
Chris Down0e4b01d2019-09-23 15:34:55 -07002368 * proposed penalty in order to reduce to a reasonable number of jiffies, and
2369 * to produce a reasonable delay curve.
2370 *
2371 * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a
2372 * reasonable delay curve compared to precision-adjusted overage, not
2373 * penalising heavily at first, but still making sure that growth beyond the
2374 * limit penalises misbehaviour cgroups by slowing them down exponentially. For
2375 * example, with a high of 100 megabytes:
2376 *
2377 * +-------+------------------------+
2378 * | usage | time to allocate in ms |
2379 * +-------+------------------------+
2380 * | 100M | 0 |
2381 * | 101M | 6 |
2382 * | 102M | 25 |
2383 * | 103M | 57 |
2384 * | 104M | 102 |
2385 * | 105M | 159 |
2386 * | 106M | 230 |
2387 * | 107M | 313 |
2388 * | 108M | 409 |
2389 * | 109M | 518 |
2390 * | 110M | 639 |
2391 * | 111M | 774 |
2392 * | 112M | 921 |
2393 * | 113M | 1081 |
2394 * | 114M | 1254 |
2395 * | 115M | 1439 |
2396 * | 116M | 1638 |
2397 * | 117M | 1849 |
2398 * | 118M | 2000 |
2399 * | 119M | 2000 |
2400 * | 120M | 2000 |
2401 * +-------+------------------------+
2402 */
2403 #define MEMCG_DELAY_PRECISION_SHIFT 20
2404 #define MEMCG_DELAY_SCALING_SHIFT 14
2405
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002406static u64 calculate_overage(unsigned long usage, unsigned long high)
2407{
2408 u64 overage;
2409
2410 if (usage <= high)
2411 return 0;
2412
2413 /*
2414 * Prevent division by 0 in overage calculation by acting as if
2415 * it was a threshold of 1 page
2416 */
2417 high = max(high, 1UL);
2418
2419 overage = usage - high;
2420 overage <<= MEMCG_DELAY_PRECISION_SHIFT;
2421 return div64_u64(overage, high);
2422}
2423
2424static u64 mem_find_max_overage(struct mem_cgroup *memcg)
2425{
2426 u64 overage, max_overage = 0;
2427
2428 do {
2429 overage = calculate_overage(page_counter_read(&memcg->memory),
Jakub Kicinskid1663a92020-06-01 21:49:49 -07002430 READ_ONCE(memcg->memory.high));
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002431 max_overage = max(overage, max_overage);
2432 } while ((memcg = parent_mem_cgroup(memcg)) &&
2433 !mem_cgroup_is_root(memcg));
2434
2435 return max_overage;
2436}
2437
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002438static u64 swap_find_max_overage(struct mem_cgroup *memcg)
2439{
2440 u64 overage, max_overage = 0;
2441
2442 do {
2443 overage = calculate_overage(page_counter_read(&memcg->swap),
2444 READ_ONCE(memcg->swap.high));
2445 if (overage)
2446 memcg_memory_event(memcg, MEMCG_SWAP_HIGH);
2447 max_overage = max(overage, max_overage);
2448 } while ((memcg = parent_mem_cgroup(memcg)) &&
2449 !mem_cgroup_is_root(memcg));
2450
2451 return max_overage;
2452}
2453
Chris Down0e4b01d2019-09-23 15:34:55 -07002454/*
Chris Downe26733e2020-03-21 18:22:23 -07002455 * Get the number of jiffies that we should penalise a mischievous cgroup which
2456 * is exceeding its memory.high by checking both it and its ancestors.
Tejun Heob23afb92015-11-05 18:46:11 -08002457 */
Chris Downe26733e2020-03-21 18:22:23 -07002458static unsigned long calculate_high_delay(struct mem_cgroup *memcg,
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002459 unsigned int nr_pages,
2460 u64 max_overage)
Tejun Heob23afb92015-11-05 18:46:11 -08002461{
Chris Downe26733e2020-03-21 18:22:23 -07002462 unsigned long penalty_jiffies;
Chris Downe26733e2020-03-21 18:22:23 -07002463
2464 if (!max_overage)
2465 return 0;
Chris Down0e4b01d2019-09-23 15:34:55 -07002466
2467 /*
Chris Down0e4b01d2019-09-23 15:34:55 -07002468 * We use overage compared to memory.high to calculate the number of
2469 * jiffies to sleep (penalty_jiffies). Ideally this value should be
2470 * fairly lenient on small overages, and increasingly harsh when the
2471 * memcg in question makes it clear that it has no intention of stopping
2472 * its crazy behaviour, so we exponentially increase the delay based on
2473 * overage amount.
2474 */
Chris Downe26733e2020-03-21 18:22:23 -07002475 penalty_jiffies = max_overage * max_overage * HZ;
2476 penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT;
2477 penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT;
Chris Down0e4b01d2019-09-23 15:34:55 -07002478
2479 /*
2480 * Factor in the task's own contribution to the overage, such that four
2481 * N-sized allocations are throttled approximately the same as one
2482 * 4N-sized allocation.
2483 *
2484 * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or
2485 * larger the current charge patch is than that.
2486 */
Jakub Kicinskiff144e62020-06-01 21:49:45 -07002487 return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH;
Chris Downe26733e2020-03-21 18:22:23 -07002488}
2489
2490/*
2491 * Scheduled by try_charge() to be executed from the userland return path
2492 * and reclaims memory over the high limit.
2493 */
2494void mem_cgroup_handle_over_high(void)
2495{
2496 unsigned long penalty_jiffies;
2497 unsigned long pflags;
Chris Downb3ff9292020-08-06 23:21:54 -07002498 unsigned long nr_reclaimed;
Chris Downe26733e2020-03-21 18:22:23 -07002499 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Chris Downd977aa92020-08-06 23:21:58 -07002500 int nr_retries = MAX_RECLAIM_RETRIES;
Chris Downe26733e2020-03-21 18:22:23 -07002501 struct mem_cgroup *memcg;
Chris Downb3ff9292020-08-06 23:21:54 -07002502 bool in_retry = false;
Chris Downe26733e2020-03-21 18:22:23 -07002503
2504 if (likely(!nr_pages))
2505 return;
2506
2507 memcg = get_mem_cgroup_from_mm(current->mm);
Chris Downe26733e2020-03-21 18:22:23 -07002508 current->memcg_nr_pages_over_high = 0;
2509
Chris Downb3ff9292020-08-06 23:21:54 -07002510retry_reclaim:
2511 /*
2512 * The allocating task should reclaim at least the batch size, but for
2513 * subsequent retries we only want to do what's necessary to prevent oom
2514 * or breaching resource isolation.
2515 *
2516 * This is distinct from memory.max or page allocator behaviour because
2517 * memory.high is currently batched, whereas memory.max and the page
2518 * allocator run every time an allocation is made.
2519 */
2520 nr_reclaimed = reclaim_high(memcg,
2521 in_retry ? SWAP_CLUSTER_MAX : nr_pages,
2522 GFP_KERNEL);
2523
Chris Downe26733e2020-03-21 18:22:23 -07002524 /*
2525 * memory.high is breached and reclaim is unable to keep up. Throttle
2526 * allocators proactively to slow down excessive growth.
2527 */
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002528 penalty_jiffies = calculate_high_delay(memcg, nr_pages,
2529 mem_find_max_overage(memcg));
Chris Down0e4b01d2019-09-23 15:34:55 -07002530
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002531 penalty_jiffies += calculate_high_delay(memcg, nr_pages,
2532 swap_find_max_overage(memcg));
2533
Chris Down0e4b01d2019-09-23 15:34:55 -07002534 /*
Jakub Kicinskiff144e62020-06-01 21:49:45 -07002535 * Clamp the max delay per usermode return so as to still keep the
2536 * application moving forwards and also permit diagnostics, albeit
2537 * extremely slowly.
2538 */
2539 penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES);
2540
2541 /*
Chris Down0e4b01d2019-09-23 15:34:55 -07002542 * Don't sleep if the amount of jiffies this memcg owes us is so low
2543 * that it's not even worth doing, in an attempt to be nice to those who
2544 * go only a small amount over their memory.high value and maybe haven't
2545 * been aggressively reclaimed enough yet.
2546 */
2547 if (penalty_jiffies <= HZ / 100)
2548 goto out;
2549
2550 /*
Chris Downb3ff9292020-08-06 23:21:54 -07002551 * If reclaim is making forward progress but we're still over
2552 * memory.high, we want to encourage that rather than doing allocator
2553 * throttling.
2554 */
2555 if (nr_reclaimed || nr_retries--) {
2556 in_retry = true;
2557 goto retry_reclaim;
2558 }
2559
2560 /*
Chris Down0e4b01d2019-09-23 15:34:55 -07002561 * If we exit early, we're guaranteed to die (since
2562 * schedule_timeout_killable sets TASK_KILLABLE). This means we don't
2563 * need to account for any ill-begotten jiffies to pay them off later.
2564 */
2565 psi_memstall_enter(&pflags);
2566 schedule_timeout_killable(penalty_jiffies);
2567 psi_memstall_leave(&pflags);
2568
2569out:
2570 css_put(&memcg->css);
Tejun Heob23afb92015-11-05 18:46:11 -08002571}
2572
Muchun Songc5c8b162021-06-28 19:37:44 -07002573static int try_charge_memcg(struct mem_cgroup *memcg, gfp_t gfp_mask,
2574 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002575{
Johannes Weinera983b5e2018-01-31 16:16:45 -08002576 unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages);
Chris Downd977aa92020-08-06 23:21:58 -07002577 int nr_retries = MAX_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002578 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002579 struct page_counter *counter;
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002580 enum oom_status oom_status;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002581 unsigned long nr_reclaimed;
Vasily Averina4ebf1b2021-11-05 13:38:09 -07002582 bool passed_oom = false;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002583 bool may_swap = true;
2584 bool drained = false;
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002585 unsigned long pflags;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002586
Johannes Weiner6539cc02014-08-06 16:05:42 -07002587retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07002588 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08002589 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002590
Johannes Weiner7941d212016-01-14 15:21:23 -08002591 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08002592 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
2593 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002594 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08002595 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002596 page_counter_uncharge(&memcg->memsw, batch);
2597 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002598 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002599 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002600 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002601 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002602
Johannes Weiner6539cc02014-08-06 16:05:42 -07002603 if (batch > nr_pages) {
2604 batch = nr_pages;
2605 goto retry;
2606 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002607
Johannes Weiner06b078f2014-08-06 16:05:44 -07002608 /*
Johannes Weiner869712f2019-11-05 21:17:13 -08002609 * Memcg doesn't have a dedicated reserve for atomic
2610 * allocations. But like the global atomic pool, we need to
2611 * put the burden of reclaim on regular allocation requests
2612 * and let these go through as privileged allocations.
2613 */
2614 if (gfp_mask & __GFP_ATOMIC)
2615 goto force;
2616
2617 /*
Johannes Weiner89a28482016-10-27 17:46:56 -07002618 * Prevent unbounded recursion when reclaim operations need to
2619 * allocate memory. This might exceed the limits temporarily,
2620 * but we prefer facilitating memory reclaim and getting back
2621 * under the limit over triggering OOM kills in these cases.
2622 */
2623 if (unlikely(current->flags & PF_MEMALLOC))
2624 goto force;
2625
Johannes Weiner06b078f2014-08-06 16:05:44 -07002626 if (unlikely(task_in_memcg_oom(current)))
2627 goto nomem;
2628
Mel Gormand0164ad2015-11-06 16:28:21 -08002629 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002630 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002631
Johannes Weinere27be242018-04-10 16:29:45 -07002632 memcg_memory_event(mem_over_limit, MEMCG_MAX);
Johannes Weiner241994ed2015-02-11 15:26:06 -08002633
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002634 psi_memstall_enter(&pflags);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002635 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
2636 gfp_mask, may_swap);
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002637 psi_memstall_leave(&pflags);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002638
Johannes Weiner61e02c72014-08-06 16:08:16 -07002639 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07002640 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07002641
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002642 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002643 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002644 drained = true;
2645 goto retry;
2646 }
2647
Johannes Weiner28c34c22014-08-06 16:05:47 -07002648 if (gfp_mask & __GFP_NORETRY)
2649 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002650 /*
2651 * Even though the limit is exceeded at this point, reclaim
2652 * may have been able to free some pages. Retry the charge
2653 * before killing the task.
2654 *
2655 * Only for regular pages, though: huge pages are rather
2656 * unlikely to succeed so close to the limit, and we fall back
2657 * to regular pages anyway in case of failure.
2658 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002659 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002660 goto retry;
2661 /*
2662 * At task move, charge accounts can be doubly counted. So, it's
2663 * better to wait until the end of task_move if something is going on.
2664 */
2665 if (mem_cgroup_wait_acct_move(mem_over_limit))
2666 goto retry;
2667
Johannes Weiner9b130612014-08-06 16:05:51 -07002668 if (nr_retries--)
2669 goto retry;
2670
Shakeel Butt38d38492019-07-11 20:55:48 -07002671 if (gfp_mask & __GFP_RETRY_MAYFAIL)
Michal Hocko29ef6802018-08-17 15:47:11 -07002672 goto nomem;
2673
Vasily Averina4ebf1b2021-11-05 13:38:09 -07002674 /* Avoid endless loop for tasks bypassed by the oom killer */
2675 if (passed_oom && task_is_dying())
2676 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002677
Michal Hocko29ef6802018-08-17 15:47:11 -07002678 /*
2679 * keep retrying as long as the memcg oom killer is able to make
2680 * a forward progress or bypass the charge if the oom killer
2681 * couldn't make any progress.
2682 */
2683 oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask,
Jerome Marchand3608de02015-11-05 18:47:29 -08002684 get_order(nr_pages * PAGE_SIZE));
Vasily Averina4ebf1b2021-11-05 13:38:09 -07002685 if (oom_status == OOM_SUCCESS) {
2686 passed_oom = true;
Chris Downd977aa92020-08-06 23:21:58 -07002687 nr_retries = MAX_RECLAIM_RETRIES;
Michal Hocko29ef6802018-08-17 15:47:11 -07002688 goto retry;
Michal Hocko29ef6802018-08-17 15:47:11 -07002689 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002690nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002691 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002692 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08002693force:
2694 /*
2695 * The allocation either can't fail or will lead to more memory
2696 * being freed very soon. Allow memory usage go over the limit
2697 * temporarily by force charging it.
2698 */
2699 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002700 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08002701 page_counter_charge(&memcg->memsw, nr_pages);
Tejun Heo10d53c72015-11-05 18:46:17 -08002702
2703 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002704
2705done_restock:
2706 if (batch > nr_pages)
2707 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002708
Johannes Weiner241994ed2015-02-11 15:26:06 -08002709 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002710 * If the hierarchy is above the normal consumption range, schedule
2711 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002712 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002713 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2714 * not recorded as it most likely matches current's and won't
2715 * change in the meantime. As high limit is checked again before
2716 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994ed2015-02-11 15:26:06 -08002717 */
2718 do {
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002719 bool mem_high, swap_high;
2720
2721 mem_high = page_counter_read(&memcg->memory) >
2722 READ_ONCE(memcg->memory.high);
2723 swap_high = page_counter_read(&memcg->swap) >
2724 READ_ONCE(memcg->swap.high);
2725
2726 /* Don't bother a random interrupted task */
2727 if (in_interrupt()) {
2728 if (mem_high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002729 schedule_work(&memcg->high_work);
2730 break;
2731 }
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002732 continue;
2733 }
2734
2735 if (mem_high || swap_high) {
2736 /*
2737 * The allocating tasks in this cgroup will need to do
2738 * reclaim or be throttled to prevent further growth
2739 * of the memory or swap footprints.
2740 *
2741 * Target some best-effort fairness between the tasks,
2742 * and distribute reclaim work and delay penalties
2743 * based on how much each task is actually allocating.
2744 */
Vladimir Davydov9516a182015-12-11 13:40:24 -08002745 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002746 set_notify_resume(current);
2747 break;
2748 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08002749 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002750
2751 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002752}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002753
Muchun Songc5c8b162021-06-28 19:37:44 -07002754static inline int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
2755 unsigned int nr_pages)
2756{
2757 if (mem_cgroup_is_root(memcg))
2758 return 0;
2759
2760 return try_charge_memcg(memcg, gfp_mask, nr_pages);
2761}
2762
Shakeel Butt58056f72021-11-05 13:37:44 -07002763static inline void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002764{
Johannes Weinerce00a962014-09-05 08:43:57 -04002765 if (mem_cgroup_is_root(memcg))
2766 return;
2767
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002768 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002769 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002770 page_counter_uncharge(&memcg->memsw, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002771}
2772
Johannes Weinerd9eb1ea2020-06-03 16:02:24 -07002773static void commit_charge(struct page *page, struct mem_cgroup *memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002774{
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08002775 VM_BUG_ON_PAGE(page_memcg(page), page);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002776 /*
Alex Shia5eb0112020-12-14 19:06:42 -08002777 * Any of the following ensures page's memcg stability:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002778 *
Johannes Weinera0b5b412020-06-03 16:02:27 -07002779 * - the page lock
2780 * - LRU isolation
2781 * - lock_page_memcg()
2782 * - exclusive reference
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002783 */
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08002784 page->memcg_data = (unsigned long)memcg;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002785}
2786
Muchun Songe74d2252021-04-29 22:56:42 -07002787static struct mem_cgroup *get_mem_cgroup_from_objcg(struct obj_cgroup *objcg)
2788{
2789 struct mem_cgroup *memcg;
2790
2791 rcu_read_lock();
2792retry:
2793 memcg = obj_cgroup_memcg(objcg);
2794 if (unlikely(!css_tryget(&memcg->css)))
2795 goto retry;
2796 rcu_read_unlock();
2797
2798 return memcg;
2799}
2800
Kirill Tkhai84c07d12018-08-17 15:47:25 -07002801#ifdef CONFIG_MEMCG_KMEM
Waiman Long41eb5df2021-06-28 19:37:34 -07002802/*
2803 * The allocated objcg pointers array is not accounted directly.
2804 * Moreover, it should not come from DMA buffer and is not readily
2805 * reclaimable. So those GFP bits should be masked off.
2806 */
2807#define OBJCGS_CLEAR_MASK (__GFP_DMA | __GFP_RECLAIMABLE | __GFP_ACCOUNT)
2808
Roman Gushchin10befea2020-08-06 23:21:27 -07002809int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
Roman Gushchin2e9bd482021-02-24 12:03:11 -08002810 gfp_t gfp, bool new_page)
Roman Gushchin10befea2020-08-06 23:21:27 -07002811{
2812 unsigned int objects = objs_per_slab_page(s, page);
Roman Gushchin2e9bd482021-02-24 12:03:11 -08002813 unsigned long memcg_data;
Roman Gushchin10befea2020-08-06 23:21:27 -07002814 void *vec;
2815
Waiman Long41eb5df2021-06-28 19:37:34 -07002816 gfp &= ~OBJCGS_CLEAR_MASK;
Roman Gushchin10befea2020-08-06 23:21:27 -07002817 vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp,
2818 page_to_nid(page));
2819 if (!vec)
2820 return -ENOMEM;
2821
Roman Gushchin2e9bd482021-02-24 12:03:11 -08002822 memcg_data = (unsigned long) vec | MEMCG_DATA_OBJCGS;
2823 if (new_page) {
2824 /*
2825 * If the slab page is brand new and nobody can yet access
2826 * it's memcg_data, no synchronization is required and
2827 * memcg_data can be simply assigned.
2828 */
2829 page->memcg_data = memcg_data;
2830 } else if (cmpxchg(&page->memcg_data, 0, memcg_data)) {
2831 /*
2832 * If the slab page is already in use, somebody can allocate
2833 * and assign obj_cgroups in parallel. In this case the existing
2834 * objcg vector should be reused.
2835 */
Roman Gushchin10befea2020-08-06 23:21:27 -07002836 kfree(vec);
Roman Gushchin2e9bd482021-02-24 12:03:11 -08002837 return 0;
2838 }
Roman Gushchin10befea2020-08-06 23:21:27 -07002839
Roman Gushchin2e9bd482021-02-24 12:03:11 -08002840 kmemleak_not_leak(vec);
Roman Gushchin10befea2020-08-06 23:21:27 -07002841 return 0;
2842}
2843
Roman Gushchin8380ce42020-03-28 19:17:25 -07002844/*
2845 * Returns a pointer to the memory cgroup to which the kernel object is charged.
2846 *
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08002847 * A passed kernel object can be a slab object or a generic kernel page, so
2848 * different mechanisms for getting the memory cgroup pointer should be used.
2849 * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller
2850 * can not know for sure how the kernel object is implemented.
2851 * mem_cgroup_from_obj() can be safely used in such cases.
2852 *
Roman Gushchin8380ce42020-03-28 19:17:25 -07002853 * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(),
2854 * cgroup_mutex, etc.
2855 */
2856struct mem_cgroup *mem_cgroup_from_obj(void *p)
2857{
2858 struct page *page;
2859
2860 if (mem_cgroup_disabled())
2861 return NULL;
2862
2863 page = virt_to_head_page(p);
2864
2865 /*
Roman Gushchin98556092020-08-06 23:21:10 -07002866 * Slab objects are accounted individually, not per-page.
2867 * Memcg membership data for each individual object is saved in
2868 * the page->obj_cgroups.
Roman Gushchin8380ce42020-03-28 19:17:25 -07002869 */
Roman Gushchin270c6a72020-12-01 13:58:28 -08002870 if (page_objcgs_check(page)) {
Roman Gushchin98556092020-08-06 23:21:10 -07002871 struct obj_cgroup *objcg;
2872 unsigned int off;
2873
2874 off = obj_to_index(page->slab_cache, page, p);
Roman Gushchin270c6a72020-12-01 13:58:28 -08002875 objcg = page_objcgs(page)[off];
Roman Gushchin10befea2020-08-06 23:21:27 -07002876 if (objcg)
2877 return obj_cgroup_memcg(objcg);
2878
2879 return NULL;
Roman Gushchin98556092020-08-06 23:21:10 -07002880 }
Roman Gushchin8380ce42020-03-28 19:17:25 -07002881
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08002882 /*
2883 * page_memcg_check() is used here, because page_has_obj_cgroups()
2884 * check above could fail because the object cgroups vector wasn't set
2885 * at that moment, but it can be set concurrently.
2886 * page_memcg_check(page) will guarantee that a proper memory
2887 * cgroup pointer or NULL will be returned.
2888 */
2889 return page_memcg_check(page);
Roman Gushchin8380ce42020-03-28 19:17:25 -07002890}
2891
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002892__always_inline struct obj_cgroup *get_obj_cgroup_from_current(void)
2893{
2894 struct obj_cgroup *objcg = NULL;
2895 struct mem_cgroup *memcg;
2896
Roman Gushchin279c3392020-10-17 16:13:44 -07002897 if (memcg_kmem_bypass())
2898 return NULL;
2899
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002900 rcu_read_lock();
Roman Gushchin37d59852020-10-17 16:13:50 -07002901 if (unlikely(active_memcg()))
2902 memcg = active_memcg();
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002903 else
2904 memcg = mem_cgroup_from_task(current);
2905
2906 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) {
2907 objcg = rcu_dereference(memcg->objcg);
2908 if (objcg && obj_cgroup_tryget(objcg))
2909 break;
Muchun Song2f7659a2020-12-14 19:06:31 -08002910 objcg = NULL;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002911 }
2912 rcu_read_unlock();
2913
2914 return objcg;
2915}
2916
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002917static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002918{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002919 int id, size;
2920 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002921
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002922 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002923 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2924 if (id < 0)
2925 return id;
2926
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002927 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002928 return id;
2929
2930 /*
2931 * There's no space for the new id in memcg_caches arrays,
2932 * so we have to grow them.
2933 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08002934 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002935
2936 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08002937 if (size < MEMCG_CACHES_MIN_SIZE)
2938 size = MEMCG_CACHES_MIN_SIZE;
2939 else if (size > MEMCG_CACHES_MAX_SIZE)
2940 size = MEMCG_CACHES_MAX_SIZE;
2941
Roman Gushchin98556092020-08-06 23:21:10 -07002942 err = memcg_update_all_list_lrus(size);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08002943 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08002944 memcg_nr_cache_ids = size;
2945
2946 up_write(&memcg_cache_ids_sem);
2947
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002948 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002949 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002950 return err;
2951 }
2952 return id;
2953}
2954
2955static void memcg_free_cache_id(int id)
2956{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002957 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08002958}
2959
Muchun Songf1286fa2021-04-29 22:56:55 -07002960/*
2961 * obj_cgroup_uncharge_pages: uncharge a number of kernel pages from a objcg
2962 * @objcg: object cgroup to uncharge
2963 * @nr_pages: number of pages to uncharge
2964 */
Muchun Songe74d2252021-04-29 22:56:42 -07002965static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg,
2966 unsigned int nr_pages)
2967{
2968 struct mem_cgroup *memcg;
2969
2970 memcg = get_mem_cgroup_from_objcg(objcg);
Muchun Songf1286fa2021-04-29 22:56:55 -07002971
2972 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2973 page_counter_uncharge(&memcg->kmem, nr_pages);
2974 refill_stock(memcg, nr_pages);
2975
Muchun Songe74d2252021-04-29 22:56:42 -07002976 css_put(&memcg->css);
2977}
2978
Muchun Songf1286fa2021-04-29 22:56:55 -07002979/*
2980 * obj_cgroup_charge_pages: charge a number of kernel pages to a objcg
2981 * @objcg: object cgroup to charge
Vladimir Davydov45264772016-07-26 15:24:21 -07002982 * @gfp: reclaim mode
Roman Gushchin92d0510c2020-04-01 21:06:49 -07002983 * @nr_pages: number of pages to charge
Vladimir Davydov45264772016-07-26 15:24:21 -07002984 *
2985 * Returns 0 on success, an error code on failure.
2986 */
Muchun Songf1286fa2021-04-29 22:56:55 -07002987static int obj_cgroup_charge_pages(struct obj_cgroup *objcg, gfp_t gfp,
2988 unsigned int nr_pages)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002989{
Muchun Songf1286fa2021-04-29 22:56:55 -07002990 struct mem_cgroup *memcg;
Johannes Weiner6071ca52015-11-05 18:50:26 -08002991 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08002992
Muchun Songf1286fa2021-04-29 22:56:55 -07002993 memcg = get_mem_cgroup_from_objcg(objcg);
2994
Muchun Songc5c8b162021-06-28 19:37:44 -07002995 ret = try_charge_memcg(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08002996 if (ret)
Muchun Songf1286fa2021-04-29 22:56:55 -07002997 goto out;
Johannes Weiner52c29b02016-01-20 15:02:35 -08002998
Shakeel Butt58056f72021-11-05 13:37:44 -07002999 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
3000 page_counter_charge(&memcg->kmem, nr_pages);
Muchun Songf1286fa2021-04-29 22:56:55 -07003001out:
3002 css_put(&memcg->css);
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003003
Muchun Songf1286fa2021-04-29 22:56:55 -07003004 return ret;
Roman Gushchin4b13f642020-04-01 21:06:56 -07003005}
3006
3007/**
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003008 * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup
Vladimir Davydov45264772016-07-26 15:24:21 -07003009 * @page: page to charge
3010 * @gfp: reclaim mode
3011 * @order: allocation order
3012 *
3013 * Returns 0 on success, an error code on failure.
3014 */
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003015int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003016{
Muchun Songb4e0b682021-04-29 22:56:52 -07003017 struct obj_cgroup *objcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07003018 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003019
Muchun Songb4e0b682021-04-29 22:56:52 -07003020 objcg = get_obj_cgroup_from_current();
3021 if (objcg) {
3022 ret = obj_cgroup_charge_pages(objcg, gfp, 1 << order);
Roman Gushchin4d96ba32019-07-11 20:56:31 -07003023 if (!ret) {
Muchun Songb4e0b682021-04-29 22:56:52 -07003024 page->memcg_data = (unsigned long)objcg |
Roman Gushchin18b2db32020-12-01 13:58:30 -08003025 MEMCG_DATA_KMEM;
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07003026 return 0;
Roman Gushchin4d96ba32019-07-11 20:56:31 -07003027 }
Muchun Songb4e0b682021-04-29 22:56:52 -07003028 obj_cgroup_put(objcg);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03003029 }
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08003030 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003031}
Roman Gushchin49a18ea2019-07-11 20:56:13 -07003032
3033/**
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003034 * __memcg_kmem_uncharge_page: uncharge a kmem page
Vladimir Davydov45264772016-07-26 15:24:21 -07003035 * @page: page to uncharge
3036 * @order: allocation order
3037 */
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003038void __memcg_kmem_uncharge_page(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003039{
Muchun Songb4e0b682021-04-29 22:56:52 -07003040 struct obj_cgroup *objcg;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003041 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003042
Muchun Songb4e0b682021-04-29 22:56:52 -07003043 if (!PageMemcgKmem(page))
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003044 return;
3045
Muchun Songb4e0b682021-04-29 22:56:52 -07003046 objcg = __page_objcg(page);
3047 obj_cgroup_uncharge_pages(objcg, nr_pages);
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08003048 page->memcg_data = 0;
Muchun Songb4e0b682021-04-29 22:56:52 -07003049 obj_cgroup_put(objcg);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08003050}
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003051
Waiman Long68ac5b32021-06-28 19:37:23 -07003052void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
3053 enum node_stat_item idx, int nr)
3054{
Waiman Long68ac5b32021-06-28 19:37:23 -07003055 unsigned long flags;
Waiman Long55927112021-06-28 19:37:30 -07003056 struct obj_stock *stock = get_obj_stock(&flags);
Waiman Long68ac5b32021-06-28 19:37:23 -07003057 int *bytes;
3058
Waiman Long68ac5b32021-06-28 19:37:23 -07003059 /*
3060 * Save vmstat data in stock and skip vmstat array update unless
3061 * accumulating over a page of vmstat data or when pgdat or idx
3062 * changes.
3063 */
3064 if (stock->cached_objcg != objcg) {
3065 drain_obj_stock(stock);
3066 obj_cgroup_get(objcg);
3067 stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes)
3068 ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0;
3069 stock->cached_objcg = objcg;
3070 stock->cached_pgdat = pgdat;
3071 } else if (stock->cached_pgdat != pgdat) {
3072 /* Flush the existing cached vmstat data */
Waiman Long7fa0dac2021-08-13 16:54:41 -07003073 struct pglist_data *oldpg = stock->cached_pgdat;
3074
Waiman Long68ac5b32021-06-28 19:37:23 -07003075 if (stock->nr_slab_reclaimable_b) {
Waiman Long7fa0dac2021-08-13 16:54:41 -07003076 mod_objcg_mlstate(objcg, oldpg, NR_SLAB_RECLAIMABLE_B,
Waiman Long68ac5b32021-06-28 19:37:23 -07003077 stock->nr_slab_reclaimable_b);
3078 stock->nr_slab_reclaimable_b = 0;
3079 }
3080 if (stock->nr_slab_unreclaimable_b) {
Waiman Long7fa0dac2021-08-13 16:54:41 -07003081 mod_objcg_mlstate(objcg, oldpg, NR_SLAB_UNRECLAIMABLE_B,
Waiman Long68ac5b32021-06-28 19:37:23 -07003082 stock->nr_slab_unreclaimable_b);
3083 stock->nr_slab_unreclaimable_b = 0;
3084 }
3085 stock->cached_pgdat = pgdat;
3086 }
3087
3088 bytes = (idx == NR_SLAB_RECLAIMABLE_B) ? &stock->nr_slab_reclaimable_b
3089 : &stock->nr_slab_unreclaimable_b;
3090 /*
3091 * Even for large object >= PAGE_SIZE, the vmstat data will still be
3092 * cached locally at least once before pushing it out.
3093 */
3094 if (!*bytes) {
3095 *bytes = nr;
3096 nr = 0;
3097 } else {
3098 *bytes += nr;
3099 if (abs(*bytes) > PAGE_SIZE) {
3100 nr = *bytes;
3101 *bytes = 0;
3102 } else {
3103 nr = 0;
3104 }
3105 }
3106 if (nr)
3107 mod_objcg_mlstate(objcg, pgdat, idx, nr);
3108
Waiman Long55927112021-06-28 19:37:30 -07003109 put_obj_stock(flags);
Waiman Long68ac5b32021-06-28 19:37:23 -07003110}
3111
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003112static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes)
3113{
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003114 unsigned long flags;
Waiman Long55927112021-06-28 19:37:30 -07003115 struct obj_stock *stock = get_obj_stock(&flags);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003116 bool ret = false;
3117
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003118 if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) {
3119 stock->nr_bytes -= nr_bytes;
3120 ret = true;
3121 }
3122
Waiman Long55927112021-06-28 19:37:30 -07003123 put_obj_stock(flags);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003124
3125 return ret;
3126}
3127
Waiman Long55927112021-06-28 19:37:30 -07003128static void drain_obj_stock(struct obj_stock *stock)
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003129{
3130 struct obj_cgroup *old = stock->cached_objcg;
3131
3132 if (!old)
3133 return;
3134
3135 if (stock->nr_bytes) {
3136 unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT;
3137 unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1);
3138
Muchun Songe74d2252021-04-29 22:56:42 -07003139 if (nr_pages)
3140 obj_cgroup_uncharge_pages(old, nr_pages);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003141
3142 /*
3143 * The leftover is flushed to the centralized per-memcg value.
3144 * On the next attempt to refill obj stock it will be moved
3145 * to a per-cpu stock (probably, on an other CPU), see
3146 * refill_obj_stock().
3147 *
3148 * How often it's flushed is a trade-off between the memory
3149 * limit enforcement accuracy and potential CPU contention,
3150 * so it might be changed in the future.
3151 */
3152 atomic_add(nr_bytes, &old->nr_charged_bytes);
3153 stock->nr_bytes = 0;
3154 }
3155
Waiman Long68ac5b32021-06-28 19:37:23 -07003156 /*
3157 * Flush the vmstat data in current stock
3158 */
3159 if (stock->nr_slab_reclaimable_b || stock->nr_slab_unreclaimable_b) {
3160 if (stock->nr_slab_reclaimable_b) {
3161 mod_objcg_mlstate(old, stock->cached_pgdat,
3162 NR_SLAB_RECLAIMABLE_B,
3163 stock->nr_slab_reclaimable_b);
3164 stock->nr_slab_reclaimable_b = 0;
3165 }
3166 if (stock->nr_slab_unreclaimable_b) {
3167 mod_objcg_mlstate(old, stock->cached_pgdat,
3168 NR_SLAB_UNRECLAIMABLE_B,
3169 stock->nr_slab_unreclaimable_b);
3170 stock->nr_slab_unreclaimable_b = 0;
3171 }
3172 stock->cached_pgdat = NULL;
3173 }
3174
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003175 obj_cgroup_put(old);
3176 stock->cached_objcg = NULL;
3177}
3178
3179static bool obj_stock_flush_required(struct memcg_stock_pcp *stock,
3180 struct mem_cgroup *root_memcg)
3181{
3182 struct mem_cgroup *memcg;
3183
Waiman Long55927112021-06-28 19:37:30 -07003184 if (in_task() && stock->task_obj.cached_objcg) {
3185 memcg = obj_cgroup_memcg(stock->task_obj.cached_objcg);
3186 if (memcg && mem_cgroup_is_descendant(memcg, root_memcg))
3187 return true;
3188 }
3189 if (stock->irq_obj.cached_objcg) {
3190 memcg = obj_cgroup_memcg(stock->irq_obj.cached_objcg);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003191 if (memcg && mem_cgroup_is_descendant(memcg, root_memcg))
3192 return true;
3193 }
3194
3195 return false;
3196}
3197
Waiman Long5387c902021-06-28 19:37:27 -07003198static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes,
3199 bool allow_uncharge)
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003200{
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003201 unsigned long flags;
Waiman Long55927112021-06-28 19:37:30 -07003202 struct obj_stock *stock = get_obj_stock(&flags);
Waiman Long5387c902021-06-28 19:37:27 -07003203 unsigned int nr_pages = 0;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003204
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003205 if (stock->cached_objcg != objcg) { /* reset if necessary */
3206 drain_obj_stock(stock);
3207 obj_cgroup_get(objcg);
3208 stock->cached_objcg = objcg;
Waiman Long5387c902021-06-28 19:37:27 -07003209 stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes)
3210 ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0;
3211 allow_uncharge = true; /* Allow uncharge when objcg changes */
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003212 }
3213 stock->nr_bytes += nr_bytes;
3214
Waiman Long5387c902021-06-28 19:37:27 -07003215 if (allow_uncharge && (stock->nr_bytes > PAGE_SIZE)) {
3216 nr_pages = stock->nr_bytes >> PAGE_SHIFT;
3217 stock->nr_bytes &= (PAGE_SIZE - 1);
3218 }
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003219
Waiman Long55927112021-06-28 19:37:30 -07003220 put_obj_stock(flags);
Waiman Long5387c902021-06-28 19:37:27 -07003221
3222 if (nr_pages)
3223 obj_cgroup_uncharge_pages(objcg, nr_pages);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003224}
3225
3226int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size)
3227{
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003228 unsigned int nr_pages, nr_bytes;
3229 int ret;
3230
3231 if (consume_obj_stock(objcg, size))
3232 return 0;
3233
3234 /*
Waiman Long5387c902021-06-28 19:37:27 -07003235 * In theory, objcg->nr_charged_bytes can have enough
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003236 * pre-charged bytes to satisfy the allocation. However,
Waiman Long5387c902021-06-28 19:37:27 -07003237 * flushing objcg->nr_charged_bytes requires two atomic
3238 * operations, and objcg->nr_charged_bytes can't be big.
3239 * The shared objcg->nr_charged_bytes can also become a
3240 * performance bottleneck if all tasks of the same memcg are
3241 * trying to update it. So it's better to ignore it and try
3242 * grab some new pages. The stock's nr_bytes will be flushed to
3243 * objcg->nr_charged_bytes later on when objcg changes.
3244 *
3245 * The stock's nr_bytes may contain enough pre-charged bytes
3246 * to allow one less page from being charged, but we can't rely
3247 * on the pre-charged bytes not being changed outside of
3248 * consume_obj_stock() or refill_obj_stock(). So ignore those
3249 * pre-charged bytes as well when charging pages. To avoid a
3250 * page uncharge right after a page charge, we set the
3251 * allow_uncharge flag to false when calling refill_obj_stock()
3252 * to temporarily allow the pre-charged bytes to exceed the page
3253 * size limit. The maximum reachable value of the pre-charged
3254 * bytes is (sizeof(object) + PAGE_SIZE - 2) if there is no data
3255 * race.
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003256 */
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003257 nr_pages = size >> PAGE_SHIFT;
3258 nr_bytes = size & (PAGE_SIZE - 1);
3259
3260 if (nr_bytes)
3261 nr_pages += 1;
3262
Muchun Songe74d2252021-04-29 22:56:42 -07003263 ret = obj_cgroup_charge_pages(objcg, gfp, nr_pages);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003264 if (!ret && nr_bytes)
Waiman Long5387c902021-06-28 19:37:27 -07003265 refill_obj_stock(objcg, PAGE_SIZE - nr_bytes, false);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003266
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003267 return ret;
3268}
3269
3270void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size)
3271{
Waiman Long5387c902021-06-28 19:37:27 -07003272 refill_obj_stock(objcg, size, true);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003273}
3274
Kirill Tkhai84c07d12018-08-17 15:47:25 -07003275#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003276
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003277/*
Zhou Guanghuibe6c8982021-03-12 21:08:30 -08003278 * Because page_memcg(head) is not set on tails, set it now.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003279 */
Zhou Guanghuibe6c8982021-03-12 21:08:30 -08003280void split_page_memcg(struct page *head, unsigned int nr)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003281{
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08003282 struct mem_cgroup *memcg = page_memcg(head);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003283 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003284
Zhou Guanghuibe6c8982021-03-12 21:08:30 -08003285 if (mem_cgroup_disabled() || !memcg)
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08003286 return;
David Rientjesb070e652013-05-07 16:18:09 -07003287
Zhou Guanghuibe6c8982021-03-12 21:08:30 -08003288 for (i = 1; i < nr; i++)
3289 head[i].memcg_data = head->memcg_data;
Muchun Songb4e0b682021-04-29 22:56:52 -07003290
3291 if (PageMemcgKmem(head))
3292 obj_cgroup_get_many(__page_objcg(head), nr - 1);
3293 else
3294 css_get_many(&memcg->css, nr - 1);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003295}
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003296
Andrew Mortonc255a452012-07-31 16:43:02 -07003297#ifdef CONFIG_MEMCG_SWAP
Daisuke Nishimura02491442010-03-10 15:22:17 -08003298/**
3299 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3300 * @entry: swap entry to be moved
3301 * @from: mem_cgroup which the entry is moved from
3302 * @to: mem_cgroup which the entry is moved to
3303 *
3304 * It succeeds only when the swap_cgroup's record for this entry is the same
3305 * as the mem_cgroup's id of @from.
3306 *
3307 * Returns 0 on success, -EINVAL on failure.
3308 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003309 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08003310 * both res and memsw, and called css_get().
3311 */
3312static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003313 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003314{
3315 unsigned short old_id, new_id;
3316
Li Zefan34c00c32013-09-23 16:56:01 +08003317 old_id = mem_cgroup_id(from);
3318 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003319
3320 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Johannes Weinerc9019e92018-01-31 16:16:37 -08003321 mod_memcg_state(from, MEMCG_SWAP, -1);
3322 mod_memcg_state(to, MEMCG_SWAP, 1);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003323 return 0;
3324 }
3325 return -EINVAL;
3326}
3327#else
3328static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003329 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003330{
3331 return -EINVAL;
3332}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003333#endif
3334
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003335static DEFINE_MUTEX(memcg_max_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003336
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003337static int mem_cgroup_resize_max(struct mem_cgroup *memcg,
3338 unsigned long max, bool memsw)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003339{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003340 bool enlarge = false;
Shakeel Buttbb4a7ea2018-06-07 17:07:27 -07003341 bool drained = false;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003342 int ret;
Yu Zhaoc054a782018-01-31 16:20:02 -08003343 bool limits_invariant;
3344 struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003345
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003346 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003347 if (signal_pending(current)) {
3348 ret = -EINTR;
3349 break;
3350 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003351
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003352 mutex_lock(&memcg_max_mutex);
Yu Zhaoc054a782018-01-31 16:20:02 -08003353 /*
3354 * Make sure that the new limit (memsw or memory limit) doesn't
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003355 * break our basic invariant rule memory.max <= memsw.max.
Yu Zhaoc054a782018-01-31 16:20:02 -08003356 */
Chris Down15b42562020-04-01 21:07:20 -07003357 limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) :
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003358 max <= memcg->memsw.max;
Yu Zhaoc054a782018-01-31 16:20:02 -08003359 if (!limits_invariant) {
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003360 mutex_unlock(&memcg_max_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003361 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003362 break;
3363 }
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003364 if (max > counter->max)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003365 enlarge = true;
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003366 ret = page_counter_set_max(counter, max);
3367 mutex_unlock(&memcg_max_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003368
3369 if (!ret)
3370 break;
3371
Shakeel Buttbb4a7ea2018-06-07 17:07:27 -07003372 if (!drained) {
3373 drain_all_stock(memcg);
3374 drained = true;
3375 continue;
3376 }
3377
Andrey Ryabinin1ab5c052018-01-31 16:20:37 -08003378 if (!try_to_free_mem_cgroup_pages(memcg, 1,
3379 GFP_KERNEL, !memsw)) {
3380 ret = -EBUSY;
3381 break;
3382 }
3383 } while (true);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003384
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003385 if (!ret && enlarge)
3386 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003387
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003388 return ret;
3389}
3390
Mel Gormanef8f2322016-07-28 15:46:05 -07003391unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
Andrew Morton0608f432013-09-24 15:27:41 -07003392 gfp_t gfp_mask,
3393 unsigned long *total_scanned)
3394{
3395 unsigned long nr_reclaimed = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07003396 struct mem_cgroup_per_node *mz, *next_mz = NULL;
Andrew Morton0608f432013-09-24 15:27:41 -07003397 unsigned long reclaimed;
3398 int loop = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07003399 struct mem_cgroup_tree_per_node *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003400 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07003401 unsigned long nr_scanned;
3402
3403 if (order > 0)
3404 return 0;
3405
Mel Gormanef8f2322016-07-28 15:46:05 -07003406 mctz = soft_limit_tree_node(pgdat->node_id);
Michal Hockod6507ff2016-08-02 14:02:37 -07003407
3408 /*
3409 * Do not even bother to check the largest node if the root
3410 * is empty. Do it lockless to prevent lock bouncing. Races
3411 * are acceptable as soft limit is best effort anyway.
3412 */
Laurent Dufourbfc72282017-03-09 16:17:06 -08003413 if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root))
Michal Hockod6507ff2016-08-02 14:02:37 -07003414 return 0;
3415
Andrew Morton0608f432013-09-24 15:27:41 -07003416 /*
3417 * This loop can run a while, specially if mem_cgroup's continuously
3418 * keep exceeding their soft limit and putting the system under
3419 * pressure
3420 */
3421 do {
3422 if (next_mz)
3423 mz = next_mz;
3424 else
3425 mz = mem_cgroup_largest_soft_limit_node(mctz);
3426 if (!mz)
3427 break;
3428
3429 nr_scanned = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07003430 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07003431 gfp_mask, &nr_scanned);
3432 nr_reclaimed += reclaimed;
3433 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003434 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08003435 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07003436
3437 /*
3438 * If we failed to reclaim anything from this memory cgroup
3439 * it is time to move on to the next cgroup
3440 */
3441 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08003442 if (!reclaimed)
3443 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
3444
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003445 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07003446 /*
3447 * One school of thought says that we should not add
3448 * back the node to the tree if reclaim returns 0.
3449 * But our reclaim could return 0, simply because due
3450 * to priority we are exposing a smaller subset of
3451 * memory to reclaim from. Consider this as a longer
3452 * term TODO.
3453 */
3454 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07003455 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003456 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07003457 css_put(&mz->memcg->css);
3458 loop++;
3459 /*
3460 * Could not reclaim anything and there are no more
3461 * mem cgroups to try or we seem to be looping without
3462 * reclaiming anything.
3463 */
3464 if (!nr_reclaimed &&
3465 (next_mz == NULL ||
3466 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3467 break;
3468 } while (!nr_reclaimed);
3469 if (next_mz)
3470 css_put(&next_mz->memcg->css);
3471 return nr_reclaimed;
3472}
3473
Tejun Heoea280e72014-05-16 13:22:48 -04003474/*
Greg Thelen51038172016-05-20 16:58:18 -07003475 * Reclaims as many pages from the given memcg as possible.
Michal Hockoc26251f2012-10-26 13:37:28 +02003476 *
3477 * Caller is responsible for holding css reference for memcg.
3478 */
3479static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
3480{
Chris Downd977aa92020-08-06 23:21:58 -07003481 int nr_retries = MAX_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02003482
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003483 /* we call try-to-free pages for make this cgroup empty */
3484 lru_add_drain_all();
Junaid Shahidd12c60f2018-06-07 17:07:31 -07003485
3486 drain_all_stock(memcg);
3487
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003488 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003489 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003490 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003491
Michal Hockoc26251f2012-10-26 13:37:28 +02003492 if (signal_pending(current))
3493 return -EINTR;
3494
Johannes Weinerb70a2a22014-10-09 15:28:56 -07003495 progress = try_to_free_mem_cgroup_pages(memcg, 1,
3496 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003497 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003498 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003499 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003500 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003501 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003502
3503 }
Michal Hockoab5196c2012-10-26 13:37:32 +02003504
3505 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003506}
3507
Tejun Heo6770c642014-05-13 12:16:21 -04003508static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
3509 char *buf, size_t nbytes,
3510 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003511{
Tejun Heo6770c642014-05-13 12:16:21 -04003512 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02003513
Michal Hockod8423012012-10-26 13:37:29 +02003514 if (mem_cgroup_is_root(memcg))
3515 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04003516 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003517}
3518
Tejun Heo182446d2013-08-08 20:11:24 -04003519static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
3520 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08003521{
Roman Gushchinbef86202020-12-14 19:06:49 -08003522 return 1;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003523}
3524
Tejun Heo182446d2013-08-08 20:11:24 -04003525static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
3526 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08003527{
Roman Gushchinbef86202020-12-14 19:06:49 -08003528 if (val == 1)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003529 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07003530
Roman Gushchinbef86202020-12-14 19:06:49 -08003531 pr_warn_once("Non-hierarchical mode is deprecated. "
3532 "Please report your usecase to linux-mm@kvack.org if you "
3533 "depend on this functionality.\n");
Glauber Costa567fb432012-07-31 16:43:07 -07003534
Roman Gushchinbef86202020-12-14 19:06:49 -08003535 return -EINVAL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003536}
3537
Andrew Morton6f646152015-11-06 16:28:58 -08003538static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04003539{
Johannes Weiner42a30032019-05-14 15:47:12 -07003540 unsigned long val;
Johannes Weinerce00a962014-09-05 08:43:57 -04003541
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003542 if (mem_cgroup_is_root(memcg)) {
Shakeel Buttfd25a9e2021-11-05 13:37:34 -07003543 mem_cgroup_flush_stats();
Johannes Weiner0d1c2072020-06-03 16:01:54 -07003544 val = memcg_page_state(memcg, NR_FILE_PAGES) +
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07003545 memcg_page_state(memcg, NR_ANON_MAPPED);
Johannes Weiner42a30032019-05-14 15:47:12 -07003546 if (swap)
3547 val += memcg_page_state(memcg, MEMCG_SWAP);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003548 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04003549 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003550 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04003551 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003552 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04003553 }
Michal Hockoc12176d2015-11-05 18:50:29 -08003554 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04003555}
3556
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003557enum {
3558 RES_USAGE,
3559 RES_LIMIT,
3560 RES_MAX_USAGE,
3561 RES_FAILCNT,
3562 RES_SOFT_LIMIT,
3563};
Johannes Weinerce00a962014-09-05 08:43:57 -04003564
Tejun Heo791badb2013-12-05 12:28:02 -05003565static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07003566 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003567{
Tejun Heo182446d2013-08-08 20:11:24 -04003568 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003569 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07003570
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003571 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003572 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003573 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08003574 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003575 case _MEMSWAP:
3576 counter = &memcg->memsw;
3577 break;
3578 case _KMEM:
3579 counter = &memcg->kmem;
3580 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003581 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003582 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003583 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003584 default:
3585 BUG();
3586 }
3587
3588 switch (MEMFILE_ATTR(cft->private)) {
3589 case RES_USAGE:
3590 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08003591 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003592 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08003593 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003594 return (u64)page_counter_read(counter) * PAGE_SIZE;
3595 case RES_LIMIT:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003596 return (u64)counter->max * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003597 case RES_MAX_USAGE:
3598 return (u64)counter->watermark * PAGE_SIZE;
3599 case RES_FAILCNT:
3600 return counter->failcnt;
3601 case RES_SOFT_LIMIT:
3602 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003603 default:
3604 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003605 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003606}
Glauber Costa510fc4e2012-12-18 14:21:47 -08003607
Kirill Tkhai84c07d12018-08-17 15:47:25 -07003608#ifdef CONFIG_MEMCG_KMEM
Johannes Weiner567e9ab2016-01-20 15:02:24 -08003609static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08003610{
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003611 struct obj_cgroup *objcg;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003612 int memcg_id;
3613
Vladimir Davydovb313aee2016-03-17 14:18:27 -07003614 if (cgroup_memory_nokmem)
3615 return 0;
3616
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003617 BUG_ON(memcg->kmemcg_id >= 0);
Vladimir Davydovd6441632014-01-23 15:53:09 -08003618
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07003619 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003620 if (memcg_id < 0)
3621 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003622
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003623 objcg = obj_cgroup_alloc();
3624 if (!objcg) {
3625 memcg_free_cache_id(memcg_id);
3626 return -ENOMEM;
3627 }
3628 objcg->memcg = memcg;
3629 rcu_assign_pointer(memcg->objcg, objcg);
3630
Roman Gushchind648bcc2020-08-06 23:20:28 -07003631 static_branch_enable(&memcg_kmem_enabled_key);
3632
Vladimir Davydov900a38f2014-12-12 16:55:10 -08003633 memcg->kmemcg_id = memcg_id;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003634
3635 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003636}
3637
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003638static void memcg_offline_kmem(struct mem_cgroup *memcg)
3639{
Muchun Song64268862021-11-05 13:37:53 -07003640 struct mem_cgroup *parent;
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003641 int kmemcg_id;
3642
Muchun Songe80216d2021-11-05 13:37:56 -07003643 if (memcg->kmemcg_id == -1)
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003644 return;
Roman Gushchin98556092020-08-06 23:21:10 -07003645
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003646 parent = parent_mem_cgroup(memcg);
3647 if (!parent)
3648 parent = root_mem_cgroup;
3649
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003650 memcg_reparent_objcgs(memcg, parent);
Roman Gushchinfb2f2b02019-07-11 20:56:34 -07003651
3652 kmemcg_id = memcg->kmemcg_id;
3653 BUG_ON(kmemcg_id < 0);
3654
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003655 /*
Muchun Song64268862021-11-05 13:37:53 -07003656 * After we have finished memcg_reparent_objcgs(), all list_lrus
3657 * corresponding to this cgroup are guaranteed to remain empty.
3658 * The ordering is imposed by list_lru_node->lock taken by
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003659 * memcg_drain_all_list_lrus().
3660 */
Kirill Tkhai9bec5c32018-08-17 15:47:58 -07003661 memcg_drain_all_list_lrus(kmemcg_id, parent);
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003662
3663 memcg_free_cache_id(kmemcg_id);
Muchun Songe80216d2021-11-05 13:37:56 -07003664 memcg->kmemcg_id = -1;
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003665}
Vladimir Davydovd6441632014-01-23 15:53:09 -08003666#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003667static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08003668{
3669 return 0;
3670}
3671static void memcg_offline_kmem(struct mem_cgroup *memcg)
3672{
3673}
Kirill Tkhai84c07d12018-08-17 15:47:25 -07003674#endif /* CONFIG_MEMCG_KMEM */
Johannes Weiner127424c2016-01-20 15:02:32 -08003675
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003676static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003677{
3678 int ret;
3679
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003680 mutex_lock(&memcg_max_mutex);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003681
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003682 ret = page_counter_set_max(&memcg->tcpmem, max);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003683 if (ret)
3684 goto out;
3685
Johannes Weiner0db15292016-01-20 15:02:50 -08003686 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003687 /*
3688 * The active flag needs to be written after the static_key
3689 * update. This is what guarantees that the socket activation
Johannes Weiner2d758072016-10-07 17:00:58 -07003690 * function is the last one to run. See mem_cgroup_sk_alloc()
3691 * for details, and note that we don't mark any socket as
3692 * belonging to this memcg until that flag is up.
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003693 *
3694 * We need to do this, because static_keys will span multiple
3695 * sites, but we can't control their order. If we mark a socket
3696 * as accounted, but the accounting functions are not patched in
3697 * yet, we'll lose accounting.
3698 *
Johannes Weiner2d758072016-10-07 17:00:58 -07003699 * We never race with the readers in mem_cgroup_sk_alloc(),
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003700 * because when this value change, the code to process it is not
3701 * patched in yet.
3702 */
3703 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08003704 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003705 }
3706out:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003707 mutex_unlock(&memcg_max_mutex);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003708 return ret;
3709}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003710
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003711/*
3712 * The user of this function is...
3713 * RES_LIMIT.
3714 */
Tejun Heo451af502014-05-13 12:16:21 -04003715static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
3716 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003717{
Tejun Heo451af502014-05-13 12:16:21 -04003718 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003719 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003720 int ret;
3721
Tejun Heo451af502014-05-13 12:16:21 -04003722 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08003723 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003724 if (ret)
3725 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07003726
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003727 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003728 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003729 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3730 ret = -EINVAL;
3731 break;
3732 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003733 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3734 case _MEM:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003735 ret = mem_cgroup_resize_max(memcg, nr_pages, false);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003736 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003737 case _MEMSWAP:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003738 ret = mem_cgroup_resize_max(memcg, nr_pages, true);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003739 break;
3740 case _KMEM:
Shakeel Butt58056f72021-11-05 13:37:44 -07003741 /* kmem.limit_in_bytes is deprecated. */
3742 ret = -EOPNOTSUPP;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003743 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003744 case _TCP:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003745 ret = memcg_update_tcp_max(memcg, nr_pages);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003746 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003747 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003748 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003749 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003750 memcg->soft_limit = nr_pages;
3751 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003752 break;
3753 }
Tejun Heo451af502014-05-13 12:16:21 -04003754 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003755}
3756
Tejun Heo6770c642014-05-13 12:16:21 -04003757static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3758 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003759{
Tejun Heo6770c642014-05-13 12:16:21 -04003760 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003761 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003762
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003763 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3764 case _MEM:
3765 counter = &memcg->memory;
3766 break;
3767 case _MEMSWAP:
3768 counter = &memcg->memsw;
3769 break;
3770 case _KMEM:
3771 counter = &memcg->kmem;
3772 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003773 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003774 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003775 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003776 default:
3777 BUG();
3778 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003779
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003780 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003781 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003782 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003783 break;
3784 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003785 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003786 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003787 default:
3788 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003789 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003790
Tejun Heo6770c642014-05-13 12:16:21 -04003791 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003792}
3793
Tejun Heo182446d2013-08-08 20:11:24 -04003794static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003795 struct cftype *cft)
3796{
Tejun Heo182446d2013-08-08 20:11:24 -04003797 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003798}
3799
Daisuke Nishimura02491442010-03-10 15:22:17 -08003800#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003801static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003802 struct cftype *cft, u64 val)
3803{
Tejun Heo182446d2013-08-08 20:11:24 -04003804 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003805
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003806 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003807 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003808
Glauber Costaee5e8472013-02-22 16:34:50 -08003809 /*
3810 * No kind of locking is needed in here, because ->can_attach() will
3811 * check this value once in the beginning of the process, and then carry
3812 * on with stale data. This means that changes to this value will only
3813 * affect task migrations starting after the change.
3814 */
3815 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003816 return 0;
3817}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003818#else
Tejun Heo182446d2013-08-08 20:11:24 -04003819static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003820 struct cftype *cft, u64 val)
3821{
3822 return -ENOSYS;
3823}
3824#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003825
Ying Han406eb0c2011-05-26 16:25:37 -07003826#ifdef CONFIG_NUMA
Johannes Weiner113b7df2019-05-13 17:18:11 -07003827
3828#define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
3829#define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
3830#define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
3831
3832static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003833 int nid, unsigned int lru_mask, bool tree)
Johannes Weiner113b7df2019-05-13 17:18:11 -07003834{
Johannes Weiner867e5e12019-11-30 17:55:34 -08003835 struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid));
Johannes Weiner113b7df2019-05-13 17:18:11 -07003836 unsigned long nr = 0;
3837 enum lru_list lru;
3838
3839 VM_BUG_ON((unsigned)nid >= nr_node_ids);
3840
3841 for_each_lru(lru) {
3842 if (!(BIT(lru) & lru_mask))
3843 continue;
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003844 if (tree)
3845 nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru);
3846 else
3847 nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru);
Johannes Weiner113b7df2019-05-13 17:18:11 -07003848 }
3849 return nr;
3850}
3851
3852static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003853 unsigned int lru_mask,
3854 bool tree)
Johannes Weiner113b7df2019-05-13 17:18:11 -07003855{
3856 unsigned long nr = 0;
3857 enum lru_list lru;
3858
3859 for_each_lru(lru) {
3860 if (!(BIT(lru) & lru_mask))
3861 continue;
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003862 if (tree)
3863 nr += memcg_page_state(memcg, NR_LRU_BASE + lru);
3864 else
3865 nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru);
Johannes Weiner113b7df2019-05-13 17:18:11 -07003866 }
3867 return nr;
3868}
3869
Tejun Heo2da8ca82013-12-05 12:28:04 -05003870static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003871{
Greg Thelen25485de2013-11-12 15:07:40 -08003872 struct numa_stat {
3873 const char *name;
3874 unsigned int lru_mask;
3875 };
3876
3877 static const struct numa_stat stats[] = {
3878 { "total", LRU_ALL },
3879 { "file", LRU_ALL_FILE },
3880 { "anon", LRU_ALL_ANON },
3881 { "unevictable", BIT(LRU_UNEVICTABLE) },
3882 };
3883 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003884 int nid;
Chris Downaa9694b2019-03-05 15:45:52 -08003885 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Ying Han406eb0c2011-05-26 16:25:37 -07003886
Shakeel Buttfd25a9e2021-11-05 13:37:34 -07003887 mem_cgroup_flush_stats();
Johannes Weiner2d146aa2021-04-29 22:56:26 -07003888
Greg Thelen25485de2013-11-12 15:07:40 -08003889 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003890 seq_printf(m, "%s=%lu", stat->name,
3891 mem_cgroup_nr_lru_pages(memcg, stat->lru_mask,
3892 false));
3893 for_each_node_state(nid, N_MEMORY)
3894 seq_printf(m, " N%d=%lu", nid,
3895 mem_cgroup_node_nr_lru_pages(memcg, nid,
3896 stat->lru_mask, false));
Greg Thelen25485de2013-11-12 15:07:40 -08003897 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003898 }
Ying Han406eb0c2011-05-26 16:25:37 -07003899
Ying Han071aee12013-11-12 15:07:41 -08003900 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
Ying Han406eb0c2011-05-26 16:25:37 -07003901
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003902 seq_printf(m, "hierarchical_%s=%lu", stat->name,
3903 mem_cgroup_nr_lru_pages(memcg, stat->lru_mask,
3904 true));
3905 for_each_node_state(nid, N_MEMORY)
3906 seq_printf(m, " N%d=%lu", nid,
3907 mem_cgroup_node_nr_lru_pages(memcg, nid,
3908 stat->lru_mask, true));
Ying Han071aee12013-11-12 15:07:41 -08003909 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07003910 }
Ying Han406eb0c2011-05-26 16:25:37 -07003911
Ying Han406eb0c2011-05-26 16:25:37 -07003912 return 0;
3913}
3914#endif /* CONFIG_NUMA */
3915
Johannes Weinerc8713d02019-07-11 20:55:59 -07003916static const unsigned int memcg1_stats[] = {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07003917 NR_FILE_PAGES,
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07003918 NR_ANON_MAPPED,
Johannes Weiner468c3982020-06-03 16:02:01 -07003919#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3920 NR_ANON_THPS,
3921#endif
Johannes Weinerc8713d02019-07-11 20:55:59 -07003922 NR_SHMEM,
3923 NR_FILE_MAPPED,
3924 NR_FILE_DIRTY,
3925 NR_WRITEBACK,
3926 MEMCG_SWAP,
3927};
3928
3929static const char *const memcg1_stat_names[] = {
3930 "cache",
3931 "rss",
Johannes Weiner468c3982020-06-03 16:02:01 -07003932#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Johannes Weinerc8713d02019-07-11 20:55:59 -07003933 "rss_huge",
Johannes Weiner468c3982020-06-03 16:02:01 -07003934#endif
Johannes Weinerc8713d02019-07-11 20:55:59 -07003935 "shmem",
3936 "mapped_file",
3937 "dirty",
3938 "writeback",
3939 "swap",
3940};
3941
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07003942/* Universal VM events cgroup1 shows, original sort order */
Greg Thelen8dd53fd2018-06-07 17:07:23 -07003943static const unsigned int memcg1_events[] = {
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07003944 PGPGIN,
3945 PGPGOUT,
3946 PGFAULT,
3947 PGMAJFAULT,
3948};
3949
Tejun Heo2da8ca82013-12-05 12:28:04 -05003950static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003951{
Chris Downaa9694b2019-03-05 15:45:52 -08003952 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003953 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003954 struct mem_cgroup *mi;
3955 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08003956
Johannes Weiner71cd3112017-05-03 14:55:13 -07003957 BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats));
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08003958
Shakeel Buttfd25a9e2021-11-05 13:37:34 -07003959 mem_cgroup_flush_stats();
Johannes Weiner2d146aa2021-04-29 22:56:26 -07003960
Johannes Weiner71cd3112017-05-03 14:55:13 -07003961 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
Johannes Weiner468c3982020-06-03 16:02:01 -07003962 unsigned long nr;
3963
Johannes Weiner71cd3112017-05-03 14:55:13 -07003964 if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003965 continue;
Johannes Weiner468c3982020-06-03 16:02:01 -07003966 nr = memcg_page_state_local(memcg, memcg1_stats[i]);
Johannes Weiner468c3982020-06-03 16:02:01 -07003967 seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003968 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08003969
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07003970 for (i = 0; i < ARRAY_SIZE(memcg1_events); i++)
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08003971 seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]),
Johannes Weiner205b20c2019-05-14 15:47:06 -07003972 memcg_events_local(memcg, memcg1_events[i]));
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003973
3974 for (i = 0; i < NR_LRU_LISTS; i++)
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08003975 seq_printf(m, "%s %lu\n", lru_list_name(i),
Johannes Weiner205b20c2019-05-14 15:47:06 -07003976 memcg_page_state_local(memcg, NR_LRU_BASE + i) *
Johannes Weiner21d89d12019-05-13 17:18:08 -07003977 PAGE_SIZE);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003978
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07003979 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003980 memory = memsw = PAGE_COUNTER_MAX;
3981 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
Chris Down15b42562020-04-01 21:07:20 -07003982 memory = min(memory, READ_ONCE(mi->memory.max));
3983 memsw = min(memsw, READ_ONCE(mi->memsw.max));
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08003984 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003985 seq_printf(m, "hierarchical_memory_limit %llu\n",
3986 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08003987 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003988 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3989 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08003990
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07003991 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
zhongjiang-ali7de2e9f2020-11-01 17:07:30 -08003992 unsigned long nr;
3993
Johannes Weiner71cd3112017-05-03 14:55:13 -07003994 if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07003995 continue;
zhongjiang-ali7de2e9f2020-11-01 17:07:30 -08003996 nr = memcg_page_state(memcg, memcg1_stats[i]);
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07003997 seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i],
zhongjiang-ali7de2e9f2020-11-01 17:07:30 -08003998 (u64)nr * PAGE_SIZE);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07003999 }
4000
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07004001 for (i = 0; i < ARRAY_SIZE(memcg1_events); i++)
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08004002 seq_printf(m, "total_%s %llu\n",
4003 vm_event_name(memcg1_events[i]),
Yafang Shaodd923992019-07-11 20:52:11 -07004004 (u64)memcg_events(memcg, memcg1_events[i]));
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004005
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07004006 for (i = 0; i < NR_LRU_LISTS; i++)
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08004007 seq_printf(m, "total_%s %llu\n", lru_list_name(i),
Johannes Weiner42a30032019-05-14 15:47:12 -07004008 (u64)memcg_page_state(memcg, NR_LRU_BASE + i) *
4009 PAGE_SIZE);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004010
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004011#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004012 {
Mel Gormanef8f2322016-07-28 15:46:05 -07004013 pg_data_t *pgdat;
4014 struct mem_cgroup_per_node *mz;
Johannes Weiner1431d4d2020-06-03 16:02:53 -07004015 unsigned long anon_cost = 0;
4016 unsigned long file_cost = 0;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004017
Mel Gormanef8f2322016-07-28 15:46:05 -07004018 for_each_online_pgdat(pgdat) {
Johannes Weinera3747b52021-04-29 22:56:14 -07004019 mz = memcg->nodeinfo[pgdat->node_id];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004020
Johannes Weiner1431d4d2020-06-03 16:02:53 -07004021 anon_cost += mz->lruvec.anon_cost;
4022 file_cost += mz->lruvec.file_cost;
Mel Gormanef8f2322016-07-28 15:46:05 -07004023 }
Johannes Weiner1431d4d2020-06-03 16:02:53 -07004024 seq_printf(m, "anon_cost %lu\n", anon_cost);
4025 seq_printf(m, "file_cost %lu\n", file_cost);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004026 }
4027#endif
4028
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004029 return 0;
4030}
4031
Tejun Heo182446d2013-08-08 20:11:24 -04004032static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
4033 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004034{
Tejun Heo182446d2013-08-08 20:11:24 -04004035 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004036
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004037 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004038}
4039
Tejun Heo182446d2013-08-08 20:11:24 -04004040static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
4041 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004042{
Tejun Heo182446d2013-08-08 20:11:24 -04004043 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08004044
Baolin Wang37bc3cb2021-09-02 14:55:53 -07004045 if (val > 200)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004046 return -EINVAL;
4047
Shakeel Butta4792032021-04-29 22:56:05 -07004048 if (!mem_cgroup_is_root(memcg))
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07004049 memcg->swappiness = val;
4050 else
4051 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08004052
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004053 return 0;
4054}
4055
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004056static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4057{
4058 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004059 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004060 int i;
4061
4062 rcu_read_lock();
4063 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004064 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004065 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004066 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004067
4068 if (!t)
4069 goto unlock;
4070
Johannes Weinerce00a962014-09-05 08:43:57 -04004071 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004072
4073 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07004074 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004075 * If it's not true, a threshold was crossed after last
4076 * call of __mem_cgroup_threshold().
4077 */
Phil Carmody5407a562010-05-26 14:42:42 -07004078 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004079
4080 /*
4081 * Iterate backward over array of thresholds starting from
4082 * current_threshold and check if a threshold is crossed.
4083 * If none of thresholds below usage is crossed, we read
4084 * only one element of the array here.
4085 */
4086 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4087 eventfd_signal(t->entries[i].eventfd, 1);
4088
4089 /* i = current_threshold + 1 */
4090 i++;
4091
4092 /*
4093 * Iterate forward over array of thresholds starting from
4094 * current_threshold+1 and check if a threshold is crossed.
4095 * If none of thresholds above usage is crossed, we read
4096 * only one element of the array here.
4097 */
4098 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4099 eventfd_signal(t->entries[i].eventfd, 1);
4100
4101 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004102 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004103unlock:
4104 rcu_read_unlock();
4105}
4106
4107static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4108{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004109 while (memcg) {
4110 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08004111 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004112 __mem_cgroup_threshold(memcg, true);
4113
4114 memcg = parent_mem_cgroup(memcg);
4115 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004116}
4117
4118static int compare_thresholds(const void *a, const void *b)
4119{
4120 const struct mem_cgroup_threshold *_a = a;
4121 const struct mem_cgroup_threshold *_b = b;
4122
Greg Thelen2bff24a2013-09-11 14:23:08 -07004123 if (_a->threshold > _b->threshold)
4124 return 1;
4125
4126 if (_a->threshold < _b->threshold)
4127 return -1;
4128
4129 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004130}
4131
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004132static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004133{
4134 struct mem_cgroup_eventfd_list *ev;
4135
Michal Hocko2bcf2e92014-07-30 16:08:33 -07004136 spin_lock(&memcg_oom_lock);
4137
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004138 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004139 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07004140
4141 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004142 return 0;
4143}
4144
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004145static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004146{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004147 struct mem_cgroup *iter;
4148
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004149 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004150 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004151}
4152
Tejun Heo59b6f872013-11-22 18:20:43 -05004153static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004154 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004155{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004156 struct mem_cgroup_thresholds *thresholds;
4157 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004158 unsigned long threshold;
4159 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004160 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004161
Johannes Weiner650c5e52015-02-11 15:26:03 -08004162 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004163 if (ret)
4164 return ret;
4165
4166 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004167
Johannes Weiner05b84302014-08-06 16:05:59 -07004168 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004169 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004170 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07004171 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004172 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004173 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07004174 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004175 BUG();
4176
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004177 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004178 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004179 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4180
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004181 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004182
4183 /* Allocate memory for new array of thresholds */
Gustavo A. R. Silva67b80462019-03-05 15:44:05 -08004184 new = kmalloc(struct_size(new, entries, size), GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004185 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004186 ret = -ENOMEM;
4187 goto unlock;
4188 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004189 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004190
4191 /* Copy thresholds (if any) to new array */
Gustavo A. R. Silvae90342e2020-10-13 16:52:36 -07004192 if (thresholds->primary)
4193 memcpy(new->entries, thresholds->primary->entries,
4194 flex_array_size(new, entries, size - 1));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004195
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004196 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004197 new->entries[size - 1].eventfd = eventfd;
4198 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004199
4200 /* Sort thresholds. Registering of new threshold isn't time-critical */
Gustavo A. R. Silva61e604e62020-10-13 16:52:39 -07004201 sort(new->entries, size, sizeof(*new->entries),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004202 compare_thresholds, NULL);
4203
4204 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004205 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004206 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07004207 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004208 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004209 * new->current_threshold will not be used until
4210 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004211 * it here.
4212 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004213 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07004214 } else
4215 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004216 }
4217
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004218 /* Free old spare buffer and save old primary buffer as spare */
4219 kfree(thresholds->spare);
4220 thresholds->spare = thresholds->primary;
4221
4222 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004223
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004224 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004225 synchronize_rcu();
4226
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004227unlock:
4228 mutex_unlock(&memcg->thresholds_lock);
4229
4230 return ret;
4231}
4232
Tejun Heo59b6f872013-11-22 18:20:43 -05004233static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004234 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004235{
Tejun Heo59b6f872013-11-22 18:20:43 -05004236 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05004237}
4238
Tejun Heo59b6f872013-11-22 18:20:43 -05004239static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004240 struct eventfd_ctx *eventfd, const char *args)
4241{
Tejun Heo59b6f872013-11-22 18:20:43 -05004242 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05004243}
4244
Tejun Heo59b6f872013-11-22 18:20:43 -05004245static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004246 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004247{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004248 struct mem_cgroup_thresholds *thresholds;
4249 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004250 unsigned long usage;
Chunguang Xu7d366652020-03-21 18:22:10 -07004251 int i, j, size, entries;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004252
4253 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07004254
4255 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004256 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004257 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07004258 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004259 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004260 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07004261 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004262 BUG();
4263
Anton Vorontsov371528c2012-02-24 05:14:46 +04004264 if (!thresholds->primary)
4265 goto unlock;
4266
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004267 /* Check if a threshold crossed before removing */
4268 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4269
4270 /* Calculate new number of threshold */
Chunguang Xu7d366652020-03-21 18:22:10 -07004271 size = entries = 0;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004272 for (i = 0; i < thresholds->primary->size; i++) {
4273 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004274 size++;
Chunguang Xu7d366652020-03-21 18:22:10 -07004275 else
4276 entries++;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004277 }
4278
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004279 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004280
Chunguang Xu7d366652020-03-21 18:22:10 -07004281 /* If no items related to eventfd have been cleared, nothing to do */
4282 if (!entries)
4283 goto unlock;
4284
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004285 /* Set thresholds array to NULL if we don't have thresholds */
4286 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004287 kfree(new);
4288 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004289 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004290 }
4291
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004292 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004293
4294 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004295 new->current_threshold = -1;
4296 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4297 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004298 continue;
4299
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004300 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07004301 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004302 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004303 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004304 * until rcu_assign_pointer(), so it's safe to increment
4305 * it here.
4306 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004307 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004308 }
4309 j++;
4310 }
4311
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004312swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004313 /* Swap primary and spare array */
4314 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07004315
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004316 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004317
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004318 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004319 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08004320
4321 /* If all events are unregistered, free the spare array */
4322 if (!new) {
4323 kfree(thresholds->spare);
4324 thresholds->spare = NULL;
4325 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04004326unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004327 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004328}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004329
Tejun Heo59b6f872013-11-22 18:20:43 -05004330static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004331 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004332{
Tejun Heo59b6f872013-11-22 18:20:43 -05004333 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05004334}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004335
Tejun Heo59b6f872013-11-22 18:20:43 -05004336static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004337 struct eventfd_ctx *eventfd)
4338{
Tejun Heo59b6f872013-11-22 18:20:43 -05004339 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05004340}
4341
Tejun Heo59b6f872013-11-22 18:20:43 -05004342static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004343 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004344{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004345 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004346
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004347 event = kmalloc(sizeof(*event), GFP_KERNEL);
4348 if (!event)
4349 return -ENOMEM;
4350
Michal Hocko1af8efe2011-07-26 16:08:24 -07004351 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004352
4353 event->eventfd = eventfd;
4354 list_add(&event->list, &memcg->oom_notify);
4355
4356 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07004357 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004358 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004359 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004360
4361 return 0;
4362}
4363
Tejun Heo59b6f872013-11-22 18:20:43 -05004364static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004365 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004366{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004367 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004368
Michal Hocko1af8efe2011-07-26 16:08:24 -07004369 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004370
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004371 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004372 if (ev->eventfd == eventfd) {
4373 list_del(&ev->list);
4374 kfree(ev);
4375 }
4376 }
4377
Michal Hocko1af8efe2011-07-26 16:08:24 -07004378 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004379}
4380
Tejun Heo2da8ca82013-12-05 12:28:04 -05004381static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004382{
Chris Downaa9694b2019-03-05 15:45:52 -08004383 struct mem_cgroup *memcg = mem_cgroup_from_seq(sf);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004384
Tejun Heo791badb2013-12-05 12:28:02 -05004385 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07004386 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
Roman Gushchinfe6bdfc2018-06-14 15:28:05 -07004387 seq_printf(sf, "oom_kill %lu\n",
4388 atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL]));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004389 return 0;
4390}
4391
Tejun Heo182446d2013-08-08 20:11:24 -04004392static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004393 struct cftype *cft, u64 val)
4394{
Tejun Heo182446d2013-08-08 20:11:24 -04004395 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004396
4397 /* cannot set to root cgroup and only 0 and 1 are allowed */
Shakeel Butta4792032021-04-29 22:56:05 -07004398 if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004399 return -EINVAL;
4400
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004401 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004402 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004403 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07004404
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004405 return 0;
4406}
4407
Tejun Heo52ebea72015-05-22 17:13:37 -04004408#ifdef CONFIG_CGROUP_WRITEBACK
4409
Tejun Heo3a8e9ac2019-08-29 15:47:19 -07004410#include <trace/events/writeback.h>
4411
Tejun Heo841710a2015-05-22 18:23:33 -04004412static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
4413{
4414 return wb_domain_init(&memcg->cgwb_domain, gfp);
4415}
4416
4417static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
4418{
4419 wb_domain_exit(&memcg->cgwb_domain);
4420}
4421
Tejun Heo2529bb32015-05-22 18:23:34 -04004422static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
4423{
4424 wb_domain_size_changed(&memcg->cgwb_domain);
4425}
4426
Tejun Heo841710a2015-05-22 18:23:33 -04004427struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
4428{
4429 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
4430
4431 if (!memcg->css.parent)
4432 return NULL;
4433
4434 return &memcg->cgwb_domain;
4435}
4436
Tejun Heoc2aa7232015-05-22 18:23:35 -04004437/**
4438 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
4439 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004440 * @pfilepages: out parameter for number of file pages
4441 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04004442 * @pdirty: out parameter for number of dirty pages
4443 * @pwriteback: out parameter for number of pages under writeback
4444 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004445 * Determine the numbers of file, headroom, dirty, and writeback pages in
4446 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
4447 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04004448 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004449 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
4450 * headroom is calculated as the lowest headroom of itself and the
4451 * ancestors. Note that this doesn't consider the actual amount of
4452 * available memory in the system. The caller should further cap
4453 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04004454 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004455void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
4456 unsigned long *pheadroom, unsigned long *pdirty,
4457 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04004458{
4459 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
4460 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04004461
Shakeel Buttfd25a9e2021-11-05 13:37:34 -07004462 mem_cgroup_flush_stats();
Tejun Heoc2aa7232015-05-22 18:23:35 -04004463
Johannes Weiner2d146aa2021-04-29 22:56:26 -07004464 *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY);
4465 *pwriteback = memcg_page_state(memcg, NR_WRITEBACK);
4466 *pfilepages = memcg_page_state(memcg, NR_INACTIVE_FILE) +
4467 memcg_page_state(memcg, NR_ACTIVE_FILE);
4468
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004469 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04004470 while ((parent = parent_mem_cgroup(memcg))) {
Chris Down15b42562020-04-01 21:07:20 -07004471 unsigned long ceiling = min(READ_ONCE(memcg->memory.max),
Jakub Kicinskid1663a92020-06-01 21:49:49 -07004472 READ_ONCE(memcg->memory.high));
Tejun Heoc2aa7232015-05-22 18:23:35 -04004473 unsigned long used = page_counter_read(&memcg->memory);
4474
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004475 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04004476 memcg = parent;
4477 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04004478}
4479
Tejun Heo97b27822019-08-26 09:06:56 -07004480/*
4481 * Foreign dirty flushing
4482 *
4483 * There's an inherent mismatch between memcg and writeback. The former
Ingo Molnarf0953a12021-05-06 18:06:47 -07004484 * tracks ownership per-page while the latter per-inode. This was a
Tejun Heo97b27822019-08-26 09:06:56 -07004485 * deliberate design decision because honoring per-page ownership in the
4486 * writeback path is complicated, may lead to higher CPU and IO overheads
4487 * and deemed unnecessary given that write-sharing an inode across
4488 * different cgroups isn't a common use-case.
4489 *
4490 * Combined with inode majority-writer ownership switching, this works well
4491 * enough in most cases but there are some pathological cases. For
4492 * example, let's say there are two cgroups A and B which keep writing to
4493 * different but confined parts of the same inode. B owns the inode and
4494 * A's memory is limited far below B's. A's dirty ratio can rise enough to
4495 * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid
4496 * triggering background writeback. A will be slowed down without a way to
4497 * make writeback of the dirty pages happen.
4498 *
Ingo Molnarf0953a12021-05-06 18:06:47 -07004499 * Conditions like the above can lead to a cgroup getting repeatedly and
Tejun Heo97b27822019-08-26 09:06:56 -07004500 * severely throttled after making some progress after each
Ingo Molnarf0953a12021-05-06 18:06:47 -07004501 * dirty_expire_interval while the underlying IO device is almost
Tejun Heo97b27822019-08-26 09:06:56 -07004502 * completely idle.
4503 *
4504 * Solving this problem completely requires matching the ownership tracking
4505 * granularities between memcg and writeback in either direction. However,
4506 * the more egregious behaviors can be avoided by simply remembering the
4507 * most recent foreign dirtying events and initiating remote flushes on
4508 * them when local writeback isn't enough to keep the memory clean enough.
4509 *
4510 * The following two functions implement such mechanism. When a foreign
4511 * page - a page whose memcg and writeback ownerships don't match - is
4512 * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning
4513 * bdi_writeback on the page owning memcg. When balance_dirty_pages()
4514 * decides that the memcg needs to sleep due to high dirty ratio, it calls
4515 * mem_cgroup_flush_foreign() which queues writeback on the recorded
4516 * foreign bdi_writebacks which haven't expired. Both the numbers of
4517 * recorded bdi_writebacks and concurrent in-flight foreign writebacks are
4518 * limited to MEMCG_CGWB_FRN_CNT.
4519 *
4520 * The mechanism only remembers IDs and doesn't hold any object references.
4521 * As being wrong occasionally doesn't matter, updates and accesses to the
4522 * records are lockless and racy.
4523 */
4524void mem_cgroup_track_foreign_dirty_slowpath(struct page *page,
4525 struct bdi_writeback *wb)
4526{
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08004527 struct mem_cgroup *memcg = page_memcg(page);
Tejun Heo97b27822019-08-26 09:06:56 -07004528 struct memcg_cgwb_frn *frn;
4529 u64 now = get_jiffies_64();
4530 u64 oldest_at = now;
4531 int oldest = -1;
4532 int i;
4533
Tejun Heo3a8e9ac2019-08-29 15:47:19 -07004534 trace_track_foreign_dirty(page, wb);
4535
Tejun Heo97b27822019-08-26 09:06:56 -07004536 /*
4537 * Pick the slot to use. If there is already a slot for @wb, keep
4538 * using it. If not replace the oldest one which isn't being
4539 * written out.
4540 */
4541 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) {
4542 frn = &memcg->cgwb_frn[i];
4543 if (frn->bdi_id == wb->bdi->id &&
4544 frn->memcg_id == wb->memcg_css->id)
4545 break;
4546 if (time_before64(frn->at, oldest_at) &&
4547 atomic_read(&frn->done.cnt) == 1) {
4548 oldest = i;
4549 oldest_at = frn->at;
4550 }
4551 }
4552
4553 if (i < MEMCG_CGWB_FRN_CNT) {
4554 /*
4555 * Re-using an existing one. Update timestamp lazily to
4556 * avoid making the cacheline hot. We want them to be
4557 * reasonably up-to-date and significantly shorter than
4558 * dirty_expire_interval as that's what expires the record.
4559 * Use the shorter of 1s and dirty_expire_interval / 8.
4560 */
4561 unsigned long update_intv =
4562 min_t(unsigned long, HZ,
4563 msecs_to_jiffies(dirty_expire_interval * 10) / 8);
4564
4565 if (time_before64(frn->at, now - update_intv))
4566 frn->at = now;
4567 } else if (oldest >= 0) {
4568 /* replace the oldest free one */
4569 frn = &memcg->cgwb_frn[oldest];
4570 frn->bdi_id = wb->bdi->id;
4571 frn->memcg_id = wb->memcg_css->id;
4572 frn->at = now;
4573 }
4574}
4575
4576/* issue foreign writeback flushes for recorded foreign dirtying events */
4577void mem_cgroup_flush_foreign(struct bdi_writeback *wb)
4578{
4579 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
4580 unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10);
4581 u64 now = jiffies_64;
4582 int i;
4583
4584 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) {
4585 struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i];
4586
4587 /*
4588 * If the record is older than dirty_expire_interval,
4589 * writeback on it has already started. No need to kick it
4590 * off again. Also, don't start a new one if there's
4591 * already one in flight.
4592 */
4593 if (time_after64(frn->at, now - intv) &&
4594 atomic_read(&frn->done.cnt) == 1) {
4595 frn->at = 0;
Tejun Heo3a8e9ac2019-08-29 15:47:19 -07004596 trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id);
Shakeel Butt7490a2d2021-09-02 14:53:27 -07004597 cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id,
Tejun Heo97b27822019-08-26 09:06:56 -07004598 WB_REASON_FOREIGN_FLUSH,
4599 &frn->done);
4600 }
4601 }
4602}
4603
Tejun Heo841710a2015-05-22 18:23:33 -04004604#else /* CONFIG_CGROUP_WRITEBACK */
4605
4606static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
4607{
4608 return 0;
4609}
4610
4611static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
4612{
4613}
4614
Tejun Heo2529bb32015-05-22 18:23:34 -04004615static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
4616{
4617}
4618
Tejun Heo52ebea72015-05-22 17:13:37 -04004619#endif /* CONFIG_CGROUP_WRITEBACK */
4620
Tejun Heo79bd9812013-11-22 18:20:42 -05004621/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05004622 * DO NOT USE IN NEW FILES.
4623 *
4624 * "cgroup.event_control" implementation.
4625 *
4626 * This is way over-engineered. It tries to support fully configurable
4627 * events for each user. Such level of flexibility is completely
4628 * unnecessary especially in the light of the planned unified hierarchy.
4629 *
4630 * Please deprecate this and replace with something simpler if at all
4631 * possible.
4632 */
4633
4634/*
Tejun Heo79bd9812013-11-22 18:20:42 -05004635 * Unregister event and free resources.
4636 *
4637 * Gets called from workqueue.
4638 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05004639static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05004640{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004641 struct mem_cgroup_event *event =
4642 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05004643 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05004644
4645 remove_wait_queue(event->wqh, &event->wait);
4646
Tejun Heo59b6f872013-11-22 18:20:43 -05004647 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05004648
4649 /* Notify userspace the event is going away. */
4650 eventfd_signal(event->eventfd, 1);
4651
4652 eventfd_ctx_put(event->eventfd);
4653 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05004654 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05004655}
4656
4657/*
Linus Torvaldsa9a08842018-02-11 14:34:03 -08004658 * Gets called on EPOLLHUP on eventfd when user closes it.
Tejun Heo79bd9812013-11-22 18:20:42 -05004659 *
4660 * Called with wqh->lock held and interrupts disabled.
4661 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02004662static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode,
Tejun Heo3bc942f2013-11-22 18:20:44 -05004663 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05004664{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004665 struct mem_cgroup_event *event =
4666 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05004667 struct mem_cgroup *memcg = event->memcg;
Al Viro3ad6f932017-07-03 20:14:56 -04004668 __poll_t flags = key_to_poll(key);
Tejun Heo79bd9812013-11-22 18:20:42 -05004669
Linus Torvaldsa9a08842018-02-11 14:34:03 -08004670 if (flags & EPOLLHUP) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004671 /*
4672 * If the event has been detached at cgroup removal, we
4673 * can simply return knowing the other side will cleanup
4674 * for us.
4675 *
4676 * We can't race against event freeing since the other
4677 * side will require wqh->lock via remove_wait_queue(),
4678 * which we hold.
4679 */
Tejun Heofba94802013-11-22 18:20:43 -05004680 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004681 if (!list_empty(&event->list)) {
4682 list_del_init(&event->list);
4683 /*
4684 * We are in atomic context, but cgroup_event_remove()
4685 * may sleep, so we have to call it in workqueue.
4686 */
4687 schedule_work(&event->remove);
4688 }
Tejun Heofba94802013-11-22 18:20:43 -05004689 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004690 }
4691
4692 return 0;
4693}
4694
Tejun Heo3bc942f2013-11-22 18:20:44 -05004695static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05004696 wait_queue_head_t *wqh, poll_table *pt)
4697{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004698 struct mem_cgroup_event *event =
4699 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05004700
4701 event->wqh = wqh;
4702 add_wait_queue(wqh, &event->wait);
4703}
4704
4705/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05004706 * DO NOT USE IN NEW FILES.
4707 *
Tejun Heo79bd9812013-11-22 18:20:42 -05004708 * Parse input and register new cgroup event handler.
4709 *
4710 * Input must be in format '<event_fd> <control_fd> <args>'.
4711 * Interpretation of args is defined by control file implementation.
4712 */
Tejun Heo451af502014-05-13 12:16:21 -04004713static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
4714 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05004715{
Tejun Heo451af502014-05-13 12:16:21 -04004716 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05004717 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004718 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05004719 struct cgroup_subsys_state *cfile_css;
4720 unsigned int efd, cfd;
4721 struct fd efile;
4722 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05004723 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05004724 char *endp;
4725 int ret;
4726
Tejun Heo451af502014-05-13 12:16:21 -04004727 buf = strstrip(buf);
4728
4729 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05004730 if (*endp != ' ')
4731 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04004732 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05004733
Tejun Heo451af502014-05-13 12:16:21 -04004734 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05004735 if ((*endp != ' ') && (*endp != '\0'))
4736 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04004737 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05004738
4739 event = kzalloc(sizeof(*event), GFP_KERNEL);
4740 if (!event)
4741 return -ENOMEM;
4742
Tejun Heo59b6f872013-11-22 18:20:43 -05004743 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05004744 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004745 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
4746 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
4747 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05004748
4749 efile = fdget(efd);
4750 if (!efile.file) {
4751 ret = -EBADF;
4752 goto out_kfree;
4753 }
4754
4755 event->eventfd = eventfd_ctx_fileget(efile.file);
4756 if (IS_ERR(event->eventfd)) {
4757 ret = PTR_ERR(event->eventfd);
4758 goto out_put_efile;
4759 }
4760
4761 cfile = fdget(cfd);
4762 if (!cfile.file) {
4763 ret = -EBADF;
4764 goto out_put_eventfd;
4765 }
4766
4767 /* the process need read permission on control file */
4768 /* AV: shouldn't we check that it's been opened for read instead? */
Christian Brauner02f92b32021-01-21 14:19:22 +01004769 ret = file_permission(cfile.file, MAY_READ);
Tejun Heo79bd9812013-11-22 18:20:42 -05004770 if (ret < 0)
4771 goto out_put_cfile;
4772
Tejun Heo79bd9812013-11-22 18:20:42 -05004773 /*
Tejun Heofba94802013-11-22 18:20:43 -05004774 * Determine the event callbacks and set them in @event. This used
4775 * to be done via struct cftype but cgroup core no longer knows
4776 * about these events. The following is crude but the whole thing
4777 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05004778 *
4779 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05004780 */
Al Virob5830432014-10-31 01:22:04 -04004781 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05004782
4783 if (!strcmp(name, "memory.usage_in_bytes")) {
4784 event->register_event = mem_cgroup_usage_register_event;
4785 event->unregister_event = mem_cgroup_usage_unregister_event;
4786 } else if (!strcmp(name, "memory.oom_control")) {
4787 event->register_event = mem_cgroup_oom_register_event;
4788 event->unregister_event = mem_cgroup_oom_unregister_event;
4789 } else if (!strcmp(name, "memory.pressure_level")) {
4790 event->register_event = vmpressure_register_event;
4791 event->unregister_event = vmpressure_unregister_event;
4792 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05004793 event->register_event = memsw_cgroup_usage_register_event;
4794 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05004795 } else {
4796 ret = -EINVAL;
4797 goto out_put_cfile;
4798 }
4799
4800 /*
Tejun Heob5557c42013-11-22 18:20:42 -05004801 * Verify @cfile should belong to @css. Also, remaining events are
4802 * automatically removed on cgroup destruction but the removal is
4803 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05004804 */
Al Virob5830432014-10-31 01:22:04 -04004805 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04004806 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05004807 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05004808 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05004809 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05004810 if (cfile_css != css) {
4811 css_put(cfile_css);
4812 goto out_put_cfile;
4813 }
Tejun Heo79bd9812013-11-22 18:20:42 -05004814
Tejun Heo451af502014-05-13 12:16:21 -04004815 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05004816 if (ret)
4817 goto out_put_css;
4818
Christoph Hellwig9965ed172018-03-05 07:26:05 -08004819 vfs_poll(efile.file, &event->pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05004820
Shakeel Butt4ba95152021-09-02 14:56:05 -07004821 spin_lock_irq(&memcg->event_list_lock);
Tejun Heofba94802013-11-22 18:20:43 -05004822 list_add(&event->list, &memcg->event_list);
Shakeel Butt4ba95152021-09-02 14:56:05 -07004823 spin_unlock_irq(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004824
4825 fdput(cfile);
4826 fdput(efile);
4827
Tejun Heo451af502014-05-13 12:16:21 -04004828 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05004829
4830out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05004831 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05004832out_put_cfile:
4833 fdput(cfile);
4834out_put_eventfd:
4835 eventfd_ctx_put(event->eventfd);
4836out_put_efile:
4837 fdput(efile);
4838out_kfree:
4839 kfree(event);
4840
4841 return ret;
4842}
4843
Johannes Weiner241994ed2015-02-11 15:26:06 -08004844static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004845 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004846 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004847 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004848 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004849 },
4850 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004851 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004852 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004853 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004854 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004855 },
4856 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004857 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004858 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004859 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004860 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004861 },
4862 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004863 .name = "soft_limit_in_bytes",
4864 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004865 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004866 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07004867 },
4868 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004869 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004870 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004871 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004872 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004873 },
Balbir Singh8697d332008-02-07 00:13:59 -08004874 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004875 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004876 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004877 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004878 {
4879 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04004880 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004881 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08004882 {
4883 .name = "use_hierarchy",
4884 .write_u64 = mem_cgroup_hierarchy_write,
4885 .read_u64 = mem_cgroup_hierarchy_read,
4886 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004887 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05004888 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04004889 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04004890 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05004891 },
4892 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004893 .name = "swappiness",
4894 .read_u64 = mem_cgroup_swappiness_read,
4895 .write_u64 = mem_cgroup_swappiness_write,
4896 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004897 {
4898 .name = "move_charge_at_immigrate",
4899 .read_u64 = mem_cgroup_move_charge_read,
4900 .write_u64 = mem_cgroup_move_charge_write,
4901 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004902 {
4903 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004904 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004905 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004906 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4907 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004908 {
4909 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07004910 },
Ying Han406eb0c2011-05-26 16:25:37 -07004911#ifdef CONFIG_NUMA
4912 {
4913 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05004914 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07004915 },
4916#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08004917 {
4918 .name = "kmem.limit_in_bytes",
4919 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004920 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004921 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004922 },
4923 {
4924 .name = "kmem.usage_in_bytes",
4925 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004926 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004927 },
4928 {
4929 .name = "kmem.failcnt",
4930 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04004931 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004932 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004933 },
4934 {
4935 .name = "kmem.max_usage_in_bytes",
4936 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004937 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004938 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08004939 },
Yafang Shaoa87425a2020-04-01 21:06:30 -07004940#if defined(CONFIG_MEMCG_KMEM) && \
4941 (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG))
Glauber Costa749c5412012-12-18 14:23:01 -08004942 {
4943 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08004944 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08004945 },
4946#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004947 {
4948 .name = "kmem.tcp.limit_in_bytes",
4949 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4950 .write = mem_cgroup_write,
4951 .read_u64 = mem_cgroup_read_u64,
4952 },
4953 {
4954 .name = "kmem.tcp.usage_in_bytes",
4955 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4956 .read_u64 = mem_cgroup_read_u64,
4957 },
4958 {
4959 .name = "kmem.tcp.failcnt",
4960 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4961 .write = mem_cgroup_reset,
4962 .read_u64 = mem_cgroup_read_u64,
4963 },
4964 {
4965 .name = "kmem.tcp.max_usage_in_bytes",
4966 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4967 .write = mem_cgroup_reset,
4968 .read_u64 = mem_cgroup_read_u64,
4969 },
Tejun Heo6bc10342012-04-01 12:09:55 -07004970 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07004971};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004972
Johannes Weiner73f576c2016-07-20 15:44:57 -07004973/*
4974 * Private memory cgroup IDR
4975 *
4976 * Swap-out records and page cache shadow entries need to store memcg
4977 * references in constrained space, so we maintain an ID space that is
4978 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4979 * memory-controlled cgroups to 64k.
4980 *
Ethon Paulb8f29352020-06-04 16:49:28 -07004981 * However, there usually are many references to the offline CSS after
Johannes Weiner73f576c2016-07-20 15:44:57 -07004982 * the cgroup has been destroyed, such as page cache or reclaimable
4983 * slab objects, that don't need to hang on to the ID. We want to keep
4984 * those dead CSS from occupying IDs, or we might quickly exhaust the
4985 * relatively small ID space and prevent the creation of new cgroups
4986 * even when there are much fewer than 64k cgroups - possibly none.
4987 *
4988 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4989 * be freed and recycled when it's no longer needed, which is usually
4990 * when the CSS is offlined.
4991 *
4992 * The only exception to that are records of swapped out tmpfs/shmem
4993 * pages that need to be attributed to live ancestors on swapin. But
4994 * those references are manageable from userspace.
4995 */
4996
4997static DEFINE_IDR(mem_cgroup_idr);
4998
Kirill Tkhai7e97de02018-08-02 15:36:01 -07004999static void mem_cgroup_id_remove(struct mem_cgroup *memcg)
5000{
5001 if (memcg->id.id > 0) {
5002 idr_remove(&mem_cgroup_idr, memcg->id.id);
5003 memcg->id.id = 0;
5004 }
5005}
5006
Vincenzo Frascinoc1514c02020-04-01 21:07:13 -07005007static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg,
5008 unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07005009{
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07005010 refcount_add(n, &memcg->id.ref);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005011}
5012
Vladimir Davydov615d66c2016-08-11 15:33:03 -07005013static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07005014{
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07005015 if (refcount_sub_and_test(n, &memcg->id.ref)) {
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005016 mem_cgroup_id_remove(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005017
5018 /* Memcg ID pins CSS */
5019 css_put(&memcg->css);
5020 }
5021}
5022
Vladimir Davydov615d66c2016-08-11 15:33:03 -07005023static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
5024{
5025 mem_cgroup_id_put_many(memcg, 1);
5026}
5027
Johannes Weiner73f576c2016-07-20 15:44:57 -07005028/**
5029 * mem_cgroup_from_id - look up a memcg from a memcg id
5030 * @id: the memcg id to look up
5031 *
5032 * Caller must hold rcu_read_lock().
5033 */
5034struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
5035{
5036 WARN_ON_ONCE(!rcu_read_lock_held());
5037 return idr_find(&mem_cgroup_idr, id);
5038}
5039
Mel Gormanef8f2322016-07-28 15:46:05 -07005040static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005041{
5042 struct mem_cgroup_per_node *pn;
Mel Gormanef8f2322016-07-28 15:46:05 -07005043 int tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005044 /*
5045 * This routine is called against possible nodes.
5046 * But it's BUG to call kmalloc() against offline node.
5047 *
5048 * TODO: this routine can waste much memory for nodes which will
5049 * never be onlined. It's better to use memory hotplug callback
5050 * function.
5051 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07005052 if (!node_state(node, N_NORMAL_MEMORY))
5053 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08005054 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005055 if (!pn)
5056 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005057
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07005058 pn->lruvec_stats_percpu = alloc_percpu_gfp(struct lruvec_stats_percpu,
5059 GFP_KERNEL_ACCOUNT);
5060 if (!pn->lruvec_stats_percpu) {
Johannes Weiner00f3ca22017-07-06 15:40:52 -07005061 kfree(pn);
5062 return 1;
5063 }
5064
Mel Gormanef8f2322016-07-28 15:46:05 -07005065 lruvec_init(&pn->lruvec);
5066 pn->usage_in_excess = 0;
5067 pn->on_tree = false;
5068 pn->memcg = memcg;
5069
Johannes Weiner54f72fe2013-07-08 15:59:49 -07005070 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005071 return 0;
5072}
5073
Mel Gormanef8f2322016-07-28 15:46:05 -07005074static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005075{
Johannes Weiner00f3ca22017-07-06 15:40:52 -07005076 struct mem_cgroup_per_node *pn = memcg->nodeinfo[node];
5077
Michal Hocko4eaf4312018-04-10 16:29:52 -07005078 if (!pn)
5079 return;
5080
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07005081 free_percpu(pn->lruvec_stats_percpu);
Johannes Weiner00f3ca22017-07-06 15:40:52 -07005082 kfree(pn);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005083}
5084
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005085static void __mem_cgroup_free(struct mem_cgroup *memcg)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005086{
5087 int node;
5088
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005089 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07005090 free_mem_cgroup_per_node_info(memcg, node);
Chris Down871789d2019-05-14 15:46:57 -07005091 free_percpu(memcg->vmstats_percpu);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005092 kfree(memcg);
5093}
5094
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005095static void mem_cgroup_free(struct mem_cgroup *memcg)
5096{
5097 memcg_wb_domain_exit(memcg);
5098 __mem_cgroup_free(memcg);
5099}
5100
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005101static struct mem_cgroup *mem_cgroup_alloc(void)
5102{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005103 struct mem_cgroup *memcg;
Alexey Dobriyanb9726c22019-03-05 15:48:26 -08005104 unsigned int size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005105 int node;
Tejun Heo97b27822019-08-26 09:06:56 -07005106 int __maybe_unused i;
Yafang Shao11d67612020-05-07 18:35:43 -07005107 long error = -ENOMEM;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005108
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08005109 size = sizeof(struct mem_cgroup);
5110 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005111
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08005112 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005113 if (!memcg)
Yafang Shao11d67612020-05-07 18:35:43 -07005114 return ERR_PTR(error);
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07005115
Johannes Weiner73f576c2016-07-20 15:44:57 -07005116 memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
5117 1, MEM_CGROUP_ID_MAX,
5118 GFP_KERNEL);
Yafang Shao11d67612020-05-07 18:35:43 -07005119 if (memcg->id.id < 0) {
5120 error = memcg->id.id;
Johannes Weiner73f576c2016-07-20 15:44:57 -07005121 goto fail;
Yafang Shao11d67612020-05-07 18:35:43 -07005122 }
Johannes Weiner73f576c2016-07-20 15:44:57 -07005123
Roman Gushchin3e38e0a2020-08-11 18:30:25 -07005124 memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu,
5125 GFP_KERNEL_ACCOUNT);
Chris Down871789d2019-05-14 15:46:57 -07005126 if (!memcg->vmstats_percpu)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005127 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08005128
Bob Liu3ed28fa2012-01-12 17:19:04 -08005129 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07005130 if (alloc_mem_cgroup_per_node_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005131 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07005132
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005133 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
5134 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08005135
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005136 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08005137 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08005138 mutex_init(&memcg->thresholds_lock);
5139 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07005140 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05005141 INIT_LIST_HEAD(&memcg->event_list);
5142 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005143 memcg->socket_pressure = jiffies;
Kirill Tkhai84c07d12018-08-17 15:47:25 -07005144#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydov900a38f2014-12-12 16:55:10 -08005145 memcg->kmemcg_id = -1;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07005146 INIT_LIST_HEAD(&memcg->objcg_list);
Vladimir Davydov900a38f2014-12-12 16:55:10 -08005147#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04005148#ifdef CONFIG_CGROUP_WRITEBACK
5149 INIT_LIST_HEAD(&memcg->cgwb_list);
Tejun Heo97b27822019-08-26 09:06:56 -07005150 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++)
5151 memcg->cgwb_frn[i].done =
5152 __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq);
Tejun Heo52ebea72015-05-22 17:13:37 -04005153#endif
Yang Shi87eaceb2019-09-23 15:38:15 -07005154#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5155 spin_lock_init(&memcg->deferred_split_queue.split_queue_lock);
5156 INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue);
5157 memcg->deferred_split_queue.split_queue_len = 0;
5158#endif
Johannes Weiner73f576c2016-07-20 15:44:57 -07005159 idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005160 return memcg;
5161fail:
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005162 mem_cgroup_id_remove(memcg);
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005163 __mem_cgroup_free(memcg);
Yafang Shao11d67612020-05-07 18:35:43 -07005164 return ERR_PTR(error);
Glauber Costad142e3e2013-02-22 16:34:52 -08005165}
5166
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005167static struct cgroup_subsys_state * __ref
5168mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08005169{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005170 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
Roman Gushchinb87d8ce2020-10-17 16:13:40 -07005171 struct mem_cgroup *memcg, *old_memcg;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005172 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08005173
Roman Gushchinb87d8ce2020-10-17 16:13:40 -07005174 old_memcg = set_active_memcg(parent);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005175 memcg = mem_cgroup_alloc();
Roman Gushchinb87d8ce2020-10-17 16:13:40 -07005176 set_active_memcg(old_memcg);
Yafang Shao11d67612020-05-07 18:35:43 -07005177 if (IS_ERR(memcg))
5178 return ERR_CAST(memcg);
Li Zefan4219b2d2013-09-23 16:56:29 +08005179
Jakub Kicinskid1663a92020-06-01 21:49:49 -07005180 page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005181 memcg->soft_limit = PAGE_COUNTER_MAX;
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07005182 page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005183 if (parent) {
5184 memcg->swappiness = mem_cgroup_swappiness(parent);
5185 memcg->oom_kill_disable = parent->oom_kill_disable;
Roman Gushchinbef86202020-12-14 19:06:49 -08005186
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005187 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08005188 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005189 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08005190 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005191 } else {
Roman Gushchinbef86202020-12-14 19:06:49 -08005192 page_counter_init(&memcg->memory, NULL);
5193 page_counter_init(&memcg->swap, NULL);
5194 page_counter_init(&memcg->kmem, NULL);
5195 page_counter_init(&memcg->tcpmem, NULL);
Vladimir Davydovd6441632014-01-23 15:53:09 -08005196
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005197 root_mem_cgroup = memcg;
5198 return &memcg->css;
5199 }
5200
Roman Gushchinbef86202020-12-14 19:06:49 -08005201 /* The following stuff does not apply to the root */
Vladimir Davydovb313aee2016-03-17 14:18:27 -07005202 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005203 if (error)
5204 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08005205
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005206 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08005207 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005208
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005209 return &memcg->css;
5210fail:
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005211 mem_cgroup_id_remove(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005212 mem_cgroup_free(memcg);
Yafang Shao11d67612020-05-07 18:35:43 -07005213 return ERR_PTR(error);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005214}
5215
Johannes Weiner73f576c2016-07-20 15:44:57 -07005216static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005217{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07005218 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5219
Kirill Tkhai0a4465d2018-08-17 15:47:37 -07005220 /*
Yang Shie4262c42021-05-04 18:36:23 -07005221 * A memcg must be visible for expand_shrinker_info()
Kirill Tkhai0a4465d2018-08-17 15:47:37 -07005222 * by the time the maps are allocated. So, we allocate maps
5223 * here, when for_each_mem_cgroup() can't skip it.
5224 */
Yang Shie4262c42021-05-04 18:36:23 -07005225 if (alloc_shrinker_info(memcg)) {
Kirill Tkhai0a4465d2018-08-17 15:47:37 -07005226 mem_cgroup_id_remove(memcg);
5227 return -ENOMEM;
5228 }
5229
Johannes Weiner73f576c2016-07-20 15:44:57 -07005230 /* Online state pins memcg ID, memcg ID pins CSS */
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07005231 refcount_set(&memcg->id.ref, 1);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005232 css_get(css);
Shakeel Buttaa48e472021-09-02 14:55:04 -07005233
5234 if (unlikely(mem_cgroup_is_root(memcg)))
5235 queue_delayed_work(system_unbound_wq, &stats_flush_dwork,
5236 2UL*HZ);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07005237 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005238}
5239
Tejun Heoeb954192013-08-08 20:11:23 -04005240static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005241{
Tejun Heoeb954192013-08-08 20:11:23 -04005242 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05005243 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05005244
5245 /*
5246 * Unregister events and notify userspace.
5247 * Notify userspace about cgroup removing only after rmdir of cgroup
5248 * directory to avoid race between userspace and kernelspace.
5249 */
Shakeel Butt4ba95152021-09-02 14:56:05 -07005250 spin_lock_irq(&memcg->event_list_lock);
Tejun Heofba94802013-11-22 18:20:43 -05005251 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05005252 list_del_init(&event->list);
5253 schedule_work(&event->remove);
5254 }
Shakeel Butt4ba95152021-09-02 14:56:05 -07005255 spin_unlock_irq(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07005256
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07005257 page_counter_set_min(&memcg->memory, 0);
Roman Gushchin23067152018-06-07 17:06:22 -07005258 page_counter_set_low(&memcg->memory, 0);
Roman Gushchin63677c742017-09-06 16:21:47 -07005259
Johannes Weiner567e9ab2016-01-20 15:02:24 -08005260 memcg_offline_kmem(memcg);
Yang Shia1780152021-05-04 18:36:42 -07005261 reparent_shrinker_deferred(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04005262 wb_memcg_offline(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005263
Roman Gushchin591edfb2018-10-26 15:03:23 -07005264 drain_all_stock(memcg);
5265
Johannes Weiner73f576c2016-07-20 15:44:57 -07005266 mem_cgroup_id_put(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005267}
5268
Vladimir Davydov6df38682015-12-29 14:54:10 -08005269static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
5270{
5271 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5272
5273 invalidate_reclaim_iterators(memcg);
5274}
5275
Tejun Heoeb954192013-08-08 20:11:23 -04005276static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005277{
Tejun Heoeb954192013-08-08 20:11:23 -04005278 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo97b27822019-08-26 09:06:56 -07005279 int __maybe_unused i;
Daisuke Nishimurac268e992009-01-15 13:51:13 -08005280
Tejun Heo97b27822019-08-26 09:06:56 -07005281#ifdef CONFIG_CGROUP_WRITEBACK
5282 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++)
5283 wb_wait_for_completion(&memcg->cgwb_frn[i].done);
5284#endif
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005285 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08005286 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08005287
Johannes Weiner0db15292016-01-20 15:02:50 -08005288 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08005289 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08005290
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005291 vmpressure_cleanup(&memcg->vmpressure);
5292 cancel_work_sync(&memcg->high_work);
5293 mem_cgroup_remove_from_trees(memcg);
Yang Shie4262c42021-05-04 18:36:23 -07005294 free_shrinker_info(memcg);
Waiman Long38d4ef42021-11-05 13:37:37 -07005295
5296 /* Need to offline kmem if online_css() fails */
5297 memcg_offline_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005298 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005299}
5300
Tejun Heo1ced9532014-07-08 18:02:57 -04005301/**
5302 * mem_cgroup_css_reset - reset the states of a mem_cgroup
5303 * @css: the target css
5304 *
5305 * Reset the states of the mem_cgroup associated with @css. This is
5306 * invoked when the userland requests disabling on the default hierarchy
5307 * but the memcg is pinned through dependency. The memcg should stop
5308 * applying policies and should revert to the vanilla state as it may be
5309 * made visible again.
5310 *
5311 * The current implementation only resets the essential configurations.
5312 * This needs to be expanded to cover all the visible parts.
5313 */
5314static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
5315{
5316 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5317
Roman Gushchinbbec2e12018-06-07 17:06:18 -07005318 page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX);
5319 page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX);
Roman Gushchinbbec2e12018-06-07 17:06:18 -07005320 page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX);
5321 page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX);
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07005322 page_counter_set_min(&memcg->memory, 0);
Roman Gushchin23067152018-06-07 17:06:22 -07005323 page_counter_set_low(&memcg->memory, 0);
Jakub Kicinskid1663a92020-06-01 21:49:49 -07005324 page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX);
Johannes Weiner24d404d2015-01-08 14:32:35 -08005325 memcg->soft_limit = PAGE_COUNTER_MAX;
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07005326 page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX);
Tejun Heo2529bb32015-05-22 18:23:34 -04005327 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04005328}
5329
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005330static void mem_cgroup_css_rstat_flush(struct cgroup_subsys_state *css, int cpu)
5331{
5332 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5333 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
5334 struct memcg_vmstats_percpu *statc;
5335 long delta, v;
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07005336 int i, nid;
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005337
5338 statc = per_cpu_ptr(memcg->vmstats_percpu, cpu);
5339
5340 for (i = 0; i < MEMCG_NR_STAT; i++) {
5341 /*
5342 * Collect the aggregated propagation counts of groups
5343 * below us. We're in a per-cpu loop here and this is
5344 * a global counter, so the first cycle will get them.
5345 */
5346 delta = memcg->vmstats.state_pending[i];
5347 if (delta)
5348 memcg->vmstats.state_pending[i] = 0;
5349
5350 /* Add CPU changes on this level since the last flush */
5351 v = READ_ONCE(statc->state[i]);
5352 if (v != statc->state_prev[i]) {
5353 delta += v - statc->state_prev[i];
5354 statc->state_prev[i] = v;
5355 }
5356
5357 if (!delta)
5358 continue;
5359
5360 /* Aggregate counts on this level and propagate upwards */
5361 memcg->vmstats.state[i] += delta;
5362 if (parent)
5363 parent->vmstats.state_pending[i] += delta;
5364 }
5365
5366 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
5367 delta = memcg->vmstats.events_pending[i];
5368 if (delta)
5369 memcg->vmstats.events_pending[i] = 0;
5370
5371 v = READ_ONCE(statc->events[i]);
5372 if (v != statc->events_prev[i]) {
5373 delta += v - statc->events_prev[i];
5374 statc->events_prev[i] = v;
5375 }
5376
5377 if (!delta)
5378 continue;
5379
5380 memcg->vmstats.events[i] += delta;
5381 if (parent)
5382 parent->vmstats.events_pending[i] += delta;
5383 }
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07005384
5385 for_each_node_state(nid, N_MEMORY) {
5386 struct mem_cgroup_per_node *pn = memcg->nodeinfo[nid];
5387 struct mem_cgroup_per_node *ppn = NULL;
5388 struct lruvec_stats_percpu *lstatc;
5389
5390 if (parent)
5391 ppn = parent->nodeinfo[nid];
5392
5393 lstatc = per_cpu_ptr(pn->lruvec_stats_percpu, cpu);
5394
5395 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
5396 delta = pn->lruvec_stats.state_pending[i];
5397 if (delta)
5398 pn->lruvec_stats.state_pending[i] = 0;
5399
5400 v = READ_ONCE(lstatc->state[i]);
5401 if (v != lstatc->state_prev[i]) {
5402 delta += v - lstatc->state_prev[i];
5403 lstatc->state_prev[i] = v;
5404 }
5405
5406 if (!delta)
5407 continue;
5408
5409 pn->lruvec_stats.state[i] += delta;
5410 if (ppn)
5411 ppn->lruvec_stats.state_pending[i] += delta;
5412 }
5413 }
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005414}
5415
Daisuke Nishimura02491442010-03-10 15:22:17 -08005416#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005417/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005418static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005419{
Johannes Weiner05b84302014-08-06 16:05:59 -07005420 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07005421
Mel Gormand0164ad2015-11-06 16:28:21 -08005422 /* Try a single bulk charge without reclaim first, kswapd may wake */
5423 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07005424 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005425 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005426 return ret;
5427 }
Johannes Weiner9476db92014-08-06 16:05:55 -07005428
David Rientjes36745342017-01-24 15:18:10 -08005429 /* Try charges one by one with reclaim, but do not retry */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005430 while (count--) {
David Rientjes36745342017-01-24 15:18:10 -08005431 ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005432 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005433 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005434 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07005435 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005436 }
Johannes Weiner9476db92014-08-06 16:05:55 -07005437 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005438}
5439
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005440union mc_target {
5441 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005442 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005443};
5444
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005445enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005446 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005447 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005448 MC_TARGET_SWAP,
Jérôme Glissec733a822017-09-08 16:11:54 -07005449 MC_TARGET_DEVICE,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005450};
5451
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005452static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5453 unsigned long addr, pte_t ptent)
5454{
Christoph Hellwig25b29952019-06-13 22:50:49 +02005455 struct page *page = vm_normal_page(vma, addr, ptent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005456
5457 if (!page || !page_mapped(page))
5458 return NULL;
5459 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005460 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005461 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005462 } else {
5463 if (!(mc.flags & MOVE_FILE))
5464 return NULL;
5465 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005466 if (!get_page_unless_zero(page))
5467 return NULL;
5468
5469 return page;
5470}
5471
Jérôme Glissec733a822017-09-08 16:11:54 -07005472#if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005473static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07005474 pte_t ptent, swp_entry_t *entry)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005475{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005476 struct page *page = NULL;
5477 swp_entry_t ent = pte_to_swp_entry(ptent);
5478
Ralph Campbell9a137152020-10-13 16:53:13 -07005479 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005480 return NULL;
Jérôme Glissec733a822017-09-08 16:11:54 -07005481
5482 /*
5483 * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to
5484 * a device and because they are not accessible by CPU they are store
5485 * as special swap entry in the CPU page table.
5486 */
5487 if (is_device_private_entry(ent)) {
Alistair Poppleaf5cdaf2021-06-30 18:54:06 -07005488 page = pfn_swap_entry_to_page(ent);
Jérôme Glissec733a822017-09-08 16:11:54 -07005489 /*
5490 * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have
5491 * a refcount of 1 when free (unlike normal page)
5492 */
5493 if (!page_ref_add_unless(page, 1, 1))
5494 return NULL;
5495 return page;
5496 }
5497
Ralph Campbell9a137152020-10-13 16:53:13 -07005498 if (non_swap_entry(ent))
5499 return NULL;
5500
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005501 /*
5502 * Because lookup_swap_cache() updates some statistics counter,
5503 * we call find_get_page() with swapper_space directly.
5504 */
Huang Yingf6ab1f72016-10-07 17:00:21 -07005505 page = find_get_page(swap_address_space(ent), swp_offset(ent));
Johannes Weiner2d1c4982020-06-03 16:02:14 -07005506 entry->val = ent.val;
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005507
5508 return page;
5509}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005510#else
5511static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07005512 pte_t ptent, swp_entry_t *entry)
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005513{
5514 return NULL;
5515}
5516#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005517
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005518static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
Peter Xu48384b02021-11-05 13:37:28 -07005519 unsigned long addr, pte_t ptent)
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005520{
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005521 if (!vma->vm_file) /* anonymous vma */
5522 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005523 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005524 return NULL;
5525
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005526 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005527 /* shmem/tmpfs may report page out on swap: account for that too. */
Matthew Wilcox (Oracle)f5df8632020-10-13 16:51:21 -07005528 return find_get_incore_page(vma->vm_file->f_mapping,
5529 linear_page_index(vma, addr));
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005530}
5531
Chen Gangb1b0dea2015-04-14 15:47:35 -07005532/**
5533 * mem_cgroup_move_account - move account of the page
5534 * @page: the page
Li RongQing25843c22016-07-26 15:26:56 -07005535 * @compound: charge the page as compound or small page
Chen Gangb1b0dea2015-04-14 15:47:35 -07005536 * @from: mem_cgroup which the page is moved from.
5537 * @to: mem_cgroup which the page is moved to. @from != @to.
5538 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08005539 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07005540 *
5541 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
5542 * from old cgroup.
5543 */
5544static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005545 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07005546 struct mem_cgroup *from,
5547 struct mem_cgroup *to)
5548{
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005549 struct lruvec *from_vec, *to_vec;
5550 struct pglist_data *pgdat;
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07005551 unsigned int nr_pages = compound ? thp_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07005552 int ret;
5553
5554 VM_BUG_ON(from == to);
5555 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005556 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07005557
5558 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005559 * Prevent mem_cgroup_migrate() from looking at
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08005560 * page's memory cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07005561 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005562 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07005563 if (!trylock_page(page))
5564 goto out;
5565
5566 ret = -EINVAL;
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08005567 if (page_memcg(page) != from)
Chen Gangb1b0dea2015-04-14 15:47:35 -07005568 goto out_unlock;
5569
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005570 pgdat = page_pgdat(page);
Johannes Weiner867e5e12019-11-30 17:55:34 -08005571 from_vec = mem_cgroup_lruvec(from, pgdat);
5572 to_vec = mem_cgroup_lruvec(to, pgdat);
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005573
Johannes Weinerabb242f2020-06-03 16:01:28 -07005574 lock_page_memcg(page);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005575
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07005576 if (PageAnon(page)) {
5577 if (page_mapped(page)) {
5578 __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages);
5579 __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages);
Johannes Weiner468c3982020-06-03 16:02:01 -07005580 if (PageTransHuge(page)) {
Muchun Song69473e52021-02-24 12:03:23 -08005581 __mod_lruvec_state(from_vec, NR_ANON_THPS,
5582 -nr_pages);
5583 __mod_lruvec_state(to_vec, NR_ANON_THPS,
5584 nr_pages);
Johannes Weiner468c3982020-06-03 16:02:01 -07005585 }
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07005586 }
5587 } else {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07005588 __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages);
5589 __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages);
5590
5591 if (PageSwapBacked(page)) {
5592 __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages);
5593 __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages);
5594 }
5595
Johannes Weiner49e50d22020-06-03 16:01:47 -07005596 if (page_mapped(page)) {
5597 __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages);
5598 __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages);
5599 }
Chen Gangb1b0dea2015-04-14 15:47:35 -07005600
Johannes Weiner49e50d22020-06-03 16:01:47 -07005601 if (PageDirty(page)) {
5602 struct address_space *mapping = page_mapping(page);
Greg Thelenc4843a72015-05-22 17:13:16 -04005603
Christoph Hellwigf56753a2020-09-24 08:51:40 +02005604 if (mapping_can_writeback(mapping)) {
Johannes Weiner49e50d22020-06-03 16:01:47 -07005605 __mod_lruvec_state(from_vec, NR_FILE_DIRTY,
5606 -nr_pages);
5607 __mod_lruvec_state(to_vec, NR_FILE_DIRTY,
5608 nr_pages);
5609 }
Greg Thelenc4843a72015-05-22 17:13:16 -04005610 }
5611 }
5612
Chen Gangb1b0dea2015-04-14 15:47:35 -07005613 if (PageWriteback(page)) {
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005614 __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages);
5615 __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005616 }
5617
5618 /*
Johannes Weinerabb242f2020-06-03 16:01:28 -07005619 * All state has been migrated, let's switch to the new memcg.
5620 *
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08005621 * It is safe to change page's memcg here because the page
Johannes Weinerabb242f2020-06-03 16:01:28 -07005622 * is referenced, charged, isolated, and locked: we can't race
5623 * with (un)charging, migration, LRU putback, or anything else
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08005624 * that would rely on a stable page's memory cgroup.
Johannes Weinerabb242f2020-06-03 16:01:28 -07005625 *
5626 * Note that lock_page_memcg is a memcg lock, not a page lock,
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08005627 * to save space. As soon as we switch page's memory cgroup to a
Johannes Weinerabb242f2020-06-03 16:01:28 -07005628 * new memcg that isn't locked, the above state can change
5629 * concurrently again. Make sure we're truly done with it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07005630 */
Johannes Weinerabb242f2020-06-03 16:01:28 -07005631 smp_mb();
Chen Gangb1b0dea2015-04-14 15:47:35 -07005632
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07005633 css_get(&to->css);
5634 css_put(&from->css);
5635
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08005636 page->memcg_data = (unsigned long)to;
Yang Shi87eaceb2019-09-23 15:38:15 -07005637
Johannes Weinerabb242f2020-06-03 16:01:28 -07005638 __unlock_page_memcg(from);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005639
5640 ret = 0;
5641
5642 local_irq_disable();
Johannes Weiner3fba69a2020-06-03 16:01:31 -07005643 mem_cgroup_charge_statistics(to, page, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005644 memcg_check_events(to, page);
Johannes Weiner3fba69a2020-06-03 16:01:31 -07005645 mem_cgroup_charge_statistics(from, page, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005646 memcg_check_events(from, page);
5647 local_irq_enable();
5648out_unlock:
5649 unlock_page(page);
5650out:
5651 return ret;
5652}
5653
Li RongQing7cf78062016-05-27 14:27:46 -07005654/**
5655 * get_mctgt_type - get target type of moving charge
5656 * @vma: the vma the pte to be checked belongs
5657 * @addr: the address corresponding to the pte to be checked
5658 * @ptent: the pte to be checked
5659 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5660 *
5661 * Returns
5662 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5663 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5664 * move charge. if @target is not NULL, the page is stored in target->page
5665 * with extra refcnt got(Callers should handle it).
5666 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5667 * target for charge migration. if @target is not NULL, the entry is stored
5668 * in target->ent.
Christoph Hellwig25b29952019-06-13 22:50:49 +02005669 * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE
5670 * (so ZONE_DEVICE page and thus not on the lru).
Jérôme Glissedf6ad692017-09-08 16:12:24 -07005671 * For now we such page is charge like a regular page would be as for all
5672 * intent and purposes it is just special memory taking the place of a
5673 * regular page.
Jérôme Glissec733a822017-09-08 16:11:54 -07005674 *
5675 * See Documentations/vm/hmm.txt and include/linux/hmm.h
Li RongQing7cf78062016-05-27 14:27:46 -07005676 *
5677 * Called with pte lock held.
5678 */
5679
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005680static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005681 unsigned long addr, pte_t ptent, union mc_target *target)
5682{
Daisuke Nishimura02491442010-03-10 15:22:17 -08005683 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005684 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005685 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005686
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005687 if (pte_present(ptent))
5688 page = mc_handle_present_pte(vma, addr, ptent);
5689 else if (is_swap_pte(ptent))
Li RongQing48406ef2016-07-26 15:22:14 -07005690 page = mc_handle_swap_pte(vma, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08005691 else if (pte_none(ptent))
Peter Xu48384b02021-11-05 13:37:28 -07005692 page = mc_handle_file_pte(vma, addr, ptent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005693
5694 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005695 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005696 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005697 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005698 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08005699 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005700 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08005701 */
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08005702 if (page_memcg(page) == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005703 ret = MC_TARGET_PAGE;
Christoph Hellwig25b29952019-06-13 22:50:49 +02005704 if (is_device_private_page(page))
Jérôme Glissec733a822017-09-08 16:11:54 -07005705 ret = MC_TARGET_DEVICE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005706 if (target)
5707 target->page = page;
5708 }
5709 if (!ret || !target)
5710 put_page(page);
5711 }
Huang Ying3e14a572017-09-06 16:22:37 -07005712 /*
5713 * There is a swap entry and a page doesn't exist or isn't charged.
5714 * But we cannot move a tail-page in a THP.
5715 */
5716 if (ent.val && !ret && (!page || !PageTransCompound(page)) &&
Li Zefan34c00c32013-09-23 16:56:01 +08005717 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07005718 ret = MC_TARGET_SWAP;
5719 if (target)
5720 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005721 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005722 return ret;
5723}
5724
Naoya Horiguchi12724852012-03-21 16:34:28 -07005725#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5726/*
Huang Yingd6810d72017-09-06 16:22:45 -07005727 * We don't consider PMD mapped swapping or file mapped pages because THP does
5728 * not support them for now.
Naoya Horiguchi12724852012-03-21 16:34:28 -07005729 * Caller should make sure that pmd_trans_huge(pmd) is true.
5730 */
5731static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5732 unsigned long addr, pmd_t pmd, union mc_target *target)
5733{
5734 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005735 enum mc_target_type ret = MC_TARGET_NONE;
5736
Zi Yan84c3fc42017-09-08 16:11:01 -07005737 if (unlikely(is_swap_pmd(pmd))) {
5738 VM_BUG_ON(thp_migration_supported() &&
5739 !is_pmd_migration_entry(pmd));
5740 return ret;
5741 }
Naoya Horiguchi12724852012-03-21 16:34:28 -07005742 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08005743 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005744 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07005745 return ret;
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08005746 if (page_memcg(page) == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005747 ret = MC_TARGET_PAGE;
5748 if (target) {
5749 get_page(page);
5750 target->page = page;
5751 }
5752 }
5753 return ret;
5754}
5755#else
5756static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5757 unsigned long addr, pmd_t pmd, union mc_target *target)
5758{
5759 return MC_TARGET_NONE;
5760}
5761#endif
5762
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005763static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5764 unsigned long addr, unsigned long end,
5765 struct mm_walk *walk)
5766{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005767 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005768 pte_t *pte;
5769 spinlock_t *ptl;
5770
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08005771 ptl = pmd_trans_huge_lock(pmd, vma);
5772 if (ptl) {
Jérôme Glissec733a822017-09-08 16:11:54 -07005773 /*
5774 * Note their can not be MC_TARGET_DEVICE for now as we do not
Christoph Hellwig25b29952019-06-13 22:50:49 +02005775 * support transparent huge page with MEMORY_DEVICE_PRIVATE but
5776 * this might change.
Jérôme Glissec733a822017-09-08 16:11:54 -07005777 */
Naoya Horiguchi12724852012-03-21 16:34:28 -07005778 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
5779 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08005780 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005781 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005782 }
Dave Hansen03319322011-03-22 16:32:56 -07005783
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005784 if (pmd_trans_unstable(pmd))
5785 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005786 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5787 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005788 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005789 mc.precharge++; /* increment precharge temporarily */
5790 pte_unmap_unlock(pte - 1, ptl);
5791 cond_resched();
5792
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005793 return 0;
5794}
5795
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02005796static const struct mm_walk_ops precharge_walk_ops = {
5797 .pmd_entry = mem_cgroup_count_precharge_pte_range,
5798};
5799
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005800static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5801{
5802 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005803
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07005804 mmap_read_lock(mm);
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02005805 walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL);
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07005806 mmap_read_unlock(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005807
5808 precharge = mc.precharge;
5809 mc.precharge = 0;
5810
5811 return precharge;
5812}
5813
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005814static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5815{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005816 unsigned long precharge = mem_cgroup_count_precharge(mm);
5817
5818 VM_BUG_ON(mc.moving_task);
5819 mc.moving_task = current;
5820 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005821}
5822
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005823/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5824static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005825{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005826 struct mem_cgroup *from = mc.from;
5827 struct mem_cgroup *to = mc.to;
5828
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005829 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005830 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005831 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005832 mc.precharge = 0;
5833 }
5834 /*
5835 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5836 * we must uncharge here.
5837 */
5838 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005839 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005840 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005841 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005842 /* we must fixup refcnts and charges */
5843 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005844 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04005845 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005846 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005847
Vladimir Davydov615d66c2016-08-11 15:33:03 -07005848 mem_cgroup_id_put_many(mc.from, mc.moved_swap);
5849
Johannes Weiner05b84302014-08-06 16:05:59 -07005850 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005851 * we charged both to->memory and to->memsw, so we
5852 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07005853 */
Johannes Weinerce00a962014-09-05 08:43:57 -04005854 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005855 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005856
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005857 mc.moved_swap = 0;
5858 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005859 memcg_oom_recover(from);
5860 memcg_oom_recover(to);
5861 wake_up_all(&mc.waitq);
5862}
5863
5864static void mem_cgroup_clear_mc(void)
5865{
Tejun Heo264a0ae2016-04-21 19:09:02 -04005866 struct mm_struct *mm = mc.mm;
5867
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005868 /*
5869 * we must clear moving_task before waking up waiters at the end of
5870 * task migration.
5871 */
5872 mc.moving_task = NULL;
5873 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005874 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005875 mc.from = NULL;
5876 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04005877 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005878 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04005879
5880 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005881}
5882
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005883static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005884{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005885 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07005886 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07005887 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04005888 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07005889 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005890 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07005891 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005892
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005893 /* charge immigration isn't supported on the default hierarchy */
5894 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07005895 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005896
Tejun Heo4530edd2015-09-11 15:00:19 -04005897 /*
5898 * Multi-process migrations only happen on the default hierarchy
5899 * where charge immigration is not used. Perform charge
5900 * immigration if @tset contains a leader and whine if there are
5901 * multiple.
5902 */
5903 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005904 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04005905 WARN_ON_ONCE(p);
5906 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005907 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04005908 }
5909 if (!p)
5910 return 0;
5911
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005912 /*
Ingo Molnarf0953a12021-05-06 18:06:47 -07005913 * We are now committed to this value whatever it is. Changes in this
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005914 * tunable will only affect upcoming migrations, not the current one.
5915 * So we need to save it, and keep it going.
5916 */
5917 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
5918 if (!move_flags)
5919 return 0;
5920
Tejun Heo9f2115f2015-09-08 15:01:10 -07005921 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005922
Tejun Heo9f2115f2015-09-08 15:01:10 -07005923 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08005924
Tejun Heo9f2115f2015-09-08 15:01:10 -07005925 mm = get_task_mm(p);
5926 if (!mm)
5927 return 0;
5928 /* We move charges only when we move a owner of the mm */
5929 if (mm->owner == p) {
5930 VM_BUG_ON(mc.from);
5931 VM_BUG_ON(mc.to);
5932 VM_BUG_ON(mc.precharge);
5933 VM_BUG_ON(mc.moved_charge);
5934 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005935
Tejun Heo9f2115f2015-09-08 15:01:10 -07005936 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04005937 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07005938 mc.from = from;
5939 mc.to = memcg;
5940 mc.flags = move_flags;
5941 spin_unlock(&mc.lock);
5942 /* We set mc.moving_task later */
5943
5944 ret = mem_cgroup_precharge_mc(mm);
5945 if (ret)
5946 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04005947 } else {
5948 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005949 }
5950 return ret;
5951}
5952
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005953static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005954{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08005955 if (mc.to)
5956 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005957}
5958
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005959static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5960 unsigned long addr, unsigned long end,
5961 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005962{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005963 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005964 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005965 pte_t *pte;
5966 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005967 enum mc_target_type target_type;
5968 union mc_target target;
5969 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005970
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08005971 ptl = pmd_trans_huge_lock(pmd, vma);
5972 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07005973 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08005974 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07005975 return 0;
5976 }
5977 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
5978 if (target_type == MC_TARGET_PAGE) {
5979 page = target.page;
5980 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005981 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08005982 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005983 mc.precharge -= HPAGE_PMD_NR;
5984 mc.moved_charge += HPAGE_PMD_NR;
5985 }
5986 putback_lru_page(page);
5987 }
5988 put_page(page);
Jérôme Glissec733a822017-09-08 16:11:54 -07005989 } else if (target_type == MC_TARGET_DEVICE) {
5990 page = target.page;
5991 if (!mem_cgroup_move_account(page, true,
5992 mc.from, mc.to)) {
5993 mc.precharge -= HPAGE_PMD_NR;
5994 mc.moved_charge += HPAGE_PMD_NR;
5995 }
5996 put_page(page);
Naoya Horiguchi12724852012-03-21 16:34:28 -07005997 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08005998 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005999 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006000 }
6001
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006002 if (pmd_trans_unstable(pmd))
6003 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006004retry:
6005 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6006 for (; addr != end; addr += PAGE_SIZE) {
6007 pte_t ptent = *(pte++);
Jérôme Glissec733a822017-09-08 16:11:54 -07006008 bool device = false;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006009 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006010
6011 if (!mc.precharge)
6012 break;
6013
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006014 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Jérôme Glissec733a822017-09-08 16:11:54 -07006015 case MC_TARGET_DEVICE:
6016 device = true;
Joe Perchese4a9bc52020-04-06 20:08:39 -07006017 fallthrough;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006018 case MC_TARGET_PAGE:
6019 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08006020 /*
6021 * We can have a part of the split pmd here. Moving it
6022 * can be done but it would be too convoluted so simply
6023 * ignore such a partial THP and keep it in original
6024 * memcg. There should be somebody mapping the head.
6025 */
6026 if (PageTransCompound(page))
6027 goto put;
Jérôme Glissec733a822017-09-08 16:11:54 -07006028 if (!device && isolate_lru_page(page))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006029 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08006030 if (!mem_cgroup_move_account(page, false,
6031 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006032 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006033 /* we uncharge from mc.from later. */
6034 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006035 }
Jérôme Glissec733a822017-09-08 16:11:54 -07006036 if (!device)
6037 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006038put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006039 put_page(page);
6040 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006041 case MC_TARGET_SWAP:
6042 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07006043 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08006044 mc.precharge--;
Hugh Dickins8d22a932020-07-23 21:15:24 -07006045 mem_cgroup_id_get_many(mc.to, 1);
6046 /* we fixup other refcnts and charges later. */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006047 mc.moved_swap++;
6048 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08006049 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006050 default:
6051 break;
6052 }
6053 }
6054 pte_unmap_unlock(pte - 1, ptl);
6055 cond_resched();
6056
6057 if (addr != end) {
6058 /*
6059 * We have consumed all precharges we got in can_attach().
6060 * We try charge one by one, but don't do any additional
6061 * charges to mc.to if we have failed in charge once in attach()
6062 * phase.
6063 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006064 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006065 if (!ret)
6066 goto retry;
6067 }
6068
6069 return ret;
6070}
6071
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02006072static const struct mm_walk_ops charge_walk_ops = {
6073 .pmd_entry = mem_cgroup_move_charge_pte_range,
6074};
6075
Tejun Heo264a0ae2016-04-21 19:09:02 -04006076static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006077{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006078 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08006079 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07006080 * Signal lock_page_memcg() to take the memcg's move_lock
6081 * while we're moving its pages to another memcg. Then wait
6082 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08006083 */
6084 atomic_inc(&mc.from->moving_account);
6085 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006086retry:
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07006087 if (unlikely(!mmap_read_trylock(mc.mm))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006088 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07006089 * Someone who are holding the mmap_lock might be waiting in
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006090 * waitq. So we cancel all extra charges, wake up all waiters,
6091 * and retry. Because we cancel precharges, we might not be able
6092 * to move enough charges, but moving charge is a best-effort
6093 * feature anyway, so it wouldn't be a big problem.
6094 */
6095 __mem_cgroup_clear_mc();
6096 cond_resched();
6097 goto retry;
6098 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08006099 /*
6100 * When we have consumed all precharges and failed in doing
6101 * additional charge, the page walk just aborts.
6102 */
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02006103 walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops,
6104 NULL);
James Morse0247f3f2016-10-07 17:00:12 -07006105
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07006106 mmap_read_unlock(mc.mm);
Johannes Weiner312722c2014-12-10 15:44:25 -08006107 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006108}
6109
Tejun Heo264a0ae2016-04-21 19:09:02 -04006110static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08006111{
Tejun Heo264a0ae2016-04-21 19:09:02 -04006112 if (mc.to) {
6113 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006114 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04006115 }
Balbir Singh67e465a2008-02-07 00:13:54 -08006116}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006117#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006118static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006119{
6120 return 0;
6121}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006122static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006123{
6124}
Tejun Heo264a0ae2016-04-21 19:09:02 -04006125static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006126{
6127}
6128#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08006129
Chris Down677dc972019-03-05 15:45:55 -08006130static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value)
6131{
6132 if (value == PAGE_COUNTER_MAX)
6133 seq_puts(m, "max\n");
6134 else
6135 seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE);
6136
6137 return 0;
6138}
6139
Johannes Weiner241994ed2015-02-11 15:26:06 -08006140static u64 memory_current_read(struct cgroup_subsys_state *css,
6141 struct cftype *cft)
6142{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08006143 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6144
6145 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006146}
6147
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006148static int memory_min_show(struct seq_file *m, void *v)
6149{
Chris Down677dc972019-03-05 15:45:55 -08006150 return seq_puts_memcg_tunable(m,
6151 READ_ONCE(mem_cgroup_from_seq(m)->memory.min));
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006152}
6153
6154static ssize_t memory_min_write(struct kernfs_open_file *of,
6155 char *buf, size_t nbytes, loff_t off)
6156{
6157 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6158 unsigned long min;
6159 int err;
6160
6161 buf = strstrip(buf);
6162 err = page_counter_memparse(buf, "max", &min);
6163 if (err)
6164 return err;
6165
6166 page_counter_set_min(&memcg->memory, min);
6167
6168 return nbytes;
6169}
6170
Johannes Weiner241994ed2015-02-11 15:26:06 -08006171static int memory_low_show(struct seq_file *m, void *v)
6172{
Chris Down677dc972019-03-05 15:45:55 -08006173 return seq_puts_memcg_tunable(m,
6174 READ_ONCE(mem_cgroup_from_seq(m)->memory.low));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006175}
6176
6177static ssize_t memory_low_write(struct kernfs_open_file *of,
6178 char *buf, size_t nbytes, loff_t off)
6179{
6180 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6181 unsigned long low;
6182 int err;
6183
6184 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08006185 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006186 if (err)
6187 return err;
6188
Roman Gushchin23067152018-06-07 17:06:22 -07006189 page_counter_set_low(&memcg->memory, low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006190
6191 return nbytes;
6192}
6193
6194static int memory_high_show(struct seq_file *m, void *v)
6195{
Jakub Kicinskid1663a92020-06-01 21:49:49 -07006196 return seq_puts_memcg_tunable(m,
6197 READ_ONCE(mem_cgroup_from_seq(m)->memory.high));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006198}
6199
6200static ssize_t memory_high_write(struct kernfs_open_file *of,
6201 char *buf, size_t nbytes, loff_t off)
6202{
6203 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Chris Downd977aa92020-08-06 23:21:58 -07006204 unsigned int nr_retries = MAX_RECLAIM_RETRIES;
Johannes Weiner8c8c3832019-11-30 17:50:09 -08006205 bool drained = false;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006206 unsigned long high;
6207 int err;
6208
6209 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08006210 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006211 if (err)
6212 return err;
6213
Johannes Weinere82553c2021-02-09 13:42:28 -08006214 page_counter_set_high(&memcg->memory, high);
6215
Johannes Weiner8c8c3832019-11-30 17:50:09 -08006216 for (;;) {
6217 unsigned long nr_pages = page_counter_read(&memcg->memory);
6218 unsigned long reclaimed;
Johannes Weiner588083b2016-03-17 14:20:25 -07006219
Johannes Weiner8c8c3832019-11-30 17:50:09 -08006220 if (nr_pages <= high)
6221 break;
6222
6223 if (signal_pending(current))
6224 break;
6225
6226 if (!drained) {
6227 drain_all_stock(memcg);
6228 drained = true;
6229 continue;
6230 }
6231
6232 reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
6233 GFP_KERNEL, true);
6234
6235 if (!reclaimed && !nr_retries--)
6236 break;
6237 }
6238
Johannes Weiner19ce33a2020-08-06 23:22:12 -07006239 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006240 return nbytes;
6241}
6242
6243static int memory_max_show(struct seq_file *m, void *v)
6244{
Chris Down677dc972019-03-05 15:45:55 -08006245 return seq_puts_memcg_tunable(m,
6246 READ_ONCE(mem_cgroup_from_seq(m)->memory.max));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006247}
6248
6249static ssize_t memory_max_write(struct kernfs_open_file *of,
6250 char *buf, size_t nbytes, loff_t off)
6251{
6252 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Chris Downd977aa92020-08-06 23:21:58 -07006253 unsigned int nr_reclaims = MAX_RECLAIM_RETRIES;
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006254 bool drained = false;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006255 unsigned long max;
6256 int err;
6257
6258 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08006259 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006260 if (err)
6261 return err;
6262
Roman Gushchinbbec2e12018-06-07 17:06:18 -07006263 xchg(&memcg->memory.max, max);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006264
6265 for (;;) {
6266 unsigned long nr_pages = page_counter_read(&memcg->memory);
6267
6268 if (nr_pages <= max)
6269 break;
6270
Johannes Weiner7249c9f2019-11-30 17:50:06 -08006271 if (signal_pending(current))
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006272 break;
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006273
6274 if (!drained) {
6275 drain_all_stock(memcg);
6276 drained = true;
6277 continue;
6278 }
6279
6280 if (nr_reclaims) {
6281 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
6282 GFP_KERNEL, true))
6283 nr_reclaims--;
6284 continue;
6285 }
6286
Johannes Weinere27be242018-04-10 16:29:45 -07006287 memcg_memory_event(memcg, MEMCG_OOM);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006288 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
6289 break;
6290 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08006291
Tejun Heo2529bb32015-05-22 18:23:34 -04006292 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006293 return nbytes;
6294}
6295
Shakeel Butt1e577f92019-07-11 20:55:55 -07006296static void __memory_events_show(struct seq_file *m, atomic_long_t *events)
6297{
6298 seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW]));
6299 seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH]));
6300 seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX]));
6301 seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM]));
6302 seq_printf(m, "oom_kill %lu\n",
6303 atomic_long_read(&events[MEMCG_OOM_KILL]));
6304}
6305
Johannes Weiner241994ed2015-02-11 15:26:06 -08006306static int memory_events_show(struct seq_file *m, void *v)
6307{
Chris Downaa9694b2019-03-05 15:45:52 -08006308 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006309
Shakeel Butt1e577f92019-07-11 20:55:55 -07006310 __memory_events_show(m, memcg->memory_events);
6311 return 0;
6312}
Johannes Weiner241994ed2015-02-11 15:26:06 -08006313
Shakeel Butt1e577f92019-07-11 20:55:55 -07006314static int memory_events_local_show(struct seq_file *m, void *v)
6315{
6316 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
6317
6318 __memory_events_show(m, memcg->memory_events_local);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006319 return 0;
6320}
6321
Johannes Weiner587d9f72016-01-20 15:03:19 -08006322static int memory_stat_show(struct seq_file *m, void *v)
6323{
Chris Downaa9694b2019-03-05 15:45:52 -08006324 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Johannes Weinerc8713d02019-07-11 20:55:59 -07006325 char *buf;
Johannes Weiner587d9f72016-01-20 15:03:19 -08006326
Johannes Weinerc8713d02019-07-11 20:55:59 -07006327 buf = memory_stat_format(memcg);
6328 if (!buf)
6329 return -ENOMEM;
6330 seq_puts(m, buf);
6331 kfree(buf);
Johannes Weiner587d9f72016-01-20 15:03:19 -08006332 return 0;
6333}
6334
Muchun Song5f9a4f42020-10-13 16:52:59 -07006335#ifdef CONFIG_NUMA
Muchun Songfff66b72021-02-24 12:03:43 -08006336static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec,
6337 int item)
6338{
6339 return lruvec_page_state(lruvec, item) * memcg_page_state_unit(item);
6340}
6341
Muchun Song5f9a4f42020-10-13 16:52:59 -07006342static int memory_numa_stat_show(struct seq_file *m, void *v)
6343{
6344 int i;
6345 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
6346
Shakeel Buttfd25a9e2021-11-05 13:37:34 -07006347 mem_cgroup_flush_stats();
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07006348
Muchun Song5f9a4f42020-10-13 16:52:59 -07006349 for (i = 0; i < ARRAY_SIZE(memory_stats); i++) {
6350 int nid;
6351
6352 if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS)
6353 continue;
6354
6355 seq_printf(m, "%s", memory_stats[i].name);
6356 for_each_node_state(nid, N_MEMORY) {
6357 u64 size;
6358 struct lruvec *lruvec;
6359
6360 lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid));
Muchun Songfff66b72021-02-24 12:03:43 -08006361 size = lruvec_page_state_output(lruvec,
6362 memory_stats[i].idx);
Muchun Song5f9a4f42020-10-13 16:52:59 -07006363 seq_printf(m, " N%d=%llu", nid, size);
6364 }
6365 seq_putc(m, '\n');
6366 }
6367
6368 return 0;
6369}
6370#endif
6371
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006372static int memory_oom_group_show(struct seq_file *m, void *v)
6373{
Chris Downaa9694b2019-03-05 15:45:52 -08006374 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006375
6376 seq_printf(m, "%d\n", memcg->oom_group);
6377
6378 return 0;
6379}
6380
6381static ssize_t memory_oom_group_write(struct kernfs_open_file *of,
6382 char *buf, size_t nbytes, loff_t off)
6383{
6384 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6385 int ret, oom_group;
6386
6387 buf = strstrip(buf);
6388 if (!buf)
6389 return -EINVAL;
6390
6391 ret = kstrtoint(buf, 0, &oom_group);
6392 if (ret)
6393 return ret;
6394
6395 if (oom_group != 0 && oom_group != 1)
6396 return -EINVAL;
6397
6398 memcg->oom_group = oom_group;
6399
6400 return nbytes;
6401}
6402
Johannes Weiner241994ed2015-02-11 15:26:06 -08006403static struct cftype memory_files[] = {
6404 {
6405 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08006406 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994ed2015-02-11 15:26:06 -08006407 .read_u64 = memory_current_read,
6408 },
6409 {
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006410 .name = "min",
6411 .flags = CFTYPE_NOT_ON_ROOT,
6412 .seq_show = memory_min_show,
6413 .write = memory_min_write,
6414 },
6415 {
Johannes Weiner241994ed2015-02-11 15:26:06 -08006416 .name = "low",
6417 .flags = CFTYPE_NOT_ON_ROOT,
6418 .seq_show = memory_low_show,
6419 .write = memory_low_write,
6420 },
6421 {
6422 .name = "high",
6423 .flags = CFTYPE_NOT_ON_ROOT,
6424 .seq_show = memory_high_show,
6425 .write = memory_high_write,
6426 },
6427 {
6428 .name = "max",
6429 .flags = CFTYPE_NOT_ON_ROOT,
6430 .seq_show = memory_max_show,
6431 .write = memory_max_write,
6432 },
6433 {
6434 .name = "events",
6435 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04006436 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994ed2015-02-11 15:26:06 -08006437 .seq_show = memory_events_show,
6438 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08006439 {
Shakeel Butt1e577f92019-07-11 20:55:55 -07006440 .name = "events.local",
6441 .flags = CFTYPE_NOT_ON_ROOT,
6442 .file_offset = offsetof(struct mem_cgroup, events_local_file),
6443 .seq_show = memory_events_local_show,
6444 },
6445 {
Johannes Weiner587d9f72016-01-20 15:03:19 -08006446 .name = "stat",
Johannes Weiner587d9f72016-01-20 15:03:19 -08006447 .seq_show = memory_stat_show,
6448 },
Muchun Song5f9a4f42020-10-13 16:52:59 -07006449#ifdef CONFIG_NUMA
6450 {
6451 .name = "numa_stat",
6452 .seq_show = memory_numa_stat_show,
6453 },
6454#endif
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006455 {
6456 .name = "oom.group",
6457 .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE,
6458 .seq_show = memory_oom_group_show,
6459 .write = memory_oom_group_write,
6460 },
Johannes Weiner241994ed2015-02-11 15:26:06 -08006461 { } /* terminate */
6462};
6463
Tejun Heo073219e2014-02-08 10:36:58 -05006464struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08006465 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08006466 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08006467 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08006468 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08006469 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04006470 .css_reset = mem_cgroup_css_reset,
Johannes Weiner2d146aa2021-04-29 22:56:26 -07006471 .css_rstat_flush = mem_cgroup_css_rstat_flush,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006472 .can_attach = mem_cgroup_can_attach,
6473 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04006474 .post_attach = mem_cgroup_move_task,
Johannes Weiner241994ed2015-02-11 15:26:06 -08006475 .dfl_cftypes = memory_files,
6476 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006477 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006478};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006479
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006480/*
6481 * This function calculates an individual cgroup's effective
6482 * protection which is derived from its own memory.min/low, its
6483 * parent's and siblings' settings, as well as the actual memory
6484 * distribution in the tree.
6485 *
6486 * The following rules apply to the effective protection values:
6487 *
6488 * 1. At the first level of reclaim, effective protection is equal to
6489 * the declared protection in memory.min and memory.low.
6490 *
6491 * 2. To enable safe delegation of the protection configuration, at
6492 * subsequent levels the effective protection is capped to the
6493 * parent's effective protection.
6494 *
6495 * 3. To make complex and dynamic subtrees easier to configure, the
6496 * user is allowed to overcommit the declared protection at a given
6497 * level. If that is the case, the parent's effective protection is
6498 * distributed to the children in proportion to how much protection
6499 * they have declared and how much of it they are utilizing.
6500 *
6501 * This makes distribution proportional, but also work-conserving:
6502 * if one cgroup claims much more protection than it uses memory,
6503 * the unused remainder is available to its siblings.
6504 *
6505 * 4. Conversely, when the declared protection is undercommitted at a
6506 * given level, the distribution of the larger parental protection
6507 * budget is NOT proportional. A cgroup's protection from a sibling
6508 * is capped to its own memory.min/low setting.
6509 *
Johannes Weiner8a931f82020-04-01 21:07:07 -07006510 * 5. However, to allow protecting recursive subtrees from each other
6511 * without having to declare each individual cgroup's fixed share
6512 * of the ancestor's claim to protection, any unutilized -
6513 * "floating" - protection from up the tree is distributed in
6514 * proportion to each cgroup's *usage*. This makes the protection
6515 * neutral wrt sibling cgroups and lets them compete freely over
6516 * the shared parental protection budget, but it protects the
6517 * subtree as a whole from neighboring subtrees.
6518 *
6519 * Note that 4. and 5. are not in conflict: 4. is about protecting
6520 * against immediate siblings whereas 5. is about protecting against
6521 * neighboring subtrees.
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006522 */
6523static unsigned long effective_protection(unsigned long usage,
Johannes Weiner8a931f82020-04-01 21:07:07 -07006524 unsigned long parent_usage,
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006525 unsigned long setting,
6526 unsigned long parent_effective,
6527 unsigned long siblings_protected)
6528{
6529 unsigned long protected;
Johannes Weiner8a931f82020-04-01 21:07:07 -07006530 unsigned long ep;
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006531
6532 protected = min(usage, setting);
6533 /*
6534 * If all cgroups at this level combined claim and use more
6535 * protection then what the parent affords them, distribute
6536 * shares in proportion to utilization.
6537 *
6538 * We are using actual utilization rather than the statically
6539 * claimed protection in order to be work-conserving: claimed
6540 * but unused protection is available to siblings that would
6541 * otherwise get a smaller chunk than what they claimed.
6542 */
6543 if (siblings_protected > parent_effective)
6544 return protected * parent_effective / siblings_protected;
6545
6546 /*
6547 * Ok, utilized protection of all children is within what the
6548 * parent affords them, so we know whatever this child claims
6549 * and utilizes is effectively protected.
6550 *
6551 * If there is unprotected usage beyond this value, reclaim
6552 * will apply pressure in proportion to that amount.
6553 *
6554 * If there is unutilized protection, the cgroup will be fully
6555 * shielded from reclaim, but we do return a smaller value for
6556 * protection than what the group could enjoy in theory. This
6557 * is okay. With the overcommit distribution above, effective
6558 * protection is always dependent on how memory is actually
6559 * consumed among the siblings anyway.
6560 */
Johannes Weiner8a931f82020-04-01 21:07:07 -07006561 ep = protected;
6562
6563 /*
6564 * If the children aren't claiming (all of) the protection
6565 * afforded to them by the parent, distribute the remainder in
6566 * proportion to the (unprotected) memory of each cgroup. That
6567 * way, cgroups that aren't explicitly prioritized wrt each
6568 * other compete freely over the allowance, but they are
6569 * collectively protected from neighboring trees.
6570 *
6571 * We're using unprotected memory for the weight so that if
6572 * some cgroups DO claim explicit protection, we don't protect
6573 * the same bytes twice.
Johannes Weinercd324ed2020-06-25 20:30:16 -07006574 *
6575 * Check both usage and parent_usage against the respective
6576 * protected values. One should imply the other, but they
6577 * aren't read atomically - make sure the division is sane.
Johannes Weiner8a931f82020-04-01 21:07:07 -07006578 */
6579 if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT))
6580 return ep;
Johannes Weinercd324ed2020-06-25 20:30:16 -07006581 if (parent_effective > siblings_protected &&
6582 parent_usage > siblings_protected &&
6583 usage > protected) {
Johannes Weiner8a931f82020-04-01 21:07:07 -07006584 unsigned long unclaimed;
6585
6586 unclaimed = parent_effective - siblings_protected;
6587 unclaimed *= usage - protected;
6588 unclaimed /= parent_usage - siblings_protected;
6589
6590 ep += unclaimed;
6591 }
6592
6593 return ep;
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006594}
6595
Johannes Weiner241994ed2015-02-11 15:26:06 -08006596/**
Mel Gorman05395712021-06-30 18:53:32 -07006597 * mem_cgroup_calculate_protection - check if memory consumption is in the normal range
Sean Christopherson34c81052017-07-10 15:48:05 -07006598 * @root: the top ancestor of the sub-tree being checked
Johannes Weiner241994ed2015-02-11 15:26:06 -08006599 * @memcg: the memory cgroup to check
6600 *
Roman Gushchin23067152018-06-07 17:06:22 -07006601 * WARNING: This function is not stateless! It can only be used as part
6602 * of a top-down tree iteration, not for isolated queries.
Johannes Weiner241994ed2015-02-11 15:26:06 -08006603 */
Chris Down45c7f7e2020-08-06 23:22:05 -07006604void mem_cgroup_calculate_protection(struct mem_cgroup *root,
6605 struct mem_cgroup *memcg)
Johannes Weiner241994ed2015-02-11 15:26:06 -08006606{
Johannes Weiner8a931f82020-04-01 21:07:07 -07006607 unsigned long usage, parent_usage;
Roman Gushchin23067152018-06-07 17:06:22 -07006608 struct mem_cgroup *parent;
6609
Johannes Weiner241994ed2015-02-11 15:26:06 -08006610 if (mem_cgroup_disabled())
Chris Down45c7f7e2020-08-06 23:22:05 -07006611 return;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006612
Sean Christopherson34c81052017-07-10 15:48:05 -07006613 if (!root)
6614 root = root_mem_cgroup;
Yafang Shao22f74962020-08-06 23:22:01 -07006615
6616 /*
6617 * Effective values of the reclaim targets are ignored so they
6618 * can be stale. Have a look at mem_cgroup_protection for more
6619 * details.
6620 * TODO: calculation should be more robust so that we do not need
6621 * that special casing.
6622 */
Sean Christopherson34c81052017-07-10 15:48:05 -07006623 if (memcg == root)
Chris Down45c7f7e2020-08-06 23:22:05 -07006624 return;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006625
Roman Gushchin23067152018-06-07 17:06:22 -07006626 usage = page_counter_read(&memcg->memory);
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006627 if (!usage)
Chris Down45c7f7e2020-08-06 23:22:05 -07006628 return;
Sean Christopherson34c81052017-07-10 15:48:05 -07006629
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006630 parent = parent_mem_cgroup(memcg);
Roman Gushchindf2a4192018-06-14 15:26:17 -07006631 /* No parent means a non-hierarchical mode on v1 memcg */
6632 if (!parent)
Chris Down45c7f7e2020-08-06 23:22:05 -07006633 return;
Roman Gushchindf2a4192018-06-14 15:26:17 -07006634
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006635 if (parent == root) {
Chris Downc3d53202020-04-01 21:07:27 -07006636 memcg->memory.emin = READ_ONCE(memcg->memory.min);
Chris Down03960e32020-06-25 20:30:22 -07006637 memcg->memory.elow = READ_ONCE(memcg->memory.low);
Chris Down45c7f7e2020-08-06 23:22:05 -07006638 return;
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006639 }
6640
Johannes Weiner8a931f82020-04-01 21:07:07 -07006641 parent_usage = page_counter_read(&parent->memory);
6642
Chris Downb3a78222020-04-01 21:07:33 -07006643 WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage,
Chris Downc3d53202020-04-01 21:07:27 -07006644 READ_ONCE(memcg->memory.min),
6645 READ_ONCE(parent->memory.emin),
Chris Downb3a78222020-04-01 21:07:33 -07006646 atomic_long_read(&parent->memory.children_min_usage)));
Roman Gushchin23067152018-06-07 17:06:22 -07006647
Chris Downb3a78222020-04-01 21:07:33 -07006648 WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage,
Chris Down03960e32020-06-25 20:30:22 -07006649 READ_ONCE(memcg->memory.low),
6650 READ_ONCE(parent->memory.elow),
Chris Downb3a78222020-04-01 21:07:33 -07006651 atomic_long_read(&parent->memory.children_low_usage)));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006652}
6653
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07006654static int charge_memcg(struct page *page, struct mem_cgroup *memcg, gfp_t gfp)
Shakeel Butt0add0c72021-04-29 22:56:36 -07006655{
6656 unsigned int nr_pages = thp_nr_pages(page);
6657 int ret;
6658
6659 ret = try_charge(memcg, gfp, nr_pages);
6660 if (ret)
6661 goto out;
6662
6663 css_get(&memcg->css);
6664 commit_charge(page, memcg);
6665
6666 local_irq_disable();
6667 mem_cgroup_charge_statistics(memcg, page, nr_pages);
6668 memcg_check_events(memcg, page);
6669 local_irq_enable();
6670out:
6671 return ret;
6672}
6673
Johannes Weiner00501b52014-08-08 14:19:20 -07006674/**
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07006675 * __mem_cgroup_charge - charge a newly allocated page to a cgroup
Johannes Weiner00501b52014-08-08 14:19:20 -07006676 * @page: page to charge
6677 * @mm: mm context of the victim
6678 * @gfp_mask: reclaim mode
Johannes Weiner00501b52014-08-08 14:19:20 -07006679 *
6680 * Try to charge @page to the memcg that @mm belongs to, reclaiming
Dan Schatzberg04f94e32021-06-28 19:38:18 -07006681 * pages according to @gfp_mask if necessary. if @mm is NULL, try to
6682 * charge to the active memcg.
Johannes Weiner00501b52014-08-08 14:19:20 -07006683 *
Shakeel Butt0add0c72021-04-29 22:56:36 -07006684 * Do not use this for pages allocated for swapin.
6685 *
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07006686 * Returns 0 on success. Otherwise, an error code is returned.
Johannes Weiner00501b52014-08-08 14:19:20 -07006687 */
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07006688int __mem_cgroup_charge(struct page *page, struct mm_struct *mm,
6689 gfp_t gfp_mask)
Johannes Weiner00501b52014-08-08 14:19:20 -07006690{
Shakeel Butt0add0c72021-04-29 22:56:36 -07006691 struct mem_cgroup *memcg;
6692 int ret;
Johannes Weiner00501b52014-08-08 14:19:20 -07006693
Shakeel Butt0add0c72021-04-29 22:56:36 -07006694 memcg = get_mem_cgroup_from_mm(mm);
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07006695 ret = charge_memcg(page, memcg, gfp_mask);
Shakeel Butt0add0c72021-04-29 22:56:36 -07006696 css_put(&memcg->css);
Johannes Weiner2d1c4982020-06-03 16:02:14 -07006697
Shakeel Butt0add0c72021-04-29 22:56:36 -07006698 return ret;
6699}
Vladimir Davydove993d902015-09-09 15:35:35 -07006700
Shakeel Butt0add0c72021-04-29 22:56:36 -07006701/**
6702 * mem_cgroup_swapin_charge_page - charge a newly allocated page for swapin
6703 * @page: page to charge
6704 * @mm: mm context of the victim
6705 * @gfp: reclaim mode
6706 * @entry: swap entry for which the page is allocated
6707 *
6708 * This function charges a page allocated for swapin. Please call this before
6709 * adding the page to the swapcache.
6710 *
6711 * Returns 0 on success. Otherwise, an error code is returned.
6712 */
6713int mem_cgroup_swapin_charge_page(struct page *page, struct mm_struct *mm,
6714 gfp_t gfp, swp_entry_t entry)
6715{
6716 struct mem_cgroup *memcg;
6717 unsigned short id;
6718 int ret;
Johannes Weiner00501b52014-08-08 14:19:20 -07006719
Shakeel Butt0add0c72021-04-29 22:56:36 -07006720 if (mem_cgroup_disabled())
6721 return 0;
6722
6723 id = lookup_swap_cgroup_id(entry);
6724 rcu_read_lock();
6725 memcg = mem_cgroup_from_id(id);
6726 if (!memcg || !css_tryget_online(&memcg->css))
Johannes Weiner00501b52014-08-08 14:19:20 -07006727 memcg = get_mem_cgroup_from_mm(mm);
Shakeel Butt0add0c72021-04-29 22:56:36 -07006728 rcu_read_unlock();
Johannes Weiner00501b52014-08-08 14:19:20 -07006729
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07006730 ret = charge_memcg(page, memcg, gfp);
Johannes Weiner00501b52014-08-08 14:19:20 -07006731
Shakeel Butt0add0c72021-04-29 22:56:36 -07006732 css_put(&memcg->css);
6733 return ret;
6734}
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006735
Shakeel Butt0add0c72021-04-29 22:56:36 -07006736/*
6737 * mem_cgroup_swapin_uncharge_swap - uncharge swap slot
6738 * @entry: swap entry for which the page is charged
6739 *
6740 * Call this function after successfully adding the charged page to swapcache.
6741 *
6742 * Note: This function assumes the page for which swap slot is being uncharged
6743 * is order 0 page.
6744 */
6745void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry)
6746{
Muchun Songcae3af62021-02-24 12:04:19 -08006747 /*
6748 * Cgroup1's unified memory+swap counter has been charged with the
6749 * new swapcache page, finish the transfer by uncharging the swap
6750 * slot. The swap slot would also get uncharged when it dies, but
6751 * it can stick around indefinitely and we'd count the page twice
6752 * the entire time.
6753 *
6754 * Cgroup2 has separate resource counters for memory and swap,
6755 * so this is a non-issue here. Memory and swap charge lifetimes
6756 * correspond 1:1 to page and swap slot lifetimes: we charge the
6757 * page to memory here, and uncharge swap when the slot is freed.
6758 */
Shakeel Butt0add0c72021-04-29 22:56:36 -07006759 if (!mem_cgroup_disabled() && do_memsw_account()) {
Johannes Weiner00501b52014-08-08 14:19:20 -07006760 /*
6761 * The swap entry might not get freed for a long time,
6762 * let's not wait for it. The page already received a
6763 * memory+swap charge, drop the swap entry duplicate.
6764 */
Shakeel Butt0add0c72021-04-29 22:56:36 -07006765 mem_cgroup_uncharge_swap(entry, 1);
Johannes Weiner00501b52014-08-08 14:19:20 -07006766 }
Johannes Weiner3fea5a42020-06-03 16:01:41 -07006767}
6768
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006769struct uncharge_gather {
6770 struct mem_cgroup *memcg;
Muchun Songb4e0b682021-04-29 22:56:52 -07006771 unsigned long nr_memory;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006772 unsigned long pgpgout;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006773 unsigned long nr_kmem;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006774 struct page *dummy_page;
6775};
6776
6777static inline void uncharge_gather_clear(struct uncharge_gather *ug)
Johannes Weiner747db952014-08-08 14:19:24 -07006778{
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006779 memset(ug, 0, sizeof(*ug));
6780}
6781
6782static void uncharge_batch(const struct uncharge_gather *ug)
6783{
Johannes Weiner747db952014-08-08 14:19:24 -07006784 unsigned long flags;
6785
Muchun Songb4e0b682021-04-29 22:56:52 -07006786 if (ug->nr_memory) {
6787 page_counter_uncharge(&ug->memcg->memory, ug->nr_memory);
Johannes Weiner7941d212016-01-14 15:21:23 -08006788 if (do_memsw_account())
Muchun Songb4e0b682021-04-29 22:56:52 -07006789 page_counter_uncharge(&ug->memcg->memsw, ug->nr_memory);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006790 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem)
6791 page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem);
6792 memcg_oom_recover(ug->memcg);
Johannes Weinerce00a962014-09-05 08:43:57 -04006793 }
Johannes Weiner747db952014-08-08 14:19:24 -07006794
6795 local_irq_save(flags);
Johannes Weinerc9019e92018-01-31 16:16:37 -08006796 __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout);
Muchun Songb4e0b682021-04-29 22:56:52 -07006797 __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_memory);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006798 memcg_check_events(ug->memcg, ug->dummy_page);
Johannes Weiner747db952014-08-08 14:19:24 -07006799 local_irq_restore(flags);
Michal Hockof1796542020-09-04 16:35:24 -07006800
6801 /* drop reference from uncharge_page */
6802 css_put(&ug->memcg->css);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006803}
6804
6805static void uncharge_page(struct page *page, struct uncharge_gather *ug)
6806{
Johannes Weiner9f762db2020-06-03 16:01:44 -07006807 unsigned long nr_pages;
Muchun Songb4e0b682021-04-29 22:56:52 -07006808 struct mem_cgroup *memcg;
6809 struct obj_cgroup *objcg;
Waiman Long55927112021-06-28 19:37:30 -07006810 bool use_objcg = PageMemcgKmem(page);
Johannes Weiner9f762db2020-06-03 16:01:44 -07006811
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006812 VM_BUG_ON_PAGE(PageLRU(page), page);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006813
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006814 /*
6815 * Nobody should be changing or seriously looking at
Muchun Songb4e0b682021-04-29 22:56:52 -07006816 * page memcg or objcg at this point, we have fully
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006817 * exclusive access to the page.
6818 */
Waiman Long55927112021-06-28 19:37:30 -07006819 if (use_objcg) {
Muchun Songb4e0b682021-04-29 22:56:52 -07006820 objcg = __page_objcg(page);
6821 /*
6822 * This get matches the put at the end of the function and
6823 * kmem pages do not hold memcg references anymore.
6824 */
6825 memcg = get_mem_cgroup_from_objcg(objcg);
6826 } else {
6827 memcg = __page_memcg(page);
6828 }
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006829
Muchun Songb4e0b682021-04-29 22:56:52 -07006830 if (!memcg)
6831 return;
6832
6833 if (ug->memcg != memcg) {
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006834 if (ug->memcg) {
6835 uncharge_batch(ug);
6836 uncharge_gather_clear(ug);
6837 }
Muchun Songb4e0b682021-04-29 22:56:52 -07006838 ug->memcg = memcg;
Muchun Song7ab345a2021-04-29 22:56:48 -07006839 ug->dummy_page = page;
Michal Hockof1796542020-09-04 16:35:24 -07006840
6841 /* pairs with css_put in uncharge_batch */
Muchun Songb4e0b682021-04-29 22:56:52 -07006842 css_get(&memcg->css);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006843 }
6844
Johannes Weiner9f762db2020-06-03 16:01:44 -07006845 nr_pages = compound_nr(page);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006846
Waiman Long55927112021-06-28 19:37:30 -07006847 if (use_objcg) {
Muchun Songb4e0b682021-04-29 22:56:52 -07006848 ug->nr_memory += nr_pages;
Johannes Weiner9f762db2020-06-03 16:01:44 -07006849 ug->nr_kmem += nr_pages;
Muchun Songb4e0b682021-04-29 22:56:52 -07006850
6851 page->memcg_data = 0;
6852 obj_cgroup_put(objcg);
6853 } else {
6854 /* LRU pages aren't accounted at the root level */
6855 if (!mem_cgroup_is_root(memcg))
6856 ug->nr_memory += nr_pages;
Roman Gushchin18b2db32020-12-01 13:58:30 -08006857 ug->pgpgout++;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006858
Muchun Songb4e0b682021-04-29 22:56:52 -07006859 page->memcg_data = 0;
6860 }
6861
6862 css_put(&memcg->css);
Johannes Weiner747db952014-08-08 14:19:24 -07006863}
6864
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006865/**
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07006866 * __mem_cgroup_uncharge - uncharge a page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006867 * @page: page to uncharge
6868 *
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07006869 * Uncharge a page previously charged with __mem_cgroup_charge().
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006870 */
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07006871void __mem_cgroup_uncharge(struct page *page)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006872{
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006873 struct uncharge_gather ug;
6874
Johannes Weiner747db952014-08-08 14:19:24 -07006875 /* Don't touch page->lru of any random page, pre-check: */
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08006876 if (!page_memcg(page))
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006877 return;
6878
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006879 uncharge_gather_clear(&ug);
6880 uncharge_page(page, &ug);
6881 uncharge_batch(&ug);
Johannes Weiner747db952014-08-08 14:19:24 -07006882}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006883
Johannes Weiner747db952014-08-08 14:19:24 -07006884/**
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07006885 * __mem_cgroup_uncharge_list - uncharge a list of page
Johannes Weiner747db952014-08-08 14:19:24 -07006886 * @page_list: list of pages to uncharge
6887 *
6888 * Uncharge a list of pages previously charged with
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07006889 * __mem_cgroup_charge().
Johannes Weiner747db952014-08-08 14:19:24 -07006890 */
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07006891void __mem_cgroup_uncharge_list(struct list_head *page_list)
Johannes Weiner747db952014-08-08 14:19:24 -07006892{
Muchun Songc41a40b2021-02-24 12:04:08 -08006893 struct uncharge_gather ug;
6894 struct page *page;
6895
Muchun Songc41a40b2021-02-24 12:04:08 -08006896 uncharge_gather_clear(&ug);
6897 list_for_each_entry(page, page_list, lru)
6898 uncharge_page(page, &ug);
6899 if (ug.memcg)
6900 uncharge_batch(&ug);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006901}
6902
6903/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07006904 * mem_cgroup_migrate - charge a page's replacement
6905 * @oldpage: currently circulating page
6906 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006907 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07006908 * Charge @newpage as a replacement page for @oldpage. @oldpage will
6909 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006910 *
6911 * Both pages must be locked, @newpage->mapping must be set up.
6912 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07006913void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006914{
Johannes Weiner29833312014-12-10 15:44:02 -08006915 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08006916 unsigned int nr_pages;
Tejun Heod93c4132016-06-24 14:49:54 -07006917 unsigned long flags;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006918
6919 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
6920 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006921 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006922 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
6923 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006924
6925 if (mem_cgroup_disabled())
6926 return;
6927
6928 /* Page cache replacement: new page already charged? */
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08006929 if (page_memcg(newpage))
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006930 return;
6931
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08006932 memcg = page_memcg(oldpage);
Alex Shia4055882020-12-18 14:01:31 -08006933 VM_WARN_ON_ONCE_PAGE(!memcg, oldpage);
Johannes Weiner29833312014-12-10 15:44:02 -08006934 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006935 return;
6936
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08006937 /* Force-charge the new page. The old one will be freed soon */
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07006938 nr_pages = thp_nr_pages(newpage);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08006939
Muchun Song8dc87c72021-06-28 19:37:47 -07006940 if (!mem_cgroup_is_root(memcg)) {
6941 page_counter_charge(&memcg->memory, nr_pages);
6942 if (do_memsw_account())
6943 page_counter_charge(&memcg->memsw, nr_pages);
6944 }
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006945
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07006946 css_get(&memcg->css);
Johannes Weinerd9eb1ea2020-06-03 16:02:24 -07006947 commit_charge(newpage, memcg);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08006948
Tejun Heod93c4132016-06-24 14:49:54 -07006949 local_irq_save(flags);
Johannes Weiner3fba69a2020-06-03 16:01:31 -07006950 mem_cgroup_charge_statistics(memcg, newpage, nr_pages);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08006951 memcg_check_events(memcg, newpage);
Tejun Heod93c4132016-06-24 14:49:54 -07006952 local_irq_restore(flags);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006953}
6954
Johannes Weineref129472016-01-14 15:21:34 -08006955DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08006956EXPORT_SYMBOL(memcg_sockets_enabled_key);
6957
Johannes Weiner2d758072016-10-07 17:00:58 -07006958void mem_cgroup_sk_alloc(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08006959{
6960 struct mem_cgroup *memcg;
6961
Johannes Weiner2d758072016-10-07 17:00:58 -07006962 if (!mem_cgroup_sockets_enabled)
6963 return;
6964
Shakeel Butte876ecc2020-03-09 22:16:05 -07006965 /* Do not associate the sock with unrelated interrupted task's memcg. */
6966 if (in_interrupt())
6967 return;
6968
Johannes Weiner11092082016-01-14 15:21:26 -08006969 rcu_read_lock();
6970 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006971 if (memcg == root_mem_cgroup)
6972 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08006973 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006974 goto out;
Shakeel Butt8965aa22020-04-01 21:07:10 -07006975 if (css_tryget(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08006976 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006977out:
Johannes Weiner11092082016-01-14 15:21:26 -08006978 rcu_read_unlock();
6979}
Johannes Weiner11092082016-01-14 15:21:26 -08006980
Johannes Weiner2d758072016-10-07 17:00:58 -07006981void mem_cgroup_sk_free(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08006982{
Johannes Weiner2d758072016-10-07 17:00:58 -07006983 if (sk->sk_memcg)
6984 css_put(&sk->sk_memcg->css);
Johannes Weiner11092082016-01-14 15:21:26 -08006985}
6986
6987/**
6988 * mem_cgroup_charge_skmem - charge socket memory
6989 * @memcg: memcg to charge
6990 * @nr_pages: number of pages to charge
Wei Wang4b1327b2021-08-17 12:40:03 -07006991 * @gfp_mask: reclaim mode
Johannes Weiner11092082016-01-14 15:21:26 -08006992 *
6993 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
Wei Wang4b1327b2021-08-17 12:40:03 -07006994 * @memcg's configured limit, %false if it doesn't.
Johannes Weiner11092082016-01-14 15:21:26 -08006995 */
Wei Wang4b1327b2021-08-17 12:40:03 -07006996bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages,
6997 gfp_t gfp_mask)
Johannes Weiner11092082016-01-14 15:21:26 -08006998{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08006999 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08007000 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007001
Johannes Weiner0db15292016-01-20 15:02:50 -08007002 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
7003 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007004 return true;
7005 }
Johannes Weiner0db15292016-01-20 15:02:50 -08007006 memcg->tcpmem_pressure = 1;
Wei Wang4b1327b2021-08-17 12:40:03 -07007007 if (gfp_mask & __GFP_NOFAIL) {
7008 page_counter_charge(&memcg->tcpmem, nr_pages);
7009 return true;
7010 }
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007011 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08007012 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08007013
Wei Wang4b1327b2021-08-17 12:40:03 -07007014 if (try_charge(memcg, gfp_mask, nr_pages) == 0) {
7015 mod_memcg_state(memcg, MEMCG_SOCK, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007016 return true;
Wei Wang4b1327b2021-08-17 12:40:03 -07007017 }
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007018
Johannes Weiner11092082016-01-14 15:21:26 -08007019 return false;
7020}
7021
7022/**
7023 * mem_cgroup_uncharge_skmem - uncharge socket memory
Mike Rapoportb7701a52018-02-06 15:42:13 -08007024 * @memcg: memcg to uncharge
7025 * @nr_pages: number of pages to uncharge
Johannes Weiner11092082016-01-14 15:21:26 -08007026 */
7027void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
7028{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007029 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08007030 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007031 return;
7032 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08007033
Johannes Weinerc9019e92018-01-31 16:16:37 -08007034 mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages);
Johannes Weinerb2807f02016-01-20 15:03:22 -08007035
Roman Gushchin475d0482017-09-08 16:13:09 -07007036 refill_stock(memcg, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08007037}
7038
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007039static int __init cgroup_memory(char *s)
7040{
7041 char *token;
7042
7043 while ((token = strsep(&s, ",")) != NULL) {
7044 if (!*token)
7045 continue;
7046 if (!strcmp(token, "nosocket"))
7047 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08007048 if (!strcmp(token, "nokmem"))
7049 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007050 }
7051 return 0;
7052}
7053__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08007054
Michal Hocko2d110852013-02-22 16:34:43 -08007055/*
Michal Hocko10813122013-02-22 16:35:41 -08007056 * subsys_initcall() for memory controller.
7057 *
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01007058 * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this
7059 * context because of lock dependencies (cgroup_lock -> cpu hotplug) but
7060 * basically everything that doesn't depend on a specific mem_cgroup structure
7061 * should be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08007062 */
7063static int __init mem_cgroup_init(void)
7064{
Johannes Weiner95a045f2015-02-11 15:26:33 -08007065 int cpu, node;
7066
Muchun Songf3344ad2021-02-24 12:03:15 -08007067 /*
7068 * Currently s32 type (can refer to struct batched_lruvec_stat) is
7069 * used for per-memcg-per-cpu caching of per-node statistics. In order
7070 * to work fine, we should make sure that the overfill threshold can't
7071 * exceed S32_MAX / PAGE_SIZE.
7072 */
7073 BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE);
7074
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01007075 cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL,
7076 memcg_hotplug_cpu_dead);
Johannes Weiner95a045f2015-02-11 15:26:33 -08007077
7078 for_each_possible_cpu(cpu)
7079 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
7080 drain_local_stock);
7081
7082 for_each_node(node) {
7083 struct mem_cgroup_tree_per_node *rtpn;
Johannes Weiner95a045f2015-02-11 15:26:33 -08007084
7085 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
7086 node_online(node) ? node : NUMA_NO_NODE);
7087
Mel Gormanef8f2322016-07-28 15:46:05 -07007088 rtpn->rb_root = RB_ROOT;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -07007089 rtpn->rb_rightmost = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -07007090 spin_lock_init(&rtpn->lock);
Johannes Weiner95a045f2015-02-11 15:26:33 -08007091 soft_limit_tree.rb_tree_per_node[node] = rtpn;
7092 }
7093
Michal Hocko2d110852013-02-22 16:34:43 -08007094 return 0;
7095}
7096subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08007097
7098#ifdef CONFIG_MEMCG_SWAP
Arnd Bergmann358c07f2016-08-25 15:17:08 -07007099static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
7100{
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07007101 while (!refcount_inc_not_zero(&memcg->id.ref)) {
Arnd Bergmann358c07f2016-08-25 15:17:08 -07007102 /*
7103 * The root cgroup cannot be destroyed, so it's refcount must
7104 * always be >= 1.
7105 */
7106 if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
7107 VM_BUG_ON(1);
7108 break;
7109 }
7110 memcg = parent_mem_cgroup(memcg);
7111 if (!memcg)
7112 memcg = root_mem_cgroup;
7113 }
7114 return memcg;
7115}
7116
Johannes Weiner21afa382015-02-11 15:26:36 -08007117/**
7118 * mem_cgroup_swapout - transfer a memsw charge to swap
7119 * @page: page whose memsw charge to transfer
7120 * @entry: swap entry to move the charge to
7121 *
7122 * Transfer the memsw charge of @page to @entry.
7123 */
7124void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
7125{
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007126 struct mem_cgroup *memcg, *swap_memcg;
Huang Yingd6810d72017-09-06 16:22:45 -07007127 unsigned int nr_entries;
Johannes Weiner21afa382015-02-11 15:26:36 -08007128 unsigned short oldid;
7129
7130 VM_BUG_ON_PAGE(PageLRU(page), page);
7131 VM_BUG_ON_PAGE(page_count(page), page);
7132
Alex Shi76358ab2020-12-18 14:01:28 -08007133 if (mem_cgroup_disabled())
7134 return;
7135
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007136 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Johannes Weiner21afa382015-02-11 15:26:36 -08007137 return;
7138
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08007139 memcg = page_memcg(page);
Johannes Weiner21afa382015-02-11 15:26:36 -08007140
Alex Shia4055882020-12-18 14:01:31 -08007141 VM_WARN_ON_ONCE_PAGE(!memcg, page);
Johannes Weiner21afa382015-02-11 15:26:36 -08007142 if (!memcg)
7143 return;
7144
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007145 /*
7146 * In case the memcg owning these pages has been offlined and doesn't
7147 * have an ID allocated to it anymore, charge the closest online
7148 * ancestor for the swap instead and transfer the memory+swap charge.
7149 */
7150 swap_memcg = mem_cgroup_id_get_online(memcg);
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07007151 nr_entries = thp_nr_pages(page);
Huang Yingd6810d72017-09-06 16:22:45 -07007152 /* Get references for the tail pages, too */
7153 if (nr_entries > 1)
7154 mem_cgroup_id_get_many(swap_memcg, nr_entries - 1);
7155 oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg),
7156 nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007157 VM_BUG_ON_PAGE(oldid, page);
Johannes Weinerc9019e92018-01-31 16:16:37 -08007158 mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007159
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08007160 page->memcg_data = 0;
Johannes Weiner21afa382015-02-11 15:26:36 -08007161
7162 if (!mem_cgroup_is_root(memcg))
Huang Yingd6810d72017-09-06 16:22:45 -07007163 page_counter_uncharge(&memcg->memory, nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007164
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007165 if (!cgroup_memory_noswap && memcg != swap_memcg) {
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007166 if (!mem_cgroup_is_root(swap_memcg))
Huang Yingd6810d72017-09-06 16:22:45 -07007167 page_counter_charge(&swap_memcg->memsw, nr_entries);
7168 page_counter_uncharge(&memcg->memsw, nr_entries);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007169 }
7170
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07007171 /*
7172 * Interrupts should be disabled here because the caller holds the
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07007173 * i_pages lock which is taken with interrupts-off. It is
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07007174 * important here to have the interrupts disabled because it is the
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07007175 * only synchronisation we have for updating the per-CPU variables.
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07007176 */
7177 VM_BUG_ON(!irqs_disabled());
Johannes Weiner3fba69a2020-06-03 16:01:31 -07007178 mem_cgroup_charge_statistics(memcg, page, -nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007179 memcg_check_events(memcg, page);
Johannes Weiner73f576c2016-07-20 15:44:57 -07007180
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07007181 css_put(&memcg->css);
Johannes Weiner21afa382015-02-11 15:26:36 -08007182}
7183
Huang Ying38d8b4e2017-07-06 15:37:18 -07007184/**
Suren Baghdasaryan01c4b282021-09-02 14:54:54 -07007185 * __mem_cgroup_try_charge_swap - try charging swap space for a page
Vladimir Davydov37e84352016-01-20 15:02:56 -08007186 * @page: page being added to swap
7187 * @entry: swap entry to charge
7188 *
Huang Ying38d8b4e2017-07-06 15:37:18 -07007189 * Try to charge @page's memcg for the swap space at @entry.
Vladimir Davydov37e84352016-01-20 15:02:56 -08007190 *
7191 * Returns 0 on success, -ENOMEM on failure.
7192 */
Suren Baghdasaryan01c4b282021-09-02 14:54:54 -07007193int __mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
Vladimir Davydov37e84352016-01-20 15:02:56 -08007194{
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07007195 unsigned int nr_pages = thp_nr_pages(page);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007196 struct page_counter *counter;
Huang Ying38d8b4e2017-07-06 15:37:18 -07007197 struct mem_cgroup *memcg;
Vladimir Davydov37e84352016-01-20 15:02:56 -08007198 unsigned short oldid;
7199
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007200 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov37e84352016-01-20 15:02:56 -08007201 return 0;
7202
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08007203 memcg = page_memcg(page);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007204
Alex Shia4055882020-12-18 14:01:31 -08007205 VM_WARN_ON_ONCE_PAGE(!memcg, page);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007206 if (!memcg)
7207 return 0;
7208
Tejun Heof3a53a32018-06-07 17:05:35 -07007209 if (!entry.val) {
7210 memcg_memory_event(memcg, MEMCG_SWAP_FAIL);
Tejun Heobb98f2c2018-06-07 17:05:31 -07007211 return 0;
Tejun Heof3a53a32018-06-07 17:05:35 -07007212 }
Tejun Heobb98f2c2018-06-07 17:05:31 -07007213
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007214 memcg = mem_cgroup_id_get_online(memcg);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007215
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007216 if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) &&
Huang Ying38d8b4e2017-07-06 15:37:18 -07007217 !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) {
Tejun Heof3a53a32018-06-07 17:05:35 -07007218 memcg_memory_event(memcg, MEMCG_SWAP_MAX);
7219 memcg_memory_event(memcg, MEMCG_SWAP_FAIL);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007220 mem_cgroup_id_put(memcg);
7221 return -ENOMEM;
7222 }
7223
Huang Ying38d8b4e2017-07-06 15:37:18 -07007224 /* Get references for the tail pages, too */
7225 if (nr_pages > 1)
7226 mem_cgroup_id_get_many(memcg, nr_pages - 1);
7227 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007228 VM_BUG_ON_PAGE(oldid, page);
Johannes Weinerc9019e92018-01-31 16:16:37 -08007229 mod_memcg_state(memcg, MEMCG_SWAP, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007230
Vladimir Davydov37e84352016-01-20 15:02:56 -08007231 return 0;
7232}
7233
Johannes Weiner21afa382015-02-11 15:26:36 -08007234/**
Suren Baghdasaryan01c4b282021-09-02 14:54:54 -07007235 * __mem_cgroup_uncharge_swap - uncharge swap space
Johannes Weiner21afa382015-02-11 15:26:36 -08007236 * @entry: swap entry to uncharge
Huang Ying38d8b4e2017-07-06 15:37:18 -07007237 * @nr_pages: the amount of swap space to uncharge
Johannes Weiner21afa382015-02-11 15:26:36 -08007238 */
Suren Baghdasaryan01c4b282021-09-02 14:54:54 -07007239void __mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages)
Johannes Weiner21afa382015-02-11 15:26:36 -08007240{
7241 struct mem_cgroup *memcg;
7242 unsigned short id;
7243
Huang Ying38d8b4e2017-07-06 15:37:18 -07007244 id = swap_cgroup_record(entry, 0, nr_pages);
Johannes Weiner21afa382015-02-11 15:26:36 -08007245 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07007246 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08007247 if (memcg) {
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007248 if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08007249 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Huang Ying38d8b4e2017-07-06 15:37:18 -07007250 page_counter_uncharge(&memcg->swap, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007251 else
Huang Ying38d8b4e2017-07-06 15:37:18 -07007252 page_counter_uncharge(&memcg->memsw, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007253 }
Johannes Weinerc9019e92018-01-31 16:16:37 -08007254 mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages);
Huang Ying38d8b4e2017-07-06 15:37:18 -07007255 mem_cgroup_id_put_many(memcg, nr_pages);
Johannes Weiner21afa382015-02-11 15:26:36 -08007256 }
7257 rcu_read_unlock();
7258}
7259
Vladimir Davydovd8b38432016-01-20 15:03:07 -08007260long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
7261{
7262 long nr_swap_pages = get_nr_swap_pages();
7263
Johannes Weinereccb52e2020-06-03 16:02:11 -07007264 if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydovd8b38432016-01-20 15:03:07 -08007265 return nr_swap_pages;
7266 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
7267 nr_swap_pages = min_t(long, nr_swap_pages,
Roman Gushchinbbec2e12018-06-07 17:06:18 -07007268 READ_ONCE(memcg->swap.max) -
Vladimir Davydovd8b38432016-01-20 15:03:07 -08007269 page_counter_read(&memcg->swap));
7270 return nr_swap_pages;
7271}
7272
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007273bool mem_cgroup_swap_full(struct page *page)
7274{
7275 struct mem_cgroup *memcg;
7276
7277 VM_BUG_ON_PAGE(!PageLocked(page), page);
7278
7279 if (vm_swap_full())
7280 return true;
Johannes Weinereccb52e2020-06-03 16:02:11 -07007281 if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007282 return false;
7283
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08007284 memcg = page_memcg(page);
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007285 if (!memcg)
7286 return false;
7287
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007288 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) {
7289 unsigned long usage = page_counter_read(&memcg->swap);
7290
7291 if (usage * 2 >= READ_ONCE(memcg->swap.high) ||
7292 usage * 2 >= READ_ONCE(memcg->swap.max))
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007293 return true;
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007294 }
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007295
7296 return false;
7297}
7298
Johannes Weinereccb52e2020-06-03 16:02:11 -07007299static int __init setup_swap_account(char *s)
Johannes Weiner21afa382015-02-11 15:26:36 -08007300{
7301 if (!strcmp(s, "1"))
Kaixu Xia5ab92902020-12-14 19:07:07 -08007302 cgroup_memory_noswap = false;
Johannes Weiner21afa382015-02-11 15:26:36 -08007303 else if (!strcmp(s, "0"))
Kaixu Xia5ab92902020-12-14 19:07:07 -08007304 cgroup_memory_noswap = true;
Johannes Weiner21afa382015-02-11 15:26:36 -08007305 return 1;
7306}
Johannes Weinereccb52e2020-06-03 16:02:11 -07007307__setup("swapaccount=", setup_swap_account);
Johannes Weiner21afa382015-02-11 15:26:36 -08007308
Vladimir Davydov37e84352016-01-20 15:02:56 -08007309static u64 swap_current_read(struct cgroup_subsys_state *css,
7310 struct cftype *cft)
7311{
7312 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
7313
7314 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
7315}
7316
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007317static int swap_high_show(struct seq_file *m, void *v)
7318{
7319 return seq_puts_memcg_tunable(m,
7320 READ_ONCE(mem_cgroup_from_seq(m)->swap.high));
7321}
7322
7323static ssize_t swap_high_write(struct kernfs_open_file *of,
7324 char *buf, size_t nbytes, loff_t off)
7325{
7326 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
7327 unsigned long high;
7328 int err;
7329
7330 buf = strstrip(buf);
7331 err = page_counter_memparse(buf, "max", &high);
7332 if (err)
7333 return err;
7334
7335 page_counter_set_high(&memcg->swap, high);
7336
7337 return nbytes;
7338}
7339
Vladimir Davydov37e84352016-01-20 15:02:56 -08007340static int swap_max_show(struct seq_file *m, void *v)
7341{
Chris Down677dc972019-03-05 15:45:55 -08007342 return seq_puts_memcg_tunable(m,
7343 READ_ONCE(mem_cgroup_from_seq(m)->swap.max));
Vladimir Davydov37e84352016-01-20 15:02:56 -08007344}
7345
7346static ssize_t swap_max_write(struct kernfs_open_file *of,
7347 char *buf, size_t nbytes, loff_t off)
7348{
7349 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
7350 unsigned long max;
7351 int err;
7352
7353 buf = strstrip(buf);
7354 err = page_counter_memparse(buf, "max", &max);
7355 if (err)
7356 return err;
7357
Tejun Heobe091022018-06-07 17:09:21 -07007358 xchg(&memcg->swap.max, max);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007359
7360 return nbytes;
7361}
7362
Tejun Heof3a53a32018-06-07 17:05:35 -07007363static int swap_events_show(struct seq_file *m, void *v)
7364{
Chris Downaa9694b2019-03-05 15:45:52 -08007365 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Tejun Heof3a53a32018-06-07 17:05:35 -07007366
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007367 seq_printf(m, "high %lu\n",
7368 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH]));
Tejun Heof3a53a32018-06-07 17:05:35 -07007369 seq_printf(m, "max %lu\n",
7370 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX]));
7371 seq_printf(m, "fail %lu\n",
7372 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL]));
7373
7374 return 0;
7375}
7376
Vladimir Davydov37e84352016-01-20 15:02:56 -08007377static struct cftype swap_files[] = {
7378 {
7379 .name = "swap.current",
7380 .flags = CFTYPE_NOT_ON_ROOT,
7381 .read_u64 = swap_current_read,
7382 },
7383 {
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007384 .name = "swap.high",
7385 .flags = CFTYPE_NOT_ON_ROOT,
7386 .seq_show = swap_high_show,
7387 .write = swap_high_write,
7388 },
7389 {
Vladimir Davydov37e84352016-01-20 15:02:56 -08007390 .name = "swap.max",
7391 .flags = CFTYPE_NOT_ON_ROOT,
7392 .seq_show = swap_max_show,
7393 .write = swap_max_write,
7394 },
Tejun Heof3a53a32018-06-07 17:05:35 -07007395 {
7396 .name = "swap.events",
7397 .flags = CFTYPE_NOT_ON_ROOT,
7398 .file_offset = offsetof(struct mem_cgroup, swap_events_file),
7399 .seq_show = swap_events_show,
7400 },
Vladimir Davydov37e84352016-01-20 15:02:56 -08007401 { } /* terminate */
7402};
7403
Johannes Weinereccb52e2020-06-03 16:02:11 -07007404static struct cftype memsw_files[] = {
Johannes Weiner21afa382015-02-11 15:26:36 -08007405 {
7406 .name = "memsw.usage_in_bytes",
7407 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
7408 .read_u64 = mem_cgroup_read_u64,
7409 },
7410 {
7411 .name = "memsw.max_usage_in_bytes",
7412 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
7413 .write = mem_cgroup_reset,
7414 .read_u64 = mem_cgroup_read_u64,
7415 },
7416 {
7417 .name = "memsw.limit_in_bytes",
7418 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
7419 .write = mem_cgroup_write,
7420 .read_u64 = mem_cgroup_read_u64,
7421 },
7422 {
7423 .name = "memsw.failcnt",
7424 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
7425 .write = mem_cgroup_reset,
7426 .read_u64 = mem_cgroup_read_u64,
7427 },
7428 { }, /* terminate */
7429};
7430
Bhupesh Sharma82ff1652020-07-23 21:15:21 -07007431/*
7432 * If mem_cgroup_swap_init() is implemented as a subsys_initcall()
7433 * instead of a core_initcall(), this could mean cgroup_memory_noswap still
7434 * remains set to false even when memcg is disabled via "cgroup_disable=memory"
7435 * boot parameter. This may result in premature OOPS inside
7436 * mem_cgroup_get_nr_swap_pages() function in corner cases.
7437 */
Johannes Weiner21afa382015-02-11 15:26:36 -08007438static int __init mem_cgroup_swap_init(void)
7439{
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007440 /* No memory control -> no swap control */
7441 if (mem_cgroup_disabled())
7442 cgroup_memory_noswap = true;
7443
7444 if (cgroup_memory_noswap)
Johannes Weinereccb52e2020-06-03 16:02:11 -07007445 return 0;
7446
7447 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files));
7448 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files));
7449
Johannes Weiner21afa382015-02-11 15:26:36 -08007450 return 0;
7451}
Bhupesh Sharma82ff1652020-07-23 21:15:21 -07007452core_initcall(mem_cgroup_swap_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08007453
7454#endif /* CONFIG_MEMCG_SWAP */