blob: 1c1061df9cd17cb664d5d6b9faf1ac79db2cef6a [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 Hellwigdc90f082022-02-16 15:31:36 +110056#include <linux/memremap.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070057#include <linux/mm_inline.h>
Johannes Weiner5d1ea48b2014-12-10 15:44:55 -080058#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080059#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070060#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070061#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050062#include <linux/file.h>
Eric W. Biederman03248ad2022-02-09 12:20:45 -060063#include <linux/resume_user_mode.h>
Chris Down0e4b01d2019-09-23 15:34:55 -070064#include <linux/psi.h>
Johannes Weinerc8713d02019-07-11 20:55:59 -070065#include <linux/seq_buf.h>
Michal Hocko6a792692023-03-17 14:44:48 +010066#include <linux/sched/isolation.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080067#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000068#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070069#include <net/ip.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080070#include "slab.h"
NeilBrown014bb1d2022-05-09 18:20:47 -070071#include "swap.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080072
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080073#include <linux/uaccess.h>
Balbir Singh8697d332008-02-07 00:13:59 -080074
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070075#include <trace/events/vmscan.h>
76
Tejun Heo073219e2014-02-08 10:36:58 -050077struct cgroup_subsys memory_cgrp_subsys __read_mostly;
78EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080079
Johannes Weiner7d828602016-01-14 15:20:56 -080080struct mem_cgroup *root_mem_cgroup __read_mostly;
81
Roman Gushchin37d59852020-10-17 16:13:50 -070082/* Active memory cgroup to use from an interrupt context */
83DEFINE_PER_CPU(struct mem_cgroup *, int_active_memcg);
Dan Schatzbergc74d40e2021-06-28 19:38:21 -070084EXPORT_PER_CPU_SYMBOL_GPL(int_active_memcg);
Roman Gushchin37d59852020-10-17 16:13:50 -070085
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080086/* Socket memory accounting disabled? */
Roman Gushchin0f0cace2021-06-02 18:09:29 -070087static bool cgroup_memory_nosocket __ro_after_init;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080088
Vladimir Davydov04823c82016-01-20 15:02:38 -080089/* Kernel memory accounting disabled? */
Muchun Song17c17362022-01-14 14:05:29 -080090static bool cgroup_memory_nokmem __ro_after_init;
Vladimir Davydov04823c82016-01-20 15:02:38 -080091
Yafang Shaob6c1a8a2023-02-10 15:47:31 +000092/* BPF memory accounting disabled? */
93static bool cgroup_memory_nobpf __ro_after_init;
94
Tejun Heo97b27822019-08-26 09:06:56 -070095#ifdef CONFIG_CGROUP_WRITEBACK
96static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq);
97#endif
98
Johannes Weiner7941d212016-01-14 15:21:23 -080099/* Whether legacy memory+swap accounting is active */
100static bool do_memsw_account(void)
101{
Johannes Weinerb25806d2022-09-26 09:57:02 -0400102 return !cgroup_subsys_on_dfl(memory_cgrp_subsys);
Johannes Weiner7941d212016-01-14 15:21:23 -0800103}
104
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700105#define THRESHOLDS_EVENTS_TARGET 128
106#define SOFTLIMIT_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700107
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700108/*
109 * Cgroups above their limits are maintained in a RB-Tree, independent of
110 * their hierarchy representation
111 */
112
Mel Gormanef8f2322016-07-28 15:46:05 -0700113struct mem_cgroup_tree_per_node {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700114 struct rb_root rb_root;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700115 struct rb_node *rb_rightmost;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700116 spinlock_t lock;
117};
118
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700119struct mem_cgroup_tree {
120 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
121};
122
123static struct mem_cgroup_tree soft_limit_tree __read_mostly;
124
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700125/* for OOM */
126struct mem_cgroup_eventfd_list {
127 struct list_head list;
128 struct eventfd_ctx *eventfd;
129};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800130
Tejun Heo79bd9812013-11-22 18:20:42 -0500131/*
132 * cgroup_event represents events which userspace want to receive.
133 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500134struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500135 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500136 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500137 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500138 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500139 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500140 * eventfd to signal userspace about the event.
141 */
142 struct eventfd_ctx *eventfd;
143 /*
144 * Each of these stored in a list by the cgroup.
145 */
146 struct list_head list;
147 /*
Tejun Heofba94802013-11-22 18:20:43 -0500148 * register_event() callback will be used to add new userspace
149 * waiter for changes related to this event. Use eventfd_signal()
150 * on eventfd to send notification to userspace.
151 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500152 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500153 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500154 /*
155 * unregister_event() callback will be called when userspace closes
156 * the eventfd or on cgroup removing. This callback must be set,
157 * if you want provide notification functionality.
158 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500159 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500160 struct eventfd_ctx *eventfd);
161 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500162 * All fields below needed to unregister event when
163 * userspace closes eventfd.
164 */
165 poll_table pt;
166 wait_queue_head_t *wqh;
Ingo Molnarac6424b2017-06-20 12:06:13 +0200167 wait_queue_entry_t wait;
Tejun Heo79bd9812013-11-22 18:20:42 -0500168 struct work_struct remove;
169};
170
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700171static void mem_cgroup_threshold(struct mem_cgroup *memcg);
172static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800173
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800174/* Stuffs for move charges at task migration. */
175/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800176 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800177 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800178#define MOVE_ANON 0x1U
179#define MOVE_FILE 0x2U
180#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800181
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800182/* "mc" and its members are protected by cgroup_mutex */
183static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800184 spinlock_t lock; /* for from, to */
Tejun Heo264a0ae2016-04-21 19:09:02 -0400185 struct mm_struct *mm;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800186 struct mem_cgroup *from;
187 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800188 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800189 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800190 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800191 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800192 struct task_struct *moving_task; /* a task moving charges */
193 wait_queue_head_t waitq; /* a waitq for other context */
194} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700195 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800196 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
197};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800198
Balbir Singh4e416952009-09-23 15:56:39 -0700199/*
Miaohe Linf4d005af2023-07-13 20:14:32 +0800200 * Maximum loops in mem_cgroup_soft_reclaim(), used for soft
Balbir Singh4e416952009-09-23 15:56:39 -0700201 * limit reclaim to prevent infinite loops, if they ever occur.
202 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700203#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700204#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700205
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800206/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800207enum res_type {
208 _MEM,
209 _MEMSWAP,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800210 _KMEM,
Vladimir Davydovd55f90b2016-01-20 15:02:44 -0800211 _TCP,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800212};
213
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700214#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
215#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800216#define MEMFILE_ATTR(val) ((val) & 0xffff)
217
Kirill Tkhaib05706f2018-08-17 15:47:33 -0700218/*
219 * Iteration constructs for visiting all cgroups (under a tree). If
220 * loops are exited prematurely (break), mem_cgroup_iter_break() must
221 * be used for reference counting.
222 */
223#define for_each_mem_cgroup_tree(iter, root) \
224 for (iter = mem_cgroup_iter(root, NULL, NULL); \
225 iter != NULL; \
226 iter = mem_cgroup_iter(root, iter, NULL))
227
228#define for_each_mem_cgroup(iter) \
229 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
230 iter != NULL; \
231 iter = mem_cgroup_iter(NULL, iter, NULL))
232
Vasily Averina4ebf1b2021-11-05 13:38:09 -0700233static inline bool task_is_dying(void)
Tetsuo Handa7775fac2019-03-05 15:46:47 -0800234{
235 return tsk_is_oom_victim(current) || fatal_signal_pending(current) ||
236 (current->flags & PF_EXITING);
237}
238
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700239/* Some nice accessors for the vmpressure. */
240struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
241{
242 if (!memcg)
243 memcg = root_mem_cgroup;
244 return &memcg->vmpressure;
245}
246
Hui Su96478752021-09-02 14:59:36 -0700247struct mem_cgroup *vmpressure_to_memcg(struct vmpressure *vmpr)
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700248{
Hui Su96478752021-09-02 14:59:36 -0700249 return container_of(vmpr, struct mem_cgroup, vmpressure);
Anton Vorontsov70ddf632013-04-29 15:08:31 -0700250}
251
Roman Gushchin1aacbd32023-10-19 15:53:42 -0700252#define CURRENT_OBJCG_UPDATE_BIT 0
253#define CURRENT_OBJCG_UPDATE_FLAG (1UL << CURRENT_OBJCG_UPDATE_BIT)
254
Kirill Tkhai84c07d12018-08-17 15:47:25 -0700255#ifdef CONFIG_MEMCG_KMEM
Roman Gushchin0764db92022-02-11 16:32:32 -0800256static DEFINE_SPINLOCK(objcg_lock);
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700257
Roman Gushchin4d5c8ae2021-06-02 18:09:30 -0700258bool mem_cgroup_kmem_disabled(void)
259{
260 return cgroup_memory_nokmem;
261}
262
Muchun Songf1286fa2021-04-29 22:56:55 -0700263static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg,
264 unsigned int nr_pages);
Roman Gushchinc1a660d2021-02-24 12:03:58 -0800265
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700266static void obj_cgroup_release(struct percpu_ref *ref)
267{
268 struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt);
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700269 unsigned int nr_bytes;
270 unsigned int nr_pages;
271 unsigned long flags;
272
273 /*
274 * At this point all allocated objects are freed, and
275 * objcg->nr_charged_bytes can't have an arbitrary byte value.
276 * However, it can be PAGE_SIZE or (x * PAGE_SIZE).
277 *
278 * The following sequence can lead to it:
279 * 1) CPU0: objcg == stock->cached_objcg
280 * 2) CPU1: we do a small allocation (e.g. 92 bytes),
281 * PAGE_SIZE bytes are charged
282 * 3) CPU1: a process from another memcg is allocating something,
283 * the stock if flushed,
284 * objcg->nr_charged_bytes = PAGE_SIZE - 92
285 * 5) CPU0: we do release this object,
286 * 92 bytes are added to stock->nr_bytes
287 * 6) CPU0: stock is flushed,
288 * 92 bytes are added to objcg->nr_charged_bytes
289 *
290 * In the result, nr_charged_bytes == PAGE_SIZE.
291 * This page will be uncharged in obj_cgroup_release().
292 */
293 nr_bytes = atomic_read(&objcg->nr_charged_bytes);
294 WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1));
295 nr_pages = nr_bytes >> PAGE_SHIFT;
296
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700297 if (nr_pages)
Muchun Songf1286fa2021-04-29 22:56:55 -0700298 obj_cgroup_uncharge_pages(objcg, nr_pages);
Muchun Song271dd6b2021-06-28 19:38:06 -0700299
Roman Gushchin0764db92022-02-11 16:32:32 -0800300 spin_lock_irqsave(&objcg_lock, flags);
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700301 list_del(&objcg->list);
Roman Gushchin0764db92022-02-11 16:32:32 -0800302 spin_unlock_irqrestore(&objcg_lock, flags);
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700303
304 percpu_ref_exit(ref);
305 kfree_rcu(objcg, rcu);
306}
307
308static struct obj_cgroup *obj_cgroup_alloc(void)
309{
310 struct obj_cgroup *objcg;
311 int ret;
312
313 objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL);
314 if (!objcg)
315 return NULL;
316
317 ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0,
318 GFP_KERNEL);
319 if (ret) {
320 kfree(objcg);
321 return NULL;
322 }
323 INIT_LIST_HEAD(&objcg->list);
324 return objcg;
325}
326
327static void memcg_reparent_objcgs(struct mem_cgroup *memcg,
328 struct mem_cgroup *parent)
329{
330 struct obj_cgroup *objcg, *iter;
331
332 objcg = rcu_replace_pointer(memcg->objcg, NULL, true);
333
Roman Gushchin0764db92022-02-11 16:32:32 -0800334 spin_lock_irq(&objcg_lock);
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700335
Muchun Song98383542021-06-28 19:38:03 -0700336 /* 1) Ready to reparent active objcg. */
337 list_add(&objcg->list, &memcg->objcg_list);
338 /* 2) Reparent active objcg and already reparented objcgs to parent. */
339 list_for_each_entry(iter, &memcg->objcg_list, list)
340 WRITE_ONCE(iter->memcg, parent);
341 /* 3) Move already reparented objcgs to the parent's list */
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700342 list_splice(&memcg->objcg_list, &parent->objcg_list);
343
Roman Gushchin0764db92022-02-11 16:32:32 -0800344 spin_unlock_irq(&objcg_lock);
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700345
346 percpu_ref_kill(&objcg->refcnt);
347}
348
Glauber Costa55007d82012-12-18 14:22:38 -0800349/*
Glauber Costad7f25f82012-12-18 14:22:40 -0800350 * A lot of the calls to the cache allocation functions are expected to be
Roman Gushchin272911a2020-08-06 23:21:17 -0700351 * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are
Glauber Costad7f25f82012-12-18 14:22:40 -0800352 * conditional to this static branch, we'll have to allow modules that does
353 * kmem_cache_alloc and the such to see this symbol as well
354 */
Roman Gushchinf7a449f2023-02-13 11:29:22 -0800355DEFINE_STATIC_KEY_FALSE(memcg_kmem_online_key);
356EXPORT_SYMBOL(memcg_kmem_online_key);
Yafang Shaob6c1a8a2023-02-10 15:47:31 +0000357
358DEFINE_STATIC_KEY_FALSE(memcg_bpf_enabled_key);
359EXPORT_SYMBOL(memcg_bpf_enabled_key);
Yang Shi0a432dcb2019-09-23 15:38:12 -0700360#endif
Tejun Heo17cc4df2017-02-22 15:41:36 -0800361
Tejun Heoad7fa852015-05-27 20:00:02 -0400362/**
Matthew Wilcox (Oracle)75376c62023-01-16 19:25:07 +0000363 * mem_cgroup_css_from_folio - css of the memcg associated with a folio
364 * @folio: folio of interest
Tejun Heoad7fa852015-05-27 20:00:02 -0400365 *
366 * If memcg is bound to the default hierarchy, css of the memcg associated
Matthew Wilcox (Oracle)75376c62023-01-16 19:25:07 +0000367 * with @folio is returned. The returned css remains associated with @folio
Tejun Heoad7fa852015-05-27 20:00:02 -0400368 * until it is released.
369 *
370 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
371 * is returned.
Tejun Heoad7fa852015-05-27 20:00:02 -0400372 */
Matthew Wilcox (Oracle)75376c62023-01-16 19:25:07 +0000373struct cgroup_subsys_state *mem_cgroup_css_from_folio(struct folio *folio)
Tejun Heoad7fa852015-05-27 20:00:02 -0400374{
Matthew Wilcox (Oracle)75376c62023-01-16 19:25:07 +0000375 struct mem_cgroup *memcg = folio_memcg(folio);
Tejun Heoad7fa852015-05-27 20:00:02 -0400376
Tejun Heo9e10a132015-09-18 11:56:28 -0400377 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400378 memcg = root_mem_cgroup;
379
Tejun Heoad7fa852015-05-27 20:00:02 -0400380 return &memcg->css;
381}
382
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700383/**
384 * page_cgroup_ino - return inode number of the memcg a page is charged to
385 * @page: the page
386 *
387 * Look up the closest online ancestor of the memory cgroup @page is charged to
388 * and return its inode number or 0 if @page is not charged to any cgroup. It
389 * is safe to call this function without holding a reference to @page.
390 *
391 * Note, this function is inherently racy, because there is nothing to prevent
392 * the cgroup inode from getting torn down and potentially reallocated a moment
393 * after page_cgroup_ino() returns, so it only should be used by callers that
394 * do not care (such as procfs interfaces).
395 */
396ino_t page_cgroup_ino(struct page *page)
397{
398 struct mem_cgroup *memcg;
399 unsigned long ino = 0;
400
401 rcu_read_lock();
Yosry Ahmedec342602023-04-12 00:34:51 +0000402 /* page_folio() is racy here, but the entire function is racy anyway */
403 memcg = folio_memcg_check(page_folio(page));
Roman Gushchin286e04b2020-08-06 23:20:52 -0700404
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700405 while (memcg && !(memcg->css.flags & CSS_ONLINE))
406 memcg = parent_mem_cgroup(memcg);
407 if (memcg)
408 ino = cgroup_ino(memcg->css.cgroup);
409 rcu_read_unlock();
410 return ino;
411}
412
Mel Gormanef8f2322016-07-28 15:46:05 -0700413static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz,
414 struct mem_cgroup_tree_per_node *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800415 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700416{
417 struct rb_node **p = &mctz->rb_root.rb_node;
418 struct rb_node *parent = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700419 struct mem_cgroup_per_node *mz_node;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700420 bool rightmost = true;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700421
422 if (mz->on_tree)
423 return;
424
425 mz->usage_in_excess = new_usage_in_excess;
426 if (!mz->usage_in_excess)
427 return;
428 while (*p) {
429 parent = *p;
Mel Gormanef8f2322016-07-28 15:46:05 -0700430 mz_node = rb_entry(parent, struct mem_cgroup_per_node,
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700431 tree_node);
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700432 if (mz->usage_in_excess < mz_node->usage_in_excess) {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700433 p = &(*p)->rb_left;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700434 rightmost = false;
Miaohe Lin378876b02020-12-14 19:06:28 -0800435 } else {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700436 p = &(*p)->rb_right;
Miaohe Lin378876b02020-12-14 19:06:28 -0800437 }
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700438 }
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700439
440 if (rightmost)
441 mctz->rb_rightmost = &mz->tree_node;
442
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700443 rb_link_node(&mz->tree_node, parent, p);
444 rb_insert_color(&mz->tree_node, &mctz->rb_root);
445 mz->on_tree = true;
446}
447
Mel Gormanef8f2322016-07-28 15:46:05 -0700448static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
449 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700450{
451 if (!mz->on_tree)
452 return;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700453
454 if (&mz->tree_node == mctz->rb_rightmost)
455 mctz->rb_rightmost = rb_prev(&mz->tree_node);
456
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700457 rb_erase(&mz->tree_node, &mctz->rb_root);
458 mz->on_tree = false;
459}
460
Mel Gormanef8f2322016-07-28 15:46:05 -0700461static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
462 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700463{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700464 unsigned long flags;
465
466 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700467 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700468 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700469}
470
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800471static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
472{
473 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700474 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800475 unsigned long excess = 0;
476
477 if (nr_pages > soft_limit)
478 excess = nr_pages - soft_limit;
479
480 return excess;
481}
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700482
Matthew Wilcox (Oracle)658b69c2021-04-29 14:51:26 -0400483static void mem_cgroup_update_tree(struct mem_cgroup *memcg, int nid)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700484{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800485 unsigned long excess;
Mel Gormanef8f2322016-07-28 15:46:05 -0700486 struct mem_cgroup_per_node *mz;
487 struct mem_cgroup_tree_per_node *mctz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700488
Yu Zhaoe4dde562022-12-21 21:19:04 -0700489 if (lru_gen_enabled()) {
T.J. Alumbaugh36c7b4d2023-01-18 00:18:24 +0000490 if (soft_limit_excess(memcg))
T.J. Alumbaugh5c7e7a02023-05-22 11:20:56 +0000491 lru_gen_soft_reclaim(memcg, nid);
Yu Zhaoe4dde562022-12-21 21:19:04 -0700492 return;
493 }
494
Matthew Wilcox (Oracle)2ab082b2021-06-25 08:57:02 -0400495 mctz = soft_limit_tree.rb_tree_per_node[nid];
Laurent Dufourbfc72282017-03-09 16:17:06 -0800496 if (!mctz)
497 return;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700498 /*
499 * Necessary to update all ancestors when hierarchy is used.
500 * because their event counter is not touched.
501 */
502 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Matthew Wilcox (Oracle)658b69c2021-04-29 14:51:26 -0400503 mz = memcg->nodeinfo[nid];
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800504 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700505 /*
506 * We have to update the tree if mz is on RB-tree or
507 * mem is over its softlimit.
508 */
509 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700510 unsigned long flags;
511
512 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700513 /* if on-tree, remove it */
514 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700515 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700516 /*
517 * Insert again. mz->usage_in_excess will be updated.
518 * If excess is 0, no tree ops.
519 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700520 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700521 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700522 }
523 }
524}
525
526static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
527{
Mel Gormanef8f2322016-07-28 15:46:05 -0700528 struct mem_cgroup_tree_per_node *mctz;
529 struct mem_cgroup_per_node *mz;
530 int nid;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700531
Jianyu Zhane2318752014-06-06 14:38:20 -0700532 for_each_node(nid) {
Johannes Weinera3747b52021-04-29 22:56:14 -0700533 mz = memcg->nodeinfo[nid];
Matthew Wilcox (Oracle)2ab082b2021-06-25 08:57:02 -0400534 mctz = soft_limit_tree.rb_tree_per_node[nid];
Laurent Dufourbfc72282017-03-09 16:17:06 -0800535 if (mctz)
536 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700537 }
538}
539
Mel Gormanef8f2322016-07-28 15:46:05 -0700540static struct mem_cgroup_per_node *
541__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700542{
Mel Gormanef8f2322016-07-28 15:46:05 -0700543 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700544
545retry:
546 mz = NULL;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700547 if (!mctz->rb_rightmost)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700548 goto done; /* Nothing to reclaim from */
549
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700550 mz = rb_entry(mctz->rb_rightmost,
551 struct mem_cgroup_per_node, tree_node);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700552 /*
553 * Remove the node now but someone else can add it back,
554 * we will to add it back at the end of reclaim to its correct
555 * position in the tree.
556 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700557 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800558 if (!soft_limit_excess(mz->memcg) ||
Shakeel Butt8965aa22020-04-01 21:07:10 -0700559 !css_tryget(&mz->memcg->css))
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700560 goto retry;
561done:
562 return mz;
563}
564
Mel Gormanef8f2322016-07-28 15:46:05 -0700565static struct mem_cgroup_per_node *
566mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700567{
Mel Gormanef8f2322016-07-28 15:46:05 -0700568 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700569
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700570 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700571 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700572 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700573 return mz;
574}
575
Shakeel Butt11192d92021-11-05 13:37:31 -0700576/*
577 * memcg and lruvec stats flushing
578 *
579 * Many codepaths leading to stats update or read are performance sensitive and
580 * adding stats flushing in such codepaths is not desirable. So, to optimize the
581 * flushing the kernel does:
582 *
583 * 1) Periodically and asynchronously flush the stats every 2 seconds to not let
584 * rstat update tree grow unbounded.
585 *
586 * 2) Flush the stats synchronously on reader side only when there are more than
587 * (MEMCG_CHARGE_BATCH * nr_cpus) update events. Though this optimization
588 * will let stats be out of sync by atmost (MEMCG_CHARGE_BATCH * nr_cpus) but
589 * only for 2 seconds due to (1).
590 */
591static void flush_memcg_stats_dwork(struct work_struct *w);
592static DECLARE_DEFERRABLE_WORK(stats_flush_dwork, flush_memcg_stats_dwork);
Shakeel Butt11192d92021-11-05 13:37:31 -0700593static DEFINE_PER_CPU(unsigned int, stats_updates);
Yosry Ahmed3cd99922023-03-30 19:17:57 +0000594static atomic_t stats_flush_ongoing = ATOMIC_INIT(0);
Shakeel Butt11192d92021-11-05 13:37:31 -0700595static atomic_t stats_flush_threshold = ATOMIC_INIT(0);
Shakeel Butt9b301612022-04-21 16:35:40 -0700596static u64 flush_next_time;
597
598#define FLUSH_TIME (2UL*HZ)
Shakeel Butt11192d92021-11-05 13:37:31 -0700599
Sebastian Andrzej Siewiorbe3e67b2022-03-22 14:40:41 -0700600/*
601 * Accessors to ensure that preemption is disabled on PREEMPT_RT because it can
602 * not rely on this as part of an acquired spinlock_t lock. These functions are
603 * never used in hardirq context on PREEMPT_RT and therefore disabling preemtion
604 * is sufficient.
605 */
606static void memcg_stats_lock(void)
607{
Thomas Gleixnere575d402022-08-25 18:41:28 +0200608 preempt_disable_nested();
609 VM_WARN_ON_IRQS_ENABLED();
Sebastian Andrzej Siewiorbe3e67b2022-03-22 14:40:41 -0700610}
611
612static void __memcg_stats_lock(void)
613{
Thomas Gleixnere575d402022-08-25 18:41:28 +0200614 preempt_disable_nested();
Sebastian Andrzej Siewiorbe3e67b2022-03-22 14:40:41 -0700615}
616
617static void memcg_stats_unlock(void)
618{
Thomas Gleixnere575d402022-08-25 18:41:28 +0200619 preempt_enable_nested();
Sebastian Andrzej Siewiorbe3e67b2022-03-22 14:40:41 -0700620}
621
Shakeel Butt5b3be692022-01-14 14:05:39 -0800622static inline void memcg_rstat_updated(struct mem_cgroup *memcg, int val)
Shakeel Butt11192d92021-11-05 13:37:31 -0700623{
Shakeel Butt5b3be692022-01-14 14:05:39 -0800624 unsigned int x;
625
Yosry Ahmedf9d911c2023-03-30 19:18:01 +0000626 if (!val)
627 return;
628
Shakeel Butt11192d92021-11-05 13:37:31 -0700629 cgroup_rstat_updated(memcg->css.cgroup, smp_processor_id());
Shakeel Butt5b3be692022-01-14 14:05:39 -0800630
631 x = __this_cpu_add_return(stats_updates, abs(val));
632 if (x > MEMCG_CHARGE_BATCH) {
Jiebin Sun873f64b2022-07-23 00:49:49 +0800633 /*
634 * If stats_flush_threshold exceeds the threshold
635 * (>num_online_cpus()), cgroup stats update will be triggered
636 * in __mem_cgroup_flush_stats(). Increasing this var further
637 * is redundant and simply adds overhead in atomic update.
638 */
639 if (atomic_read(&stats_flush_threshold) <= num_online_cpus())
640 atomic_add(x / MEMCG_CHARGE_BATCH, &stats_flush_threshold);
Shakeel Butt5b3be692022-01-14 14:05:39 -0800641 __this_cpu_write(stats_updates, 0);
642 }
Shakeel Butt11192d92021-11-05 13:37:31 -0700643}
644
Yosry Ahmed35822fd2023-04-21 17:40:19 +0000645static void do_flush_stats(void)
Shakeel Butt11192d92021-11-05 13:37:31 -0700646{
Yosry Ahmed3cd99922023-03-30 19:17:57 +0000647 /*
648 * We always flush the entire tree, so concurrent flushers can just
649 * skip. This avoids a thundering herd problem on the rstat global lock
650 * from memcg flushers (e.g. reclaim, refault, etc).
651 */
652 if (atomic_read(&stats_flush_ongoing) ||
653 atomic_xchg(&stats_flush_ongoing, 1))
Shakeel Butt11192d92021-11-05 13:37:31 -0700654 return;
655
Yosry Ahmed3cd99922023-03-30 19:17:57 +0000656 WRITE_ONCE(flush_next_time, jiffies_64 + 2*FLUSH_TIME);
Yosry Ahmed9fad9ae2023-03-30 19:17:58 +0000657
Yosry Ahmed35822fd2023-04-21 17:40:19 +0000658 cgroup_rstat_flush(root_mem_cgroup->css.cgroup);
Yosry Ahmed9fad9ae2023-03-30 19:17:58 +0000659
Shakeel Butt11192d92021-11-05 13:37:31 -0700660 atomic_set(&stats_flush_threshold, 0);
Yosry Ahmed3cd99922023-03-30 19:17:57 +0000661 atomic_set(&stats_flush_ongoing, 0);
Shakeel Butt11192d92021-11-05 13:37:31 -0700662}
663
Yosry Ahmed9fad9ae2023-03-30 19:17:58 +0000664void mem_cgroup_flush_stats(void)
665{
Yosry Ahmed35822fd2023-04-21 17:40:19 +0000666 if (atomic_read(&stats_flush_threshold) > num_online_cpus())
667 do_flush_stats();
Yosry Ahmed9fad9ae2023-03-30 19:17:58 +0000668}
669
Yosry Ahmed4009b2f2023-03-30 19:17:59 +0000670void mem_cgroup_flush_stats_ratelimited(void)
Shakeel Butt9b301612022-04-21 16:35:40 -0700671{
Yosry Ahmed3cd99922023-03-30 19:17:57 +0000672 if (time_after64(jiffies_64, READ_ONCE(flush_next_time)))
Yosry Ahmed4009b2f2023-03-30 19:17:59 +0000673 mem_cgroup_flush_stats();
Shakeel Butt9b301612022-04-21 16:35:40 -0700674}
675
Shakeel Butt11192d92021-11-05 13:37:31 -0700676static void flush_memcg_stats_dwork(struct work_struct *w)
677{
Yosry Ahmed9fad9ae2023-03-30 19:17:58 +0000678 /*
679 * Always flush here so that flushing in latency-sensitive paths is
680 * as cheap as possible.
681 */
Yosry Ahmed35822fd2023-04-21 17:40:19 +0000682 do_flush_stats();
Shakeel Butt9b301612022-04-21 16:35:40 -0700683 queue_delayed_work(system_unbound_wq, &stats_flush_dwork, FLUSH_TIME);
Shakeel Butt11192d92021-11-05 13:37:31 -0700684}
685
Shakeel Buttd396def2022-09-07 04:35:36 +0000686/* Subset of vm_event_item to report for memcg event stats */
687static const unsigned int memcg_vm_event_stat[] = {
Shakeel Butt8278f1c2022-09-07 04:35:37 +0000688 PGPGIN,
689 PGPGOUT,
Shakeel Buttd396def2022-09-07 04:35:36 +0000690 PGSCAN_KSWAPD,
691 PGSCAN_DIRECT,
Johannes Weiner57e9cc52022-10-26 14:01:33 -0400692 PGSCAN_KHUGEPAGED,
Shakeel Buttd396def2022-09-07 04:35:36 +0000693 PGSTEAL_KSWAPD,
694 PGSTEAL_DIRECT,
Johannes Weiner57e9cc52022-10-26 14:01:33 -0400695 PGSTEAL_KHUGEPAGED,
Shakeel Buttd396def2022-09-07 04:35:36 +0000696 PGFAULT,
697 PGMAJFAULT,
698 PGREFILL,
699 PGACTIVATE,
700 PGDEACTIVATE,
701 PGLAZYFREE,
702 PGLAZYFREED,
703#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP)
704 ZSWPIN,
705 ZSWPOUT,
706#endif
707#ifdef CONFIG_TRANSPARENT_HUGEPAGE
708 THP_FAULT_ALLOC,
709 THP_COLLAPSE_ALLOC,
Xin Hao811244a2023-09-14 00:49:37 +0800710 THP_SWPOUT,
711 THP_SWPOUT_FALLBACK,
Shakeel Buttd396def2022-09-07 04:35:36 +0000712#endif
713};
714
Shakeel Butt8278f1c2022-09-07 04:35:37 +0000715#define NR_MEMCG_EVENTS ARRAY_SIZE(memcg_vm_event_stat)
716static int mem_cgroup_events_index[NR_VM_EVENT_ITEMS] __read_mostly;
717
718static void init_memcg_events(void)
719{
720 int i;
721
722 for (i = 0; i < NR_MEMCG_EVENTS; ++i)
723 mem_cgroup_events_index[memcg_vm_event_stat[i]] = i + 1;
724}
725
726static inline int memcg_events_index(enum vm_event_item idx)
727{
728 return mem_cgroup_events_index[idx] - 1;
729}
730
Shakeel Butt410f8e82022-09-07 04:35:35 +0000731struct memcg_vmstats_percpu {
732 /* Local (CPU and cgroup) page state & events */
733 long state[MEMCG_NR_STAT];
Shakeel Butt8278f1c2022-09-07 04:35:37 +0000734 unsigned long events[NR_MEMCG_EVENTS];
Shakeel Butt410f8e82022-09-07 04:35:35 +0000735
736 /* Delta calculation for lockless upward propagation */
737 long state_prev[MEMCG_NR_STAT];
Shakeel Butt8278f1c2022-09-07 04:35:37 +0000738 unsigned long events_prev[NR_MEMCG_EVENTS];
Shakeel Butt410f8e82022-09-07 04:35:35 +0000739
740 /* Cgroup1: threshold notifications & softlimit tree updates */
741 unsigned long nr_page_events;
742 unsigned long targets[MEM_CGROUP_NTARGETS];
743};
744
745struct memcg_vmstats {
746 /* Aggregated (CPU and subtree) page state & events */
747 long state[MEMCG_NR_STAT];
Shakeel Butt8278f1c2022-09-07 04:35:37 +0000748 unsigned long events[NR_MEMCG_EVENTS];
Shakeel Butt410f8e82022-09-07 04:35:35 +0000749
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +0000750 /* Non-hierarchical (CPU aggregated) page state & events */
751 long state_local[MEMCG_NR_STAT];
752 unsigned long events_local[NR_MEMCG_EVENTS];
753
Shakeel Butt410f8e82022-09-07 04:35:35 +0000754 /* Pending child counts during tree propagation */
755 long state_pending[MEMCG_NR_STAT];
Shakeel Butt8278f1c2022-09-07 04:35:37 +0000756 unsigned long events_pending[NR_MEMCG_EVENTS];
Shakeel Butt410f8e82022-09-07 04:35:35 +0000757};
758
759unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)
760{
761 long x = READ_ONCE(memcg->vmstats->state[idx]);
762#ifdef CONFIG_SMP
763 if (x < 0)
764 x = 0;
765#endif
766 return x;
767}
768
Yosry Ahmed7bd5bc32023-09-22 17:57:40 +0000769static int memcg_page_state_unit(int item);
770
771/*
772 * Normalize the value passed into memcg_rstat_updated() to be in pages. Round
773 * up non-zero sub-page updates to 1 page as zero page updates are ignored.
774 */
775static int memcg_state_val_in_pages(int idx, int val)
776{
777 int unit = memcg_page_state_unit(idx);
778
779 if (!val || unit == PAGE_SIZE)
780 return val;
781 else
782 return max(val * unit / PAGE_SIZE, 1UL);
783}
784
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700785/**
786 * __mod_memcg_state - update cgroup memory statistics
787 * @memcg: the memory cgroup
788 * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item
789 * @val: delta to add to the counter, can be negative
790 */
791void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val)
792{
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700793 if (mem_cgroup_disabled())
794 return;
795
Johannes Weiner2d146aa2021-04-29 22:56:26 -0700796 __this_cpu_add(memcg->vmstats_percpu->state[idx], val);
Yosry Ahmed7bd5bc32023-09-22 17:57:40 +0000797 memcg_rstat_updated(memcg, memcg_state_val_in_pages(idx, val));
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700798}
799
Johannes Weiner2d146aa2021-04-29 22:56:26 -0700800/* idx can be of type enum memcg_stat_item or node_stat_item. */
Johannes Weinera18e6e62021-04-29 22:56:17 -0700801static unsigned long memcg_page_state_local(struct mem_cgroup *memcg, int idx)
802{
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +0000803 long x = READ_ONCE(memcg->vmstats->state_local[idx]);
Johannes Weinera18e6e62021-04-29 22:56:17 -0700804
Johannes Weinera18e6e62021-04-29 22:56:17 -0700805#ifdef CONFIG_SMP
806 if (x < 0)
807 x = 0;
808#endif
809 return x;
810}
811
Roman Gushchineedc4e52020-08-06 23:20:32 -0700812void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
813 int val)
814{
815 struct mem_cgroup_per_node *pn;
816 struct mem_cgroup *memcg;
Roman Gushchineedc4e52020-08-06 23:20:32 -0700817
818 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
819 memcg = pn->memcg;
820
Sebastian Andrzej Siewiorbe3e67b2022-03-22 14:40:41 -0700821 /*
Muhammad Muzammilbe16dd762023-10-23 17:44:05 +0500822 * The caller from rmap relies on disabled preemption because they never
Sebastian Andrzej Siewiorbe3e67b2022-03-22 14:40:41 -0700823 * update their counter from in-interrupt context. For these two
824 * counters we check that the update is never performed from an
825 * interrupt context while other caller need to have disabled interrupt.
826 */
827 __memcg_stats_lock();
Thomas Gleixnere575d402022-08-25 18:41:28 +0200828 if (IS_ENABLED(CONFIG_DEBUG_VM)) {
Sebastian Andrzej Siewiorbe3e67b2022-03-22 14:40:41 -0700829 switch (idx) {
830 case NR_ANON_MAPPED:
831 case NR_FILE_MAPPED:
832 case NR_ANON_THPS:
833 case NR_SHMEM_PMDMAPPED:
834 case NR_FILE_PMDMAPPED:
835 WARN_ON_ONCE(!in_task());
836 break;
837 default:
Thomas Gleixnere575d402022-08-25 18:41:28 +0200838 VM_WARN_ON_IRQS_ENABLED();
Sebastian Andrzej Siewiorbe3e67b2022-03-22 14:40:41 -0700839 }
840 }
841
Roman Gushchineedc4e52020-08-06 23:20:32 -0700842 /* Update memcg */
Shakeel Butt11192d92021-11-05 13:37:31 -0700843 __this_cpu_add(memcg->vmstats_percpu->state[idx], val);
Roman Gushchineedc4e52020-08-06 23:20:32 -0700844
845 /* Update lruvec */
Shakeel Butt7e1c0d62021-09-02 14:55:00 -0700846 __this_cpu_add(pn->lruvec_stats_percpu->state[idx], val);
Shakeel Butt11192d92021-11-05 13:37:31 -0700847
Yosry Ahmed7bd5bc32023-09-22 17:57:40 +0000848 memcg_rstat_updated(memcg, memcg_state_val_in_pages(idx, val));
Sebastian Andrzej Siewiorbe3e67b2022-03-22 14:40:41 -0700849 memcg_stats_unlock();
Roman Gushchineedc4e52020-08-06 23:20:32 -0700850}
851
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700852/**
853 * __mod_lruvec_state - update lruvec memory statistics
854 * @lruvec: the lruvec
855 * @idx: the stat item
856 * @val: delta to add to the counter, can be negative
857 *
858 * The lruvec is the intersection of the NUMA node and a cgroup. This
859 * function updates the all three counters that are affected by a
860 * change of state at this level: per-node, per-cgroup, per-lruvec.
861 */
862void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
863 int val)
864{
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700865 /* Update node */
Roman Gushchineedc4e52020-08-06 23:20:32 -0700866 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700867
Roman Gushchineedc4e52020-08-06 23:20:32 -0700868 /* Update memcg and lruvec */
869 if (!mem_cgroup_disabled())
870 __mod_memcg_lruvec_state(lruvec, idx, val);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700871}
872
Shakeel Buttc47d5032020-12-14 19:07:14 -0800873void __mod_lruvec_page_state(struct page *page, enum node_stat_item idx,
874 int val)
875{
876 struct page *head = compound_head(page); /* rmap on tail pages */
Muchun Songb4e0b682021-04-29 22:56:52 -0700877 struct mem_cgroup *memcg;
Shakeel Buttc47d5032020-12-14 19:07:14 -0800878 pg_data_t *pgdat = page_pgdat(page);
879 struct lruvec *lruvec;
880
Muchun Songb4e0b682021-04-29 22:56:52 -0700881 rcu_read_lock();
882 memcg = page_memcg(head);
Shakeel Buttc47d5032020-12-14 19:07:14 -0800883 /* Untracked pages have no memcg, no lruvec. Update only the node */
Linus Torvaldsd635a69d2020-12-15 13:22:29 -0800884 if (!memcg) {
Muchun Songb4e0b682021-04-29 22:56:52 -0700885 rcu_read_unlock();
Shakeel Buttc47d5032020-12-14 19:07:14 -0800886 __mod_node_page_state(pgdat, idx, val);
887 return;
888 }
889
Linus Torvaldsd635a69d2020-12-15 13:22:29 -0800890 lruvec = mem_cgroup_lruvec(memcg, pgdat);
Shakeel Buttc47d5032020-12-14 19:07:14 -0800891 __mod_lruvec_state(lruvec, idx, val);
Muchun Songb4e0b682021-04-29 22:56:52 -0700892 rcu_read_unlock();
Shakeel Buttc47d5032020-12-14 19:07:14 -0800893}
Shakeel Buttf0c0c112020-12-14 19:07:17 -0800894EXPORT_SYMBOL(__mod_lruvec_page_state);
Shakeel Buttc47d5032020-12-14 19:07:14 -0800895
Muchun Songda3ceef2020-12-14 19:07:04 -0800896void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val)
Roman Gushchinec9f0232019-08-13 15:37:41 -0700897{
Roman Gushchin4f103c62020-04-01 21:06:36 -0700898 pg_data_t *pgdat = page_pgdat(virt_to_page(p));
Roman Gushchinec9f0232019-08-13 15:37:41 -0700899 struct mem_cgroup *memcg;
900 struct lruvec *lruvec;
901
902 rcu_read_lock();
Roman Gushchinfc4db902022-06-10 11:03:10 -0700903 memcg = mem_cgroup_from_slab_obj(p);
Roman Gushchinec9f0232019-08-13 15:37:41 -0700904
Muchun Song8faeb1f2020-11-21 22:17:12 -0800905 /*
906 * Untracked pages have no memcg, no lruvec. Update only the
907 * node. If we reparent the slab objects to the root memcg,
908 * when we free the slab object, we need to update the per-memcg
909 * vmstats to keep it correct for the root memcg.
910 */
911 if (!memcg) {
Roman Gushchinec9f0232019-08-13 15:37:41 -0700912 __mod_node_page_state(pgdat, idx, val);
913 } else {
Johannes Weiner867e5e12019-11-30 17:55:34 -0800914 lruvec = mem_cgroup_lruvec(memcg, pgdat);
Roman Gushchinec9f0232019-08-13 15:37:41 -0700915 __mod_lruvec_state(lruvec, idx, val);
916 }
917 rcu_read_unlock();
918}
919
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700920/**
921 * __count_memcg_events - account VM events in a cgroup
922 * @memcg: the memory cgroup
923 * @idx: the event item
Ingo Molnarf0953a12021-05-06 18:06:47 -0700924 * @count: the number of events that occurred
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700925 */
926void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx,
927 unsigned long count)
928{
Shakeel Butt8278f1c2022-09-07 04:35:37 +0000929 int index = memcg_events_index(idx);
930
931 if (mem_cgroup_disabled() || index < 0)
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700932 return;
933
Sebastian Andrzej Siewiorbe3e67b2022-03-22 14:40:41 -0700934 memcg_stats_lock();
Shakeel Butt8278f1c2022-09-07 04:35:37 +0000935 __this_cpu_add(memcg->vmstats_percpu->events[index], count);
Shakeel Butt5b3be692022-01-14 14:05:39 -0800936 memcg_rstat_updated(memcg, count);
Sebastian Andrzej Siewiorbe3e67b2022-03-22 14:40:41 -0700937 memcg_stats_unlock();
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700938}
939
Johannes Weiner42a30032019-05-14 15:47:12 -0700940static unsigned long memcg_events(struct mem_cgroup *memcg, int event)
Johannes Weinere9f89742011-03-23 16:42:37 -0700941{
Shakeel Butt8278f1c2022-09-07 04:35:37 +0000942 int index = memcg_events_index(event);
943
944 if (index < 0)
945 return 0;
946 return READ_ONCE(memcg->vmstats->events[index]);
Johannes Weinere9f89742011-03-23 16:42:37 -0700947}
948
Johannes Weiner42a30032019-05-14 15:47:12 -0700949static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event)
950{
Shakeel Butt8278f1c2022-09-07 04:35:37 +0000951 int index = memcg_events_index(event);
952
953 if (index < 0)
954 return 0;
Johannes Weiner815744d2019-06-13 15:55:46 -0700955
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +0000956 return READ_ONCE(memcg->vmstats->events_local[index]);
Johannes Weiner42a30032019-05-14 15:47:12 -0700957}
958
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700959static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
Johannes Weiner3fba69a2020-06-03 16:01:31 -0700960 int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800961{
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800962 /* pagein of a big page is an event. So, ignore page size */
963 if (nr_pages > 0)
Johannes Weinerc9019e92018-01-31 16:16:37 -0800964 __count_memcg_events(memcg, PGPGIN, 1);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800965 else {
Johannes Weinerc9019e92018-01-31 16:16:37 -0800966 __count_memcg_events(memcg, PGPGOUT, 1);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800967 nr_pages = -nr_pages; /* for event */
968 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800969
Chris Down871789d2019-05-14 15:46:57 -0700970 __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800971}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800972
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800973static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
974 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800975{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700976 unsigned long val, next;
977
Chris Down871789d2019-05-14 15:46:57 -0700978 val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events);
979 next = __this_cpu_read(memcg->vmstats_percpu->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700980 /* from time_after() in jiffies.h */
Michal Hocko6a1a8b82017-07-10 15:48:53 -0700981 if ((long)(next - val) < 0) {
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800982 switch (target) {
983 case MEM_CGROUP_TARGET_THRESH:
984 next = val + THRESHOLDS_EVENTS_TARGET;
985 break;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700986 case MEM_CGROUP_TARGET_SOFTLIMIT:
987 next = val + SOFTLIMIT_EVENTS_TARGET;
988 break;
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800989 default:
990 break;
991 }
Chris Down871789d2019-05-14 15:46:57 -0700992 __this_cpu_write(memcg->vmstats_percpu->targets[target], next);
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800993 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700994 }
Johannes Weinerf53d7ce32012-01-12 17:18:23 -0800995 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800996}
997
998/*
999 * Check events in order.
1000 *
1001 */
Matthew Wilcox (Oracle)8e88bd22021-06-25 09:05:47 -04001002static void memcg_check_events(struct mem_cgroup *memcg, int nid)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001003{
Sebastian Andrzej Siewior2343e882022-03-22 14:40:38 -07001004 if (IS_ENABLED(CONFIG_PREEMPT_RT))
1005 return;
1006
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001007 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce32012-01-12 17:18:23 -08001008 if (unlikely(mem_cgroup_event_ratelimit(memcg,
1009 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -07001010 bool do_softlimit;
Johannes Weinerf53d7ce32012-01-12 17:18:23 -08001011
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -07001012 do_softlimit = mem_cgroup_event_ratelimit(memcg,
1013 MEM_CGROUP_TARGET_SOFTLIMIT);
Johannes Weinerf53d7ce32012-01-12 17:18:23 -08001014 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -07001015 if (unlikely(do_softlimit))
Matthew Wilcox (Oracle)8e88bd22021-06-25 09:05:47 -04001016 mem_cgroup_update_tree(memcg, nid);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07001017 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -08001018}
1019
Balbir Singhcf475ad2008-04-29 01:00:16 -07001020struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001021{
Balbir Singh31a78f22008-09-28 23:09:31 +01001022 /*
1023 * mm_update_next_owner() may clear mm->owner to NULL
1024 * if it races with swapoff, page migration, etc.
1025 * So this can be called with p == NULL.
1026 */
1027 if (unlikely(!p))
1028 return NULL;
1029
Tejun Heo073219e2014-02-08 10:36:58 -05001030 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001031}
Michal Hocko33398cf2015-09-08 15:01:02 -07001032EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -08001033
Dan Schatzberg04f94e32021-06-28 19:38:18 -07001034static __always_inline struct mem_cgroup *active_memcg(void)
1035{
Vasily Averin55a68c82021-09-02 14:55:49 -07001036 if (!in_task())
Dan Schatzberg04f94e32021-06-28 19:38:18 -07001037 return this_cpu_read(int_active_memcg);
1038 else
1039 return current->active_memcg;
1040}
1041
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001042/**
1043 * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg.
1044 * @mm: mm from which memcg should be extracted. It can be NULL.
1045 *
Dan Schatzberg04f94e32021-06-28 19:38:18 -07001046 * Obtain a reference on mm->memcg and returns it if successful. If mm
1047 * is NULL, then the memcg is chosen as follows:
1048 * 1) The active memcg, if set.
1049 * 2) current->mm->memcg, if available
1050 * 3) root memcg
1051 * If mem_cgroup is disabled, NULL is returned.
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001052 */
1053struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001054{
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001055 struct mem_cgroup *memcg;
1056
1057 if (mem_cgroup_disabled())
1058 return NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001059
Muchun Song2884b6b2021-06-28 19:37:50 -07001060 /*
1061 * Page cache insertions can happen without an
1062 * actual mm context, e.g. during disk probing
1063 * on boot, loopback IO, acct() writes etc.
1064 *
1065 * No need to css_get on root memcg as the reference
1066 * counting is disabled on the root level in the
1067 * cgroup core. See CSS_NO_REF.
1068 */
Dan Schatzberg04f94e32021-06-28 19:38:18 -07001069 if (unlikely(!mm)) {
1070 memcg = active_memcg();
1071 if (unlikely(memcg)) {
1072 /* remote memcg must hold a ref */
1073 css_get(&memcg->css);
1074 return memcg;
1075 }
1076 mm = current->mm;
1077 if (unlikely(!mm))
1078 return root_mem_cgroup;
1079 }
Muchun Song2884b6b2021-06-28 19:37:50 -07001080
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001081 rcu_read_lock();
1082 do {
Muchun Song2884b6b2021-06-28 19:37:50 -07001083 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1084 if (unlikely(!memcg))
Johannes Weinerdf381972014-04-07 15:37:43 -07001085 memcg = root_mem_cgroup;
Roman Gushchin00d484f2019-11-15 17:34:43 -08001086 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001087 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001088 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001089}
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001090EXPORT_SYMBOL(get_mem_cgroup_from_mm);
1091
Shakeel Buttf745c6f2018-08-17 15:46:44 -07001092/**
Nhat Pham4b569382023-10-06 11:46:26 -07001093 * get_mem_cgroup_from_current - Obtain a reference on current task's memcg.
1094 */
1095struct mem_cgroup *get_mem_cgroup_from_current(void)
1096{
1097 struct mem_cgroup *memcg;
1098
1099 if (mem_cgroup_disabled())
1100 return NULL;
1101
1102again:
1103 rcu_read_lock();
1104 memcg = mem_cgroup_from_task(current);
1105 if (!css_tryget(&memcg->css)) {
1106 rcu_read_unlock();
1107 goto again;
1108 }
1109 rcu_read_unlock();
1110 return memcg;
1111}
1112
1113/**
Johannes Weiner56600482012-01-12 17:17:59 -08001114 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1115 * @root: hierarchy root
1116 * @prev: previously returned memcg, NULL on first invocation
1117 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1118 *
1119 * Returns references to children of the hierarchy below @root, or
1120 * @root itself, or %NULL after a full round-trip.
1121 *
1122 * Caller must pass the return value in @prev on subsequent
1123 * invocations for reference counting, or use mem_cgroup_iter_break()
1124 * to cancel a hierarchy walk before the round-trip is complete.
1125 *
Miaohe Lin05bdc522020-10-13 16:52:45 -07001126 * Reclaimers can specify a node in @reclaim to divide up the memcgs
1127 * in the hierarchy among all concurrent reclaimers operating on the
1128 * same node.
Johannes Weiner56600482012-01-12 17:17:59 -08001129 */
Andrew Morton694fbc02013-09-24 15:27:37 -07001130struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -08001131 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -07001132 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001133{
Kees Cook3f649ab2020-06-03 13:09:38 -07001134 struct mem_cgroup_reclaim_iter *iter;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001135 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001136 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001137 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001138
Andrew Morton694fbc02013-09-24 15:27:37 -07001139 if (mem_cgroup_disabled())
1140 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001141
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001142 if (!root)
1143 root = root_mem_cgroup;
1144
Michal Hocko542f85f2013-04-29 15:07:15 -07001145 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001146
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001147 if (reclaim) {
Mel Gormanef8f2322016-07-28 15:46:05 -07001148 struct mem_cgroup_per_node *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001149
Johannes Weinera3747b52021-04-29 22:56:14 -07001150 mz = root->nodeinfo[reclaim->pgdat->node_id];
Yafang Shao9da83f32019-11-30 17:50:03 -08001151 iter = &mz->iter;
Michal Hocko5f578162013-04-29 15:07:17 -07001152
Wei Yanga9320aa2022-04-28 23:15:59 -07001153 /*
1154 * On start, join the current reclaim iteration cycle.
1155 * Exit when a concurrent walker completes it.
1156 */
1157 if (!prev)
1158 reclaim->generation = iter->generation;
1159 else if (reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -07001160 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001161
Vladimir Davydov6df38682015-12-29 14:54:10 -08001162 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001163 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -08001164 if (!pos || css_tryget(&pos->css))
1165 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001166 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -08001167 * css reference reached zero, so iter->position will
1168 * be cleared by ->css_released. However, we should not
1169 * rely on this happening soon, because ->css_released
1170 * is called from a work queue, and by busy-waiting we
1171 * might block it. So we clear iter->position right
1172 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001173 */
Vladimir Davydov6df38682015-12-29 14:54:10 -08001174 (void)cmpxchg(&iter->position, pos, NULL);
1175 }
Wei Yang89d83302022-04-28 23:15:58 -07001176 } else if (prev) {
1177 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001178 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001179
1180 if (pos)
1181 css = &pos->css;
1182
1183 for (;;) {
1184 css = css_next_descendant_pre(css, &root->css);
1185 if (!css) {
1186 /*
1187 * Reclaimers share the hierarchy walk, and a
1188 * new one might jump in right at the end of
1189 * the hierarchy - make sure they see at least
1190 * one group and restart from the beginning.
1191 */
1192 if (!prev)
1193 continue;
1194 break;
1195 }
1196
1197 /*
1198 * Verify the css and acquire a reference. The root
1199 * is provided by the caller, so we know it's alive
1200 * and kicking, and don't take an extra reference.
1201 */
Wei Yang41555da2022-04-28 23:15:58 -07001202 if (css == &root->css || css_tryget(css)) {
1203 memcg = mem_cgroup_from_css(css);
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001204 break;
Wei Yang41555da2022-04-28 23:15:58 -07001205 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001206 }
1207
1208 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001209 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -08001210 * The position could have already been updated by a competing
1211 * thread, so check that the value hasn't changed since we read
1212 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001213 */
Vladimir Davydov6df38682015-12-29 14:54:10 -08001214 (void)cmpxchg(&iter->position, pos, memcg);
1215
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001216 if (pos)
1217 css_put(&pos->css);
1218
1219 if (!memcg)
1220 iter->generation++;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001221 }
1222
Michal Hocko542f85f2013-04-29 15:07:15 -07001223out_unlock:
1224 rcu_read_unlock();
Michal Hockoc40046f2013-04-29 15:07:14 -07001225 if (prev && prev != root)
1226 css_put(&prev->css);
1227
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001228 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001229}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001230
Johannes Weiner56600482012-01-12 17:17:59 -08001231/**
1232 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1233 * @root: hierarchy root
1234 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1235 */
1236void mem_cgroup_iter_break(struct mem_cgroup *root,
1237 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001238{
1239 if (!root)
1240 root = root_mem_cgroup;
1241 if (prev && prev != root)
1242 css_put(&prev->css);
1243}
1244
Miles Chen54a83d62019-08-13 15:37:28 -07001245static void __invalidate_reclaim_iterators(struct mem_cgroup *from,
1246 struct mem_cgroup *dead_memcg)
Vladimir Davydov6df38682015-12-29 14:54:10 -08001247{
Vladimir Davydov6df38682015-12-29 14:54:10 -08001248 struct mem_cgroup_reclaim_iter *iter;
Mel Gormanef8f2322016-07-28 15:46:05 -07001249 struct mem_cgroup_per_node *mz;
1250 int nid;
Vladimir Davydov6df38682015-12-29 14:54:10 -08001251
Miles Chen54a83d62019-08-13 15:37:28 -07001252 for_each_node(nid) {
Johannes Weinera3747b52021-04-29 22:56:14 -07001253 mz = from->nodeinfo[nid];
Yafang Shao9da83f32019-11-30 17:50:03 -08001254 iter = &mz->iter;
1255 cmpxchg(&iter->position, dead_memcg, NULL);
Vladimir Davydov6df38682015-12-29 14:54:10 -08001256 }
1257}
1258
Miles Chen54a83d62019-08-13 15:37:28 -07001259static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
1260{
1261 struct mem_cgroup *memcg = dead_memcg;
1262 struct mem_cgroup *last;
1263
1264 do {
1265 __invalidate_reclaim_iterators(memcg, dead_memcg);
1266 last = memcg;
1267 } while ((memcg = parent_mem_cgroup(memcg)));
1268
1269 /*
xupandab8dd3ee2022-08-15 06:51:04 +00001270 * When cgroup1 non-hierarchy mode is used,
Miles Chen54a83d62019-08-13 15:37:28 -07001271 * parent_mem_cgroup() does not walk all the way up to the
1272 * cgroup root (root_mem_cgroup). So we have to handle
1273 * dead_memcg from cgroup root separately.
1274 */
Kamalesh Babulal7848ed62022-09-30 19:14:33 +05301275 if (!mem_cgroup_is_root(last))
Miles Chen54a83d62019-08-13 15:37:28 -07001276 __invalidate_reclaim_iterators(root_mem_cgroup,
1277 dead_memcg);
1278}
1279
Johannes Weiner925b7672012-01-12 17:18:15 -08001280/**
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001281 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
1282 * @memcg: hierarchy root
1283 * @fn: function to call for each task
1284 * @arg: argument passed to @fn
1285 *
1286 * This function iterates over tasks attached to @memcg or to any of its
1287 * descendants and calls @fn for each task. If @fn returns a non-zero
ZhangPeng025b7792023-06-16 14:30:30 +08001288 * value, the function breaks the iteration loop. Otherwise, it will iterate
1289 * over all tasks and return 0.
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001290 *
1291 * This function must not be called for the root memory cgroup.
1292 */
ZhangPeng025b7792023-06-16 14:30:30 +08001293void mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
1294 int (*fn)(struct task_struct *, void *), void *arg)
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001295{
1296 struct mem_cgroup *iter;
1297 int ret = 0;
1298
Kamalesh Babulal7848ed62022-09-30 19:14:33 +05301299 BUG_ON(mem_cgroup_is_root(memcg));
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001300
1301 for_each_mem_cgroup_tree(iter, memcg) {
1302 struct css_task_iter it;
1303 struct task_struct *task;
1304
Tetsuo Handaf168a9a2019-07-11 21:00:20 -07001305 css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001306 while (!ret && (task = css_task_iter_next(&it)))
1307 ret = fn(task, arg);
1308 css_task_iter_end(&it);
1309 if (ret) {
1310 mem_cgroup_iter_break(memcg, iter);
1311 break;
1312 }
1313 }
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001314}
1315
Alex Shi6168d0d2020-12-15 12:34:29 -08001316#ifdef CONFIG_DEBUG_VM
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001317void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio)
Alex Shi6168d0d2020-12-15 12:34:29 -08001318{
1319 struct mem_cgroup *memcg;
1320
1321 if (mem_cgroup_disabled())
1322 return;
1323
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001324 memcg = folio_memcg(folio);
Alex Shi6168d0d2020-12-15 12:34:29 -08001325
1326 if (!memcg)
Kamalesh Babulal7848ed62022-09-30 19:14:33 +05301327 VM_BUG_ON_FOLIO(!mem_cgroup_is_root(lruvec_memcg(lruvec)), folio);
Alex Shi6168d0d2020-12-15 12:34:29 -08001328 else
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001329 VM_BUG_ON_FOLIO(lruvec_memcg(lruvec) != memcg, folio);
Alex Shi6168d0d2020-12-15 12:34:29 -08001330}
1331#endif
1332
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001333/**
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001334 * folio_lruvec_lock - Lock the lruvec for a folio.
1335 * @folio: Pointer to the folio.
Alex Shi6168d0d2020-12-15 12:34:29 -08001336 *
Alex Shid7e3aba2021-02-24 12:03:47 -08001337 * These functions are safe to use under any of the following conditions:
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001338 * - folio locked
1339 * - folio_test_lru false
1340 * - folio_memcg_lock()
1341 * - folio frozen (refcount of 0)
1342 *
1343 * Return: The lruvec this folio is on with its lock held.
Alex Shi6168d0d2020-12-15 12:34:29 -08001344 */
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001345struct lruvec *folio_lruvec_lock(struct folio *folio)
Alex Shi6168d0d2020-12-15 12:34:29 -08001346{
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001347 struct lruvec *lruvec = folio_lruvec(folio);
Alex Shi6168d0d2020-12-15 12:34:29 -08001348
Alex Shi6168d0d2020-12-15 12:34:29 -08001349 spin_lock(&lruvec->lru_lock);
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001350 lruvec_memcg_debug(lruvec, folio);
Alex Shi6168d0d2020-12-15 12:34:29 -08001351
1352 return lruvec;
1353}
1354
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001355/**
1356 * folio_lruvec_lock_irq - Lock the lruvec for a folio.
1357 * @folio: Pointer to the folio.
1358 *
1359 * These functions are safe to use under any of the following conditions:
1360 * - folio locked
1361 * - folio_test_lru false
1362 * - folio_memcg_lock()
1363 * - folio frozen (refcount of 0)
1364 *
1365 * Return: The lruvec this folio is on with its lock held and interrupts
1366 * disabled.
1367 */
1368struct lruvec *folio_lruvec_lock_irq(struct folio *folio)
Alex Shi6168d0d2020-12-15 12:34:29 -08001369{
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001370 struct lruvec *lruvec = folio_lruvec(folio);
Alex Shi6168d0d2020-12-15 12:34:29 -08001371
Alex Shi6168d0d2020-12-15 12:34:29 -08001372 spin_lock_irq(&lruvec->lru_lock);
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001373 lruvec_memcg_debug(lruvec, folio);
Alex Shi6168d0d2020-12-15 12:34:29 -08001374
1375 return lruvec;
1376}
1377
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001378/**
1379 * folio_lruvec_lock_irqsave - Lock the lruvec for a folio.
1380 * @folio: Pointer to the folio.
1381 * @flags: Pointer to irqsave flags.
1382 *
1383 * These functions are safe to use under any of the following conditions:
1384 * - folio locked
1385 * - folio_test_lru false
1386 * - folio_memcg_lock()
1387 * - folio frozen (refcount of 0)
1388 *
1389 * Return: The lruvec this folio is on with its lock held and interrupts
1390 * disabled.
1391 */
1392struct lruvec *folio_lruvec_lock_irqsave(struct folio *folio,
1393 unsigned long *flags)
Alex Shi6168d0d2020-12-15 12:34:29 -08001394{
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001395 struct lruvec *lruvec = folio_lruvec(folio);
Alex Shi6168d0d2020-12-15 12:34:29 -08001396
Alex Shi6168d0d2020-12-15 12:34:29 -08001397 spin_lock_irqsave(&lruvec->lru_lock, *flags);
Matthew Wilcox (Oracle)e809c3f2021-06-28 21:59:47 -04001398 lruvec_memcg_debug(lruvec, folio);
Alex Shi6168d0d2020-12-15 12:34:29 -08001399
1400 return lruvec;
1401}
1402
1403/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001404 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1405 * @lruvec: mem_cgroup per zone lru vector
1406 * @lru: index of lru list the page is sitting on
Michal Hockob4536f0c82017-01-10 16:58:04 -08001407 * @zid: zone id of the accounted pages
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001408 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001409 *
Hugh Dickinsca707232016-05-19 17:12:35 -07001410 * This function must be called under lru_lock, just before a page is added
Hugh Dickins07ca7602022-02-14 18:29:54 -08001411 * to or just after a page is removed from an lru list.
Johannes Weiner925b7672012-01-12 17:18:15 -08001412 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001413void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
Michal Hockob4536f0c82017-01-10 16:58:04 -08001414 int zid, int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001415{
Mel Gormanef8f2322016-07-28 15:46:05 -07001416 struct mem_cgroup_per_node *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001417 unsigned long *lru_size;
Hugh Dickinsca707232016-05-19 17:12:35 -07001418 long size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001419
1420 if (mem_cgroup_disabled())
1421 return;
1422
Mel Gormanef8f2322016-07-28 15:46:05 -07001423 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
Michal Hockob4536f0c82017-01-10 16:58:04 -08001424 lru_size = &mz->lru_zone_size[zid][lru];
Hugh Dickinsca707232016-05-19 17:12:35 -07001425
1426 if (nr_pages < 0)
1427 *lru_size += nr_pages;
1428
1429 size = *lru_size;
Michal Hockob4536f0c82017-01-10 16:58:04 -08001430 if (WARN_ONCE(size < 0,
1431 "%s(%p, %d, %d): lru_size %ld\n",
1432 __func__, lruvec, lru, nr_pages, size)) {
Hugh Dickinsca707232016-05-19 17:12:35 -07001433 VM_BUG_ON(1);
1434 *lru_size = 0;
1435 }
1436
1437 if (nr_pages > 0)
1438 *lru_size += nr_pages;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001439}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001440
Johannes Weiner19942822011-02-01 15:52:43 -08001441/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001442 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001443 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001444 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001445 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001446 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001447 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001448static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001449{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001450 unsigned long margin = 0;
1451 unsigned long count;
1452 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001453
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001454 count = page_counter_read(&memcg->memory);
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001455 limit = READ_ONCE(memcg->memory.max);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001456 if (count < limit)
1457 margin = limit - count;
1458
Johannes Weiner7941d212016-01-14 15:21:23 -08001459 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001460 count = page_counter_read(&memcg->memsw);
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001461 limit = READ_ONCE(memcg->memsw.max);
Kaixu Xia1c4448e2020-06-01 21:49:36 -07001462 if (count < limit)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001463 margin = min(margin, limit - count);
Li RongQingcbedbac2016-05-27 14:27:43 -07001464 else
1465 margin = 0;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001466 }
1467
1468 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001469}
1470
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001471/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001472 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001473 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001474 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1475 * moving cgroups. This is for waiting at high-memory pressure
1476 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001477 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001478static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001479{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001480 struct mem_cgroup *from;
1481 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001482 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001483 /*
1484 * Unlike task_move routines, we access mc.to, mc.from not under
1485 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1486 */
1487 spin_lock(&mc.lock);
1488 from = mc.from;
1489 to = mc.to;
1490 if (!from)
1491 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001492
Johannes Weiner2314b422014-12-10 15:44:33 -08001493 ret = mem_cgroup_is_descendant(from, memcg) ||
1494 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001495unlock:
1496 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001497 return ret;
1498}
1499
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001500static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001501{
1502 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001503 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001504 DEFINE_WAIT(wait);
1505 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1506 /* moving charge context might have finished. */
1507 if (mc.moving_task)
1508 schedule();
1509 finish_wait(&mc.waitq, &wait);
1510 return true;
1511 }
1512 }
1513 return false;
1514}
1515
Muchun Song5f9a4f42020-10-13 16:52:59 -07001516struct memory_stat {
1517 const char *name;
Muchun Song5f9a4f42020-10-13 16:52:59 -07001518 unsigned int idx;
1519};
1520
Muchun Song57b28472021-02-24 12:03:31 -08001521static const struct memory_stat memory_stats[] = {
Muchun Songfff66b72021-02-24 12:03:43 -08001522 { "anon", NR_ANON_MAPPED },
1523 { "file", NR_FILE_PAGES },
Yosry Ahmeda8c49af2022-03-22 14:40:10 -07001524 { "kernel", MEMCG_KMEM },
Muchun Songfff66b72021-02-24 12:03:43 -08001525 { "kernel_stack", NR_KERNEL_STACK_KB },
1526 { "pagetables", NR_PAGETABLE },
Yosry Ahmedebc97a52022-08-23 00:46:36 +00001527 { "sec_pagetables", NR_SECONDARY_PAGETABLE },
Muchun Songfff66b72021-02-24 12:03:43 -08001528 { "percpu", MEMCG_PERCPU_B },
1529 { "sock", MEMCG_SOCK },
Shakeel Butt4e5aa1f2022-01-14 14:05:45 -08001530 { "vmalloc", MEMCG_VMALLOC },
Muchun Songfff66b72021-02-24 12:03:43 -08001531 { "shmem", NR_SHMEM },
Johannes Weinerf4840cc2022-05-19 14:08:53 -07001532#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP)
1533 { "zswap", MEMCG_ZSWAP_B },
1534 { "zswapped", MEMCG_ZSWAPPED },
1535#endif
Muchun Songfff66b72021-02-24 12:03:43 -08001536 { "file_mapped", NR_FILE_MAPPED },
1537 { "file_dirty", NR_FILE_DIRTY },
1538 { "file_writeback", NR_WRITEBACK },
Shakeel Buttb6038942021-02-24 12:03:55 -08001539#ifdef CONFIG_SWAP
1540 { "swapcached", NR_SWAPCACHE },
1541#endif
Muchun Song5f9a4f42020-10-13 16:52:59 -07001542#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Muchun Songfff66b72021-02-24 12:03:43 -08001543 { "anon_thp", NR_ANON_THPS },
1544 { "file_thp", NR_FILE_THPS },
1545 { "shmem_thp", NR_SHMEM_THPS },
Muchun Song5f9a4f42020-10-13 16:52:59 -07001546#endif
Muchun Songfff66b72021-02-24 12:03:43 -08001547 { "inactive_anon", NR_INACTIVE_ANON },
1548 { "active_anon", NR_ACTIVE_ANON },
1549 { "inactive_file", NR_INACTIVE_FILE },
1550 { "active_file", NR_ACTIVE_FILE },
1551 { "unevictable", NR_UNEVICTABLE },
1552 { "slab_reclaimable", NR_SLAB_RECLAIMABLE_B },
1553 { "slab_unreclaimable", NR_SLAB_UNRECLAIMABLE_B },
Muchun Song5f9a4f42020-10-13 16:52:59 -07001554
1555 /* The memory events */
Muchun Songfff66b72021-02-24 12:03:43 -08001556 { "workingset_refault_anon", WORKINGSET_REFAULT_ANON },
1557 { "workingset_refault_file", WORKINGSET_REFAULT_FILE },
1558 { "workingset_activate_anon", WORKINGSET_ACTIVATE_ANON },
1559 { "workingset_activate_file", WORKINGSET_ACTIVATE_FILE },
1560 { "workingset_restore_anon", WORKINGSET_RESTORE_ANON },
1561 { "workingset_restore_file", WORKINGSET_RESTORE_FILE },
1562 { "workingset_nodereclaim", WORKINGSET_NODERECLAIM },
Muchun Song5f9a4f42020-10-13 16:52:59 -07001563};
1564
Yosry Ahmedff841a02023-09-22 17:57:39 +00001565/* The actual unit of the state item, not the same as the output unit */
Muchun Songfff66b72021-02-24 12:03:43 -08001566static int memcg_page_state_unit(int item)
1567{
1568 switch (item) {
1569 case MEMCG_PERCPU_B:
Johannes Weinerf4840cc2022-05-19 14:08:53 -07001570 case MEMCG_ZSWAP_B:
Muchun Songfff66b72021-02-24 12:03:43 -08001571 case NR_SLAB_RECLAIMABLE_B:
1572 case NR_SLAB_UNRECLAIMABLE_B:
Muchun Songfff66b72021-02-24 12:03:43 -08001573 return 1;
1574 case NR_KERNEL_STACK_KB:
1575 return SZ_1K;
1576 default:
1577 return PAGE_SIZE;
1578 }
1579}
1580
Yosry Ahmedff841a02023-09-22 17:57:39 +00001581/* Translate stat items to the correct unit for memory.stat output */
1582static int memcg_page_state_output_unit(int item)
1583{
1584 /*
1585 * Workingset state is actually in pages, but we export it to userspace
1586 * as a scalar count of events, so special case it here.
1587 */
1588 switch (item) {
1589 case WORKINGSET_REFAULT_ANON:
1590 case WORKINGSET_REFAULT_FILE:
1591 case WORKINGSET_ACTIVATE_ANON:
1592 case WORKINGSET_ACTIVATE_FILE:
1593 case WORKINGSET_RESTORE_ANON:
1594 case WORKINGSET_RESTORE_FILE:
1595 case WORKINGSET_NODERECLAIM:
1596 return 1;
1597 default:
1598 return memcg_page_state_unit(item);
1599 }
1600}
1601
Muchun Songfff66b72021-02-24 12:03:43 -08001602static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg,
1603 int item)
1604{
Yosry Ahmedff841a02023-09-22 17:57:39 +00001605 return memcg_page_state(memcg, item) *
1606 memcg_page_state_output_unit(item);
1607}
1608
1609static inline unsigned long memcg_page_state_local_output(
1610 struct mem_cgroup *memcg, int item)
1611{
1612 return memcg_page_state_local(memcg, item) *
1613 memcg_page_state_output_unit(item);
Muchun Songfff66b72021-02-24 12:03:43 -08001614}
1615
Yosry Ahmeddddb44f2023-04-28 13:24:06 +00001616static void memcg_stat_format(struct mem_cgroup *memcg, struct seq_buf *s)
Johannes Weinerc8713d02019-07-11 20:55:59 -07001617{
Johannes Weinerc8713d02019-07-11 20:55:59 -07001618 int i;
Johannes Weiner71cd3112017-05-03 14:55:13 -07001619
Johannes Weinerc8713d02019-07-11 20:55:59 -07001620 /*
1621 * Provide statistics on the state of the memory subsystem as
1622 * well as cumulative event counters that show past behavior.
1623 *
1624 * This list is ordered following a combination of these gradients:
1625 * 1) generic big picture -> specifics and details
1626 * 2) reflecting userspace activity -> reflecting kernel heuristics
1627 *
1628 * Current memory state:
1629 */
Shakeel Buttfd25a9e2021-11-05 13:37:34 -07001630 mem_cgroup_flush_stats();
Johannes Weinerc8713d02019-07-11 20:55:59 -07001631
Muchun Song5f9a4f42020-10-13 16:52:59 -07001632 for (i = 0; i < ARRAY_SIZE(memory_stats); i++) {
1633 u64 size;
Johannes Weinerc8713d02019-07-11 20:55:59 -07001634
Muchun Songfff66b72021-02-24 12:03:43 -08001635 size = memcg_page_state_output(memcg, memory_stats[i].idx);
Yosry Ahmed5b423602023-04-28 13:24:05 +00001636 seq_buf_printf(s, "%s %llu\n", memory_stats[i].name, size);
Johannes Weinerc8713d02019-07-11 20:55:59 -07001637
Muchun Song5f9a4f42020-10-13 16:52:59 -07001638 if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) {
Muchun Songfff66b72021-02-24 12:03:43 -08001639 size += memcg_page_state_output(memcg,
1640 NR_SLAB_RECLAIMABLE_B);
Yosry Ahmed5b423602023-04-28 13:24:05 +00001641 seq_buf_printf(s, "slab %llu\n", size);
Muchun Song5f9a4f42020-10-13 16:52:59 -07001642 }
1643 }
Johannes Weinerc8713d02019-07-11 20:55:59 -07001644
1645 /* Accumulated memory events */
Yosry Ahmed5b423602023-04-28 13:24:05 +00001646 seq_buf_printf(s, "pgscan %lu\n",
Johannes Weinerc8713d02019-07-11 20:55:59 -07001647 memcg_events(memcg, PGSCAN_KSWAPD) +
Johannes Weiner57e9cc52022-10-26 14:01:33 -04001648 memcg_events(memcg, PGSCAN_DIRECT) +
1649 memcg_events(memcg, PGSCAN_KHUGEPAGED));
Yosry Ahmed5b423602023-04-28 13:24:05 +00001650 seq_buf_printf(s, "pgsteal %lu\n",
Johannes Weinerc8713d02019-07-11 20:55:59 -07001651 memcg_events(memcg, PGSTEAL_KSWAPD) +
Johannes Weiner57e9cc52022-10-26 14:01:33 -04001652 memcg_events(memcg, PGSTEAL_DIRECT) +
1653 memcg_events(memcg, PGSTEAL_KHUGEPAGED));
Johannes Weinerc8713d02019-07-11 20:55:59 -07001654
Shakeel Butt8278f1c2022-09-07 04:35:37 +00001655 for (i = 0; i < ARRAY_SIZE(memcg_vm_event_stat); i++) {
1656 if (memcg_vm_event_stat[i] == PGPGIN ||
1657 memcg_vm_event_stat[i] == PGPGOUT)
1658 continue;
1659
Yosry Ahmed5b423602023-04-28 13:24:05 +00001660 seq_buf_printf(s, "%s %lu\n",
Qi Zheng673520f2022-06-04 16:22:09 +08001661 vm_event_name(memcg_vm_event_stat[i]),
1662 memcg_events(memcg, memcg_vm_event_stat[i]));
Shakeel Butt8278f1c2022-09-07 04:35:37 +00001663 }
Johannes Weinerc8713d02019-07-11 20:55:59 -07001664
1665 /* The above should easily fit into one page */
Yosry Ahmed5b423602023-04-28 13:24:05 +00001666 WARN_ON_ONCE(seq_buf_has_overflowed(s));
Johannes Weinerc8713d02019-07-11 20:55:59 -07001667}
Johannes Weiner71cd3112017-05-03 14:55:13 -07001668
Yosry Ahmeddddb44f2023-04-28 13:24:06 +00001669static void memcg1_stat_format(struct mem_cgroup *memcg, struct seq_buf *s);
1670
1671static void memory_stat_format(struct mem_cgroup *memcg, struct seq_buf *s)
1672{
1673 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
1674 memcg_stat_format(memcg, s);
1675 else
1676 memcg1_stat_format(memcg, s);
1677 WARN_ON_ONCE(seq_buf_has_overflowed(s));
1678}
1679
Balbir Singhe2224322009-04-02 16:57:39 -07001680/**
yuzhoujianf0c867d2018-12-28 00:36:10 -08001681 * mem_cgroup_print_oom_context: Print OOM information relevant to
1682 * memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001683 * @memcg: The memory cgroup that went over limit
1684 * @p: Task that is going to be killed
1685 *
1686 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1687 * enabled
1688 */
yuzhoujianf0c867d2018-12-28 00:36:10 -08001689void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p)
1690{
1691 rcu_read_lock();
1692
1693 if (memcg) {
1694 pr_cont(",oom_memcg=");
1695 pr_cont_cgroup_path(memcg->css.cgroup);
1696 } else
1697 pr_cont(",global_oom");
1698 if (p) {
1699 pr_cont(",task_memcg=");
1700 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1701 }
1702 rcu_read_unlock();
1703}
1704
1705/**
1706 * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to
1707 * memory controller.
1708 * @memcg: The memory cgroup that went over limit
1709 */
1710void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
Balbir Singhe2224322009-04-02 16:57:39 -07001711{
Tetsuo Handa68aaee12022-07-22 19:45:39 +09001712 /* Use static buffer, for the caller is holding oom_lock. */
1713 static char buf[PAGE_SIZE];
Yosry Ahmed5b423602023-04-28 13:24:05 +00001714 struct seq_buf s;
Tetsuo Handa68aaee12022-07-22 19:45:39 +09001715
1716 lockdep_assert_held(&oom_lock);
Balbir Singhe2224322009-04-02 16:57:39 -07001717
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001718 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1719 K((u64)page_counter_read(&memcg->memory)),
Chris Down15b42562020-04-01 21:07:20 -07001720 K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt);
Johannes Weinerc8713d02019-07-11 20:55:59 -07001721 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
1722 pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n",
1723 K((u64)page_counter_read(&memcg->swap)),
Chris Down32d087c2020-04-01 21:07:30 -07001724 K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt);
Johannes Weinerc8713d02019-07-11 20:55:59 -07001725 else {
1726 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1727 K((u64)page_counter_read(&memcg->memsw)),
1728 K((u64)memcg->memsw.max), memcg->memsw.failcnt);
1729 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1730 K((u64)page_counter_read(&memcg->kmem)),
1731 K((u64)memcg->kmem.max), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001732 }
Johannes Weinerc8713d02019-07-11 20:55:59 -07001733
1734 pr_info("Memory cgroup stats for ");
1735 pr_cont_cgroup_path(memcg->css.cgroup);
1736 pr_cont(":");
Yosry Ahmed5b423602023-04-28 13:24:05 +00001737 seq_buf_init(&s, buf, sizeof(buf));
1738 memory_stat_format(memcg, &s);
1739 seq_buf_do_printk(&s, KERN_INFO);
Balbir Singhe2224322009-04-02 16:57:39 -07001740}
1741
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001742/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001743 * Return the memory (and swap, if configured) limit for a memcg.
1744 */
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001745unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001746{
Waiman Long8d387a52020-10-13 16:52:52 -07001747 unsigned long max = READ_ONCE(memcg->memory.max);
David Rientjesa63d83f2010-08-09 17:19:46 -07001748
Johannes Weinerb94c4e92022-09-26 09:57:03 -04001749 if (do_memsw_account()) {
Waiman Long8d387a52020-10-13 16:52:52 -07001750 if (mem_cgroup_swappiness(memcg)) {
1751 /* Calculate swap excess capacity from memsw limit */
1752 unsigned long swap = READ_ONCE(memcg->memsw.max) - max;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001753
Waiman Long8d387a52020-10-13 16:52:52 -07001754 max += min(swap, (unsigned long)total_swap_pages);
1755 }
Johannes Weinerb94c4e92022-09-26 09:57:03 -04001756 } else {
1757 if (mem_cgroup_swappiness(memcg))
1758 max += min(READ_ONCE(memcg->swap.max),
1759 (unsigned long)total_swap_pages);
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001760 }
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001761 return max;
David Rientjesa63d83f2010-08-09 17:19:46 -07001762}
1763
Chris Down9783aa92019-10-06 17:58:32 -07001764unsigned long mem_cgroup_size(struct mem_cgroup *memcg)
1765{
1766 return page_counter_read(&memcg->memory);
1767}
1768
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001769static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001770 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001771{
David Rientjes6e0fc462015-09-08 15:00:36 -07001772 struct oom_control oc = {
1773 .zonelist = NULL,
1774 .nodemask = NULL,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001775 .memcg = memcg,
David Rientjes6e0fc462015-09-08 15:00:36 -07001776 .gfp_mask = gfp_mask,
1777 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001778 };
Yafang Shao1378b372020-08-06 23:22:08 -07001779 bool ret = true;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001780
Tetsuo Handa7775fac2019-03-05 15:46:47 -08001781 if (mutex_lock_killable(&oom_lock))
1782 return true;
Yafang Shao1378b372020-08-06 23:22:08 -07001783
1784 if (mem_cgroup_margin(memcg) >= (1 << order))
1785 goto unlock;
1786
Tetsuo Handa7775fac2019-03-05 15:46:47 -08001787 /*
1788 * A few threads which were not waiting at mutex_lock_killable() can
1789 * fail to bail out. Therefore, check again after holding oom_lock.
1790 */
Vasily Averina4ebf1b2021-11-05 13:38:09 -07001791 ret = task_is_dying() || out_of_memory(&oc);
Yafang Shao1378b372020-08-06 23:22:08 -07001792
1793unlock:
Johannes Weinerdc564012015-06-24 16:57:19 -07001794 mutex_unlock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001795 return ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001796}
1797
Andrew Morton0608f432013-09-24 15:27:41 -07001798static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
Mel Gormanef8f2322016-07-28 15:46:05 -07001799 pg_data_t *pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001800 gfp_t gfp_mask,
1801 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001802{
Andrew Morton0608f432013-09-24 15:27:41 -07001803 struct mem_cgroup *victim = NULL;
1804 int total = 0;
1805 int loop = 0;
1806 unsigned long excess;
1807 unsigned long nr_scanned;
1808 struct mem_cgroup_reclaim_cookie reclaim = {
Mel Gormanef8f2322016-07-28 15:46:05 -07001809 .pgdat = pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001810 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001811
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001812 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001813
Andrew Morton0608f432013-09-24 15:27:41 -07001814 while (1) {
1815 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1816 if (!victim) {
1817 loop++;
1818 if (loop >= 2) {
1819 /*
1820 * If we have not been able to reclaim
1821 * anything, it might because there are
1822 * no reclaimable pages under this hierarchy
1823 */
1824 if (!total)
1825 break;
1826 /*
1827 * We want to do more targeted reclaim.
1828 * excess >> 2 is not to excessive so as to
1829 * reclaim too much, nor too less that we keep
1830 * coming back to reclaim from this cgroup
1831 */
1832 if (total >= (excess >> 2) ||
1833 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1834 break;
1835 }
1836 continue;
1837 }
Mel Gormana9dd0a82016-07-28 15:46:02 -07001838 total += mem_cgroup_shrink_node(victim, gfp_mask, false,
Mel Gormanef8f2322016-07-28 15:46:05 -07001839 pgdat, &nr_scanned);
Andrew Morton0608f432013-09-24 15:27:41 -07001840 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001841 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001842 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001843 }
Andrew Morton0608f432013-09-24 15:27:41 -07001844 mem_cgroup_iter_break(root_memcg, victim);
1845 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001846}
1847
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001848#ifdef CONFIG_LOCKDEP
1849static struct lockdep_map memcg_oom_lock_dep_map = {
1850 .name = "memcg_oom_lock",
1851};
1852#endif
1853
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001854static DEFINE_SPINLOCK(memcg_oom_lock);
1855
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001856/*
1857 * Check OOM-Killer is already running under our hierarchy.
1858 * If someone is running, return false.
1859 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001860static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001861{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001862 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001863
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001864 spin_lock(&memcg_oom_lock);
1865
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001866 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001867 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001868 /*
1869 * this subtree of our hierarchy is already locked
1870 * so we cannot give a lock.
1871 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001872 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001873 mem_cgroup_iter_break(memcg, iter);
1874 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001875 } else
1876 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001877 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001878
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001879 if (failed) {
1880 /*
1881 * OK, we failed to lock the whole subtree so we have
1882 * to clean up what we set up to the failing subtree
1883 */
1884 for_each_mem_cgroup_tree(iter, memcg) {
1885 if (iter == failed) {
1886 mem_cgroup_iter_break(memcg, iter);
1887 break;
1888 }
1889 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001890 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001891 } else
1892 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001893
1894 spin_unlock(&memcg_oom_lock);
1895
1896 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001897}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001898
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001899static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001900{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001901 struct mem_cgroup *iter;
1902
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001903 spin_lock(&memcg_oom_lock);
Qian Cai5facae42019-09-19 12:09:40 -04001904 mutex_release(&memcg_oom_lock_dep_map, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001905 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001906 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001907 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001908}
1909
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001910static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001911{
1912 struct mem_cgroup *iter;
1913
Tejun Heoc2b42d32015-06-24 16:58:23 -07001914 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001915 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001916 iter->under_oom++;
1917 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001918}
1919
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001920static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001921{
1922 struct mem_cgroup *iter;
1923
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001924 /*
Ingo Molnarf0953a12021-05-06 18:06:47 -07001925 * Be careful about under_oom underflows because a child memcg
Miaohe Lin7a52d4d82020-10-13 16:53:05 -07001926 * could have been added after mem_cgroup_mark_under_oom.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001927 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001928 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001929 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001930 if (iter->under_oom > 0)
1931 iter->under_oom--;
1932 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001933}
1934
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001935static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1936
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001937struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001938 struct mem_cgroup *memcg;
Ingo Molnarac6424b2017-06-20 12:06:13 +02001939 wait_queue_entry_t wait;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001940};
1941
Ingo Molnarac6424b2017-06-20 12:06:13 +02001942static int memcg_oom_wake_function(wait_queue_entry_t *wait,
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001943 unsigned mode, int sync, void *arg)
1944{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001945 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1946 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001947 struct oom_wait_info *oom_wait_info;
1948
1949 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001950 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001951
Johannes Weiner2314b422014-12-10 15:44:33 -08001952 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1953 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001954 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001955 return autoremove_wake_function(wait, mode, sync, arg);
1956}
1957
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001958static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001959{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001960 /*
1961 * For the following lockless ->under_oom test, the only required
1962 * guarantee is that it must see the state asserted by an OOM when
1963 * this function is called as a result of userland actions
1964 * triggered by the notification of the OOM. This is trivially
1965 * achieved by invoking mem_cgroup_mark_under_oom() before
1966 * triggering notification.
1967 */
1968 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001969 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001970}
1971
Shakeel Buttbecdf892022-03-22 14:40:19 -07001972/*
1973 * Returns true if successfully killed one or more processes. Though in some
1974 * corner cases it can return true even without killing any process.
1975 */
1976static bool mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001977{
Shakeel Buttbecdf892022-03-22 14:40:19 -07001978 bool locked, ret;
Michal Hocko7056d3a2018-12-28 00:39:57 -08001979
Michal Hocko29ef6802018-08-17 15:47:11 -07001980 if (order > PAGE_ALLOC_COSTLY_ORDER)
Shakeel Buttbecdf892022-03-22 14:40:19 -07001981 return false;
Michal Hocko29ef6802018-08-17 15:47:11 -07001982
Roman Gushchin7a1adfd2018-10-26 15:09:48 -07001983 memcg_memory_event(memcg, MEMCG_OOM);
1984
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001985 /*
Johannes Weiner494264202013-10-16 13:46:59 -07001986 * We are in the middle of the charge context here, so we
1987 * don't want to block when potentially sitting on a callstack
1988 * that holds all kinds of filesystem and mm locks.
1989 *
Michal Hocko29ef6802018-08-17 15:47:11 -07001990 * cgroup1 allows disabling the OOM killer and waiting for outside
1991 * handling until the charge can succeed; remember the context and put
1992 * the task to sleep at the end of the page fault when all locks are
1993 * released.
Johannes Weiner494264202013-10-16 13:46:59 -07001994 *
Michal Hocko29ef6802018-08-17 15:47:11 -07001995 * On the other hand, in-kernel OOM killer allows for an async victim
1996 * memory reclaim (oom_reaper) and that means that we are not solely
1997 * relying on the oom victim to make a forward progress and we can
1998 * invoke the oom killer here.
1999 *
2000 * Please note that mem_cgroup_out_of_memory might fail to find a
2001 * victim and then we have to bail out from the charge path.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002002 */
Yue Zhao17c56de2023-03-06 23:41:37 +08002003 if (READ_ONCE(memcg->oom_kill_disable)) {
Shakeel Buttbecdf892022-03-22 14:40:19 -07002004 if (current->in_user_fault) {
2005 css_get(&memcg->css);
2006 current->memcg_in_oom = memcg;
2007 current->memcg_oom_gfp_mask = mask;
2008 current->memcg_oom_order = order;
2009 }
2010 return false;
Michal Hocko29ef6802018-08-17 15:47:11 -07002011 }
2012
Michal Hocko7056d3a2018-12-28 00:39:57 -08002013 mem_cgroup_mark_under_oom(memcg);
Michal Hocko29ef6802018-08-17 15:47:11 -07002014
Michal Hocko7056d3a2018-12-28 00:39:57 -08002015 locked = mem_cgroup_oom_trylock(memcg);
2016
2017 if (locked)
2018 mem_cgroup_oom_notify(memcg);
2019
2020 mem_cgroup_unmark_under_oom(memcg);
Shakeel Buttbecdf892022-03-22 14:40:19 -07002021 ret = mem_cgroup_out_of_memory(memcg, mask, order);
Michal Hocko7056d3a2018-12-28 00:39:57 -08002022
2023 if (locked)
2024 mem_cgroup_oom_unlock(memcg);
2025
2026 return ret;
Johannes Weiner494264202013-10-16 13:46:59 -07002027}
2028
2029/**
2030 * mem_cgroup_oom_synchronize - complete memcg OOM handling
2031 * @handle: actually kill/wait or just clean up the OOM state
2032 *
2033 * This has to be called at the end of a page fault if the memcg OOM
2034 * handler was enabled.
2035 *
2036 * Memcg supports userspace OOM handling where failed allocations must
2037 * sleep on a waitqueue until the userspace task resolves the
2038 * situation. Sleeping directly in the charge context with all kinds
2039 * of locks held is not a good idea, instead we remember an OOM state
2040 * in the task and mem_cgroup_oom_synchronize() has to be called at
2041 * the end of the page fault to complete the OOM handling.
2042 *
2043 * Returns %true if an ongoing memcg OOM situation was detected and
2044 * completed, %false otherwise.
2045 */
2046bool mem_cgroup_oom_synchronize(bool handle)
2047{
Tejun Heo626ebc42015-11-05 18:46:09 -08002048 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner494264202013-10-16 13:46:59 -07002049 struct oom_wait_info owait;
2050 bool locked;
2051
2052 /* OOM is global, do not handle */
2053 if (!memcg)
2054 return false;
2055
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07002056 if (!handle)
Johannes Weiner494264202013-10-16 13:46:59 -07002057 goto cleanup;
2058
2059 owait.memcg = memcg;
2060 owait.wait.flags = 0;
2061 owait.wait.func = memcg_oom_wake_function;
2062 owait.wait.private = current;
Ingo Molnar2055da92017-06-20 12:06:46 +02002063 INIT_LIST_HEAD(&owait.wait.entry);
Johannes Weiner494264202013-10-16 13:46:59 -07002064
2065 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002066 mem_cgroup_mark_under_oom(memcg);
2067
2068 locked = mem_cgroup_oom_trylock(memcg);
2069
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002070 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002071 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002072
Haifeng Xu857f2132023-04-19 03:07:38 +00002073 schedule();
2074 mem_cgroup_unmark_under_oom(memcg);
2075 finish_wait(&memcg_oom_waitq, &owait.wait);
Johannes Weiner494264202013-10-16 13:46:59 -07002076
Haifeng Xu18b1d182023-04-19 03:07:39 +00002077 if (locked)
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002078 mem_cgroup_oom_unlock(memcg);
Johannes Weiner494264202013-10-16 13:46:59 -07002079cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08002080 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002081 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002082 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002083}
2084
Johannes Weinerd7365e72014-10-29 14:50:48 -07002085/**
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07002086 * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM
2087 * @victim: task to be killed by the OOM killer
2088 * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM
2089 *
2090 * Returns a pointer to a memory cgroup, which has to be cleaned up
2091 * by killing all belonging OOM-killable tasks.
2092 *
2093 * Caller has to call mem_cgroup_put() on the returned non-NULL memcg.
2094 */
2095struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
2096 struct mem_cgroup *oom_domain)
2097{
2098 struct mem_cgroup *oom_group = NULL;
2099 struct mem_cgroup *memcg;
2100
2101 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2102 return NULL;
2103
2104 if (!oom_domain)
2105 oom_domain = root_mem_cgroup;
2106
2107 rcu_read_lock();
2108
2109 memcg = mem_cgroup_from_task(victim);
Kamalesh Babulal7848ed62022-09-30 19:14:33 +05302110 if (mem_cgroup_is_root(memcg))
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07002111 goto out;
2112
2113 /*
Roman Gushchin48fe2672020-04-01 21:07:39 -07002114 * If the victim task has been asynchronously moved to a different
2115 * memory cgroup, we might end up killing tasks outside oom_domain.
2116 * In this case it's better to ignore memory.group.oom.
2117 */
2118 if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain)))
2119 goto out;
2120
2121 /*
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07002122 * Traverse the memory cgroup hierarchy from the victim task's
2123 * cgroup up to the OOMing cgroup (or root) to find the
2124 * highest-level memory cgroup with oom.group set.
2125 */
2126 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Yue Zhaoeaf7b662023-03-06 23:41:35 +08002127 if (READ_ONCE(memcg->oom_group))
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07002128 oom_group = memcg;
2129
2130 if (memcg == oom_domain)
2131 break;
2132 }
2133
2134 if (oom_group)
2135 css_get(&oom_group->css);
2136out:
2137 rcu_read_unlock();
2138
2139 return oom_group;
2140}
2141
2142void mem_cgroup_print_oom_group(struct mem_cgroup *memcg)
2143{
2144 pr_info("Tasks in ");
2145 pr_cont_cgroup_path(memcg->css.cgroup);
2146 pr_cont(" are going to be killed due to memory.oom.group set\n");
2147}
2148
2149/**
Matthew Wilcox (Oracle)f70ad442021-06-28 17:26:00 -04002150 * folio_memcg_lock - Bind a folio to its memcg.
2151 * @folio: The folio.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002152 *
Matthew Wilcox (Oracle)f70ad442021-06-28 17:26:00 -04002153 * This function prevents unlocked LRU folios from being moved to
Johannes Weiner739f79f2017-08-18 15:15:48 -07002154 * another cgroup.
2155 *
Matthew Wilcox (Oracle)f70ad442021-06-28 17:26:00 -04002156 * It ensures lifetime of the bound memcg. The caller is responsible
2157 * for the lifetime of the folio.
Balbir Singhd69b0422009-06-17 16:26:34 -07002158 */
Matthew Wilcox (Oracle)f70ad442021-06-28 17:26:00 -04002159void folio_memcg_lock(struct folio *folio)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002160{
2161 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08002162 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002163
Johannes Weiner6de22612015-02-11 15:25:01 -08002164 /*
2165 * The RCU lock is held throughout the transaction. The fast
2166 * path can get away without acquiring the memcg->move_lock
2167 * because page moving starts with an RCU grace period.
Johannes Weiner739f79f2017-08-18 15:15:48 -07002168 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07002169 rcu_read_lock();
2170
2171 if (mem_cgroup_disabled())
Johannes Weiner1c824a62021-04-29 22:55:32 -07002172 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002173again:
Matthew Wilcox (Oracle)f70ad442021-06-28 17:26:00 -04002174 memcg = folio_memcg(folio);
Johannes Weiner29833312014-12-10 15:44:02 -08002175 if (unlikely(!memcg))
Johannes Weiner1c824a62021-04-29 22:55:32 -07002176 return;
Johannes Weinerd7365e72014-10-29 14:50:48 -07002177
Alex Shi20ad50d2020-12-15 12:33:51 -08002178#ifdef CONFIG_PROVE_LOCKING
2179 local_irq_save(flags);
2180 might_lock(&memcg->move_lock);
2181 local_irq_restore(flags);
2182#endif
2183
Qiang Huangbdcbb652014-06-04 16:08:21 -07002184 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner1c824a62021-04-29 22:55:32 -07002185 return;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002186
Johannes Weiner6de22612015-02-11 15:25:01 -08002187 spin_lock_irqsave(&memcg->move_lock, flags);
Matthew Wilcox (Oracle)f70ad442021-06-28 17:26:00 -04002188 if (memcg != folio_memcg(folio)) {
Johannes Weiner6de22612015-02-11 15:25:01 -08002189 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002190 goto again;
2191 }
Johannes Weiner6de22612015-02-11 15:25:01 -08002192
2193 /*
Johannes Weiner1c824a62021-04-29 22:55:32 -07002194 * When charge migration first begins, we can have multiple
2195 * critical sections holding the fast-path RCU lock and one
2196 * holding the slowpath move_lock. Track the task who has the
Kefeng Wang6c77b602023-06-14 22:36:12 +08002197 * move_lock for folio_memcg_unlock().
Johannes Weiner6de22612015-02-11 15:25:01 -08002198 */
2199 memcg->move_lock_task = current;
2200 memcg->move_lock_flags = flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002201}
Matthew Wilcox (Oracle)f70ad442021-06-28 17:26:00 -04002202
Matthew Wilcox (Oracle)f70ad442021-06-28 17:26:00 -04002203static void __folio_memcg_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002204{
Johannes Weiner6de22612015-02-11 15:25:01 -08002205 if (memcg && memcg->move_lock_task == current) {
2206 unsigned long flags = memcg->move_lock_flags;
2207
2208 memcg->move_lock_task = NULL;
2209 memcg->move_lock_flags = 0;
2210
2211 spin_unlock_irqrestore(&memcg->move_lock, flags);
2212 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002213
Johannes Weinerd7365e72014-10-29 14:50:48 -07002214 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002215}
Johannes Weiner739f79f2017-08-18 15:15:48 -07002216
2217/**
Matthew Wilcox (Oracle)f70ad442021-06-28 17:26:00 -04002218 * folio_memcg_unlock - Release the binding between a folio and its memcg.
2219 * @folio: The folio.
2220 *
2221 * This releases the binding created by folio_memcg_lock(). This does
2222 * not change the accounting of this folio to its memcg, but it does
2223 * permit others to change it.
Johannes Weiner739f79f2017-08-18 15:15:48 -07002224 */
Matthew Wilcox (Oracle)f70ad442021-06-28 17:26:00 -04002225void folio_memcg_unlock(struct folio *folio)
2226{
2227 __folio_memcg_unlock(folio_memcg(folio));
2228}
Matthew Wilcox (Oracle)f70ad442021-06-28 17:26:00 -04002229
Michal Hockofead2b82022-03-22 14:40:35 -07002230struct memcg_stock_pcp {
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07002231 local_lock_t stock_lock;
Michal Hockofead2b82022-03-22 14:40:35 -07002232 struct mem_cgroup *cached; /* this never be root cgroup */
2233 unsigned int nr_pages;
2234
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002235#ifdef CONFIG_MEMCG_KMEM
2236 struct obj_cgroup *cached_objcg;
Waiman Long68ac5b3c2021-06-28 19:37:23 -07002237 struct pglist_data *cached_pgdat;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002238 unsigned int nr_bytes;
Waiman Long68ac5b3c2021-06-28 19:37:23 -07002239 int nr_slab_reclaimable_b;
2240 int nr_slab_unreclaimable_b;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002241#endif
2242
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002243 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002244 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002245#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002246};
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07002247static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock) = {
2248 .stock_lock = INIT_LOCAL_LOCK(stock_lock),
2249};
Michal Hocko9f50fad2011-08-09 11:56:26 +02002250static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002251
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002252#ifdef CONFIG_MEMCG_KMEM
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07002253static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002254static bool obj_stock_flush_required(struct memcg_stock_pcp *stock,
2255 struct mem_cgroup *root_memcg);
Yosry Ahmeda8c49af2022-03-22 14:40:10 -07002256static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002257
2258#else
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07002259static inline struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock)
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002260{
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07002261 return NULL;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002262}
2263static bool obj_stock_flush_required(struct memcg_stock_pcp *stock,
2264 struct mem_cgroup *root_memcg)
2265{
2266 return false;
2267}
Yosry Ahmeda8c49af2022-03-22 14:40:10 -07002268static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages)
2269{
2270}
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002271#endif
2272
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002273/**
2274 * consume_stock: Try to consume stocked charge on this cpu.
2275 * @memcg: memcg to consume from.
2276 * @nr_pages: how many pages to charge.
2277 *
2278 * The charges will only happen if @memcg matches the current cpu's memcg
2279 * stock, and at least @nr_pages are available in that stock. Failure to
2280 * service an allocation will refill the stock.
2281 *
2282 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002283 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002284static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002285{
2286 struct memcg_stock_pcp *stock;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002287 unsigned long flags;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002288 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002289
Johannes Weinera983b5e2018-01-31 16:16:45 -08002290 if (nr_pages > MEMCG_CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002291 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002292
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07002293 local_lock_irqsave(&memcg_stock.stock_lock, flags);
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002294
2295 stock = this_cpu_ptr(&memcg_stock);
Roman Gushchinf785a8f2023-05-02 09:08:39 -07002296 if (memcg == READ_ONCE(stock->cached) && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002297 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002298 ret = true;
2299 }
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002300
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07002301 local_unlock_irqrestore(&memcg_stock.stock_lock, flags);
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002302
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002303 return ret;
2304}
2305
2306/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002307 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002308 */
2309static void drain_stock(struct memcg_stock_pcp *stock)
2310{
Roman Gushchinf785a8f2023-05-02 09:08:39 -07002311 struct mem_cgroup *old = READ_ONCE(stock->cached);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002312
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07002313 if (!old)
2314 return;
2315
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002316 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002317 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002318 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002319 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002320 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002321 }
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07002322
2323 css_put(&old->css);
Roman Gushchinf785a8f2023-05-02 09:08:39 -07002324 WRITE_ONCE(stock->cached, NULL);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002325}
2326
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002327static void drain_local_stock(struct work_struct *dummy)
2328{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002329 struct memcg_stock_pcp *stock;
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07002330 struct obj_cgroup *old = NULL;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002331 unsigned long flags;
2332
Michal Hocko72f01842017-10-03 16:14:53 -07002333 /*
Michal Hocko5c49cf92021-09-02 14:56:02 -07002334 * The only protection from cpu hotplug (memcg_hotplug_cpu_dead) vs.
2335 * drain_stock races is that we always operate on local CPU stock
2336 * here with IRQ disabled
Michal Hocko72f01842017-10-03 16:14:53 -07002337 */
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07002338 local_lock_irqsave(&memcg_stock.stock_lock, flags);
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002339
2340 stock = this_cpu_ptr(&memcg_stock);
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07002341 old = drain_obj_stock(stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002342 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002343 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002344
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07002345 local_unlock_irqrestore(&memcg_stock.stock_lock, flags);
2346 if (old)
2347 obj_cgroup_put(old);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002348}
2349
2350/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002351 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002352 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002353 */
Johannes Weineraf9a3b62022-03-22 14:40:44 -07002354static void __refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002355{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002356 struct memcg_stock_pcp *stock;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002357
2358 stock = this_cpu_ptr(&memcg_stock);
Roman Gushchinf785a8f2023-05-02 09:08:39 -07002359 if (READ_ONCE(stock->cached) != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002360 drain_stock(stock);
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07002361 css_get(&memcg->css);
Roman Gushchinf785a8f2023-05-02 09:08:39 -07002362 WRITE_ONCE(stock->cached, memcg);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002363 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002364 stock->nr_pages += nr_pages;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002365
Johannes Weinera983b5e2018-01-31 16:16:45 -08002366 if (stock->nr_pages > MEMCG_CHARGE_BATCH)
Roman Gushchin475d0482017-09-08 16:13:09 -07002367 drain_stock(stock);
Johannes Weineraf9a3b62022-03-22 14:40:44 -07002368}
Roman Gushchin475d0482017-09-08 16:13:09 -07002369
Johannes Weineraf9a3b62022-03-22 14:40:44 -07002370static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
2371{
2372 unsigned long flags;
2373
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07002374 local_lock_irqsave(&memcg_stock.stock_lock, flags);
Johannes Weineraf9a3b62022-03-22 14:40:44 -07002375 __refill_stock(memcg, nr_pages);
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07002376 local_unlock_irqrestore(&memcg_stock.stock_lock, flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002377}
2378
2379/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002380 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002381 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002382 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002383static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002384{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002385 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002386
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002387 /* If someone's already draining, avoid adding running more workers. */
2388 if (!mutex_trylock(&percpu_charge_mutex))
2389 return;
Michal Hocko72f01842017-10-03 16:14:53 -07002390 /*
2391 * Notify other cpus that system-wide "drain" is running
2392 * We do not care about races with the cpu hotplug because cpu down
2393 * as well as workers from this path always operate on the local
2394 * per-cpu data. CPU up doesn't touch memcg_stock at all.
2395 */
Sebastian Andrzej Siewior0790ed62022-03-22 14:40:50 -07002396 migrate_disable();
2397 curcpu = smp_processor_id();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002398 for_each_online_cpu(cpu) {
2399 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002400 struct mem_cgroup *memcg;
Roman Gushchine1a366b2019-09-23 15:34:58 -07002401 bool flush = false;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002402
Roman Gushchine1a366b2019-09-23 15:34:58 -07002403 rcu_read_lock();
Roman Gushchinf785a8f2023-05-02 09:08:39 -07002404 memcg = READ_ONCE(stock->cached);
Roman Gushchine1a366b2019-09-23 15:34:58 -07002405 if (memcg && stock->nr_pages &&
2406 mem_cgroup_is_descendant(memcg, root_memcg))
2407 flush = true;
Miaohe Lin27fb0952021-09-02 14:55:59 -07002408 else if (obj_stock_flush_required(stock, root_memcg))
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002409 flush = true;
Roman Gushchine1a366b2019-09-23 15:34:58 -07002410 rcu_read_unlock();
2411
2412 if (flush &&
2413 !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
Michal Hockod1a05b62011-07-26 16:08:27 -07002414 if (cpu == curcpu)
2415 drain_local_stock(&stock->work);
Michal Hocko6a792692023-03-17 14:44:48 +01002416 else if (!cpu_is_isolated(cpu))
Michal Hockod1a05b62011-07-26 16:08:27 -07002417 schedule_work_on(cpu, &stock->work);
2418 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002419 }
Sebastian Andrzej Siewior0790ed62022-03-22 14:40:50 -07002420 migrate_enable();
Michal Hocko9f50fad2011-08-09 11:56:26 +02002421 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002422}
2423
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01002424static int memcg_hotplug_cpu_dead(unsigned int cpu)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002425{
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002426 struct memcg_stock_pcp *stock;
2427
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002428 stock = &per_cpu(memcg_stock, cpu);
2429 drain_stock(stock);
Johannes Weinera983b5e2018-01-31 16:16:45 -08002430
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01002431 return 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002432}
2433
Chris Downb3ff9292020-08-06 23:21:54 -07002434static unsigned long reclaim_high(struct mem_cgroup *memcg,
2435 unsigned int nr_pages,
2436 gfp_t gfp_mask)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002437{
Chris Downb3ff9292020-08-06 23:21:54 -07002438 unsigned long nr_reclaimed = 0;
2439
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002440 do {
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002441 unsigned long pflags;
2442
Jakub Kicinskid1663a92020-06-01 21:49:49 -07002443 if (page_counter_read(&memcg->memory) <=
2444 READ_ONCE(memcg->memory.high))
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002445 continue;
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002446
Johannes Weinere27be242018-04-10 16:29:45 -07002447 memcg_memory_event(memcg, MEMCG_HIGH);
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002448
2449 psi_memstall_enter(&pflags);
Chris Downb3ff9292020-08-06 23:21:54 -07002450 nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages,
Yosry Ahmed73b73ba2022-07-14 06:49:18 +00002451 gfp_mask,
Michal Hocko55ab8342022-12-16 10:46:33 +01002452 MEMCG_RECLAIM_MAY_SWAP);
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002453 psi_memstall_leave(&pflags);
Chris Down4bf17302020-04-06 20:03:30 -07002454 } while ((memcg = parent_mem_cgroup(memcg)) &&
2455 !mem_cgroup_is_root(memcg));
Chris Downb3ff9292020-08-06 23:21:54 -07002456
2457 return nr_reclaimed;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002458}
2459
2460static void high_work_func(struct work_struct *work)
2461{
2462 struct mem_cgroup *memcg;
2463
2464 memcg = container_of(work, struct mem_cgroup, high_work);
Johannes Weinera983b5e2018-01-31 16:16:45 -08002465 reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002466}
2467
Tejun Heob23afb92015-11-05 18:46:11 -08002468/*
Chris Down0e4b01d2019-09-23 15:34:55 -07002469 * Clamp the maximum sleep time per allocation batch to 2 seconds. This is
2470 * enough to still cause a significant slowdown in most cases, while still
2471 * allowing diagnostics and tracing to proceed without becoming stuck.
2472 */
2473#define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ)
2474
2475/*
2476 * When calculating the delay, we use these either side of the exponentiation to
2477 * maintain precision and scale to a reasonable number of jiffies (see the table
2478 * below.
2479 *
2480 * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the
2481 * overage ratio to a delay.
Randy Dunlapac5ddd02020-08-11 18:33:02 -07002482 * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the
Chris Down0e4b01d2019-09-23 15:34:55 -07002483 * proposed penalty in order to reduce to a reasonable number of jiffies, and
2484 * to produce a reasonable delay curve.
2485 *
2486 * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a
2487 * reasonable delay curve compared to precision-adjusted overage, not
2488 * penalising heavily at first, but still making sure that growth beyond the
2489 * limit penalises misbehaviour cgroups by slowing them down exponentially. For
2490 * example, with a high of 100 megabytes:
2491 *
2492 * +-------+------------------------+
2493 * | usage | time to allocate in ms |
2494 * +-------+------------------------+
2495 * | 100M | 0 |
2496 * | 101M | 6 |
2497 * | 102M | 25 |
2498 * | 103M | 57 |
2499 * | 104M | 102 |
2500 * | 105M | 159 |
2501 * | 106M | 230 |
2502 * | 107M | 313 |
2503 * | 108M | 409 |
2504 * | 109M | 518 |
2505 * | 110M | 639 |
2506 * | 111M | 774 |
2507 * | 112M | 921 |
2508 * | 113M | 1081 |
2509 * | 114M | 1254 |
2510 * | 115M | 1439 |
2511 * | 116M | 1638 |
2512 * | 117M | 1849 |
2513 * | 118M | 2000 |
2514 * | 119M | 2000 |
2515 * | 120M | 2000 |
2516 * +-------+------------------------+
2517 */
2518 #define MEMCG_DELAY_PRECISION_SHIFT 20
2519 #define MEMCG_DELAY_SCALING_SHIFT 14
2520
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002521static u64 calculate_overage(unsigned long usage, unsigned long high)
2522{
2523 u64 overage;
2524
2525 if (usage <= high)
2526 return 0;
2527
2528 /*
2529 * Prevent division by 0 in overage calculation by acting as if
2530 * it was a threshold of 1 page
2531 */
2532 high = max(high, 1UL);
2533
2534 overage = usage - high;
2535 overage <<= MEMCG_DELAY_PRECISION_SHIFT;
2536 return div64_u64(overage, high);
2537}
2538
2539static u64 mem_find_max_overage(struct mem_cgroup *memcg)
2540{
2541 u64 overage, max_overage = 0;
2542
2543 do {
2544 overage = calculate_overage(page_counter_read(&memcg->memory),
Jakub Kicinskid1663a92020-06-01 21:49:49 -07002545 READ_ONCE(memcg->memory.high));
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002546 max_overage = max(overage, max_overage);
2547 } while ((memcg = parent_mem_cgroup(memcg)) &&
2548 !mem_cgroup_is_root(memcg));
2549
2550 return max_overage;
2551}
2552
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002553static u64 swap_find_max_overage(struct mem_cgroup *memcg)
2554{
2555 u64 overage, max_overage = 0;
2556
2557 do {
2558 overage = calculate_overage(page_counter_read(&memcg->swap),
2559 READ_ONCE(memcg->swap.high));
2560 if (overage)
2561 memcg_memory_event(memcg, MEMCG_SWAP_HIGH);
2562 max_overage = max(overage, max_overage);
2563 } while ((memcg = parent_mem_cgroup(memcg)) &&
2564 !mem_cgroup_is_root(memcg));
2565
2566 return max_overage;
2567}
2568
Chris Down0e4b01d2019-09-23 15:34:55 -07002569/*
Chris Downe26733e2020-03-21 18:22:23 -07002570 * Get the number of jiffies that we should penalise a mischievous cgroup which
2571 * is exceeding its memory.high by checking both it and its ancestors.
Tejun Heob23afb92015-11-05 18:46:11 -08002572 */
Chris Downe26733e2020-03-21 18:22:23 -07002573static unsigned long calculate_high_delay(struct mem_cgroup *memcg,
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002574 unsigned int nr_pages,
2575 u64 max_overage)
Tejun Heob23afb92015-11-05 18:46:11 -08002576{
Chris Downe26733e2020-03-21 18:22:23 -07002577 unsigned long penalty_jiffies;
Chris Downe26733e2020-03-21 18:22:23 -07002578
2579 if (!max_overage)
2580 return 0;
Chris Down0e4b01d2019-09-23 15:34:55 -07002581
2582 /*
Chris Down0e4b01d2019-09-23 15:34:55 -07002583 * We use overage compared to memory.high to calculate the number of
2584 * jiffies to sleep (penalty_jiffies). Ideally this value should be
2585 * fairly lenient on small overages, and increasingly harsh when the
2586 * memcg in question makes it clear that it has no intention of stopping
2587 * its crazy behaviour, so we exponentially increase the delay based on
2588 * overage amount.
2589 */
Chris Downe26733e2020-03-21 18:22:23 -07002590 penalty_jiffies = max_overage * max_overage * HZ;
2591 penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT;
2592 penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT;
Chris Down0e4b01d2019-09-23 15:34:55 -07002593
2594 /*
2595 * Factor in the task's own contribution to the overage, such that four
2596 * N-sized allocations are throttled approximately the same as one
2597 * 4N-sized allocation.
2598 *
2599 * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or
2600 * larger the current charge patch is than that.
2601 */
Jakub Kicinskiff144e62020-06-01 21:49:45 -07002602 return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH;
Chris Downe26733e2020-03-21 18:22:23 -07002603}
2604
2605/*
2606 * Scheduled by try_charge() to be executed from the userland return path
2607 * and reclaims memory over the high limit.
2608 */
Johannes Weiner9ea9cb02023-09-14 11:21:39 -04002609void mem_cgroup_handle_over_high(gfp_t gfp_mask)
Chris Downe26733e2020-03-21 18:22:23 -07002610{
2611 unsigned long penalty_jiffies;
2612 unsigned long pflags;
Chris Downb3ff9292020-08-06 23:21:54 -07002613 unsigned long nr_reclaimed;
Chris Downe26733e2020-03-21 18:22:23 -07002614 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Chris Downd977aa92020-08-06 23:21:58 -07002615 int nr_retries = MAX_RECLAIM_RETRIES;
Chris Downe26733e2020-03-21 18:22:23 -07002616 struct mem_cgroup *memcg;
Chris Downb3ff9292020-08-06 23:21:54 -07002617 bool in_retry = false;
Chris Downe26733e2020-03-21 18:22:23 -07002618
2619 if (likely(!nr_pages))
2620 return;
2621
2622 memcg = get_mem_cgroup_from_mm(current->mm);
Chris Downe26733e2020-03-21 18:22:23 -07002623 current->memcg_nr_pages_over_high = 0;
2624
Chris Downb3ff9292020-08-06 23:21:54 -07002625retry_reclaim:
2626 /*
2627 * The allocating task should reclaim at least the batch size, but for
2628 * subsequent retries we only want to do what's necessary to prevent oom
2629 * or breaching resource isolation.
2630 *
2631 * This is distinct from memory.max or page allocator behaviour because
2632 * memory.high is currently batched, whereas memory.max and the page
2633 * allocator run every time an allocation is made.
2634 */
2635 nr_reclaimed = reclaim_high(memcg,
2636 in_retry ? SWAP_CLUSTER_MAX : nr_pages,
Johannes Weiner9ea9cb02023-09-14 11:21:39 -04002637 gfp_mask);
Chris Downb3ff9292020-08-06 23:21:54 -07002638
Chris Downe26733e2020-03-21 18:22:23 -07002639 /*
2640 * memory.high is breached and reclaim is unable to keep up. Throttle
2641 * allocators proactively to slow down excessive growth.
2642 */
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002643 penalty_jiffies = calculate_high_delay(memcg, nr_pages,
2644 mem_find_max_overage(memcg));
Chris Down0e4b01d2019-09-23 15:34:55 -07002645
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002646 penalty_jiffies += calculate_high_delay(memcg, nr_pages,
2647 swap_find_max_overage(memcg));
2648
Chris Down0e4b01d2019-09-23 15:34:55 -07002649 /*
Jakub Kicinskiff144e62020-06-01 21:49:45 -07002650 * Clamp the max delay per usermode return so as to still keep the
2651 * application moving forwards and also permit diagnostics, albeit
2652 * extremely slowly.
2653 */
2654 penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES);
2655
2656 /*
Chris Down0e4b01d2019-09-23 15:34:55 -07002657 * Don't sleep if the amount of jiffies this memcg owes us is so low
2658 * that it's not even worth doing, in an attempt to be nice to those who
2659 * go only a small amount over their memory.high value and maybe haven't
2660 * been aggressively reclaimed enough yet.
2661 */
2662 if (penalty_jiffies <= HZ / 100)
2663 goto out;
2664
2665 /*
Chris Downb3ff9292020-08-06 23:21:54 -07002666 * If reclaim is making forward progress but we're still over
2667 * memory.high, we want to encourage that rather than doing allocator
2668 * throttling.
2669 */
2670 if (nr_reclaimed || nr_retries--) {
2671 in_retry = true;
2672 goto retry_reclaim;
2673 }
2674
2675 /*
Chris Down0e4b01d2019-09-23 15:34:55 -07002676 * If we exit early, we're guaranteed to die (since
2677 * schedule_timeout_killable sets TASK_KILLABLE). This means we don't
2678 * need to account for any ill-begotten jiffies to pay them off later.
2679 */
2680 psi_memstall_enter(&pflags);
2681 schedule_timeout_killable(penalty_jiffies);
2682 psi_memstall_leave(&pflags);
2683
2684out:
2685 css_put(&memcg->css);
Tejun Heob23afb92015-11-05 18:46:11 -08002686}
2687
Muchun Songc5c8b162021-06-28 19:37:44 -07002688static int try_charge_memcg(struct mem_cgroup *memcg, gfp_t gfp_mask,
2689 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002690{
Johannes Weinera983b5e2018-01-31 16:16:45 -08002691 unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages);
Chris Downd977aa92020-08-06 23:21:58 -07002692 int nr_retries = MAX_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002693 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002694 struct page_counter *counter;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002695 unsigned long nr_reclaimed;
Vasily Averina4ebf1b2021-11-05 13:38:09 -07002696 bool passed_oom = false;
Yosry Ahmed73b73ba2022-07-14 06:49:18 +00002697 unsigned int reclaim_options = MEMCG_RECLAIM_MAY_SWAP;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002698 bool drained = false;
Roman Gushchind6e103a2022-07-01 20:35:21 -07002699 bool raised_max_event = false;
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002700 unsigned long pflags;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002701
Johannes Weiner6539cc02014-08-06 16:05:42 -07002702retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07002703 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08002704 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002705
Johannes Weiner7941d212016-01-14 15:21:23 -08002706 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08002707 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
2708 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002709 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08002710 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002711 page_counter_uncharge(&memcg->memsw, batch);
2712 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002713 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002714 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Yosry Ahmed73b73ba2022-07-14 06:49:18 +00002715 reclaim_options &= ~MEMCG_RECLAIM_MAY_SWAP;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002716 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002717
Johannes Weiner6539cc02014-08-06 16:05:42 -07002718 if (batch > nr_pages) {
2719 batch = nr_pages;
2720 goto retry;
2721 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002722
Johannes Weiner06b078fc02014-08-06 16:05:44 -07002723 /*
Johannes Weiner89a28482016-10-27 17:46:56 -07002724 * Prevent unbounded recursion when reclaim operations need to
2725 * allocate memory. This might exceed the limits temporarily,
2726 * but we prefer facilitating memory reclaim and getting back
2727 * under the limit over triggering OOM kills in these cases.
2728 */
2729 if (unlikely(current->flags & PF_MEMALLOC))
2730 goto force;
2731
Johannes Weiner06b078fc02014-08-06 16:05:44 -07002732 if (unlikely(task_in_memcg_oom(current)))
2733 goto nomem;
2734
Mel Gormand0164ad2015-11-06 16:28:21 -08002735 if (!gfpflags_allow_blocking(gfp_mask))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002736 goto nomem;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07002737
Johannes Weinere27be242018-04-10 16:29:45 -07002738 memcg_memory_event(mem_over_limit, MEMCG_MAX);
Roman Gushchind6e103a2022-07-01 20:35:21 -07002739 raised_max_event = true;
Johannes Weiner241994ed2015-02-11 15:26:06 -08002740
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002741 psi_memstall_enter(&pflags);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002742 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
Michal Hocko55ab8342022-12-16 10:46:33 +01002743 gfp_mask, reclaim_options);
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002744 psi_memstall_leave(&pflags);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002745
Johannes Weiner61e02c72014-08-06 16:08:16 -07002746 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07002747 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07002748
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002749 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002750 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002751 drained = true;
2752 goto retry;
2753 }
2754
Johannes Weiner28c34c22014-08-06 16:05:47 -07002755 if (gfp_mask & __GFP_NORETRY)
2756 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002757 /*
2758 * Even though the limit is exceeded at this point, reclaim
2759 * may have been able to free some pages. Retry the charge
2760 * before killing the task.
2761 *
2762 * Only for regular pages, though: huge pages are rather
2763 * unlikely to succeed so close to the limit, and we fall back
2764 * to regular pages anyway in case of failure.
2765 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002766 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002767 goto retry;
2768 /*
2769 * At task move, charge accounts can be doubly counted. So, it's
2770 * better to wait until the end of task_move if something is going on.
2771 */
2772 if (mem_cgroup_wait_acct_move(mem_over_limit))
2773 goto retry;
2774
Johannes Weiner9b130612014-08-06 16:05:51 -07002775 if (nr_retries--)
2776 goto retry;
2777
Shakeel Butt38d38492019-07-11 20:55:48 -07002778 if (gfp_mask & __GFP_RETRY_MAYFAIL)
Michal Hocko29ef6802018-08-17 15:47:11 -07002779 goto nomem;
2780
Vasily Averina4ebf1b2021-11-05 13:38:09 -07002781 /* Avoid endless loop for tasks bypassed by the oom killer */
2782 if (passed_oom && task_is_dying())
2783 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002784
Michal Hocko29ef6802018-08-17 15:47:11 -07002785 /*
2786 * keep retrying as long as the memcg oom killer is able to make
2787 * a forward progress or bypass the charge if the oom killer
2788 * couldn't make any progress.
2789 */
Shakeel Buttbecdf892022-03-22 14:40:19 -07002790 if (mem_cgroup_oom(mem_over_limit, gfp_mask,
2791 get_order(nr_pages * PAGE_SIZE))) {
Vasily Averina4ebf1b2021-11-05 13:38:09 -07002792 passed_oom = true;
Chris Downd977aa92020-08-06 23:21:58 -07002793 nr_retries = MAX_RECLAIM_RETRIES;
Michal Hocko29ef6802018-08-17 15:47:11 -07002794 goto retry;
Michal Hocko29ef6802018-08-17 15:47:11 -07002795 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002796nomem:
Shakeel Butt1461e8c2022-03-22 14:40:22 -07002797 /*
2798 * Memcg doesn't have a dedicated reserve for atomic
2799 * allocations. But like the global atomic pool, we need to
2800 * put the burden of reclaim on regular allocation requests
2801 * and let these go through as privileged allocations.
2802 */
2803 if (!(gfp_mask & (__GFP_NOFAIL | __GFP_HIGH)))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002804 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08002805force:
2806 /*
Roman Gushchind6e103a2022-07-01 20:35:21 -07002807 * If the allocation has to be enforced, don't forget to raise
2808 * a MEMCG_MAX event.
2809 */
2810 if (!raised_max_event)
2811 memcg_memory_event(mem_over_limit, MEMCG_MAX);
2812
2813 /*
Tejun Heo10d53c72015-11-05 18:46:17 -08002814 * The allocation either can't fail or will lead to more memory
2815 * being freed very soon. Allow memory usage go over the limit
2816 * temporarily by force charging it.
2817 */
2818 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002819 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08002820 page_counter_charge(&memcg->memsw, nr_pages);
Tejun Heo10d53c72015-11-05 18:46:17 -08002821
2822 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002823
2824done_restock:
2825 if (batch > nr_pages)
2826 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002827
Johannes Weiner241994ed2015-02-11 15:26:06 -08002828 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002829 * If the hierarchy is above the normal consumption range, schedule
2830 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002831 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002832 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2833 * not recorded as it most likely matches current's and won't
2834 * change in the meantime. As high limit is checked again before
2835 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994ed2015-02-11 15:26:06 -08002836 */
2837 do {
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002838 bool mem_high, swap_high;
2839
2840 mem_high = page_counter_read(&memcg->memory) >
2841 READ_ONCE(memcg->memory.high);
2842 swap_high = page_counter_read(&memcg->swap) >
2843 READ_ONCE(memcg->swap.high);
2844
2845 /* Don't bother a random interrupted task */
Shakeel Butt086f6942022-03-22 14:40:07 -07002846 if (!in_task()) {
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002847 if (mem_high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002848 schedule_work(&memcg->high_work);
2849 break;
2850 }
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002851 continue;
2852 }
2853
2854 if (mem_high || swap_high) {
2855 /*
2856 * The allocating tasks in this cgroup will need to do
2857 * reclaim or be throttled to prevent further growth
2858 * of the memory or swap footprints.
2859 *
2860 * Target some best-effort fairness between the tasks,
2861 * and distribute reclaim work and delay penalties
2862 * based on how much each task is actually allocating.
2863 */
Vladimir Davydov9516a182015-12-11 13:40:24 -08002864 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002865 set_notify_resume(current);
2866 break;
2867 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08002868 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002869
Shakeel Buttc9afe312022-03-22 14:40:28 -07002870 if (current->memcg_nr_pages_over_high > MEMCG_CHARGE_BATCH &&
2871 !(current->flags & PF_MEMALLOC) &&
2872 gfpflags_allow_blocking(gfp_mask)) {
Johannes Weiner9ea9cb02023-09-14 11:21:39 -04002873 mem_cgroup_handle_over_high(gfp_mask);
Shakeel Buttc9afe312022-03-22 14:40:28 -07002874 }
Tejun Heo10d53c72015-11-05 18:46:17 -08002875 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002876}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002877
Muchun Songc5c8b162021-06-28 19:37:44 -07002878static inline int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
2879 unsigned int nr_pages)
2880{
2881 if (mem_cgroup_is_root(memcg))
2882 return 0;
2883
2884 return try_charge_memcg(memcg, gfp_mask, nr_pages);
2885}
2886
Nhat Pham4b569382023-10-06 11:46:26 -07002887/**
2888 * mem_cgroup_cancel_charge() - cancel an uncommitted try_charge() call.
2889 * @memcg: memcg previously charged.
2890 * @nr_pages: number of pages previously charged.
2891 */
2892void mem_cgroup_cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002893{
Johannes Weinerce00a962014-09-05 08:43:57 -04002894 if (mem_cgroup_is_root(memcg))
2895 return;
2896
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002897 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002898 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002899 page_counter_uncharge(&memcg->memsw, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002900}
2901
Matthew Wilcox (Oracle)118f2872021-04-29 14:07:12 -04002902static void commit_charge(struct folio *folio, struct mem_cgroup *memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002903{
Matthew Wilcox (Oracle)118f2872021-04-29 14:07:12 -04002904 VM_BUG_ON_FOLIO(folio_memcg(folio), folio);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002905 /*
Alex Shia5eb0112020-12-14 19:06:42 -08002906 * Any of the following ensures page's memcg stability:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002907 *
Johannes Weinera0b5b412020-06-03 16:02:27 -07002908 * - the page lock
2909 * - LRU isolation
Kefeng Wang6c77b602023-06-14 22:36:12 +08002910 * - folio_memcg_lock()
Johannes Weinera0b5b412020-06-03 16:02:27 -07002911 * - exclusive reference
Yu Zhao018ee472022-09-18 02:00:04 -06002912 * - mem_cgroup_trylock_pages()
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002913 */
Matthew Wilcox (Oracle)118f2872021-04-29 14:07:12 -04002914 folio->memcg_data = (unsigned long)memcg;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002915}
2916
Nhat Pham4b569382023-10-06 11:46:26 -07002917/**
2918 * mem_cgroup_commit_charge - commit a previously successful try_charge().
2919 * @folio: folio to commit the charge to.
2920 * @memcg: memcg previously charged.
2921 */
2922void mem_cgroup_commit_charge(struct folio *folio, struct mem_cgroup *memcg)
2923{
2924 css_get(&memcg->css);
2925 commit_charge(folio, memcg);
2926
2927 local_irq_disable();
2928 mem_cgroup_charge_statistics(memcg, folio_nr_pages(folio));
2929 memcg_check_events(memcg, folio_nid(folio));
2930 local_irq_enable();
2931}
2932
Kirill Tkhai84c07d12018-08-17 15:47:25 -07002933#ifdef CONFIG_MEMCG_KMEM
Waiman Long41eb5df2021-06-28 19:37:34 -07002934/*
2935 * The allocated objcg pointers array is not accounted directly.
2936 * Moreover, it should not come from DMA buffer and is not readily
2937 * reclaimable. So those GFP bits should be masked off.
2938 */
Roman Gushchin24948e32023-11-07 09:18:02 -08002939#define OBJCGS_CLEAR_MASK (__GFP_DMA | __GFP_RECLAIMABLE | \
2940 __GFP_ACCOUNT | __GFP_NOFAIL)
Waiman Long41eb5df2021-06-28 19:37:34 -07002941
Waiman Longa7ebf562021-12-10 14:47:05 -08002942/*
Waiman Longa7ebf562021-12-10 14:47:05 -08002943 * mod_objcg_mlstate() may be called with irq enabled, so
2944 * mod_memcg_lruvec_state() should be used.
2945 */
2946static inline void mod_objcg_mlstate(struct obj_cgroup *objcg,
2947 struct pglist_data *pgdat,
2948 enum node_stat_item idx, int nr)
2949{
2950 struct mem_cgroup *memcg;
2951 struct lruvec *lruvec;
2952
2953 rcu_read_lock();
2954 memcg = obj_cgroup_memcg(objcg);
2955 lruvec = mem_cgroup_lruvec(memcg, pgdat);
2956 mod_memcg_lruvec_state(lruvec, idx, nr);
2957 rcu_read_unlock();
2958}
2959
Vlastimil Babka4b5f8d9a2021-11-02 22:42:04 +01002960int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s,
2961 gfp_t gfp, bool new_slab)
Roman Gushchin10befea2020-08-06 23:21:27 -07002962{
Vlastimil Babka4b5f8d9a2021-11-02 22:42:04 +01002963 unsigned int objects = objs_per_slab(s, slab);
Roman Gushchin2e9bd482021-02-24 12:03:11 -08002964 unsigned long memcg_data;
Roman Gushchin10befea2020-08-06 23:21:27 -07002965 void *vec;
2966
Waiman Long41eb5df2021-06-28 19:37:34 -07002967 gfp &= ~OBJCGS_CLEAR_MASK;
Roman Gushchin10befea2020-08-06 23:21:27 -07002968 vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp,
Vlastimil Babka4b5f8d9a2021-11-02 22:42:04 +01002969 slab_nid(slab));
Roman Gushchin10befea2020-08-06 23:21:27 -07002970 if (!vec)
2971 return -ENOMEM;
2972
Roman Gushchin2e9bd482021-02-24 12:03:11 -08002973 memcg_data = (unsigned long) vec | MEMCG_DATA_OBJCGS;
Vlastimil Babka4b5f8d9a2021-11-02 22:42:04 +01002974 if (new_slab) {
Roman Gushchin2e9bd482021-02-24 12:03:11 -08002975 /*
Vlastimil Babka4b5f8d9a2021-11-02 22:42:04 +01002976 * If the slab is brand new and nobody can yet access its
2977 * memcg_data, no synchronization is required and memcg_data can
2978 * be simply assigned.
Roman Gushchin2e9bd482021-02-24 12:03:11 -08002979 */
Vlastimil Babka4b5f8d9a2021-11-02 22:42:04 +01002980 slab->memcg_data = memcg_data;
2981 } else if (cmpxchg(&slab->memcg_data, 0, memcg_data)) {
Roman Gushchin2e9bd482021-02-24 12:03:11 -08002982 /*
Vlastimil Babka4b5f8d9a2021-11-02 22:42:04 +01002983 * If the slab is already in use, somebody can allocate and
2984 * assign obj_cgroups in parallel. In this case the existing
Roman Gushchin2e9bd482021-02-24 12:03:11 -08002985 * objcg vector should be reused.
2986 */
Roman Gushchin10befea2020-08-06 23:21:27 -07002987 kfree(vec);
Roman Gushchin2e9bd482021-02-24 12:03:11 -08002988 return 0;
2989 }
Roman Gushchin10befea2020-08-06 23:21:27 -07002990
Roman Gushchin2e9bd482021-02-24 12:03:11 -08002991 kmemleak_not_leak(vec);
Roman Gushchin10befea2020-08-06 23:21:27 -07002992 return 0;
2993}
2994
Roman Gushchinfc4db902022-06-10 11:03:10 -07002995static __always_inline
2996struct mem_cgroup *mem_cgroup_from_obj_folio(struct folio *folio, void *p)
Roman Gushchin8380ce42020-03-28 19:17:25 -07002997{
Roman Gushchin8380ce42020-03-28 19:17:25 -07002998 /*
Roman Gushchin98556092020-08-06 23:21:10 -07002999 * Slab objects are accounted individually, not per-page.
3000 * Memcg membership data for each individual object is saved in
Vlastimil Babka4b5f8d9a2021-11-02 22:42:04 +01003001 * slab->memcg_data.
Roman Gushchin8380ce42020-03-28 19:17:25 -07003002 */
Vlastimil Babka4b5f8d9a2021-11-02 22:42:04 +01003003 if (folio_test_slab(folio)) {
3004 struct obj_cgroup **objcgs;
3005 struct slab *slab;
Roman Gushchin98556092020-08-06 23:21:10 -07003006 unsigned int off;
3007
Vlastimil Babka4b5f8d9a2021-11-02 22:42:04 +01003008 slab = folio_slab(folio);
3009 objcgs = slab_objcgs(slab);
3010 if (!objcgs)
3011 return NULL;
3012
3013 off = obj_to_index(slab->slab_cache, slab, p);
3014 if (objcgs[off])
3015 return obj_cgroup_memcg(objcgs[off]);
Roman Gushchin10befea2020-08-06 23:21:27 -07003016
3017 return NULL;
Roman Gushchin98556092020-08-06 23:21:10 -07003018 }
Roman Gushchin8380ce42020-03-28 19:17:25 -07003019
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08003020 /*
Matthew Wilcoxbecacb02022-12-30 15:08:42 +08003021 * folio_memcg_check() is used here, because in theory we can encounter
Vlastimil Babka4b5f8d9a2021-11-02 22:42:04 +01003022 * a folio where the slab flag has been cleared already, but
3023 * slab->memcg_data has not been freed yet
Matthew Wilcoxbecacb02022-12-30 15:08:42 +08003024 * folio_memcg_check() will guarantee that a proper memory
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08003025 * cgroup pointer or NULL will be returned.
3026 */
Matthew Wilcoxbecacb02022-12-30 15:08:42 +08003027 return folio_memcg_check(folio);
Roman Gushchin8380ce42020-03-28 19:17:25 -07003028}
3029
Roman Gushchinfc4db902022-06-10 11:03:10 -07003030/*
3031 * Returns a pointer to the memory cgroup to which the kernel object is charged.
3032 *
3033 * A passed kernel object can be a slab object, vmalloc object or a generic
3034 * kernel page, so different mechanisms for getting the memory cgroup pointer
3035 * should be used.
3036 *
3037 * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller
3038 * can not know for sure how the kernel object is implemented.
3039 * mem_cgroup_from_obj() can be safely used in such cases.
3040 *
3041 * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(),
3042 * cgroup_mutex, etc.
3043 */
3044struct mem_cgroup *mem_cgroup_from_obj(void *p)
3045{
3046 struct folio *folio;
3047
3048 if (mem_cgroup_disabled())
3049 return NULL;
3050
3051 if (unlikely(is_vmalloc_addr(p)))
3052 folio = page_folio(vmalloc_to_page(p));
3053 else
3054 folio = virt_to_folio(p);
3055
3056 return mem_cgroup_from_obj_folio(folio, p);
3057}
3058
3059/*
3060 * Returns a pointer to the memory cgroup to which the kernel object is charged.
3061 * Similar to mem_cgroup_from_obj(), but faster and not suitable for objects,
3062 * allocated using vmalloc().
3063 *
3064 * A passed kernel object must be a slab object or a generic kernel page.
3065 *
3066 * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(),
3067 * cgroup_mutex, etc.
3068 */
3069struct mem_cgroup *mem_cgroup_from_slab_obj(void *p)
3070{
3071 if (mem_cgroup_disabled())
3072 return NULL;
3073
3074 return mem_cgroup_from_obj_folio(virt_to_folio(p), p);
3075}
3076
Johannes Weinerf4840cc2022-05-19 14:08:53 -07003077static struct obj_cgroup *__get_obj_cgroup_from_memcg(struct mem_cgroup *memcg)
3078{
3079 struct obj_cgroup *objcg = NULL;
3080
Kamalesh Babulal7848ed62022-09-30 19:14:33 +05303081 for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) {
Johannes Weinerf4840cc2022-05-19 14:08:53 -07003082 objcg = rcu_dereference(memcg->objcg);
Roman Gushchin7d0715d2023-10-19 15:53:41 -07003083 if (likely(objcg && obj_cgroup_tryget(objcg)))
Johannes Weinerf4840cc2022-05-19 14:08:53 -07003084 break;
3085 objcg = NULL;
3086 }
3087 return objcg;
3088}
3089
Roman Gushchin1aacbd32023-10-19 15:53:42 -07003090static struct obj_cgroup *current_objcg_update(void)
3091{
3092 struct mem_cgroup *memcg;
3093 struct obj_cgroup *old, *objcg = NULL;
3094
3095 do {
3096 /* Atomically drop the update bit. */
3097 old = xchg(&current->objcg, NULL);
3098 if (old) {
3099 old = (struct obj_cgroup *)
3100 ((unsigned long)old & ~CURRENT_OBJCG_UPDATE_FLAG);
3101 if (old)
3102 obj_cgroup_put(old);
3103
3104 old = NULL;
3105 }
3106
3107 /* If new objcg is NULL, no reason for the second atomic update. */
3108 if (!current->mm || (current->flags & PF_KTHREAD))
3109 return NULL;
3110
3111 /*
3112 * Release the objcg pointer from the previous iteration,
3113 * if try_cmpxcg() below fails.
3114 */
3115 if (unlikely(objcg)) {
3116 obj_cgroup_put(objcg);
3117 objcg = NULL;
3118 }
3119
3120 /*
3121 * Obtain the new objcg pointer. The current task can be
3122 * asynchronously moved to another memcg and the previous
3123 * memcg can be offlined. So let's get the memcg pointer
3124 * and try get a reference to objcg under a rcu read lock.
3125 */
3126
3127 rcu_read_lock();
3128 memcg = mem_cgroup_from_task(current);
3129 objcg = __get_obj_cgroup_from_memcg(memcg);
3130 rcu_read_unlock();
3131
3132 /*
3133 * Try set up a new objcg pointer atomically. If it
3134 * fails, it means the update flag was set concurrently, so
3135 * the whole procedure should be repeated.
3136 */
3137 } while (!try_cmpxchg(&current->objcg, &old, objcg));
3138
3139 return objcg;
3140}
3141
Roman Gushchine86828e2023-10-19 15:53:44 -07003142__always_inline struct obj_cgroup *current_obj_cgroup(void)
3143{
3144 struct mem_cgroup *memcg;
3145 struct obj_cgroup *objcg;
3146
3147 if (in_task()) {
3148 memcg = current->active_memcg;
3149 if (unlikely(memcg))
3150 goto from_memcg;
3151
3152 objcg = READ_ONCE(current->objcg);
3153 if (unlikely((unsigned long)objcg & CURRENT_OBJCG_UPDATE_FLAG))
3154 objcg = current_objcg_update();
3155 /*
3156 * Objcg reference is kept by the task, so it's safe
3157 * to use the objcg by the current task.
3158 */
3159 return objcg;
3160 }
3161
3162 memcg = this_cpu_read(int_active_memcg);
3163 if (unlikely(memcg))
3164 goto from_memcg;
3165
3166 return NULL;
3167
3168from_memcg:
3169 for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) {
3170 /*
3171 * Memcg pointer is protected by scope (see set_active_memcg())
3172 * and is pinning the corresponding objcg, so objcg can't go
3173 * away and can be used within the scope without any additional
3174 * protection.
3175 */
3176 objcg = rcu_dereference_check(memcg->objcg, 1);
3177 if (likely(objcg))
3178 break;
3179 objcg = NULL;
3180 }
3181
3182 return objcg;
3183}
3184
Matthew Wilcox (Oracle)074e3e22023-07-15 05:23:41 +01003185struct obj_cgroup *get_obj_cgroup_from_folio(struct folio *folio)
Johannes Weinerf4840cc2022-05-19 14:08:53 -07003186{
3187 struct obj_cgroup *objcg;
3188
Roman Gushchinf7a449f2023-02-13 11:29:22 -08003189 if (!memcg_kmem_online())
Johannes Weinerf4840cc2022-05-19 14:08:53 -07003190 return NULL;
3191
Matthew Wilcox (Oracle)074e3e22023-07-15 05:23:41 +01003192 if (folio_memcg_kmem(folio)) {
3193 objcg = __folio_objcg(folio);
Johannes Weinerf4840cc2022-05-19 14:08:53 -07003194 obj_cgroup_get(objcg);
3195 } else {
3196 struct mem_cgroup *memcg;
3197
3198 rcu_read_lock();
Matthew Wilcox (Oracle)074e3e22023-07-15 05:23:41 +01003199 memcg = __folio_memcg(folio);
Johannes Weinerf4840cc2022-05-19 14:08:53 -07003200 if (memcg)
3201 objcg = __get_obj_cgroup_from_memcg(memcg);
3202 else
3203 objcg = NULL;
3204 rcu_read_unlock();
3205 }
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003206 return objcg;
3207}
3208
Yosry Ahmeda8c49af2022-03-22 14:40:10 -07003209static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages)
3210{
3211 mod_memcg_state(memcg, MEMCG_KMEM, nr_pages);
3212 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
3213 if (nr_pages > 0)
3214 page_counter_charge(&memcg->kmem, nr_pages);
3215 else
3216 page_counter_uncharge(&memcg->kmem, -nr_pages);
3217 }
3218}
3219
3220
Muchun Songf1286fa2021-04-29 22:56:55 -07003221/*
3222 * obj_cgroup_uncharge_pages: uncharge a number of kernel pages from a objcg
3223 * @objcg: object cgroup to uncharge
3224 * @nr_pages: number of pages to uncharge
3225 */
Muchun Songe74d2252021-04-29 22:56:42 -07003226static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg,
3227 unsigned int nr_pages)
3228{
3229 struct mem_cgroup *memcg;
3230
3231 memcg = get_mem_cgroup_from_objcg(objcg);
Muchun Songf1286fa2021-04-29 22:56:55 -07003232
Yosry Ahmeda8c49af2022-03-22 14:40:10 -07003233 memcg_account_kmem(memcg, -nr_pages);
Muchun Songf1286fa2021-04-29 22:56:55 -07003234 refill_stock(memcg, nr_pages);
3235
Muchun Songe74d2252021-04-29 22:56:42 -07003236 css_put(&memcg->css);
3237}
3238
Muchun Songf1286fa2021-04-29 22:56:55 -07003239/*
3240 * obj_cgroup_charge_pages: charge a number of kernel pages to a objcg
3241 * @objcg: object cgroup to charge
Vladimir Davydov45264772016-07-26 15:24:21 -07003242 * @gfp: reclaim mode
Roman Gushchin92d0510c2020-04-01 21:06:49 -07003243 * @nr_pages: number of pages to charge
Vladimir Davydov45264772016-07-26 15:24:21 -07003244 *
3245 * Returns 0 on success, an error code on failure.
3246 */
Muchun Songf1286fa2021-04-29 22:56:55 -07003247static int obj_cgroup_charge_pages(struct obj_cgroup *objcg, gfp_t gfp,
3248 unsigned int nr_pages)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003249{
Muchun Songf1286fa2021-04-29 22:56:55 -07003250 struct mem_cgroup *memcg;
Johannes Weiner6071ca52015-11-05 18:50:26 -08003251 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003252
Muchun Songf1286fa2021-04-29 22:56:55 -07003253 memcg = get_mem_cgroup_from_objcg(objcg);
3254
Muchun Songc5c8b162021-06-28 19:37:44 -07003255 ret = try_charge_memcg(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08003256 if (ret)
Muchun Songf1286fa2021-04-29 22:56:55 -07003257 goto out;
Johannes Weiner52c29b02016-01-20 15:02:35 -08003258
Yosry Ahmeda8c49af2022-03-22 14:40:10 -07003259 memcg_account_kmem(memcg, nr_pages);
Muchun Songf1286fa2021-04-29 22:56:55 -07003260out:
3261 css_put(&memcg->css);
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003262
Muchun Songf1286fa2021-04-29 22:56:55 -07003263 return ret;
Roman Gushchin4b13f642020-04-01 21:06:56 -07003264}
3265
3266/**
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003267 * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup
Vladimir Davydov45264772016-07-26 15:24:21 -07003268 * @page: page to charge
3269 * @gfp: reclaim mode
3270 * @order: allocation order
3271 *
3272 * Returns 0 on success, an error code on failure.
3273 */
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003274int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003275{
Muchun Songb4e0b682021-04-29 22:56:52 -07003276 struct obj_cgroup *objcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07003277 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003278
Roman Gushchine86828e2023-10-19 15:53:44 -07003279 objcg = current_obj_cgroup();
Muchun Songb4e0b682021-04-29 22:56:52 -07003280 if (objcg) {
3281 ret = obj_cgroup_charge_pages(objcg, gfp, 1 << order);
Roman Gushchin4d96ba32019-07-11 20:56:31 -07003282 if (!ret) {
Roman Gushchine86828e2023-10-19 15:53:44 -07003283 obj_cgroup_get(objcg);
Muchun Songb4e0b682021-04-29 22:56:52 -07003284 page->memcg_data = (unsigned long)objcg |
Roman Gushchin18b2db32020-12-01 13:58:30 -08003285 MEMCG_DATA_KMEM;
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07003286 return 0;
Roman Gushchin4d96ba32019-07-11 20:56:31 -07003287 }
Vladimir Davydovc4159a72016-08-08 23:03:12 +03003288 }
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08003289 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003290}
Roman Gushchin49a18ea2019-07-11 20:56:13 -07003291
3292/**
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003293 * __memcg_kmem_uncharge_page: uncharge a kmem page
Vladimir Davydov45264772016-07-26 15:24:21 -07003294 * @page: page to uncharge
3295 * @order: allocation order
3296 */
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003297void __memcg_kmem_uncharge_page(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003298{
Matthew Wilcox (Oracle)1b7e4462021-06-28 14:59:26 -04003299 struct folio *folio = page_folio(page);
Muchun Songb4e0b682021-04-29 22:56:52 -07003300 struct obj_cgroup *objcg;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003301 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003302
Matthew Wilcox (Oracle)1b7e4462021-06-28 14:59:26 -04003303 if (!folio_memcg_kmem(folio))
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003304 return;
3305
Matthew Wilcox (Oracle)1b7e4462021-06-28 14:59:26 -04003306 objcg = __folio_objcg(folio);
Muchun Songb4e0b682021-04-29 22:56:52 -07003307 obj_cgroup_uncharge_pages(objcg, nr_pages);
Matthew Wilcox (Oracle)1b7e4462021-06-28 14:59:26 -04003308 folio->memcg_data = 0;
Muchun Songb4e0b682021-04-29 22:56:52 -07003309 obj_cgroup_put(objcg);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08003310}
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003311
Waiman Long68ac5b3c2021-06-28 19:37:23 -07003312void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
3313 enum node_stat_item idx, int nr)
3314{
Michal Hockofead2b82022-03-22 14:40:35 -07003315 struct memcg_stock_pcp *stock;
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07003316 struct obj_cgroup *old = NULL;
Waiman Long68ac5b3c2021-06-28 19:37:23 -07003317 unsigned long flags;
3318 int *bytes;
3319
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07003320 local_lock_irqsave(&memcg_stock.stock_lock, flags);
Michal Hockofead2b82022-03-22 14:40:35 -07003321 stock = this_cpu_ptr(&memcg_stock);
3322
Waiman Long68ac5b3c2021-06-28 19:37:23 -07003323 /*
3324 * Save vmstat data in stock and skip vmstat array update unless
3325 * accumulating over a page of vmstat data or when pgdat or idx
3326 * changes.
3327 */
Roman Gushchin3b8abb32023-05-02 09:08:38 -07003328 if (READ_ONCE(stock->cached_objcg) != objcg) {
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07003329 old = drain_obj_stock(stock);
Waiman Long68ac5b3c2021-06-28 19:37:23 -07003330 obj_cgroup_get(objcg);
3331 stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes)
3332 ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0;
Roman Gushchin3b8abb32023-05-02 09:08:38 -07003333 WRITE_ONCE(stock->cached_objcg, objcg);
Waiman Long68ac5b3c2021-06-28 19:37:23 -07003334 stock->cached_pgdat = pgdat;
3335 } else if (stock->cached_pgdat != pgdat) {
3336 /* Flush the existing cached vmstat data */
Waiman Long7fa0dac2021-08-13 16:54:41 -07003337 struct pglist_data *oldpg = stock->cached_pgdat;
3338
Waiman Long68ac5b3c2021-06-28 19:37:23 -07003339 if (stock->nr_slab_reclaimable_b) {
Waiman Long7fa0dac2021-08-13 16:54:41 -07003340 mod_objcg_mlstate(objcg, oldpg, NR_SLAB_RECLAIMABLE_B,
Waiman Long68ac5b3c2021-06-28 19:37:23 -07003341 stock->nr_slab_reclaimable_b);
3342 stock->nr_slab_reclaimable_b = 0;
3343 }
3344 if (stock->nr_slab_unreclaimable_b) {
Waiman Long7fa0dac2021-08-13 16:54:41 -07003345 mod_objcg_mlstate(objcg, oldpg, NR_SLAB_UNRECLAIMABLE_B,
Waiman Long68ac5b3c2021-06-28 19:37:23 -07003346 stock->nr_slab_unreclaimable_b);
3347 stock->nr_slab_unreclaimable_b = 0;
3348 }
3349 stock->cached_pgdat = pgdat;
3350 }
3351
3352 bytes = (idx == NR_SLAB_RECLAIMABLE_B) ? &stock->nr_slab_reclaimable_b
3353 : &stock->nr_slab_unreclaimable_b;
3354 /*
3355 * Even for large object >= PAGE_SIZE, the vmstat data will still be
3356 * cached locally at least once before pushing it out.
3357 */
3358 if (!*bytes) {
3359 *bytes = nr;
3360 nr = 0;
3361 } else {
3362 *bytes += nr;
3363 if (abs(*bytes) > PAGE_SIZE) {
3364 nr = *bytes;
3365 *bytes = 0;
3366 } else {
3367 nr = 0;
3368 }
3369 }
3370 if (nr)
3371 mod_objcg_mlstate(objcg, pgdat, idx, nr);
3372
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07003373 local_unlock_irqrestore(&memcg_stock.stock_lock, flags);
3374 if (old)
3375 obj_cgroup_put(old);
Waiman Long68ac5b3c2021-06-28 19:37:23 -07003376}
3377
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003378static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes)
3379{
Michal Hockofead2b82022-03-22 14:40:35 -07003380 struct memcg_stock_pcp *stock;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003381 unsigned long flags;
3382 bool ret = false;
3383
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07003384 local_lock_irqsave(&memcg_stock.stock_lock, flags);
Michal Hockofead2b82022-03-22 14:40:35 -07003385
3386 stock = this_cpu_ptr(&memcg_stock);
Roman Gushchin3b8abb32023-05-02 09:08:38 -07003387 if (objcg == READ_ONCE(stock->cached_objcg) && stock->nr_bytes >= nr_bytes) {
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003388 stock->nr_bytes -= nr_bytes;
3389 ret = true;
3390 }
3391
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07003392 local_unlock_irqrestore(&memcg_stock.stock_lock, flags);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003393
3394 return ret;
3395}
3396
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07003397static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock)
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003398{
Roman Gushchin3b8abb32023-05-02 09:08:38 -07003399 struct obj_cgroup *old = READ_ONCE(stock->cached_objcg);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003400
3401 if (!old)
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07003402 return NULL;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003403
3404 if (stock->nr_bytes) {
3405 unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT;
3406 unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1);
3407
Johannes Weineraf9a3b62022-03-22 14:40:44 -07003408 if (nr_pages) {
3409 struct mem_cgroup *memcg;
3410
3411 memcg = get_mem_cgroup_from_objcg(old);
3412
3413 memcg_account_kmem(memcg, -nr_pages);
3414 __refill_stock(memcg, nr_pages);
3415
3416 css_put(&memcg->css);
3417 }
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003418
3419 /*
3420 * The leftover is flushed to the centralized per-memcg value.
3421 * On the next attempt to refill obj stock it will be moved
3422 * to a per-cpu stock (probably, on an other CPU), see
3423 * refill_obj_stock().
3424 *
3425 * How often it's flushed is a trade-off between the memory
3426 * limit enforcement accuracy and potential CPU contention,
3427 * so it might be changed in the future.
3428 */
3429 atomic_add(nr_bytes, &old->nr_charged_bytes);
3430 stock->nr_bytes = 0;
3431 }
3432
Waiman Long68ac5b3c2021-06-28 19:37:23 -07003433 /*
3434 * Flush the vmstat data in current stock
3435 */
3436 if (stock->nr_slab_reclaimable_b || stock->nr_slab_unreclaimable_b) {
3437 if (stock->nr_slab_reclaimable_b) {
3438 mod_objcg_mlstate(old, stock->cached_pgdat,
3439 NR_SLAB_RECLAIMABLE_B,
3440 stock->nr_slab_reclaimable_b);
3441 stock->nr_slab_reclaimable_b = 0;
3442 }
3443 if (stock->nr_slab_unreclaimable_b) {
3444 mod_objcg_mlstate(old, stock->cached_pgdat,
3445 NR_SLAB_UNRECLAIMABLE_B,
3446 stock->nr_slab_unreclaimable_b);
3447 stock->nr_slab_unreclaimable_b = 0;
3448 }
3449 stock->cached_pgdat = NULL;
3450 }
3451
Roman Gushchin3b8abb32023-05-02 09:08:38 -07003452 WRITE_ONCE(stock->cached_objcg, NULL);
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07003453 /*
3454 * The `old' objects needs to be released by the caller via
3455 * obj_cgroup_put() outside of memcg_stock_pcp::stock_lock.
3456 */
3457 return old;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003458}
3459
3460static bool obj_stock_flush_required(struct memcg_stock_pcp *stock,
3461 struct mem_cgroup *root_memcg)
3462{
Roman Gushchin3b8abb32023-05-02 09:08:38 -07003463 struct obj_cgroup *objcg = READ_ONCE(stock->cached_objcg);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003464 struct mem_cgroup *memcg;
3465
Roman Gushchin3b8abb32023-05-02 09:08:38 -07003466 if (objcg) {
3467 memcg = obj_cgroup_memcg(objcg);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003468 if (memcg && mem_cgroup_is_descendant(memcg, root_memcg))
3469 return true;
3470 }
3471
3472 return false;
3473}
3474
Waiman Long5387c9042021-06-28 19:37:27 -07003475static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes,
3476 bool allow_uncharge)
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003477{
Michal Hockofead2b82022-03-22 14:40:35 -07003478 struct memcg_stock_pcp *stock;
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07003479 struct obj_cgroup *old = NULL;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003480 unsigned long flags;
Waiman Long5387c9042021-06-28 19:37:27 -07003481 unsigned int nr_pages = 0;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003482
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07003483 local_lock_irqsave(&memcg_stock.stock_lock, flags);
Michal Hockofead2b82022-03-22 14:40:35 -07003484
3485 stock = this_cpu_ptr(&memcg_stock);
Roman Gushchin3b8abb32023-05-02 09:08:38 -07003486 if (READ_ONCE(stock->cached_objcg) != objcg) { /* reset if necessary */
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07003487 old = drain_obj_stock(stock);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003488 obj_cgroup_get(objcg);
Roman Gushchin3b8abb32023-05-02 09:08:38 -07003489 WRITE_ONCE(stock->cached_objcg, objcg);
Waiman Long5387c9042021-06-28 19:37:27 -07003490 stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes)
3491 ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0;
3492 allow_uncharge = true; /* Allow uncharge when objcg changes */
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003493 }
3494 stock->nr_bytes += nr_bytes;
3495
Waiman Long5387c9042021-06-28 19:37:27 -07003496 if (allow_uncharge && (stock->nr_bytes > PAGE_SIZE)) {
3497 nr_pages = stock->nr_bytes >> PAGE_SHIFT;
3498 stock->nr_bytes &= (PAGE_SIZE - 1);
3499 }
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003500
Sebastian Andrzej Siewior56751142022-03-22 14:40:47 -07003501 local_unlock_irqrestore(&memcg_stock.stock_lock, flags);
3502 if (old)
3503 obj_cgroup_put(old);
Waiman Long5387c9042021-06-28 19:37:27 -07003504
3505 if (nr_pages)
3506 obj_cgroup_uncharge_pages(objcg, nr_pages);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003507}
3508
3509int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size)
3510{
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003511 unsigned int nr_pages, nr_bytes;
3512 int ret;
3513
3514 if (consume_obj_stock(objcg, size))
3515 return 0;
3516
3517 /*
Waiman Long5387c9042021-06-28 19:37:27 -07003518 * In theory, objcg->nr_charged_bytes can have enough
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003519 * pre-charged bytes to satisfy the allocation. However,
Waiman Long5387c9042021-06-28 19:37:27 -07003520 * flushing objcg->nr_charged_bytes requires two atomic
3521 * operations, and objcg->nr_charged_bytes can't be big.
3522 * The shared objcg->nr_charged_bytes can also become a
3523 * performance bottleneck if all tasks of the same memcg are
3524 * trying to update it. So it's better to ignore it and try
3525 * grab some new pages. The stock's nr_bytes will be flushed to
3526 * objcg->nr_charged_bytes later on when objcg changes.
3527 *
3528 * The stock's nr_bytes may contain enough pre-charged bytes
3529 * to allow one less page from being charged, but we can't rely
3530 * on the pre-charged bytes not being changed outside of
3531 * consume_obj_stock() or refill_obj_stock(). So ignore those
3532 * pre-charged bytes as well when charging pages. To avoid a
3533 * page uncharge right after a page charge, we set the
3534 * allow_uncharge flag to false when calling refill_obj_stock()
3535 * to temporarily allow the pre-charged bytes to exceed the page
3536 * size limit. The maximum reachable value of the pre-charged
3537 * bytes is (sizeof(object) + PAGE_SIZE - 2) if there is no data
3538 * race.
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003539 */
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003540 nr_pages = size >> PAGE_SHIFT;
3541 nr_bytes = size & (PAGE_SIZE - 1);
3542
3543 if (nr_bytes)
3544 nr_pages += 1;
3545
Muchun Songe74d2252021-04-29 22:56:42 -07003546 ret = obj_cgroup_charge_pages(objcg, gfp, nr_pages);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003547 if (!ret && nr_bytes)
Waiman Long5387c9042021-06-28 19:37:27 -07003548 refill_obj_stock(objcg, PAGE_SIZE - nr_bytes, false);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003549
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003550 return ret;
3551}
3552
3553void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size)
3554{
Waiman Long5387c9042021-06-28 19:37:27 -07003555 refill_obj_stock(objcg, size, true);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003556}
3557
Kirill Tkhai84c07d12018-08-17 15:47:25 -07003558#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003559
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003560/*
Zhou Guanghuibe6c8982021-03-12 21:08:30 -08003561 * Because page_memcg(head) is not set on tails, set it now.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003562 */
Zhou Guanghuibe6c8982021-03-12 21:08:30 -08003563void split_page_memcg(struct page *head, unsigned int nr)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003564{
Matthew Wilcox (Oracle)1b7e4462021-06-28 14:59:26 -04003565 struct folio *folio = page_folio(head);
3566 struct mem_cgroup *memcg = folio_memcg(folio);
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003567 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003568
Zhou Guanghuibe6c8982021-03-12 21:08:30 -08003569 if (mem_cgroup_disabled() || !memcg)
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08003570 return;
David Rientjesb070e652013-05-07 16:18:09 -07003571
Zhou Guanghuibe6c8982021-03-12 21:08:30 -08003572 for (i = 1; i < nr; i++)
Matthew Wilcox (Oracle)1b7e4462021-06-28 14:59:26 -04003573 folio_page(folio, i)->memcg_data = folio->memcg_data;
Muchun Songb4e0b682021-04-29 22:56:52 -07003574
Matthew Wilcox (Oracle)1b7e4462021-06-28 14:59:26 -04003575 if (folio_memcg_kmem(folio))
3576 obj_cgroup_get_many(__folio_objcg(folio), nr - 1);
Muchun Songb4e0b682021-04-29 22:56:52 -07003577 else
3578 css_get_many(&memcg->css, nr - 1);
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003579}
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003580
Johannes Weinere55b9f92022-09-26 09:57:04 -04003581#ifdef CONFIG_SWAP
Daisuke Nishimura02491442010-03-10 15:22:17 -08003582/**
3583 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3584 * @entry: swap entry to be moved
3585 * @from: mem_cgroup which the entry is moved from
3586 * @to: mem_cgroup which the entry is moved to
3587 *
3588 * It succeeds only when the swap_cgroup's record for this entry is the same
3589 * as the mem_cgroup's id of @from.
3590 *
3591 * Returns 0 on success, -EINVAL on failure.
3592 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003593 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08003594 * both res and memsw, and called css_get().
3595 */
3596static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003597 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003598{
3599 unsigned short old_id, new_id;
3600
Li Zefan34c00c32013-09-23 16:56:01 +08003601 old_id = mem_cgroup_id(from);
3602 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003603
3604 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Johannes Weinerc9019e92018-01-31 16:16:37 -08003605 mod_memcg_state(from, MEMCG_SWAP, -1);
3606 mod_memcg_state(to, MEMCG_SWAP, 1);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003607 return 0;
3608 }
3609 return -EINVAL;
3610}
3611#else
3612static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003613 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003614{
3615 return -EINVAL;
3616}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003617#endif
3618
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003619static DEFINE_MUTEX(memcg_max_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003620
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003621static int mem_cgroup_resize_max(struct mem_cgroup *memcg,
3622 unsigned long max, bool memsw)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003623{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003624 bool enlarge = false;
Shakeel Buttbb4a7ea2018-06-07 17:07:27 -07003625 bool drained = false;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003626 int ret;
Yu Zhaoc054a782018-01-31 16:20:02 -08003627 bool limits_invariant;
3628 struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003629
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003630 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003631 if (signal_pending(current)) {
3632 ret = -EINTR;
3633 break;
3634 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003635
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003636 mutex_lock(&memcg_max_mutex);
Yu Zhaoc054a782018-01-31 16:20:02 -08003637 /*
3638 * Make sure that the new limit (memsw or memory limit) doesn't
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003639 * break our basic invariant rule memory.max <= memsw.max.
Yu Zhaoc054a782018-01-31 16:20:02 -08003640 */
Chris Down15b42562020-04-01 21:07:20 -07003641 limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) :
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003642 max <= memcg->memsw.max;
Yu Zhaoc054a782018-01-31 16:20:02 -08003643 if (!limits_invariant) {
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003644 mutex_unlock(&memcg_max_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003645 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003646 break;
3647 }
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003648 if (max > counter->max)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003649 enlarge = true;
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003650 ret = page_counter_set_max(counter, max);
3651 mutex_unlock(&memcg_max_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003652
3653 if (!ret)
3654 break;
3655
Shakeel Buttbb4a7ea2018-06-07 17:07:27 -07003656 if (!drained) {
3657 drain_all_stock(memcg);
3658 drained = true;
3659 continue;
3660 }
3661
Yosry Ahmed73b73ba2022-07-14 06:49:18 +00003662 if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL,
Michal Hocko55ab8342022-12-16 10:46:33 +01003663 memsw ? 0 : MEMCG_RECLAIM_MAY_SWAP)) {
Andrey Ryabinin1ab5c052018-01-31 16:20:37 -08003664 ret = -EBUSY;
3665 break;
3666 }
3667 } while (true);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003668
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003669 if (!ret && enlarge)
3670 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003671
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003672 return ret;
3673}
3674
Mel Gormanef8f2322016-07-28 15:46:05 -07003675unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
Andrew Morton0608f432013-09-24 15:27:41 -07003676 gfp_t gfp_mask,
3677 unsigned long *total_scanned)
3678{
3679 unsigned long nr_reclaimed = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07003680 struct mem_cgroup_per_node *mz, *next_mz = NULL;
Andrew Morton0608f432013-09-24 15:27:41 -07003681 unsigned long reclaimed;
3682 int loop = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07003683 struct mem_cgroup_tree_per_node *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003684 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07003685
Yu Zhaoe4dde562022-12-21 21:19:04 -07003686 if (lru_gen_enabled())
3687 return 0;
3688
Andrew Morton0608f432013-09-24 15:27:41 -07003689 if (order > 0)
3690 return 0;
3691
Matthew Wilcox (Oracle)2ab082b2021-06-25 08:57:02 -04003692 mctz = soft_limit_tree.rb_tree_per_node[pgdat->node_id];
Michal Hockod6507ff2016-08-02 14:02:37 -07003693
3694 /*
3695 * Do not even bother to check the largest node if the root
3696 * is empty. Do it lockless to prevent lock bouncing. Races
3697 * are acceptable as soft limit is best effort anyway.
3698 */
Laurent Dufourbfc72282017-03-09 16:17:06 -08003699 if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root))
Michal Hockod6507ff2016-08-02 14:02:37 -07003700 return 0;
3701
Andrew Morton0608f432013-09-24 15:27:41 -07003702 /*
3703 * This loop can run a while, specially if mem_cgroup's continuously
3704 * keep exceeding their soft limit and putting the system under
3705 * pressure
3706 */
3707 do {
3708 if (next_mz)
3709 mz = next_mz;
3710 else
3711 mz = mem_cgroup_largest_soft_limit_node(mctz);
3712 if (!mz)
3713 break;
3714
Mel Gormanef8f2322016-07-28 15:46:05 -07003715 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
Miaohe Lind8f65332022-04-28 23:15:58 -07003716 gfp_mask, total_scanned);
Andrew Morton0608f432013-09-24 15:27:41 -07003717 nr_reclaimed += reclaimed;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003718 spin_lock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07003719
3720 /*
3721 * If we failed to reclaim anything from this memory cgroup
3722 * it is time to move on to the next cgroup
3723 */
3724 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08003725 if (!reclaimed)
3726 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
3727
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003728 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07003729 /*
3730 * One school of thought says that we should not add
3731 * back the node to the tree if reclaim returns 0.
3732 * But our reclaim could return 0, simply because due
3733 * to priority we are exposing a smaller subset of
3734 * memory to reclaim from. Consider this as a longer
3735 * term TODO.
3736 */
3737 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07003738 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003739 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07003740 css_put(&mz->memcg->css);
3741 loop++;
3742 /*
3743 * Could not reclaim anything and there are no more
3744 * mem cgroups to try or we seem to be looping without
3745 * reclaiming anything.
3746 */
3747 if (!nr_reclaimed &&
3748 (next_mz == NULL ||
3749 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3750 break;
3751 } while (!nr_reclaimed);
3752 if (next_mz)
3753 css_put(&next_mz->memcg->css);
3754 return nr_reclaimed;
3755}
3756
Tejun Heoea280e72014-05-16 13:22:48 -04003757/*
Greg Thelen51038172016-05-20 16:58:18 -07003758 * Reclaims as many pages from the given memcg as possible.
Michal Hockoc26251f2012-10-26 13:37:28 +02003759 *
3760 * Caller is responsible for holding css reference for memcg.
3761 */
3762static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
3763{
Chris Downd977aa92020-08-06 23:21:58 -07003764 int nr_retries = MAX_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02003765
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003766 /* we call try-to-free pages for make this cgroup empty */
3767 lru_add_drain_all();
Junaid Shahidd12c60f2018-06-07 17:07:31 -07003768
3769 drain_all_stock(memcg);
3770
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003771 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003772 while (nr_retries && page_counter_read(&memcg->memory)) {
Michal Hockoc26251f2012-10-26 13:37:28 +02003773 if (signal_pending(current))
3774 return -EINTR;
3775
Yosry Ahmed73b73ba2022-07-14 06:49:18 +00003776 if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL,
Michal Hocko55ab8342022-12-16 10:46:33 +01003777 MEMCG_RECLAIM_MAY_SWAP))
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003778 nr_retries--;
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003779 }
Michal Hockoab5196c2012-10-26 13:37:32 +02003780
3781 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003782}
3783
Tejun Heo6770c642014-05-13 12:16:21 -04003784static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
3785 char *buf, size_t nbytes,
3786 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003787{
Tejun Heo6770c642014-05-13 12:16:21 -04003788 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02003789
Michal Hockod8423012012-10-26 13:37:29 +02003790 if (mem_cgroup_is_root(memcg))
3791 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04003792 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003793}
3794
Tejun Heo182446d2013-08-08 20:11:24 -04003795static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
3796 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08003797{
Roman Gushchinbef86202020-12-14 19:06:49 -08003798 return 1;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003799}
3800
Tejun Heo182446d2013-08-08 20:11:24 -04003801static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
3802 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08003803{
Roman Gushchinbef86202020-12-14 19:06:49 -08003804 if (val == 1)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003805 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07003806
Roman Gushchinbef86202020-12-14 19:06:49 -08003807 pr_warn_once("Non-hierarchical mode is deprecated. "
3808 "Please report your usecase to linux-mm@kvack.org if you "
3809 "depend on this functionality.\n");
Glauber Costa567fb432012-07-31 16:43:07 -07003810
Roman Gushchinbef86202020-12-14 19:06:49 -08003811 return -EINVAL;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003812}
3813
Andrew Morton6f646152015-11-06 16:28:58 -08003814static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04003815{
Johannes Weiner42a30032019-05-14 15:47:12 -07003816 unsigned long val;
Johannes Weinerce00a962014-09-05 08:43:57 -04003817
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003818 if (mem_cgroup_is_root(memcg)) {
Yosry Ahmeda2174e92023-03-30 19:17:56 +00003819 /*
Yosry Ahmedf82a7a82023-04-21 17:40:18 +00003820 * Approximate root's usage from global state. This isn't
3821 * perfect, but the root usage was always an approximation.
Yosry Ahmeda2174e92023-03-30 19:17:56 +00003822 */
Yosry Ahmedf82a7a82023-04-21 17:40:18 +00003823 val = global_node_page_state(NR_FILE_PAGES) +
3824 global_node_page_state(NR_ANON_MAPPED);
Johannes Weiner42a30032019-05-14 15:47:12 -07003825 if (swap)
Yosry Ahmedf82a7a82023-04-21 17:40:18 +00003826 val += total_swap_pages - get_nr_swap_pages();
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003827 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04003828 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003829 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04003830 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003831 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04003832 }
Michal Hockoc12176d2015-11-05 18:50:29 -08003833 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04003834}
3835
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003836enum {
3837 RES_USAGE,
3838 RES_LIMIT,
3839 RES_MAX_USAGE,
3840 RES_FAILCNT,
3841 RES_SOFT_LIMIT,
3842};
Johannes Weinerce00a962014-09-05 08:43:57 -04003843
Tejun Heo791badb2013-12-05 12:28:02 -05003844static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07003845 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003846{
Tejun Heo182446d2013-08-08 20:11:24 -04003847 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003848 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07003849
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003850 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003851 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003852 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08003853 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003854 case _MEMSWAP:
3855 counter = &memcg->memsw;
3856 break;
3857 case _KMEM:
3858 counter = &memcg->kmem;
3859 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003860 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003861 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003862 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003863 default:
3864 BUG();
3865 }
3866
3867 switch (MEMFILE_ATTR(cft->private)) {
3868 case RES_USAGE:
3869 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08003870 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003871 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08003872 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003873 return (u64)page_counter_read(counter) * PAGE_SIZE;
3874 case RES_LIMIT:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003875 return (u64)counter->max * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003876 case RES_MAX_USAGE:
3877 return (u64)counter->watermark * PAGE_SIZE;
3878 case RES_FAILCNT:
3879 return counter->failcnt;
3880 case RES_SOFT_LIMIT:
Yue Zhao2178e202023-03-06 23:41:38 +08003881 return (u64)READ_ONCE(memcg->soft_limit) * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003882 default:
3883 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003884 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003885}
Glauber Costa510fc4e2012-12-18 14:21:47 -08003886
Florian Schmidt6b0ba2a2023-04-04 10:58:59 +00003887/*
3888 * This function doesn't do anything useful. Its only job is to provide a read
3889 * handler for a file so that cgroup_file_mode() will add read permissions.
3890 */
3891static int mem_cgroup_dummy_seq_show(__always_unused struct seq_file *m,
3892 __always_unused void *v)
3893{
3894 return -EINVAL;
3895}
3896
Kirill Tkhai84c07d12018-08-17 15:47:25 -07003897#ifdef CONFIG_MEMCG_KMEM
Johannes Weiner567e9ab2016-01-20 15:02:24 -08003898static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08003899{
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003900 struct obj_cgroup *objcg;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003901
Xiang Yang9c94bef2022-06-25 14:18:44 +08003902 if (mem_cgroup_kmem_disabled())
Vladimir Davydovb313aee2016-03-17 14:18:27 -07003903 return 0;
3904
Muchun Songda0efe32022-03-22 14:41:15 -07003905 if (unlikely(mem_cgroup_is_root(memcg)))
3906 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003907
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003908 objcg = obj_cgroup_alloc();
Muchun Songf9c69d62022-03-22 14:41:28 -07003909 if (!objcg)
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003910 return -ENOMEM;
Muchun Songf9c69d62022-03-22 14:41:28 -07003911
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003912 objcg->memcg = memcg;
3913 rcu_assign_pointer(memcg->objcg, objcg);
Roman Gushchin675d6c92023-10-19 15:53:43 -07003914 obj_cgroup_get(objcg);
3915 memcg->orig_objcg = objcg;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003916
Roman Gushchinf7a449f2023-02-13 11:29:22 -08003917 static_branch_enable(&memcg_kmem_online_key);
Roman Gushchind648bcc2020-08-06 23:20:28 -07003918
Muchun Songf9c69d62022-03-22 14:41:28 -07003919 memcg->kmemcg_id = memcg->id.id;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003920
3921 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003922}
3923
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003924static void memcg_offline_kmem(struct mem_cgroup *memcg)
3925{
Muchun Song64268862021-11-05 13:37:53 -07003926 struct mem_cgroup *parent;
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003927
Xiang Yang9c94bef2022-06-25 14:18:44 +08003928 if (mem_cgroup_kmem_disabled())
Muchun Songda0efe32022-03-22 14:41:15 -07003929 return;
3930
3931 if (unlikely(mem_cgroup_is_root(memcg)))
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003932 return;
Roman Gushchin98556092020-08-06 23:21:10 -07003933
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003934 parent = parent_mem_cgroup(memcg);
3935 if (!parent)
3936 parent = root_mem_cgroup;
3937
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003938 memcg_reparent_objcgs(memcg, parent);
Roman Gushchinfb2f2b02019-07-11 20:56:34 -07003939
Muchun Song5abc1e32022-03-22 14:41:19 -07003940 /*
Muchun Song64268862021-11-05 13:37:53 -07003941 * After we have finished memcg_reparent_objcgs(), all list_lrus
3942 * corresponding to this cgroup are guaranteed to remain empty.
3943 * The ordering is imposed by list_lru_node->lock taken by
Muchun Song1f391eb2022-03-22 14:41:22 -07003944 * memcg_reparent_list_lrus().
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003945 */
Muchun Song1f391eb2022-03-22 14:41:22 -07003946 memcg_reparent_list_lrus(memcg, parent);
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003947}
Vladimir Davydovd6441632014-01-23 15:53:09 -08003948#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003949static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08003950{
3951 return 0;
3952}
3953static void memcg_offline_kmem(struct mem_cgroup *memcg)
3954{
3955}
Kirill Tkhai84c07d12018-08-17 15:47:25 -07003956#endif /* CONFIG_MEMCG_KMEM */
Johannes Weiner127424c2016-01-20 15:02:32 -08003957
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003958static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003959{
3960 int ret;
3961
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003962 mutex_lock(&memcg_max_mutex);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003963
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003964 ret = page_counter_set_max(&memcg->tcpmem, max);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003965 if (ret)
3966 goto out;
3967
Johannes Weiner0db15292016-01-20 15:02:50 -08003968 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003969 /*
3970 * The active flag needs to be written after the static_key
3971 * update. This is what guarantees that the socket activation
Johannes Weiner2d758072016-10-07 17:00:58 -07003972 * function is the last one to run. See mem_cgroup_sk_alloc()
3973 * for details, and note that we don't mark any socket as
3974 * belonging to this memcg until that flag is up.
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003975 *
3976 * We need to do this, because static_keys will span multiple
3977 * sites, but we can't control their order. If we mark a socket
3978 * as accounted, but the accounting functions are not patched in
3979 * yet, we'll lose accounting.
3980 *
Johannes Weiner2d758072016-10-07 17:00:58 -07003981 * We never race with the readers in mem_cgroup_sk_alloc(),
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003982 * because when this value change, the code to process it is not
3983 * patched in yet.
3984 */
3985 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08003986 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003987 }
3988out:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003989 mutex_unlock(&memcg_max_mutex);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003990 return ret;
3991}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003992
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003993/*
3994 * The user of this function is...
3995 * RES_LIMIT.
3996 */
Tejun Heo451af5042014-05-13 12:16:21 -04003997static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
3998 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003999{
Tejun Heo451af5042014-05-13 12:16:21 -04004000 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004001 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004002 int ret;
4003
Tejun Heo451af5042014-05-13 12:16:21 -04004004 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08004005 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004006 if (ret)
4007 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07004008
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004009 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004010 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07004011 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
4012 ret = -EINVAL;
4013 break;
4014 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004015 switch (MEMFILE_TYPE(of_cft(of)->private)) {
4016 case _MEM:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07004017 ret = mem_cgroup_resize_max(memcg, nr_pages, false);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004018 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004019 case _MEMSWAP:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07004020 ret = mem_cgroup_resize_max(memcg, nr_pages, true);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004021 break;
Michal Hocko45976482023-09-21 09:38:29 +02004022 case _KMEM:
4023 pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. "
4024 "Writing any value to this file has no effect. "
4025 "Please report your usecase to linux-mm@kvack.org if you "
4026 "depend on this functionality.\n");
4027 ret = 0;
4028 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004029 case _TCP:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07004030 ret = memcg_update_tcp_max(memcg, nr_pages);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004031 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004032 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004033 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07004034 case RES_SOFT_LIMIT:
Sebastian Andrzej Siewior2343e882022-03-22 14:40:38 -07004035 if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
4036 ret = -EOPNOTSUPP;
4037 } else {
Yue Zhao2178e202023-03-06 23:41:38 +08004038 WRITE_ONCE(memcg->soft_limit, nr_pages);
Sebastian Andrzej Siewior2343e882022-03-22 14:40:38 -07004039 ret = 0;
4040 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07004041 break;
4042 }
Tejun Heo451af5042014-05-13 12:16:21 -04004043 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004044}
4045
Tejun Heo6770c642014-05-13 12:16:21 -04004046static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
4047 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004048{
Tejun Heo6770c642014-05-13 12:16:21 -04004049 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004050 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004051
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004052 switch (MEMFILE_TYPE(of_cft(of)->private)) {
4053 case _MEM:
4054 counter = &memcg->memory;
4055 break;
4056 case _MEMSWAP:
4057 counter = &memcg->memsw;
4058 break;
4059 case _KMEM:
4060 counter = &memcg->kmem;
4061 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004062 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08004063 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08004064 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004065 default:
4066 BUG();
4067 }
Tejun Heoaf36f902012-04-01 12:09:55 -07004068
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004069 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004070 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004071 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004072 break;
4073 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004074 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004075 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004076 default:
4077 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07004078 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07004079
Tejun Heo6770c642014-05-13 12:16:21 -04004080 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004081}
4082
Tejun Heo182446d2013-08-08 20:11:24 -04004083static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004084 struct cftype *cft)
4085{
Tejun Heo182446d2013-08-08 20:11:24 -04004086 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004087}
4088
Daisuke Nishimura02491442010-03-10 15:22:17 -08004089#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04004090static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004091 struct cftype *cft, u64 val)
4092{
Tejun Heo182446d2013-08-08 20:11:24 -04004093 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004094
Johannes Weinerda34a842022-12-07 14:00:39 +01004095 pr_warn_once("Cgroup memory moving (move_charge_at_immigrate) is deprecated. "
4096 "Please report your usecase to linux-mm@kvack.org if you "
4097 "depend on this functionality.\n");
4098
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08004099 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004100 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004101
Glauber Costaee5e8472013-02-22 16:34:50 -08004102 /*
4103 * No kind of locking is needed in here, because ->can_attach() will
4104 * check this value once in the beginning of the process, and then carry
4105 * on with stale data. This means that changes to this value will only
4106 * affect task migrations starting after the change.
4107 */
4108 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004109 return 0;
4110}
Daisuke Nishimura02491442010-03-10 15:22:17 -08004111#else
Tejun Heo182446d2013-08-08 20:11:24 -04004112static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08004113 struct cftype *cft, u64 val)
4114{
4115 return -ENOSYS;
4116}
4117#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08004118
Ying Han406eb0c2011-05-26 16:25:37 -07004119#ifdef CONFIG_NUMA
Johannes Weiner113b7df2019-05-13 17:18:11 -07004120
4121#define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
4122#define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
4123#define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
4124
4125static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
Shakeel Buttdd8657b2020-06-03 15:56:24 -07004126 int nid, unsigned int lru_mask, bool tree)
Johannes Weiner113b7df2019-05-13 17:18:11 -07004127{
Johannes Weiner867e5e12019-11-30 17:55:34 -08004128 struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid));
Johannes Weiner113b7df2019-05-13 17:18:11 -07004129 unsigned long nr = 0;
4130 enum lru_list lru;
4131
4132 VM_BUG_ON((unsigned)nid >= nr_node_ids);
4133
4134 for_each_lru(lru) {
4135 if (!(BIT(lru) & lru_mask))
4136 continue;
Shakeel Buttdd8657b2020-06-03 15:56:24 -07004137 if (tree)
4138 nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru);
4139 else
4140 nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru);
Johannes Weiner113b7df2019-05-13 17:18:11 -07004141 }
4142 return nr;
4143}
4144
4145static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
Shakeel Buttdd8657b2020-06-03 15:56:24 -07004146 unsigned int lru_mask,
4147 bool tree)
Johannes Weiner113b7df2019-05-13 17:18:11 -07004148{
4149 unsigned long nr = 0;
4150 enum lru_list lru;
4151
4152 for_each_lru(lru) {
4153 if (!(BIT(lru) & lru_mask))
4154 continue;
Shakeel Buttdd8657b2020-06-03 15:56:24 -07004155 if (tree)
4156 nr += memcg_page_state(memcg, NR_LRU_BASE + lru);
4157 else
4158 nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru);
Johannes Weiner113b7df2019-05-13 17:18:11 -07004159 }
4160 return nr;
4161}
4162
Tejun Heo2da8ca82013-12-05 12:28:04 -05004163static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07004164{
Greg Thelen25485de2013-11-12 15:07:40 -08004165 struct numa_stat {
4166 const char *name;
4167 unsigned int lru_mask;
4168 };
4169
4170 static const struct numa_stat stats[] = {
4171 { "total", LRU_ALL },
4172 { "file", LRU_ALL_FILE },
4173 { "anon", LRU_ALL_ANON },
4174 { "unevictable", BIT(LRU_UNEVICTABLE) },
4175 };
4176 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07004177 int nid;
Chris Downaa9694b2019-03-05 15:45:52 -08004178 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Ying Han406eb0c2011-05-26 16:25:37 -07004179
Shakeel Buttfd25a9e2021-11-05 13:37:34 -07004180 mem_cgroup_flush_stats();
Johannes Weiner2d146aa2021-04-29 22:56:26 -07004181
Greg Thelen25485de2013-11-12 15:07:40 -08004182 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
Shakeel Buttdd8657b2020-06-03 15:56:24 -07004183 seq_printf(m, "%s=%lu", stat->name,
4184 mem_cgroup_nr_lru_pages(memcg, stat->lru_mask,
4185 false));
4186 for_each_node_state(nid, N_MEMORY)
4187 seq_printf(m, " N%d=%lu", nid,
4188 mem_cgroup_node_nr_lru_pages(memcg, nid,
4189 stat->lru_mask, false));
Greg Thelen25485de2013-11-12 15:07:40 -08004190 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07004191 }
Ying Han406eb0c2011-05-26 16:25:37 -07004192
Ying Han071aee12013-11-12 15:07:41 -08004193 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
Ying Han406eb0c2011-05-26 16:25:37 -07004194
Shakeel Buttdd8657b2020-06-03 15:56:24 -07004195 seq_printf(m, "hierarchical_%s=%lu", stat->name,
4196 mem_cgroup_nr_lru_pages(memcg, stat->lru_mask,
4197 true));
4198 for_each_node_state(nid, N_MEMORY)
4199 seq_printf(m, " N%d=%lu", nid,
4200 mem_cgroup_node_nr_lru_pages(memcg, nid,
4201 stat->lru_mask, true));
Ying Han071aee12013-11-12 15:07:41 -08004202 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07004203 }
Ying Han406eb0c2011-05-26 16:25:37 -07004204
Ying Han406eb0c2011-05-26 16:25:37 -07004205 return 0;
4206}
4207#endif /* CONFIG_NUMA */
4208
Johannes Weinerc8713d02019-07-11 20:55:59 -07004209static const unsigned int memcg1_stats[] = {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07004210 NR_FILE_PAGES,
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07004211 NR_ANON_MAPPED,
Johannes Weiner468c3982020-06-03 16:02:01 -07004212#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4213 NR_ANON_THPS,
4214#endif
Johannes Weinerc8713d02019-07-11 20:55:59 -07004215 NR_SHMEM,
4216 NR_FILE_MAPPED,
4217 NR_FILE_DIRTY,
4218 NR_WRITEBACK,
Yang Shie09b0b62022-08-16 11:58:01 -07004219 WORKINGSET_REFAULT_ANON,
4220 WORKINGSET_REFAULT_FILE,
Liu Shixin72a14e82023-09-15 18:58:44 +08004221#ifdef CONFIG_SWAP
Johannes Weinerc8713d02019-07-11 20:55:59 -07004222 MEMCG_SWAP,
Liu Shixin72a14e82023-09-15 18:58:44 +08004223 NR_SWAPCACHE,
4224#endif
Johannes Weinerc8713d02019-07-11 20:55:59 -07004225};
4226
4227static const char *const memcg1_stat_names[] = {
4228 "cache",
4229 "rss",
Johannes Weiner468c3982020-06-03 16:02:01 -07004230#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Johannes Weinerc8713d02019-07-11 20:55:59 -07004231 "rss_huge",
Johannes Weiner468c3982020-06-03 16:02:01 -07004232#endif
Johannes Weinerc8713d02019-07-11 20:55:59 -07004233 "shmem",
4234 "mapped_file",
4235 "dirty",
4236 "writeback",
Yang Shie09b0b62022-08-16 11:58:01 -07004237 "workingset_refault_anon",
4238 "workingset_refault_file",
Liu Shixin72a14e82023-09-15 18:58:44 +08004239#ifdef CONFIG_SWAP
Johannes Weinerc8713d02019-07-11 20:55:59 -07004240 "swap",
Liu Shixin72a14e82023-09-15 18:58:44 +08004241 "swapcached",
4242#endif
Johannes Weinerc8713d02019-07-11 20:55:59 -07004243};
4244
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07004245/* Universal VM events cgroup1 shows, original sort order */
Greg Thelen8dd53fd2018-06-07 17:07:23 -07004246static const unsigned int memcg1_events[] = {
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07004247 PGPGIN,
4248 PGPGOUT,
4249 PGFAULT,
4250 PGMAJFAULT,
4251};
4252
Yosry Ahmeddddb44f2023-04-28 13:24:06 +00004253static void memcg1_stat_format(struct mem_cgroup *memcg, struct seq_buf *s)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004254{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004255 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004256 struct mem_cgroup *mi;
4257 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004258
Johannes Weiner71cd3112017-05-03 14:55:13 -07004259 BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats));
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08004260
Shakeel Buttfd25a9e2021-11-05 13:37:34 -07004261 mem_cgroup_flush_stats();
Johannes Weiner2d146aa2021-04-29 22:56:26 -07004262
Johannes Weiner71cd3112017-05-03 14:55:13 -07004263 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
Johannes Weiner468c3982020-06-03 16:02:01 -07004264 unsigned long nr;
4265
Yosry Ahmedff841a02023-09-22 17:57:39 +00004266 nr = memcg_page_state_local_output(memcg, memcg1_stats[i]);
4267 seq_buf_printf(s, "%s %lu\n", memcg1_stat_names[i], nr);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004268 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004269
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07004270 for (i = 0; i < ARRAY_SIZE(memcg1_events); i++)
Yosry Ahmeddddb44f2023-04-28 13:24:06 +00004271 seq_buf_printf(s, "%s %lu\n", vm_event_name(memcg1_events[i]),
4272 memcg_events_local(memcg, memcg1_events[i]));
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004273
4274 for (i = 0; i < NR_LRU_LISTS; i++)
Yosry Ahmeddddb44f2023-04-28 13:24:06 +00004275 seq_buf_printf(s, "%s %lu\n", lru_list_name(i),
4276 memcg_page_state_local(memcg, NR_LRU_BASE + i) *
4277 PAGE_SIZE);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004278
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004279 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004280 memory = memsw = PAGE_COUNTER_MAX;
4281 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
Chris Down15b42562020-04-01 21:07:20 -07004282 memory = min(memory, READ_ONCE(mi->memory.max));
4283 memsw = min(memsw, READ_ONCE(mi->memsw.max));
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004284 }
Yosry Ahmeddddb44f2023-04-28 13:24:06 +00004285 seq_buf_printf(s, "hierarchical_memory_limit %llu\n",
4286 (u64)memory * PAGE_SIZE);
Liu Shixin840ea532023-09-15 18:58:45 +08004287 seq_buf_printf(s, "hierarchical_memsw_limit %llu\n",
4288 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004289
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07004290 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
zhongjiang-ali7de2e9f2020-11-01 17:07:30 -08004291 unsigned long nr;
4292
Yosry Ahmedff841a02023-09-22 17:57:39 +00004293 nr = memcg_page_state_output(memcg, memcg1_stats[i]);
Yosry Ahmeddddb44f2023-04-28 13:24:06 +00004294 seq_buf_printf(s, "total_%s %llu\n", memcg1_stat_names[i],
Yosry Ahmedff841a02023-09-22 17:57:39 +00004295 (u64)nr);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004296 }
4297
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07004298 for (i = 0; i < ARRAY_SIZE(memcg1_events); i++)
Yosry Ahmeddddb44f2023-04-28 13:24:06 +00004299 seq_buf_printf(s, "total_%s %llu\n",
4300 vm_event_name(memcg1_events[i]),
4301 (u64)memcg_events(memcg, memcg1_events[i]));
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004302
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07004303 for (i = 0; i < NR_LRU_LISTS; i++)
Yosry Ahmeddddb44f2023-04-28 13:24:06 +00004304 seq_buf_printf(s, "total_%s %llu\n", lru_list_name(i),
4305 (u64)memcg_page_state(memcg, NR_LRU_BASE + i) *
4306 PAGE_SIZE);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004307
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004308#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004309 {
Mel Gormanef8f2322016-07-28 15:46:05 -07004310 pg_data_t *pgdat;
4311 struct mem_cgroup_per_node *mz;
Johannes Weiner1431d4d2020-06-03 16:02:53 -07004312 unsigned long anon_cost = 0;
4313 unsigned long file_cost = 0;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004314
Mel Gormanef8f2322016-07-28 15:46:05 -07004315 for_each_online_pgdat(pgdat) {
Johannes Weinera3747b52021-04-29 22:56:14 -07004316 mz = memcg->nodeinfo[pgdat->node_id];
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004317
Johannes Weiner1431d4d2020-06-03 16:02:53 -07004318 anon_cost += mz->lruvec.anon_cost;
4319 file_cost += mz->lruvec.file_cost;
Mel Gormanef8f2322016-07-28 15:46:05 -07004320 }
Yosry Ahmeddddb44f2023-04-28 13:24:06 +00004321 seq_buf_printf(s, "anon_cost %lu\n", anon_cost);
4322 seq_buf_printf(s, "file_cost %lu\n", file_cost);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004323 }
4324#endif
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004325}
4326
Tejun Heo182446d2013-08-08 20:11:24 -04004327static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
4328 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004329{
Tejun Heo182446d2013-08-08 20:11:24 -04004330 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004331
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004332 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004333}
4334
Tejun Heo182446d2013-08-08 20:11:24 -04004335static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
4336 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004337{
Tejun Heo182446d2013-08-08 20:11:24 -04004338 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08004339
Baolin Wang37bc3cb2021-09-02 14:55:53 -07004340 if (val > 200)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004341 return -EINVAL;
4342
Shakeel Butta4792032021-04-29 22:56:05 -07004343 if (!mem_cgroup_is_root(memcg))
Yue Zhao82b3aa22023-03-06 23:41:36 +08004344 WRITE_ONCE(memcg->swappiness, val);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07004345 else
Yue Zhao82b3aa22023-03-06 23:41:36 +08004346 WRITE_ONCE(vm_swappiness, val);
Li Zefan068b38c2009-01-15 13:51:26 -08004347
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004348 return 0;
4349}
4350
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004351static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4352{
4353 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004354 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004355 int i;
4356
4357 rcu_read_lock();
4358 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004359 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004360 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004361 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004362
4363 if (!t)
4364 goto unlock;
4365
Johannes Weinerce00a962014-09-05 08:43:57 -04004366 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004367
4368 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07004369 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004370 * If it's not true, a threshold was crossed after last
4371 * call of __mem_cgroup_threshold().
4372 */
Phil Carmody5407a562010-05-26 14:42:42 -07004373 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004374
4375 /*
4376 * Iterate backward over array of thresholds starting from
4377 * current_threshold and check if a threshold is crossed.
4378 * If none of thresholds below usage is crossed, we read
4379 * only one element of the array here.
4380 */
4381 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4382 eventfd_signal(t->entries[i].eventfd, 1);
4383
4384 /* i = current_threshold + 1 */
4385 i++;
4386
4387 /*
4388 * Iterate forward over array of thresholds starting from
4389 * current_threshold+1 and check if a threshold is crossed.
4390 * If none of thresholds above usage is crossed, we read
4391 * only one element of the array here.
4392 */
4393 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4394 eventfd_signal(t->entries[i].eventfd, 1);
4395
4396 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004397 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004398unlock:
4399 rcu_read_unlock();
4400}
4401
4402static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4403{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004404 while (memcg) {
4405 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08004406 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004407 __mem_cgroup_threshold(memcg, true);
4408
4409 memcg = parent_mem_cgroup(memcg);
4410 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004411}
4412
4413static int compare_thresholds(const void *a, const void *b)
4414{
4415 const struct mem_cgroup_threshold *_a = a;
4416 const struct mem_cgroup_threshold *_b = b;
4417
Greg Thelen2bff24a2013-09-11 14:23:08 -07004418 if (_a->threshold > _b->threshold)
4419 return 1;
4420
4421 if (_a->threshold < _b->threshold)
4422 return -1;
4423
4424 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004425}
4426
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004427static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004428{
4429 struct mem_cgroup_eventfd_list *ev;
4430
Michal Hocko2bcf2e92014-07-30 16:08:33 -07004431 spin_lock(&memcg_oom_lock);
4432
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004433 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004434 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07004435
4436 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004437 return 0;
4438}
4439
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004440static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004441{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004442 struct mem_cgroup *iter;
4443
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004444 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004445 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004446}
4447
Tejun Heo59b6f872013-11-22 18:20:43 -05004448static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004449 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004450{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004451 struct mem_cgroup_thresholds *thresholds;
4452 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004453 unsigned long threshold;
4454 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004455 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004456
Johannes Weiner650c5e52015-02-11 15:26:03 -08004457 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004458 if (ret)
4459 return ret;
4460
4461 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004462
Johannes Weiner05b84302014-08-06 16:05:59 -07004463 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004464 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004465 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07004466 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004467 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004468 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07004469 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004470 BUG();
4471
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004472 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004473 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004474 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4475
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004476 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004477
4478 /* Allocate memory for new array of thresholds */
Gustavo A. R. Silva67b80462019-03-05 15:44:05 -08004479 new = kmalloc(struct_size(new, entries, size), GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004480 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004481 ret = -ENOMEM;
4482 goto unlock;
4483 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004484 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004485
4486 /* Copy thresholds (if any) to new array */
Gustavo A. R. Silvae90342e2020-10-13 16:52:36 -07004487 if (thresholds->primary)
4488 memcpy(new->entries, thresholds->primary->entries,
4489 flex_array_size(new, entries, size - 1));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004490
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004491 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004492 new->entries[size - 1].eventfd = eventfd;
4493 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004494
4495 /* Sort thresholds. Registering of new threshold isn't time-critical */
Gustavo A. R. Silva61e604e62020-10-13 16:52:39 -07004496 sort(new->entries, size, sizeof(*new->entries),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004497 compare_thresholds, NULL);
4498
4499 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004500 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004501 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07004502 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004503 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004504 * new->current_threshold will not be used until
4505 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004506 * it here.
4507 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004508 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07004509 } else
4510 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004511 }
4512
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004513 /* Free old spare buffer and save old primary buffer as spare */
4514 kfree(thresholds->spare);
4515 thresholds->spare = thresholds->primary;
4516
4517 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004518
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004519 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004520 synchronize_rcu();
4521
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004522unlock:
4523 mutex_unlock(&memcg->thresholds_lock);
4524
4525 return ret;
4526}
4527
Tejun Heo59b6f872013-11-22 18:20:43 -05004528static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004529 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004530{
Tejun Heo59b6f872013-11-22 18:20:43 -05004531 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05004532}
4533
Tejun Heo59b6f872013-11-22 18:20:43 -05004534static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004535 struct eventfd_ctx *eventfd, const char *args)
4536{
Tejun Heo59b6f872013-11-22 18:20:43 -05004537 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05004538}
4539
Tejun Heo59b6f872013-11-22 18:20:43 -05004540static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004541 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004542{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004543 struct mem_cgroup_thresholds *thresholds;
4544 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004545 unsigned long usage;
Chunguang Xu7d366652020-03-21 18:22:10 -07004546 int i, j, size, entries;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004547
4548 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07004549
4550 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004551 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004552 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07004553 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004554 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004555 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07004556 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004557 BUG();
4558
Anton Vorontsov371528c2012-02-24 05:14:46 +04004559 if (!thresholds->primary)
4560 goto unlock;
4561
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004562 /* Check if a threshold crossed before removing */
4563 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4564
4565 /* Calculate new number of threshold */
Chunguang Xu7d366652020-03-21 18:22:10 -07004566 size = entries = 0;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004567 for (i = 0; i < thresholds->primary->size; i++) {
4568 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004569 size++;
Chunguang Xu7d366652020-03-21 18:22:10 -07004570 else
4571 entries++;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004572 }
4573
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004574 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004575
Chunguang Xu7d366652020-03-21 18:22:10 -07004576 /* If no items related to eventfd have been cleared, nothing to do */
4577 if (!entries)
4578 goto unlock;
4579
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004580 /* Set thresholds array to NULL if we don't have thresholds */
4581 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004582 kfree(new);
4583 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004584 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004585 }
4586
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004587 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004588
4589 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004590 new->current_threshold = -1;
4591 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4592 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004593 continue;
4594
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004595 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07004596 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004597 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004598 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004599 * until rcu_assign_pointer(), so it's safe to increment
4600 * it here.
4601 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004602 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004603 }
4604 j++;
4605 }
4606
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004607swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004608 /* Swap primary and spare array */
4609 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07004610
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004611 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004612
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004613 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004614 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08004615
4616 /* If all events are unregistered, free the spare array */
4617 if (!new) {
4618 kfree(thresholds->spare);
4619 thresholds->spare = NULL;
4620 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04004621unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004622 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004623}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004624
Tejun Heo59b6f872013-11-22 18:20:43 -05004625static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004626 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004627{
Tejun Heo59b6f872013-11-22 18:20:43 -05004628 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05004629}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004630
Tejun Heo59b6f872013-11-22 18:20:43 -05004631static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004632 struct eventfd_ctx *eventfd)
4633{
Tejun Heo59b6f872013-11-22 18:20:43 -05004634 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05004635}
4636
Tejun Heo59b6f872013-11-22 18:20:43 -05004637static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004638 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004639{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004640 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004641
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004642 event = kmalloc(sizeof(*event), GFP_KERNEL);
4643 if (!event)
4644 return -ENOMEM;
4645
Michal Hocko1af8efe2011-07-26 16:08:24 -07004646 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004647
4648 event->eventfd = eventfd;
4649 list_add(&event->list, &memcg->oom_notify);
4650
4651 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07004652 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004653 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004654 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004655
4656 return 0;
4657}
4658
Tejun Heo59b6f872013-11-22 18:20:43 -05004659static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004660 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004661{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004662 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004663
Michal Hocko1af8efe2011-07-26 16:08:24 -07004664 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004665
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004666 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004667 if (ev->eventfd == eventfd) {
4668 list_del(&ev->list);
4669 kfree(ev);
4670 }
4671 }
4672
Michal Hocko1af8efe2011-07-26 16:08:24 -07004673 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004674}
4675
Tejun Heo2da8ca82013-12-05 12:28:04 -05004676static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004677{
Chris Downaa9694b2019-03-05 15:45:52 -08004678 struct mem_cgroup *memcg = mem_cgroup_from_seq(sf);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004679
Yue Zhao17c56de2023-03-06 23:41:37 +08004680 seq_printf(sf, "oom_kill_disable %d\n", READ_ONCE(memcg->oom_kill_disable));
Tejun Heoc2b42d32015-06-24 16:58:23 -07004681 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
Roman Gushchinfe6bdfc2018-06-14 15:28:05 -07004682 seq_printf(sf, "oom_kill %lu\n",
4683 atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL]));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004684 return 0;
4685}
4686
Tejun Heo182446d2013-08-08 20:11:24 -04004687static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004688 struct cftype *cft, u64 val)
4689{
Tejun Heo182446d2013-08-08 20:11:24 -04004690 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004691
4692 /* cannot set to root cgroup and only 0 and 1 are allowed */
Shakeel Butta4792032021-04-29 22:56:05 -07004693 if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004694 return -EINVAL;
4695
Yue Zhao17c56de2023-03-06 23:41:37 +08004696 WRITE_ONCE(memcg->oom_kill_disable, val);
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004697 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004698 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07004699
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004700 return 0;
4701}
4702
Tejun Heo52ebea72015-05-22 17:13:37 -04004703#ifdef CONFIG_CGROUP_WRITEBACK
4704
Tejun Heo3a8e9ac2019-08-29 15:47:19 -07004705#include <trace/events/writeback.h>
4706
Tejun Heo841710a2015-05-22 18:23:33 -04004707static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
4708{
4709 return wb_domain_init(&memcg->cgwb_domain, gfp);
4710}
4711
4712static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
4713{
4714 wb_domain_exit(&memcg->cgwb_domain);
4715}
4716
Tejun Heo2529bb32015-05-22 18:23:34 -04004717static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
4718{
4719 wb_domain_size_changed(&memcg->cgwb_domain);
4720}
4721
Tejun Heo841710a2015-05-22 18:23:33 -04004722struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
4723{
4724 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
4725
4726 if (!memcg->css.parent)
4727 return NULL;
4728
4729 return &memcg->cgwb_domain;
4730}
4731
Tejun Heoc2aa7232015-05-22 18:23:35 -04004732/**
4733 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
4734 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004735 * @pfilepages: out parameter for number of file pages
4736 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04004737 * @pdirty: out parameter for number of dirty pages
4738 * @pwriteback: out parameter for number of pages under writeback
4739 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004740 * Determine the numbers of file, headroom, dirty, and writeback pages in
4741 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
4742 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04004743 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004744 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
4745 * headroom is calculated as the lowest headroom of itself and the
4746 * ancestors. Note that this doesn't consider the actual amount of
4747 * available memory in the system. The caller should further cap
4748 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04004749 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004750void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
4751 unsigned long *pheadroom, unsigned long *pdirty,
4752 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04004753{
4754 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
4755 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04004756
Yosry Ahmed190409c2023-04-21 17:40:17 +00004757 mem_cgroup_flush_stats();
Tejun Heoc2aa7232015-05-22 18:23:35 -04004758
Johannes Weiner2d146aa2021-04-29 22:56:26 -07004759 *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY);
4760 *pwriteback = memcg_page_state(memcg, NR_WRITEBACK);
4761 *pfilepages = memcg_page_state(memcg, NR_INACTIVE_FILE) +
4762 memcg_page_state(memcg, NR_ACTIVE_FILE);
4763
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004764 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04004765 while ((parent = parent_mem_cgroup(memcg))) {
Chris Down15b42562020-04-01 21:07:20 -07004766 unsigned long ceiling = min(READ_ONCE(memcg->memory.max),
Jakub Kicinskid1663a92020-06-01 21:49:49 -07004767 READ_ONCE(memcg->memory.high));
Tejun Heoc2aa7232015-05-22 18:23:35 -04004768 unsigned long used = page_counter_read(&memcg->memory);
4769
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004770 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04004771 memcg = parent;
4772 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04004773}
4774
Tejun Heo97b27822019-08-26 09:06:56 -07004775/*
4776 * Foreign dirty flushing
4777 *
4778 * There's an inherent mismatch between memcg and writeback. The former
Ingo Molnarf0953a12021-05-06 18:06:47 -07004779 * tracks ownership per-page while the latter per-inode. This was a
Tejun Heo97b27822019-08-26 09:06:56 -07004780 * deliberate design decision because honoring per-page ownership in the
4781 * writeback path is complicated, may lead to higher CPU and IO overheads
4782 * and deemed unnecessary given that write-sharing an inode across
4783 * different cgroups isn't a common use-case.
4784 *
4785 * Combined with inode majority-writer ownership switching, this works well
4786 * enough in most cases but there are some pathological cases. For
4787 * example, let's say there are two cgroups A and B which keep writing to
4788 * different but confined parts of the same inode. B owns the inode and
4789 * A's memory is limited far below B's. A's dirty ratio can rise enough to
4790 * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid
4791 * triggering background writeback. A will be slowed down without a way to
4792 * make writeback of the dirty pages happen.
4793 *
Ingo Molnarf0953a12021-05-06 18:06:47 -07004794 * Conditions like the above can lead to a cgroup getting repeatedly and
Tejun Heo97b27822019-08-26 09:06:56 -07004795 * severely throttled after making some progress after each
Ingo Molnarf0953a12021-05-06 18:06:47 -07004796 * dirty_expire_interval while the underlying IO device is almost
Tejun Heo97b27822019-08-26 09:06:56 -07004797 * completely idle.
4798 *
4799 * Solving this problem completely requires matching the ownership tracking
4800 * granularities between memcg and writeback in either direction. However,
4801 * the more egregious behaviors can be avoided by simply remembering the
4802 * most recent foreign dirtying events and initiating remote flushes on
4803 * them when local writeback isn't enough to keep the memory clean enough.
4804 *
4805 * The following two functions implement such mechanism. When a foreign
4806 * page - a page whose memcg and writeback ownerships don't match - is
4807 * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning
4808 * bdi_writeback on the page owning memcg. When balance_dirty_pages()
4809 * decides that the memcg needs to sleep due to high dirty ratio, it calls
4810 * mem_cgroup_flush_foreign() which queues writeback on the recorded
4811 * foreign bdi_writebacks which haven't expired. Both the numbers of
4812 * recorded bdi_writebacks and concurrent in-flight foreign writebacks are
4813 * limited to MEMCG_CGWB_FRN_CNT.
4814 *
4815 * The mechanism only remembers IDs and doesn't hold any object references.
4816 * As being wrong occasionally doesn't matter, updates and accesses to the
4817 * records are lockless and racy.
4818 */
Matthew Wilcox (Oracle)9d8053f2021-05-04 11:43:01 -04004819void mem_cgroup_track_foreign_dirty_slowpath(struct folio *folio,
Tejun Heo97b27822019-08-26 09:06:56 -07004820 struct bdi_writeback *wb)
4821{
Matthew Wilcox (Oracle)9d8053f2021-05-04 11:43:01 -04004822 struct mem_cgroup *memcg = folio_memcg(folio);
Tejun Heo97b27822019-08-26 09:06:56 -07004823 struct memcg_cgwb_frn *frn;
4824 u64 now = get_jiffies_64();
4825 u64 oldest_at = now;
4826 int oldest = -1;
4827 int i;
4828
Matthew Wilcox (Oracle)9d8053f2021-05-04 11:43:01 -04004829 trace_track_foreign_dirty(folio, wb);
Tejun Heo3a8e9ac2019-08-29 15:47:19 -07004830
Tejun Heo97b27822019-08-26 09:06:56 -07004831 /*
4832 * Pick the slot to use. If there is already a slot for @wb, keep
4833 * using it. If not replace the oldest one which isn't being
4834 * written out.
4835 */
4836 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) {
4837 frn = &memcg->cgwb_frn[i];
4838 if (frn->bdi_id == wb->bdi->id &&
4839 frn->memcg_id == wb->memcg_css->id)
4840 break;
4841 if (time_before64(frn->at, oldest_at) &&
4842 atomic_read(&frn->done.cnt) == 1) {
4843 oldest = i;
4844 oldest_at = frn->at;
4845 }
4846 }
4847
4848 if (i < MEMCG_CGWB_FRN_CNT) {
4849 /*
4850 * Re-using an existing one. Update timestamp lazily to
4851 * avoid making the cacheline hot. We want them to be
4852 * reasonably up-to-date and significantly shorter than
4853 * dirty_expire_interval as that's what expires the record.
4854 * Use the shorter of 1s and dirty_expire_interval / 8.
4855 */
4856 unsigned long update_intv =
4857 min_t(unsigned long, HZ,
4858 msecs_to_jiffies(dirty_expire_interval * 10) / 8);
4859
4860 if (time_before64(frn->at, now - update_intv))
4861 frn->at = now;
4862 } else if (oldest >= 0) {
4863 /* replace the oldest free one */
4864 frn = &memcg->cgwb_frn[oldest];
4865 frn->bdi_id = wb->bdi->id;
4866 frn->memcg_id = wb->memcg_css->id;
4867 frn->at = now;
4868 }
4869}
4870
4871/* issue foreign writeback flushes for recorded foreign dirtying events */
4872void mem_cgroup_flush_foreign(struct bdi_writeback *wb)
4873{
4874 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
4875 unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10);
4876 u64 now = jiffies_64;
4877 int i;
4878
4879 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) {
4880 struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i];
4881
4882 /*
4883 * If the record is older than dirty_expire_interval,
4884 * writeback on it has already started. No need to kick it
4885 * off again. Also, don't start a new one if there's
4886 * already one in flight.
4887 */
4888 if (time_after64(frn->at, now - intv) &&
4889 atomic_read(&frn->done.cnt) == 1) {
4890 frn->at = 0;
Tejun Heo3a8e9ac2019-08-29 15:47:19 -07004891 trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id);
Shakeel Butt7490a2d2021-09-02 14:53:27 -07004892 cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id,
Tejun Heo97b27822019-08-26 09:06:56 -07004893 WB_REASON_FOREIGN_FLUSH,
4894 &frn->done);
4895 }
4896 }
4897}
4898
Tejun Heo841710a2015-05-22 18:23:33 -04004899#else /* CONFIG_CGROUP_WRITEBACK */
4900
4901static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
4902{
4903 return 0;
4904}
4905
4906static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
4907{
4908}
4909
Tejun Heo2529bb32015-05-22 18:23:34 -04004910static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
4911{
4912}
4913
Tejun Heo52ebea72015-05-22 17:13:37 -04004914#endif /* CONFIG_CGROUP_WRITEBACK */
4915
Tejun Heo79bd9812013-11-22 18:20:42 -05004916/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05004917 * DO NOT USE IN NEW FILES.
4918 *
4919 * "cgroup.event_control" implementation.
4920 *
4921 * This is way over-engineered. It tries to support fully configurable
4922 * events for each user. Such level of flexibility is completely
4923 * unnecessary especially in the light of the planned unified hierarchy.
4924 *
4925 * Please deprecate this and replace with something simpler if at all
4926 * possible.
4927 */
4928
4929/*
Tejun Heo79bd9812013-11-22 18:20:42 -05004930 * Unregister event and free resources.
4931 *
4932 * Gets called from workqueue.
4933 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05004934static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05004935{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004936 struct mem_cgroup_event *event =
4937 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05004938 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05004939
4940 remove_wait_queue(event->wqh, &event->wait);
4941
Tejun Heo59b6f872013-11-22 18:20:43 -05004942 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05004943
4944 /* Notify userspace the event is going away. */
4945 eventfd_signal(event->eventfd, 1);
4946
4947 eventfd_ctx_put(event->eventfd);
4948 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05004949 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05004950}
4951
4952/*
Linus Torvaldsa9a08842018-02-11 14:34:03 -08004953 * Gets called on EPOLLHUP on eventfd when user closes it.
Tejun Heo79bd9812013-11-22 18:20:42 -05004954 *
4955 * Called with wqh->lock held and interrupts disabled.
4956 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02004957static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode,
Tejun Heo3bc942f2013-11-22 18:20:44 -05004958 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05004959{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004960 struct mem_cgroup_event *event =
4961 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05004962 struct mem_cgroup *memcg = event->memcg;
Al Viro3ad6f932017-07-03 20:14:56 -04004963 __poll_t flags = key_to_poll(key);
Tejun Heo79bd9812013-11-22 18:20:42 -05004964
Linus Torvaldsa9a08842018-02-11 14:34:03 -08004965 if (flags & EPOLLHUP) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004966 /*
4967 * If the event has been detached at cgroup removal, we
4968 * can simply return knowing the other side will cleanup
4969 * for us.
4970 *
4971 * We can't race against event freeing since the other
4972 * side will require wqh->lock via remove_wait_queue(),
4973 * which we hold.
4974 */
Tejun Heofba94802013-11-22 18:20:43 -05004975 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004976 if (!list_empty(&event->list)) {
4977 list_del_init(&event->list);
4978 /*
4979 * We are in atomic context, but cgroup_event_remove()
4980 * may sleep, so we have to call it in workqueue.
4981 */
4982 schedule_work(&event->remove);
4983 }
Tejun Heofba94802013-11-22 18:20:43 -05004984 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004985 }
4986
4987 return 0;
4988}
4989
Tejun Heo3bc942f2013-11-22 18:20:44 -05004990static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05004991 wait_queue_head_t *wqh, poll_table *pt)
4992{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004993 struct mem_cgroup_event *event =
4994 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05004995
4996 event->wqh = wqh;
4997 add_wait_queue(wqh, &event->wait);
4998}
4999
5000/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05005001 * DO NOT USE IN NEW FILES.
5002 *
Tejun Heo79bd9812013-11-22 18:20:42 -05005003 * Parse input and register new cgroup event handler.
5004 *
5005 * Input must be in format '<event_fd> <control_fd> <args>'.
5006 * Interpretation of args is defined by control file implementation.
5007 */
Tejun Heo451af5042014-05-13 12:16:21 -04005008static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
5009 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05005010{
Tejun Heo451af5042014-05-13 12:16:21 -04005011 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05005012 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05005013 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05005014 struct cgroup_subsys_state *cfile_css;
5015 unsigned int efd, cfd;
5016 struct fd efile;
5017 struct fd cfile;
Tejun Heo4a7ba452022-12-07 16:53:15 -10005018 struct dentry *cdentry;
Tejun Heofba94802013-11-22 18:20:43 -05005019 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05005020 char *endp;
5021 int ret;
5022
Sebastian Andrzej Siewior2343e882022-03-22 14:40:38 -07005023 if (IS_ENABLED(CONFIG_PREEMPT_RT))
5024 return -EOPNOTSUPP;
5025
Tejun Heo451af5042014-05-13 12:16:21 -04005026 buf = strstrip(buf);
5027
5028 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05005029 if (*endp != ' ')
5030 return -EINVAL;
Tejun Heo451af5042014-05-13 12:16:21 -04005031 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05005032
Tejun Heo451af5042014-05-13 12:16:21 -04005033 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05005034 if ((*endp != ' ') && (*endp != '\0'))
5035 return -EINVAL;
Tejun Heo451af5042014-05-13 12:16:21 -04005036 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05005037
5038 event = kzalloc(sizeof(*event), GFP_KERNEL);
5039 if (!event)
5040 return -ENOMEM;
5041
Tejun Heo59b6f872013-11-22 18:20:43 -05005042 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05005043 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05005044 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
5045 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
5046 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05005047
5048 efile = fdget(efd);
5049 if (!efile.file) {
5050 ret = -EBADF;
5051 goto out_kfree;
5052 }
5053
5054 event->eventfd = eventfd_ctx_fileget(efile.file);
5055 if (IS_ERR(event->eventfd)) {
5056 ret = PTR_ERR(event->eventfd);
5057 goto out_put_efile;
5058 }
5059
5060 cfile = fdget(cfd);
5061 if (!cfile.file) {
5062 ret = -EBADF;
5063 goto out_put_eventfd;
5064 }
5065
5066 /* the process need read permission on control file */
5067 /* AV: shouldn't we check that it's been opened for read instead? */
Christian Brauner02f92b32021-01-21 14:19:22 +01005068 ret = file_permission(cfile.file, MAY_READ);
Tejun Heo79bd9812013-11-22 18:20:42 -05005069 if (ret < 0)
5070 goto out_put_cfile;
5071
Tejun Heo79bd9812013-11-22 18:20:42 -05005072 /*
Tejun Heo4a7ba452022-12-07 16:53:15 -10005073 * The control file must be a regular cgroup1 file. As a regular cgroup
5074 * file can't be renamed, it's safe to access its name afterwards.
5075 */
5076 cdentry = cfile.file->f_path.dentry;
5077 if (cdentry->d_sb->s_type != &cgroup_fs_type || !d_is_reg(cdentry)) {
5078 ret = -EINVAL;
5079 goto out_put_cfile;
5080 }
5081
5082 /*
Tejun Heofba94802013-11-22 18:20:43 -05005083 * Determine the event callbacks and set them in @event. This used
5084 * to be done via struct cftype but cgroup core no longer knows
5085 * about these events. The following is crude but the whole thing
5086 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05005087 *
5088 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05005089 */
Tejun Heo4a7ba452022-12-07 16:53:15 -10005090 name = cdentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05005091
5092 if (!strcmp(name, "memory.usage_in_bytes")) {
5093 event->register_event = mem_cgroup_usage_register_event;
5094 event->unregister_event = mem_cgroup_usage_unregister_event;
5095 } else if (!strcmp(name, "memory.oom_control")) {
5096 event->register_event = mem_cgroup_oom_register_event;
5097 event->unregister_event = mem_cgroup_oom_unregister_event;
5098 } else if (!strcmp(name, "memory.pressure_level")) {
5099 event->register_event = vmpressure_register_event;
5100 event->unregister_event = vmpressure_unregister_event;
5101 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05005102 event->register_event = memsw_cgroup_usage_register_event;
5103 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05005104 } else {
5105 ret = -EINVAL;
5106 goto out_put_cfile;
5107 }
5108
5109 /*
Tejun Heob5557c42013-11-22 18:20:42 -05005110 * Verify @cfile should belong to @css. Also, remaining events are
5111 * automatically removed on cgroup destruction but the removal is
5112 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05005113 */
Tejun Heo4a7ba452022-12-07 16:53:15 -10005114 cfile_css = css_tryget_online_from_dir(cdentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04005115 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05005116 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05005117 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05005118 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05005119 if (cfile_css != css) {
5120 css_put(cfile_css);
5121 goto out_put_cfile;
5122 }
Tejun Heo79bd9812013-11-22 18:20:42 -05005123
Tejun Heo451af5042014-05-13 12:16:21 -04005124 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05005125 if (ret)
5126 goto out_put_css;
5127
Christoph Hellwig9965ed172018-03-05 07:26:05 -08005128 vfs_poll(efile.file, &event->pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05005129
Shakeel Butt4ba95152021-09-02 14:56:05 -07005130 spin_lock_irq(&memcg->event_list_lock);
Tejun Heofba94802013-11-22 18:20:43 -05005131 list_add(&event->list, &memcg->event_list);
Shakeel Butt4ba95152021-09-02 14:56:05 -07005132 spin_unlock_irq(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05005133
5134 fdput(cfile);
5135 fdput(efile);
5136
Tejun Heo451af5042014-05-13 12:16:21 -04005137 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05005138
5139out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05005140 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05005141out_put_cfile:
5142 fdput(cfile);
5143out_put_eventfd:
5144 eventfd_ctx_put(event->eventfd);
5145out_put_efile:
5146 fdput(efile);
5147out_kfree:
5148 kfree(event);
5149
5150 return ret;
5151}
5152
Muchun Songc29b5b32022-01-14 14:04:01 -08005153#if defined(CONFIG_MEMCG_KMEM) && (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG))
5154static int mem_cgroup_slab_show(struct seq_file *m, void *p)
5155{
5156 /*
5157 * Deprecated.
Yang Yangdf4ae282022-06-10 02:44:52 +00005158 * Please, take a look at tools/cgroup/memcg_slabinfo.py .
Muchun Songc29b5b32022-01-14 14:04:01 -08005159 */
5160 return 0;
5161}
5162#endif
5163
Yosry Ahmeddddb44f2023-04-28 13:24:06 +00005164static int memory_stat_show(struct seq_file *m, void *v);
5165
Johannes Weiner241994ed2015-02-11 15:26:06 -08005166static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005167 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005168 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005169 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05005170 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005171 },
5172 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005173 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005174 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04005175 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005176 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07005177 },
5178 {
Balbir Singh0eea1032008-02-07 00:13:57 -08005179 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005180 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af5042014-05-13 12:16:21 -04005181 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05005182 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005183 },
5184 {
Balbir Singh296c81d2009-09-23 15:56:36 -07005185 .name = "soft_limit_in_bytes",
5186 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af5042014-05-13 12:16:21 -04005187 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05005188 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07005189 },
5190 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005191 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005192 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04005193 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005194 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005195 },
Balbir Singh8697d332008-02-07 00:13:59 -08005196 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005197 .name = "stat",
Yosry Ahmeddddb44f2023-04-28 13:24:06 +00005198 .seq_show = memory_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005199 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005200 {
5201 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04005202 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005203 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08005204 {
5205 .name = "use_hierarchy",
5206 .write_u64 = mem_cgroup_hierarchy_write,
5207 .read_u64 = mem_cgroup_hierarchy_read,
5208 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005209 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05005210 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af5042014-05-13 12:16:21 -04005211 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04005212 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05005213 },
5214 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005215 .name = "swappiness",
5216 .read_u64 = mem_cgroup_swappiness_read,
5217 .write_u64 = mem_cgroup_swappiness_write,
5218 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005219 {
5220 .name = "move_charge_at_immigrate",
5221 .read_u64 = mem_cgroup_move_charge_read,
5222 .write_u64 = mem_cgroup_move_charge_write,
5223 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005224 {
5225 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05005226 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005227 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005228 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07005229 {
5230 .name = "pressure_level",
Florian Schmidt6b0ba2a2023-04-04 10:58:59 +00005231 .seq_show = mem_cgroup_dummy_seq_show,
Anton Vorontsov70ddf632013-04-29 15:08:31 -07005232 },
Ying Han406eb0c2011-05-26 16:25:37 -07005233#ifdef CONFIG_NUMA
5234 {
5235 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05005236 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07005237 },
5238#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08005239 {
Michal Hocko45976482023-09-21 09:38:29 +02005240 .name = "kmem.limit_in_bytes",
5241 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
5242 .write = mem_cgroup_write,
5243 .read_u64 = mem_cgroup_read_u64,
5244 },
5245 {
Glauber Costa510fc4e2012-12-18 14:21:47 -08005246 .name = "kmem.usage_in_bytes",
5247 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05005248 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08005249 },
5250 {
5251 .name = "kmem.failcnt",
5252 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04005253 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005254 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08005255 },
5256 {
5257 .name = "kmem.max_usage_in_bytes",
5258 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04005259 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005260 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08005261 },
Yafang Shaoa87425a2020-04-01 21:06:30 -07005262#if defined(CONFIG_MEMCG_KMEM) && \
5263 (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG))
Glauber Costa749c5412012-12-18 14:23:01 -08005264 {
5265 .name = "kmem.slabinfo",
Muchun Songc29b5b32022-01-14 14:04:01 -08005266 .seq_show = mem_cgroup_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08005267 },
5268#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08005269 {
5270 .name = "kmem.tcp.limit_in_bytes",
5271 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
5272 .write = mem_cgroup_write,
5273 .read_u64 = mem_cgroup_read_u64,
5274 },
5275 {
5276 .name = "kmem.tcp.usage_in_bytes",
5277 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
5278 .read_u64 = mem_cgroup_read_u64,
5279 },
5280 {
5281 .name = "kmem.tcp.failcnt",
5282 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
5283 .write = mem_cgroup_reset,
5284 .read_u64 = mem_cgroup_read_u64,
5285 },
5286 {
5287 .name = "kmem.tcp.max_usage_in_bytes",
5288 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
5289 .write = mem_cgroup_reset,
5290 .read_u64 = mem_cgroup_read_u64,
5291 },
Tejun Heo6bc10342012-04-01 12:09:55 -07005292 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07005293};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005294
Johannes Weiner73f576c2016-07-20 15:44:57 -07005295/*
5296 * Private memory cgroup IDR
5297 *
5298 * Swap-out records and page cache shadow entries need to store memcg
5299 * references in constrained space, so we maintain an ID space that is
5300 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
5301 * memory-controlled cgroups to 64k.
5302 *
Ethon Paulb8f29352020-06-04 16:49:28 -07005303 * However, there usually are many references to the offline CSS after
Johannes Weiner73f576c2016-07-20 15:44:57 -07005304 * the cgroup has been destroyed, such as page cache or reclaimable
5305 * slab objects, that don't need to hang on to the ID. We want to keep
5306 * those dead CSS from occupying IDs, or we might quickly exhaust the
5307 * relatively small ID space and prevent the creation of new cgroups
5308 * even when there are much fewer than 64k cgroups - possibly none.
5309 *
5310 * Maintain a private 16-bit ID space for memcg, and allow the ID to
5311 * be freed and recycled when it's no longer needed, which is usually
5312 * when the CSS is offlined.
5313 *
5314 * The only exception to that are records of swapped out tmpfs/shmem
5315 * pages that need to be attributed to live ancestors on swapin. But
5316 * those references are manageable from userspace.
5317 */
5318
Miaohe Lin60b1e242023-07-08 10:33:04 +08005319#define MEM_CGROUP_ID_MAX ((1UL << MEM_CGROUP_ID_SHIFT) - 1)
Johannes Weiner73f576c2016-07-20 15:44:57 -07005320static DEFINE_IDR(mem_cgroup_idr);
5321
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005322static void mem_cgroup_id_remove(struct mem_cgroup *memcg)
5323{
5324 if (memcg->id.id > 0) {
5325 idr_remove(&mem_cgroup_idr, memcg->id.id);
5326 memcg->id.id = 0;
5327 }
5328}
5329
Vincenzo Frascinoc1514c02020-04-01 21:07:13 -07005330static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg,
5331 unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07005332{
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07005333 refcount_add(n, &memcg->id.ref);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005334}
5335
Vladimir Davydov615d66c2016-08-11 15:33:03 -07005336static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07005337{
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07005338 if (refcount_sub_and_test(n, &memcg->id.ref)) {
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005339 mem_cgroup_id_remove(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005340
5341 /* Memcg ID pins CSS */
5342 css_put(&memcg->css);
5343 }
5344}
5345
Vladimir Davydov615d66c2016-08-11 15:33:03 -07005346static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
5347{
5348 mem_cgroup_id_put_many(memcg, 1);
5349}
5350
Johannes Weiner73f576c2016-07-20 15:44:57 -07005351/**
5352 * mem_cgroup_from_id - look up a memcg from a memcg id
5353 * @id: the memcg id to look up
5354 *
5355 * Caller must hold rcu_read_lock().
5356 */
5357struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
5358{
5359 WARN_ON_ONCE(!rcu_read_lock_held());
5360 return idr_find(&mem_cgroup_idr, id);
5361}
5362
Roman Gushchinc15187a2022-05-31 20:22:22 -07005363#ifdef CONFIG_SHRINKER_DEBUG
5364struct mem_cgroup *mem_cgroup_get_from_ino(unsigned long ino)
5365{
5366 struct cgroup *cgrp;
5367 struct cgroup_subsys_state *css;
5368 struct mem_cgroup *memcg;
5369
5370 cgrp = cgroup_get_from_id(ino);
Michal Koutnýfa7e4392022-08-26 18:52:37 +02005371 if (IS_ERR(cgrp))
Tejun Heoc0f2df42022-08-28 17:54:15 -10005372 return ERR_CAST(cgrp);
Roman Gushchinc15187a2022-05-31 20:22:22 -07005373
5374 css = cgroup_get_e_css(cgrp, &memory_cgrp_subsys);
5375 if (css)
5376 memcg = container_of(css, struct mem_cgroup, css);
5377 else
5378 memcg = ERR_PTR(-ENOENT);
5379
5380 cgroup_put(cgrp);
5381
5382 return memcg;
5383}
5384#endif
5385
Mel Gormanef8f2322016-07-28 15:46:05 -07005386static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005387{
5388 struct mem_cgroup_per_node *pn;
Wei Yang8c9bb392022-03-22 14:47:06 -07005389
5390 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, node);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005391 if (!pn)
5392 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005393
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07005394 pn->lruvec_stats_percpu = alloc_percpu_gfp(struct lruvec_stats_percpu,
5395 GFP_KERNEL_ACCOUNT);
5396 if (!pn->lruvec_stats_percpu) {
Johannes Weiner00f3ca22017-07-06 15:40:52 -07005397 kfree(pn);
5398 return 1;
5399 }
5400
Mel Gormanef8f2322016-07-28 15:46:05 -07005401 lruvec_init(&pn->lruvec);
Mel Gormanef8f2322016-07-28 15:46:05 -07005402 pn->memcg = memcg;
5403
Johannes Weiner54f72fe2013-07-08 15:59:49 -07005404 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005405 return 0;
5406}
5407
Mel Gormanef8f2322016-07-28 15:46:05 -07005408static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005409{
Johannes Weiner00f3ca22017-07-06 15:40:52 -07005410 struct mem_cgroup_per_node *pn = memcg->nodeinfo[node];
5411
Michal Hocko4eaf4312018-04-10 16:29:52 -07005412 if (!pn)
5413 return;
5414
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07005415 free_percpu(pn->lruvec_stats_percpu);
Johannes Weiner00f3ca22017-07-06 15:40:52 -07005416 kfree(pn);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005417}
5418
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005419static void __mem_cgroup_free(struct mem_cgroup *memcg)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005420{
5421 int node;
5422
Roman Gushchin675d6c92023-10-19 15:53:43 -07005423 if (memcg->orig_objcg)
5424 obj_cgroup_put(memcg->orig_objcg);
5425
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005426 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07005427 free_mem_cgroup_per_node_info(memcg, node);
Shakeel Butt410f8e82022-09-07 04:35:35 +00005428 kfree(memcg->vmstats);
Chris Down871789d2019-05-14 15:46:57 -07005429 free_percpu(memcg->vmstats_percpu);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005430 kfree(memcg);
5431}
5432
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005433static void mem_cgroup_free(struct mem_cgroup *memcg)
5434{
Yu Zhaoec1c86b2022-09-18 02:00:02 -06005435 lru_gen_exit_memcg(memcg);
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005436 memcg_wb_domain_exit(memcg);
5437 __mem_cgroup_free(memcg);
5438}
5439
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005440static struct mem_cgroup *mem_cgroup_alloc(void)
5441{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005442 struct mem_cgroup *memcg;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005443 int node;
Tejun Heo97b27822019-08-26 09:06:56 -07005444 int __maybe_unused i;
Yafang Shao11d67612020-05-07 18:35:43 -07005445 long error = -ENOMEM;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005446
Wang Weiyang06b2c3b2022-01-14 14:05:42 -08005447 memcg = kzalloc(struct_size(memcg, nodeinfo, nr_node_ids), GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005448 if (!memcg)
Yafang Shao11d67612020-05-07 18:35:43 -07005449 return ERR_PTR(error);
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07005450
Johannes Weiner73f576c2016-07-20 15:44:57 -07005451 memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
Muchun Songbe740502022-03-22 14:41:31 -07005452 1, MEM_CGROUP_ID_MAX + 1, GFP_KERNEL);
Yafang Shao11d67612020-05-07 18:35:43 -07005453 if (memcg->id.id < 0) {
5454 error = memcg->id.id;
Johannes Weiner73f576c2016-07-20 15:44:57 -07005455 goto fail;
Yafang Shao11d67612020-05-07 18:35:43 -07005456 }
Johannes Weiner73f576c2016-07-20 15:44:57 -07005457
Shakeel Butt410f8e82022-09-07 04:35:35 +00005458 memcg->vmstats = kzalloc(sizeof(struct memcg_vmstats), GFP_KERNEL);
5459 if (!memcg->vmstats)
5460 goto fail;
5461
Roman Gushchin3e38e0a2020-08-11 18:30:25 -07005462 memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu,
5463 GFP_KERNEL_ACCOUNT);
Chris Down871789d2019-05-14 15:46:57 -07005464 if (!memcg->vmstats_percpu)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005465 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08005466
Bob Liu3ed28fa2012-01-12 17:19:04 -08005467 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07005468 if (alloc_mem_cgroup_per_node_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005469 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07005470
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005471 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
5472 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08005473
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005474 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08005475 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08005476 mutex_init(&memcg->thresholds_lock);
5477 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07005478 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05005479 INIT_LIST_HEAD(&memcg->event_list);
5480 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005481 memcg->socket_pressure = jiffies;
Kirill Tkhai84c07d12018-08-17 15:47:25 -07005482#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydov900a38f2014-12-12 16:55:10 -08005483 memcg->kmemcg_id = -1;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07005484 INIT_LIST_HEAD(&memcg->objcg_list);
Vladimir Davydov900a38f2014-12-12 16:55:10 -08005485#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04005486#ifdef CONFIG_CGROUP_WRITEBACK
5487 INIT_LIST_HEAD(&memcg->cgwb_list);
Tejun Heo97b27822019-08-26 09:06:56 -07005488 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++)
5489 memcg->cgwb_frn[i].done =
5490 __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq);
Tejun Heo52ebea72015-05-22 17:13:37 -04005491#endif
Yang Shi87eaceb2019-09-23 15:38:15 -07005492#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5493 spin_lock_init(&memcg->deferred_split_queue.split_queue_lock);
5494 INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue);
5495 memcg->deferred_split_queue.split_queue_len = 0;
5496#endif
Yu Zhaoec1c86b2022-09-18 02:00:02 -06005497 lru_gen_init_memcg(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005498 return memcg;
5499fail:
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005500 mem_cgroup_id_remove(memcg);
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005501 __mem_cgroup_free(memcg);
Yafang Shao11d67612020-05-07 18:35:43 -07005502 return ERR_PTR(error);
Glauber Costad142e3e2013-02-22 16:34:52 -08005503}
5504
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005505static struct cgroup_subsys_state * __ref
5506mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08005507{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005508 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
Roman Gushchinb87d8ce2020-10-17 16:13:40 -07005509 struct mem_cgroup *memcg, *old_memcg;
Glauber Costad142e3e2013-02-22 16:34:52 -08005510
Roman Gushchinb87d8ce2020-10-17 16:13:40 -07005511 old_memcg = set_active_memcg(parent);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005512 memcg = mem_cgroup_alloc();
Roman Gushchinb87d8ce2020-10-17 16:13:40 -07005513 set_active_memcg(old_memcg);
Yafang Shao11d67612020-05-07 18:35:43 -07005514 if (IS_ERR(memcg))
5515 return ERR_CAST(memcg);
Li Zefan4219b2d2013-09-23 16:56:29 +08005516
Jakub Kicinskid1663a92020-06-01 21:49:49 -07005517 page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX);
Yue Zhao2178e202023-03-06 23:41:38 +08005518 WRITE_ONCE(memcg->soft_limit, PAGE_COUNTER_MAX);
Johannes Weinerf4840cc2022-05-19 14:08:53 -07005519#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP)
5520 memcg->zswap_max = PAGE_COUNTER_MAX;
5521#endif
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07005522 page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005523 if (parent) {
Yue Zhao82b3aa22023-03-06 23:41:36 +08005524 WRITE_ONCE(memcg->swappiness, mem_cgroup_swappiness(parent));
Yue Zhao17c56de2023-03-06 23:41:37 +08005525 WRITE_ONCE(memcg->oom_kill_disable, READ_ONCE(parent->oom_kill_disable));
Roman Gushchinbef86202020-12-14 19:06:49 -08005526
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005527 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08005528 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005529 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08005530 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005531 } else {
Shakeel Butt8278f1c2022-09-07 04:35:37 +00005532 init_memcg_events();
Roman Gushchinbef86202020-12-14 19:06:49 -08005533 page_counter_init(&memcg->memory, NULL);
5534 page_counter_init(&memcg->swap, NULL);
5535 page_counter_init(&memcg->kmem, NULL);
5536 page_counter_init(&memcg->tcpmem, NULL);
Vladimir Davydovd6441632014-01-23 15:53:09 -08005537
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005538 root_mem_cgroup = memcg;
5539 return &memcg->css;
5540 }
5541
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005542 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08005543 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005544
Yafang Shaob6c1a8a2023-02-10 15:47:31 +00005545#if defined(CONFIG_MEMCG_KMEM)
5546 if (!cgroup_memory_nobpf)
5547 static_branch_inc(&memcg_bpf_enabled_key);
5548#endif
5549
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005550 return &memcg->css;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005551}
5552
Johannes Weiner73f576c2016-07-20 15:44:57 -07005553static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005554{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07005555 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5556
Muchun Songda0efe32022-03-22 14:41:15 -07005557 if (memcg_online_kmem(memcg))
5558 goto remove_id;
5559
Kirill Tkhai0a4465d2018-08-17 15:47:37 -07005560 /*
Yang Shie4262c42021-05-04 18:36:23 -07005561 * A memcg must be visible for expand_shrinker_info()
Kirill Tkhai0a4465d2018-08-17 15:47:37 -07005562 * by the time the maps are allocated. So, we allocate maps
5563 * here, when for_each_mem_cgroup() can't skip it.
5564 */
Muchun Songda0efe32022-03-22 14:41:15 -07005565 if (alloc_shrinker_info(memcg))
5566 goto offline_kmem;
Kirill Tkhai0a4465d2018-08-17 15:47:37 -07005567
Shakeel Buttaa48e472021-09-02 14:55:04 -07005568 if (unlikely(mem_cgroup_is_root(memcg)))
5569 queue_delayed_work(system_unbound_wq, &stats_flush_dwork,
Miaohe Lin396faf82023-06-03 15:21:16 +08005570 FLUSH_TIME);
Yu Zhaoe4dde562022-12-21 21:19:04 -07005571 lru_gen_online_memcg(memcg);
Johannes Weiner6f0df8e2023-08-23 15:54:30 -07005572
5573 /* Online state pins memcg ID, memcg ID pins CSS */
5574 refcount_set(&memcg->id.ref, 1);
5575 css_get(css);
5576
5577 /*
5578 * Ensure mem_cgroup_from_id() works once we're fully online.
5579 *
5580 * We could do this earlier and require callers to filter with
5581 * css_tryget_online(). But right now there are no users that
5582 * need earlier access, and the workingset code relies on the
5583 * cgroup tree linkage (mem_cgroup_get_nr_swap_pages()). So
5584 * publish it here at the end of onlining. This matches the
5585 * regular ID destruction during offlining.
5586 */
5587 idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
5588
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07005589 return 0;
Muchun Songda0efe32022-03-22 14:41:15 -07005590offline_kmem:
5591 memcg_offline_kmem(memcg);
5592remove_id:
5593 mem_cgroup_id_remove(memcg);
5594 return -ENOMEM;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005595}
5596
Tejun Heoeb954192013-08-08 20:11:23 -04005597static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005598{
Tejun Heoeb954192013-08-08 20:11:23 -04005599 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05005600 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05005601
5602 /*
5603 * Unregister events and notify userspace.
5604 * Notify userspace about cgroup removing only after rmdir of cgroup
5605 * directory to avoid race between userspace and kernelspace.
5606 */
Shakeel Butt4ba95152021-09-02 14:56:05 -07005607 spin_lock_irq(&memcg->event_list_lock);
Tejun Heofba94802013-11-22 18:20:43 -05005608 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05005609 list_del_init(&event->list);
5610 schedule_work(&event->remove);
5611 }
Shakeel Butt4ba95152021-09-02 14:56:05 -07005612 spin_unlock_irq(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07005613
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07005614 page_counter_set_min(&memcg->memory, 0);
Roman Gushchin23067152018-06-07 17:06:22 -07005615 page_counter_set_low(&memcg->memory, 0);
Roman Gushchin63677c742017-09-06 16:21:47 -07005616
Johannes Weiner567e9ab2016-01-20 15:02:24 -08005617 memcg_offline_kmem(memcg);
Yang Shia1780152021-05-04 18:36:42 -07005618 reparent_shrinker_deferred(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04005619 wb_memcg_offline(memcg);
Yu Zhaoe4dde562022-12-21 21:19:04 -07005620 lru_gen_offline_memcg(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005621
Roman Gushchin591edfb2018-10-26 15:03:23 -07005622 drain_all_stock(memcg);
5623
Johannes Weiner73f576c2016-07-20 15:44:57 -07005624 mem_cgroup_id_put(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005625}
5626
Vladimir Davydov6df38682015-12-29 14:54:10 -08005627static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
5628{
5629 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5630
5631 invalidate_reclaim_iterators(memcg);
Yu Zhaoe4dde562022-12-21 21:19:04 -07005632 lru_gen_release_memcg(memcg);
Vladimir Davydov6df38682015-12-29 14:54:10 -08005633}
5634
Tejun Heoeb954192013-08-08 20:11:23 -04005635static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005636{
Tejun Heoeb954192013-08-08 20:11:23 -04005637 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo97b27822019-08-26 09:06:56 -07005638 int __maybe_unused i;
Daisuke Nishimurac268e992009-01-15 13:51:13 -08005639
Tejun Heo97b27822019-08-26 09:06:56 -07005640#ifdef CONFIG_CGROUP_WRITEBACK
5641 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++)
5642 wb_wait_for_completion(&memcg->cgwb_frn[i].done);
5643#endif
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005644 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08005645 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08005646
Johannes Weiner0db15292016-01-20 15:02:50 -08005647 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08005648 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08005649
Yafang Shaob6c1a8a2023-02-10 15:47:31 +00005650#if defined(CONFIG_MEMCG_KMEM)
5651 if (!cgroup_memory_nobpf)
5652 static_branch_dec(&memcg_bpf_enabled_key);
5653#endif
5654
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005655 vmpressure_cleanup(&memcg->vmpressure);
5656 cancel_work_sync(&memcg->high_work);
5657 mem_cgroup_remove_from_trees(memcg);
Yang Shie4262c42021-05-04 18:36:23 -07005658 free_shrinker_info(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005659 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005660}
5661
Tejun Heo1ced953b2014-07-08 18:02:57 -04005662/**
5663 * mem_cgroup_css_reset - reset the states of a mem_cgroup
5664 * @css: the target css
5665 *
5666 * Reset the states of the mem_cgroup associated with @css. This is
5667 * invoked when the userland requests disabling on the default hierarchy
5668 * but the memcg is pinned through dependency. The memcg should stop
5669 * applying policies and should revert to the vanilla state as it may be
5670 * made visible again.
5671 *
5672 * The current implementation only resets the essential configurations.
5673 * This needs to be expanded to cover all the visible parts.
5674 */
5675static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
5676{
5677 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5678
Roman Gushchinbbec2e12018-06-07 17:06:18 -07005679 page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX);
5680 page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX);
Roman Gushchinbbec2e12018-06-07 17:06:18 -07005681 page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX);
5682 page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX);
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07005683 page_counter_set_min(&memcg->memory, 0);
Roman Gushchin23067152018-06-07 17:06:22 -07005684 page_counter_set_low(&memcg->memory, 0);
Jakub Kicinskid1663a92020-06-01 21:49:49 -07005685 page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX);
Yue Zhao2178e202023-03-06 23:41:38 +08005686 WRITE_ONCE(memcg->soft_limit, PAGE_COUNTER_MAX);
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07005687 page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX);
Tejun Heo2529bb32015-05-22 18:23:34 -04005688 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced953b2014-07-08 18:02:57 -04005689}
5690
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005691static void mem_cgroup_css_rstat_flush(struct cgroup_subsys_state *css, int cpu)
5692{
5693 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5694 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
5695 struct memcg_vmstats_percpu *statc;
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +00005696 long delta, delta_cpu, v;
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07005697 int i, nid;
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005698
5699 statc = per_cpu_ptr(memcg->vmstats_percpu, cpu);
5700
5701 for (i = 0; i < MEMCG_NR_STAT; i++) {
5702 /*
5703 * Collect the aggregated propagation counts of groups
5704 * below us. We're in a per-cpu loop here and this is
5705 * a global counter, so the first cycle will get them.
5706 */
Shakeel Butt410f8e82022-09-07 04:35:35 +00005707 delta = memcg->vmstats->state_pending[i];
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005708 if (delta)
Shakeel Butt410f8e82022-09-07 04:35:35 +00005709 memcg->vmstats->state_pending[i] = 0;
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005710
5711 /* Add CPU changes on this level since the last flush */
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +00005712 delta_cpu = 0;
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005713 v = READ_ONCE(statc->state[i]);
5714 if (v != statc->state_prev[i]) {
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +00005715 delta_cpu = v - statc->state_prev[i];
5716 delta += delta_cpu;
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005717 statc->state_prev[i] = v;
5718 }
5719
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005720 /* Aggregate counts on this level and propagate upwards */
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +00005721 if (delta_cpu)
5722 memcg->vmstats->state_local[i] += delta_cpu;
5723
5724 if (delta) {
5725 memcg->vmstats->state[i] += delta;
5726 if (parent)
5727 parent->vmstats->state_pending[i] += delta;
5728 }
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005729 }
5730
Shakeel Butt8278f1c2022-09-07 04:35:37 +00005731 for (i = 0; i < NR_MEMCG_EVENTS; i++) {
Shakeel Butt410f8e82022-09-07 04:35:35 +00005732 delta = memcg->vmstats->events_pending[i];
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005733 if (delta)
Shakeel Butt410f8e82022-09-07 04:35:35 +00005734 memcg->vmstats->events_pending[i] = 0;
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005735
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +00005736 delta_cpu = 0;
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005737 v = READ_ONCE(statc->events[i]);
5738 if (v != statc->events_prev[i]) {
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +00005739 delta_cpu = v - statc->events_prev[i];
5740 delta += delta_cpu;
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005741 statc->events_prev[i] = v;
5742 }
5743
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +00005744 if (delta_cpu)
5745 memcg->vmstats->events_local[i] += delta_cpu;
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005746
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +00005747 if (delta) {
5748 memcg->vmstats->events[i] += delta;
5749 if (parent)
5750 parent->vmstats->events_pending[i] += delta;
5751 }
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005752 }
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07005753
5754 for_each_node_state(nid, N_MEMORY) {
5755 struct mem_cgroup_per_node *pn = memcg->nodeinfo[nid];
5756 struct mem_cgroup_per_node *ppn = NULL;
5757 struct lruvec_stats_percpu *lstatc;
5758
5759 if (parent)
5760 ppn = parent->nodeinfo[nid];
5761
5762 lstatc = per_cpu_ptr(pn->lruvec_stats_percpu, cpu);
5763
5764 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) {
5765 delta = pn->lruvec_stats.state_pending[i];
5766 if (delta)
5767 pn->lruvec_stats.state_pending[i] = 0;
5768
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +00005769 delta_cpu = 0;
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07005770 v = READ_ONCE(lstatc->state[i]);
5771 if (v != lstatc->state_prev[i]) {
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +00005772 delta_cpu = v - lstatc->state_prev[i];
5773 delta += delta_cpu;
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07005774 lstatc->state_prev[i] = v;
5775 }
5776
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +00005777 if (delta_cpu)
5778 pn->lruvec_stats.state_local[i] += delta_cpu;
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07005779
Yosry Ahmedf82e6bf2023-07-26 15:32:23 +00005780 if (delta) {
5781 pn->lruvec_stats.state[i] += delta;
5782 if (ppn)
5783 ppn->lruvec_stats.state_pending[i] += delta;
5784 }
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07005785 }
5786 }
Johannes Weiner2d146aa2021-04-29 22:56:26 -07005787}
5788
Daisuke Nishimura02491442010-03-10 15:22:17 -08005789#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005790/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005791static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005792{
Johannes Weiner05b84302014-08-06 16:05:59 -07005793 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07005794
Mel Gormand0164ad2015-11-06 16:28:21 -08005795 /* Try a single bulk charge without reclaim first, kswapd may wake */
5796 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07005797 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005798 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005799 return ret;
5800 }
Johannes Weiner9476db92014-08-06 16:05:55 -07005801
David Rientjes36745342017-01-24 15:18:10 -08005802 /* Try charges one by one with reclaim, but do not retry */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005803 while (count--) {
David Rientjes36745342017-01-24 15:18:10 -08005804 ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005805 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005806 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005807 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07005808 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005809 }
Johannes Weiner9476db92014-08-06 16:05:55 -07005810 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005811}
5812
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005813union mc_target {
5814 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005815 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005816};
5817
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005818enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005819 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005820 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005821 MC_TARGET_SWAP,
Jérôme Glissec733a822017-09-08 16:11:54 -07005822 MC_TARGET_DEVICE,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005823};
5824
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005825static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5826 unsigned long addr, pte_t ptent)
5827{
Christoph Hellwig25b29952019-06-13 22:50:49 +02005828 struct page *page = vm_normal_page(vma, addr, ptent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005829
Miaohe Lin58f341f2023-07-17 19:36:44 +08005830 if (!page)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005831 return NULL;
5832 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005833 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005834 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005835 } else {
5836 if (!(mc.flags & MOVE_FILE))
5837 return NULL;
5838 }
Miaohe Lin58f341f2023-07-17 19:36:44 +08005839 get_page(page);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005840
5841 return page;
5842}
5843
Jérôme Glissec733a822017-09-08 16:11:54 -07005844#if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005845static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07005846 pte_t ptent, swp_entry_t *entry)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005847{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005848 struct page *page = NULL;
5849 swp_entry_t ent = pte_to_swp_entry(ptent);
5850
Ralph Campbell9a137152020-10-13 16:53:13 -07005851 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005852 return NULL;
Jérôme Glissec733a822017-09-08 16:11:54 -07005853
5854 /*
Christoph Hellwig27674ef2022-02-16 15:31:36 +11005855 * Handle device private pages that are not accessible by the CPU, but
5856 * stored as special swap entries in the page table.
Jérôme Glissec733a822017-09-08 16:11:54 -07005857 */
5858 if (is_device_private_entry(ent)) {
Alistair Poppleaf5cdaf2021-06-30 18:54:06 -07005859 page = pfn_swap_entry_to_page(ent);
Christoph Hellwig27674ef2022-02-16 15:31:36 +11005860 if (!get_page_unless_zero(page))
Jérôme Glissec733a822017-09-08 16:11:54 -07005861 return NULL;
5862 return page;
5863 }
5864
Ralph Campbell9a137152020-10-13 16:53:13 -07005865 if (non_swap_entry(ent))
5866 return NULL;
5867
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005868 /*
Matthew Wilcox (Oracle)cb691e22022-09-02 20:46:34 +01005869 * Because swap_cache_get_folio() updates some statistics counter,
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005870 * we call find_get_page() with swapper_space directly.
5871 */
Huang Yingf6ab1f72016-10-07 17:00:21 -07005872 page = find_get_page(swap_address_space(ent), swp_offset(ent));
Johannes Weiner2d1c4982020-06-03 16:02:14 -07005873 entry->val = ent.val;
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005874
5875 return page;
5876}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005877#else
5878static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07005879 pte_t ptent, swp_entry_t *entry)
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005880{
5881 return NULL;
5882}
5883#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005884
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005885static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
Peter Xu48384b02021-11-05 13:37:28 -07005886 unsigned long addr, pte_t ptent)
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005887{
Matthew Wilcox (Oracle)524984f2022-10-19 19:33:31 +01005888 unsigned long index;
5889 struct folio *folio;
5890
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005891 if (!vma->vm_file) /* anonymous vma */
5892 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005893 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005894 return NULL;
5895
Matthew Wilcox (Oracle)524984f2022-10-19 19:33:31 +01005896 /* folio is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005897 /* shmem/tmpfs may report page out on swap: account for that too. */
Matthew Wilcox (Oracle)524984f2022-10-19 19:33:31 +01005898 index = linear_page_index(vma, addr);
5899 folio = filemap_get_incore_folio(vma->vm_file->f_mapping, index);
Christoph Hellwig66dabbb2023-03-07 15:34:10 +01005900 if (IS_ERR(folio))
Matthew Wilcox (Oracle)524984f2022-10-19 19:33:31 +01005901 return NULL;
5902 return folio_file_page(folio, index);
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005903}
5904
Chen Gangb1b0dea2015-04-14 15:47:35 -07005905/**
5906 * mem_cgroup_move_account - move account of the page
5907 * @page: the page
Li RongQing25843c22016-07-26 15:26:56 -07005908 * @compound: charge the page as compound or small page
Chen Gangb1b0dea2015-04-14 15:47:35 -07005909 * @from: mem_cgroup which the page is moved from.
5910 * @to: mem_cgroup which the page is moved to. @from != @to.
5911 *
Johannes Weiner4e0cf052022-12-06 18:13:39 +01005912 * The page must be locked and not on the LRU.
Chen Gangb1b0dea2015-04-14 15:47:35 -07005913 *
5914 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
5915 * from old cgroup.
5916 */
5917static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005918 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07005919 struct mem_cgroup *from,
5920 struct mem_cgroup *to)
5921{
Matthew Wilcox (Oracle)fcce4672021-03-01 16:34:06 -05005922 struct folio *folio = page_folio(page);
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005923 struct lruvec *from_vec, *to_vec;
5924 struct pglist_data *pgdat;
Matthew Wilcox (Oracle)fcce4672021-03-01 16:34:06 -05005925 unsigned int nr_pages = compound ? folio_nr_pages(folio) : 1;
Matthew Wilcox (Oracle)8e88bd22021-06-25 09:05:47 -04005926 int nid, ret;
Chen Gangb1b0dea2015-04-14 15:47:35 -07005927
5928 VM_BUG_ON(from == to);
Johannes Weiner4e0cf052022-12-06 18:13:39 +01005929 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
Matthew Wilcox (Oracle)fcce4672021-03-01 16:34:06 -05005930 VM_BUG_ON_FOLIO(folio_test_lru(folio), folio);
Matthew Wilcox (Oracle)9c325212021-11-16 21:17:14 -05005931 VM_BUG_ON(compound && !folio_test_large(folio));
Chen Gangb1b0dea2015-04-14 15:47:35 -07005932
Chen Gangb1b0dea2015-04-14 15:47:35 -07005933 ret = -EINVAL;
Matthew Wilcox (Oracle)fcce4672021-03-01 16:34:06 -05005934 if (folio_memcg(folio) != from)
Johannes Weiner4e0cf052022-12-06 18:13:39 +01005935 goto out;
Chen Gangb1b0dea2015-04-14 15:47:35 -07005936
Matthew Wilcox (Oracle)fcce4672021-03-01 16:34:06 -05005937 pgdat = folio_pgdat(folio);
Johannes Weiner867e5e12019-11-30 17:55:34 -08005938 from_vec = mem_cgroup_lruvec(from, pgdat);
5939 to_vec = mem_cgroup_lruvec(to, pgdat);
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005940
Matthew Wilcox (Oracle)fcce4672021-03-01 16:34:06 -05005941 folio_memcg_lock(folio);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005942
Matthew Wilcox (Oracle)fcce4672021-03-01 16:34:06 -05005943 if (folio_test_anon(folio)) {
5944 if (folio_mapped(folio)) {
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07005945 __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages);
5946 __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages);
Matthew Wilcox (Oracle)61992772023-08-16 16:11:59 +01005947 if (folio_test_pmd_mappable(folio)) {
Muchun Song69473e52021-02-24 12:03:23 -08005948 __mod_lruvec_state(from_vec, NR_ANON_THPS,
5949 -nr_pages);
5950 __mod_lruvec_state(to_vec, NR_ANON_THPS,
5951 nr_pages);
Johannes Weiner468c3982020-06-03 16:02:01 -07005952 }
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07005953 }
5954 } else {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07005955 __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages);
5956 __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages);
5957
Matthew Wilcox (Oracle)fcce4672021-03-01 16:34:06 -05005958 if (folio_test_swapbacked(folio)) {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07005959 __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages);
5960 __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages);
5961 }
5962
Matthew Wilcox (Oracle)fcce4672021-03-01 16:34:06 -05005963 if (folio_mapped(folio)) {
Johannes Weiner49e50d22020-06-03 16:01:47 -07005964 __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages);
5965 __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages);
5966 }
Chen Gangb1b0dea2015-04-14 15:47:35 -07005967
Matthew Wilcox (Oracle)fcce4672021-03-01 16:34:06 -05005968 if (folio_test_dirty(folio)) {
5969 struct address_space *mapping = folio_mapping(folio);
Greg Thelenc4843a72015-05-22 17:13:16 -04005970
Christoph Hellwigf56753a2020-09-24 08:51:40 +02005971 if (mapping_can_writeback(mapping)) {
Johannes Weiner49e50d22020-06-03 16:01:47 -07005972 __mod_lruvec_state(from_vec, NR_FILE_DIRTY,
5973 -nr_pages);
5974 __mod_lruvec_state(to_vec, NR_FILE_DIRTY,
5975 nr_pages);
5976 }
Greg Thelenc4843a72015-05-22 17:13:16 -04005977 }
5978 }
5979
Hugh Dickinsc449deb2022-12-04 17:01:03 -08005980#ifdef CONFIG_SWAP
5981 if (folio_test_swapcache(folio)) {
5982 __mod_lruvec_state(from_vec, NR_SWAPCACHE, -nr_pages);
5983 __mod_lruvec_state(to_vec, NR_SWAPCACHE, nr_pages);
5984 }
5985#endif
Matthew Wilcox (Oracle)fcce4672021-03-01 16:34:06 -05005986 if (folio_test_writeback(folio)) {
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005987 __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages);
5988 __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005989 }
5990
5991 /*
Johannes Weinerabb242f2020-06-03 16:01:28 -07005992 * All state has been migrated, let's switch to the new memcg.
5993 *
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08005994 * It is safe to change page's memcg here because the page
Johannes Weinerabb242f2020-06-03 16:01:28 -07005995 * is referenced, charged, isolated, and locked: we can't race
5996 * with (un)charging, migration, LRU putback, or anything else
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08005997 * that would rely on a stable page's memory cgroup.
Johannes Weinerabb242f2020-06-03 16:01:28 -07005998 *
Kefeng Wang6c77b602023-06-14 22:36:12 +08005999 * Note that folio_memcg_lock is a memcg lock, not a page lock,
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08006000 * to save space. As soon as we switch page's memory cgroup to a
Johannes Weinerabb242f2020-06-03 16:01:28 -07006001 * new memcg that isn't locked, the above state can change
6002 * concurrently again. Make sure we're truly done with it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07006003 */
Johannes Weinerabb242f2020-06-03 16:01:28 -07006004 smp_mb();
Chen Gangb1b0dea2015-04-14 15:47:35 -07006005
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07006006 css_get(&to->css);
6007 css_put(&from->css);
6008
Matthew Wilcox (Oracle)fcce4672021-03-01 16:34:06 -05006009 folio->memcg_data = (unsigned long)to;
Yang Shi87eaceb2019-09-23 15:38:15 -07006010
Matthew Wilcox (Oracle)f70ad442021-06-28 17:26:00 -04006011 __folio_memcg_unlock(from);
Chen Gangb1b0dea2015-04-14 15:47:35 -07006012
6013 ret = 0;
Matthew Wilcox (Oracle)fcce4672021-03-01 16:34:06 -05006014 nid = folio_nid(folio);
Chen Gangb1b0dea2015-04-14 15:47:35 -07006015
6016 local_irq_disable();
Matthew Wilcox (Oracle)6e0110c2021-04-29 13:40:11 -04006017 mem_cgroup_charge_statistics(to, nr_pages);
Matthew Wilcox (Oracle)8e88bd22021-06-25 09:05:47 -04006018 memcg_check_events(to, nid);
Matthew Wilcox (Oracle)6e0110c2021-04-29 13:40:11 -04006019 mem_cgroup_charge_statistics(from, -nr_pages);
Matthew Wilcox (Oracle)8e88bd22021-06-25 09:05:47 -04006020 memcg_check_events(from, nid);
Chen Gangb1b0dea2015-04-14 15:47:35 -07006021 local_irq_enable();
Chen Gangb1b0dea2015-04-14 15:47:35 -07006022out:
6023 return ret;
6024}
6025
Li RongQing7cf78062016-05-27 14:27:46 -07006026/**
6027 * get_mctgt_type - get target type of moving charge
6028 * @vma: the vma the pte to be checked belongs
6029 * @addr: the address corresponding to the pte to be checked
6030 * @ptent: the pte to be checked
6031 * @target: the pointer the target page or swap ent will be stored(can be NULL)
6032 *
Matthew Wilcox (Oracle)853f62a2023-08-18 21:06:28 +01006033 * Context: Called with pte lock held.
6034 * Return:
6035 * * MC_TARGET_NONE - If the pte is not a target for move charge.
6036 * * MC_TARGET_PAGE - If the page corresponding to this pte is a target for
6037 * move charge. If @target is not NULL, the page is stored in target->page
6038 * with extra refcnt taken (Caller should release it).
6039 * * MC_TARGET_SWAP - If the swap entry corresponding to this pte is a
6040 * target for charge migration. If @target is not NULL, the entry is
6041 * stored in target->ent.
6042 * * MC_TARGET_DEVICE - Like MC_TARGET_PAGE but page is device memory and
6043 * thus not on the lru. For now such page is charged like a regular page
6044 * would be as it is just special memory taking the place of a regular page.
6045 * See Documentations/vm/hmm.txt and include/linux/hmm.h
Li RongQing7cf78062016-05-27 14:27:46 -07006046 */
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006047static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006048 unsigned long addr, pte_t ptent, union mc_target *target)
6049{
Daisuke Nishimura02491442010-03-10 15:22:17 -08006050 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006051 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006052 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006053
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006054 if (pte_present(ptent))
6055 page = mc_handle_present_pte(vma, addr, ptent);
Peter Xu5c041f52022-05-12 20:22:52 -07006056 else if (pte_none_mostly(ptent))
6057 /*
6058 * PTE markers should be treated as a none pte here, separated
6059 * from other swap handling below.
6060 */
6061 page = mc_handle_file_pte(vma, addr, ptent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006062 else if (is_swap_pte(ptent))
Li RongQing48406ef2016-07-26 15:22:14 -07006063 page = mc_handle_swap_pte(vma, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006064
Johannes Weiner4e0cf052022-12-06 18:13:39 +01006065 if (target && page) {
6066 if (!trylock_page(page)) {
6067 put_page(page);
6068 return ret;
6069 }
6070 /*
6071 * page_mapped() must be stable during the move. This
6072 * pte is locked, so if it's present, the page cannot
6073 * become unmapped. If it isn't, we have only partial
6074 * control over the mapped state: the page lock will
6075 * prevent new faults against pagecache and swapcache,
6076 * so an unmapped page cannot become mapped. However,
6077 * if the page is already mapped elsewhere, it can
6078 * unmap, and there is nothing we can do about it.
6079 * Alas, skip moving the page in this case.
6080 */
6081 if (!pte_present(ptent) && page_mapped(page)) {
6082 unlock_page(page);
6083 put_page(page);
6084 return ret;
6085 }
6086 }
6087
Daisuke Nishimura90254a62010-05-26 14:42:38 -07006088 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006089 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006090 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08006091 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006092 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08006093 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006094 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08006095 */
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08006096 if (page_memcg(page) == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08006097 ret = MC_TARGET_PAGE;
Alex Sierraf25cbb72022-07-15 10:05:10 -05006098 if (is_device_private_page(page) ||
6099 is_device_coherent_page(page))
Jérôme Glissec733a822017-09-08 16:11:54 -07006100 ret = MC_TARGET_DEVICE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006101 if (target)
6102 target->page = page;
6103 }
Johannes Weiner4e0cf052022-12-06 18:13:39 +01006104 if (!ret || !target) {
6105 if (target)
6106 unlock_page(page);
Daisuke Nishimura02491442010-03-10 15:22:17 -08006107 put_page(page);
Johannes Weiner4e0cf052022-12-06 18:13:39 +01006108 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08006109 }
Huang Ying3e14a572017-09-06 16:22:37 -07006110 /*
6111 * There is a swap entry and a page doesn't exist or isn't charged.
6112 * But we cannot move a tail-page in a THP.
6113 */
6114 if (ent.val && !ret && (!page || !PageTransCompound(page)) &&
Li Zefan34c00c32013-09-23 16:56:01 +08006115 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07006116 ret = MC_TARGET_SWAP;
6117 if (target)
6118 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006119 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006120 return ret;
6121}
6122
Naoya Horiguchi12724852012-03-21 16:34:28 -07006123#ifdef CONFIG_TRANSPARENT_HUGEPAGE
6124/*
Huang Yingd6810d732017-09-06 16:22:45 -07006125 * We don't consider PMD mapped swapping or file mapped pages because THP does
6126 * not support them for now.
Naoya Horiguchi12724852012-03-21 16:34:28 -07006127 * Caller should make sure that pmd_trans_huge(pmd) is true.
6128 */
6129static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6130 unsigned long addr, pmd_t pmd, union mc_target *target)
6131{
6132 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006133 enum mc_target_type ret = MC_TARGET_NONE;
6134
Zi Yan84c3fc42017-09-08 16:11:01 -07006135 if (unlikely(is_swap_pmd(pmd))) {
6136 VM_BUG_ON(thp_migration_supported() &&
6137 !is_pmd_migration_entry(pmd));
6138 return ret;
6139 }
Naoya Horiguchi12724852012-03-21 16:34:28 -07006140 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08006141 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08006142 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07006143 return ret;
Roman Gushchinbcfe06b2020-12-01 13:58:27 -08006144 if (page_memcg(page) == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006145 ret = MC_TARGET_PAGE;
6146 if (target) {
6147 get_page(page);
Johannes Weiner4e0cf052022-12-06 18:13:39 +01006148 if (!trylock_page(page)) {
6149 put_page(page);
6150 return MC_TARGET_NONE;
6151 }
Naoya Horiguchi12724852012-03-21 16:34:28 -07006152 target->page = page;
6153 }
6154 }
6155 return ret;
6156}
6157#else
6158static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
6159 unsigned long addr, pmd_t pmd, union mc_target *target)
6160{
6161 return MC_TARGET_NONE;
6162}
6163#endif
6164
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006165static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
6166 unsigned long addr, unsigned long end,
6167 struct mm_walk *walk)
6168{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08006169 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006170 pte_t *pte;
6171 spinlock_t *ptl;
6172
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08006173 ptl = pmd_trans_huge_lock(pmd, vma);
6174 if (ptl) {
Jérôme Glissec733a822017-09-08 16:11:54 -07006175 /*
6176 * Note their can not be MC_TARGET_DEVICE for now as we do not
Christoph Hellwig25b29952019-06-13 22:50:49 +02006177 * support transparent huge page with MEMORY_DEVICE_PRIVATE but
6178 * this might change.
Jérôme Glissec733a822017-09-08 16:11:54 -07006179 */
Naoya Horiguchi12724852012-03-21 16:34:28 -07006180 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
6181 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006182 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006183 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006184 }
Dave Hansen03319322011-03-22 16:32:56 -07006185
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006186 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins04dee9e2023-06-08 18:29:22 -07006187 if (!pte)
6188 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006189 for (; addr != end; pte++, addr += PAGE_SIZE)
Ryan Robertsc33c7942023-06-12 16:15:45 +01006190 if (get_mctgt_type(vma, addr, ptep_get(pte), NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006191 mc.precharge++; /* increment precharge temporarily */
6192 pte_unmap_unlock(pte - 1, ptl);
6193 cond_resched();
6194
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006195 return 0;
6196}
6197
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02006198static const struct mm_walk_ops precharge_walk_ops = {
6199 .pmd_entry = mem_cgroup_count_precharge_pte_range,
Suren Baghdasaryan49b06382023-08-04 08:27:19 -07006200 .walk_lock = PGWALK_RDLOCK,
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02006201};
6202
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006203static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
6204{
6205 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006206
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07006207 mmap_read_lock(mm);
Liam R. Howlettba0aff82022-09-06 19:49:01 +00006208 walk_page_range(mm, 0, ULONG_MAX, &precharge_walk_ops, NULL);
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07006209 mmap_read_unlock(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006210
6211 precharge = mc.precharge;
6212 mc.precharge = 0;
6213
6214 return precharge;
6215}
6216
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006217static int mem_cgroup_precharge_mc(struct mm_struct *mm)
6218{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006219 unsigned long precharge = mem_cgroup_count_precharge(mm);
6220
6221 VM_BUG_ON(mc.moving_task);
6222 mc.moving_task = current;
6223 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006224}
6225
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006226/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
6227static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006228{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006229 struct mem_cgroup *from = mc.from;
6230 struct mem_cgroup *to = mc.to;
6231
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006232 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006233 if (mc.precharge) {
Nhat Pham4b569382023-10-06 11:46:26 -07006234 mem_cgroup_cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006235 mc.precharge = 0;
6236 }
6237 /*
6238 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
6239 * we must uncharge here.
6240 */
6241 if (mc.moved_charge) {
Nhat Pham4b569382023-10-06 11:46:26 -07006242 mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006243 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006244 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006245 /* we must fixup refcnts and charges */
6246 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006247 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04006248 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08006249 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006250
Vladimir Davydov615d66c2016-08-11 15:33:03 -07006251 mem_cgroup_id_put_many(mc.from, mc.moved_swap);
6252
Johannes Weiner05b84302014-08-06 16:05:59 -07006253 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08006254 * we charged both to->memory and to->memsw, so we
6255 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07006256 */
Johannes Weinerce00a962014-09-05 08:43:57 -04006257 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08006258 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006259
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006260 mc.moved_swap = 0;
6261 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006262 memcg_oom_recover(from);
6263 memcg_oom_recover(to);
6264 wake_up_all(&mc.waitq);
6265}
6266
6267static void mem_cgroup_clear_mc(void)
6268{
Tejun Heo264a0ae2016-04-21 19:09:02 -04006269 struct mm_struct *mm = mc.mm;
6270
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006271 /*
6272 * we must clear moving_task before waking up waiters at the end of
6273 * task migration.
6274 */
6275 mc.moving_task = NULL;
6276 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006277 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006278 mc.from = NULL;
6279 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04006280 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07006281 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04006282
6283 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006284}
6285
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006286static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006287{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006288 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07006289 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07006290 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04006291 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07006292 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08006293 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07006294 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006295
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006296 /* charge immigration isn't supported on the default hierarchy */
6297 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07006298 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006299
Tejun Heo4530edd2015-09-11 15:00:19 -04006300 /*
6301 * Multi-process migrations only happen on the default hierarchy
6302 * where charge immigration is not used. Perform charge
6303 * immigration if @tset contains a leader and whine if there are
6304 * multiple.
6305 */
6306 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006307 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04006308 WARN_ON_ONCE(p);
6309 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006310 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04006311 }
6312 if (!p)
6313 return 0;
6314
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006315 /*
Ingo Molnarf0953a12021-05-06 18:06:47 -07006316 * We are now committed to this value whatever it is. Changes in this
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006317 * tunable will only affect upcoming migrations, not the current one.
6318 * So we need to save it, and keep it going.
6319 */
6320 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
6321 if (!move_flags)
6322 return 0;
6323
Tejun Heo9f2115f2015-09-08 15:01:10 -07006324 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006325
Tejun Heo9f2115f2015-09-08 15:01:10 -07006326 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08006327
Tejun Heo9f2115f2015-09-08 15:01:10 -07006328 mm = get_task_mm(p);
6329 if (!mm)
6330 return 0;
6331 /* We move charges only when we move a owner of the mm */
6332 if (mm->owner == p) {
6333 VM_BUG_ON(mc.from);
6334 VM_BUG_ON(mc.to);
6335 VM_BUG_ON(mc.precharge);
6336 VM_BUG_ON(mc.moved_charge);
6337 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006338
Tejun Heo9f2115f2015-09-08 15:01:10 -07006339 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04006340 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07006341 mc.from = from;
6342 mc.to = memcg;
6343 mc.flags = move_flags;
6344 spin_unlock(&mc.lock);
6345 /* We set mc.moving_task later */
6346
6347 ret = mem_cgroup_precharge_mc(mm);
6348 if (ret)
6349 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04006350 } else {
6351 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006352 }
6353 return ret;
6354}
6355
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006356static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006357{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08006358 if (mc.to)
6359 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006360}
6361
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006362static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
6363 unsigned long addr, unsigned long end,
6364 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006365{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006366 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08006367 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006368 pte_t *pte;
6369 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006370 enum mc_target_type target_type;
6371 union mc_target target;
6372 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006373
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08006374 ptl = pmd_trans_huge_lock(pmd, vma);
6375 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07006376 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006377 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07006378 return 0;
6379 }
6380 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
6381 if (target_type == MC_TARGET_PAGE) {
6382 page = target.page;
Baolin Wangf7f9c002023-02-15 18:39:35 +08006383 if (isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08006384 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08006385 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006386 mc.precharge -= HPAGE_PMD_NR;
6387 mc.moved_charge += HPAGE_PMD_NR;
6388 }
6389 putback_lru_page(page);
6390 }
Johannes Weiner4e0cf052022-12-06 18:13:39 +01006391 unlock_page(page);
Naoya Horiguchi12724852012-03-21 16:34:28 -07006392 put_page(page);
Jérôme Glissec733a822017-09-08 16:11:54 -07006393 } else if (target_type == MC_TARGET_DEVICE) {
6394 page = target.page;
6395 if (!mem_cgroup_move_account(page, true,
6396 mc.from, mc.to)) {
6397 mc.precharge -= HPAGE_PMD_NR;
6398 mc.moved_charge += HPAGE_PMD_NR;
6399 }
Johannes Weiner4e0cf052022-12-06 18:13:39 +01006400 unlock_page(page);
Jérôme Glissec733a822017-09-08 16:11:54 -07006401 put_page(page);
Naoya Horiguchi12724852012-03-21 16:34:28 -07006402 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006403 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006404 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006405 }
6406
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006407retry:
6408 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins04dee9e2023-06-08 18:29:22 -07006409 if (!pte)
6410 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006411 for (; addr != end; addr += PAGE_SIZE) {
Ryan Robertsc33c7942023-06-12 16:15:45 +01006412 pte_t ptent = ptep_get(pte++);
Jérôme Glissec733a822017-09-08 16:11:54 -07006413 bool device = false;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006414 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006415
6416 if (!mc.precharge)
6417 break;
6418
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006419 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Jérôme Glissec733a822017-09-08 16:11:54 -07006420 case MC_TARGET_DEVICE:
6421 device = true;
Joe Perchese4a9bc52020-04-06 20:08:39 -07006422 fallthrough;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006423 case MC_TARGET_PAGE:
6424 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08006425 /*
6426 * We can have a part of the split pmd here. Moving it
6427 * can be done but it would be too convoluted so simply
6428 * ignore such a partial THP and keep it in original
6429 * memcg. There should be somebody mapping the head.
6430 */
6431 if (PageTransCompound(page))
6432 goto put;
Baolin Wangf7f9c002023-02-15 18:39:35 +08006433 if (!device && !isolate_lru_page(page))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006434 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08006435 if (!mem_cgroup_move_account(page, false,
6436 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006437 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006438 /* we uncharge from mc.from later. */
6439 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006440 }
Jérôme Glissec733a822017-09-08 16:11:54 -07006441 if (!device)
6442 putback_lru_page(page);
Johannes Weiner4e0cf052022-12-06 18:13:39 +01006443put: /* get_mctgt_type() gets & locks the page */
6444 unlock_page(page);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006445 put_page(page);
6446 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006447 case MC_TARGET_SWAP:
6448 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07006449 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08006450 mc.precharge--;
Hugh Dickins8d22a932020-07-23 21:15:24 -07006451 mem_cgroup_id_get_many(mc.to, 1);
6452 /* we fixup other refcnts and charges later. */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006453 mc.moved_swap++;
6454 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08006455 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006456 default:
6457 break;
6458 }
6459 }
6460 pte_unmap_unlock(pte - 1, ptl);
6461 cond_resched();
6462
6463 if (addr != end) {
6464 /*
6465 * We have consumed all precharges we got in can_attach().
6466 * We try charge one by one, but don't do any additional
6467 * charges to mc.to if we have failed in charge once in attach()
6468 * phase.
6469 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006470 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006471 if (!ret)
6472 goto retry;
6473 }
6474
6475 return ret;
6476}
6477
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02006478static const struct mm_walk_ops charge_walk_ops = {
6479 .pmd_entry = mem_cgroup_move_charge_pte_range,
Suren Baghdasaryan49b06382023-08-04 08:27:19 -07006480 .walk_lock = PGWALK_RDLOCK,
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02006481};
6482
Tejun Heo264a0ae2016-04-21 19:09:02 -04006483static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006484{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006485 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08006486 /*
Kefeng Wang6c77b602023-06-14 22:36:12 +08006487 * Signal folio_memcg_lock() to take the memcg's move_lock
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07006488 * while we're moving its pages to another memcg. Then wait
6489 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08006490 */
6491 atomic_inc(&mc.from->moving_account);
6492 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006493retry:
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07006494 if (unlikely(!mmap_read_trylock(mc.mm))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006495 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07006496 * Someone who are holding the mmap_lock might be waiting in
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006497 * waitq. So we cancel all extra charges, wake up all waiters,
6498 * and retry. Because we cancel precharges, we might not be able
6499 * to move enough charges, but moving charge is a best-effort
6500 * feature anyway, so it wouldn't be a big problem.
6501 */
6502 __mem_cgroup_clear_mc();
6503 cond_resched();
6504 goto retry;
6505 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08006506 /*
6507 * When we have consumed all precharges and failed in doing
6508 * additional charge, the page walk just aborts.
6509 */
Liam R. Howlettba0aff82022-09-06 19:49:01 +00006510 walk_page_range(mc.mm, 0, ULONG_MAX, &charge_walk_ops, NULL);
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07006511 mmap_read_unlock(mc.mm);
Johannes Weiner312722c2014-12-10 15:44:25 -08006512 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006513}
6514
Tejun Heo264a0ae2016-04-21 19:09:02 -04006515static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08006516{
Tejun Heo264a0ae2016-04-21 19:09:02 -04006517 if (mc.to) {
6518 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006519 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04006520 }
Balbir Singh67e465a2008-02-07 00:13:54 -08006521}
Roman Gushchin1aacbd32023-10-19 15:53:42 -07006522
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006523#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006524static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006525{
6526 return 0;
6527}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006528static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006529{
6530}
Tejun Heo264a0ae2016-04-21 19:09:02 -04006531static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006532{
6533}
6534#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08006535
Roman Gushchin1aacbd32023-10-19 15:53:42 -07006536#ifdef CONFIG_MEMCG_KMEM
6537static void mem_cgroup_fork(struct task_struct *task)
6538{
6539 /*
6540 * Set the update flag to cause task->objcg to be initialized lazily
6541 * on the first allocation. It can be done without any synchronization
6542 * because it's always performed on the current task, so does
6543 * current_objcg_update().
6544 */
6545 task->objcg = (struct obj_cgroup *)CURRENT_OBJCG_UPDATE_FLAG;
6546}
6547
6548static void mem_cgroup_exit(struct task_struct *task)
6549{
6550 struct obj_cgroup *objcg = task->objcg;
6551
6552 objcg = (struct obj_cgroup *)
6553 ((unsigned long)objcg & ~CURRENT_OBJCG_UPDATE_FLAG);
6554 if (objcg)
6555 obj_cgroup_put(objcg);
6556
6557 /*
6558 * Some kernel allocations can happen after this point,
6559 * but let's ignore them. It can be done without any synchronization
6560 * because it's always performed on the current task, so does
6561 * current_objcg_update().
6562 */
6563 task->objcg = NULL;
6564}
6565#endif
6566
Yu Zhaobd74fda2022-09-18 02:00:05 -06006567#ifdef CONFIG_LRU_GEN
Roman Gushchin1aacbd32023-10-19 15:53:42 -07006568static void mem_cgroup_lru_gen_attach(struct cgroup_taskset *tset)
Yu Zhaobd74fda2022-09-18 02:00:05 -06006569{
6570 struct task_struct *task;
6571 struct cgroup_subsys_state *css;
6572
6573 /* find the first leader if there is any */
6574 cgroup_taskset_for_each_leader(task, css, tset)
6575 break;
6576
6577 if (!task)
6578 return;
6579
6580 task_lock(task);
6581 if (task->mm && READ_ONCE(task->mm->owner) == task)
6582 lru_gen_migrate_mm(task->mm);
6583 task_unlock(task);
6584}
6585#else
Roman Gushchin1aacbd32023-10-19 15:53:42 -07006586static void mem_cgroup_lru_gen_attach(struct cgroup_taskset *tset) {}
6587#endif /* CONFIG_LRU_GEN */
6588
6589#ifdef CONFIG_MEMCG_KMEM
6590static void mem_cgroup_kmem_attach(struct cgroup_taskset *tset)
6591{
6592 struct task_struct *task;
6593 struct cgroup_subsys_state *css;
6594
6595 cgroup_taskset_for_each(task, css, tset) {
6596 /* atomically set the update bit */
6597 set_bit(CURRENT_OBJCG_UPDATE_BIT, (unsigned long *)&task->objcg);
6598 }
6599}
6600#else
6601static void mem_cgroup_kmem_attach(struct cgroup_taskset *tset) {}
6602#endif /* CONFIG_MEMCG_KMEM */
6603
6604#if defined(CONFIG_LRU_GEN) || defined(CONFIG_MEMCG_KMEM)
Yu Zhaobd74fda2022-09-18 02:00:05 -06006605static void mem_cgroup_attach(struct cgroup_taskset *tset)
6606{
Roman Gushchin1aacbd32023-10-19 15:53:42 -07006607 mem_cgroup_lru_gen_attach(tset);
6608 mem_cgroup_kmem_attach(tset);
Yu Zhaobd74fda2022-09-18 02:00:05 -06006609}
Roman Gushchin1aacbd32023-10-19 15:53:42 -07006610#endif
Yu Zhaobd74fda2022-09-18 02:00:05 -06006611
Chris Down677dc972019-03-05 15:45:55 -08006612static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value)
6613{
6614 if (value == PAGE_COUNTER_MAX)
6615 seq_puts(m, "max\n");
6616 else
6617 seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE);
6618
6619 return 0;
6620}
6621
Johannes Weiner241994ed2015-02-11 15:26:06 -08006622static u64 memory_current_read(struct cgroup_subsys_state *css,
6623 struct cftype *cft)
6624{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08006625 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6626
6627 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006628}
6629
Ganesan Rajagopal8e20d4b2022-05-13 16:48:57 -07006630static u64 memory_peak_read(struct cgroup_subsys_state *css,
6631 struct cftype *cft)
6632{
6633 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6634
6635 return (u64)memcg->memory.watermark * PAGE_SIZE;
6636}
6637
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006638static int memory_min_show(struct seq_file *m, void *v)
6639{
Chris Down677dc972019-03-05 15:45:55 -08006640 return seq_puts_memcg_tunable(m,
6641 READ_ONCE(mem_cgroup_from_seq(m)->memory.min));
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006642}
6643
6644static ssize_t memory_min_write(struct kernfs_open_file *of,
6645 char *buf, size_t nbytes, loff_t off)
6646{
6647 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6648 unsigned long min;
6649 int err;
6650
6651 buf = strstrip(buf);
6652 err = page_counter_memparse(buf, "max", &min);
6653 if (err)
6654 return err;
6655
6656 page_counter_set_min(&memcg->memory, min);
6657
6658 return nbytes;
6659}
6660
Johannes Weiner241994ed2015-02-11 15:26:06 -08006661static int memory_low_show(struct seq_file *m, void *v)
6662{
Chris Down677dc972019-03-05 15:45:55 -08006663 return seq_puts_memcg_tunable(m,
6664 READ_ONCE(mem_cgroup_from_seq(m)->memory.low));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006665}
6666
6667static ssize_t memory_low_write(struct kernfs_open_file *of,
6668 char *buf, size_t nbytes, loff_t off)
6669{
6670 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6671 unsigned long low;
6672 int err;
6673
6674 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08006675 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006676 if (err)
6677 return err;
6678
Roman Gushchin23067152018-06-07 17:06:22 -07006679 page_counter_set_low(&memcg->memory, low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006680
6681 return nbytes;
6682}
6683
6684static int memory_high_show(struct seq_file *m, void *v)
6685{
Jakub Kicinskid1663a92020-06-01 21:49:49 -07006686 return seq_puts_memcg_tunable(m,
6687 READ_ONCE(mem_cgroup_from_seq(m)->memory.high));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006688}
6689
6690static ssize_t memory_high_write(struct kernfs_open_file *of,
6691 char *buf, size_t nbytes, loff_t off)
6692{
6693 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Chris Downd977aa92020-08-06 23:21:58 -07006694 unsigned int nr_retries = MAX_RECLAIM_RETRIES;
Johannes Weiner8c8c3832019-11-30 17:50:09 -08006695 bool drained = false;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006696 unsigned long high;
6697 int err;
6698
6699 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08006700 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006701 if (err)
6702 return err;
6703
Johannes Weinere82553c2021-02-09 13:42:28 -08006704 page_counter_set_high(&memcg->memory, high);
6705
Johannes Weiner8c8c3832019-11-30 17:50:09 -08006706 for (;;) {
6707 unsigned long nr_pages = page_counter_read(&memcg->memory);
6708 unsigned long reclaimed;
Johannes Weiner588083b2016-03-17 14:20:25 -07006709
Johannes Weiner8c8c3832019-11-30 17:50:09 -08006710 if (nr_pages <= high)
6711 break;
6712
6713 if (signal_pending(current))
6714 break;
6715
6716 if (!drained) {
6717 drain_all_stock(memcg);
6718 drained = true;
6719 continue;
6720 }
6721
6722 reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
Michal Hocko55ab8342022-12-16 10:46:33 +01006723 GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP);
Johannes Weiner8c8c3832019-11-30 17:50:09 -08006724
6725 if (!reclaimed && !nr_retries--)
6726 break;
6727 }
6728
Johannes Weiner19ce33a2020-08-06 23:22:12 -07006729 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006730 return nbytes;
6731}
6732
6733static int memory_max_show(struct seq_file *m, void *v)
6734{
Chris Down677dc972019-03-05 15:45:55 -08006735 return seq_puts_memcg_tunable(m,
6736 READ_ONCE(mem_cgroup_from_seq(m)->memory.max));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006737}
6738
6739static ssize_t memory_max_write(struct kernfs_open_file *of,
6740 char *buf, size_t nbytes, loff_t off)
6741{
6742 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Chris Downd977aa92020-08-06 23:21:58 -07006743 unsigned int nr_reclaims = MAX_RECLAIM_RETRIES;
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006744 bool drained = false;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006745 unsigned long max;
6746 int err;
6747
6748 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08006749 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006750 if (err)
6751 return err;
6752
Roman Gushchinbbec2e12018-06-07 17:06:18 -07006753 xchg(&memcg->memory.max, max);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006754
6755 for (;;) {
6756 unsigned long nr_pages = page_counter_read(&memcg->memory);
6757
6758 if (nr_pages <= max)
6759 break;
6760
Johannes Weiner7249c9f2019-11-30 17:50:06 -08006761 if (signal_pending(current))
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006762 break;
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006763
6764 if (!drained) {
6765 drain_all_stock(memcg);
6766 drained = true;
6767 continue;
6768 }
6769
6770 if (nr_reclaims) {
6771 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
Michal Hocko55ab8342022-12-16 10:46:33 +01006772 GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP))
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006773 nr_reclaims--;
6774 continue;
6775 }
6776
Johannes Weinere27be242018-04-10 16:29:45 -07006777 memcg_memory_event(memcg, MEMCG_OOM);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006778 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
6779 break;
6780 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08006781
Tejun Heo2529bb32015-05-22 18:23:34 -04006782 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006783 return nbytes;
6784}
6785
Shakeel Butt1e577f92019-07-11 20:55:55 -07006786static void __memory_events_show(struct seq_file *m, atomic_long_t *events)
6787{
6788 seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW]));
6789 seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH]));
6790 seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX]));
6791 seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM]));
6792 seq_printf(m, "oom_kill %lu\n",
6793 atomic_long_read(&events[MEMCG_OOM_KILL]));
Dan Schatzbergb6bf9ab2022-01-14 14:05:35 -08006794 seq_printf(m, "oom_group_kill %lu\n",
6795 atomic_long_read(&events[MEMCG_OOM_GROUP_KILL]));
Shakeel Butt1e577f92019-07-11 20:55:55 -07006796}
6797
Johannes Weiner241994ed2015-02-11 15:26:06 -08006798static int memory_events_show(struct seq_file *m, void *v)
6799{
Chris Downaa9694b2019-03-05 15:45:52 -08006800 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006801
Shakeel Butt1e577f92019-07-11 20:55:55 -07006802 __memory_events_show(m, memcg->memory_events);
6803 return 0;
6804}
Johannes Weiner241994ed2015-02-11 15:26:06 -08006805
Shakeel Butt1e577f92019-07-11 20:55:55 -07006806static int memory_events_local_show(struct seq_file *m, void *v)
6807{
6808 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
6809
6810 __memory_events_show(m, memcg->memory_events_local);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006811 return 0;
6812}
6813
Johannes Weiner587d9f72016-01-20 15:03:19 -08006814static int memory_stat_show(struct seq_file *m, void *v)
6815{
Chris Downaa9694b2019-03-05 15:45:52 -08006816 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Tetsuo Handa68aaee12022-07-22 19:45:39 +09006817 char *buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
Yosry Ahmed5b423602023-04-28 13:24:05 +00006818 struct seq_buf s;
Johannes Weiner587d9f72016-01-20 15:03:19 -08006819
Johannes Weinerc8713d02019-07-11 20:55:59 -07006820 if (!buf)
6821 return -ENOMEM;
Yosry Ahmed5b423602023-04-28 13:24:05 +00006822 seq_buf_init(&s, buf, PAGE_SIZE);
6823 memory_stat_format(memcg, &s);
Johannes Weinerc8713d02019-07-11 20:55:59 -07006824 seq_puts(m, buf);
6825 kfree(buf);
Johannes Weiner587d9f72016-01-20 15:03:19 -08006826 return 0;
6827}
6828
Muchun Song5f9a4f42020-10-13 16:52:59 -07006829#ifdef CONFIG_NUMA
Muchun Songfff66b72021-02-24 12:03:43 -08006830static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec,
6831 int item)
6832{
Yosry Ahmedff841a02023-09-22 17:57:39 +00006833 return lruvec_page_state(lruvec, item) *
6834 memcg_page_state_output_unit(item);
Muchun Songfff66b72021-02-24 12:03:43 -08006835}
6836
Muchun Song5f9a4f42020-10-13 16:52:59 -07006837static int memory_numa_stat_show(struct seq_file *m, void *v)
6838{
6839 int i;
6840 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
6841
Shakeel Buttfd25a9e2021-11-05 13:37:34 -07006842 mem_cgroup_flush_stats();
Shakeel Butt7e1c0d62021-09-02 14:55:00 -07006843
Muchun Song5f9a4f42020-10-13 16:52:59 -07006844 for (i = 0; i < ARRAY_SIZE(memory_stats); i++) {
6845 int nid;
6846
6847 if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS)
6848 continue;
6849
6850 seq_printf(m, "%s", memory_stats[i].name);
6851 for_each_node_state(nid, N_MEMORY) {
6852 u64 size;
6853 struct lruvec *lruvec;
6854
6855 lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid));
Muchun Songfff66b72021-02-24 12:03:43 -08006856 size = lruvec_page_state_output(lruvec,
6857 memory_stats[i].idx);
Muchun Song5f9a4f42020-10-13 16:52:59 -07006858 seq_printf(m, " N%d=%llu", nid, size);
6859 }
6860 seq_putc(m, '\n');
6861 }
6862
6863 return 0;
6864}
6865#endif
6866
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006867static int memory_oom_group_show(struct seq_file *m, void *v)
6868{
Chris Downaa9694b2019-03-05 15:45:52 -08006869 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006870
Yue Zhaoeaf7b662023-03-06 23:41:35 +08006871 seq_printf(m, "%d\n", READ_ONCE(memcg->oom_group));
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006872
6873 return 0;
6874}
6875
6876static ssize_t memory_oom_group_write(struct kernfs_open_file *of,
6877 char *buf, size_t nbytes, loff_t off)
6878{
6879 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6880 int ret, oom_group;
6881
6882 buf = strstrip(buf);
6883 if (!buf)
6884 return -EINVAL;
6885
6886 ret = kstrtoint(buf, 0, &oom_group);
6887 if (ret)
6888 return ret;
6889
6890 if (oom_group != 0 && oom_group != 1)
6891 return -EINVAL;
6892
Yue Zhaoeaf7b662023-03-06 23:41:35 +08006893 WRITE_ONCE(memcg->oom_group, oom_group);
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006894
6895 return nbytes;
6896}
6897
Shakeel Butt94968382022-04-29 14:36:59 -07006898static ssize_t memory_reclaim(struct kernfs_open_file *of, char *buf,
6899 size_t nbytes, loff_t off)
6900{
6901 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6902 unsigned int nr_retries = MAX_RECLAIM_RETRIES;
6903 unsigned long nr_to_reclaim, nr_reclaimed = 0;
Michal Hocko55ab8342022-12-16 10:46:33 +01006904 unsigned int reclaim_options;
6905 int err;
Shakeel Butt94968382022-04-29 14:36:59 -07006906
6907 buf = strstrip(buf);
Michal Hocko55ab8342022-12-16 10:46:33 +01006908 err = page_counter_memparse(buf, "", &nr_to_reclaim);
6909 if (err)
6910 return err;
Shakeel Butt94968382022-04-29 14:36:59 -07006911
Michal Hocko55ab8342022-12-16 10:46:33 +01006912 reclaim_options = MEMCG_RECLAIM_MAY_SWAP | MEMCG_RECLAIM_PROACTIVE;
Shakeel Butt94968382022-04-29 14:36:59 -07006913 while (nr_reclaimed < nr_to_reclaim) {
6914 unsigned long reclaimed;
6915
6916 if (signal_pending(current))
6917 return -EINTR;
6918
6919 /*
6920 * This is the final attempt, drain percpu lru caches in the
6921 * hope of introducing more evictable pages for
6922 * try_to_free_mem_cgroup_pages().
6923 */
6924 if (!nr_retries)
6925 lru_add_drain_all();
6926
6927 reclaimed = try_to_free_mem_cgroup_pages(memcg,
Efly Young03885362023-07-21 09:41:16 +08006928 min(nr_to_reclaim - nr_reclaimed, SWAP_CLUSTER_MAX),
6929 GFP_KERNEL, reclaim_options);
Shakeel Butt94968382022-04-29 14:36:59 -07006930
6931 if (!reclaimed && !nr_retries--)
6932 return -EAGAIN;
6933
6934 nr_reclaimed += reclaimed;
6935 }
6936
6937 return nbytes;
6938}
6939
Johannes Weiner241994ed2015-02-11 15:26:06 -08006940static struct cftype memory_files[] = {
6941 {
6942 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08006943 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994ed2015-02-11 15:26:06 -08006944 .read_u64 = memory_current_read,
6945 },
6946 {
Ganesan Rajagopal8e20d4b2022-05-13 16:48:57 -07006947 .name = "peak",
6948 .flags = CFTYPE_NOT_ON_ROOT,
6949 .read_u64 = memory_peak_read,
6950 },
6951 {
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006952 .name = "min",
6953 .flags = CFTYPE_NOT_ON_ROOT,
6954 .seq_show = memory_min_show,
6955 .write = memory_min_write,
6956 },
6957 {
Johannes Weiner241994ed2015-02-11 15:26:06 -08006958 .name = "low",
6959 .flags = CFTYPE_NOT_ON_ROOT,
6960 .seq_show = memory_low_show,
6961 .write = memory_low_write,
6962 },
6963 {
6964 .name = "high",
6965 .flags = CFTYPE_NOT_ON_ROOT,
6966 .seq_show = memory_high_show,
6967 .write = memory_high_write,
6968 },
6969 {
6970 .name = "max",
6971 .flags = CFTYPE_NOT_ON_ROOT,
6972 .seq_show = memory_max_show,
6973 .write = memory_max_write,
6974 },
6975 {
6976 .name = "events",
6977 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04006978 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994ed2015-02-11 15:26:06 -08006979 .seq_show = memory_events_show,
6980 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08006981 {
Shakeel Butt1e577f92019-07-11 20:55:55 -07006982 .name = "events.local",
6983 .flags = CFTYPE_NOT_ON_ROOT,
6984 .file_offset = offsetof(struct mem_cgroup, events_local_file),
6985 .seq_show = memory_events_local_show,
6986 },
6987 {
Johannes Weiner587d9f72016-01-20 15:03:19 -08006988 .name = "stat",
Johannes Weiner587d9f72016-01-20 15:03:19 -08006989 .seq_show = memory_stat_show,
6990 },
Muchun Song5f9a4f42020-10-13 16:52:59 -07006991#ifdef CONFIG_NUMA
6992 {
6993 .name = "numa_stat",
6994 .seq_show = memory_numa_stat_show,
6995 },
6996#endif
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006997 {
6998 .name = "oom.group",
6999 .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE,
7000 .seq_show = memory_oom_group_show,
7001 .write = memory_oom_group_write,
7002 },
Shakeel Butt94968382022-04-29 14:36:59 -07007003 {
7004 .name = "reclaim",
7005 .flags = CFTYPE_NS_DELEGATABLE,
7006 .write = memory_reclaim,
7007 },
Johannes Weiner241994ed2015-02-11 15:26:06 -08007008 { } /* terminate */
7009};
7010
Tejun Heo073219e2014-02-08 10:36:58 -05007011struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08007012 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08007013 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08007014 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08007015 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08007016 .css_free = mem_cgroup_css_free,
Tejun Heo1ced953b2014-07-08 18:02:57 -04007017 .css_reset = mem_cgroup_css_reset,
Johannes Weiner2d146aa2021-04-29 22:56:26 -07007018 .css_rstat_flush = mem_cgroup_css_rstat_flush,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007019 .can_attach = mem_cgroup_can_attach,
Roman Gushchin1aacbd32023-10-19 15:53:42 -07007020#if defined(CONFIG_LRU_GEN) || defined(CONFIG_MEMCG_KMEM)
Yu Zhaobd74fda2022-09-18 02:00:05 -06007021 .attach = mem_cgroup_attach,
Roman Gushchin1aacbd32023-10-19 15:53:42 -07007022#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08007023 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04007024 .post_attach = mem_cgroup_move_task,
Roman Gushchin1aacbd32023-10-19 15:53:42 -07007025#ifdef CONFIG_MEMCG_KMEM
7026 .fork = mem_cgroup_fork,
7027 .exit = mem_cgroup_exit,
7028#endif
Johannes Weiner241994ed2015-02-11 15:26:06 -08007029 .dfl_cftypes = memory_files,
7030 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08007031 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08007032};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08007033
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07007034/*
7035 * This function calculates an individual cgroup's effective
7036 * protection which is derived from its own memory.min/low, its
7037 * parent's and siblings' settings, as well as the actual memory
7038 * distribution in the tree.
7039 *
7040 * The following rules apply to the effective protection values:
7041 *
7042 * 1. At the first level of reclaim, effective protection is equal to
7043 * the declared protection in memory.min and memory.low.
7044 *
7045 * 2. To enable safe delegation of the protection configuration, at
7046 * subsequent levels the effective protection is capped to the
7047 * parent's effective protection.
7048 *
7049 * 3. To make complex and dynamic subtrees easier to configure, the
7050 * user is allowed to overcommit the declared protection at a given
7051 * level. If that is the case, the parent's effective protection is
7052 * distributed to the children in proportion to how much protection
7053 * they have declared and how much of it they are utilizing.
7054 *
7055 * This makes distribution proportional, but also work-conserving:
7056 * if one cgroup claims much more protection than it uses memory,
7057 * the unused remainder is available to its siblings.
7058 *
7059 * 4. Conversely, when the declared protection is undercommitted at a
7060 * given level, the distribution of the larger parental protection
7061 * budget is NOT proportional. A cgroup's protection from a sibling
7062 * is capped to its own memory.min/low setting.
7063 *
Johannes Weiner8a931f82020-04-01 21:07:07 -07007064 * 5. However, to allow protecting recursive subtrees from each other
7065 * without having to declare each individual cgroup's fixed share
7066 * of the ancestor's claim to protection, any unutilized -
7067 * "floating" - protection from up the tree is distributed in
7068 * proportion to each cgroup's *usage*. This makes the protection
7069 * neutral wrt sibling cgroups and lets them compete freely over
7070 * the shared parental protection budget, but it protects the
7071 * subtree as a whole from neighboring subtrees.
7072 *
7073 * Note that 4. and 5. are not in conflict: 4. is about protecting
7074 * against immediate siblings whereas 5. is about protecting against
7075 * neighboring subtrees.
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07007076 */
7077static unsigned long effective_protection(unsigned long usage,
Johannes Weiner8a931f82020-04-01 21:07:07 -07007078 unsigned long parent_usage,
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07007079 unsigned long setting,
7080 unsigned long parent_effective,
7081 unsigned long siblings_protected)
7082{
7083 unsigned long protected;
Johannes Weiner8a931f82020-04-01 21:07:07 -07007084 unsigned long ep;
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07007085
7086 protected = min(usage, setting);
7087 /*
7088 * If all cgroups at this level combined claim and use more
Haifeng Xu08e0f492023-05-22 09:52:32 +00007089 * protection than what the parent affords them, distribute
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07007090 * shares in proportion to utilization.
7091 *
7092 * We are using actual utilization rather than the statically
7093 * claimed protection in order to be work-conserving: claimed
7094 * but unused protection is available to siblings that would
7095 * otherwise get a smaller chunk than what they claimed.
7096 */
7097 if (siblings_protected > parent_effective)
7098 return protected * parent_effective / siblings_protected;
7099
7100 /*
7101 * Ok, utilized protection of all children is within what the
7102 * parent affords them, so we know whatever this child claims
7103 * and utilizes is effectively protected.
7104 *
7105 * If there is unprotected usage beyond this value, reclaim
7106 * will apply pressure in proportion to that amount.
7107 *
7108 * If there is unutilized protection, the cgroup will be fully
7109 * shielded from reclaim, but we do return a smaller value for
7110 * protection than what the group could enjoy in theory. This
7111 * is okay. With the overcommit distribution above, effective
7112 * protection is always dependent on how memory is actually
7113 * consumed among the siblings anyway.
7114 */
Johannes Weiner8a931f82020-04-01 21:07:07 -07007115 ep = protected;
7116
7117 /*
7118 * If the children aren't claiming (all of) the protection
7119 * afforded to them by the parent, distribute the remainder in
7120 * proportion to the (unprotected) memory of each cgroup. That
7121 * way, cgroups that aren't explicitly prioritized wrt each
7122 * other compete freely over the allowance, but they are
7123 * collectively protected from neighboring trees.
7124 *
7125 * We're using unprotected memory for the weight so that if
7126 * some cgroups DO claim explicit protection, we don't protect
7127 * the same bytes twice.
Johannes Weinercd324ed2020-06-25 20:30:16 -07007128 *
7129 * Check both usage and parent_usage against the respective
7130 * protected values. One should imply the other, but they
7131 * aren't read atomically - make sure the division is sane.
Johannes Weiner8a931f82020-04-01 21:07:07 -07007132 */
7133 if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT))
7134 return ep;
Johannes Weinercd324ed2020-06-25 20:30:16 -07007135 if (parent_effective > siblings_protected &&
7136 parent_usage > siblings_protected &&
7137 usage > protected) {
Johannes Weiner8a931f82020-04-01 21:07:07 -07007138 unsigned long unclaimed;
7139
7140 unclaimed = parent_effective - siblings_protected;
7141 unclaimed *= usage - protected;
7142 unclaimed /= parent_usage - siblings_protected;
7143
7144 ep += unclaimed;
7145 }
7146
7147 return ep;
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07007148}
7149
Johannes Weiner241994ed2015-02-11 15:26:06 -08007150/**
Mel Gorman053957182021-06-30 18:53:32 -07007151 * mem_cgroup_calculate_protection - check if memory consumption is in the normal range
Sean Christopherson34c81052017-07-10 15:48:05 -07007152 * @root: the top ancestor of the sub-tree being checked
Johannes Weiner241994ed2015-02-11 15:26:06 -08007153 * @memcg: the memory cgroup to check
7154 *
Roman Gushchin23067152018-06-07 17:06:22 -07007155 * WARNING: This function is not stateless! It can only be used as part
7156 * of a top-down tree iteration, not for isolated queries.
Johannes Weiner241994ed2015-02-11 15:26:06 -08007157 */
Chris Down45c7f7e2020-08-06 23:22:05 -07007158void mem_cgroup_calculate_protection(struct mem_cgroup *root,
7159 struct mem_cgroup *memcg)
Johannes Weiner241994ed2015-02-11 15:26:06 -08007160{
Johannes Weiner8a931f82020-04-01 21:07:07 -07007161 unsigned long usage, parent_usage;
Roman Gushchin23067152018-06-07 17:06:22 -07007162 struct mem_cgroup *parent;
7163
Johannes Weiner241994ed2015-02-11 15:26:06 -08007164 if (mem_cgroup_disabled())
Chris Down45c7f7e2020-08-06 23:22:05 -07007165 return;
Johannes Weiner241994ed2015-02-11 15:26:06 -08007166
Sean Christopherson34c81052017-07-10 15:48:05 -07007167 if (!root)
7168 root = root_mem_cgroup;
Yafang Shao22f74962020-08-06 23:22:01 -07007169
7170 /*
7171 * Effective values of the reclaim targets are ignored so they
7172 * can be stale. Have a look at mem_cgroup_protection for more
7173 * details.
7174 * TODO: calculation should be more robust so that we do not need
7175 * that special casing.
7176 */
Sean Christopherson34c81052017-07-10 15:48:05 -07007177 if (memcg == root)
Chris Down45c7f7e2020-08-06 23:22:05 -07007178 return;
Johannes Weiner241994ed2015-02-11 15:26:06 -08007179
Roman Gushchin23067152018-06-07 17:06:22 -07007180 usage = page_counter_read(&memcg->memory);
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07007181 if (!usage)
Chris Down45c7f7e2020-08-06 23:22:05 -07007182 return;
Sean Christopherson34c81052017-07-10 15:48:05 -07007183
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07007184 parent = parent_mem_cgroup(memcg);
Roman Gushchindf2a4192018-06-14 15:26:17 -07007185
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07007186 if (parent == root) {
Chris Downc3d53202020-04-01 21:07:27 -07007187 memcg->memory.emin = READ_ONCE(memcg->memory.min);
Chris Down03960e32020-06-25 20:30:22 -07007188 memcg->memory.elow = READ_ONCE(memcg->memory.low);
Chris Down45c7f7e2020-08-06 23:22:05 -07007189 return;
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07007190 }
7191
Johannes Weiner8a931f82020-04-01 21:07:07 -07007192 parent_usage = page_counter_read(&parent->memory);
7193
Chris Downb3a78222020-04-01 21:07:33 -07007194 WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage,
Chris Downc3d53202020-04-01 21:07:27 -07007195 READ_ONCE(memcg->memory.min),
7196 READ_ONCE(parent->memory.emin),
Chris Downb3a78222020-04-01 21:07:33 -07007197 atomic_long_read(&parent->memory.children_min_usage)));
Roman Gushchin23067152018-06-07 17:06:22 -07007198
Chris Downb3a78222020-04-01 21:07:33 -07007199 WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage,
Chris Down03960e32020-06-25 20:30:22 -07007200 READ_ONCE(memcg->memory.low),
7201 READ_ONCE(parent->memory.elow),
Chris Downb3a78222020-04-01 21:07:33 -07007202 atomic_long_read(&parent->memory.children_low_usage)));
Johannes Weiner241994ed2015-02-11 15:26:06 -08007203}
7204
Matthew Wilcox (Oracle)8f425e42021-06-25 09:27:04 -04007205static int charge_memcg(struct folio *folio, struct mem_cgroup *memcg,
7206 gfp_t gfp)
Shakeel Butt0add0c72021-04-29 22:56:36 -07007207{
Shakeel Butt0add0c72021-04-29 22:56:36 -07007208 int ret;
7209
Nhat Pham4b569382023-10-06 11:46:26 -07007210 ret = try_charge(memcg, gfp, folio_nr_pages(folio));
Shakeel Butt0add0c72021-04-29 22:56:36 -07007211 if (ret)
7212 goto out;
7213
Nhat Pham4b569382023-10-06 11:46:26 -07007214 mem_cgroup_commit_charge(folio, memcg);
Shakeel Butt0add0c72021-04-29 22:56:36 -07007215out:
7216 return ret;
7217}
7218
Matthew Wilcox (Oracle)8f425e42021-06-25 09:27:04 -04007219int __mem_cgroup_charge(struct folio *folio, struct mm_struct *mm, gfp_t gfp)
Johannes Weiner00501b52014-08-08 14:19:20 -07007220{
Shakeel Butt0add0c72021-04-29 22:56:36 -07007221 struct mem_cgroup *memcg;
7222 int ret;
Johannes Weiner00501b52014-08-08 14:19:20 -07007223
Shakeel Butt0add0c72021-04-29 22:56:36 -07007224 memcg = get_mem_cgroup_from_mm(mm);
Matthew Wilcox (Oracle)8f425e42021-06-25 09:27:04 -04007225 ret = charge_memcg(folio, memcg, gfp);
Shakeel Butt0add0c72021-04-29 22:56:36 -07007226 css_put(&memcg->css);
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007227
Shakeel Butt0add0c72021-04-29 22:56:36 -07007228 return ret;
7229}
Vladimir Davydove993d902015-09-09 15:35:35 -07007230
Shakeel Butt0add0c72021-04-29 22:56:36 -07007231/**
Nhat Pham8cba9572023-10-06 11:46:28 -07007232 * mem_cgroup_hugetlb_try_charge - try to charge the memcg for a hugetlb folio
7233 * @memcg: memcg to charge.
7234 * @gfp: reclaim mode.
7235 * @nr_pages: number of pages to charge.
7236 *
7237 * This function is called when allocating a huge page folio to determine if
7238 * the memcg has the capacity for it. It does not commit the charge yet,
7239 * as the hugetlb folio itself has not been obtained from the hugetlb pool.
7240 *
7241 * Once we have obtained the hugetlb folio, we can call
7242 * mem_cgroup_commit_charge() to commit the charge. If we fail to obtain the
7243 * folio, we should instead call mem_cgroup_cancel_charge() to undo the effect
7244 * of try_charge().
7245 *
7246 * Returns 0 on success. Otherwise, an error code is returned.
7247 */
7248int mem_cgroup_hugetlb_try_charge(struct mem_cgroup *memcg, gfp_t gfp,
7249 long nr_pages)
7250{
7251 /*
7252 * If hugetlb memcg charging is not enabled, do not fail hugetlb allocation,
7253 * but do not attempt to commit charge later (or cancel on error) either.
7254 */
7255 if (mem_cgroup_disabled() || !memcg ||
7256 !cgroup_subsys_on_dfl(memory_cgrp_subsys) ||
7257 !(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING))
7258 return -EOPNOTSUPP;
7259
7260 if (try_charge(memcg, gfp, nr_pages))
7261 return -ENOMEM;
7262
7263 return 0;
7264}
7265
7266/**
Matthew Wilcox (Oracle)65995912022-09-02 20:46:12 +01007267 * mem_cgroup_swapin_charge_folio - Charge a newly allocated folio for swapin.
7268 * @folio: folio to charge.
Shakeel Butt0add0c72021-04-29 22:56:36 -07007269 * @mm: mm context of the victim
7270 * @gfp: reclaim mode
Matthew Wilcox (Oracle)65995912022-09-02 20:46:12 +01007271 * @entry: swap entry for which the folio is allocated
Shakeel Butt0add0c72021-04-29 22:56:36 -07007272 *
Matthew Wilcox (Oracle)65995912022-09-02 20:46:12 +01007273 * This function charges a folio allocated for swapin. Please call this before
7274 * adding the folio to the swapcache.
Shakeel Butt0add0c72021-04-29 22:56:36 -07007275 *
7276 * Returns 0 on success. Otherwise, an error code is returned.
7277 */
Matthew Wilcox (Oracle)65995912022-09-02 20:46:12 +01007278int mem_cgroup_swapin_charge_folio(struct folio *folio, struct mm_struct *mm,
Shakeel Butt0add0c72021-04-29 22:56:36 -07007279 gfp_t gfp, swp_entry_t entry)
7280{
7281 struct mem_cgroup *memcg;
7282 unsigned short id;
7283 int ret;
Johannes Weiner00501b52014-08-08 14:19:20 -07007284
Shakeel Butt0add0c72021-04-29 22:56:36 -07007285 if (mem_cgroup_disabled())
7286 return 0;
7287
7288 id = lookup_swap_cgroup_id(entry);
7289 rcu_read_lock();
7290 memcg = mem_cgroup_from_id(id);
7291 if (!memcg || !css_tryget_online(&memcg->css))
Johannes Weiner00501b52014-08-08 14:19:20 -07007292 memcg = get_mem_cgroup_from_mm(mm);
Shakeel Butt0add0c72021-04-29 22:56:36 -07007293 rcu_read_unlock();
Johannes Weiner00501b52014-08-08 14:19:20 -07007294
Matthew Wilcox (Oracle)8f425e42021-06-25 09:27:04 -04007295 ret = charge_memcg(folio, memcg, gfp);
Johannes Weiner00501b52014-08-08 14:19:20 -07007296
Shakeel Butt0add0c72021-04-29 22:56:36 -07007297 css_put(&memcg->css);
7298 return ret;
7299}
Johannes Weiner6abb5a82014-08-08 14:19:33 -07007300
Shakeel Butt0add0c72021-04-29 22:56:36 -07007301/*
7302 * mem_cgroup_swapin_uncharge_swap - uncharge swap slot
7303 * @entry: swap entry for which the page is charged
7304 *
7305 * Call this function after successfully adding the charged page to swapcache.
7306 *
7307 * Note: This function assumes the page for which swap slot is being uncharged
7308 * is order 0 page.
7309 */
7310void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry)
7311{
Muchun Songcae3af62021-02-24 12:04:19 -08007312 /*
7313 * Cgroup1's unified memory+swap counter has been charged with the
7314 * new swapcache page, finish the transfer by uncharging the swap
7315 * slot. The swap slot would also get uncharged when it dies, but
7316 * it can stick around indefinitely and we'd count the page twice
7317 * the entire time.
7318 *
7319 * Cgroup2 has separate resource counters for memory and swap,
7320 * so this is a non-issue here. Memory and swap charge lifetimes
7321 * correspond 1:1 to page and swap slot lifetimes: we charge the
7322 * page to memory here, and uncharge swap when the slot is freed.
7323 */
Shakeel Butt0add0c72021-04-29 22:56:36 -07007324 if (!mem_cgroup_disabled() && do_memsw_account()) {
Johannes Weiner00501b52014-08-08 14:19:20 -07007325 /*
7326 * The swap entry might not get freed for a long time,
7327 * let's not wait for it. The page already received a
7328 * memory+swap charge, drop the swap entry duplicate.
7329 */
Shakeel Butt0add0c72021-04-29 22:56:36 -07007330 mem_cgroup_uncharge_swap(entry, 1);
Johannes Weiner00501b52014-08-08 14:19:20 -07007331 }
Johannes Weiner3fea5a42020-06-03 16:01:41 -07007332}
7333
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007334struct uncharge_gather {
7335 struct mem_cgroup *memcg;
Muchun Songb4e0b682021-04-29 22:56:52 -07007336 unsigned long nr_memory;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007337 unsigned long pgpgout;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007338 unsigned long nr_kmem;
Matthew Wilcox (Oracle)8e88bd22021-06-25 09:05:47 -04007339 int nid;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007340};
7341
7342static inline void uncharge_gather_clear(struct uncharge_gather *ug)
Johannes Weiner747db952014-08-08 14:19:24 -07007343{
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007344 memset(ug, 0, sizeof(*ug));
7345}
7346
7347static void uncharge_batch(const struct uncharge_gather *ug)
7348{
Johannes Weiner747db952014-08-08 14:19:24 -07007349 unsigned long flags;
7350
Muchun Songb4e0b682021-04-29 22:56:52 -07007351 if (ug->nr_memory) {
7352 page_counter_uncharge(&ug->memcg->memory, ug->nr_memory);
Johannes Weiner7941d212016-01-14 15:21:23 -08007353 if (do_memsw_account())
Muchun Songb4e0b682021-04-29 22:56:52 -07007354 page_counter_uncharge(&ug->memcg->memsw, ug->nr_memory);
Yosry Ahmeda8c49af2022-03-22 14:40:10 -07007355 if (ug->nr_kmem)
7356 memcg_account_kmem(ug->memcg, -ug->nr_kmem);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007357 memcg_oom_recover(ug->memcg);
Johannes Weinerce00a962014-09-05 08:43:57 -04007358 }
Johannes Weiner747db952014-08-08 14:19:24 -07007359
7360 local_irq_save(flags);
Johannes Weinerc9019e92018-01-31 16:16:37 -08007361 __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout);
Muchun Songb4e0b682021-04-29 22:56:52 -07007362 __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_memory);
Matthew Wilcox (Oracle)8e88bd22021-06-25 09:05:47 -04007363 memcg_check_events(ug->memcg, ug->nid);
Johannes Weiner747db952014-08-08 14:19:24 -07007364 local_irq_restore(flags);
Michal Hockof1796542020-09-04 16:35:24 -07007365
Matthew Wilcox (Oracle)c4ed6eb2021-06-29 21:47:12 -04007366 /* drop reference from uncharge_folio */
Michal Hockof1796542020-09-04 16:35:24 -07007367 css_put(&ug->memcg->css);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007368}
7369
Matthew Wilcox (Oracle)c4ed6eb2021-06-29 21:47:12 -04007370static void uncharge_folio(struct folio *folio, struct uncharge_gather *ug)
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007371{
Matthew Wilcox (Oracle)c4ed6eb2021-06-29 21:47:12 -04007372 long nr_pages;
Muchun Songb4e0b682021-04-29 22:56:52 -07007373 struct mem_cgroup *memcg;
7374 struct obj_cgroup *objcg;
Johannes Weiner9f762db2020-06-03 16:01:44 -07007375
Matthew Wilcox (Oracle)c4ed6eb2021-06-29 21:47:12 -04007376 VM_BUG_ON_FOLIO(folio_test_lru(folio), folio);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007377
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007378 /*
7379 * Nobody should be changing or seriously looking at
Matthew Wilcox (Oracle)c4ed6eb2021-06-29 21:47:12 -04007380 * folio memcg or objcg at this point, we have fully
7381 * exclusive access to the folio.
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007382 */
Michal Hockofead2b82022-03-22 14:40:35 -07007383 if (folio_memcg_kmem(folio)) {
Matthew Wilcox (Oracle)1b7e4462021-06-28 14:59:26 -04007384 objcg = __folio_objcg(folio);
Muchun Songb4e0b682021-04-29 22:56:52 -07007385 /*
7386 * This get matches the put at the end of the function and
7387 * kmem pages do not hold memcg references anymore.
7388 */
7389 memcg = get_mem_cgroup_from_objcg(objcg);
7390 } else {
Matthew Wilcox (Oracle)1b7e4462021-06-28 14:59:26 -04007391 memcg = __folio_memcg(folio);
Muchun Songb4e0b682021-04-29 22:56:52 -07007392 }
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007393
Muchun Songb4e0b682021-04-29 22:56:52 -07007394 if (!memcg)
7395 return;
7396
7397 if (ug->memcg != memcg) {
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007398 if (ug->memcg) {
7399 uncharge_batch(ug);
7400 uncharge_gather_clear(ug);
7401 }
Muchun Songb4e0b682021-04-29 22:56:52 -07007402 ug->memcg = memcg;
Matthew Wilcox (Oracle)c4ed6eb2021-06-29 21:47:12 -04007403 ug->nid = folio_nid(folio);
Michal Hockof1796542020-09-04 16:35:24 -07007404
7405 /* pairs with css_put in uncharge_batch */
Muchun Songb4e0b682021-04-29 22:56:52 -07007406 css_get(&memcg->css);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007407 }
7408
Matthew Wilcox (Oracle)c4ed6eb2021-06-29 21:47:12 -04007409 nr_pages = folio_nr_pages(folio);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007410
Michal Hockofead2b82022-03-22 14:40:35 -07007411 if (folio_memcg_kmem(folio)) {
Muchun Songb4e0b682021-04-29 22:56:52 -07007412 ug->nr_memory += nr_pages;
Johannes Weiner9f762db2020-06-03 16:01:44 -07007413 ug->nr_kmem += nr_pages;
Muchun Songb4e0b682021-04-29 22:56:52 -07007414
Matthew Wilcox (Oracle)c4ed6eb2021-06-29 21:47:12 -04007415 folio->memcg_data = 0;
Muchun Songb4e0b682021-04-29 22:56:52 -07007416 obj_cgroup_put(objcg);
7417 } else {
7418 /* LRU pages aren't accounted at the root level */
7419 if (!mem_cgroup_is_root(memcg))
7420 ug->nr_memory += nr_pages;
Roman Gushchin18b2db32020-12-01 13:58:30 -08007421 ug->pgpgout++;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007422
Matthew Wilcox (Oracle)c4ed6eb2021-06-29 21:47:12 -04007423 folio->memcg_data = 0;
Muchun Songb4e0b682021-04-29 22:56:52 -07007424 }
7425
7426 css_put(&memcg->css);
Johannes Weiner747db952014-08-08 14:19:24 -07007427}
7428
Matthew Wilcox (Oracle)bbc6b702021-05-01 20:42:23 -04007429void __mem_cgroup_uncharge(struct folio *folio)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007430{
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007431 struct uncharge_gather ug;
7432
Matthew Wilcox (Oracle)bbc6b702021-05-01 20:42:23 -04007433 /* Don't touch folio->lru of any random page, pre-check: */
7434 if (!folio_memcg(folio))
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007435 return;
7436
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007437 uncharge_gather_clear(&ug);
Matthew Wilcox (Oracle)bbc6b702021-05-01 20:42:23 -04007438 uncharge_folio(folio, &ug);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07007439 uncharge_batch(&ug);
Johannes Weiner747db952014-08-08 14:19:24 -07007440}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007441
Johannes Weiner747db952014-08-08 14:19:24 -07007442/**
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07007443 * __mem_cgroup_uncharge_list - uncharge a list of page
Johannes Weiner747db952014-08-08 14:19:24 -07007444 * @page_list: list of pages to uncharge
7445 *
7446 * Uncharge a list of pages previously charged with
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07007447 * __mem_cgroup_charge().
Johannes Weiner747db952014-08-08 14:19:24 -07007448 */
Suren Baghdasaryan2c8d8f92021-09-02 14:54:50 -07007449void __mem_cgroup_uncharge_list(struct list_head *page_list)
Johannes Weiner747db952014-08-08 14:19:24 -07007450{
Muchun Songc41a40b2021-02-24 12:04:08 -08007451 struct uncharge_gather ug;
Matthew Wilcox (Oracle)c4ed6eb2021-06-29 21:47:12 -04007452 struct folio *folio;
Muchun Songc41a40b2021-02-24 12:04:08 -08007453
Muchun Songc41a40b2021-02-24 12:04:08 -08007454 uncharge_gather_clear(&ug);
Matthew Wilcox (Oracle)c4ed6eb2021-06-29 21:47:12 -04007455 list_for_each_entry(folio, page_list, lru)
7456 uncharge_folio(folio, &ug);
Muchun Songc41a40b2021-02-24 12:04:08 -08007457 if (ug.memcg)
7458 uncharge_batch(&ug);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007459}
7460
7461/**
Nhat Pham85ce2c52023-10-06 11:46:27 -07007462 * mem_cgroup_replace_folio - Charge a folio's replacement.
Matthew Wilcox (Oracle)d21bba22021-05-06 18:14:59 -04007463 * @old: Currently circulating folio.
7464 * @new: Replacement folio.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007465 *
Matthew Wilcox (Oracle)d21bba22021-05-06 18:14:59 -04007466 * Charge @new as a replacement folio for @old. @old will
Nhat Pham85ce2c52023-10-06 11:46:27 -07007467 * be uncharged upon free. This is only used by the page cache
7468 * (in replace_page_cache_folio()).
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007469 *
Matthew Wilcox (Oracle)d21bba22021-05-06 18:14:59 -04007470 * Both folios must be locked, @new->mapping must be set up.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007471 */
Nhat Pham85ce2c52023-10-06 11:46:27 -07007472void mem_cgroup_replace_folio(struct folio *old, struct folio *new)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007473{
Johannes Weiner29833312014-12-10 15:44:02 -08007474 struct mem_cgroup *memcg;
Matthew Wilcox (Oracle)d21bba22021-05-06 18:14:59 -04007475 long nr_pages = folio_nr_pages(new);
Tejun Heod93c4132016-06-24 14:49:54 -07007476 unsigned long flags;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007477
Matthew Wilcox (Oracle)d21bba22021-05-06 18:14:59 -04007478 VM_BUG_ON_FOLIO(!folio_test_locked(old), old);
7479 VM_BUG_ON_FOLIO(!folio_test_locked(new), new);
7480 VM_BUG_ON_FOLIO(folio_test_anon(old) != folio_test_anon(new), new);
7481 VM_BUG_ON_FOLIO(folio_nr_pages(old) != nr_pages, new);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007482
7483 if (mem_cgroup_disabled())
7484 return;
7485
Matthew Wilcox (Oracle)d21bba22021-05-06 18:14:59 -04007486 /* Page cache replacement: new folio already charged? */
7487 if (folio_memcg(new))
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007488 return;
7489
Matthew Wilcox (Oracle)d21bba22021-05-06 18:14:59 -04007490 memcg = folio_memcg(old);
7491 VM_WARN_ON_ONCE_FOLIO(!memcg, old);
Johannes Weiner29833312014-12-10 15:44:02 -08007492 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007493 return;
7494
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08007495 /* Force-charge the new page. The old one will be freed soon */
Muchun Song8dc87c72021-06-28 19:37:47 -07007496 if (!mem_cgroup_is_root(memcg)) {
7497 page_counter_charge(&memcg->memory, nr_pages);
7498 if (do_memsw_account())
7499 page_counter_charge(&memcg->memsw, nr_pages);
7500 }
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007501
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07007502 css_get(&memcg->css);
Matthew Wilcox (Oracle)d21bba22021-05-06 18:14:59 -04007503 commit_charge(new, memcg);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08007504
Tejun Heod93c4132016-06-24 14:49:54 -07007505 local_irq_save(flags);
Matthew Wilcox (Oracle)6e0110c2021-04-29 13:40:11 -04007506 mem_cgroup_charge_statistics(memcg, nr_pages);
Matthew Wilcox (Oracle)d21bba22021-05-06 18:14:59 -04007507 memcg_check_events(memcg, folio_nid(new));
Tejun Heod93c4132016-06-24 14:49:54 -07007508 local_irq_restore(flags);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007509}
7510
Nhat Pham85ce2c52023-10-06 11:46:27 -07007511/**
7512 * mem_cgroup_migrate - Transfer the memcg data from the old to the new folio.
7513 * @old: Currently circulating folio.
7514 * @new: Replacement folio.
7515 *
7516 * Transfer the memcg data from the old folio to the new folio for migration.
7517 * The old folio's data info will be cleared. Note that the memory counters
7518 * will remain unchanged throughout the process.
7519 *
7520 * Both folios must be locked, @new->mapping must be set up.
7521 */
7522void mem_cgroup_migrate(struct folio *old, struct folio *new)
7523{
7524 struct mem_cgroup *memcg;
7525
7526 VM_BUG_ON_FOLIO(!folio_test_locked(old), old);
7527 VM_BUG_ON_FOLIO(!folio_test_locked(new), new);
7528 VM_BUG_ON_FOLIO(folio_test_anon(old) != folio_test_anon(new), new);
7529 VM_BUG_ON_FOLIO(folio_nr_pages(old) != folio_nr_pages(new), new);
7530
7531 if (mem_cgroup_disabled())
7532 return;
7533
7534 memcg = folio_memcg(old);
Nhat Pham8cba9572023-10-06 11:46:28 -07007535 /*
7536 * Note that it is normal to see !memcg for a hugetlb folio.
7537 * For e.g, itt could have been allocated when memory_hugetlb_accounting
7538 * was not selected.
7539 */
7540 VM_WARN_ON_ONCE_FOLIO(!folio_test_hugetlb(old) && !memcg, old);
Nhat Pham85ce2c52023-10-06 11:46:27 -07007541 if (!memcg)
7542 return;
7543
7544 /* Transfer the charge and the css ref */
7545 commit_charge(new, memcg);
7546 old->memcg_data = 0;
7547}
7548
Johannes Weineref129472016-01-14 15:21:34 -08007549DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08007550EXPORT_SYMBOL(memcg_sockets_enabled_key);
7551
Johannes Weiner2d758072016-10-07 17:00:58 -07007552void mem_cgroup_sk_alloc(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08007553{
7554 struct mem_cgroup *memcg;
7555
Johannes Weiner2d758072016-10-07 17:00:58 -07007556 if (!mem_cgroup_sockets_enabled)
7557 return;
7558
Shakeel Butte876ecc2020-03-09 22:16:05 -07007559 /* Do not associate the sock with unrelated interrupted task's memcg. */
Shakeel Butt086f6942022-03-22 14:40:07 -07007560 if (!in_task())
Shakeel Butte876ecc2020-03-09 22:16:05 -07007561 return;
7562
Johannes Weiner11092082016-01-14 15:21:26 -08007563 rcu_read_lock();
7564 memcg = mem_cgroup_from_task(current);
Kamalesh Babulal7848ed62022-09-30 19:14:33 +05307565 if (mem_cgroup_is_root(memcg))
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007566 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08007567 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007568 goto out;
Shakeel Butt8965aa22020-04-01 21:07:10 -07007569 if (css_tryget(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08007570 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007571out:
Johannes Weiner11092082016-01-14 15:21:26 -08007572 rcu_read_unlock();
7573}
Johannes Weiner11092082016-01-14 15:21:26 -08007574
Johannes Weiner2d758072016-10-07 17:00:58 -07007575void mem_cgroup_sk_free(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08007576{
Johannes Weiner2d758072016-10-07 17:00:58 -07007577 if (sk->sk_memcg)
7578 css_put(&sk->sk_memcg->css);
Johannes Weiner11092082016-01-14 15:21:26 -08007579}
7580
7581/**
7582 * mem_cgroup_charge_skmem - charge socket memory
7583 * @memcg: memcg to charge
7584 * @nr_pages: number of pages to charge
Wei Wang4b1327b2021-08-17 12:40:03 -07007585 * @gfp_mask: reclaim mode
Johannes Weiner11092082016-01-14 15:21:26 -08007586 *
7587 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
Wei Wang4b1327b2021-08-17 12:40:03 -07007588 * @memcg's configured limit, %false if it doesn't.
Johannes Weiner11092082016-01-14 15:21:26 -08007589 */
Wei Wang4b1327b2021-08-17 12:40:03 -07007590bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages,
7591 gfp_t gfp_mask)
Johannes Weiner11092082016-01-14 15:21:26 -08007592{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007593 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08007594 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007595
Johannes Weiner0db15292016-01-20 15:02:50 -08007596 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
7597 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007598 return true;
7599 }
Johannes Weiner0db15292016-01-20 15:02:50 -08007600 memcg->tcpmem_pressure = 1;
Wei Wang4b1327b2021-08-17 12:40:03 -07007601 if (gfp_mask & __GFP_NOFAIL) {
7602 page_counter_charge(&memcg->tcpmem, nr_pages);
7603 return true;
7604 }
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007605 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08007606 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08007607
Wei Wang4b1327b2021-08-17 12:40:03 -07007608 if (try_charge(memcg, gfp_mask, nr_pages) == 0) {
7609 mod_memcg_state(memcg, MEMCG_SOCK, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007610 return true;
Wei Wang4b1327b2021-08-17 12:40:03 -07007611 }
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007612
Johannes Weiner11092082016-01-14 15:21:26 -08007613 return false;
7614}
7615
7616/**
7617 * mem_cgroup_uncharge_skmem - uncharge socket memory
Mike Rapoportb7701a52018-02-06 15:42:13 -08007618 * @memcg: memcg to uncharge
7619 * @nr_pages: number of pages to uncharge
Johannes Weiner11092082016-01-14 15:21:26 -08007620 */
7621void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
7622{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007623 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08007624 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007625 return;
7626 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08007627
Johannes Weinerc9019e92018-01-31 16:16:37 -08007628 mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages);
Johannes Weinerb2807f02016-01-20 15:03:22 -08007629
Roman Gushchin475d0482017-09-08 16:13:09 -07007630 refill_stock(memcg, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08007631}
7632
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007633static int __init cgroup_memory(char *s)
7634{
7635 char *token;
7636
7637 while ((token = strsep(&s, ",")) != NULL) {
7638 if (!*token)
7639 continue;
7640 if (!strcmp(token, "nosocket"))
7641 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08007642 if (!strcmp(token, "nokmem"))
7643 cgroup_memory_nokmem = true;
Yafang Shaob6c1a8a2023-02-10 15:47:31 +00007644 if (!strcmp(token, "nobpf"))
7645 cgroup_memory_nobpf = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007646 }
Randy Dunlap460a79e2022-03-22 14:40:31 -07007647 return 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007648}
7649__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08007650
Michal Hocko2d110852013-02-22 16:34:43 -08007651/*
Michal Hocko10813122013-02-22 16:35:41 -08007652 * subsys_initcall() for memory controller.
7653 *
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01007654 * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this
7655 * context because of lock dependencies (cgroup_lock -> cpu hotplug) but
7656 * basically everything that doesn't depend on a specific mem_cgroup structure
7657 * should be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08007658 */
7659static int __init mem_cgroup_init(void)
7660{
Johannes Weiner95a045f2015-02-11 15:26:33 -08007661 int cpu, node;
7662
Muchun Songf3344ad2021-02-24 12:03:15 -08007663 /*
7664 * Currently s32 type (can refer to struct batched_lruvec_stat) is
7665 * used for per-memcg-per-cpu caching of per-node statistics. In order
7666 * to work fine, we should make sure that the overfill threshold can't
7667 * exceed S32_MAX / PAGE_SIZE.
7668 */
7669 BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE);
7670
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01007671 cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL,
7672 memcg_hotplug_cpu_dead);
Johannes Weiner95a045f2015-02-11 15:26:33 -08007673
7674 for_each_possible_cpu(cpu)
7675 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
7676 drain_local_stock);
7677
7678 for_each_node(node) {
7679 struct mem_cgroup_tree_per_node *rtpn;
Johannes Weiner95a045f2015-02-11 15:26:33 -08007680
Haifeng Xu91f0dcc2023-06-19 13:04:42 +00007681 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, node);
Johannes Weiner95a045f2015-02-11 15:26:33 -08007682
Mel Gormanef8f2322016-07-28 15:46:05 -07007683 rtpn->rb_root = RB_ROOT;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -07007684 rtpn->rb_rightmost = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -07007685 spin_lock_init(&rtpn->lock);
Johannes Weiner95a045f2015-02-11 15:26:33 -08007686 soft_limit_tree.rb_tree_per_node[node] = rtpn;
7687 }
7688
Michal Hocko2d110852013-02-22 16:34:43 -08007689 return 0;
7690}
7691subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08007692
Johannes Weinere55b9f92022-09-26 09:57:04 -04007693#ifdef CONFIG_SWAP
Arnd Bergmann358c07f2016-08-25 15:17:08 -07007694static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
7695{
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07007696 while (!refcount_inc_not_zero(&memcg->id.ref)) {
Arnd Bergmann358c07f2016-08-25 15:17:08 -07007697 /*
7698 * The root cgroup cannot be destroyed, so it's refcount must
7699 * always be >= 1.
7700 */
Kamalesh Babulal7848ed62022-09-30 19:14:33 +05307701 if (WARN_ON_ONCE(mem_cgroup_is_root(memcg))) {
Arnd Bergmann358c07f2016-08-25 15:17:08 -07007702 VM_BUG_ON(1);
7703 break;
7704 }
7705 memcg = parent_mem_cgroup(memcg);
7706 if (!memcg)
7707 memcg = root_mem_cgroup;
7708 }
7709 return memcg;
7710}
7711
Johannes Weiner21afa382015-02-11 15:26:36 -08007712/**
7713 * mem_cgroup_swapout - transfer a memsw charge to swap
Matthew Wilcox (Oracle)3ecb0082021-12-27 21:11:34 -05007714 * @folio: folio whose memsw charge to transfer
Johannes Weiner21afa382015-02-11 15:26:36 -08007715 * @entry: swap entry to move the charge to
7716 *
Matthew Wilcox (Oracle)3ecb0082021-12-27 21:11:34 -05007717 * Transfer the memsw charge of @folio to @entry.
Johannes Weiner21afa382015-02-11 15:26:36 -08007718 */
Matthew Wilcox (Oracle)3ecb0082021-12-27 21:11:34 -05007719void mem_cgroup_swapout(struct folio *folio, swp_entry_t entry)
Johannes Weiner21afa382015-02-11 15:26:36 -08007720{
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007721 struct mem_cgroup *memcg, *swap_memcg;
Huang Yingd6810d732017-09-06 16:22:45 -07007722 unsigned int nr_entries;
Johannes Weiner21afa382015-02-11 15:26:36 -08007723 unsigned short oldid;
7724
Matthew Wilcox (Oracle)3ecb0082021-12-27 21:11:34 -05007725 VM_BUG_ON_FOLIO(folio_test_lru(folio), folio);
7726 VM_BUG_ON_FOLIO(folio_ref_count(folio), folio);
Johannes Weiner21afa382015-02-11 15:26:36 -08007727
Alex Shi76358ab2020-12-18 14:01:28 -08007728 if (mem_cgroup_disabled())
7729 return;
7730
Johannes Weinerb94c4e92022-09-26 09:57:03 -04007731 if (!do_memsw_account())
Johannes Weiner21afa382015-02-11 15:26:36 -08007732 return;
7733
Matthew Wilcox (Oracle)3ecb0082021-12-27 21:11:34 -05007734 memcg = folio_memcg(folio);
Johannes Weiner21afa382015-02-11 15:26:36 -08007735
Matthew Wilcox (Oracle)3ecb0082021-12-27 21:11:34 -05007736 VM_WARN_ON_ONCE_FOLIO(!memcg, folio);
Johannes Weiner21afa382015-02-11 15:26:36 -08007737 if (!memcg)
7738 return;
7739
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007740 /*
7741 * In case the memcg owning these pages has been offlined and doesn't
7742 * have an ID allocated to it anymore, charge the closest online
7743 * ancestor for the swap instead and transfer the memory+swap charge.
7744 */
7745 swap_memcg = mem_cgroup_id_get_online(memcg);
Matthew Wilcox (Oracle)3ecb0082021-12-27 21:11:34 -05007746 nr_entries = folio_nr_pages(folio);
Huang Yingd6810d732017-09-06 16:22:45 -07007747 /* Get references for the tail pages, too */
7748 if (nr_entries > 1)
7749 mem_cgroup_id_get_many(swap_memcg, nr_entries - 1);
7750 oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg),
7751 nr_entries);
Matthew Wilcox (Oracle)3ecb0082021-12-27 21:11:34 -05007752 VM_BUG_ON_FOLIO(oldid, folio);
Johannes Weinerc9019e92018-01-31 16:16:37 -08007753 mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007754
Matthew Wilcox (Oracle)3ecb0082021-12-27 21:11:34 -05007755 folio->memcg_data = 0;
Johannes Weiner21afa382015-02-11 15:26:36 -08007756
7757 if (!mem_cgroup_is_root(memcg))
Huang Yingd6810d732017-09-06 16:22:45 -07007758 page_counter_uncharge(&memcg->memory, nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007759
Johannes Weinerb25806d2022-09-26 09:57:02 -04007760 if (memcg != swap_memcg) {
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007761 if (!mem_cgroup_is_root(swap_memcg))
Huang Yingd6810d732017-09-06 16:22:45 -07007762 page_counter_charge(&swap_memcg->memsw, nr_entries);
7763 page_counter_uncharge(&memcg->memsw, nr_entries);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007764 }
7765
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07007766 /*
7767 * Interrupts should be disabled here because the caller holds the
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07007768 * i_pages lock which is taken with interrupts-off. It is
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07007769 * important here to have the interrupts disabled because it is the
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07007770 * only synchronisation we have for updating the per-CPU variables.
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07007771 */
Sebastian Andrzej Siewiorbe3e67b2022-03-22 14:40:41 -07007772 memcg_stats_lock();
Matthew Wilcox (Oracle)6e0110c2021-04-29 13:40:11 -04007773 mem_cgroup_charge_statistics(memcg, -nr_entries);
Sebastian Andrzej Siewiorbe3e67b2022-03-22 14:40:41 -07007774 memcg_stats_unlock();
Matthew Wilcox (Oracle)3ecb0082021-12-27 21:11:34 -05007775 memcg_check_events(memcg, folio_nid(folio));
Johannes Weiner73f576c2016-07-20 15:44:57 -07007776
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07007777 css_put(&memcg->css);
Johannes Weiner21afa382015-02-11 15:26:36 -08007778}
7779
Huang Ying38d8b4e2017-07-06 15:37:18 -07007780/**
Matthew Wilcox (Oracle)e2e3fdc2022-05-12 20:23:02 -07007781 * __mem_cgroup_try_charge_swap - try charging swap space for a folio
7782 * @folio: folio being added to swap
Vladimir Davydov37e84352016-01-20 15:02:56 -08007783 * @entry: swap entry to charge
7784 *
Matthew Wilcox (Oracle)e2e3fdc2022-05-12 20:23:02 -07007785 * Try to charge @folio's memcg for the swap space at @entry.
Vladimir Davydov37e84352016-01-20 15:02:56 -08007786 *
7787 * Returns 0 on success, -ENOMEM on failure.
7788 */
Matthew Wilcox (Oracle)e2e3fdc2022-05-12 20:23:02 -07007789int __mem_cgroup_try_charge_swap(struct folio *folio, swp_entry_t entry)
Vladimir Davydov37e84352016-01-20 15:02:56 -08007790{
Matthew Wilcox (Oracle)e2e3fdc2022-05-12 20:23:02 -07007791 unsigned int nr_pages = folio_nr_pages(folio);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007792 struct page_counter *counter;
Huang Ying38d8b4e2017-07-06 15:37:18 -07007793 struct mem_cgroup *memcg;
Vladimir Davydov37e84352016-01-20 15:02:56 -08007794 unsigned short oldid;
7795
Johannes Weinerb94c4e92022-09-26 09:57:03 -04007796 if (do_memsw_account())
Vladimir Davydov37e84352016-01-20 15:02:56 -08007797 return 0;
7798
Matthew Wilcox (Oracle)e2e3fdc2022-05-12 20:23:02 -07007799 memcg = folio_memcg(folio);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007800
Matthew Wilcox (Oracle)e2e3fdc2022-05-12 20:23:02 -07007801 VM_WARN_ON_ONCE_FOLIO(!memcg, folio);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007802 if (!memcg)
7803 return 0;
7804
Tejun Heof3a53a32018-06-07 17:05:35 -07007805 if (!entry.val) {
7806 memcg_memory_event(memcg, MEMCG_SWAP_FAIL);
Tejun Heobb98f2c2018-06-07 17:05:31 -07007807 return 0;
Tejun Heof3a53a32018-06-07 17:05:35 -07007808 }
Tejun Heobb98f2c2018-06-07 17:05:31 -07007809
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007810 memcg = mem_cgroup_id_get_online(memcg);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007811
Johannes Weinerb25806d2022-09-26 09:57:02 -04007812 if (!mem_cgroup_is_root(memcg) &&
Huang Ying38d8b4e2017-07-06 15:37:18 -07007813 !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) {
Tejun Heof3a53a32018-06-07 17:05:35 -07007814 memcg_memory_event(memcg, MEMCG_SWAP_MAX);
7815 memcg_memory_event(memcg, MEMCG_SWAP_FAIL);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007816 mem_cgroup_id_put(memcg);
7817 return -ENOMEM;
7818 }
7819
Huang Ying38d8b4e2017-07-06 15:37:18 -07007820 /* Get references for the tail pages, too */
7821 if (nr_pages > 1)
7822 mem_cgroup_id_get_many(memcg, nr_pages - 1);
7823 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages);
Matthew Wilcox (Oracle)e2e3fdc2022-05-12 20:23:02 -07007824 VM_BUG_ON_FOLIO(oldid, folio);
Johannes Weinerc9019e92018-01-31 16:16:37 -08007825 mod_memcg_state(memcg, MEMCG_SWAP, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007826
Vladimir Davydov37e84352016-01-20 15:02:56 -08007827 return 0;
7828}
7829
Johannes Weiner21afa382015-02-11 15:26:36 -08007830/**
Suren Baghdasaryan01c4b282021-09-02 14:54:54 -07007831 * __mem_cgroup_uncharge_swap - uncharge swap space
Johannes Weiner21afa382015-02-11 15:26:36 -08007832 * @entry: swap entry to uncharge
Huang Ying38d8b4e2017-07-06 15:37:18 -07007833 * @nr_pages: the amount of swap space to uncharge
Johannes Weiner21afa382015-02-11 15:26:36 -08007834 */
Suren Baghdasaryan01c4b282021-09-02 14:54:54 -07007835void __mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages)
Johannes Weiner21afa382015-02-11 15:26:36 -08007836{
7837 struct mem_cgroup *memcg;
7838 unsigned short id;
7839
Huang Ying38d8b4e2017-07-06 15:37:18 -07007840 id = swap_cgroup_record(entry, 0, nr_pages);
Johannes Weiner21afa382015-02-11 15:26:36 -08007841 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07007842 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08007843 if (memcg) {
Johannes Weinerb25806d2022-09-26 09:57:02 -04007844 if (!mem_cgroup_is_root(memcg)) {
Johannes Weinerb94c4e92022-09-26 09:57:03 -04007845 if (do_memsw_account())
Huang Ying38d8b4e2017-07-06 15:37:18 -07007846 page_counter_uncharge(&memcg->memsw, nr_pages);
Johannes Weinerb94c4e92022-09-26 09:57:03 -04007847 else
7848 page_counter_uncharge(&memcg->swap, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007849 }
Johannes Weinerc9019e92018-01-31 16:16:37 -08007850 mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages);
Huang Ying38d8b4e2017-07-06 15:37:18 -07007851 mem_cgroup_id_put_many(memcg, nr_pages);
Johannes Weiner21afa382015-02-11 15:26:36 -08007852 }
7853 rcu_read_unlock();
7854}
7855
Vladimir Davydovd8b38432016-01-20 15:03:07 -08007856long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
7857{
7858 long nr_swap_pages = get_nr_swap_pages();
7859
Johannes Weinerb25806d2022-09-26 09:57:02 -04007860 if (mem_cgroup_disabled() || do_memsw_account())
Vladimir Davydovd8b38432016-01-20 15:03:07 -08007861 return nr_swap_pages;
Kamalesh Babulal7848ed62022-09-30 19:14:33 +05307862 for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg))
Vladimir Davydovd8b38432016-01-20 15:03:07 -08007863 nr_swap_pages = min_t(long, nr_swap_pages,
Roman Gushchinbbec2e12018-06-07 17:06:18 -07007864 READ_ONCE(memcg->swap.max) -
Vladimir Davydovd8b38432016-01-20 15:03:07 -08007865 page_counter_read(&memcg->swap));
7866 return nr_swap_pages;
7867}
7868
Matthew Wilcox (Oracle)9202d527b2022-09-02 20:46:43 +01007869bool mem_cgroup_swap_full(struct folio *folio)
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007870{
7871 struct mem_cgroup *memcg;
7872
Matthew Wilcox (Oracle)9202d527b2022-09-02 20:46:43 +01007873 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007874
7875 if (vm_swap_full())
7876 return true;
Johannes Weinerb25806d2022-09-26 09:57:02 -04007877 if (do_memsw_account())
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007878 return false;
7879
Matthew Wilcox (Oracle)9202d527b2022-09-02 20:46:43 +01007880 memcg = folio_memcg(folio);
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007881 if (!memcg)
7882 return false;
7883
Kamalesh Babulal7848ed62022-09-30 19:14:33 +05307884 for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) {
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007885 unsigned long usage = page_counter_read(&memcg->swap);
7886
7887 if (usage * 2 >= READ_ONCE(memcg->swap.high) ||
7888 usage * 2 >= READ_ONCE(memcg->swap.max))
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007889 return true;
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007890 }
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007891
7892 return false;
7893}
7894
Johannes Weinereccb52e2020-06-03 16:02:11 -07007895static int __init setup_swap_account(char *s)
Johannes Weiner21afa382015-02-11 15:26:36 -08007896{
Johannes Weinerb25806d2022-09-26 09:57:02 -04007897 pr_warn_once("The swapaccount= commandline option is deprecated. "
7898 "Please report your usecase to linux-mm@kvack.org if you "
7899 "depend on this functionality.\n");
Johannes Weiner21afa382015-02-11 15:26:36 -08007900 return 1;
7901}
Johannes Weinereccb52e2020-06-03 16:02:11 -07007902__setup("swapaccount=", setup_swap_account);
Johannes Weiner21afa382015-02-11 15:26:36 -08007903
Vladimir Davydov37e84352016-01-20 15:02:56 -08007904static u64 swap_current_read(struct cgroup_subsys_state *css,
7905 struct cftype *cft)
7906{
7907 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
7908
7909 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
7910}
7911
Lars R. Damerowe0e0b412023-05-24 11:17:33 -07007912static u64 swap_peak_read(struct cgroup_subsys_state *css,
7913 struct cftype *cft)
7914{
7915 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
7916
7917 return (u64)memcg->swap.watermark * PAGE_SIZE;
7918}
7919
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007920static int swap_high_show(struct seq_file *m, void *v)
7921{
7922 return seq_puts_memcg_tunable(m,
7923 READ_ONCE(mem_cgroup_from_seq(m)->swap.high));
7924}
7925
7926static ssize_t swap_high_write(struct kernfs_open_file *of,
7927 char *buf, size_t nbytes, loff_t off)
7928{
7929 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
7930 unsigned long high;
7931 int err;
7932
7933 buf = strstrip(buf);
7934 err = page_counter_memparse(buf, "max", &high);
7935 if (err)
7936 return err;
7937
7938 page_counter_set_high(&memcg->swap, high);
7939
7940 return nbytes;
7941}
7942
Vladimir Davydov37e84352016-01-20 15:02:56 -08007943static int swap_max_show(struct seq_file *m, void *v)
7944{
Chris Down677dc972019-03-05 15:45:55 -08007945 return seq_puts_memcg_tunable(m,
7946 READ_ONCE(mem_cgroup_from_seq(m)->swap.max));
Vladimir Davydov37e84352016-01-20 15:02:56 -08007947}
7948
7949static ssize_t swap_max_write(struct kernfs_open_file *of,
7950 char *buf, size_t nbytes, loff_t off)
7951{
7952 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
7953 unsigned long max;
7954 int err;
7955
7956 buf = strstrip(buf);
7957 err = page_counter_memparse(buf, "max", &max);
7958 if (err)
7959 return err;
7960
Tejun Heobe091022018-06-07 17:09:21 -07007961 xchg(&memcg->swap.max, max);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007962
7963 return nbytes;
7964}
7965
Tejun Heof3a53a32018-06-07 17:05:35 -07007966static int swap_events_show(struct seq_file *m, void *v)
7967{
Chris Downaa9694b2019-03-05 15:45:52 -08007968 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Tejun Heof3a53a32018-06-07 17:05:35 -07007969
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007970 seq_printf(m, "high %lu\n",
7971 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH]));
Tejun Heof3a53a32018-06-07 17:05:35 -07007972 seq_printf(m, "max %lu\n",
7973 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX]));
7974 seq_printf(m, "fail %lu\n",
7975 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL]));
7976
7977 return 0;
7978}
7979
Vladimir Davydov37e84352016-01-20 15:02:56 -08007980static struct cftype swap_files[] = {
7981 {
7982 .name = "swap.current",
7983 .flags = CFTYPE_NOT_ON_ROOT,
7984 .read_u64 = swap_current_read,
7985 },
7986 {
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007987 .name = "swap.high",
7988 .flags = CFTYPE_NOT_ON_ROOT,
7989 .seq_show = swap_high_show,
7990 .write = swap_high_write,
7991 },
7992 {
Vladimir Davydov37e84352016-01-20 15:02:56 -08007993 .name = "swap.max",
7994 .flags = CFTYPE_NOT_ON_ROOT,
7995 .seq_show = swap_max_show,
7996 .write = swap_max_write,
7997 },
Tejun Heof3a53a32018-06-07 17:05:35 -07007998 {
Lars R. Damerowe0e0b412023-05-24 11:17:33 -07007999 .name = "swap.peak",
8000 .flags = CFTYPE_NOT_ON_ROOT,
8001 .read_u64 = swap_peak_read,
8002 },
8003 {
Tejun Heof3a53a32018-06-07 17:05:35 -07008004 .name = "swap.events",
8005 .flags = CFTYPE_NOT_ON_ROOT,
8006 .file_offset = offsetof(struct mem_cgroup, swap_events_file),
8007 .seq_show = swap_events_show,
8008 },
Vladimir Davydov37e84352016-01-20 15:02:56 -08008009 { } /* terminate */
8010};
8011
Johannes Weinereccb52e2020-06-03 16:02:11 -07008012static struct cftype memsw_files[] = {
Johannes Weiner21afa382015-02-11 15:26:36 -08008013 {
8014 .name = "memsw.usage_in_bytes",
8015 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
8016 .read_u64 = mem_cgroup_read_u64,
8017 },
8018 {
8019 .name = "memsw.max_usage_in_bytes",
8020 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
8021 .write = mem_cgroup_reset,
8022 .read_u64 = mem_cgroup_read_u64,
8023 },
8024 {
8025 .name = "memsw.limit_in_bytes",
8026 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
8027 .write = mem_cgroup_write,
8028 .read_u64 = mem_cgroup_read_u64,
8029 },
8030 {
8031 .name = "memsw.failcnt",
8032 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
8033 .write = mem_cgroup_reset,
8034 .read_u64 = mem_cgroup_read_u64,
8035 },
8036 { }, /* terminate */
8037};
8038
Johannes Weinerf4840cc2022-05-19 14:08:53 -07008039#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP)
8040/**
8041 * obj_cgroup_may_zswap - check if this cgroup can zswap
8042 * @objcg: the object cgroup
8043 *
8044 * Check if the hierarchical zswap limit has been reached.
8045 *
8046 * This doesn't check for specific headroom, and it is not atomic
8047 * either. But with zswap, the size of the allocation is only known
Muhammad Muzammilbe16dd762023-10-23 17:44:05 +05008048 * once compression has occurred, and this optimistic pre-check avoids
Johannes Weinerf4840cc2022-05-19 14:08:53 -07008049 * spending cycles on compression when there is already no room left
8050 * or zswap is disabled altogether somewhere in the hierarchy.
8051 */
8052bool obj_cgroup_may_zswap(struct obj_cgroup *objcg)
8053{
8054 struct mem_cgroup *memcg, *original_memcg;
8055 bool ret = true;
8056
8057 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
8058 return true;
8059
8060 original_memcg = get_mem_cgroup_from_objcg(objcg);
Kamalesh Babulal7848ed62022-09-30 19:14:33 +05308061 for (memcg = original_memcg; !mem_cgroup_is_root(memcg);
Johannes Weinerf4840cc2022-05-19 14:08:53 -07008062 memcg = parent_mem_cgroup(memcg)) {
8063 unsigned long max = READ_ONCE(memcg->zswap_max);
8064 unsigned long pages;
8065
8066 if (max == PAGE_COUNTER_MAX)
8067 continue;
8068 if (max == 0) {
8069 ret = false;
8070 break;
8071 }
8072
8073 cgroup_rstat_flush(memcg->css.cgroup);
8074 pages = memcg_page_state(memcg, MEMCG_ZSWAP_B) / PAGE_SIZE;
8075 if (pages < max)
8076 continue;
8077 ret = false;
8078 break;
8079 }
8080 mem_cgroup_put(original_memcg);
8081 return ret;
8082}
8083
8084/**
8085 * obj_cgroup_charge_zswap - charge compression backend memory
8086 * @objcg: the object cgroup
8087 * @size: size of compressed object
8088 *
Miaohe Lin3a1060c2023-08-03 20:00:21 +08008089 * This forces the charge after obj_cgroup_may_zswap() allowed
Johannes Weinerf4840cc2022-05-19 14:08:53 -07008090 * compression and storage in zwap for this cgroup to go ahead.
8091 */
8092void obj_cgroup_charge_zswap(struct obj_cgroup *objcg, size_t size)
8093{
8094 struct mem_cgroup *memcg;
8095
8096 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
8097 return;
8098
8099 VM_WARN_ON_ONCE(!(current->flags & PF_MEMALLOC));
8100
8101 /* PF_MEMALLOC context, charging must succeed */
8102 if (obj_cgroup_charge(objcg, GFP_KERNEL, size))
8103 VM_WARN_ON_ONCE(1);
8104
8105 rcu_read_lock();
8106 memcg = obj_cgroup_memcg(objcg);
8107 mod_memcg_state(memcg, MEMCG_ZSWAP_B, size);
8108 mod_memcg_state(memcg, MEMCG_ZSWAPPED, 1);
8109 rcu_read_unlock();
8110}
8111
8112/**
8113 * obj_cgroup_uncharge_zswap - uncharge compression backend memory
8114 * @objcg: the object cgroup
8115 * @size: size of compressed object
8116 *
8117 * Uncharges zswap memory on page in.
8118 */
8119void obj_cgroup_uncharge_zswap(struct obj_cgroup *objcg, size_t size)
8120{
8121 struct mem_cgroup *memcg;
8122
8123 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
8124 return;
8125
8126 obj_cgroup_uncharge(objcg, size);
8127
8128 rcu_read_lock();
8129 memcg = obj_cgroup_memcg(objcg);
8130 mod_memcg_state(memcg, MEMCG_ZSWAP_B, -size);
8131 mod_memcg_state(memcg, MEMCG_ZSWAPPED, -1);
8132 rcu_read_unlock();
8133}
8134
8135static u64 zswap_current_read(struct cgroup_subsys_state *css,
8136 struct cftype *cft)
8137{
8138 cgroup_rstat_flush(css->cgroup);
8139 return memcg_page_state(mem_cgroup_from_css(css), MEMCG_ZSWAP_B);
8140}
8141
8142static int zswap_max_show(struct seq_file *m, void *v)
8143{
8144 return seq_puts_memcg_tunable(m,
8145 READ_ONCE(mem_cgroup_from_seq(m)->zswap_max));
8146}
8147
8148static ssize_t zswap_max_write(struct kernfs_open_file *of,
8149 char *buf, size_t nbytes, loff_t off)
8150{
8151 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
8152 unsigned long max;
8153 int err;
8154
8155 buf = strstrip(buf);
8156 err = page_counter_memparse(buf, "max", &max);
8157 if (err)
8158 return err;
8159
8160 xchg(&memcg->zswap_max, max);
8161
8162 return nbytes;
8163}
8164
8165static struct cftype zswap_files[] = {
8166 {
8167 .name = "zswap.current",
8168 .flags = CFTYPE_NOT_ON_ROOT,
8169 .read_u64 = zswap_current_read,
8170 },
8171 {
8172 .name = "zswap.max",
8173 .flags = CFTYPE_NOT_ON_ROOT,
8174 .seq_show = zswap_max_show,
8175 .write = zswap_max_write,
8176 },
8177 { } /* terminate */
8178};
8179#endif /* CONFIG_MEMCG_KMEM && CONFIG_ZSWAP */
8180
Johannes Weiner21afa382015-02-11 15:26:36 -08008181static int __init mem_cgroup_swap_init(void)
8182{
Johannes Weiner2d1c4982020-06-03 16:02:14 -07008183 if (mem_cgroup_disabled())
Johannes Weinereccb52e2020-06-03 16:02:11 -07008184 return 0;
8185
8186 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files));
8187 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files));
Johannes Weinerf4840cc2022-05-19 14:08:53 -07008188#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP)
8189 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, zswap_files));
8190#endif
Johannes Weiner21afa382015-02-11 15:26:36 -08008191 return 0;
8192}
Johannes Weinerb25806d2022-09-26 09:57:02 -04008193subsys_initcall(mem_cgroup_swap_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08008194
Johannes Weinere55b9f92022-09-26 09:57:04 -04008195#endif /* CONFIG_SWAP */