blob: 2e13c0a419fc86e346408499fba1b3748a1453b1 [file] [log] [blame]
Thomas Gleixnerc942fdd2019-05-27 08:55:06 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Balbir Singh8cdea7c2008-02-07 00:13:50 -08002/* memcontrol.c - Memory Controller
3 *
4 * Copyright IBM Corporation, 2007
5 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08007 * Copyright 2007 OpenVZ SWsoft Inc
8 * Author: Pavel Emelianov <xemul@openvz.org>
9 *
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080010 * Memory thresholds
11 * Copyright (C) 2009 Nokia Corporation
12 * Author: Kirill A. Shutemov
13 *
Glauber Costa7ae1e1d2012-12-18 14:21:56 -080014 * Kernel Memory Controller
15 * Copyright (C) 2012 Parallels Inc. and Google Inc.
16 * Authors: Glauber Costa and Suleiman Souhlal
17 *
Johannes Weiner1575e682015-04-14 15:44:51 -070018 * Native page reclaim
19 * Charge lifetime sanitation
20 * Lockless page tracking & accounting
21 * Unified hierarchy configuration model
22 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
Balbir Singh8cdea7c2008-02-07 00:13:50 -080023 */
24
Johannes Weiner3e32cb22014-12-10 15:42:31 -080025#include <linux/page_counter.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080026#include <linux/memcontrol.h>
27#include <linux/cgroup.h>
Christoph Hellwiga5201102019-08-28 16:19:53 +020028#include <linux/pagewalk.h>
Ingo Molnar6e84f312017-02-08 18:51:29 +010029#include <linux/sched/mm.h>
Hugh Dickins3a4f8a02017-02-24 14:59:36 -080030#include <linux/shmem_fs.h>
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -080031#include <linux/hugetlb.h>
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -080032#include <linux/pagemap.h>
Chris Down1ff9e6e2019-03-05 15:48:09 -080033#include <linux/vm_event_item.h>
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -080034#include <linux/smp.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080036#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080037#include <linux/bit_spinlock.h>
38#include <linux/rcupdate.h>
Balbir Singhe2224322009-04-02 16:57:39 -070039#include <linux/limits.h>
Paul Gortmakerb9e15ba2011-05-26 16:00:52 -040040#include <linux/export.h>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080041#include <linux/mutex.h>
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -070042#include <linux/rbtree.h>
Balbir Singhb6ac57d2008-04-29 01:00:19 -070043#include <linux/slab.h>
Balbir Singh66e17072008-02-07 00:13:56 -080044#include <linux/swap.h>
Daisuke Nishimura02491442010-03-10 15:22:17 -080045#include <linux/swapops.h>
Balbir Singh66e17072008-02-07 00:13:56 -080046#include <linux/spinlock.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080047#include <linux/eventfd.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050048#include <linux/poll.h>
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -080049#include <linux/sort.h>
Balbir Singh66e17072008-02-07 00:13:56 -080050#include <linux/fs.h>
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -080051#include <linux/seq_file.h>
Anton Vorontsov70ddf632013-04-29 15:08:31 -070052#include <linux/vmpressure.h>
Christoph Lameterb69408e2008-10-18 20:26:14 -070053#include <linux/mm_inline.h>
Johannes Weiner5d1ea482014-12-10 15:44:55 -080054#include <linux/swap_cgroup.h>
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -080055#include <linux/cpu.h>
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070056#include <linux/oom.h>
Johannes Weiner0056f4e2013-10-31 16:34:14 -070057#include <linux/lockdep.h>
Tejun Heo79bd9812013-11-22 18:20:42 -050058#include <linux/file.h>
Tejun Heob23afb92015-11-05 18:46:11 -080059#include <linux/tracehook.h>
Chris Down0e4b01d2019-09-23 15:34:55 -070060#include <linux/psi.h>
Johannes Weinerc8713d02019-07-11 20:55:59 -070061#include <linux/seq_buf.h>
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -080062#include "internal.h"
Glauber Costad1a4c0b2011-12-11 21:47:04 +000063#include <net/sock.h>
Michal Hocko4bd2c1e2012-10-08 16:33:10 -070064#include <net/ip.h>
Qiang Huangf35c3a82013-11-12 15:08:22 -080065#include "slab.h"
Balbir Singh8cdea7c2008-02-07 00:13:50 -080066
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080067#include <linux/uaccess.h>
Balbir Singh8697d332008-02-07 00:13:59 -080068
KOSAKI Motohirocc8e9702010-08-09 17:19:57 -070069#include <trace/events/vmscan.h>
70
Tejun Heo073219e2014-02-08 10:36:58 -050071struct cgroup_subsys memory_cgrp_subsys __read_mostly;
72EXPORT_SYMBOL(memory_cgrp_subsys);
David Rientjes68ae5642012-12-12 13:51:57 -080073
Johannes Weiner7d828602016-01-14 15:20:56 -080074struct mem_cgroup *root_mem_cgroup __read_mostly;
75
Roman Gushchin37d59852020-10-17 16:13:50 -070076/* Active memory cgroup to use from an interrupt context */
77DEFINE_PER_CPU(struct mem_cgroup *, int_active_memcg);
78
Johannes Weinerf7e1cb62016-01-14 15:21:29 -080079/* Socket memory accounting disabled? */
80static bool cgroup_memory_nosocket;
81
Vladimir Davydov04823c82016-01-20 15:02:38 -080082/* Kernel memory accounting disabled? */
83static bool cgroup_memory_nokmem;
84
Johannes Weiner21afa382015-02-11 15:26:36 -080085/* Whether the swap controller is active */
Andrew Mortonc255a452012-07-31 16:43:02 -070086#ifdef CONFIG_MEMCG_SWAP
Johannes Weinereccb52e2020-06-03 16:02:11 -070087bool cgroup_memory_noswap __read_mostly;
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080088#else
Johannes Weinereccb52e2020-06-03 16:02:11 -070089#define cgroup_memory_noswap 1
Johannes Weiner2d1c4982020-06-03 16:02:14 -070090#endif
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -080091
Tejun Heo97b27822019-08-26 09:06:56 -070092#ifdef CONFIG_CGROUP_WRITEBACK
93static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq);
94#endif
95
Johannes Weiner7941d212016-01-14 15:21:23 -080096/* Whether legacy memory+swap accounting is active */
97static bool do_memsw_account(void)
98{
Johannes Weinereccb52e2020-06-03 16:02:11 -070099 return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_noswap;
Johannes Weiner7941d212016-01-14 15:21:23 -0800100}
101
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700102#define THRESHOLDS_EVENTS_TARGET 128
103#define SOFTLIMIT_EVENTS_TARGET 1024
Johannes Weinere9f89742011-03-23 16:42:37 -0700104
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700105/*
106 * Cgroups above their limits are maintained in a RB-Tree, independent of
107 * their hierarchy representation
108 */
109
Mel Gormanef8f2322016-07-28 15:46:05 -0700110struct mem_cgroup_tree_per_node {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700111 struct rb_root rb_root;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700112 struct rb_node *rb_rightmost;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700113 spinlock_t lock;
114};
115
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700116struct mem_cgroup_tree {
117 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
118};
119
120static struct mem_cgroup_tree soft_limit_tree __read_mostly;
121
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700122/* for OOM */
123struct mem_cgroup_eventfd_list {
124 struct list_head list;
125 struct eventfd_ctx *eventfd;
126};
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800127
Tejun Heo79bd9812013-11-22 18:20:42 -0500128/*
129 * cgroup_event represents events which userspace want to receive.
130 */
Tejun Heo3bc942f2013-11-22 18:20:44 -0500131struct mem_cgroup_event {
Tejun Heo79bd9812013-11-22 18:20:42 -0500132 /*
Tejun Heo59b6f872013-11-22 18:20:43 -0500133 * memcg which the event belongs to.
Tejun Heo79bd9812013-11-22 18:20:42 -0500134 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500135 struct mem_cgroup *memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -0500136 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500137 * eventfd to signal userspace about the event.
138 */
139 struct eventfd_ctx *eventfd;
140 /*
141 * Each of these stored in a list by the cgroup.
142 */
143 struct list_head list;
144 /*
Tejun Heofba94802013-11-22 18:20:43 -0500145 * register_event() callback will be used to add new userspace
146 * waiter for changes related to this event. Use eventfd_signal()
147 * on eventfd to send notification to userspace.
148 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500149 int (*register_event)(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -0500150 struct eventfd_ctx *eventfd, const char *args);
Tejun Heofba94802013-11-22 18:20:43 -0500151 /*
152 * unregister_event() callback will be called when userspace closes
153 * the eventfd or on cgroup removing. This callback must be set,
154 * if you want provide notification functionality.
155 */
Tejun Heo59b6f872013-11-22 18:20:43 -0500156 void (*unregister_event)(struct mem_cgroup *memcg,
Tejun Heofba94802013-11-22 18:20:43 -0500157 struct eventfd_ctx *eventfd);
158 /*
Tejun Heo79bd9812013-11-22 18:20:42 -0500159 * All fields below needed to unregister event when
160 * userspace closes eventfd.
161 */
162 poll_table pt;
163 wait_queue_head_t *wqh;
Ingo Molnarac6424b2017-06-20 12:06:13 +0200164 wait_queue_entry_t wait;
Tejun Heo79bd9812013-11-22 18:20:42 -0500165 struct work_struct remove;
166};
167
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700168static void mem_cgroup_threshold(struct mem_cgroup *memcg);
169static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -0800170
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800171/* Stuffs for move charges at task migration. */
172/*
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800173 * Types of charges to be moved.
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800174 */
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800175#define MOVE_ANON 0x1U
176#define MOVE_FILE 0x2U
177#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -0800178
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800179/* "mc" and its members are protected by cgroup_mutex */
180static struct move_charge_struct {
Daisuke Nishimurab1dd6932010-11-24 12:57:06 -0800181 spinlock_t lock; /* for from, to */
Tejun Heo264a0ae2016-04-21 19:09:02 -0400182 struct mm_struct *mm;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800183 struct mem_cgroup *from;
184 struct mem_cgroup *to;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -0800185 unsigned long flags;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800186 unsigned long precharge;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -0800187 unsigned long moved_charge;
Daisuke Nishimura483c30b2010-03-10 15:22:18 -0800188 unsigned long moved_swap;
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800189 struct task_struct *moving_task; /* a task moving charges */
190 wait_queue_head_t waitq; /* a waitq for other context */
191} mc = {
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -0700192 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
Daisuke Nishimura8033b972010-03-10 15:22:16 -0800193 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
194};
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -0800195
Balbir Singh4e416952009-09-23 15:56:39 -0700196/*
197 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
198 * limit reclaim to prevent infinite loops, if they ever occur.
199 */
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700200#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700201#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
Balbir Singh4e416952009-09-23 15:56:39 -0700202
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800203/* for encoding cft->private value on file */
Glauber Costa86ae53e2012-12-18 14:21:45 -0800204enum res_type {
205 _MEM,
206 _MEMSWAP,
207 _OOM_TYPE,
Glauber Costa510fc4e2012-12-18 14:21:47 -0800208 _KMEM,
Vladimir Davydovd55f90b2016-01-20 15:02:44 -0800209 _TCP,
Glauber Costa86ae53e2012-12-18 14:21:45 -0800210};
211
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -0700212#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
213#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800214#define MEMFILE_ATTR(val) ((val) & 0xffff)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -0700215/* Used for OOM nofiier */
216#define OOM_CONTROL (0)
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -0800217
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
Tetsuo Handa7775fac2019-03-05 15:46:47 -0800233static inline bool should_force_charge(void)
234{
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
247struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
248{
249 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
250}
251
Kirill Tkhai84c07d12018-08-17 15:47:25 -0700252#ifdef CONFIG_MEMCG_KMEM
Roman Gushchinbf4f0592020-08-06 23:20:49 -0700253extern spinlock_t css_set_lock;
254
255static void obj_cgroup_release(struct percpu_ref *ref)
256{
257 struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt);
258 struct mem_cgroup *memcg;
259 unsigned int nr_bytes;
260 unsigned int nr_pages;
261 unsigned long flags;
262
263 /*
264 * At this point all allocated objects are freed, and
265 * objcg->nr_charged_bytes can't have an arbitrary byte value.
266 * However, it can be PAGE_SIZE or (x * PAGE_SIZE).
267 *
268 * The following sequence can lead to it:
269 * 1) CPU0: objcg == stock->cached_objcg
270 * 2) CPU1: we do a small allocation (e.g. 92 bytes),
271 * PAGE_SIZE bytes are charged
272 * 3) CPU1: a process from another memcg is allocating something,
273 * the stock if flushed,
274 * objcg->nr_charged_bytes = PAGE_SIZE - 92
275 * 5) CPU0: we do release this object,
276 * 92 bytes are added to stock->nr_bytes
277 * 6) CPU0: stock is flushed,
278 * 92 bytes are added to objcg->nr_charged_bytes
279 *
280 * In the result, nr_charged_bytes == PAGE_SIZE.
281 * This page will be uncharged in obj_cgroup_release().
282 */
283 nr_bytes = atomic_read(&objcg->nr_charged_bytes);
284 WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1));
285 nr_pages = nr_bytes >> PAGE_SHIFT;
286
287 spin_lock_irqsave(&css_set_lock, flags);
288 memcg = obj_cgroup_memcg(objcg);
289 if (nr_pages)
290 __memcg_kmem_uncharge(memcg, nr_pages);
291 list_del(&objcg->list);
292 mem_cgroup_put(memcg);
293 spin_unlock_irqrestore(&css_set_lock, flags);
294
295 percpu_ref_exit(ref);
296 kfree_rcu(objcg, rcu);
297}
298
299static struct obj_cgroup *obj_cgroup_alloc(void)
300{
301 struct obj_cgroup *objcg;
302 int ret;
303
304 objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL);
305 if (!objcg)
306 return NULL;
307
308 ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0,
309 GFP_KERNEL);
310 if (ret) {
311 kfree(objcg);
312 return NULL;
313 }
314 INIT_LIST_HEAD(&objcg->list);
315 return objcg;
316}
317
318static void memcg_reparent_objcgs(struct mem_cgroup *memcg,
319 struct mem_cgroup *parent)
320{
321 struct obj_cgroup *objcg, *iter;
322
323 objcg = rcu_replace_pointer(memcg->objcg, NULL, true);
324
325 spin_lock_irq(&css_set_lock);
326
327 /* Move active objcg to the parent's list */
328 xchg(&objcg->memcg, parent);
329 css_get(&parent->css);
330 list_add(&objcg->list, &parent->objcg_list);
331
332 /* Move already reparented objcgs to the parent's list */
333 list_for_each_entry(iter, &memcg->objcg_list, list) {
334 css_get(&parent->css);
335 xchg(&iter->memcg, parent);
336 css_put(&memcg->css);
337 }
338 list_splice(&memcg->objcg_list, &parent->objcg_list);
339
340 spin_unlock_irq(&css_set_lock);
341
342 percpu_ref_kill(&objcg->refcnt);
343}
344
Glauber Costa55007d82012-12-18 14:22:38 -0800345/*
Roman Gushchin98556092020-08-06 23:21:10 -0700346 * This will be used as a shrinker list's index.
Li Zefanb8627832013-09-23 16:56:47 +0800347 * The main reason for not using cgroup id for this:
348 * this works better in sparse environments, where we have a lot of memcgs,
349 * but only a few kmem-limited. Or also, if we have, for instance, 200
350 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
351 * 200 entry array for that.
Glauber Costa55007d82012-12-18 14:22:38 -0800352 *
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800353 * The current size of the caches array is stored in memcg_nr_cache_ids. It
354 * will double each time we have to increase it.
Glauber Costa55007d82012-12-18 14:22:38 -0800355 */
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -0800356static DEFINE_IDA(memcg_cache_ida);
357int memcg_nr_cache_ids;
Glauber Costa749c5412012-12-18 14:23:01 -0800358
Vladimir Davydov05257a12015-02-12 14:59:01 -0800359/* Protects memcg_nr_cache_ids */
360static DECLARE_RWSEM(memcg_cache_ids_sem);
361
362void memcg_get_cache_ids(void)
363{
364 down_read(&memcg_cache_ids_sem);
365}
366
367void memcg_put_cache_ids(void)
368{
369 up_read(&memcg_cache_ids_sem);
370}
371
Glauber Costa55007d82012-12-18 14:22:38 -0800372/*
373 * MIN_SIZE is different than 1, because we would like to avoid going through
374 * the alloc/free process all the time. In a small machine, 4 kmem-limited
375 * cgroups is a reasonable guess. In the future, it could be a parameter or
376 * tunable, but that is strictly not necessary.
377 *
Li Zefanb8627832013-09-23 16:56:47 +0800378 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
Glauber Costa55007d82012-12-18 14:22:38 -0800379 * this constant directly from cgroup, but it is understandable that this is
380 * better kept as an internal representation in cgroup.c. In any case, the
Li Zefanb8627832013-09-23 16:56:47 +0800381 * cgrp_id space is not getting any smaller, and we don't have to necessarily
Glauber Costa55007d82012-12-18 14:22:38 -0800382 * increase ours as well if it increases.
383 */
384#define MEMCG_CACHES_MIN_SIZE 4
Li Zefanb8627832013-09-23 16:56:47 +0800385#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
Glauber Costa55007d82012-12-18 14:22:38 -0800386
Glauber Costad7f25f82012-12-18 14:22:40 -0800387/*
388 * A lot of the calls to the cache allocation functions are expected to be
Roman Gushchin272911a2020-08-06 23:21:17 -0700389 * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are
Glauber Costad7f25f82012-12-18 14:22:40 -0800390 * conditional to this static branch, we'll have to allow modules that does
391 * kmem_cache_alloc and the such to see this symbol as well
392 */
Johannes Weineref129472016-01-14 15:21:34 -0800393DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
Glauber Costad7f25f82012-12-18 14:22:40 -0800394EXPORT_SYMBOL(memcg_kmem_enabled_key);
Yang Shi0a432dc2019-09-23 15:38:12 -0700395#endif
Tejun Heo17cc4df2017-02-22 15:41:36 -0800396
Kirill Tkhai0a4465d2018-08-17 15:47:37 -0700397static int memcg_shrinker_map_size;
398static DEFINE_MUTEX(memcg_shrinker_map_mutex);
399
400static void memcg_free_shrinker_map_rcu(struct rcu_head *head)
401{
402 kvfree(container_of(head, struct memcg_shrinker_map, rcu));
403}
404
405static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg,
406 int size, int old_size)
407{
408 struct memcg_shrinker_map *new, *old;
409 int nid;
410
411 lockdep_assert_held(&memcg_shrinker_map_mutex);
412
413 for_each_node(nid) {
414 old = rcu_dereference_protected(
415 mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true);
416 /* Not yet online memcg */
417 if (!old)
418 return 0;
419
Kirill Tkhai86daf942020-04-01 21:06:33 -0700420 new = kvmalloc_node(sizeof(*new) + size, GFP_KERNEL, nid);
Kirill Tkhai0a4465d2018-08-17 15:47:37 -0700421 if (!new)
422 return -ENOMEM;
423
424 /* Set all old bits, clear all new bits */
425 memset(new->map, (int)0xff, old_size);
426 memset((void *)new->map + old_size, 0, size - old_size);
427
428 rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new);
429 call_rcu(&old->rcu, memcg_free_shrinker_map_rcu);
430 }
431
432 return 0;
433}
434
435static void memcg_free_shrinker_maps(struct mem_cgroup *memcg)
436{
437 struct mem_cgroup_per_node *pn;
438 struct memcg_shrinker_map *map;
439 int nid;
440
441 if (mem_cgroup_is_root(memcg))
442 return;
443
444 for_each_node(nid) {
445 pn = mem_cgroup_nodeinfo(memcg, nid);
446 map = rcu_dereference_protected(pn->shrinker_map, true);
447 if (map)
448 kvfree(map);
449 rcu_assign_pointer(pn->shrinker_map, NULL);
450 }
451}
452
453static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg)
454{
455 struct memcg_shrinker_map *map;
456 int nid, size, ret = 0;
457
458 if (mem_cgroup_is_root(memcg))
459 return 0;
460
461 mutex_lock(&memcg_shrinker_map_mutex);
462 size = memcg_shrinker_map_size;
463 for_each_node(nid) {
Kirill Tkhai86daf942020-04-01 21:06:33 -0700464 map = kvzalloc_node(sizeof(*map) + size, GFP_KERNEL, nid);
Kirill Tkhai0a4465d2018-08-17 15:47:37 -0700465 if (!map) {
466 memcg_free_shrinker_maps(memcg);
467 ret = -ENOMEM;
468 break;
469 }
470 rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map);
471 }
472 mutex_unlock(&memcg_shrinker_map_mutex);
473
474 return ret;
475}
476
477int memcg_expand_shrinker_maps(int new_id)
478{
479 int size, old_size, ret = 0;
480 struct mem_cgroup *memcg;
481
482 size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long);
483 old_size = memcg_shrinker_map_size;
484 if (size <= old_size)
485 return 0;
486
487 mutex_lock(&memcg_shrinker_map_mutex);
488 if (!root_mem_cgroup)
489 goto unlock;
490
491 for_each_mem_cgroup(memcg) {
492 if (mem_cgroup_is_root(memcg))
493 continue;
494 ret = memcg_expand_one_shrinker_map(memcg, size, old_size);
Vasily Averin75866af2020-02-20 20:04:18 -0800495 if (ret) {
496 mem_cgroup_iter_break(NULL, memcg);
Kirill Tkhai0a4465d2018-08-17 15:47:37 -0700497 goto unlock;
Vasily Averin75866af2020-02-20 20:04:18 -0800498 }
Kirill Tkhai0a4465d2018-08-17 15:47:37 -0700499 }
500unlock:
501 if (!ret)
502 memcg_shrinker_map_size = size;
503 mutex_unlock(&memcg_shrinker_map_mutex);
504 return ret;
505}
Kirill Tkhaifae91d62018-08-17 15:48:10 -0700506
507void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id)
508{
509 if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) {
510 struct memcg_shrinker_map *map;
511
512 rcu_read_lock();
513 map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map);
Kirill Tkhaif90280d2018-08-17 15:48:25 -0700514 /* Pairs with smp mb in shrink_slab() */
515 smp_mb__before_atomic();
Kirill Tkhaifae91d62018-08-17 15:48:10 -0700516 set_bit(shrinker_id, map->map);
517 rcu_read_unlock();
518 }
519}
520
Tejun Heoad7fa852015-05-27 20:00:02 -0400521/**
522 * mem_cgroup_css_from_page - css of the memcg associated with a page
523 * @page: page of interest
524 *
525 * If memcg is bound to the default hierarchy, css of the memcg associated
526 * with @page is returned. The returned css remains associated with @page
527 * until it is released.
528 *
529 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
530 * is returned.
Tejun Heoad7fa852015-05-27 20:00:02 -0400531 */
532struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
533{
534 struct mem_cgroup *memcg;
535
Tejun Heoad7fa852015-05-27 20:00:02 -0400536 memcg = page->mem_cgroup;
537
Tejun Heo9e10a132015-09-18 11:56:28 -0400538 if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heoad7fa852015-05-27 20:00:02 -0400539 memcg = root_mem_cgroup;
540
Tejun Heoad7fa852015-05-27 20:00:02 -0400541 return &memcg->css;
542}
543
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700544/**
545 * page_cgroup_ino - return inode number of the memcg a page is charged to
546 * @page: the page
547 *
548 * Look up the closest online ancestor of the memory cgroup @page is charged to
549 * and return its inode number or 0 if @page is not charged to any cgroup. It
550 * is safe to call this function without holding a reference to @page.
551 *
552 * Note, this function is inherently racy, because there is nothing to prevent
553 * the cgroup inode from getting torn down and potentially reallocated a moment
554 * after page_cgroup_ino() returns, so it only should be used by callers that
555 * do not care (such as procfs interfaces).
556 */
557ino_t page_cgroup_ino(struct page *page)
558{
559 struct mem_cgroup *memcg;
560 unsigned long ino = 0;
561
562 rcu_read_lock();
Roman Gushchin98556092020-08-06 23:21:10 -0700563 memcg = page->mem_cgroup;
Roman Gushchin286e04b2020-08-06 23:20:52 -0700564
Roman Gushchin98556092020-08-06 23:21:10 -0700565 /*
566 * The lowest bit set means that memcg isn't a valid
567 * memcg pointer, but a obj_cgroups pointer.
568 * In this case the page is shared and doesn't belong
569 * to any specific memory cgroup.
570 */
571 if ((unsigned long) memcg & 0x1UL)
572 memcg = NULL;
Roman Gushchin286e04b2020-08-06 23:20:52 -0700573
Vladimir Davydov2fc04522015-09-09 15:35:28 -0700574 while (memcg && !(memcg->css.flags & CSS_ONLINE))
575 memcg = parent_mem_cgroup(memcg);
576 if (memcg)
577 ino = cgroup_ino(memcg->css.cgroup);
578 rcu_read_unlock();
579 return ino;
580}
581
Mel Gormanef8f2322016-07-28 15:46:05 -0700582static struct mem_cgroup_per_node *
583mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page)
Balbir Singhf64c3f52009-09-23 15:56:37 -0700584{
Johannes Weiner97a6c372011-03-23 16:42:27 -0700585 int nid = page_to_nid(page);
Balbir Singhf64c3f52009-09-23 15:56:37 -0700586
Mel Gormanef8f2322016-07-28 15:46:05 -0700587 return memcg->nodeinfo[nid];
Balbir Singhf64c3f52009-09-23 15:56:37 -0700588}
589
Mel Gormanef8f2322016-07-28 15:46:05 -0700590static struct mem_cgroup_tree_per_node *
591soft_limit_tree_node(int nid)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700592{
Mel Gormanef8f2322016-07-28 15:46:05 -0700593 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700594}
595
Mel Gormanef8f2322016-07-28 15:46:05 -0700596static struct mem_cgroup_tree_per_node *
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700597soft_limit_tree_from_page(struct page *page)
598{
599 int nid = page_to_nid(page);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700600
Mel Gormanef8f2322016-07-28 15:46:05 -0700601 return soft_limit_tree.rb_tree_per_node[nid];
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700602}
603
Mel Gormanef8f2322016-07-28 15:46:05 -0700604static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz,
605 struct mem_cgroup_tree_per_node *mctz,
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800606 unsigned long new_usage_in_excess)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700607{
608 struct rb_node **p = &mctz->rb_root.rb_node;
609 struct rb_node *parent = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -0700610 struct mem_cgroup_per_node *mz_node;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700611 bool rightmost = true;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700612
613 if (mz->on_tree)
614 return;
615
616 mz->usage_in_excess = new_usage_in_excess;
617 if (!mz->usage_in_excess)
618 return;
619 while (*p) {
620 parent = *p;
Mel Gormanef8f2322016-07-28 15:46:05 -0700621 mz_node = rb_entry(parent, struct mem_cgroup_per_node,
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700622 tree_node);
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700623 if (mz->usage_in_excess < mz_node->usage_in_excess) {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700624 p = &(*p)->rb_left;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700625 rightmost = false;
Miaohe Lin378876b2020-12-14 19:06:28 -0800626 } else {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700627 p = &(*p)->rb_right;
Miaohe Lin378876b2020-12-14 19:06:28 -0800628 }
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700629 }
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700630
631 if (rightmost)
632 mctz->rb_rightmost = &mz->tree_node;
633
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700634 rb_link_node(&mz->tree_node, parent, p);
635 rb_insert_color(&mz->tree_node, &mctz->rb_root);
636 mz->on_tree = true;
637}
638
Mel Gormanef8f2322016-07-28 15:46:05 -0700639static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
640 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700641{
642 if (!mz->on_tree)
643 return;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700644
645 if (&mz->tree_node == mctz->rb_rightmost)
646 mctz->rb_rightmost = rb_prev(&mz->tree_node);
647
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700648 rb_erase(&mz->tree_node, &mctz->rb_root);
649 mz->on_tree = false;
650}
651
Mel Gormanef8f2322016-07-28 15:46:05 -0700652static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
653 struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700654{
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700655 unsigned long flags;
656
657 spin_lock_irqsave(&mctz->lock, flags);
Johannes Weinercf2c8122014-06-06 14:38:21 -0700658 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700659 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700660}
661
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800662static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
663{
664 unsigned long nr_pages = page_counter_read(&memcg->memory);
Jason Low4db0c3c2015-04-15 16:14:08 -0700665 unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800666 unsigned long excess = 0;
667
668 if (nr_pages > soft_limit)
669 excess = nr_pages - soft_limit;
670
671 return excess;
672}
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700673
674static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
675{
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800676 unsigned long excess;
Mel Gormanef8f2322016-07-28 15:46:05 -0700677 struct mem_cgroup_per_node *mz;
678 struct mem_cgroup_tree_per_node *mctz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700679
Jianyu Zhane2318752014-06-06 14:38:20 -0700680 mctz = soft_limit_tree_from_page(page);
Laurent Dufourbfc72282017-03-09 16:17:06 -0800681 if (!mctz)
682 return;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700683 /*
684 * Necessary to update all ancestors when hierarchy is used.
685 * because their event counter is not touched.
686 */
687 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700688 mz = mem_cgroup_page_nodeinfo(memcg, page);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800689 excess = soft_limit_excess(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700690 /*
691 * We have to update the tree if mz is on RB-tree or
692 * mem is over its softlimit.
693 */
694 if (excess || mz->on_tree) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700695 unsigned long flags;
696
697 spin_lock_irqsave(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700698 /* if on-tree, remove it */
699 if (mz->on_tree)
Johannes Weinercf2c8122014-06-06 14:38:21 -0700700 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700701 /*
702 * Insert again. mz->usage_in_excess will be updated.
703 * If excess is 0, no tree ops.
704 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700705 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700706 spin_unlock_irqrestore(&mctz->lock, flags);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700707 }
708 }
709}
710
711static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
712{
Mel Gormanef8f2322016-07-28 15:46:05 -0700713 struct mem_cgroup_tree_per_node *mctz;
714 struct mem_cgroup_per_node *mz;
715 int nid;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700716
Jianyu Zhane2318752014-06-06 14:38:20 -0700717 for_each_node(nid) {
Mel Gormanef8f2322016-07-28 15:46:05 -0700718 mz = mem_cgroup_nodeinfo(memcg, nid);
719 mctz = soft_limit_tree_node(nid);
Laurent Dufourbfc72282017-03-09 16:17:06 -0800720 if (mctz)
721 mem_cgroup_remove_exceeded(mz, mctz);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700722 }
723}
724
Mel Gormanef8f2322016-07-28 15:46:05 -0700725static struct mem_cgroup_per_node *
726__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700727{
Mel Gormanef8f2322016-07-28 15:46:05 -0700728 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700729
730retry:
731 mz = NULL;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700732 if (!mctz->rb_rightmost)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700733 goto done; /* Nothing to reclaim from */
734
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -0700735 mz = rb_entry(mctz->rb_rightmost,
736 struct mem_cgroup_per_node, tree_node);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700737 /*
738 * Remove the node now but someone else can add it back,
739 * we will to add it back at the end of reclaim to its correct
740 * position in the tree.
741 */
Johannes Weinercf2c8122014-06-06 14:38:21 -0700742 __mem_cgroup_remove_exceeded(mz, mctz);
Johannes Weiner3e32cb22014-12-10 15:42:31 -0800743 if (!soft_limit_excess(mz->memcg) ||
Shakeel Butt8965aa22020-04-01 21:07:10 -0700744 !css_tryget(&mz->memcg->css))
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700745 goto retry;
746done:
747 return mz;
748}
749
Mel Gormanef8f2322016-07-28 15:46:05 -0700750static struct mem_cgroup_per_node *
751mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700752{
Mel Gormanef8f2322016-07-28 15:46:05 -0700753 struct mem_cgroup_per_node *mz;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700754
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700755 spin_lock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700756 mz = __mem_cgroup_largest_soft_limit_node(mctz);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700757 spin_unlock_irq(&mctz->lock);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700758 return mz;
759}
760
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700761/**
762 * __mod_memcg_state - update cgroup memory statistics
763 * @memcg: the memory cgroup
764 * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item
765 * @val: delta to add to the counter, can be negative
766 */
767void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val)
768{
Roman Gushchinea426c22020-08-06 23:20:35 -0700769 long x, threshold = MEMCG_CHARGE_BATCH;
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700770
771 if (mem_cgroup_disabled())
772 return;
773
Roman Gushchin772616b2020-08-11 18:30:21 -0700774 if (memcg_stat_item_in_bytes(idx))
Roman Gushchinea426c22020-08-06 23:20:35 -0700775 threshold <<= PAGE_SHIFT;
776
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700777 x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]);
Roman Gushchinea426c22020-08-06 23:20:35 -0700778 if (unlikely(abs(x) > threshold)) {
Johannes Weiner42a30032019-05-14 15:47:12 -0700779 struct mem_cgroup *mi;
780
Yafang Shao766a4c12019-07-16 16:26:06 -0700781 /*
782 * Batch local counters to keep them in sync with
783 * the hierarchical ones.
784 */
785 __this_cpu_add(memcg->vmstats_local->stat[idx], x);
Johannes Weiner42a30032019-05-14 15:47:12 -0700786 for (mi = memcg; mi; mi = parent_mem_cgroup(mi))
787 atomic_long_add(x, &mi->vmstats[idx]);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700788 x = 0;
789 }
790 __this_cpu_write(memcg->vmstats_percpu->stat[idx], x);
791}
792
Johannes Weiner42a30032019-05-14 15:47:12 -0700793static struct mem_cgroup_per_node *
794parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid)
795{
796 struct mem_cgroup *parent;
797
798 parent = parent_mem_cgroup(pn->memcg);
799 if (!parent)
800 return NULL;
801 return mem_cgroup_nodeinfo(parent, nid);
802}
803
Roman Gushchineedc4e52020-08-06 23:20:32 -0700804void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
805 int val)
806{
807 struct mem_cgroup_per_node *pn;
808 struct mem_cgroup *memcg;
Roman Gushchinea426c22020-08-06 23:20:35 -0700809 long x, threshold = MEMCG_CHARGE_BATCH;
Roman Gushchineedc4e52020-08-06 23:20:32 -0700810
811 pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
812 memcg = pn->memcg;
813
814 /* Update memcg */
815 __mod_memcg_state(memcg, idx, val);
816
817 /* Update lruvec */
818 __this_cpu_add(pn->lruvec_stat_local->count[idx], val);
819
Roman Gushchinea426c22020-08-06 23:20:35 -0700820 if (vmstat_item_in_bytes(idx))
821 threshold <<= PAGE_SHIFT;
822
Roman Gushchineedc4e52020-08-06 23:20:32 -0700823 x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]);
Roman Gushchinea426c22020-08-06 23:20:35 -0700824 if (unlikely(abs(x) > threshold)) {
Roman Gushchineedc4e52020-08-06 23:20:32 -0700825 pg_data_t *pgdat = lruvec_pgdat(lruvec);
826 struct mem_cgroup_per_node *pi;
827
828 for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id))
829 atomic_long_add(x, &pi->lruvec_stat[idx]);
830 x = 0;
831 }
832 __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x);
833}
834
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700835/**
836 * __mod_lruvec_state - update lruvec memory statistics
837 * @lruvec: the lruvec
838 * @idx: the stat item
839 * @val: delta to add to the counter, can be negative
840 *
841 * The lruvec is the intersection of the NUMA node and a cgroup. This
842 * function updates the all three counters that are affected by a
843 * change of state at this level: per-node, per-cgroup, per-lruvec.
844 */
845void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,
846 int val)
847{
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700848 /* Update node */
Roman Gushchineedc4e52020-08-06 23:20:32 -0700849 __mod_node_page_state(lruvec_pgdat(lruvec), idx, val);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700850
Roman Gushchineedc4e52020-08-06 23:20:32 -0700851 /* Update memcg and lruvec */
852 if (!mem_cgroup_disabled())
853 __mod_memcg_lruvec_state(lruvec, idx, val);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700854}
855
Roman Gushchinec9f0232019-08-13 15:37:41 -0700856void __mod_lruvec_slab_state(void *p, enum node_stat_item idx, int val)
857{
Roman Gushchin4f103c62020-04-01 21:06:36 -0700858 pg_data_t *pgdat = page_pgdat(virt_to_page(p));
Roman Gushchinec9f0232019-08-13 15:37:41 -0700859 struct mem_cgroup *memcg;
860 struct lruvec *lruvec;
861
862 rcu_read_lock();
Roman Gushchin4f103c62020-04-01 21:06:36 -0700863 memcg = mem_cgroup_from_obj(p);
Roman Gushchinec9f0232019-08-13 15:37:41 -0700864
Muchun Song8faeb1f2020-11-21 22:17:12 -0800865 /*
866 * Untracked pages have no memcg, no lruvec. Update only the
867 * node. If we reparent the slab objects to the root memcg,
868 * when we free the slab object, we need to update the per-memcg
869 * vmstats to keep it correct for the root memcg.
870 */
871 if (!memcg) {
Roman Gushchinec9f0232019-08-13 15:37:41 -0700872 __mod_node_page_state(pgdat, idx, val);
873 } else {
Johannes Weiner867e5e12019-11-30 17:55:34 -0800874 lruvec = mem_cgroup_lruvec(memcg, pgdat);
Roman Gushchinec9f0232019-08-13 15:37:41 -0700875 __mod_lruvec_state(lruvec, idx, val);
876 }
877 rcu_read_unlock();
878}
879
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700880/**
881 * __count_memcg_events - account VM events in a cgroup
882 * @memcg: the memory cgroup
883 * @idx: the event item
884 * @count: the number of events that occured
885 */
886void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx,
887 unsigned long count)
888{
889 unsigned long x;
890
891 if (mem_cgroup_disabled())
892 return;
893
894 x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]);
895 if (unlikely(x > MEMCG_CHARGE_BATCH)) {
Johannes Weiner42a30032019-05-14 15:47:12 -0700896 struct mem_cgroup *mi;
897
Yafang Shao766a4c12019-07-16 16:26:06 -0700898 /*
899 * Batch local counters to keep them in sync with
900 * the hierarchical ones.
901 */
902 __this_cpu_add(memcg->vmstats_local->events[idx], x);
Johannes Weiner42a30032019-05-14 15:47:12 -0700903 for (mi = memcg; mi; mi = parent_mem_cgroup(mi))
904 atomic_long_add(x, &mi->vmevents[idx]);
Johannes Weinerdb9adbc2019-05-14 15:47:09 -0700905 x = 0;
906 }
907 __this_cpu_write(memcg->vmstats_percpu->events[idx], x);
908}
909
Johannes Weiner42a30032019-05-14 15:47:12 -0700910static unsigned long memcg_events(struct mem_cgroup *memcg, int event)
Johannes Weinere9f89742011-03-23 16:42:37 -0700911{
Chris Down871789d2019-05-14 15:46:57 -0700912 return atomic_long_read(&memcg->vmevents[event]);
Johannes Weinere9f89742011-03-23 16:42:37 -0700913}
914
Johannes Weiner42a30032019-05-14 15:47:12 -0700915static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event)
916{
Johannes Weiner815744d2019-06-13 15:55:46 -0700917 long x = 0;
918 int cpu;
919
920 for_each_possible_cpu(cpu)
921 x += per_cpu(memcg->vmstats_local->events[event], cpu);
922 return x;
Johannes Weiner42a30032019-05-14 15:47:12 -0700923}
924
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700925static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
David Rientjesb070e652013-05-07 16:18:09 -0700926 struct page *page,
Johannes Weiner3fba69a2020-06-03 16:01:31 -0700927 int nr_pages)
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800928{
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800929 /* pagein of a big page is an event. So, ignore page size */
930 if (nr_pages > 0)
Johannes Weinerc9019e92018-01-31 16:16:37 -0800931 __count_memcg_events(memcg, PGPGIN, 1);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800932 else {
Johannes Weinerc9019e92018-01-31 16:16:37 -0800933 __count_memcg_events(memcg, PGPGOUT, 1);
KAMEZAWA Hiroyuki3751d602011-02-01 15:52:45 -0800934 nr_pages = -nr_pages; /* for event */
935 }
KAMEZAWA Hiroyukie401f172011-01-20 14:44:23 -0800936
Chris Down871789d2019-05-14 15:46:57 -0700937 __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -0800938}
KAMEZAWA Hiroyukid52aa412008-02-07 00:14:24 -0800939
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800940static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
941 enum mem_cgroup_events_target target)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800942{
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700943 unsigned long val, next;
944
Chris Down871789d2019-05-14 15:46:57 -0700945 val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events);
946 next = __this_cpu_read(memcg->vmstats_percpu->targets[target]);
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700947 /* from time_after() in jiffies.h */
Michal Hocko6a1a8b82017-07-10 15:48:53 -0700948 if ((long)(next - val) < 0) {
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800949 switch (target) {
950 case MEM_CGROUP_TARGET_THRESH:
951 next = val + THRESHOLDS_EVENTS_TARGET;
952 break;
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700953 case MEM_CGROUP_TARGET_SOFTLIMIT:
954 next = val + SOFTLIMIT_EVENTS_TARGET;
955 break;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800956 default:
957 break;
958 }
Chris Down871789d2019-05-14 15:46:57 -0700959 __this_cpu_write(memcg->vmstats_percpu->targets[target], next);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800960 return true;
Johannes Weiner7a159cc2011-03-23 16:42:38 -0700961 }
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800962 return false;
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800963}
964
965/*
966 * Check events in order.
967 *
968 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -0700969static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800970{
971 /* threshold event is triggered in finer grain than soft limit */
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800972 if (unlikely(mem_cgroup_event_ratelimit(memcg,
973 MEM_CGROUP_TARGET_THRESH))) {
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700974 bool do_softlimit;
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800975
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700976 do_softlimit = mem_cgroup_event_ratelimit(memcg,
977 MEM_CGROUP_TARGET_SOFTLIMIT);
Johannes Weinerf53d7ce2012-01-12 17:18:23 -0800978 mem_cgroup_threshold(memcg);
Andrew Mortonbb4cc1a82013-09-24 15:27:40 -0700979 if (unlikely(do_softlimit))
980 mem_cgroup_update_tree(memcg, page);
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700981 }
KAMEZAWA Hiroyukid2265e62010-03-10 15:22:31 -0800982}
983
Balbir Singhcf475ad2008-04-29 01:00:16 -0700984struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800985{
Balbir Singh31a78f22008-09-28 23:09:31 +0100986 /*
987 * mm_update_next_owner() may clear mm->owner to NULL
988 * if it races with swapoff, page migration, etc.
989 * So this can be called with p == NULL.
990 */
991 if (unlikely(!p))
992 return NULL;
993
Tejun Heo073219e2014-02-08 10:36:58 -0500994 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800995}
Michal Hocko33398cf2015-09-08 15:01:02 -0700996EXPORT_SYMBOL(mem_cgroup_from_task);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800997
Shakeel Buttd46eb14b2018-08-17 15:46:39 -0700998/**
999 * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg.
1000 * @mm: mm from which memcg should be extracted. It can be NULL.
1001 *
1002 * Obtain a reference on mm->memcg and returns it if successful. Otherwise
1003 * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is
1004 * returned.
1005 */
1006struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001007{
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001008 struct mem_cgroup *memcg;
1009
1010 if (mem_cgroup_disabled())
1011 return NULL;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001012
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001013 rcu_read_lock();
1014 do {
Michal Hocko6f6acb02014-05-22 11:54:19 -07001015 /*
1016 * Page cache insertions can happen withou an
1017 * actual mm context, e.g. during disk probing
1018 * on boot, loopback IO, acct() writes etc.
1019 */
1020 if (unlikely(!mm))
Johannes Weinerdf381972014-04-07 15:37:43 -07001021 memcg = root_mem_cgroup;
Michal Hocko6f6acb02014-05-22 11:54:19 -07001022 else {
1023 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1024 if (unlikely(!memcg))
1025 memcg = root_mem_cgroup;
1026 }
Roman Gushchin00d484f2019-11-15 17:34:43 -08001027 } while (!css_tryget(&memcg->css));
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001028 rcu_read_unlock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001029 return memcg;
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001030}
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001031EXPORT_SYMBOL(get_mem_cgroup_from_mm);
1032
1033/**
Shakeel Buttf745c6f2018-08-17 15:46:44 -07001034 * get_mem_cgroup_from_page: Obtain a reference on given page's memcg.
1035 * @page: page from which memcg should be extracted.
1036 *
1037 * Obtain a reference on page->memcg and returns it if successful. Otherwise
1038 * root_mem_cgroup is returned.
1039 */
1040struct mem_cgroup *get_mem_cgroup_from_page(struct page *page)
1041{
1042 struct mem_cgroup *memcg = page->mem_cgroup;
1043
1044 if (mem_cgroup_disabled())
1045 return NULL;
1046
1047 rcu_read_lock();
Shakeel Butt8965aa22020-04-01 21:07:10 -07001048 /* Page should not get uncharged and freed memcg under us. */
1049 if (!memcg || WARN_ON_ONCE(!css_tryget(&memcg->css)))
Shakeel Buttf745c6f2018-08-17 15:46:44 -07001050 memcg = root_mem_cgroup;
1051 rcu_read_unlock();
1052 return memcg;
1053}
1054EXPORT_SYMBOL(get_mem_cgroup_from_page);
1055
Roman Gushchin37d59852020-10-17 16:13:50 -07001056static __always_inline struct mem_cgroup *active_memcg(void)
1057{
1058 if (in_interrupt())
1059 return this_cpu_read(int_active_memcg);
1060 else
1061 return current->active_memcg;
1062}
1063
1064static __always_inline struct mem_cgroup *get_active_memcg(void)
1065{
1066 struct mem_cgroup *memcg;
1067
1068 rcu_read_lock();
1069 memcg = active_memcg();
1070 if (memcg) {
1071 /* current->active_memcg must hold a ref. */
1072 if (WARN_ON_ONCE(!css_tryget(&memcg->css)))
1073 memcg = root_mem_cgroup;
1074 else
1075 memcg = current->active_memcg;
1076 }
1077 rcu_read_unlock();
1078
1079 return memcg;
1080}
1081
Roman Gushchin4127c652020-10-17 16:13:53 -07001082static __always_inline bool memcg_kmem_bypass(void)
1083{
1084 /* Allow remote memcg charging from any context. */
1085 if (unlikely(active_memcg()))
1086 return false;
1087
1088 /* Memcg to charge can't be determined. */
1089 if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
1090 return true;
1091
1092 return false;
1093}
1094
Shakeel Buttf745c6f2018-08-17 15:46:44 -07001095/**
Roman Gushchin37d59852020-10-17 16:13:50 -07001096 * If active memcg is set, do not fallback to current->mm->memcg.
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001097 */
1098static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void)
1099{
Roman Gushchin279c3392020-10-17 16:13:44 -07001100 if (memcg_kmem_bypass())
1101 return NULL;
1102
Roman Gushchin37d59852020-10-17 16:13:50 -07001103 if (unlikely(active_memcg()))
1104 return get_active_memcg();
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001105
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07001106 return get_mem_cgroup_from_mm(current->mm);
1107}
KAMEZAWA Hiroyuki54595fe2009-01-07 18:08:33 -08001108
Johannes Weiner56600482012-01-12 17:17:59 -08001109/**
1110 * mem_cgroup_iter - iterate over memory cgroup hierarchy
1111 * @root: hierarchy root
1112 * @prev: previously returned memcg, NULL on first invocation
1113 * @reclaim: cookie for shared reclaim walks, NULL for full walks
1114 *
1115 * Returns references to children of the hierarchy below @root, or
1116 * @root itself, or %NULL after a full round-trip.
1117 *
1118 * Caller must pass the return value in @prev on subsequent
1119 * invocations for reference counting, or use mem_cgroup_iter_break()
1120 * to cancel a hierarchy walk before the round-trip is complete.
1121 *
Miaohe Lin05bdc522020-10-13 16:52:45 -07001122 * Reclaimers can specify a node in @reclaim to divide up the memcgs
1123 * in the hierarchy among all concurrent reclaimers operating on the
1124 * same node.
Johannes Weiner56600482012-01-12 17:17:59 -08001125 */
Andrew Morton694fbc02013-09-24 15:27:37 -07001126struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
Johannes Weiner56600482012-01-12 17:17:59 -08001127 struct mem_cgroup *prev,
Andrew Morton694fbc02013-09-24 15:27:37 -07001128 struct mem_cgroup_reclaim_cookie *reclaim)
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07001129{
Kees Cook3f649ab2020-06-03 13:09:38 -07001130 struct mem_cgroup_reclaim_iter *iter;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001131 struct cgroup_subsys_state *css = NULL;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001132 struct mem_cgroup *memcg = NULL;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001133 struct mem_cgroup *pos = NULL;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001134
Andrew Morton694fbc02013-09-24 15:27:37 -07001135 if (mem_cgroup_disabled())
1136 return NULL;
Johannes Weiner56600482012-01-12 17:17:59 -08001137
KAMEZAWA Hiroyuki711d3d22010-10-27 15:33:42 -07001138 if (!root)
1139 root = root_mem_cgroup;
1140
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001141 if (prev && !reclaim)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001142 pos = prev;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001143
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001144 if (!root->use_hierarchy && root != root_mem_cgroup) {
1145 if (prev)
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001146 goto out;
Andrew Morton694fbc02013-09-24 15:27:37 -07001147 return root;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001148 }
1149
Michal Hocko542f85f2013-04-29 15:07:15 -07001150 rcu_read_lock();
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001151
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001152 if (reclaim) {
Mel Gormanef8f2322016-07-28 15:46:05 -07001153 struct mem_cgroup_per_node *mz;
Johannes Weiner527a5ec2012-01-12 17:17:55 -08001154
Mel Gormanef8f2322016-07-28 15:46:05 -07001155 mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id);
Yafang Shao9da83f32019-11-30 17:50:03 -08001156 iter = &mz->iter;
Michal Hocko5f578162013-04-29 15:07:17 -07001157
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001158 if (prev && reclaim->generation != iter->generation)
Michal Hocko542f85f2013-04-29 15:07:15 -07001159 goto out_unlock;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001160
Vladimir Davydov6df38682015-12-29 14:54:10 -08001161 while (1) {
Jason Low4db0c3c2015-04-15 16:14:08 -07001162 pos = READ_ONCE(iter->position);
Vladimir Davydov6df38682015-12-29 14:54:10 -08001163 if (!pos || css_tryget(&pos->css))
1164 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001165 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -08001166 * css reference reached zero, so iter->position will
1167 * be cleared by ->css_released. However, we should not
1168 * rely on this happening soon, because ->css_released
1169 * is called from a work queue, and by busy-waiting we
1170 * might block it. So we clear iter->position right
1171 * away.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001172 */
Vladimir Davydov6df38682015-12-29 14:54:10 -08001173 (void)cmpxchg(&iter->position, pos, NULL);
1174 }
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001175 }
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001176
1177 if (pos)
1178 css = &pos->css;
1179
1180 for (;;) {
1181 css = css_next_descendant_pre(css, &root->css);
1182 if (!css) {
1183 /*
1184 * Reclaimers share the hierarchy walk, and a
1185 * new one might jump in right at the end of
1186 * the hierarchy - make sure they see at least
1187 * one group and restart from the beginning.
1188 */
1189 if (!prev)
1190 continue;
1191 break;
1192 }
1193
1194 /*
1195 * Verify the css and acquire a reference. The root
1196 * is provided by the caller, so we know it's alive
1197 * and kicking, and don't take an extra reference.
1198 */
1199 memcg = mem_cgroup_from_css(css);
1200
1201 if (css == &root->css)
1202 break;
1203
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08001204 if (css_tryget(css))
1205 break;
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001206
1207 memcg = NULL;
1208 }
1209
1210 if (reclaim) {
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001211 /*
Vladimir Davydov6df38682015-12-29 14:54:10 -08001212 * The position could have already been updated by a competing
1213 * thread, so check that the value hasn't changed since we read
1214 * it to avoid reclaiming from the same cgroup twice.
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001215 */
Vladimir Davydov6df38682015-12-29 14:54:10 -08001216 (void)cmpxchg(&iter->position, pos, memcg);
1217
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001218 if (pos)
1219 css_put(&pos->css);
1220
1221 if (!memcg)
1222 iter->generation++;
1223 else if (!prev)
1224 reclaim->generation = iter->generation;
1225 }
1226
Michal Hocko542f85f2013-04-29 15:07:15 -07001227out_unlock:
1228 rcu_read_unlock();
Johannes Weiner5ac8fb32014-12-10 15:42:39 -08001229out:
Michal Hockoc40046f2013-04-29 15:07:14 -07001230 if (prev && prev != root)
1231 css_put(&prev->css);
1232
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001233 return memcg;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001234}
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001235
Johannes Weiner56600482012-01-12 17:17:59 -08001236/**
1237 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1238 * @root: hierarchy root
1239 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1240 */
1241void mem_cgroup_iter_break(struct mem_cgroup *root,
1242 struct mem_cgroup *prev)
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001243{
1244 if (!root)
1245 root = root_mem_cgroup;
1246 if (prev && prev != root)
1247 css_put(&prev->css);
1248}
1249
Miles Chen54a83d62019-08-13 15:37:28 -07001250static void __invalidate_reclaim_iterators(struct mem_cgroup *from,
1251 struct mem_cgroup *dead_memcg)
Vladimir Davydov6df38682015-12-29 14:54:10 -08001252{
Vladimir Davydov6df38682015-12-29 14:54:10 -08001253 struct mem_cgroup_reclaim_iter *iter;
Mel Gormanef8f2322016-07-28 15:46:05 -07001254 struct mem_cgroup_per_node *mz;
1255 int nid;
Vladimir Davydov6df38682015-12-29 14:54:10 -08001256
Miles Chen54a83d62019-08-13 15:37:28 -07001257 for_each_node(nid) {
1258 mz = mem_cgroup_nodeinfo(from, nid);
Yafang Shao9da83f32019-11-30 17:50:03 -08001259 iter = &mz->iter;
1260 cmpxchg(&iter->position, dead_memcg, NULL);
Vladimir Davydov6df38682015-12-29 14:54:10 -08001261 }
1262}
1263
Miles Chen54a83d62019-08-13 15:37:28 -07001264static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
1265{
1266 struct mem_cgroup *memcg = dead_memcg;
1267 struct mem_cgroup *last;
1268
1269 do {
1270 __invalidate_reclaim_iterators(memcg, dead_memcg);
1271 last = memcg;
1272 } while ((memcg = parent_mem_cgroup(memcg)));
1273
1274 /*
1275 * When cgruop1 non-hierarchy mode is used,
1276 * parent_mem_cgroup() does not walk all the way up to the
1277 * cgroup root (root_mem_cgroup). So we have to handle
1278 * dead_memcg from cgroup root separately.
1279 */
1280 if (last != root_mem_cgroup)
1281 __invalidate_reclaim_iterators(root_mem_cgroup,
1282 dead_memcg);
1283}
1284
Johannes Weiner925b7672012-01-12 17:18:15 -08001285/**
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001286 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
1287 * @memcg: hierarchy root
1288 * @fn: function to call for each task
1289 * @arg: argument passed to @fn
1290 *
1291 * This function iterates over tasks attached to @memcg or to any of its
1292 * descendants and calls @fn for each task. If @fn returns a non-zero
1293 * value, the function breaks the iteration loop and returns the value.
1294 * Otherwise, it will iterate over all tasks and return 0.
1295 *
1296 * This function must not be called for the root memory cgroup.
1297 */
1298int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
1299 int (*fn)(struct task_struct *, void *), void *arg)
1300{
1301 struct mem_cgroup *iter;
1302 int ret = 0;
1303
1304 BUG_ON(memcg == root_mem_cgroup);
1305
1306 for_each_mem_cgroup_tree(iter, memcg) {
1307 struct css_task_iter it;
1308 struct task_struct *task;
1309
Tetsuo Handaf168a9a2019-07-11 21:00:20 -07001310 css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001311 while (!ret && (task = css_task_iter_next(&it)))
1312 ret = fn(task, arg);
1313 css_task_iter_end(&it);
1314 if (ret) {
1315 mem_cgroup_iter_break(memcg, iter);
1316 break;
1317 }
1318 }
1319 return ret;
1320}
1321
1322/**
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001323 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
Johannes Weiner925b7672012-01-12 17:18:15 -08001324 * @page: the page
Mike Rapoportf144c392018-02-06 15:42:16 -08001325 * @pgdat: pgdat of the page
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001326 *
Johannes Weinera0b5b412020-06-03 16:02:27 -07001327 * This function relies on page->mem_cgroup being stable - see the
1328 * access rules in commit_charge().
Minchan Kim3f58a822011-03-22 16:32:53 -07001329 */
Mel Gorman599d0c92016-07-28 15:45:31 -07001330struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat)
Minchan Kim3f58a822011-03-22 16:32:53 -07001331{
Mel Gormanef8f2322016-07-28 15:46:05 -07001332 struct mem_cgroup_per_node *mz;
Johannes Weiner925b7672012-01-12 17:18:15 -08001333 struct mem_cgroup *memcg;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001334 struct lruvec *lruvec;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08001335
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001336 if (mem_cgroup_disabled()) {
Johannes Weiner867e5e12019-11-30 17:55:34 -08001337 lruvec = &pgdat->__lruvec;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001338 goto out;
1339 }
Christoph Lameterb69408e2008-10-18 20:26:14 -07001340
Johannes Weiner1306a852014-12-10 15:44:52 -08001341 memcg = page->mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001342 /*
Johannes Weinerdfe0e772014-12-10 15:43:43 -08001343 * Swapcache readahead pages are added to the LRU - and
Johannes Weiner29833312014-12-10 15:44:02 -08001344 * possibly migrated - before they are charged.
Hugh Dickins75121022012-03-05 14:59:18 -08001345 */
Johannes Weiner29833312014-12-10 15:44:02 -08001346 if (!memcg)
1347 memcg = root_mem_cgroup;
Hugh Dickins75121022012-03-05 14:59:18 -08001348
Mel Gormanef8f2322016-07-28 15:46:05 -07001349 mz = mem_cgroup_page_nodeinfo(memcg, page);
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001350 lruvec = &mz->lruvec;
1351out:
1352 /*
1353 * Since a node can be onlined after the mem_cgroup was created,
1354 * we have to be prepared to initialize lruvec->zone here;
1355 * and if offlined then reonlined, we need to reinitialize it.
1356 */
Mel Gorman599d0c92016-07-28 15:45:31 -07001357 if (unlikely(lruvec->pgdat != pgdat))
1358 lruvec->pgdat = pgdat;
Hugh Dickinsbea8c152012-11-16 14:14:54 -08001359 return lruvec;
Johannes Weiner925b7672012-01-12 17:18:15 -08001360}
1361
1362/**
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001363 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1364 * @lruvec: mem_cgroup per zone lru vector
1365 * @lru: index of lru list the page is sitting on
Michal Hockob4536f0c82017-01-10 16:58:04 -08001366 * @zid: zone id of the accounted pages
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001367 * @nr_pages: positive when adding or negative when removing
Johannes Weiner925b7672012-01-12 17:18:15 -08001368 *
Hugh Dickinsca707232016-05-19 17:12:35 -07001369 * This function must be called under lru_lock, just before a page is added
1370 * to or just after a page is removed from an lru list (that ordering being
1371 * so as to allow it to check that lru_size 0 is consistent with list_empty).
Johannes Weiner925b7672012-01-12 17:18:15 -08001372 */
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001373void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
Michal Hockob4536f0c82017-01-10 16:58:04 -08001374 int zid, int nr_pages)
Johannes Weiner925b7672012-01-12 17:18:15 -08001375{
Mel Gormanef8f2322016-07-28 15:46:05 -07001376 struct mem_cgroup_per_node *mz;
Hugh Dickinsfa9add62012-05-29 15:07:09 -07001377 unsigned long *lru_size;
Hugh Dickinsca707232016-05-19 17:12:35 -07001378 long size;
Johannes Weiner925b7672012-01-12 17:18:15 -08001379
1380 if (mem_cgroup_disabled())
1381 return;
1382
Mel Gormanef8f2322016-07-28 15:46:05 -07001383 mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
Michal Hockob4536f0c82017-01-10 16:58:04 -08001384 lru_size = &mz->lru_zone_size[zid][lru];
Hugh Dickinsca707232016-05-19 17:12:35 -07001385
1386 if (nr_pages < 0)
1387 *lru_size += nr_pages;
1388
1389 size = *lru_size;
Michal Hockob4536f0c82017-01-10 16:58:04 -08001390 if (WARN_ONCE(size < 0,
1391 "%s(%p, %d, %d): lru_size %ld\n",
1392 __func__, lruvec, lru, nr_pages, size)) {
Hugh Dickinsca707232016-05-19 17:12:35 -07001393 VM_BUG_ON(1);
1394 *lru_size = 0;
1395 }
1396
1397 if (nr_pages > 0)
1398 *lru_size += nr_pages;
KAMEZAWA Hiroyuki08e552c2009-01-07 18:08:01 -08001399}
KAMEZAWA Hiroyuki544122e2009-01-07 18:08:34 -08001400
Johannes Weiner19942822011-02-01 15:52:43 -08001401/**
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001402 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
Wanpeng Lidad75572012-06-20 12:53:01 -07001403 * @memcg: the memory cgroup
Johannes Weiner19942822011-02-01 15:52:43 -08001404 *
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001405 * Returns the maximum amount of memory @mem can be charged with, in
Johannes Weiner7ec99d62011-03-23 16:42:36 -07001406 * pages.
Johannes Weiner19942822011-02-01 15:52:43 -08001407 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001408static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
Johannes Weiner19942822011-02-01 15:52:43 -08001409{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001410 unsigned long margin = 0;
1411 unsigned long count;
1412 unsigned long limit;
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001413
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001414 count = page_counter_read(&memcg->memory);
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001415 limit = READ_ONCE(memcg->memory.max);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001416 if (count < limit)
1417 margin = limit - count;
1418
Johannes Weiner7941d212016-01-14 15:21:23 -08001419 if (do_memsw_account()) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001420 count = page_counter_read(&memcg->memsw);
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001421 limit = READ_ONCE(memcg->memsw.max);
Kaixu Xia1c4448e2020-06-01 21:49:36 -07001422 if (count < limit)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001423 margin = min(margin, limit - count);
Li RongQingcbedbac2016-05-27 14:27:43 -07001424 else
1425 margin = 0;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001426 }
1427
1428 return margin;
Johannes Weiner19942822011-02-01 15:52:43 -08001429}
1430
KAMEZAWA Hiroyuki619d0942012-03-21 16:34:23 -07001431/*
Qiang Huangbdcbb652014-06-04 16:08:21 -07001432 * A routine for checking "mem" is under move_account() or not.
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001433 *
Qiang Huangbdcbb652014-06-04 16:08:21 -07001434 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1435 * moving cgroups. This is for waiting at high-memory pressure
1436 * caused by "move".
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07001437 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001438static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001439{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001440 struct mem_cgroup *from;
1441 struct mem_cgroup *to;
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001442 bool ret = false;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001443 /*
1444 * Unlike task_move routines, we access mc.to, mc.from not under
1445 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1446 */
1447 spin_lock(&mc.lock);
1448 from = mc.from;
1449 to = mc.to;
1450 if (!from)
1451 goto unlock;
Michal Hocko3e920412011-07-26 16:08:29 -07001452
Johannes Weiner2314b422014-12-10 15:44:33 -08001453 ret = mem_cgroup_is_descendant(from, memcg) ||
1454 mem_cgroup_is_descendant(to, memcg);
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07001455unlock:
1456 spin_unlock(&mc.lock);
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001457 return ret;
1458}
1459
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001460static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001461{
1462 if (mc.moving_task && current != mc.moving_task) {
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001463 if (mem_cgroup_under_move(memcg)) {
KAMEZAWA Hiroyuki4b534332010-08-10 18:02:57 -07001464 DEFINE_WAIT(wait);
1465 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1466 /* moving charge context might have finished. */
1467 if (mc.moving_task)
1468 schedule();
1469 finish_wait(&mc.waitq, &wait);
1470 return true;
1471 }
1472 }
1473 return false;
1474}
1475
Muchun Song5f9a4f42020-10-13 16:52:59 -07001476struct memory_stat {
1477 const char *name;
1478 unsigned int ratio;
1479 unsigned int idx;
1480};
1481
1482static struct memory_stat memory_stats[] = {
1483 { "anon", PAGE_SIZE, NR_ANON_MAPPED },
1484 { "file", PAGE_SIZE, NR_FILE_PAGES },
1485 { "kernel_stack", 1024, NR_KERNEL_STACK_KB },
1486 { "percpu", 1, MEMCG_PERCPU_B },
1487 { "sock", PAGE_SIZE, MEMCG_SOCK },
1488 { "shmem", PAGE_SIZE, NR_SHMEM },
1489 { "file_mapped", PAGE_SIZE, NR_FILE_MAPPED },
1490 { "file_dirty", PAGE_SIZE, NR_FILE_DIRTY },
1491 { "file_writeback", PAGE_SIZE, NR_WRITEBACK },
1492#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1493 /*
1494 * The ratio will be initialized in memory_stats_init(). Because
1495 * on some architectures, the macro of HPAGE_PMD_SIZE is not
1496 * constant(e.g. powerpc).
1497 */
1498 { "anon_thp", 0, NR_ANON_THPS },
Johannes Weinerb8eddff2020-12-14 19:06:20 -08001499 { "file_thp", 0, NR_FILE_THPS },
1500 { "shmem_thp", 0, NR_SHMEM_THPS },
Muchun Song5f9a4f42020-10-13 16:52:59 -07001501#endif
1502 { "inactive_anon", PAGE_SIZE, NR_INACTIVE_ANON },
1503 { "active_anon", PAGE_SIZE, NR_ACTIVE_ANON },
1504 { "inactive_file", PAGE_SIZE, NR_INACTIVE_FILE },
1505 { "active_file", PAGE_SIZE, NR_ACTIVE_FILE },
1506 { "unevictable", PAGE_SIZE, NR_UNEVICTABLE },
1507
1508 /*
1509 * Note: The slab_reclaimable and slab_unreclaimable must be
1510 * together and slab_reclaimable must be in front.
1511 */
1512 { "slab_reclaimable", 1, NR_SLAB_RECLAIMABLE_B },
1513 { "slab_unreclaimable", 1, NR_SLAB_UNRECLAIMABLE_B },
1514
1515 /* The memory events */
1516 { "workingset_refault_anon", 1, WORKINGSET_REFAULT_ANON },
1517 { "workingset_refault_file", 1, WORKINGSET_REFAULT_FILE },
1518 { "workingset_activate_anon", 1, WORKINGSET_ACTIVATE_ANON },
1519 { "workingset_activate_file", 1, WORKINGSET_ACTIVATE_FILE },
1520 { "workingset_restore_anon", 1, WORKINGSET_RESTORE_ANON },
1521 { "workingset_restore_file", 1, WORKINGSET_RESTORE_FILE },
1522 { "workingset_nodereclaim", 1, WORKINGSET_NODERECLAIM },
1523};
1524
1525static int __init memory_stats_init(void)
1526{
1527 int i;
1528
1529 for (i = 0; i < ARRAY_SIZE(memory_stats); i++) {
1530#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Johannes Weinerb8eddff2020-12-14 19:06:20 -08001531 if (memory_stats[i].idx == NR_ANON_THPS ||
1532 memory_stats[i].idx == NR_FILE_THPS ||
1533 memory_stats[i].idx == NR_SHMEM_THPS)
Muchun Song5f9a4f42020-10-13 16:52:59 -07001534 memory_stats[i].ratio = HPAGE_PMD_SIZE;
1535#endif
1536 VM_BUG_ON(!memory_stats[i].ratio);
1537 VM_BUG_ON(memory_stats[i].idx >= MEMCG_NR_STAT);
1538 }
1539
1540 return 0;
1541}
1542pure_initcall(memory_stats_init);
1543
Johannes Weinerc8713d02019-07-11 20:55:59 -07001544static char *memory_stat_format(struct mem_cgroup *memcg)
1545{
1546 struct seq_buf s;
1547 int i;
Johannes Weiner71cd3112017-05-03 14:55:13 -07001548
Johannes Weinerc8713d02019-07-11 20:55:59 -07001549 seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE);
1550 if (!s.buffer)
1551 return NULL;
1552
1553 /*
1554 * Provide statistics on the state of the memory subsystem as
1555 * well as cumulative event counters that show past behavior.
1556 *
1557 * This list is ordered following a combination of these gradients:
1558 * 1) generic big picture -> specifics and details
1559 * 2) reflecting userspace activity -> reflecting kernel heuristics
1560 *
1561 * Current memory state:
1562 */
1563
Muchun Song5f9a4f42020-10-13 16:52:59 -07001564 for (i = 0; i < ARRAY_SIZE(memory_stats); i++) {
1565 u64 size;
Johannes Weinerc8713d02019-07-11 20:55:59 -07001566
Muchun Song5f9a4f42020-10-13 16:52:59 -07001567 size = memcg_page_state(memcg, memory_stats[i].idx);
1568 size *= memory_stats[i].ratio;
1569 seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size);
Johannes Weinerc8713d02019-07-11 20:55:59 -07001570
Muchun Song5f9a4f42020-10-13 16:52:59 -07001571 if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) {
1572 size = memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B) +
1573 memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B);
1574 seq_buf_printf(&s, "slab %llu\n", size);
1575 }
1576 }
Johannes Weinerc8713d02019-07-11 20:55:59 -07001577
1578 /* Accumulated memory events */
1579
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001580 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT),
1581 memcg_events(memcg, PGFAULT));
1582 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT),
1583 memcg_events(memcg, PGMAJFAULT));
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001584 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL),
1585 memcg_events(memcg, PGREFILL));
Johannes Weinerc8713d02019-07-11 20:55:59 -07001586 seq_buf_printf(&s, "pgscan %lu\n",
1587 memcg_events(memcg, PGSCAN_KSWAPD) +
1588 memcg_events(memcg, PGSCAN_DIRECT));
1589 seq_buf_printf(&s, "pgsteal %lu\n",
1590 memcg_events(memcg, PGSTEAL_KSWAPD) +
1591 memcg_events(memcg, PGSTEAL_DIRECT));
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001592 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE),
1593 memcg_events(memcg, PGACTIVATE));
1594 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE),
1595 memcg_events(memcg, PGDEACTIVATE));
1596 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE),
1597 memcg_events(memcg, PGLAZYFREE));
1598 seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED),
1599 memcg_events(memcg, PGLAZYFREED));
Johannes Weinerc8713d02019-07-11 20:55:59 -07001600
1601#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001602 seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC),
Johannes Weinerc8713d02019-07-11 20:55:59 -07001603 memcg_events(memcg, THP_FAULT_ALLOC));
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08001604 seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC),
Johannes Weinerc8713d02019-07-11 20:55:59 -07001605 memcg_events(memcg, THP_COLLAPSE_ALLOC));
1606#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1607
1608 /* The above should easily fit into one page */
1609 WARN_ON_ONCE(seq_buf_has_overflowed(&s));
1610
1611 return s.buffer;
1612}
Johannes Weiner71cd3112017-05-03 14:55:13 -07001613
Sha Zhengju58cf1882013-02-22 16:32:05 -08001614#define K(x) ((x) << (PAGE_SHIFT-10))
Balbir Singhe2224322009-04-02 16:57:39 -07001615/**
yuzhoujianf0c867d2018-12-28 00:36:10 -08001616 * mem_cgroup_print_oom_context: Print OOM information relevant to
1617 * memory controller.
Balbir Singhe2224322009-04-02 16:57:39 -07001618 * @memcg: The memory cgroup that went over limit
1619 * @p: Task that is going to be killed
1620 *
1621 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1622 * enabled
1623 */
yuzhoujianf0c867d2018-12-28 00:36:10 -08001624void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p)
1625{
1626 rcu_read_lock();
1627
1628 if (memcg) {
1629 pr_cont(",oom_memcg=");
1630 pr_cont_cgroup_path(memcg->css.cgroup);
1631 } else
1632 pr_cont(",global_oom");
1633 if (p) {
1634 pr_cont(",task_memcg=");
1635 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1636 }
1637 rcu_read_unlock();
1638}
1639
1640/**
1641 * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to
1642 * memory controller.
1643 * @memcg: The memory cgroup that went over limit
1644 */
1645void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg)
Balbir Singhe2224322009-04-02 16:57:39 -07001646{
Johannes Weinerc8713d02019-07-11 20:55:59 -07001647 char *buf;
Balbir Singhe2224322009-04-02 16:57:39 -07001648
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001649 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1650 K((u64)page_counter_read(&memcg->memory)),
Chris Down15b42562020-04-01 21:07:20 -07001651 K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt);
Johannes Weinerc8713d02019-07-11 20:55:59 -07001652 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
1653 pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n",
1654 K((u64)page_counter_read(&memcg->swap)),
Chris Down32d087c2020-04-01 21:07:30 -07001655 K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt);
Johannes Weinerc8713d02019-07-11 20:55:59 -07001656 else {
1657 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1658 K((u64)page_counter_read(&memcg->memsw)),
1659 K((u64)memcg->memsw.max), memcg->memsw.failcnt);
1660 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1661 K((u64)page_counter_read(&memcg->kmem)),
1662 K((u64)memcg->kmem.max), memcg->kmem.failcnt);
Sha Zhengju58cf1882013-02-22 16:32:05 -08001663 }
Johannes Weinerc8713d02019-07-11 20:55:59 -07001664
1665 pr_info("Memory cgroup stats for ");
1666 pr_cont_cgroup_path(memcg->css.cgroup);
1667 pr_cont(":");
1668 buf = memory_stat_format(memcg);
1669 if (!buf)
1670 return;
1671 pr_info("%s", buf);
1672 kfree(buf);
Balbir Singhe2224322009-04-02 16:57:39 -07001673}
1674
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07001675/*
David Rientjesa63d83f2010-08-09 17:19:46 -07001676 * Return the memory (and swap, if configured) limit for a memcg.
1677 */
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001678unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg)
David Rientjesa63d83f2010-08-09 17:19:46 -07001679{
Waiman Long8d387a52020-10-13 16:52:52 -07001680 unsigned long max = READ_ONCE(memcg->memory.max);
David Rientjesa63d83f2010-08-09 17:19:46 -07001681
Waiman Long8d387a52020-10-13 16:52:52 -07001682 if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
1683 if (mem_cgroup_swappiness(memcg))
1684 max += min(READ_ONCE(memcg->swap.max),
1685 (unsigned long)total_swap_pages);
1686 } else { /* v1 */
1687 if (mem_cgroup_swappiness(memcg)) {
1688 /* Calculate swap excess capacity from memsw limit */
1689 unsigned long swap = READ_ONCE(memcg->memsw.max) - max;
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001690
Waiman Long8d387a52020-10-13 16:52:52 -07001691 max += min(swap, (unsigned long)total_swap_pages);
1692 }
Michal Hocko9a5a8f12012-11-16 14:14:49 -08001693 }
Roman Gushchinbbec2e12018-06-07 17:06:18 -07001694 return max;
David Rientjesa63d83f2010-08-09 17:19:46 -07001695}
1696
Chris Down9783aa92019-10-06 17:58:32 -07001697unsigned long mem_cgroup_size(struct mem_cgroup *memcg)
1698{
1699 return page_counter_read(&memcg->memory);
1700}
1701
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07001702static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
David Rientjes19965462012-12-11 16:00:26 -08001703 int order)
David Rientjes9cbb78b2012-07-31 16:43:44 -07001704{
David Rientjes6e0fc462015-09-08 15:00:36 -07001705 struct oom_control oc = {
1706 .zonelist = NULL,
1707 .nodemask = NULL,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001708 .memcg = memcg,
David Rientjes6e0fc462015-09-08 15:00:36 -07001709 .gfp_mask = gfp_mask,
1710 .order = order,
David Rientjes6e0fc462015-09-08 15:00:36 -07001711 };
Yafang Shao1378b372020-08-06 23:22:08 -07001712 bool ret = true;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001713
Tetsuo Handa7775fac2019-03-05 15:46:47 -08001714 if (mutex_lock_killable(&oom_lock))
1715 return true;
Yafang Shao1378b372020-08-06 23:22:08 -07001716
1717 if (mem_cgroup_margin(memcg) >= (1 << order))
1718 goto unlock;
1719
Tetsuo Handa7775fac2019-03-05 15:46:47 -08001720 /*
1721 * A few threads which were not waiting at mutex_lock_killable() can
1722 * fail to bail out. Therefore, check again after holding oom_lock.
1723 */
1724 ret = should_force_charge() || out_of_memory(&oc);
Yafang Shao1378b372020-08-06 23:22:08 -07001725
1726unlock:
Johannes Weinerdc564012015-06-24 16:57:19 -07001727 mutex_unlock(&oom_lock);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001728 return ret;
David Rientjes9cbb78b2012-07-31 16:43:44 -07001729}
1730
Andrew Morton0608f432013-09-24 15:27:41 -07001731static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
Mel Gormanef8f2322016-07-28 15:46:05 -07001732 pg_data_t *pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001733 gfp_t gfp_mask,
1734 unsigned long *total_scanned)
Balbir Singh6d61ef42009-01-07 18:08:06 -08001735{
Andrew Morton0608f432013-09-24 15:27:41 -07001736 struct mem_cgroup *victim = NULL;
1737 int total = 0;
1738 int loop = 0;
1739 unsigned long excess;
1740 unsigned long nr_scanned;
1741 struct mem_cgroup_reclaim_cookie reclaim = {
Mel Gormanef8f2322016-07-28 15:46:05 -07001742 .pgdat = pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07001743 };
Johannes Weiner9d11ea92011-03-23 16:42:21 -07001744
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001745 excess = soft_limit_excess(root_memcg);
Balbir Singh6d61ef42009-01-07 18:08:06 -08001746
Andrew Morton0608f432013-09-24 15:27:41 -07001747 while (1) {
1748 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1749 if (!victim) {
1750 loop++;
1751 if (loop >= 2) {
1752 /*
1753 * If we have not been able to reclaim
1754 * anything, it might because there are
1755 * no reclaimable pages under this hierarchy
1756 */
1757 if (!total)
1758 break;
1759 /*
1760 * We want to do more targeted reclaim.
1761 * excess >> 2 is not to excessive so as to
1762 * reclaim too much, nor too less that we keep
1763 * coming back to reclaim from this cgroup
1764 */
1765 if (total >= (excess >> 2) ||
1766 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1767 break;
1768 }
1769 continue;
1770 }
Mel Gormana9dd0a82016-07-28 15:46:02 -07001771 total += mem_cgroup_shrink_node(victim, gfp_mask, false,
Mel Gormanef8f2322016-07-28 15:46:05 -07001772 pgdat, &nr_scanned);
Andrew Morton0608f432013-09-24 15:27:41 -07001773 *total_scanned += nr_scanned;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08001774 if (!soft_limit_excess(root_memcg))
Andrew Morton0608f432013-09-24 15:27:41 -07001775 break;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001776 }
Andrew Morton0608f432013-09-24 15:27:41 -07001777 mem_cgroup_iter_break(root_memcg, victim);
1778 return total;
Balbir Singh6d61ef42009-01-07 18:08:06 -08001779}
1780
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001781#ifdef CONFIG_LOCKDEP
1782static struct lockdep_map memcg_oom_lock_dep_map = {
1783 .name = "memcg_oom_lock",
1784};
1785#endif
1786
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001787static DEFINE_SPINLOCK(memcg_oom_lock);
1788
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001789/*
1790 * Check OOM-Killer is already running under our hierarchy.
1791 * If someone is running, return false.
1792 */
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001793static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001794{
Michal Hocko79dfdac2011-07-26 16:08:23 -07001795 struct mem_cgroup *iter, *failed = NULL;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001796
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001797 spin_lock(&memcg_oom_lock);
1798
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001799 for_each_mem_cgroup_tree(iter, memcg) {
Johannes Weiner23751be2011-08-25 15:59:16 -07001800 if (iter->oom_lock) {
Michal Hocko79dfdac2011-07-26 16:08:23 -07001801 /*
1802 * this subtree of our hierarchy is already locked
1803 * so we cannot give a lock.
1804 */
Michal Hocko79dfdac2011-07-26 16:08:23 -07001805 failed = iter;
Johannes Weiner9f3a0d02012-01-12 17:17:48 -08001806 mem_cgroup_iter_break(memcg, iter);
1807 break;
Johannes Weiner23751be2011-08-25 15:59:16 -07001808 } else
1809 iter->oom_lock = true;
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001810 }
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001811
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001812 if (failed) {
1813 /*
1814 * OK, we failed to lock the whole subtree so we have
1815 * to clean up what we set up to the failing subtree
1816 */
1817 for_each_mem_cgroup_tree(iter, memcg) {
1818 if (iter == failed) {
1819 mem_cgroup_iter_break(memcg, iter);
1820 break;
1821 }
1822 iter->oom_lock = false;
Michal Hocko79dfdac2011-07-26 16:08:23 -07001823 }
Johannes Weiner0056f4e2013-10-31 16:34:14 -07001824 } else
1825 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001826
1827 spin_unlock(&memcg_oom_lock);
1828
1829 return !failed;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08001830}
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001831
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001832static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001833{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07001834 struct mem_cgroup *iter;
1835
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001836 spin_lock(&memcg_oom_lock);
Qian Cai5facae42019-09-19 12:09:40 -04001837 mutex_release(&memcg_oom_lock_dep_map, _RET_IP_);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001838 for_each_mem_cgroup_tree(iter, memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001839 iter->oom_lock = false;
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07001840 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001841}
1842
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001843static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001844{
1845 struct mem_cgroup *iter;
1846
Tejun Heoc2b42d32015-06-24 16:58:23 -07001847 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001848 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001849 iter->under_oom++;
1850 spin_unlock(&memcg_oom_lock);
Michal Hocko79dfdac2011-07-26 16:08:23 -07001851}
1852
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001853static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
Michal Hocko79dfdac2011-07-26 16:08:23 -07001854{
1855 struct mem_cgroup *iter;
1856
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001857 /*
Miaohe Lin7a52d4d82020-10-13 16:53:05 -07001858 * Be careful about under_oom underflows becase a child memcg
1859 * could have been added after mem_cgroup_mark_under_oom.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001860 */
Tejun Heoc2b42d32015-06-24 16:58:23 -07001861 spin_lock(&memcg_oom_lock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001862 for_each_mem_cgroup_tree(iter, memcg)
Tejun Heoc2b42d32015-06-24 16:58:23 -07001863 if (iter->under_oom > 0)
1864 iter->under_oom--;
1865 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07001866}
1867
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001868static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1869
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001870struct oom_wait_info {
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001871 struct mem_cgroup *memcg;
Ingo Molnarac6424b2017-06-20 12:06:13 +02001872 wait_queue_entry_t wait;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001873};
1874
Ingo Molnarac6424b2017-06-20 12:06:13 +02001875static int memcg_oom_wake_function(wait_queue_entry_t *wait,
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001876 unsigned mode, int sync, void *arg)
1877{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001878 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1879 struct mem_cgroup *oom_wait_memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001880 struct oom_wait_info *oom_wait_info;
1881
1882 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07001883 oom_wait_memcg = oom_wait_info->memcg;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001884
Johannes Weiner2314b422014-12-10 15:44:33 -08001885 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1886 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001887 return 0;
KAMEZAWA Hiroyukidc98df52010-05-26 14:42:36 -07001888 return autoremove_wake_function(wait, mode, sync, arg);
1889}
1890
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07001891static void memcg_oom_recover(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001892{
Tejun Heoc2b42d32015-06-24 16:58:23 -07001893 /*
1894 * For the following lockless ->under_oom test, the only required
1895 * guarantee is that it must see the state asserted by an OOM when
1896 * this function is called as a result of userland actions
1897 * triggered by the notification of the OOM. This is trivially
1898 * achieved by invoking mem_cgroup_mark_under_oom() before
1899 * triggering notification.
1900 */
1901 if (memcg && memcg->under_oom)
Tejun Heof4b90b702015-06-24 16:58:21 -07001902 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07001903}
1904
Michal Hocko29ef6802018-08-17 15:47:11 -07001905enum oom_status {
1906 OOM_SUCCESS,
1907 OOM_FAILED,
1908 OOM_ASYNC,
1909 OOM_SKIPPED
1910};
1911
1912static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001913{
Michal Hocko7056d3a2018-12-28 00:39:57 -08001914 enum oom_status ret;
1915 bool locked;
1916
Michal Hocko29ef6802018-08-17 15:47:11 -07001917 if (order > PAGE_ALLOC_COSTLY_ORDER)
1918 return OOM_SKIPPED;
1919
Roman Gushchin7a1adfd2018-10-26 15:09:48 -07001920 memcg_memory_event(memcg, MEMCG_OOM);
1921
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001922 /*
Johannes Weiner49426422013-10-16 13:46:59 -07001923 * We are in the middle of the charge context here, so we
1924 * don't want to block when potentially sitting on a callstack
1925 * that holds all kinds of filesystem and mm locks.
1926 *
Michal Hocko29ef6802018-08-17 15:47:11 -07001927 * cgroup1 allows disabling the OOM killer and waiting for outside
1928 * handling until the charge can succeed; remember the context and put
1929 * the task to sleep at the end of the page fault when all locks are
1930 * released.
Johannes Weiner49426422013-10-16 13:46:59 -07001931 *
Michal Hocko29ef6802018-08-17 15:47:11 -07001932 * On the other hand, in-kernel OOM killer allows for an async victim
1933 * memory reclaim (oom_reaper) and that means that we are not solely
1934 * relying on the oom victim to make a forward progress and we can
1935 * invoke the oom killer here.
1936 *
1937 * Please note that mem_cgroup_out_of_memory might fail to find a
1938 * victim and then we have to bail out from the charge path.
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08001939 */
Michal Hocko29ef6802018-08-17 15:47:11 -07001940 if (memcg->oom_kill_disable) {
1941 if (!current->in_user_fault)
1942 return OOM_SKIPPED;
1943 css_get(&memcg->css);
1944 current->memcg_in_oom = memcg;
1945 current->memcg_oom_gfp_mask = mask;
1946 current->memcg_oom_order = order;
1947
1948 return OOM_ASYNC;
1949 }
1950
Michal Hocko7056d3a2018-12-28 00:39:57 -08001951 mem_cgroup_mark_under_oom(memcg);
Michal Hocko29ef6802018-08-17 15:47:11 -07001952
Michal Hocko7056d3a2018-12-28 00:39:57 -08001953 locked = mem_cgroup_oom_trylock(memcg);
1954
1955 if (locked)
1956 mem_cgroup_oom_notify(memcg);
1957
1958 mem_cgroup_unmark_under_oom(memcg);
1959 if (mem_cgroup_out_of_memory(memcg, mask, order))
1960 ret = OOM_SUCCESS;
1961 else
1962 ret = OOM_FAILED;
1963
1964 if (locked)
1965 mem_cgroup_oom_unlock(memcg);
1966
1967 return ret;
Johannes Weiner49426422013-10-16 13:46:59 -07001968}
1969
1970/**
1971 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1972 * @handle: actually kill/wait or just clean up the OOM state
1973 *
1974 * This has to be called at the end of a page fault if the memcg OOM
1975 * handler was enabled.
1976 *
1977 * Memcg supports userspace OOM handling where failed allocations must
1978 * sleep on a waitqueue until the userspace task resolves the
1979 * situation. Sleeping directly in the charge context with all kinds
1980 * of locks held is not a good idea, instead we remember an OOM state
1981 * in the task and mem_cgroup_oom_synchronize() has to be called at
1982 * the end of the page fault to complete the OOM handling.
1983 *
1984 * Returns %true if an ongoing memcg OOM situation was detected and
1985 * completed, %false otherwise.
1986 */
1987bool mem_cgroup_oom_synchronize(bool handle)
1988{
Tejun Heo626ebc42015-11-05 18:46:09 -08001989 struct mem_cgroup *memcg = current->memcg_in_oom;
Johannes Weiner49426422013-10-16 13:46:59 -07001990 struct oom_wait_info owait;
1991 bool locked;
1992
1993 /* OOM is global, do not handle */
1994 if (!memcg)
1995 return false;
1996
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001997 if (!handle)
Johannes Weiner49426422013-10-16 13:46:59 -07001998 goto cleanup;
1999
2000 owait.memcg = memcg;
2001 owait.wait.flags = 0;
2002 owait.wait.func = memcg_oom_wake_function;
2003 owait.wait.private = current;
Ingo Molnar2055da92017-06-20 12:06:46 +02002004 INIT_LIST_HEAD(&owait.wait.entry);
Johannes Weiner49426422013-10-16 13:46:59 -07002005
2006 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002007 mem_cgroup_mark_under_oom(memcg);
2008
2009 locked = mem_cgroup_oom_trylock(memcg);
2010
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002011 if (locked)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002012 mem_cgroup_oom_notify(memcg);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002013
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002014 if (locked && !memcg->oom_kill_disable) {
2015 mem_cgroup_unmark_under_oom(memcg);
Johannes Weiner49426422013-10-16 13:46:59 -07002016 finish_wait(&memcg_oom_waitq, &owait.wait);
Tejun Heo626ebc42015-11-05 18:46:09 -08002017 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
2018 current->memcg_oom_order);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07002019 } else {
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002020 schedule();
Johannes Weiner49426422013-10-16 13:46:59 -07002021 mem_cgroup_unmark_under_oom(memcg);
2022 finish_wait(&memcg_oom_waitq, &owait.wait);
2023 }
2024
2025 if (locked) {
Johannes Weinerfb2a6fc2013-09-12 15:13:43 -07002026 mem_cgroup_oom_unlock(memcg);
2027 /*
2028 * There is no guarantee that an OOM-lock contender
2029 * sees the wakeups triggered by the OOM kill
2030 * uncharges. Wake any sleepers explicitely.
2031 */
2032 memcg_oom_recover(memcg);
2033 }
Johannes Weiner49426422013-10-16 13:46:59 -07002034cleanup:
Tejun Heo626ebc42015-11-05 18:46:09 -08002035 current->memcg_in_oom = NULL;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07002036 css_put(&memcg->css);
KAMEZAWA Hiroyuki867578c2010-03-10 15:22:39 -08002037 return true;
KAMEZAWA Hiroyuki0b7f5692009-04-02 16:57:38 -07002038}
2039
Johannes Weinerd7365e72014-10-29 14:50:48 -07002040/**
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07002041 * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM
2042 * @victim: task to be killed by the OOM killer
2043 * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM
2044 *
2045 * Returns a pointer to a memory cgroup, which has to be cleaned up
2046 * by killing all belonging OOM-killable tasks.
2047 *
2048 * Caller has to call mem_cgroup_put() on the returned non-NULL memcg.
2049 */
2050struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim,
2051 struct mem_cgroup *oom_domain)
2052{
2053 struct mem_cgroup *oom_group = NULL;
2054 struct mem_cgroup *memcg;
2055
2056 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2057 return NULL;
2058
2059 if (!oom_domain)
2060 oom_domain = root_mem_cgroup;
2061
2062 rcu_read_lock();
2063
2064 memcg = mem_cgroup_from_task(victim);
2065 if (memcg == root_mem_cgroup)
2066 goto out;
2067
2068 /*
Roman Gushchin48fe2672020-04-01 21:07:39 -07002069 * If the victim task has been asynchronously moved to a different
2070 * memory cgroup, we might end up killing tasks outside oom_domain.
2071 * In this case it's better to ignore memory.group.oom.
2072 */
2073 if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain)))
2074 goto out;
2075
2076 /*
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07002077 * Traverse the memory cgroup hierarchy from the victim task's
2078 * cgroup up to the OOMing cgroup (or root) to find the
2079 * highest-level memory cgroup with oom.group set.
2080 */
2081 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
2082 if (memcg->oom_group)
2083 oom_group = memcg;
2084
2085 if (memcg == oom_domain)
2086 break;
2087 }
2088
2089 if (oom_group)
2090 css_get(&oom_group->css);
2091out:
2092 rcu_read_unlock();
2093
2094 return oom_group;
2095}
2096
2097void mem_cgroup_print_oom_group(struct mem_cgroup *memcg)
2098{
2099 pr_info("Tasks in ");
2100 pr_cont_cgroup_path(memcg->css.cgroup);
2101 pr_cont(" are going to be killed due to memory.oom.group set\n");
2102}
2103
2104/**
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002105 * lock_page_memcg - lock a page->mem_cgroup binding
2106 * @page: the page
KAMEZAWA Hiroyuki32047e22010-10-27 15:33:40 -07002107 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002108 * This function protects unlocked LRU pages from being moved to
Johannes Weiner739f79f2017-08-18 15:15:48 -07002109 * another cgroup.
2110 *
2111 * It ensures lifetime of the returned memcg. Caller is responsible
2112 * for the lifetime of the page; __unlock_page_memcg() is available
2113 * when @page might get freed inside the locked section.
Balbir Singhd69b0422009-06-17 16:26:34 -07002114 */
Johannes Weiner739f79f2017-08-18 15:15:48 -07002115struct mem_cgroup *lock_page_memcg(struct page *page)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002116{
Johannes Weiner9da7b522020-06-03 16:01:51 -07002117 struct page *head = compound_head(page); /* rmap on tail pages */
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002118 struct mem_cgroup *memcg;
Johannes Weiner6de22612015-02-11 15:25:01 -08002119 unsigned long flags;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002120
Johannes Weiner6de22612015-02-11 15:25:01 -08002121 /*
2122 * The RCU lock is held throughout the transaction. The fast
2123 * path can get away without acquiring the memcg->move_lock
2124 * because page moving starts with an RCU grace period.
Johannes Weiner739f79f2017-08-18 15:15:48 -07002125 *
2126 * The RCU lock also protects the memcg from being freed when
2127 * the page state that is going to change is the only thing
2128 * preventing the page itself from being freed. E.g. writeback
2129 * doesn't hold a page reference and relies on PG_writeback to
2130 * keep off truncation, migration and so forth.
2131 */
Johannes Weinerd7365e72014-10-29 14:50:48 -07002132 rcu_read_lock();
2133
2134 if (mem_cgroup_disabled())
Johannes Weiner739f79f2017-08-18 15:15:48 -07002135 return NULL;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002136again:
Johannes Weiner9da7b522020-06-03 16:01:51 -07002137 memcg = head->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08002138 if (unlikely(!memcg))
Johannes Weiner739f79f2017-08-18 15:15:48 -07002139 return NULL;
Johannes Weinerd7365e72014-10-29 14:50:48 -07002140
Qiang Huangbdcbb652014-06-04 16:08:21 -07002141 if (atomic_read(&memcg->moving_account) <= 0)
Johannes Weiner739f79f2017-08-18 15:15:48 -07002142 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002143
Johannes Weiner6de22612015-02-11 15:25:01 -08002144 spin_lock_irqsave(&memcg->move_lock, flags);
Johannes Weiner9da7b522020-06-03 16:01:51 -07002145 if (memcg != head->mem_cgroup) {
Johannes Weiner6de22612015-02-11 15:25:01 -08002146 spin_unlock_irqrestore(&memcg->move_lock, flags);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002147 goto again;
2148 }
Johannes Weiner6de22612015-02-11 15:25:01 -08002149
2150 /*
2151 * When charge migration first begins, we can have locked and
2152 * unlocked page stat updates happening concurrently. Track
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002153 * the task who has the lock for unlock_page_memcg().
Johannes Weiner6de22612015-02-11 15:25:01 -08002154 */
2155 memcg->move_lock_task = current;
2156 memcg->move_lock_flags = flags;
Johannes Weinerd7365e72014-10-29 14:50:48 -07002157
Johannes Weiner739f79f2017-08-18 15:15:48 -07002158 return memcg;
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002159}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002160EXPORT_SYMBOL(lock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002161
Johannes Weinerd7365e72014-10-29 14:50:48 -07002162/**
Johannes Weiner739f79f2017-08-18 15:15:48 -07002163 * __unlock_page_memcg - unlock and unpin a memcg
2164 * @memcg: the memcg
2165 *
2166 * Unlock and unpin a memcg returned by lock_page_memcg().
Johannes Weinerd7365e72014-10-29 14:50:48 -07002167 */
Johannes Weiner739f79f2017-08-18 15:15:48 -07002168void __unlock_page_memcg(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002169{
Johannes Weiner6de22612015-02-11 15:25:01 -08002170 if (memcg && memcg->move_lock_task == current) {
2171 unsigned long flags = memcg->move_lock_flags;
2172
2173 memcg->move_lock_task = NULL;
2174 memcg->move_lock_flags = 0;
2175
2176 spin_unlock_irqrestore(&memcg->move_lock, flags);
2177 }
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002178
Johannes Weinerd7365e72014-10-29 14:50:48 -07002179 rcu_read_unlock();
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002180}
Johannes Weiner739f79f2017-08-18 15:15:48 -07002181
2182/**
2183 * unlock_page_memcg - unlock a page->mem_cgroup binding
2184 * @page: the page
2185 */
2186void unlock_page_memcg(struct page *page)
2187{
Johannes Weiner9da7b522020-06-03 16:01:51 -07002188 struct page *head = compound_head(page);
2189
2190 __unlock_page_memcg(head->mem_cgroup);
Johannes Weiner739f79f2017-08-18 15:15:48 -07002191}
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07002192EXPORT_SYMBOL(unlock_page_memcg);
KAMEZAWA Hiroyuki89c06bd2012-03-21 16:34:25 -07002193
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002194struct memcg_stock_pcp {
2195 struct mem_cgroup *cached; /* this never be root cgroup */
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002196 unsigned int nr_pages;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002197
2198#ifdef CONFIG_MEMCG_KMEM
2199 struct obj_cgroup *cached_objcg;
2200 unsigned int nr_bytes;
2201#endif
2202
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002203 struct work_struct work;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002204 unsigned long flags;
Kirill A. Shutemova0db00f2012-05-29 15:06:56 -07002205#define FLUSHING_CACHED_CHARGE 0
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002206};
2207static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
Michal Hocko9f50fad2011-08-09 11:56:26 +02002208static DEFINE_MUTEX(percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002209
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002210#ifdef CONFIG_MEMCG_KMEM
2211static void drain_obj_stock(struct memcg_stock_pcp *stock);
2212static bool obj_stock_flush_required(struct memcg_stock_pcp *stock,
2213 struct mem_cgroup *root_memcg);
2214
2215#else
2216static inline void drain_obj_stock(struct memcg_stock_pcp *stock)
2217{
2218}
2219static bool obj_stock_flush_required(struct memcg_stock_pcp *stock,
2220 struct mem_cgroup *root_memcg)
2221{
2222 return false;
2223}
2224#endif
2225
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002226/**
2227 * consume_stock: Try to consume stocked charge on this cpu.
2228 * @memcg: memcg to consume from.
2229 * @nr_pages: how many pages to charge.
2230 *
2231 * The charges will only happen if @memcg matches the current cpu's memcg
2232 * stock, and at least @nr_pages are available in that stock. Failure to
2233 * service an allocation will refill the stock.
2234 *
2235 * returns true if successful, false otherwise.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002236 */
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002237static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002238{
2239 struct memcg_stock_pcp *stock;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002240 unsigned long flags;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002241 bool ret = false;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002242
Johannes Weinera983b5e2018-01-31 16:16:45 -08002243 if (nr_pages > MEMCG_CHARGE_BATCH)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002244 return ret;
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002245
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002246 local_irq_save(flags);
2247
2248 stock = this_cpu_ptr(&memcg_stock);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002249 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
Suleiman Souhlala0956d52012-12-18 14:21:36 -08002250 stock->nr_pages -= nr_pages;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002251 ret = true;
2252 }
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002253
2254 local_irq_restore(flags);
2255
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002256 return ret;
2257}
2258
2259/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002260 * Returns stocks cached in percpu and reset cached information.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002261 */
2262static void drain_stock(struct memcg_stock_pcp *stock)
2263{
2264 struct mem_cgroup *old = stock->cached;
2265
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07002266 if (!old)
2267 return;
2268
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002269 if (stock->nr_pages) {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002270 page_counter_uncharge(&old->memory, stock->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002271 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002272 page_counter_uncharge(&old->memsw, stock->nr_pages);
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002273 stock->nr_pages = 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002274 }
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07002275
2276 css_put(&old->css);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002277 stock->cached = NULL;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002278}
2279
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002280static void drain_local_stock(struct work_struct *dummy)
2281{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002282 struct memcg_stock_pcp *stock;
2283 unsigned long flags;
2284
Michal Hocko72f01842017-10-03 16:14:53 -07002285 /*
2286 * The only protection from memory hotplug vs. drain_stock races is
2287 * that we always operate on local CPU stock here with IRQ disabled
2288 */
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002289 local_irq_save(flags);
2290
2291 stock = this_cpu_ptr(&memcg_stock);
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002292 drain_obj_stock(stock);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002293 drain_stock(stock);
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002294 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002295
2296 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002297}
2298
2299/*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002300 * Cache charges(val) to local per_cpu area.
Greg Thelen320cc512010-03-15 15:27:28 +01002301 * This will be consumed by consume_stock() function, later.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002302 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002303static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002304{
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002305 struct memcg_stock_pcp *stock;
2306 unsigned long flags;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002307
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002308 local_irq_save(flags);
2309
2310 stock = this_cpu_ptr(&memcg_stock);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002311 if (stock->cached != memcg) { /* reset if necessary */
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002312 drain_stock(stock);
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07002313 css_get(&memcg->css);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002314 stock->cached = memcg;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002315 }
Johannes Weiner11c9ea42011-03-23 16:42:34 -07002316 stock->nr_pages += nr_pages;
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002317
Johannes Weinera983b5e2018-01-31 16:16:45 -08002318 if (stock->nr_pages > MEMCG_CHARGE_BATCH)
Roman Gushchin475d0482017-09-08 16:13:09 -07002319 drain_stock(stock);
2320
Johannes Weinerdb2ba40c2016-09-19 14:44:36 -07002321 local_irq_restore(flags);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002322}
2323
2324/*
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002325 * Drains all per-CPU charge caches for given root_memcg resp. subtree
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002326 * of the hierarchy under it.
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002327 */
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002328static void drain_all_stock(struct mem_cgroup *root_memcg)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002329{
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002330 int cpu, curcpu;
Michal Hockod38144b2011-07-26 16:08:28 -07002331
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002332 /* If someone's already draining, avoid adding running more workers. */
2333 if (!mutex_trylock(&percpu_charge_mutex))
2334 return;
Michal Hocko72f01842017-10-03 16:14:53 -07002335 /*
2336 * Notify other cpus that system-wide "drain" is running
2337 * We do not care about races with the cpu hotplug because cpu down
2338 * as well as workers from this path always operate on the local
2339 * per-cpu data. CPU up doesn't touch memcg_stock at all.
2340 */
Johannes Weiner5af12d02011-08-25 15:59:07 -07002341 curcpu = get_cpu();
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002342 for_each_online_cpu(cpu) {
2343 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002344 struct mem_cgroup *memcg;
Roman Gushchine1a366b2019-09-23 15:34:58 -07002345 bool flush = false;
KAMEZAWA Hiroyuki26fe6162011-06-15 15:08:45 -07002346
Roman Gushchine1a366b2019-09-23 15:34:58 -07002347 rcu_read_lock();
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07002348 memcg = stock->cached;
Roman Gushchine1a366b2019-09-23 15:34:58 -07002349 if (memcg && stock->nr_pages &&
2350 mem_cgroup_is_descendant(memcg, root_memcg))
2351 flush = true;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002352 if (obj_stock_flush_required(stock, root_memcg))
2353 flush = true;
Roman Gushchine1a366b2019-09-23 15:34:58 -07002354 rcu_read_unlock();
2355
2356 if (flush &&
2357 !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
Michal Hockod1a05b62011-07-26 16:08:27 -07002358 if (cpu == curcpu)
2359 drain_local_stock(&stock->work);
2360 else
2361 schedule_work_on(cpu, &stock->work);
2362 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002363 }
Johannes Weiner5af12d02011-08-25 15:59:07 -07002364 put_cpu();
Michal Hocko9f50fad2011-08-09 11:56:26 +02002365 mutex_unlock(&percpu_charge_mutex);
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002366}
2367
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01002368static int memcg_hotplug_cpu_dead(unsigned int cpu)
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002369{
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002370 struct memcg_stock_pcp *stock;
Johannes Weiner42a30032019-05-14 15:47:12 -07002371 struct mem_cgroup *memcg, *mi;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002372
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002373 stock = &per_cpu(memcg_stock, cpu);
2374 drain_stock(stock);
Johannes Weinera983b5e2018-01-31 16:16:45 -08002375
2376 for_each_mem_cgroup(memcg) {
2377 int i;
2378
2379 for (i = 0; i < MEMCG_NR_STAT; i++) {
2380 int nid;
2381 long x;
2382
Chris Down871789d2019-05-14 15:46:57 -07002383 x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0);
Johannes Weiner815744d2019-06-13 15:55:46 -07002384 if (x)
Johannes Weiner42a30032019-05-14 15:47:12 -07002385 for (mi = memcg; mi; mi = parent_mem_cgroup(mi))
2386 atomic_long_add(x, &memcg->vmstats[i]);
Johannes Weinera983b5e2018-01-31 16:16:45 -08002387
2388 if (i >= NR_VM_NODE_STAT_ITEMS)
2389 continue;
2390
2391 for_each_node(nid) {
2392 struct mem_cgroup_per_node *pn;
2393
2394 pn = mem_cgroup_nodeinfo(memcg, nid);
2395 x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0);
Johannes Weiner815744d2019-06-13 15:55:46 -07002396 if (x)
Johannes Weiner42a30032019-05-14 15:47:12 -07002397 do {
2398 atomic_long_add(x, &pn->lruvec_stat[i]);
2399 } while ((pn = parent_nodeinfo(pn, nid)));
Johannes Weinera983b5e2018-01-31 16:16:45 -08002400 }
2401 }
2402
Johannes Weinere27be242018-04-10 16:29:45 -07002403 for (i = 0; i < NR_VM_EVENT_ITEMS; i++) {
Johannes Weinera983b5e2018-01-31 16:16:45 -08002404 long x;
2405
Chris Down871789d2019-05-14 15:46:57 -07002406 x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0);
Johannes Weiner815744d2019-06-13 15:55:46 -07002407 if (x)
Johannes Weiner42a30032019-05-14 15:47:12 -07002408 for (mi = memcg; mi; mi = parent_mem_cgroup(mi))
2409 atomic_long_add(x, &memcg->vmevents[i]);
Johannes Weinera983b5e2018-01-31 16:16:45 -08002410 }
2411 }
2412
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01002413 return 0;
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002414}
2415
Chris Downb3ff9292020-08-06 23:21:54 -07002416static unsigned long reclaim_high(struct mem_cgroup *memcg,
2417 unsigned int nr_pages,
2418 gfp_t gfp_mask)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002419{
Chris Downb3ff9292020-08-06 23:21:54 -07002420 unsigned long nr_reclaimed = 0;
2421
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002422 do {
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002423 unsigned long pflags;
2424
Jakub Kicinskid1663a92020-06-01 21:49:49 -07002425 if (page_counter_read(&memcg->memory) <=
2426 READ_ONCE(memcg->memory.high))
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002427 continue;
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002428
Johannes Weinere27be242018-04-10 16:29:45 -07002429 memcg_memory_event(memcg, MEMCG_HIGH);
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002430
2431 psi_memstall_enter(&pflags);
Chris Downb3ff9292020-08-06 23:21:54 -07002432 nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages,
2433 gfp_mask, true);
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002434 psi_memstall_leave(&pflags);
Chris Down4bf17302020-04-06 20:03:30 -07002435 } while ((memcg = parent_mem_cgroup(memcg)) &&
2436 !mem_cgroup_is_root(memcg));
Chris Downb3ff9292020-08-06 23:21:54 -07002437
2438 return nr_reclaimed;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002439}
2440
2441static void high_work_func(struct work_struct *work)
2442{
2443 struct mem_cgroup *memcg;
2444
2445 memcg = container_of(work, struct mem_cgroup, high_work);
Johannes Weinera983b5e2018-01-31 16:16:45 -08002446 reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002447}
2448
Tejun Heob23afb92015-11-05 18:46:11 -08002449/*
Chris Down0e4b01d2019-09-23 15:34:55 -07002450 * Clamp the maximum sleep time per allocation batch to 2 seconds. This is
2451 * enough to still cause a significant slowdown in most cases, while still
2452 * allowing diagnostics and tracing to proceed without becoming stuck.
2453 */
2454#define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ)
2455
2456/*
2457 * When calculating the delay, we use these either side of the exponentiation to
2458 * maintain precision and scale to a reasonable number of jiffies (see the table
2459 * below.
2460 *
2461 * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the
2462 * overage ratio to a delay.
Randy Dunlapac5ddd02020-08-11 18:33:02 -07002463 * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the
Chris Down0e4b01d2019-09-23 15:34:55 -07002464 * proposed penalty in order to reduce to a reasonable number of jiffies, and
2465 * to produce a reasonable delay curve.
2466 *
2467 * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a
2468 * reasonable delay curve compared to precision-adjusted overage, not
2469 * penalising heavily at first, but still making sure that growth beyond the
2470 * limit penalises misbehaviour cgroups by slowing them down exponentially. For
2471 * example, with a high of 100 megabytes:
2472 *
2473 * +-------+------------------------+
2474 * | usage | time to allocate in ms |
2475 * +-------+------------------------+
2476 * | 100M | 0 |
2477 * | 101M | 6 |
2478 * | 102M | 25 |
2479 * | 103M | 57 |
2480 * | 104M | 102 |
2481 * | 105M | 159 |
2482 * | 106M | 230 |
2483 * | 107M | 313 |
2484 * | 108M | 409 |
2485 * | 109M | 518 |
2486 * | 110M | 639 |
2487 * | 111M | 774 |
2488 * | 112M | 921 |
2489 * | 113M | 1081 |
2490 * | 114M | 1254 |
2491 * | 115M | 1439 |
2492 * | 116M | 1638 |
2493 * | 117M | 1849 |
2494 * | 118M | 2000 |
2495 * | 119M | 2000 |
2496 * | 120M | 2000 |
2497 * +-------+------------------------+
2498 */
2499 #define MEMCG_DELAY_PRECISION_SHIFT 20
2500 #define MEMCG_DELAY_SCALING_SHIFT 14
2501
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002502static u64 calculate_overage(unsigned long usage, unsigned long high)
2503{
2504 u64 overage;
2505
2506 if (usage <= high)
2507 return 0;
2508
2509 /*
2510 * Prevent division by 0 in overage calculation by acting as if
2511 * it was a threshold of 1 page
2512 */
2513 high = max(high, 1UL);
2514
2515 overage = usage - high;
2516 overage <<= MEMCG_DELAY_PRECISION_SHIFT;
2517 return div64_u64(overage, high);
2518}
2519
2520static u64 mem_find_max_overage(struct mem_cgroup *memcg)
2521{
2522 u64 overage, max_overage = 0;
2523
2524 do {
2525 overage = calculate_overage(page_counter_read(&memcg->memory),
Jakub Kicinskid1663a92020-06-01 21:49:49 -07002526 READ_ONCE(memcg->memory.high));
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002527 max_overage = max(overage, max_overage);
2528 } while ((memcg = parent_mem_cgroup(memcg)) &&
2529 !mem_cgroup_is_root(memcg));
2530
2531 return max_overage;
2532}
2533
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002534static u64 swap_find_max_overage(struct mem_cgroup *memcg)
2535{
2536 u64 overage, max_overage = 0;
2537
2538 do {
2539 overage = calculate_overage(page_counter_read(&memcg->swap),
2540 READ_ONCE(memcg->swap.high));
2541 if (overage)
2542 memcg_memory_event(memcg, MEMCG_SWAP_HIGH);
2543 max_overage = max(overage, max_overage);
2544 } while ((memcg = parent_mem_cgroup(memcg)) &&
2545 !mem_cgroup_is_root(memcg));
2546
2547 return max_overage;
2548}
2549
Chris Down0e4b01d2019-09-23 15:34:55 -07002550/*
Chris Downe26733e2020-03-21 18:22:23 -07002551 * Get the number of jiffies that we should penalise a mischievous cgroup which
2552 * is exceeding its memory.high by checking both it and its ancestors.
Tejun Heob23afb92015-11-05 18:46:11 -08002553 */
Chris Downe26733e2020-03-21 18:22:23 -07002554static unsigned long calculate_high_delay(struct mem_cgroup *memcg,
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002555 unsigned int nr_pages,
2556 u64 max_overage)
Tejun Heob23afb92015-11-05 18:46:11 -08002557{
Chris Downe26733e2020-03-21 18:22:23 -07002558 unsigned long penalty_jiffies;
Chris Downe26733e2020-03-21 18:22:23 -07002559
2560 if (!max_overage)
2561 return 0;
Chris Down0e4b01d2019-09-23 15:34:55 -07002562
2563 /*
Chris Down0e4b01d2019-09-23 15:34:55 -07002564 * We use overage compared to memory.high to calculate the number of
2565 * jiffies to sleep (penalty_jiffies). Ideally this value should be
2566 * fairly lenient on small overages, and increasingly harsh when the
2567 * memcg in question makes it clear that it has no intention of stopping
2568 * its crazy behaviour, so we exponentially increase the delay based on
2569 * overage amount.
2570 */
Chris Downe26733e2020-03-21 18:22:23 -07002571 penalty_jiffies = max_overage * max_overage * HZ;
2572 penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT;
2573 penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT;
Chris Down0e4b01d2019-09-23 15:34:55 -07002574
2575 /*
2576 * Factor in the task's own contribution to the overage, such that four
2577 * N-sized allocations are throttled approximately the same as one
2578 * 4N-sized allocation.
2579 *
2580 * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or
2581 * larger the current charge patch is than that.
2582 */
Jakub Kicinskiff144e62020-06-01 21:49:45 -07002583 return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH;
Chris Downe26733e2020-03-21 18:22:23 -07002584}
2585
2586/*
2587 * Scheduled by try_charge() to be executed from the userland return path
2588 * and reclaims memory over the high limit.
2589 */
2590void mem_cgroup_handle_over_high(void)
2591{
2592 unsigned long penalty_jiffies;
2593 unsigned long pflags;
Chris Downb3ff9292020-08-06 23:21:54 -07002594 unsigned long nr_reclaimed;
Chris Downe26733e2020-03-21 18:22:23 -07002595 unsigned int nr_pages = current->memcg_nr_pages_over_high;
Chris Downd977aa92020-08-06 23:21:58 -07002596 int nr_retries = MAX_RECLAIM_RETRIES;
Chris Downe26733e2020-03-21 18:22:23 -07002597 struct mem_cgroup *memcg;
Chris Downb3ff9292020-08-06 23:21:54 -07002598 bool in_retry = false;
Chris Downe26733e2020-03-21 18:22:23 -07002599
2600 if (likely(!nr_pages))
2601 return;
2602
2603 memcg = get_mem_cgroup_from_mm(current->mm);
Chris Downe26733e2020-03-21 18:22:23 -07002604 current->memcg_nr_pages_over_high = 0;
2605
Chris Downb3ff9292020-08-06 23:21:54 -07002606retry_reclaim:
2607 /*
2608 * The allocating task should reclaim at least the batch size, but for
2609 * subsequent retries we only want to do what's necessary to prevent oom
2610 * or breaching resource isolation.
2611 *
2612 * This is distinct from memory.max or page allocator behaviour because
2613 * memory.high is currently batched, whereas memory.max and the page
2614 * allocator run every time an allocation is made.
2615 */
2616 nr_reclaimed = reclaim_high(memcg,
2617 in_retry ? SWAP_CLUSTER_MAX : nr_pages,
2618 GFP_KERNEL);
2619
Chris Downe26733e2020-03-21 18:22:23 -07002620 /*
2621 * memory.high is breached and reclaim is unable to keep up. Throttle
2622 * allocators proactively to slow down excessive growth.
2623 */
Jakub Kicinski8a5dbc62020-06-01 21:49:42 -07002624 penalty_jiffies = calculate_high_delay(memcg, nr_pages,
2625 mem_find_max_overage(memcg));
Chris Down0e4b01d2019-09-23 15:34:55 -07002626
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002627 penalty_jiffies += calculate_high_delay(memcg, nr_pages,
2628 swap_find_max_overage(memcg));
2629
Chris Down0e4b01d2019-09-23 15:34:55 -07002630 /*
Jakub Kicinskiff144e62020-06-01 21:49:45 -07002631 * Clamp the max delay per usermode return so as to still keep the
2632 * application moving forwards and also permit diagnostics, albeit
2633 * extremely slowly.
2634 */
2635 penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES);
2636
2637 /*
Chris Down0e4b01d2019-09-23 15:34:55 -07002638 * Don't sleep if the amount of jiffies this memcg owes us is so low
2639 * that it's not even worth doing, in an attempt to be nice to those who
2640 * go only a small amount over their memory.high value and maybe haven't
2641 * been aggressively reclaimed enough yet.
2642 */
2643 if (penalty_jiffies <= HZ / 100)
2644 goto out;
2645
2646 /*
Chris Downb3ff9292020-08-06 23:21:54 -07002647 * If reclaim is making forward progress but we're still over
2648 * memory.high, we want to encourage that rather than doing allocator
2649 * throttling.
2650 */
2651 if (nr_reclaimed || nr_retries--) {
2652 in_retry = true;
2653 goto retry_reclaim;
2654 }
2655
2656 /*
Chris Down0e4b01d2019-09-23 15:34:55 -07002657 * If we exit early, we're guaranteed to die (since
2658 * schedule_timeout_killable sets TASK_KILLABLE). This means we don't
2659 * need to account for any ill-begotten jiffies to pay them off later.
2660 */
2661 psi_memstall_enter(&pflags);
2662 schedule_timeout_killable(penalty_jiffies);
2663 psi_memstall_leave(&pflags);
2664
2665out:
2666 css_put(&memcg->css);
Tejun Heob23afb92015-11-05 18:46:11 -08002667}
2668
Johannes Weiner00501b52014-08-08 14:19:20 -07002669static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
2670 unsigned int nr_pages)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002671{
Johannes Weinera983b5e2018-01-31 16:16:45 -08002672 unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages);
Chris Downd977aa92020-08-06 23:21:58 -07002673 int nr_retries = MAX_RECLAIM_RETRIES;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002674 struct mem_cgroup *mem_over_limit;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002675 struct page_counter *counter;
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002676 enum oom_status oom_status;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002677 unsigned long nr_reclaimed;
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002678 bool may_swap = true;
2679 bool drained = false;
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002680 unsigned long pflags;
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -08002681
Johannes Weinerce00a962014-09-05 08:43:57 -04002682 if (mem_cgroup_is_root(memcg))
Tejun Heo10d53c72015-11-05 18:46:17 -08002683 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002684retry:
Michal Hockob6b6cc72014-04-07 15:37:44 -07002685 if (consume_stock(memcg, nr_pages))
Tejun Heo10d53c72015-11-05 18:46:17 -08002686 return 0;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002687
Johannes Weiner7941d212016-01-14 15:21:23 -08002688 if (!do_memsw_account() ||
Johannes Weiner6071ca52015-11-05 18:50:26 -08002689 page_counter_try_charge(&memcg->memsw, batch, &counter)) {
2690 if (page_counter_try_charge(&memcg->memory, batch, &counter))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002691 goto done_restock;
Johannes Weiner7941d212016-01-14 15:21:23 -08002692 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002693 page_counter_uncharge(&memcg->memsw, batch);
2694 mem_over_limit = mem_cgroup_from_counter(counter, memory);
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002695 } else {
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002696 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002697 may_swap = false;
Johannes Weiner3fbe7242014-10-09 15:28:54 -07002698 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002699
Johannes Weiner6539cc02014-08-06 16:05:42 -07002700 if (batch > nr_pages) {
2701 batch = nr_pages;
2702 goto retry;
2703 }
KAMEZAWA Hiroyukicdec2e42009-12-15 16:47:08 -08002704
Johannes Weiner06b078f2014-08-06 16:05:44 -07002705 /*
Johannes Weiner869712f2019-11-05 21:17:13 -08002706 * Memcg doesn't have a dedicated reserve for atomic
2707 * allocations. But like the global atomic pool, we need to
2708 * put the burden of reclaim on regular allocation requests
2709 * and let these go through as privileged allocations.
2710 */
2711 if (gfp_mask & __GFP_ATOMIC)
2712 goto force;
2713
2714 /*
Johannes Weiner06b078f2014-08-06 16:05:44 -07002715 * Unlike in global OOM situations, memcg is not in a physical
2716 * memory shortage. Allow dying and OOM-killed tasks to
2717 * bypass the last charges so that they can exit quickly and
2718 * free their memory.
2719 */
Tetsuo Handa7775fac2019-03-05 15:46:47 -08002720 if (unlikely(should_force_charge()))
Tejun Heo10d53c72015-11-05 18:46:17 -08002721 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07002722
Johannes Weiner89a28482016-10-27 17:46:56 -07002723 /*
2724 * 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 Weiner06b078f2014-08-06 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);
Johannes Weiner241994ed2015-02-11 15:26:06 -08002739
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002740 psi_memstall_enter(&pflags);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002741 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
2742 gfp_mask, may_swap);
Johannes Weinere22c6ed2020-08-06 23:22:15 -07002743 psi_memstall_leave(&pflags);
Johannes Weiner6539cc02014-08-06 16:05:42 -07002744
Johannes Weiner61e02c72014-08-06 16:08:16 -07002745 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
Johannes Weiner6539cc02014-08-06 16:05:42 -07002746 goto retry;
Johannes Weiner28c34c22014-08-06 16:05:47 -07002747
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002748 if (!drained) {
Johannes Weiner6d3d6aa2014-12-10 15:42:50 -08002749 drain_all_stock(mem_over_limit);
Johannes Weinerb70a2a22014-10-09 15:28:56 -07002750 drained = true;
2751 goto retry;
2752 }
2753
Johannes Weiner28c34c22014-08-06 16:05:47 -07002754 if (gfp_mask & __GFP_NORETRY)
2755 goto nomem;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002756 /*
2757 * Even though the limit is exceeded at this point, reclaim
2758 * may have been able to free some pages. Retry the charge
2759 * before killing the task.
2760 *
2761 * Only for regular pages, though: huge pages are rather
2762 * unlikely to succeed so close to the limit, and we fall back
2763 * to regular pages anyway in case of failure.
2764 */
Johannes Weiner61e02c72014-08-06 16:08:16 -07002765 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
Johannes Weiner6539cc02014-08-06 16:05:42 -07002766 goto retry;
2767 /*
2768 * At task move, charge accounts can be doubly counted. So, it's
2769 * better to wait until the end of task_move if something is going on.
2770 */
2771 if (mem_cgroup_wait_acct_move(mem_over_limit))
2772 goto retry;
2773
Johannes Weiner9b130612014-08-06 16:05:51 -07002774 if (nr_retries--)
2775 goto retry;
2776
Shakeel Butt38d38492019-07-11 20:55:48 -07002777 if (gfp_mask & __GFP_RETRY_MAYFAIL)
Michal Hocko29ef6802018-08-17 15:47:11 -07002778 goto nomem;
2779
Johannes Weiner06b078f2014-08-06 16:05:44 -07002780 if (gfp_mask & __GFP_NOFAIL)
Tejun Heo10d53c72015-11-05 18:46:17 -08002781 goto force;
Johannes Weiner06b078f2014-08-06 16:05:44 -07002782
Johannes Weiner6539cc02014-08-06 16:05:42 -07002783 if (fatal_signal_pending(current))
Tejun Heo10d53c72015-11-05 18:46:17 -08002784 goto force;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002785
Michal Hocko29ef6802018-08-17 15:47:11 -07002786 /*
2787 * keep retrying as long as the memcg oom killer is able to make
2788 * a forward progress or bypass the charge if the oom killer
2789 * couldn't make any progress.
2790 */
2791 oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask,
Jerome Marchand3608de02015-11-05 18:47:29 -08002792 get_order(nr_pages * PAGE_SIZE));
Michal Hocko29ef6802018-08-17 15:47:11 -07002793 switch (oom_status) {
2794 case OOM_SUCCESS:
Chris Downd977aa92020-08-06 23:21:58 -07002795 nr_retries = MAX_RECLAIM_RETRIES;
Michal Hocko29ef6802018-08-17 15:47:11 -07002796 goto retry;
2797 case OOM_FAILED:
2798 goto force;
2799 default:
2800 goto nomem;
2801 }
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002802nomem:
Johannes Weiner6d1fdc42014-04-07 15:37:45 -07002803 if (!(gfp_mask & __GFP_NOFAIL))
Johannes Weiner3168ecb2013-10-31 16:34:13 -07002804 return -ENOMEM;
Tejun Heo10d53c72015-11-05 18:46:17 -08002805force:
2806 /*
2807 * The allocation either can't fail or will lead to more memory
2808 * being freed very soon. Allow memory usage go over the limit
2809 * temporarily by force charging it.
2810 */
2811 page_counter_charge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002812 if (do_memsw_account())
Tejun Heo10d53c72015-11-05 18:46:17 -08002813 page_counter_charge(&memcg->memsw, nr_pages);
Tejun Heo10d53c72015-11-05 18:46:17 -08002814
2815 return 0;
Johannes Weiner6539cc02014-08-06 16:05:42 -07002816
2817done_restock:
2818 if (batch > nr_pages)
2819 refill_stock(memcg, batch - nr_pages);
Tejun Heob23afb92015-11-05 18:46:11 -08002820
Johannes Weiner241994ed2015-02-11 15:26:06 -08002821 /*
Tejun Heob23afb92015-11-05 18:46:11 -08002822 * If the hierarchy is above the normal consumption range, schedule
2823 * reclaim on returning to userland. We can perform reclaim here
Mel Gorman71baba42015-11-06 16:28:28 -08002824 * if __GFP_RECLAIM but let's always punt for simplicity and so that
Tejun Heob23afb92015-11-05 18:46:11 -08002825 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2826 * not recorded as it most likely matches current's and won't
2827 * change in the meantime. As high limit is checked again before
2828 * reclaim, the cost of mismatch is negligible.
Johannes Weiner241994ed2015-02-11 15:26:06 -08002829 */
2830 do {
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002831 bool mem_high, swap_high;
2832
2833 mem_high = page_counter_read(&memcg->memory) >
2834 READ_ONCE(memcg->memory.high);
2835 swap_high = page_counter_read(&memcg->swap) >
2836 READ_ONCE(memcg->swap.high);
2837
2838 /* Don't bother a random interrupted task */
2839 if (in_interrupt()) {
2840 if (mem_high) {
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08002841 schedule_work(&memcg->high_work);
2842 break;
2843 }
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07002844 continue;
2845 }
2846
2847 if (mem_high || swap_high) {
2848 /*
2849 * The allocating tasks in this cgroup will need to do
2850 * reclaim or be throttled to prevent further growth
2851 * of the memory or swap footprints.
2852 *
2853 * Target some best-effort fairness between the tasks,
2854 * and distribute reclaim work and delay penalties
2855 * based on how much each task is actually allocating.
2856 */
Vladimir Davydov9516a182015-12-11 13:40:24 -08002857 current->memcg_nr_pages_over_high += batch;
Tejun Heob23afb92015-11-05 18:46:11 -08002858 set_notify_resume(current);
2859 break;
2860 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08002861 } while ((memcg = parent_mem_cgroup(memcg)));
Tejun Heo10d53c72015-11-05 18:46:17 -08002862
2863 return 0;
KAMEZAWA Hiroyuki7a81b882009-01-07 18:07:48 -08002864}
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002865
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07002866#if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU)
Johannes Weiner00501b52014-08-08 14:19:20 -07002867static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
Daisuke Nishimuraa3032a22009-12-15 16:47:10 -08002868{
Johannes Weinerce00a962014-09-05 08:43:57 -04002869 if (mem_cgroup_is_root(memcg))
2870 return;
2871
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002872 page_counter_uncharge(&memcg->memory, nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08002873 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08002874 page_counter_uncharge(&memcg->memsw, nr_pages);
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002875}
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07002876#endif
KAMEZAWA Hiroyukid01dd172012-05-29 15:07:03 -07002877
Johannes Weinerd9eb1ea2020-06-03 16:02:24 -07002878static void commit_charge(struct page *page, struct mem_cgroup *memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002879{
Johannes Weiner1306a852014-12-10 15:44:52 -08002880 VM_BUG_ON_PAGE(page->mem_cgroup, page);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002881 /*
Johannes Weinera0b5b412020-06-03 16:02:27 -07002882 * Any of the following ensures page->mem_cgroup stability:
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002883 *
Johannes Weinera0b5b412020-06-03 16:02:27 -07002884 * - the page lock
2885 * - LRU isolation
2886 * - lock_page_memcg()
2887 * - exclusive reference
Johannes Weiner0a31bc92014-08-08 14:19:22 -07002888 */
Johannes Weiner1306a852014-12-10 15:44:52 -08002889 page->mem_cgroup = memcg;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -08002890}
2891
Kirill Tkhai84c07d12018-08-17 15:47:25 -07002892#ifdef CONFIG_MEMCG_KMEM
Roman Gushchin10befea2020-08-06 23:21:27 -07002893int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
2894 gfp_t gfp)
2895{
2896 unsigned int objects = objs_per_slab_page(s, page);
2897 void *vec;
2898
2899 vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp,
2900 page_to_nid(page));
2901 if (!vec)
2902 return -ENOMEM;
2903
2904 if (cmpxchg(&page->obj_cgroups, NULL,
2905 (struct obj_cgroup **) ((unsigned long)vec | 0x1UL)))
2906 kfree(vec);
2907 else
2908 kmemleak_not_leak(vec);
2909
2910 return 0;
2911}
2912
Roman Gushchin8380ce42020-03-28 19:17:25 -07002913/*
2914 * Returns a pointer to the memory cgroup to which the kernel object is charged.
2915 *
2916 * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(),
2917 * cgroup_mutex, etc.
2918 */
2919struct mem_cgroup *mem_cgroup_from_obj(void *p)
2920{
2921 struct page *page;
2922
2923 if (mem_cgroup_disabled())
2924 return NULL;
2925
2926 page = virt_to_head_page(p);
2927
2928 /*
Roman Gushchin19b629c2020-10-13 16:52:42 -07002929 * If page->mem_cgroup is set, it's either a simple mem_cgroup pointer
2930 * or a pointer to obj_cgroup vector. In the latter case the lowest
2931 * bit of the pointer is set.
2932 * The page->mem_cgroup pointer can be asynchronously changed
2933 * from NULL to (obj_cgroup_vec | 0x1UL), but can't be changed
2934 * from a valid memcg pointer to objcg vector or back.
2935 */
2936 if (!page->mem_cgroup)
2937 return NULL;
2938
2939 /*
Roman Gushchin98556092020-08-06 23:21:10 -07002940 * Slab objects are accounted individually, not per-page.
2941 * Memcg membership data for each individual object is saved in
2942 * the page->obj_cgroups.
Roman Gushchin8380ce42020-03-28 19:17:25 -07002943 */
Roman Gushchin98556092020-08-06 23:21:10 -07002944 if (page_has_obj_cgroups(page)) {
2945 struct obj_cgroup *objcg;
2946 unsigned int off;
2947
2948 off = obj_to_index(page->slab_cache, page, p);
2949 objcg = page_obj_cgroups(page)[off];
Roman Gushchin10befea2020-08-06 23:21:27 -07002950 if (objcg)
2951 return obj_cgroup_memcg(objcg);
2952
2953 return NULL;
Roman Gushchin98556092020-08-06 23:21:10 -07002954 }
Roman Gushchin8380ce42020-03-28 19:17:25 -07002955
2956 /* All other pages use page->mem_cgroup */
2957 return page->mem_cgroup;
2958}
2959
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002960__always_inline struct obj_cgroup *get_obj_cgroup_from_current(void)
2961{
2962 struct obj_cgroup *objcg = NULL;
2963 struct mem_cgroup *memcg;
2964
Roman Gushchin279c3392020-10-17 16:13:44 -07002965 if (memcg_kmem_bypass())
2966 return NULL;
2967
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002968 rcu_read_lock();
Roman Gushchin37d59852020-10-17 16:13:50 -07002969 if (unlikely(active_memcg()))
2970 memcg = active_memcg();
Roman Gushchinbf4f0592020-08-06 23:20:49 -07002971 else
2972 memcg = mem_cgroup_from_task(current);
2973
2974 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) {
2975 objcg = rcu_dereference(memcg->objcg);
2976 if (objcg && obj_cgroup_tryget(objcg))
2977 break;
2978 }
2979 rcu_read_unlock();
2980
2981 return objcg;
2982}
2983
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002984static int memcg_alloc_cache_id(void)
Glauber Costa55007d82012-12-18 14:22:38 -08002985{
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002986 int id, size;
2987 int err;
Glauber Costa55007d82012-12-18 14:22:38 -08002988
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002989 id = ida_simple_get(&memcg_cache_ida,
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002990 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2991 if (id < 0)
2992 return id;
2993
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08002994 if (id < memcg_nr_cache_ids)
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07002995 return id;
2996
2997 /*
2998 * There's no space for the new id in memcg_caches arrays,
2999 * so we have to grow them.
3000 */
Vladimir Davydov05257a12015-02-12 14:59:01 -08003001 down_write(&memcg_cache_ids_sem);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07003002
3003 size = 2 * (id + 1);
Glauber Costa55007d82012-12-18 14:22:38 -08003004 if (size < MEMCG_CACHES_MIN_SIZE)
3005 size = MEMCG_CACHES_MIN_SIZE;
3006 else if (size > MEMCG_CACHES_MAX_SIZE)
3007 size = MEMCG_CACHES_MAX_SIZE;
3008
Roman Gushchin98556092020-08-06 23:21:10 -07003009 err = memcg_update_all_list_lrus(size);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08003010 if (!err)
Vladimir Davydov05257a12015-02-12 14:59:01 -08003011 memcg_nr_cache_ids = size;
3012
3013 up_write(&memcg_cache_ids_sem);
3014
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07003015 if (err) {
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08003016 ida_simple_remove(&memcg_cache_ida, id);
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07003017 return err;
3018 }
3019 return id;
3020}
3021
3022static void memcg_free_cache_id(int id)
3023{
Vladimir Davydovdbcf73e2015-02-12 14:58:57 -08003024 ida_simple_remove(&memcg_cache_ida, id);
Glauber Costa55007d82012-12-18 14:22:38 -08003025}
3026
Vladimir Davydov45264772016-07-26 15:24:21 -07003027/**
Roman Gushchin4b13f642020-04-01 21:06:56 -07003028 * __memcg_kmem_charge: charge a number of kernel pages to a memcg
Roman Gushchin10eaec22020-04-01 21:06:39 -07003029 * @memcg: memory cgroup to charge
Vladimir Davydov45264772016-07-26 15:24:21 -07003030 * @gfp: reclaim mode
Roman Gushchin92d0510c2020-04-01 21:06:49 -07003031 * @nr_pages: number of pages to charge
Vladimir Davydov45264772016-07-26 15:24:21 -07003032 *
3033 * Returns 0 on success, an error code on failure.
3034 */
Roman Gushchin4b13f642020-04-01 21:06:56 -07003035int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp,
3036 unsigned int nr_pages)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003037{
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003038 struct page_counter *counter;
Johannes Weiner6071ca52015-11-05 18:50:26 -08003039 int ret;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003040
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003041 ret = try_charge(memcg, gfp, nr_pages);
Johannes Weiner52c29b02016-01-20 15:02:35 -08003042 if (ret)
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003043 return ret;
Johannes Weiner52c29b02016-01-20 15:02:35 -08003044
3045 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
3046 !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
Michal Hockoe55d9d92019-09-25 16:45:53 -07003047
3048 /*
3049 * Enforce __GFP_NOFAIL allocation because callers are not
3050 * prepared to see failures and likely do not have any failure
3051 * handling code.
3052 */
3053 if (gfp & __GFP_NOFAIL) {
3054 page_counter_charge(&memcg->kmem, nr_pages);
3055 return 0;
3056 }
Johannes Weiner52c29b02016-01-20 15:02:35 -08003057 cancel_charge(memcg, nr_pages);
3058 return -ENOMEM;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003059 }
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003060 return 0;
3061}
3062
Vladimir Davydov45264772016-07-26 15:24:21 -07003063/**
Roman Gushchin4b13f642020-04-01 21:06:56 -07003064 * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg
3065 * @memcg: memcg to uncharge
3066 * @nr_pages: number of pages to uncharge
3067 */
3068void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages)
3069{
3070 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
3071 page_counter_uncharge(&memcg->kmem, nr_pages);
3072
3073 page_counter_uncharge(&memcg->memory, nr_pages);
3074 if (do_memsw_account())
3075 page_counter_uncharge(&memcg->memsw, nr_pages);
3076}
3077
3078/**
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003079 * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup
Vladimir Davydov45264772016-07-26 15:24:21 -07003080 * @page: page to charge
3081 * @gfp: reclaim mode
3082 * @order: allocation order
3083 *
3084 * Returns 0 on success, an error code on failure.
3085 */
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003086int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003087{
3088 struct mem_cgroup *memcg;
Vladimir Davydovfcff7d72016-03-17 14:17:29 -07003089 int ret = 0;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003090
Shakeel Buttd46eb14b2018-08-17 15:46:39 -07003091 memcg = get_mem_cgroup_from_current();
Roman Gushchin279c3392020-10-17 16:13:44 -07003092 if (memcg && !mem_cgroup_is_root(memcg)) {
Roman Gushchin4b13f642020-04-01 21:06:56 -07003093 ret = __memcg_kmem_charge(memcg, gfp, 1 << order);
Roman Gushchin4d96ba32019-07-11 20:56:31 -07003094 if (!ret) {
3095 page->mem_cgroup = memcg;
Vladimir Davydovc4159a72016-08-08 23:03:12 +03003096 __SetPageKmemcg(page);
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07003097 return 0;
Roman Gushchin4d96ba32019-07-11 20:56:31 -07003098 }
Roman Gushchin279c3392020-10-17 16:13:44 -07003099 css_put(&memcg->css);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03003100 }
Vladimir Davydovd05e83a2015-11-05 18:48:59 -08003101 return ret;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003102}
Roman Gushchin49a18ea2019-07-11 20:56:13 -07003103
3104/**
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003105 * __memcg_kmem_uncharge_page: uncharge a kmem page
Vladimir Davydov45264772016-07-26 15:24:21 -07003106 * @page: page to uncharge
3107 * @order: allocation order
3108 */
Roman Gushchinf4b00ea2020-04-01 21:06:46 -07003109void __memcg_kmem_uncharge_page(struct page *page, int order)
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003110{
Johannes Weiner1306a852014-12-10 15:44:52 -08003111 struct mem_cgroup *memcg = page->mem_cgroup;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08003112 unsigned int nr_pages = 1 << order;
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003113
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003114 if (!memcg)
3115 return;
3116
Sasha Levin309381fea2014-01-23 15:52:54 -08003117 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
Roman Gushchin4b13f642020-04-01 21:06:56 -07003118 __memcg_kmem_uncharge(memcg, nr_pages);
Johannes Weiner1306a852014-12-10 15:44:52 -08003119 page->mem_cgroup = NULL;
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07003120 css_put(&memcg->css);
Vladimir Davydovc4159a72016-08-08 23:03:12 +03003121
3122 /* slab pages do not have PageKmemcg flag set */
3123 if (PageKmemcg(page))
3124 __ClearPageKmemcg(page);
Vladimir Davydov60d3fd32015-02-12 14:59:10 -08003125}
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003126
3127static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes)
3128{
3129 struct memcg_stock_pcp *stock;
3130 unsigned long flags;
3131 bool ret = false;
3132
3133 local_irq_save(flags);
3134
3135 stock = this_cpu_ptr(&memcg_stock);
3136 if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) {
3137 stock->nr_bytes -= nr_bytes;
3138 ret = true;
3139 }
3140
3141 local_irq_restore(flags);
3142
3143 return ret;
3144}
3145
3146static void drain_obj_stock(struct memcg_stock_pcp *stock)
3147{
3148 struct obj_cgroup *old = stock->cached_objcg;
3149
3150 if (!old)
3151 return;
3152
3153 if (stock->nr_bytes) {
3154 unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT;
3155 unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1);
3156
3157 if (nr_pages) {
3158 rcu_read_lock();
3159 __memcg_kmem_uncharge(obj_cgroup_memcg(old), nr_pages);
3160 rcu_read_unlock();
3161 }
3162
3163 /*
3164 * The leftover is flushed to the centralized per-memcg value.
3165 * On the next attempt to refill obj stock it will be moved
3166 * to a per-cpu stock (probably, on an other CPU), see
3167 * refill_obj_stock().
3168 *
3169 * How often it's flushed is a trade-off between the memory
3170 * limit enforcement accuracy and potential CPU contention,
3171 * so it might be changed in the future.
3172 */
3173 atomic_add(nr_bytes, &old->nr_charged_bytes);
3174 stock->nr_bytes = 0;
3175 }
3176
3177 obj_cgroup_put(old);
3178 stock->cached_objcg = NULL;
3179}
3180
3181static bool obj_stock_flush_required(struct memcg_stock_pcp *stock,
3182 struct mem_cgroup *root_memcg)
3183{
3184 struct mem_cgroup *memcg;
3185
3186 if (stock->cached_objcg) {
3187 memcg = obj_cgroup_memcg(stock->cached_objcg);
3188 if (memcg && mem_cgroup_is_descendant(memcg, root_memcg))
3189 return true;
3190 }
3191
3192 return false;
3193}
3194
3195static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes)
3196{
3197 struct memcg_stock_pcp *stock;
3198 unsigned long flags;
3199
3200 local_irq_save(flags);
3201
3202 stock = this_cpu_ptr(&memcg_stock);
3203 if (stock->cached_objcg != objcg) { /* reset if necessary */
3204 drain_obj_stock(stock);
3205 obj_cgroup_get(objcg);
3206 stock->cached_objcg = objcg;
3207 stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0);
3208 }
3209 stock->nr_bytes += nr_bytes;
3210
3211 if (stock->nr_bytes > PAGE_SIZE)
3212 drain_obj_stock(stock);
3213
3214 local_irq_restore(flags);
3215}
3216
3217int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size)
3218{
3219 struct mem_cgroup *memcg;
3220 unsigned int nr_pages, nr_bytes;
3221 int ret;
3222
3223 if (consume_obj_stock(objcg, size))
3224 return 0;
3225
3226 /*
3227 * In theory, memcg->nr_charged_bytes can have enough
3228 * pre-charged bytes to satisfy the allocation. However,
3229 * flushing memcg->nr_charged_bytes requires two atomic
3230 * operations, and memcg->nr_charged_bytes can't be big,
3231 * so it's better to ignore it and try grab some new pages.
3232 * memcg->nr_charged_bytes will be flushed in
3233 * refill_obj_stock(), called from this function or
3234 * independently later.
3235 */
3236 rcu_read_lock();
3237 memcg = obj_cgroup_memcg(objcg);
3238 css_get(&memcg->css);
3239 rcu_read_unlock();
3240
3241 nr_pages = size >> PAGE_SHIFT;
3242 nr_bytes = size & (PAGE_SIZE - 1);
3243
3244 if (nr_bytes)
3245 nr_pages += 1;
3246
3247 ret = __memcg_kmem_charge(memcg, gfp, nr_pages);
3248 if (!ret && nr_bytes)
3249 refill_obj_stock(objcg, PAGE_SIZE - nr_bytes);
3250
3251 css_put(&memcg->css);
3252 return ret;
3253}
3254
3255void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size)
3256{
3257 refill_obj_stock(objcg, size);
3258}
3259
Kirill Tkhai84c07d12018-08-17 15:47:25 -07003260#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa7ae1e1d2012-12-18 14:21:56 -08003261
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003262#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3263
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003264/*
3265 * Because tail pages are not marked as "used", set it. We're under
Andrey Ryabininf4b7e272019-03-05 15:49:39 -08003266 * pgdat->lru_lock and migration entries setup in all page mappings.
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003267 */
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003268void mem_cgroup_split_huge_fixup(struct page *head)
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003269{
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07003270 struct mem_cgroup *memcg = head->mem_cgroup;
KAMEZAWA Hiroyukie94c8a92012-01-12 17:18:20 -08003271 int i;
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003272
KAMEZAWA Hiroyuki3d37c4a2011-01-25 15:07:28 -08003273 if (mem_cgroup_disabled())
3274 return;
David Rientjesb070e652013-05-07 16:18:09 -07003275
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07003276 for (i = 1; i < HPAGE_PMD_NR; i++) {
3277 css_get(&memcg->css);
3278 head[i].mem_cgroup = memcg;
3279 }
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003280}
Hugh Dickins12d27102012-01-12 17:19:52 -08003281#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
KAMEZAWA Hiroyukica3e0212011-01-20 14:44:24 -08003282
Andrew Mortonc255a452012-07-31 16:43:02 -07003283#ifdef CONFIG_MEMCG_SWAP
Daisuke Nishimura02491442010-03-10 15:22:17 -08003284/**
3285 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3286 * @entry: swap entry to be moved
3287 * @from: mem_cgroup which the entry is moved from
3288 * @to: mem_cgroup which the entry is moved to
3289 *
3290 * It succeeds only when the swap_cgroup's record for this entry is the same
3291 * as the mem_cgroup's id of @from.
3292 *
3293 * Returns 0 on success, -EINVAL on failure.
3294 *
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003295 * The caller must have charged to @to, IOW, called page_counter_charge() about
Daisuke Nishimura02491442010-03-10 15:22:17 -08003296 * both res and memsw, and called css_get().
3297 */
3298static int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003299 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003300{
3301 unsigned short old_id, new_id;
3302
Li Zefan34c00c32013-09-23 16:56:01 +08003303 old_id = mem_cgroup_id(from);
3304 new_id = mem_cgroup_id(to);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003305
3306 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
Johannes Weinerc9019e92018-01-31 16:16:37 -08003307 mod_memcg_state(from, MEMCG_SWAP, -1);
3308 mod_memcg_state(to, MEMCG_SWAP, 1);
Daisuke Nishimura02491442010-03-10 15:22:17 -08003309 return 0;
3310 }
3311 return -EINVAL;
3312}
3313#else
3314static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
Hugh Dickinse91cbb42012-05-29 15:06:51 -07003315 struct mem_cgroup *from, struct mem_cgroup *to)
Daisuke Nishimura02491442010-03-10 15:22:17 -08003316{
3317 return -EINVAL;
3318}
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003319#endif
3320
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003321static DEFINE_MUTEX(memcg_max_mutex);
Daisuke Nishimuraf212ad72011-03-23 16:42:25 -07003322
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003323static int mem_cgroup_resize_max(struct mem_cgroup *memcg,
3324 unsigned long max, bool memsw)
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003325{
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003326 bool enlarge = false;
Shakeel Buttbb4a7ea2018-06-07 17:07:27 -07003327 bool drained = false;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003328 int ret;
Yu Zhaoc054a782018-01-31 16:20:02 -08003329 bool limits_invariant;
3330 struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory;
KAMEZAWA Hiroyuki81d39c22009-04-02 16:57:36 -07003331
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003332 do {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003333 if (signal_pending(current)) {
3334 ret = -EINTR;
3335 break;
3336 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003337
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003338 mutex_lock(&memcg_max_mutex);
Yu Zhaoc054a782018-01-31 16:20:02 -08003339 /*
3340 * Make sure that the new limit (memsw or memory limit) doesn't
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003341 * break our basic invariant rule memory.max <= memsw.max.
Yu Zhaoc054a782018-01-31 16:20:02 -08003342 */
Chris Down15b42562020-04-01 21:07:20 -07003343 limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) :
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003344 max <= memcg->memsw.max;
Yu Zhaoc054a782018-01-31 16:20:02 -08003345 if (!limits_invariant) {
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003346 mutex_unlock(&memcg_max_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003347 ret = -EINVAL;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003348 break;
3349 }
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003350 if (max > counter->max)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003351 enlarge = true;
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003352 ret = page_counter_set_max(counter, max);
3353 mutex_unlock(&memcg_max_mutex);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003354
3355 if (!ret)
3356 break;
3357
Shakeel Buttbb4a7ea2018-06-07 17:07:27 -07003358 if (!drained) {
3359 drain_all_stock(memcg);
3360 drained = true;
3361 continue;
3362 }
3363
Andrey Ryabinin1ab5c052018-01-31 16:20:37 -08003364 if (!try_to_free_mem_cgroup_pages(memcg, 1,
3365 GFP_KERNEL, !memsw)) {
3366 ret = -EBUSY;
3367 break;
3368 }
3369 } while (true);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003370
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07003371 if (!ret && enlarge)
3372 memcg_oom_recover(memcg);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003373
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003374 return ret;
3375}
3376
Mel Gormanef8f2322016-07-28 15:46:05 -07003377unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
Andrew Morton0608f432013-09-24 15:27:41 -07003378 gfp_t gfp_mask,
3379 unsigned long *total_scanned)
3380{
3381 unsigned long nr_reclaimed = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07003382 struct mem_cgroup_per_node *mz, *next_mz = NULL;
Andrew Morton0608f432013-09-24 15:27:41 -07003383 unsigned long reclaimed;
3384 int loop = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07003385 struct mem_cgroup_tree_per_node *mctz;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003386 unsigned long excess;
Andrew Morton0608f432013-09-24 15:27:41 -07003387 unsigned long nr_scanned;
3388
3389 if (order > 0)
3390 return 0;
3391
Mel Gormanef8f2322016-07-28 15:46:05 -07003392 mctz = soft_limit_tree_node(pgdat->node_id);
Michal Hockod6507ff2016-08-02 14:02:37 -07003393
3394 /*
3395 * Do not even bother to check the largest node if the root
3396 * is empty. Do it lockless to prevent lock bouncing. Races
3397 * are acceptable as soft limit is best effort anyway.
3398 */
Laurent Dufourbfc72282017-03-09 16:17:06 -08003399 if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root))
Michal Hockod6507ff2016-08-02 14:02:37 -07003400 return 0;
3401
Andrew Morton0608f432013-09-24 15:27:41 -07003402 /*
3403 * This loop can run a while, specially if mem_cgroup's continuously
3404 * keep exceeding their soft limit and putting the system under
3405 * pressure
3406 */
3407 do {
3408 if (next_mz)
3409 mz = next_mz;
3410 else
3411 mz = mem_cgroup_largest_soft_limit_node(mctz);
3412 if (!mz)
3413 break;
3414
3415 nr_scanned = 0;
Mel Gormanef8f2322016-07-28 15:46:05 -07003416 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
Andrew Morton0608f432013-09-24 15:27:41 -07003417 gfp_mask, &nr_scanned);
3418 nr_reclaimed += reclaimed;
3419 *total_scanned += nr_scanned;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003420 spin_lock_irq(&mctz->lock);
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08003421 __mem_cgroup_remove_exceeded(mz, mctz);
Andrew Morton0608f432013-09-24 15:27:41 -07003422
3423 /*
3424 * If we failed to reclaim anything from this memory cgroup
3425 * it is time to move on to the next cgroup
3426 */
3427 next_mz = NULL;
Vladimir Davydovbc2f2e72014-12-10 15:43:40 -08003428 if (!reclaimed)
3429 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
3430
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003431 excess = soft_limit_excess(mz->memcg);
Andrew Morton0608f432013-09-24 15:27:41 -07003432 /*
3433 * One school of thought says that we should not add
3434 * back the node to the tree if reclaim returns 0.
3435 * But our reclaim could return 0, simply because due
3436 * to priority we are exposing a smaller subset of
3437 * memory to reclaim from. Consider this as a longer
3438 * term TODO.
3439 */
3440 /* If excess == 0, no tree ops */
Johannes Weinercf2c8122014-06-06 14:38:21 -07003441 __mem_cgroup_insert_exceeded(mz, mctz, excess);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07003442 spin_unlock_irq(&mctz->lock);
Andrew Morton0608f432013-09-24 15:27:41 -07003443 css_put(&mz->memcg->css);
3444 loop++;
3445 /*
3446 * Could not reclaim anything and there are no more
3447 * mem cgroups to try or we seem to be looping without
3448 * reclaiming anything.
3449 */
3450 if (!nr_reclaimed &&
3451 (next_mz == NULL ||
3452 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3453 break;
3454 } while (!nr_reclaimed);
3455 if (next_mz)
3456 css_put(&next_mz->memcg->css);
3457 return nr_reclaimed;
3458}
3459
Tejun Heoea280e72014-05-16 13:22:48 -04003460/*
3461 * Test whether @memcg has children, dead or alive. Note that this
3462 * function doesn't care whether @memcg has use_hierarchy enabled and
3463 * returns %true if there are child csses according to the cgroup
Ethon Paulb8f29352020-06-04 16:49:28 -07003464 * hierarchy. Testing use_hierarchy is the caller's responsibility.
Tejun Heoea280e72014-05-16 13:22:48 -04003465 */
Glauber Costab5f99b52013-02-22 16:34:53 -08003466static inline bool memcg_has_children(struct mem_cgroup *memcg)
3467{
Tejun Heoea280e72014-05-16 13:22:48 -04003468 bool ret;
3469
Tejun Heoea280e72014-05-16 13:22:48 -04003470 rcu_read_lock();
3471 ret = css_next_child(NULL, &memcg->css);
3472 rcu_read_unlock();
3473 return ret;
Glauber Costab5f99b52013-02-22 16:34:53 -08003474}
3475
3476/*
Greg Thelen51038172016-05-20 16:58:18 -07003477 * Reclaims as many pages from the given memcg as possible.
Michal Hockoc26251f2012-10-26 13:37:28 +02003478 *
3479 * Caller is responsible for holding css reference for memcg.
3480 */
3481static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
3482{
Chris Downd977aa92020-08-06 23:21:58 -07003483 int nr_retries = MAX_RECLAIM_RETRIES;
Michal Hockoc26251f2012-10-26 13:37:28 +02003484
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003485 /* we call try-to-free pages for make this cgroup empty */
3486 lru_add_drain_all();
Junaid Shahidd12c60f2018-06-07 17:07:31 -07003487
3488 drain_all_stock(memcg);
3489
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003490 /* try to free all pages in this cgroup */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003491 while (nr_retries && page_counter_read(&memcg->memory)) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003492 int progress;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003493
Michal Hockoc26251f2012-10-26 13:37:28 +02003494 if (signal_pending(current))
3495 return -EINTR;
3496
Johannes Weinerb70a2a22014-10-09 15:28:56 -07003497 progress = try_to_free_mem_cgroup_pages(memcg, 1,
3498 GFP_KERNEL, true);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003499 if (!progress) {
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003500 nr_retries--;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003501 /* maybe some writeback is necessary */
Jens Axboe8aa7e842009-07-09 14:52:32 +02003502 congestion_wait(BLK_RW_ASYNC, HZ/10);
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003503 }
KAMEZAWA Hiroyukif817ed42009-01-07 18:07:53 -08003504
3505 }
Michal Hockoab5196c2012-10-26 13:37:32 +02003506
3507 return 0;
KAMEZAWA Hiroyukicc847582008-02-07 00:14:16 -08003508}
3509
Tejun Heo6770c642014-05-13 12:16:21 -04003510static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
3511 char *buf, size_t nbytes,
3512 loff_t off)
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003513{
Tejun Heo6770c642014-05-13 12:16:21 -04003514 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Michal Hockoc26251f2012-10-26 13:37:28 +02003515
Michal Hockod8423012012-10-26 13:37:29 +02003516 if (mem_cgroup_is_root(memcg))
3517 return -EINVAL;
Tejun Heo6770c642014-05-13 12:16:21 -04003518 return mem_cgroup_force_empty(memcg) ?: nbytes;
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08003519}
3520
Tejun Heo182446d2013-08-08 20:11:24 -04003521static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
3522 struct cftype *cft)
Balbir Singh18f59ea2009-01-07 18:08:07 -08003523{
Tejun Heo182446d2013-08-08 20:11:24 -04003524 return mem_cgroup_from_css(css)->use_hierarchy;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003525}
3526
Tejun Heo182446d2013-08-08 20:11:24 -04003527static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
3528 struct cftype *cft, u64 val)
Balbir Singh18f59ea2009-01-07 18:08:07 -08003529{
3530 int retval = 0;
Tejun Heo182446d2013-08-08 20:11:24 -04003531 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo5c9d5352014-05-16 13:22:48 -04003532 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
Balbir Singh18f59ea2009-01-07 18:08:07 -08003533
Glauber Costa567fb432012-07-31 16:43:07 -07003534 if (memcg->use_hierarchy == val)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003535 return 0;
Glauber Costa567fb432012-07-31 16:43:07 -07003536
Balbir Singh18f59ea2009-01-07 18:08:07 -08003537 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02003538 * If parent's use_hierarchy is set, we can't make any modifications
Balbir Singh18f59ea2009-01-07 18:08:07 -08003539 * in the child subtrees. If it is unset, then the change can
3540 * occur, provided the current cgroup has no children.
3541 *
3542 * For the root cgroup, parent_mem is NULL, we allow value to be
3543 * set if there are no children.
3544 */
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003545 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
Balbir Singh18f59ea2009-01-07 18:08:07 -08003546 (val == 1 || val == 0)) {
Tejun Heoea280e72014-05-16 13:22:48 -04003547 if (!memcg_has_children(memcg))
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07003548 memcg->use_hierarchy = val;
Balbir Singh18f59ea2009-01-07 18:08:07 -08003549 else
3550 retval = -EBUSY;
3551 } else
3552 retval = -EINVAL;
Glauber Costa567fb432012-07-31 16:43:07 -07003553
Balbir Singh18f59ea2009-01-07 18:08:07 -08003554 return retval;
3555}
3556
Andrew Morton6f646152015-11-06 16:28:58 -08003557static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
Johannes Weinerce00a962014-09-05 08:43:57 -04003558{
Johannes Weiner42a30032019-05-14 15:47:12 -07003559 unsigned long val;
Johannes Weinerce00a962014-09-05 08:43:57 -04003560
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003561 if (mem_cgroup_is_root(memcg)) {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07003562 val = memcg_page_state(memcg, NR_FILE_PAGES) +
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07003563 memcg_page_state(memcg, NR_ANON_MAPPED);
Johannes Weiner42a30032019-05-14 15:47:12 -07003564 if (swap)
3565 val += memcg_page_state(memcg, MEMCG_SWAP);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003566 } else {
Johannes Weinerce00a962014-09-05 08:43:57 -04003567 if (!swap)
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003568 val = page_counter_read(&memcg->memory);
Johannes Weinerce00a962014-09-05 08:43:57 -04003569 else
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003570 val = page_counter_read(&memcg->memsw);
Johannes Weinerce00a962014-09-05 08:43:57 -04003571 }
Michal Hockoc12176d2015-11-05 18:50:29 -08003572 return val;
Johannes Weinerce00a962014-09-05 08:43:57 -04003573}
3574
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003575enum {
3576 RES_USAGE,
3577 RES_LIMIT,
3578 RES_MAX_USAGE,
3579 RES_FAILCNT,
3580 RES_SOFT_LIMIT,
3581};
Johannes Weinerce00a962014-09-05 08:43:57 -04003582
Tejun Heo791badb2013-12-05 12:28:02 -05003583static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
Johannes Weiner05b84302014-08-06 16:05:59 -07003584 struct cftype *cft)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003585{
Tejun Heo182446d2013-08-08 20:11:24 -04003586 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003587 struct page_counter *counter;
Tejun Heoaf36f902012-04-01 12:09:55 -07003588
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003589 switch (MEMFILE_TYPE(cft->private)) {
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003590 case _MEM:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003591 counter = &memcg->memory;
Glauber Costa510fc4e2012-12-18 14:21:47 -08003592 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003593 case _MEMSWAP:
3594 counter = &memcg->memsw;
3595 break;
3596 case _KMEM:
3597 counter = &memcg->kmem;
3598 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003599 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003600 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003601 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003602 default:
3603 BUG();
3604 }
3605
3606 switch (MEMFILE_ATTR(cft->private)) {
3607 case RES_USAGE:
3608 if (counter == &memcg->memory)
Michal Hockoc12176d2015-11-05 18:50:29 -08003609 return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003610 if (counter == &memcg->memsw)
Michal Hockoc12176d2015-11-05 18:50:29 -08003611 return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003612 return (u64)page_counter_read(counter) * PAGE_SIZE;
3613 case RES_LIMIT:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003614 return (u64)counter->max * PAGE_SIZE;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003615 case RES_MAX_USAGE:
3616 return (u64)counter->watermark * PAGE_SIZE;
3617 case RES_FAILCNT:
3618 return counter->failcnt;
3619 case RES_SOFT_LIMIT:
3620 return (u64)memcg->soft_limit * PAGE_SIZE;
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003621 default:
3622 BUG();
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003623 }
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003624}
Glauber Costa510fc4e2012-12-18 14:21:47 -08003625
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003626static void memcg_flush_percpu_vmstats(struct mem_cgroup *memcg)
Roman Gushchinc350a992019-08-24 17:54:47 -07003627{
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003628 unsigned long stat[MEMCG_NR_STAT] = {0};
Roman Gushchinc350a992019-08-24 17:54:47 -07003629 struct mem_cgroup *mi;
3630 int node, cpu, i;
Roman Gushchinc350a992019-08-24 17:54:47 -07003631
3632 for_each_online_cpu(cpu)
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003633 for (i = 0; i < MEMCG_NR_STAT; i++)
Shakeel Butt6c1c2802019-08-30 16:04:53 -07003634 stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu);
Roman Gushchinc350a992019-08-24 17:54:47 -07003635
3636 for (mi = memcg; mi; mi = parent_mem_cgroup(mi))
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003637 for (i = 0; i < MEMCG_NR_STAT; i++)
Roman Gushchinc350a992019-08-24 17:54:47 -07003638 atomic_long_add(stat[i], &mi->vmstats[i]);
3639
3640 for_each_node(node) {
3641 struct mem_cgroup_per_node *pn = memcg->nodeinfo[node];
3642 struct mem_cgroup_per_node *pi;
3643
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003644 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
Roman Gushchinc350a992019-08-24 17:54:47 -07003645 stat[i] = 0;
3646
3647 for_each_online_cpu(cpu)
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003648 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
Shakeel Butt6c1c2802019-08-30 16:04:53 -07003649 stat[i] += per_cpu(
3650 pn->lruvec_stat_cpu->count[i], cpu);
Roman Gushchinc350a992019-08-24 17:54:47 -07003651
3652 for (pi = pn; pi; pi = parent_nodeinfo(pi, node))
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08003653 for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++)
Roman Gushchinc350a992019-08-24 17:54:47 -07003654 atomic_long_add(stat[i], &pi->lruvec_stat[i]);
3655 }
3656}
3657
Roman Gushchinbb65f892019-08-24 17:54:50 -07003658static void memcg_flush_percpu_vmevents(struct mem_cgroup *memcg)
3659{
3660 unsigned long events[NR_VM_EVENT_ITEMS];
3661 struct mem_cgroup *mi;
3662 int cpu, i;
3663
3664 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
3665 events[i] = 0;
3666
3667 for_each_online_cpu(cpu)
3668 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
Shakeel Butt6c1c2802019-08-30 16:04:53 -07003669 events[i] += per_cpu(memcg->vmstats_percpu->events[i],
3670 cpu);
Roman Gushchinbb65f892019-08-24 17:54:50 -07003671
3672 for (mi = memcg; mi; mi = parent_mem_cgroup(mi))
3673 for (i = 0; i < NR_VM_EVENT_ITEMS; i++)
3674 atomic_long_add(events[i], &mi->vmevents[i]);
3675}
3676
Kirill Tkhai84c07d12018-08-17 15:47:25 -07003677#ifdef CONFIG_MEMCG_KMEM
Johannes Weiner567e9ab2016-01-20 15:02:24 -08003678static int memcg_online_kmem(struct mem_cgroup *memcg)
Vladimir Davydovd6441632014-01-23 15:53:09 -08003679{
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003680 struct obj_cgroup *objcg;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003681 int memcg_id;
3682
Vladimir Davydovb313aee2016-03-17 14:18:27 -07003683 if (cgroup_memory_nokmem)
3684 return 0;
3685
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -08003686 BUG_ON(memcg->kmemcg_id >= 0);
Johannes Weiner567e9ab2016-01-20 15:02:24 -08003687 BUG_ON(memcg->kmem_state);
Vladimir Davydovd6441632014-01-23 15:53:09 -08003688
Vladimir Davydovf3bb3042014-10-09 15:28:45 -07003689 memcg_id = memcg_alloc_cache_id();
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003690 if (memcg_id < 0)
3691 return memcg_id;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003692
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003693 objcg = obj_cgroup_alloc();
3694 if (!objcg) {
3695 memcg_free_cache_id(memcg_id);
3696 return -ENOMEM;
3697 }
3698 objcg->memcg = memcg;
3699 rcu_assign_pointer(memcg->objcg, objcg);
3700
Roman Gushchind648bcc2020-08-06 23:20:28 -07003701 static_branch_enable(&memcg_kmem_enabled_key);
3702
Vladimir Davydovd6441632014-01-23 15:53:09 -08003703 /*
Johannes Weiner567e9ab2016-01-20 15:02:24 -08003704 * A memory cgroup is considered kmem-online as soon as it gets
Vladimir Davydov900a38f2014-12-12 16:55:10 -08003705 * kmemcg_id. Setting the id after enabling static branching will
Vladimir Davydovd6441632014-01-23 15:53:09 -08003706 * guarantee no one starts accounting before all call sites are
3707 * patched.
3708 */
Vladimir Davydov900a38f2014-12-12 16:55:10 -08003709 memcg->kmemcg_id = memcg_id;
Johannes Weiner567e9ab2016-01-20 15:02:24 -08003710 memcg->kmem_state = KMEM_ONLINE;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003711
3712 return 0;
Vladimir Davydovd6441632014-01-23 15:53:09 -08003713}
3714
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003715static void memcg_offline_kmem(struct mem_cgroup *memcg)
3716{
3717 struct cgroup_subsys_state *css;
3718 struct mem_cgroup *parent, *child;
3719 int kmemcg_id;
3720
3721 if (memcg->kmem_state != KMEM_ONLINE)
3722 return;
Roman Gushchin98556092020-08-06 23:21:10 -07003723
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003724 memcg->kmem_state = KMEM_ALLOCATED;
3725
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003726 parent = parent_mem_cgroup(memcg);
3727 if (!parent)
3728 parent = root_mem_cgroup;
3729
Roman Gushchinbf4f0592020-08-06 23:20:49 -07003730 memcg_reparent_objcgs(memcg, parent);
Roman Gushchinfb2f2b02019-07-11 20:56:34 -07003731
3732 kmemcg_id = memcg->kmemcg_id;
3733 BUG_ON(kmemcg_id < 0);
3734
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003735 /*
3736 * Change kmemcg_id of this cgroup and all its descendants to the
3737 * parent's id, and then move all entries from this cgroup's list_lrus
3738 * to ones of the parent. After we have finished, all list_lrus
3739 * corresponding to this cgroup are guaranteed to remain empty. The
3740 * ordering is imposed by list_lru_node->lock taken by
3741 * memcg_drain_all_list_lrus().
3742 */
Tejun Heo3a06bb72016-06-03 14:55:44 -07003743 rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003744 css_for_each_descendant_pre(css, &memcg->css) {
3745 child = mem_cgroup_from_css(css);
3746 BUG_ON(child->kmemcg_id != kmemcg_id);
3747 child->kmemcg_id = parent->kmemcg_id;
3748 if (!memcg->use_hierarchy)
3749 break;
3750 }
Tejun Heo3a06bb72016-06-03 14:55:44 -07003751 rcu_read_unlock();
3752
Kirill Tkhai9bec5c32018-08-17 15:47:58 -07003753 memcg_drain_all_list_lrus(kmemcg_id, parent);
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003754
3755 memcg_free_cache_id(kmemcg_id);
3756}
3757
3758static void memcg_free_kmem(struct mem_cgroup *memcg)
3759{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003760 /* css_alloc() failed, offlining didn't happen */
3761 if (unlikely(memcg->kmem_state == KMEM_ONLINE))
3762 memcg_offline_kmem(memcg);
Johannes Weiner8e0a8912016-01-20 15:02:26 -08003763}
Vladimir Davydovd6441632014-01-23 15:53:09 -08003764#else
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08003765static int memcg_online_kmem(struct mem_cgroup *memcg)
Johannes Weiner127424c2016-01-20 15:02:32 -08003766{
3767 return 0;
3768}
3769static void memcg_offline_kmem(struct mem_cgroup *memcg)
3770{
3771}
3772static void memcg_free_kmem(struct mem_cgroup *memcg)
3773{
3774}
Kirill Tkhai84c07d12018-08-17 15:47:25 -07003775#endif /* CONFIG_MEMCG_KMEM */
Johannes Weiner127424c2016-01-20 15:02:32 -08003776
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003777static int memcg_update_kmem_max(struct mem_cgroup *memcg,
3778 unsigned long max)
Johannes Weiner127424c2016-01-20 15:02:32 -08003779{
Vladimir Davydovb313aee2016-03-17 14:18:27 -07003780 int ret;
Johannes Weiner127424c2016-01-20 15:02:32 -08003781
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003782 mutex_lock(&memcg_max_mutex);
3783 ret = page_counter_set_max(&memcg->kmem, max);
3784 mutex_unlock(&memcg_max_mutex);
Johannes Weiner127424c2016-01-20 15:02:32 -08003785 return ret;
3786}
Glauber Costa510fc4e2012-12-18 14:21:47 -08003787
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003788static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003789{
3790 int ret;
3791
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003792 mutex_lock(&memcg_max_mutex);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003793
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003794 ret = page_counter_set_max(&memcg->tcpmem, max);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003795 if (ret)
3796 goto out;
3797
Johannes Weiner0db15292016-01-20 15:02:50 -08003798 if (!memcg->tcpmem_active) {
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003799 /*
3800 * The active flag needs to be written after the static_key
3801 * update. This is what guarantees that the socket activation
Johannes Weiner2d758072016-10-07 17:00:58 -07003802 * function is the last one to run. See mem_cgroup_sk_alloc()
3803 * for details, and note that we don't mark any socket as
3804 * belonging to this memcg until that flag is up.
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003805 *
3806 * We need to do this, because static_keys will span multiple
3807 * sites, but we can't control their order. If we mark a socket
3808 * as accounted, but the accounting functions are not patched in
3809 * yet, we'll lose accounting.
3810 *
Johannes Weiner2d758072016-10-07 17:00:58 -07003811 * We never race with the readers in mem_cgroup_sk_alloc(),
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003812 * because when this value change, the code to process it is not
3813 * patched in yet.
3814 */
3815 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weiner0db15292016-01-20 15:02:50 -08003816 memcg->tcpmem_active = true;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003817 }
3818out:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003819 mutex_unlock(&memcg_max_mutex);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003820 return ret;
3821}
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003822
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003823/*
3824 * The user of this function is...
3825 * RES_LIMIT.
3826 */
Tejun Heo451af502014-05-13 12:16:21 -04003827static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
3828 char *buf, size_t nbytes, loff_t off)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003829{
Tejun Heo451af502014-05-13 12:16:21 -04003830 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003831 unsigned long nr_pages;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003832 int ret;
3833
Tejun Heo451af502014-05-13 12:16:21 -04003834 buf = strstrip(buf);
Johannes Weiner650c5e52015-02-11 15:26:03 -08003835 ret = page_counter_memparse(buf, "-1", &nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003836 if (ret)
3837 return ret;
Tejun Heoaf36f902012-04-01 12:09:55 -07003838
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003839 switch (MEMFILE_ATTR(of_cft(of)->private)) {
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003840 case RES_LIMIT:
Balbir Singh4b3bde42009-09-23 15:56:32 -07003841 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3842 ret = -EINVAL;
3843 break;
3844 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003845 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3846 case _MEM:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003847 ret = mem_cgroup_resize_max(memcg, nr_pages, false);
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08003848 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003849 case _MEMSWAP:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003850 ret = mem_cgroup_resize_max(memcg, nr_pages, true);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003851 break;
3852 case _KMEM:
Michal Hocko01581152019-09-23 15:37:22 -07003853 pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. "
3854 "Please report your usecase to linux-mm@kvack.org if you "
3855 "depend on this functionality.\n");
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003856 ret = memcg_update_kmem_max(memcg, nr_pages);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003857 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003858 case _TCP:
Roman Gushchinbbec2e12018-06-07 17:06:18 -07003859 ret = memcg_update_tcp_max(memcg, nr_pages);
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003860 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003861 }
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003862 break;
Balbir Singh296c81d2009-09-23 15:56:36 -07003863 case RES_SOFT_LIMIT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003864 memcg->soft_limit = nr_pages;
3865 ret = 0;
KAMEZAWA Hiroyuki628f4232008-07-25 01:47:20 -07003866 break;
3867 }
Tejun Heo451af502014-05-13 12:16:21 -04003868 return ret ?: nbytes;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08003869}
3870
Tejun Heo6770c642014-05-13 12:16:21 -04003871static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3872 size_t nbytes, loff_t off)
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003873{
Tejun Heo6770c642014-05-13 12:16:21 -04003874 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003875 struct page_counter *counter;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003876
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003877 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3878 case _MEM:
3879 counter = &memcg->memory;
3880 break;
3881 case _MEMSWAP:
3882 counter = &memcg->memsw;
3883 break;
3884 case _KMEM:
3885 counter = &memcg->kmem;
3886 break;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003887 case _TCP:
Johannes Weiner0db15292016-01-20 15:02:50 -08003888 counter = &memcg->tcpmem;
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08003889 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003890 default:
3891 BUG();
3892 }
Tejun Heoaf36f902012-04-01 12:09:55 -07003893
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003894 switch (MEMFILE_ATTR(of_cft(of)->private)) {
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003895 case RES_MAX_USAGE:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003896 page_counter_reset_watermark(counter);
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003897 break;
3898 case RES_FAILCNT:
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003899 counter->failcnt = 0;
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003900 break;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08003901 default:
3902 BUG();
Pavel Emelyanov29f2a4d2008-04-29 01:00:21 -07003903 }
Balbir Singhf64c3f52009-09-23 15:56:37 -07003904
Tejun Heo6770c642014-05-13 12:16:21 -04003905 return nbytes;
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07003906}
3907
Tejun Heo182446d2013-08-08 20:11:24 -04003908static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003909 struct cftype *cft)
3910{
Tejun Heo182446d2013-08-08 20:11:24 -04003911 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003912}
3913
Daisuke Nishimura02491442010-03-10 15:22:17 -08003914#ifdef CONFIG_MMU
Tejun Heo182446d2013-08-08 20:11:24 -04003915static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003916 struct cftype *cft, u64 val)
3917{
Tejun Heo182446d2013-08-08 20:11:24 -04003918 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003919
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08003920 if (val & ~MOVE_MASK)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003921 return -EINVAL;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003922
Glauber Costaee5e8472013-02-22 16:34:50 -08003923 /*
3924 * No kind of locking is needed in here, because ->can_attach() will
3925 * check this value once in the beginning of the process, and then carry
3926 * on with stale data. This means that changes to this value will only
3927 * affect task migrations starting after the change.
3928 */
3929 memcg->move_charge_at_immigrate = val;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003930 return 0;
3931}
Daisuke Nishimura02491442010-03-10 15:22:17 -08003932#else
Tejun Heo182446d2013-08-08 20:11:24 -04003933static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
Daisuke Nishimura02491442010-03-10 15:22:17 -08003934 struct cftype *cft, u64 val)
3935{
3936 return -ENOSYS;
3937}
3938#endif
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08003939
Ying Han406eb0c2011-05-26 16:25:37 -07003940#ifdef CONFIG_NUMA
Johannes Weiner113b7df2019-05-13 17:18:11 -07003941
3942#define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE))
3943#define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON))
3944#define LRU_ALL ((1 << NR_LRU_LISTS) - 1)
3945
3946static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003947 int nid, unsigned int lru_mask, bool tree)
Johannes Weiner113b7df2019-05-13 17:18:11 -07003948{
Johannes Weiner867e5e12019-11-30 17:55:34 -08003949 struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid));
Johannes Weiner113b7df2019-05-13 17:18:11 -07003950 unsigned long nr = 0;
3951 enum lru_list lru;
3952
3953 VM_BUG_ON((unsigned)nid >= nr_node_ids);
3954
3955 for_each_lru(lru) {
3956 if (!(BIT(lru) & lru_mask))
3957 continue;
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003958 if (tree)
3959 nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru);
3960 else
3961 nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru);
Johannes Weiner113b7df2019-05-13 17:18:11 -07003962 }
3963 return nr;
3964}
3965
3966static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003967 unsigned int lru_mask,
3968 bool tree)
Johannes Weiner113b7df2019-05-13 17:18:11 -07003969{
3970 unsigned long nr = 0;
3971 enum lru_list lru;
3972
3973 for_each_lru(lru) {
3974 if (!(BIT(lru) & lru_mask))
3975 continue;
Shakeel Buttdd8657b2020-06-03 15:56:24 -07003976 if (tree)
3977 nr += memcg_page_state(memcg, NR_LRU_BASE + lru);
3978 else
3979 nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru);
Johannes Weiner113b7df2019-05-13 17:18:11 -07003980 }
3981 return nr;
3982}
3983
Tejun Heo2da8ca82013-12-05 12:28:04 -05003984static int memcg_numa_stat_show(struct seq_file *m, void *v)
Ying Han406eb0c2011-05-26 16:25:37 -07003985{
Greg Thelen25485de2013-11-12 15:07:40 -08003986 struct numa_stat {
3987 const char *name;
3988 unsigned int lru_mask;
3989 };
3990
3991 static const struct numa_stat stats[] = {
3992 { "total", LRU_ALL },
3993 { "file", LRU_ALL_FILE },
3994 { "anon", LRU_ALL_ANON },
3995 { "unevictable", BIT(LRU_UNEVICTABLE) },
3996 };
3997 const struct numa_stat *stat;
Ying Han406eb0c2011-05-26 16:25:37 -07003998 int nid;
Chris Downaa9694b2019-03-05 15:45:52 -08003999 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Ying Han406eb0c2011-05-26 16:25:37 -07004000
Greg Thelen25485de2013-11-12 15:07:40 -08004001 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
Shakeel Buttdd8657b2020-06-03 15:56:24 -07004002 seq_printf(m, "%s=%lu", stat->name,
4003 mem_cgroup_nr_lru_pages(memcg, stat->lru_mask,
4004 false));
4005 for_each_node_state(nid, N_MEMORY)
4006 seq_printf(m, " N%d=%lu", nid,
4007 mem_cgroup_node_nr_lru_pages(memcg, nid,
4008 stat->lru_mask, false));
Greg Thelen25485de2013-11-12 15:07:40 -08004009 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07004010 }
Ying Han406eb0c2011-05-26 16:25:37 -07004011
Ying Han071aee12013-11-12 15:07:41 -08004012 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
Ying Han406eb0c2011-05-26 16:25:37 -07004013
Shakeel Buttdd8657b2020-06-03 15:56:24 -07004014 seq_printf(m, "hierarchical_%s=%lu", stat->name,
4015 mem_cgroup_nr_lru_pages(memcg, stat->lru_mask,
4016 true));
4017 for_each_node_state(nid, N_MEMORY)
4018 seq_printf(m, " N%d=%lu", nid,
4019 mem_cgroup_node_nr_lru_pages(memcg, nid,
4020 stat->lru_mask, true));
Ying Han071aee12013-11-12 15:07:41 -08004021 seq_putc(m, '\n');
Ying Han406eb0c2011-05-26 16:25:37 -07004022 }
Ying Han406eb0c2011-05-26 16:25:37 -07004023
Ying Han406eb0c2011-05-26 16:25:37 -07004024 return 0;
4025}
4026#endif /* CONFIG_NUMA */
4027
Johannes Weinerc8713d02019-07-11 20:55:59 -07004028static const unsigned int memcg1_stats[] = {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07004029 NR_FILE_PAGES,
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07004030 NR_ANON_MAPPED,
Johannes Weiner468c3982020-06-03 16:02:01 -07004031#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4032 NR_ANON_THPS,
4033#endif
Johannes Weinerc8713d02019-07-11 20:55:59 -07004034 NR_SHMEM,
4035 NR_FILE_MAPPED,
4036 NR_FILE_DIRTY,
4037 NR_WRITEBACK,
4038 MEMCG_SWAP,
4039};
4040
4041static const char *const memcg1_stat_names[] = {
4042 "cache",
4043 "rss",
Johannes Weiner468c3982020-06-03 16:02:01 -07004044#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Johannes Weinerc8713d02019-07-11 20:55:59 -07004045 "rss_huge",
Johannes Weiner468c3982020-06-03 16:02:01 -07004046#endif
Johannes Weinerc8713d02019-07-11 20:55:59 -07004047 "shmem",
4048 "mapped_file",
4049 "dirty",
4050 "writeback",
4051 "swap",
4052};
4053
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07004054/* Universal VM events cgroup1 shows, original sort order */
Greg Thelen8dd53fd2018-06-07 17:07:23 -07004055static const unsigned int memcg1_events[] = {
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07004056 PGPGIN,
4057 PGPGOUT,
4058 PGFAULT,
4059 PGMAJFAULT,
4060};
4061
Tejun Heo2da8ca82013-12-05 12:28:04 -05004062static int memcg_stat_show(struct seq_file *m, void *v)
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004063{
Chris Downaa9694b2019-03-05 15:45:52 -08004064 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004065 unsigned long memory, memsw;
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004066 struct mem_cgroup *mi;
4067 unsigned int i;
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004068
Johannes Weiner71cd3112017-05-03 14:55:13 -07004069 BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats));
Rickard Strandqvist70bc0682014-12-12 16:56:41 -08004070
Johannes Weiner71cd3112017-05-03 14:55:13 -07004071 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
Johannes Weiner468c3982020-06-03 16:02:01 -07004072 unsigned long nr;
4073
Johannes Weiner71cd3112017-05-03 14:55:13 -07004074 if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004075 continue;
Johannes Weiner468c3982020-06-03 16:02:01 -07004076 nr = memcg_page_state_local(memcg, memcg1_stats[i]);
4077#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4078 if (memcg1_stats[i] == NR_ANON_THPS)
4079 nr *= HPAGE_PMD_NR;
4080#endif
4081 seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE);
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004082 }
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08004083
Johannes Weinerdf0e53d2017-05-03 14:55:10 -07004084 for (i = 0; i < ARRAY_SIZE(memcg1_events); i++)
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08004085 seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]),
Johannes Weiner205b20c2019-05-14 15:47:06 -07004086 memcg_events_local(memcg, memcg1_events[i]));
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004087
4088 for (i = 0; i < NR_LRU_LISTS; i++)
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08004089 seq_printf(m, "%s %lu\n", lru_list_name(i),
Johannes Weiner205b20c2019-05-14 15:47:06 -07004090 memcg_page_state_local(memcg, NR_LRU_BASE + i) *
Johannes Weiner21d89d12019-05-13 17:18:08 -07004091 PAGE_SIZE);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004092
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004093 /* Hierarchical information */
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004094 memory = memsw = PAGE_COUNTER_MAX;
4095 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
Chris Down15b42562020-04-01 21:07:20 -07004096 memory = min(memory, READ_ONCE(mi->memory.max));
4097 memsw = min(memsw, READ_ONCE(mi->memsw.max));
KAMEZAWA Hiroyukifee7b542009-01-07 18:08:26 -08004098 }
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004099 seq_printf(m, "hierarchical_memory_limit %llu\n",
4100 (u64)memory * PAGE_SIZE);
Johannes Weiner7941d212016-01-14 15:21:23 -08004101 if (do_memsw_account())
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004102 seq_printf(m, "hierarchical_memsw_limit %llu\n",
4103 (u64)memsw * PAGE_SIZE);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004104
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07004105 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
zhongjiang-ali7de2e9f2020-11-01 17:07:30 -08004106 unsigned long nr;
4107
Johannes Weiner71cd3112017-05-03 14:55:13 -07004108 if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account())
Daisuke Nishimura1dd3a272009-09-23 15:56:43 -07004109 continue;
zhongjiang-ali7de2e9f2020-11-01 17:07:30 -08004110 nr = memcg_page_state(memcg, memcg1_stats[i]);
4111#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4112 if (memcg1_stats[i] == NR_ANON_THPS)
4113 nr *= HPAGE_PMD_NR;
4114#endif
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07004115 seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i],
zhongjiang-ali7de2e9f2020-11-01 17:07:30 -08004116 (u64)nr * PAGE_SIZE);
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004117 }
4118
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07004119 for (i = 0; i < ARRAY_SIZE(memcg1_events); i++)
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08004120 seq_printf(m, "total_%s %llu\n",
4121 vm_event_name(memcg1_events[i]),
Yafang Shaodd923992019-07-11 20:52:11 -07004122 (u64)memcg_events(memcg, memcg1_events[i]));
Johannes Weineraf7c4b02012-05-29 15:07:08 -07004123
Shakeel Butt8de7ecc62018-08-21 21:53:17 -07004124 for (i = 0; i < NR_LRU_LISTS; i++)
Konstantin Khlebnikovebc5d83d2019-12-04 16:49:53 -08004125 seq_printf(m, "total_%s %llu\n", lru_list_name(i),
Johannes Weiner42a30032019-05-14 15:47:12 -07004126 (u64)memcg_page_state(memcg, NR_LRU_BASE + i) *
4127 PAGE_SIZE);
KAMEZAWA Hiroyuki14067bb2009-04-02 16:57:35 -07004128
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004129#ifdef CONFIG_DEBUG_VM
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004130 {
Mel Gormanef8f2322016-07-28 15:46:05 -07004131 pg_data_t *pgdat;
4132 struct mem_cgroup_per_node *mz;
Johannes Weiner1431d4d2020-06-03 16:02:53 -07004133 unsigned long anon_cost = 0;
4134 unsigned long file_cost = 0;
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004135
Mel Gormanef8f2322016-07-28 15:46:05 -07004136 for_each_online_pgdat(pgdat) {
4137 mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004138
Johannes Weiner1431d4d2020-06-03 16:02:53 -07004139 anon_cost += mz->lruvec.anon_cost;
4140 file_cost += mz->lruvec.file_cost;
Mel Gormanef8f2322016-07-28 15:46:05 -07004141 }
Johannes Weiner1431d4d2020-06-03 16:02:53 -07004142 seq_printf(m, "anon_cost %lu\n", anon_cost);
4143 seq_printf(m, "file_cost %lu\n", file_cost);
KOSAKI Motohiro7f016ee2009-01-07 18:08:22 -08004144 }
4145#endif
4146
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08004147 return 0;
4148}
4149
Tejun Heo182446d2013-08-08 20:11:24 -04004150static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
4151 struct cftype *cft)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004152{
Tejun Heo182446d2013-08-08 20:11:24 -04004153 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004154
KAMEZAWA Hiroyuki1f4c0252011-07-26 16:08:21 -07004155 return mem_cgroup_swappiness(memcg);
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004156}
4157
Tejun Heo182446d2013-08-08 20:11:24 -04004158static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
4159 struct cftype *cft, u64 val)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004160{
Tejun Heo182446d2013-08-08 20:11:24 -04004161 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Li Zefan068b38c2009-01-15 13:51:26 -08004162
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07004163 if (val > 100)
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004164 return -EINVAL;
4165
Linus Torvalds14208b02014-06-09 15:03:33 -07004166 if (css->parent)
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07004167 memcg->swappiness = val;
4168 else
4169 vm_swappiness = val;
Li Zefan068b38c2009-01-15 13:51:26 -08004170
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08004171 return 0;
4172}
4173
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004174static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4175{
4176 struct mem_cgroup_threshold_ary *t;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004177 unsigned long usage;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004178 int i;
4179
4180 rcu_read_lock();
4181 if (!swap)
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004182 t = rcu_dereference(memcg->thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004183 else
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004184 t = rcu_dereference(memcg->memsw_thresholds.primary);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004185
4186 if (!t)
4187 goto unlock;
4188
Johannes Weinerce00a962014-09-05 08:43:57 -04004189 usage = mem_cgroup_usage(memcg, swap);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004190
4191 /*
Sha Zhengju748dad32012-05-29 15:06:57 -07004192 * current_threshold points to threshold just below or equal to usage.
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004193 * If it's not true, a threshold was crossed after last
4194 * call of __mem_cgroup_threshold().
4195 */
Phil Carmody5407a562010-05-26 14:42:42 -07004196 i = t->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004197
4198 /*
4199 * Iterate backward over array of thresholds starting from
4200 * current_threshold and check if a threshold is crossed.
4201 * If none of thresholds below usage is crossed, we read
4202 * only one element of the array here.
4203 */
4204 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4205 eventfd_signal(t->entries[i].eventfd, 1);
4206
4207 /* i = current_threshold + 1 */
4208 i++;
4209
4210 /*
4211 * Iterate forward over array of thresholds starting from
4212 * current_threshold+1 and check if a threshold is crossed.
4213 * If none of thresholds above usage is crossed, we read
4214 * only one element of the array here.
4215 */
4216 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4217 eventfd_signal(t->entries[i].eventfd, 1);
4218
4219 /* Update current_threshold */
Phil Carmody5407a562010-05-26 14:42:42 -07004220 t->current_threshold = i - 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004221unlock:
4222 rcu_read_unlock();
4223}
4224
4225static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4226{
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004227 while (memcg) {
4228 __mem_cgroup_threshold(memcg, false);
Johannes Weiner7941d212016-01-14 15:21:23 -08004229 if (do_memsw_account())
Kirill A. Shutemovad4ca5f2010-10-07 12:59:27 -07004230 __mem_cgroup_threshold(memcg, true);
4231
4232 memcg = parent_mem_cgroup(memcg);
4233 }
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004234}
4235
4236static int compare_thresholds(const void *a, const void *b)
4237{
4238 const struct mem_cgroup_threshold *_a = a;
4239 const struct mem_cgroup_threshold *_b = b;
4240
Greg Thelen2bff24a2013-09-11 14:23:08 -07004241 if (_a->threshold > _b->threshold)
4242 return 1;
4243
4244 if (_a->threshold < _b->threshold)
4245 return -1;
4246
4247 return 0;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004248}
4249
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004250static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004251{
4252 struct mem_cgroup_eventfd_list *ev;
4253
Michal Hocko2bcf2e92014-07-30 16:08:33 -07004254 spin_lock(&memcg_oom_lock);
4255
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004256 list_for_each_entry(ev, &memcg->oom_notify, list)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004257 eventfd_signal(ev->eventfd, 1);
Michal Hocko2bcf2e92014-07-30 16:08:33 -07004258
4259 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004260 return 0;
4261}
4262
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004263static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004264{
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004265 struct mem_cgroup *iter;
4266
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004267 for_each_mem_cgroup_tree(iter, memcg)
KAMEZAWA Hiroyuki7d74b062010-10-27 15:33:41 -07004268 mem_cgroup_oom_notify_cb(iter);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004269}
4270
Tejun Heo59b6f872013-11-22 18:20:43 -05004271static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004272 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004273{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004274 struct mem_cgroup_thresholds *thresholds;
4275 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004276 unsigned long threshold;
4277 unsigned long usage;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004278 int i, size, ret;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004279
Johannes Weiner650c5e52015-02-11 15:26:03 -08004280 ret = page_counter_memparse(args, "-1", &threshold);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004281 if (ret)
4282 return ret;
4283
4284 mutex_lock(&memcg->thresholds_lock);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004285
Johannes Weiner05b84302014-08-06 16:05:59 -07004286 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004287 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004288 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07004289 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004290 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004291 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07004292 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004293 BUG();
4294
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004295 /* Check if a threshold crossed before adding a new one */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004296 if (thresholds->primary)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004297 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4298
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004299 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004300
4301 /* Allocate memory for new array of thresholds */
Gustavo A. R. Silva67b80462019-03-05 15:44:05 -08004302 new = kmalloc(struct_size(new, entries, size), GFP_KERNEL);
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004303 if (!new) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004304 ret = -ENOMEM;
4305 goto unlock;
4306 }
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004307 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004308
4309 /* Copy thresholds (if any) to new array */
Gustavo A. R. Silvae90342e2020-10-13 16:52:36 -07004310 if (thresholds->primary)
4311 memcpy(new->entries, thresholds->primary->entries,
4312 flex_array_size(new, entries, size - 1));
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004313
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004314 /* Add new threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004315 new->entries[size - 1].eventfd = eventfd;
4316 new->entries[size - 1].threshold = threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004317
4318 /* Sort thresholds. Registering of new threshold isn't time-critical */
Gustavo A. R. Silva61e604e62020-10-13 16:52:39 -07004319 sort(new->entries, size, sizeof(*new->entries),
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004320 compare_thresholds, NULL);
4321
4322 /* Find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004323 new->current_threshold = -1;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004324 for (i = 0; i < size; i++) {
Sha Zhengju748dad32012-05-29 15:06:57 -07004325 if (new->entries[i].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004326 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004327 * new->current_threshold will not be used until
4328 * rcu_assign_pointer(), so it's safe to increment
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004329 * it here.
4330 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004331 ++new->current_threshold;
Sha Zhengju748dad32012-05-29 15:06:57 -07004332 } else
4333 break;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004334 }
4335
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004336 /* Free old spare buffer and save old primary buffer as spare */
4337 kfree(thresholds->spare);
4338 thresholds->spare = thresholds->primary;
4339
4340 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004341
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004342 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004343 synchronize_rcu();
4344
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004345unlock:
4346 mutex_unlock(&memcg->thresholds_lock);
4347
4348 return ret;
4349}
4350
Tejun Heo59b6f872013-11-22 18:20:43 -05004351static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004352 struct eventfd_ctx *eventfd, const char *args)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004353{
Tejun Heo59b6f872013-11-22 18:20:43 -05004354 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05004355}
4356
Tejun Heo59b6f872013-11-22 18:20:43 -05004357static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004358 struct eventfd_ctx *eventfd, const char *args)
4359{
Tejun Heo59b6f872013-11-22 18:20:43 -05004360 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05004361}
4362
Tejun Heo59b6f872013-11-22 18:20:43 -05004363static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004364 struct eventfd_ctx *eventfd, enum res_type type)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004365{
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004366 struct mem_cgroup_thresholds *thresholds;
4367 struct mem_cgroup_threshold_ary *new;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08004368 unsigned long usage;
Chunguang Xu7d366652020-03-21 18:22:10 -07004369 int i, j, size, entries;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004370
4371 mutex_lock(&memcg->thresholds_lock);
Johannes Weiner05b84302014-08-06 16:05:59 -07004372
4373 if (type == _MEM) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004374 thresholds = &memcg->thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004375 usage = mem_cgroup_usage(memcg, false);
Johannes Weiner05b84302014-08-06 16:05:59 -07004376 } else if (type == _MEMSWAP) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004377 thresholds = &memcg->memsw_thresholds;
Johannes Weinerce00a962014-09-05 08:43:57 -04004378 usage = mem_cgroup_usage(memcg, true);
Johannes Weiner05b84302014-08-06 16:05:59 -07004379 } else
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004380 BUG();
4381
Anton Vorontsov371528c2012-02-24 05:14:46 +04004382 if (!thresholds->primary)
4383 goto unlock;
4384
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004385 /* Check if a threshold crossed before removing */
4386 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4387
4388 /* Calculate new number of threshold */
Chunguang Xu7d366652020-03-21 18:22:10 -07004389 size = entries = 0;
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004390 for (i = 0; i < thresholds->primary->size; i++) {
4391 if (thresholds->primary->entries[i].eventfd != eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004392 size++;
Chunguang Xu7d366652020-03-21 18:22:10 -07004393 else
4394 entries++;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004395 }
4396
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004397 new = thresholds->spare;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004398
Chunguang Xu7d366652020-03-21 18:22:10 -07004399 /* If no items related to eventfd have been cleared, nothing to do */
4400 if (!entries)
4401 goto unlock;
4402
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004403 /* Set thresholds array to NULL if we don't have thresholds */
4404 if (!size) {
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004405 kfree(new);
4406 new = NULL;
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004407 goto swap_buffers;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004408 }
4409
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004410 new->size = size;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004411
4412 /* Copy thresholds and find current threshold */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004413 new->current_threshold = -1;
4414 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4415 if (thresholds->primary->entries[i].eventfd == eventfd)
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004416 continue;
4417
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004418 new->entries[j] = thresholds->primary->entries[i];
Sha Zhengju748dad32012-05-29 15:06:57 -07004419 if (new->entries[j].threshold <= usage) {
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004420 /*
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004421 * new->current_threshold will not be used
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004422 * until rcu_assign_pointer(), so it's safe to increment
4423 * it here.
4424 */
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004425 ++new->current_threshold;
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004426 }
4427 j++;
4428 }
4429
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004430swap_buffers:
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004431 /* Swap primary and spare array */
4432 thresholds->spare = thresholds->primary;
Sha Zhengju8c757762012-05-10 13:01:45 -07004433
Kirill A. Shutemov2c488db2010-05-26 14:42:47 -07004434 rcu_assign_pointer(thresholds->primary, new);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004435
Kirill A. Shutemov907860e2010-05-26 14:42:46 -07004436 /* To be sure that nobody uses thresholds */
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004437 synchronize_rcu();
Martijn Coenen6611d8d2016-01-15 16:57:49 -08004438
4439 /* If all events are unregistered, free the spare array */
4440 if (!new) {
4441 kfree(thresholds->spare);
4442 thresholds->spare = NULL;
4443 }
Anton Vorontsov371528c2012-02-24 05:14:46 +04004444unlock:
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004445 mutex_unlock(&memcg->thresholds_lock);
Kirill A. Shutemov2e72b632010-03-10 15:22:24 -08004446}
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08004447
Tejun Heo59b6f872013-11-22 18:20:43 -05004448static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004449 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004450{
Tejun Heo59b6f872013-11-22 18:20:43 -05004451 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
Tejun Heo347c4a82013-11-22 18:20:43 -05004452}
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004453
Tejun Heo59b6f872013-11-22 18:20:43 -05004454static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004455 struct eventfd_ctx *eventfd)
4456{
Tejun Heo59b6f872013-11-22 18:20:43 -05004457 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
Tejun Heo347c4a82013-11-22 18:20:43 -05004458}
4459
Tejun Heo59b6f872013-11-22 18:20:43 -05004460static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004461 struct eventfd_ctx *eventfd, const char *args)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004462{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004463 struct mem_cgroup_eventfd_list *event;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004464
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004465 event = kmalloc(sizeof(*event), GFP_KERNEL);
4466 if (!event)
4467 return -ENOMEM;
4468
Michal Hocko1af8efe2011-07-26 16:08:24 -07004469 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004470
4471 event->eventfd = eventfd;
4472 list_add(&event->list, &memcg->oom_notify);
4473
4474 /* already in OOM ? */
Tejun Heoc2b42d32015-06-24 16:58:23 -07004475 if (memcg->under_oom)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004476 eventfd_signal(eventfd, 1);
Michal Hocko1af8efe2011-07-26 16:08:24 -07004477 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004478
4479 return 0;
4480}
4481
Tejun Heo59b6f872013-11-22 18:20:43 -05004482static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
Tejun Heo347c4a82013-11-22 18:20:43 -05004483 struct eventfd_ctx *eventfd)
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004484{
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004485 struct mem_cgroup_eventfd_list *ev, *tmp;
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004486
Michal Hocko1af8efe2011-07-26 16:08:24 -07004487 spin_lock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004488
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004489 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004490 if (ev->eventfd == eventfd) {
4491 list_del(&ev->list);
4492 kfree(ev);
4493 }
4494 }
4495
Michal Hocko1af8efe2011-07-26 16:08:24 -07004496 spin_unlock(&memcg_oom_lock);
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07004497}
4498
Tejun Heo2da8ca82013-12-05 12:28:04 -05004499static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004500{
Chris Downaa9694b2019-03-05 15:45:52 -08004501 struct mem_cgroup *memcg = mem_cgroup_from_seq(sf);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004502
Tejun Heo791badb2013-12-05 12:28:02 -05004503 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
Tejun Heoc2b42d32015-06-24 16:58:23 -07004504 seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
Roman Gushchinfe6bdfc2018-06-14 15:28:05 -07004505 seq_printf(sf, "oom_kill %lu\n",
4506 atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL]));
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004507 return 0;
4508}
4509
Tejun Heo182446d2013-08-08 20:11:24 -04004510static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004511 struct cftype *cft, u64 val)
4512{
Tejun Heo182446d2013-08-08 20:11:24 -04004513 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004514
4515 /* cannot set to root cgroup and only 0 and 1 are allowed */
Linus Torvalds14208b02014-06-09 15:03:33 -07004516 if (!css->parent || !((val == 0) || (val == 1)))
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004517 return -EINVAL;
4518
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004519 memcg->oom_kill_disable = val;
KAMEZAWA Hiroyuki4d845eb2010-06-29 15:05:18 -07004520 if (!val)
Raghavendra K Tc0ff4b82011-11-02 13:38:15 -07004521 memcg_oom_recover(memcg);
Johannes Weiner3dae7fe2014-06-04 16:07:01 -07004522
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07004523 return 0;
4524}
4525
Tejun Heo52ebea72015-05-22 17:13:37 -04004526#ifdef CONFIG_CGROUP_WRITEBACK
4527
Tejun Heo3a8e9ac2019-08-29 15:47:19 -07004528#include <trace/events/writeback.h>
4529
Tejun Heo841710a2015-05-22 18:23:33 -04004530static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
4531{
4532 return wb_domain_init(&memcg->cgwb_domain, gfp);
4533}
4534
4535static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
4536{
4537 wb_domain_exit(&memcg->cgwb_domain);
4538}
4539
Tejun Heo2529bb32015-05-22 18:23:34 -04004540static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
4541{
4542 wb_domain_size_changed(&memcg->cgwb_domain);
4543}
4544
Tejun Heo841710a2015-05-22 18:23:33 -04004545struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
4546{
4547 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
4548
4549 if (!memcg->css.parent)
4550 return NULL;
4551
4552 return &memcg->cgwb_domain;
4553}
4554
Greg Thelen0b3d6e62019-04-05 18:39:18 -07004555/*
4556 * idx can be of type enum memcg_stat_item or node_stat_item.
4557 * Keep in sync with memcg_exact_page().
4558 */
4559static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx)
4560{
Chris Down871789d2019-05-14 15:46:57 -07004561 long x = atomic_long_read(&memcg->vmstats[idx]);
Greg Thelen0b3d6e62019-04-05 18:39:18 -07004562 int cpu;
4563
4564 for_each_online_cpu(cpu)
Chris Down871789d2019-05-14 15:46:57 -07004565 x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx];
Greg Thelen0b3d6e62019-04-05 18:39:18 -07004566 if (x < 0)
4567 x = 0;
4568 return x;
4569}
4570
Tejun Heoc2aa7232015-05-22 18:23:35 -04004571/**
4572 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
4573 * @wb: bdi_writeback in question
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004574 * @pfilepages: out parameter for number of file pages
4575 * @pheadroom: out parameter for number of allocatable pages according to memcg
Tejun Heoc2aa7232015-05-22 18:23:35 -04004576 * @pdirty: out parameter for number of dirty pages
4577 * @pwriteback: out parameter for number of pages under writeback
4578 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004579 * Determine the numbers of file, headroom, dirty, and writeback pages in
4580 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
4581 * is a bit more involved.
Tejun Heoc2aa7232015-05-22 18:23:35 -04004582 *
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004583 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
4584 * headroom is calculated as the lowest headroom of itself and the
4585 * ancestors. Note that this doesn't consider the actual amount of
4586 * available memory in the system. The caller should further cap
4587 * *@pheadroom accordingly.
Tejun Heoc2aa7232015-05-22 18:23:35 -04004588 */
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004589void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
4590 unsigned long *pheadroom, unsigned long *pdirty,
4591 unsigned long *pwriteback)
Tejun Heoc2aa7232015-05-22 18:23:35 -04004592{
4593 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
4594 struct mem_cgroup *parent;
Tejun Heoc2aa7232015-05-22 18:23:35 -04004595
Greg Thelen0b3d6e62019-04-05 18:39:18 -07004596 *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY);
Tejun Heoc2aa7232015-05-22 18:23:35 -04004597
Greg Thelen0b3d6e62019-04-05 18:39:18 -07004598 *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK);
Johannes Weiner21d89d12019-05-13 17:18:08 -07004599 *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) +
4600 memcg_exact_page_state(memcg, NR_ACTIVE_FILE);
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004601 *pheadroom = PAGE_COUNTER_MAX;
Tejun Heoc2aa7232015-05-22 18:23:35 -04004602
Tejun Heoc2aa7232015-05-22 18:23:35 -04004603 while ((parent = parent_mem_cgroup(memcg))) {
Chris Down15b42562020-04-01 21:07:20 -07004604 unsigned long ceiling = min(READ_ONCE(memcg->memory.max),
Jakub Kicinskid1663a92020-06-01 21:49:49 -07004605 READ_ONCE(memcg->memory.high));
Tejun Heoc2aa7232015-05-22 18:23:35 -04004606 unsigned long used = page_counter_read(&memcg->memory);
4607
Tejun Heoc5edf9c2015-09-29 13:04:26 -04004608 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
Tejun Heoc2aa7232015-05-22 18:23:35 -04004609 memcg = parent;
4610 }
Tejun Heoc2aa7232015-05-22 18:23:35 -04004611}
4612
Tejun Heo97b27822019-08-26 09:06:56 -07004613/*
4614 * Foreign dirty flushing
4615 *
4616 * There's an inherent mismatch between memcg and writeback. The former
4617 * trackes ownership per-page while the latter per-inode. This was a
4618 * deliberate design decision because honoring per-page ownership in the
4619 * writeback path is complicated, may lead to higher CPU and IO overheads
4620 * and deemed unnecessary given that write-sharing an inode across
4621 * different cgroups isn't a common use-case.
4622 *
4623 * Combined with inode majority-writer ownership switching, this works well
4624 * enough in most cases but there are some pathological cases. For
4625 * example, let's say there are two cgroups A and B which keep writing to
4626 * different but confined parts of the same inode. B owns the inode and
4627 * A's memory is limited far below B's. A's dirty ratio can rise enough to
4628 * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid
4629 * triggering background writeback. A will be slowed down without a way to
4630 * make writeback of the dirty pages happen.
4631 *
4632 * Conditions like the above can lead to a cgroup getting repatedly and
4633 * severely throttled after making some progress after each
4634 * dirty_expire_interval while the underyling IO device is almost
4635 * completely idle.
4636 *
4637 * Solving this problem completely requires matching the ownership tracking
4638 * granularities between memcg and writeback in either direction. However,
4639 * the more egregious behaviors can be avoided by simply remembering the
4640 * most recent foreign dirtying events and initiating remote flushes on
4641 * them when local writeback isn't enough to keep the memory clean enough.
4642 *
4643 * The following two functions implement such mechanism. When a foreign
4644 * page - a page whose memcg and writeback ownerships don't match - is
4645 * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning
4646 * bdi_writeback on the page owning memcg. When balance_dirty_pages()
4647 * decides that the memcg needs to sleep due to high dirty ratio, it calls
4648 * mem_cgroup_flush_foreign() which queues writeback on the recorded
4649 * foreign bdi_writebacks which haven't expired. Both the numbers of
4650 * recorded bdi_writebacks and concurrent in-flight foreign writebacks are
4651 * limited to MEMCG_CGWB_FRN_CNT.
4652 *
4653 * The mechanism only remembers IDs and doesn't hold any object references.
4654 * As being wrong occasionally doesn't matter, updates and accesses to the
4655 * records are lockless and racy.
4656 */
4657void mem_cgroup_track_foreign_dirty_slowpath(struct page *page,
4658 struct bdi_writeback *wb)
4659{
4660 struct mem_cgroup *memcg = page->mem_cgroup;
4661 struct memcg_cgwb_frn *frn;
4662 u64 now = get_jiffies_64();
4663 u64 oldest_at = now;
4664 int oldest = -1;
4665 int i;
4666
Tejun Heo3a8e9ac2019-08-29 15:47:19 -07004667 trace_track_foreign_dirty(page, wb);
4668
Tejun Heo97b27822019-08-26 09:06:56 -07004669 /*
4670 * Pick the slot to use. If there is already a slot for @wb, keep
4671 * using it. If not replace the oldest one which isn't being
4672 * written out.
4673 */
4674 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) {
4675 frn = &memcg->cgwb_frn[i];
4676 if (frn->bdi_id == wb->bdi->id &&
4677 frn->memcg_id == wb->memcg_css->id)
4678 break;
4679 if (time_before64(frn->at, oldest_at) &&
4680 atomic_read(&frn->done.cnt) == 1) {
4681 oldest = i;
4682 oldest_at = frn->at;
4683 }
4684 }
4685
4686 if (i < MEMCG_CGWB_FRN_CNT) {
4687 /*
4688 * Re-using an existing one. Update timestamp lazily to
4689 * avoid making the cacheline hot. We want them to be
4690 * reasonably up-to-date and significantly shorter than
4691 * dirty_expire_interval as that's what expires the record.
4692 * Use the shorter of 1s and dirty_expire_interval / 8.
4693 */
4694 unsigned long update_intv =
4695 min_t(unsigned long, HZ,
4696 msecs_to_jiffies(dirty_expire_interval * 10) / 8);
4697
4698 if (time_before64(frn->at, now - update_intv))
4699 frn->at = now;
4700 } else if (oldest >= 0) {
4701 /* replace the oldest free one */
4702 frn = &memcg->cgwb_frn[oldest];
4703 frn->bdi_id = wb->bdi->id;
4704 frn->memcg_id = wb->memcg_css->id;
4705 frn->at = now;
4706 }
4707}
4708
4709/* issue foreign writeback flushes for recorded foreign dirtying events */
4710void mem_cgroup_flush_foreign(struct bdi_writeback *wb)
4711{
4712 struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
4713 unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10);
4714 u64 now = jiffies_64;
4715 int i;
4716
4717 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) {
4718 struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i];
4719
4720 /*
4721 * If the record is older than dirty_expire_interval,
4722 * writeback on it has already started. No need to kick it
4723 * off again. Also, don't start a new one if there's
4724 * already one in flight.
4725 */
4726 if (time_after64(frn->at, now - intv) &&
4727 atomic_read(&frn->done.cnt) == 1) {
4728 frn->at = 0;
Tejun Heo3a8e9ac2019-08-29 15:47:19 -07004729 trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id);
Tejun Heo97b27822019-08-26 09:06:56 -07004730 cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0,
4731 WB_REASON_FOREIGN_FLUSH,
4732 &frn->done);
4733 }
4734 }
4735}
4736
Tejun Heo841710a2015-05-22 18:23:33 -04004737#else /* CONFIG_CGROUP_WRITEBACK */
4738
4739static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
4740{
4741 return 0;
4742}
4743
4744static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
4745{
4746}
4747
Tejun Heo2529bb32015-05-22 18:23:34 -04004748static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
4749{
4750}
4751
Tejun Heo52ebea72015-05-22 17:13:37 -04004752#endif /* CONFIG_CGROUP_WRITEBACK */
4753
Tejun Heo79bd9812013-11-22 18:20:42 -05004754/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05004755 * DO NOT USE IN NEW FILES.
4756 *
4757 * "cgroup.event_control" implementation.
4758 *
4759 * This is way over-engineered. It tries to support fully configurable
4760 * events for each user. Such level of flexibility is completely
4761 * unnecessary especially in the light of the planned unified hierarchy.
4762 *
4763 * Please deprecate this and replace with something simpler if at all
4764 * possible.
4765 */
4766
4767/*
Tejun Heo79bd9812013-11-22 18:20:42 -05004768 * Unregister event and free resources.
4769 *
4770 * Gets called from workqueue.
4771 */
Tejun Heo3bc942f2013-11-22 18:20:44 -05004772static void memcg_event_remove(struct work_struct *work)
Tejun Heo79bd9812013-11-22 18:20:42 -05004773{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004774 struct mem_cgroup_event *event =
4775 container_of(work, struct mem_cgroup_event, remove);
Tejun Heo59b6f872013-11-22 18:20:43 -05004776 struct mem_cgroup *memcg = event->memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05004777
4778 remove_wait_queue(event->wqh, &event->wait);
4779
Tejun Heo59b6f872013-11-22 18:20:43 -05004780 event->unregister_event(memcg, event->eventfd);
Tejun Heo79bd9812013-11-22 18:20:42 -05004781
4782 /* Notify userspace the event is going away. */
4783 eventfd_signal(event->eventfd, 1);
4784
4785 eventfd_ctx_put(event->eventfd);
4786 kfree(event);
Tejun Heo59b6f872013-11-22 18:20:43 -05004787 css_put(&memcg->css);
Tejun Heo79bd9812013-11-22 18:20:42 -05004788}
4789
4790/*
Linus Torvaldsa9a08842018-02-11 14:34:03 -08004791 * Gets called on EPOLLHUP on eventfd when user closes it.
Tejun Heo79bd9812013-11-22 18:20:42 -05004792 *
4793 * Called with wqh->lock held and interrupts disabled.
4794 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02004795static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode,
Tejun Heo3bc942f2013-11-22 18:20:44 -05004796 int sync, void *key)
Tejun Heo79bd9812013-11-22 18:20:42 -05004797{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004798 struct mem_cgroup_event *event =
4799 container_of(wait, struct mem_cgroup_event, wait);
Tejun Heo59b6f872013-11-22 18:20:43 -05004800 struct mem_cgroup *memcg = event->memcg;
Al Viro3ad6f932017-07-03 20:14:56 -04004801 __poll_t flags = key_to_poll(key);
Tejun Heo79bd9812013-11-22 18:20:42 -05004802
Linus Torvaldsa9a08842018-02-11 14:34:03 -08004803 if (flags & EPOLLHUP) {
Tejun Heo79bd9812013-11-22 18:20:42 -05004804 /*
4805 * If the event has been detached at cgroup removal, we
4806 * can simply return knowing the other side will cleanup
4807 * for us.
4808 *
4809 * We can't race against event freeing since the other
4810 * side will require wqh->lock via remove_wait_queue(),
4811 * which we hold.
4812 */
Tejun Heofba94802013-11-22 18:20:43 -05004813 spin_lock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004814 if (!list_empty(&event->list)) {
4815 list_del_init(&event->list);
4816 /*
4817 * We are in atomic context, but cgroup_event_remove()
4818 * may sleep, so we have to call it in workqueue.
4819 */
4820 schedule_work(&event->remove);
4821 }
Tejun Heofba94802013-11-22 18:20:43 -05004822 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004823 }
4824
4825 return 0;
4826}
4827
Tejun Heo3bc942f2013-11-22 18:20:44 -05004828static void memcg_event_ptable_queue_proc(struct file *file,
Tejun Heo79bd9812013-11-22 18:20:42 -05004829 wait_queue_head_t *wqh, poll_table *pt)
4830{
Tejun Heo3bc942f2013-11-22 18:20:44 -05004831 struct mem_cgroup_event *event =
4832 container_of(pt, struct mem_cgroup_event, pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05004833
4834 event->wqh = wqh;
4835 add_wait_queue(wqh, &event->wait);
4836}
4837
4838/*
Tejun Heo3bc942f2013-11-22 18:20:44 -05004839 * DO NOT USE IN NEW FILES.
4840 *
Tejun Heo79bd9812013-11-22 18:20:42 -05004841 * Parse input and register new cgroup event handler.
4842 *
4843 * Input must be in format '<event_fd> <control_fd> <args>'.
4844 * Interpretation of args is defined by control file implementation.
4845 */
Tejun Heo451af502014-05-13 12:16:21 -04004846static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
4847 char *buf, size_t nbytes, loff_t off)
Tejun Heo79bd9812013-11-22 18:20:42 -05004848{
Tejun Heo451af502014-05-13 12:16:21 -04004849 struct cgroup_subsys_state *css = of_css(of);
Tejun Heofba94802013-11-22 18:20:43 -05004850 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004851 struct mem_cgroup_event *event;
Tejun Heo79bd9812013-11-22 18:20:42 -05004852 struct cgroup_subsys_state *cfile_css;
4853 unsigned int efd, cfd;
4854 struct fd efile;
4855 struct fd cfile;
Tejun Heofba94802013-11-22 18:20:43 -05004856 const char *name;
Tejun Heo79bd9812013-11-22 18:20:42 -05004857 char *endp;
4858 int ret;
4859
Tejun Heo451af502014-05-13 12:16:21 -04004860 buf = strstrip(buf);
4861
4862 efd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05004863 if (*endp != ' ')
4864 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04004865 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05004866
Tejun Heo451af502014-05-13 12:16:21 -04004867 cfd = simple_strtoul(buf, &endp, 10);
Tejun Heo79bd9812013-11-22 18:20:42 -05004868 if ((*endp != ' ') && (*endp != '\0'))
4869 return -EINVAL;
Tejun Heo451af502014-05-13 12:16:21 -04004870 buf = endp + 1;
Tejun Heo79bd9812013-11-22 18:20:42 -05004871
4872 event = kzalloc(sizeof(*event), GFP_KERNEL);
4873 if (!event)
4874 return -ENOMEM;
4875
Tejun Heo59b6f872013-11-22 18:20:43 -05004876 event->memcg = memcg;
Tejun Heo79bd9812013-11-22 18:20:42 -05004877 INIT_LIST_HEAD(&event->list);
Tejun Heo3bc942f2013-11-22 18:20:44 -05004878 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
4879 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
4880 INIT_WORK(&event->remove, memcg_event_remove);
Tejun Heo79bd9812013-11-22 18:20:42 -05004881
4882 efile = fdget(efd);
4883 if (!efile.file) {
4884 ret = -EBADF;
4885 goto out_kfree;
4886 }
4887
4888 event->eventfd = eventfd_ctx_fileget(efile.file);
4889 if (IS_ERR(event->eventfd)) {
4890 ret = PTR_ERR(event->eventfd);
4891 goto out_put_efile;
4892 }
4893
4894 cfile = fdget(cfd);
4895 if (!cfile.file) {
4896 ret = -EBADF;
4897 goto out_put_eventfd;
4898 }
4899
4900 /* the process need read permission on control file */
4901 /* AV: shouldn't we check that it's been opened for read instead? */
4902 ret = inode_permission(file_inode(cfile.file), MAY_READ);
4903 if (ret < 0)
4904 goto out_put_cfile;
4905
Tejun Heo79bd9812013-11-22 18:20:42 -05004906 /*
Tejun Heofba94802013-11-22 18:20:43 -05004907 * Determine the event callbacks and set them in @event. This used
4908 * to be done via struct cftype but cgroup core no longer knows
4909 * about these events. The following is crude but the whole thing
4910 * is for compatibility anyway.
Tejun Heo3bc942f2013-11-22 18:20:44 -05004911 *
4912 * DO NOT ADD NEW FILES.
Tejun Heofba94802013-11-22 18:20:43 -05004913 */
Al Virob5830432014-10-31 01:22:04 -04004914 name = cfile.file->f_path.dentry->d_name.name;
Tejun Heofba94802013-11-22 18:20:43 -05004915
4916 if (!strcmp(name, "memory.usage_in_bytes")) {
4917 event->register_event = mem_cgroup_usage_register_event;
4918 event->unregister_event = mem_cgroup_usage_unregister_event;
4919 } else if (!strcmp(name, "memory.oom_control")) {
4920 event->register_event = mem_cgroup_oom_register_event;
4921 event->unregister_event = mem_cgroup_oom_unregister_event;
4922 } else if (!strcmp(name, "memory.pressure_level")) {
4923 event->register_event = vmpressure_register_event;
4924 event->unregister_event = vmpressure_unregister_event;
4925 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
Tejun Heo347c4a82013-11-22 18:20:43 -05004926 event->register_event = memsw_cgroup_usage_register_event;
4927 event->unregister_event = memsw_cgroup_usage_unregister_event;
Tejun Heofba94802013-11-22 18:20:43 -05004928 } else {
4929 ret = -EINVAL;
4930 goto out_put_cfile;
4931 }
4932
4933 /*
Tejun Heob5557c42013-11-22 18:20:42 -05004934 * Verify @cfile should belong to @css. Also, remaining events are
4935 * automatically removed on cgroup destruction but the removal is
4936 * asynchronous, so take an extra ref on @css.
Tejun Heo79bd9812013-11-22 18:20:42 -05004937 */
Al Virob5830432014-10-31 01:22:04 -04004938 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
Tejun Heoec903c02014-05-13 12:11:01 -04004939 &memory_cgrp_subsys);
Tejun Heo79bd9812013-11-22 18:20:42 -05004940 ret = -EINVAL;
Tejun Heo5a17f542014-02-11 11:52:47 -05004941 if (IS_ERR(cfile_css))
Tejun Heo79bd9812013-11-22 18:20:42 -05004942 goto out_put_cfile;
Tejun Heo5a17f542014-02-11 11:52:47 -05004943 if (cfile_css != css) {
4944 css_put(cfile_css);
4945 goto out_put_cfile;
4946 }
Tejun Heo79bd9812013-11-22 18:20:42 -05004947
Tejun Heo451af502014-05-13 12:16:21 -04004948 ret = event->register_event(memcg, event->eventfd, buf);
Tejun Heo79bd9812013-11-22 18:20:42 -05004949 if (ret)
4950 goto out_put_css;
4951
Christoph Hellwig9965ed172018-03-05 07:26:05 -08004952 vfs_poll(efile.file, &event->pt);
Tejun Heo79bd9812013-11-22 18:20:42 -05004953
Tejun Heofba94802013-11-22 18:20:43 -05004954 spin_lock(&memcg->event_list_lock);
4955 list_add(&event->list, &memcg->event_list);
4956 spin_unlock(&memcg->event_list_lock);
Tejun Heo79bd9812013-11-22 18:20:42 -05004957
4958 fdput(cfile);
4959 fdput(efile);
4960
Tejun Heo451af502014-05-13 12:16:21 -04004961 return nbytes;
Tejun Heo79bd9812013-11-22 18:20:42 -05004962
4963out_put_css:
Tejun Heob5557c42013-11-22 18:20:42 -05004964 css_put(css);
Tejun Heo79bd9812013-11-22 18:20:42 -05004965out_put_cfile:
4966 fdput(cfile);
4967out_put_eventfd:
4968 eventfd_ctx_put(event->eventfd);
4969out_put_efile:
4970 fdput(efile);
4971out_kfree:
4972 kfree(event);
4973
4974 return ret;
4975}
4976
Johannes Weiner241994ed2015-02-11 15:26:06 -08004977static struct cftype mem_cgroup_legacy_files[] = {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004978 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004979 .name = "usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004980 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05004981 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004982 },
4983 {
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004984 .name = "max_usage_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004985 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04004986 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05004987 .read_u64 = mem_cgroup_read_u64,
Pavel Emelyanovc84872e2008-04-29 01:00:17 -07004988 },
4989 {
Balbir Singh0eea1032008-02-07 00:13:57 -08004990 .name = "limit_in_bytes",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08004991 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004992 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004993 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08004994 },
4995 {
Balbir Singh296c81d2009-09-23 15:56:36 -07004996 .name = "soft_limit_in_bytes",
4997 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04004998 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05004999 .read_u64 = mem_cgroup_read_u64,
Balbir Singh296c81d2009-09-23 15:56:36 -07005000 },
5001 {
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005002 .name = "failcnt",
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005003 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04005004 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005005 .read_u64 = mem_cgroup_read_u64,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005006 },
Balbir Singh8697d332008-02-07 00:13:59 -08005007 {
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005008 .name = "stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05005009 .seq_show = memcg_stat_show,
KAMEZAWA Hiroyukid2ceb9b2008-02-07 00:14:25 -08005010 },
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005011 {
5012 .name = "force_empty",
Tejun Heo6770c642014-05-13 12:16:21 -04005013 .write = mem_cgroup_force_empty_write,
KAMEZAWA Hiroyukic1e862c2009-01-07 18:07:55 -08005014 },
Balbir Singh18f59ea2009-01-07 18:08:07 -08005015 {
5016 .name = "use_hierarchy",
5017 .write_u64 = mem_cgroup_hierarchy_write,
5018 .read_u64 = mem_cgroup_hierarchy_read,
5019 },
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005020 {
Tejun Heo3bc942f2013-11-22 18:20:44 -05005021 .name = "cgroup.event_control", /* XXX: for compat */
Tejun Heo451af502014-05-13 12:16:21 -04005022 .write = memcg_write_event_control,
Tejun Heo7dbdb192015-09-18 17:54:23 -04005023 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
Tejun Heo79bd9812013-11-22 18:20:42 -05005024 },
5025 {
KOSAKI Motohiroa7885eb2009-01-07 18:08:24 -08005026 .name = "swappiness",
5027 .read_u64 = mem_cgroup_swappiness_read,
5028 .write_u64 = mem_cgroup_swappiness_write,
5029 },
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005030 {
5031 .name = "move_charge_at_immigrate",
5032 .read_u64 = mem_cgroup_move_charge_read,
5033 .write_u64 = mem_cgroup_move_charge_write,
5034 },
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005035 {
5036 .name = "oom_control",
Tejun Heo2da8ca82013-12-05 12:28:04 -05005037 .seq_show = mem_cgroup_oom_control_read,
KAMEZAWA Hiroyuki3c11ecf2010-05-26 14:42:37 -07005038 .write_u64 = mem_cgroup_oom_control_write,
KAMEZAWA Hiroyuki9490ff22010-05-26 14:42:36 -07005039 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
5040 },
Anton Vorontsov70ddf632013-04-29 15:08:31 -07005041 {
5042 .name = "pressure_level",
Anton Vorontsov70ddf632013-04-29 15:08:31 -07005043 },
Ying Han406eb0c2011-05-26 16:25:37 -07005044#ifdef CONFIG_NUMA
5045 {
5046 .name = "numa_stat",
Tejun Heo2da8ca82013-12-05 12:28:04 -05005047 .seq_show = memcg_numa_stat_show,
Ying Han406eb0c2011-05-26 16:25:37 -07005048 },
5049#endif
Glauber Costa510fc4e2012-12-18 14:21:47 -08005050 {
5051 .name = "kmem.limit_in_bytes",
5052 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
Tejun Heo451af502014-05-13 12:16:21 -04005053 .write = mem_cgroup_write,
Tejun Heo791badb2013-12-05 12:28:02 -05005054 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08005055 },
5056 {
5057 .name = "kmem.usage_in_bytes",
5058 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
Tejun Heo791badb2013-12-05 12:28:02 -05005059 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08005060 },
5061 {
5062 .name = "kmem.failcnt",
5063 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
Tejun Heo6770c642014-05-13 12:16:21 -04005064 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005065 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08005066 },
5067 {
5068 .name = "kmem.max_usage_in_bytes",
5069 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
Tejun Heo6770c642014-05-13 12:16:21 -04005070 .write = mem_cgroup_reset,
Tejun Heo791badb2013-12-05 12:28:02 -05005071 .read_u64 = mem_cgroup_read_u64,
Glauber Costa510fc4e2012-12-18 14:21:47 -08005072 },
Yafang Shaoa87425a2020-04-01 21:06:30 -07005073#if defined(CONFIG_MEMCG_KMEM) && \
5074 (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG))
Glauber Costa749c5412012-12-18 14:23:01 -08005075 {
5076 .name = "kmem.slabinfo",
Vladimir Davydovb0475012014-12-10 15:44:19 -08005077 .seq_show = memcg_slab_show,
Glauber Costa749c5412012-12-18 14:23:01 -08005078 },
5079#endif
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08005080 {
5081 .name = "kmem.tcp.limit_in_bytes",
5082 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
5083 .write = mem_cgroup_write,
5084 .read_u64 = mem_cgroup_read_u64,
5085 },
5086 {
5087 .name = "kmem.tcp.usage_in_bytes",
5088 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
5089 .read_u64 = mem_cgroup_read_u64,
5090 },
5091 {
5092 .name = "kmem.tcp.failcnt",
5093 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
5094 .write = mem_cgroup_reset,
5095 .read_u64 = mem_cgroup_read_u64,
5096 },
5097 {
5098 .name = "kmem.tcp.max_usage_in_bytes",
5099 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
5100 .write = mem_cgroup_reset,
5101 .read_u64 = mem_cgroup_read_u64,
5102 },
Tejun Heo6bc10342012-04-01 12:09:55 -07005103 { }, /* terminate */
Tejun Heoaf36f902012-04-01 12:09:55 -07005104};
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -08005105
Johannes Weiner73f576c2016-07-20 15:44:57 -07005106/*
5107 * Private memory cgroup IDR
5108 *
5109 * Swap-out records and page cache shadow entries need to store memcg
5110 * references in constrained space, so we maintain an ID space that is
5111 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
5112 * memory-controlled cgroups to 64k.
5113 *
Ethon Paulb8f29352020-06-04 16:49:28 -07005114 * However, there usually are many references to the offline CSS after
Johannes Weiner73f576c2016-07-20 15:44:57 -07005115 * the cgroup has been destroyed, such as page cache or reclaimable
5116 * slab objects, that don't need to hang on to the ID. We want to keep
5117 * those dead CSS from occupying IDs, or we might quickly exhaust the
5118 * relatively small ID space and prevent the creation of new cgroups
5119 * even when there are much fewer than 64k cgroups - possibly none.
5120 *
5121 * Maintain a private 16-bit ID space for memcg, and allow the ID to
5122 * be freed and recycled when it's no longer needed, which is usually
5123 * when the CSS is offlined.
5124 *
5125 * The only exception to that are records of swapped out tmpfs/shmem
5126 * pages that need to be attributed to live ancestors on swapin. But
5127 * those references are manageable from userspace.
5128 */
5129
5130static DEFINE_IDR(mem_cgroup_idr);
5131
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005132static void mem_cgroup_id_remove(struct mem_cgroup *memcg)
5133{
5134 if (memcg->id.id > 0) {
5135 idr_remove(&mem_cgroup_idr, memcg->id.id);
5136 memcg->id.id = 0;
5137 }
5138}
5139
Vincenzo Frascinoc1514c02020-04-01 21:07:13 -07005140static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg,
5141 unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07005142{
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07005143 refcount_add(n, &memcg->id.ref);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005144}
5145
Vladimir Davydov615d66c2016-08-11 15:33:03 -07005146static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
Johannes Weiner73f576c2016-07-20 15:44:57 -07005147{
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07005148 if (refcount_sub_and_test(n, &memcg->id.ref)) {
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005149 mem_cgroup_id_remove(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005150
5151 /* Memcg ID pins CSS */
5152 css_put(&memcg->css);
5153 }
5154}
5155
Vladimir Davydov615d66c2016-08-11 15:33:03 -07005156static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
5157{
5158 mem_cgroup_id_put_many(memcg, 1);
5159}
5160
Johannes Weiner73f576c2016-07-20 15:44:57 -07005161/**
5162 * mem_cgroup_from_id - look up a memcg from a memcg id
5163 * @id: the memcg id to look up
5164 *
5165 * Caller must hold rcu_read_lock().
5166 */
5167struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
5168{
5169 WARN_ON_ONCE(!rcu_read_lock_held());
5170 return idr_find(&mem_cgroup_idr, id);
5171}
5172
Mel Gormanef8f2322016-07-28 15:46:05 -07005173static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005174{
5175 struct mem_cgroup_per_node *pn;
Mel Gormanef8f2322016-07-28 15:46:05 -07005176 int tmp = node;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005177 /*
5178 * This routine is called against possible nodes.
5179 * But it's BUG to call kmalloc() against offline node.
5180 *
5181 * TODO: this routine can waste much memory for nodes which will
5182 * never be onlined. It's better to use memory hotplug callback
5183 * function.
5184 */
KAMEZAWA Hiroyuki41e33552008-04-08 17:41:54 -07005185 if (!node_state(node, N_NORMAL_MEMORY))
5186 tmp = -1;
Jesper Juhl17295c82011-01-13 15:47:42 -08005187 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005188 if (!pn)
5189 return 1;
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005190
Roman Gushchin3e38e0a2020-08-11 18:30:25 -07005191 pn->lruvec_stat_local = alloc_percpu_gfp(struct lruvec_stat,
5192 GFP_KERNEL_ACCOUNT);
Johannes Weiner815744d2019-06-13 15:55:46 -07005193 if (!pn->lruvec_stat_local) {
5194 kfree(pn);
5195 return 1;
5196 }
5197
Roman Gushchin3e38e0a2020-08-11 18:30:25 -07005198 pn->lruvec_stat_cpu = alloc_percpu_gfp(struct lruvec_stat,
5199 GFP_KERNEL_ACCOUNT);
Johannes Weinera983b5e2018-01-31 16:16:45 -08005200 if (!pn->lruvec_stat_cpu) {
Johannes Weiner815744d2019-06-13 15:55:46 -07005201 free_percpu(pn->lruvec_stat_local);
Johannes Weiner00f3ca22017-07-06 15:40:52 -07005202 kfree(pn);
5203 return 1;
5204 }
5205
Mel Gormanef8f2322016-07-28 15:46:05 -07005206 lruvec_init(&pn->lruvec);
5207 pn->usage_in_excess = 0;
5208 pn->on_tree = false;
5209 pn->memcg = memcg;
5210
Johannes Weiner54f72fe2013-07-08 15:59:49 -07005211 memcg->nodeinfo[node] = pn;
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08005212 return 0;
5213}
5214
Mel Gormanef8f2322016-07-28 15:46:05 -07005215static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005216{
Johannes Weiner00f3ca22017-07-06 15:40:52 -07005217 struct mem_cgroup_per_node *pn = memcg->nodeinfo[node];
5218
Michal Hocko4eaf4312018-04-10 16:29:52 -07005219 if (!pn)
5220 return;
5221
Johannes Weinera983b5e2018-01-31 16:16:45 -08005222 free_percpu(pn->lruvec_stat_cpu);
Johannes Weiner815744d2019-06-13 15:55:46 -07005223 free_percpu(pn->lruvec_stat_local);
Johannes Weiner00f3ca22017-07-06 15:40:52 -07005224 kfree(pn);
KAMEZAWA Hiroyuki1ecaab22008-02-07 00:14:38 -08005225}
5226
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005227static void __mem_cgroup_free(struct mem_cgroup *memcg)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005228{
5229 int node;
5230
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005231 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07005232 free_mem_cgroup_per_node_info(memcg, node);
Chris Down871789d2019-05-14 15:46:57 -07005233 free_percpu(memcg->vmstats_percpu);
Johannes Weiner815744d2019-06-13 15:55:46 -07005234 free_percpu(memcg->vmstats_local);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005235 kfree(memcg);
5236}
5237
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005238static void mem_cgroup_free(struct mem_cgroup *memcg)
5239{
5240 memcg_wb_domain_exit(memcg);
Shakeel Butt7961eee2019-11-05 21:16:21 -08005241 /*
5242 * Flush percpu vmstats and vmevents to guarantee the value correctness
5243 * on parent's and all ancestor levels.
5244 */
Roman Gushchin4a87e2a2020-01-13 16:29:16 -08005245 memcg_flush_percpu_vmstats(memcg);
Shakeel Butt7961eee2019-11-05 21:16:21 -08005246 memcg_flush_percpu_vmevents(memcg);
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005247 __mem_cgroup_free(memcg);
5248}
5249
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005250static struct mem_cgroup *mem_cgroup_alloc(void)
5251{
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005252 struct mem_cgroup *memcg;
Alexey Dobriyanb9726c22019-03-05 15:48:26 -08005253 unsigned int size;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005254 int node;
Tejun Heo97b27822019-08-26 09:06:56 -07005255 int __maybe_unused i;
Yafang Shao11d67612020-05-07 18:35:43 -07005256 long error = -ENOMEM;
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005257
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08005258 size = sizeof(struct mem_cgroup);
5259 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
KAMEZAWA Hiroyuki33327942008-04-29 01:00:24 -07005260
Vladimir Davydov8ff69e22014-01-23 15:52:52 -08005261 memcg = kzalloc(size, GFP_KERNEL);
Hugh Dickinsd79154b2012-03-21 16:34:18 -07005262 if (!memcg)
Yafang Shao11d67612020-05-07 18:35:43 -07005263 return ERR_PTR(error);
Dan Carpentere7bbcdf2010-03-23 13:35:12 -07005264
Johannes Weiner73f576c2016-07-20 15:44:57 -07005265 memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
5266 1, MEM_CGROUP_ID_MAX,
5267 GFP_KERNEL);
Yafang Shao11d67612020-05-07 18:35:43 -07005268 if (memcg->id.id < 0) {
5269 error = memcg->id.id;
Johannes Weiner73f576c2016-07-20 15:44:57 -07005270 goto fail;
Yafang Shao11d67612020-05-07 18:35:43 -07005271 }
Johannes Weiner73f576c2016-07-20 15:44:57 -07005272
Roman Gushchin3e38e0a2020-08-11 18:30:25 -07005273 memcg->vmstats_local = alloc_percpu_gfp(struct memcg_vmstats_percpu,
5274 GFP_KERNEL_ACCOUNT);
Johannes Weiner815744d2019-06-13 15:55:46 -07005275 if (!memcg->vmstats_local)
5276 goto fail;
5277
Roman Gushchin3e38e0a2020-08-11 18:30:25 -07005278 memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu,
5279 GFP_KERNEL_ACCOUNT);
Chris Down871789d2019-05-14 15:46:57 -07005280 if (!memcg->vmstats_percpu)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005281 goto fail;
Pavel Emelianov78fb7462008-02-07 00:13:51 -08005282
Bob Liu3ed28fa2012-01-12 17:19:04 -08005283 for_each_node(node)
Mel Gormanef8f2322016-07-28 15:46:05 -07005284 if (alloc_mem_cgroup_per_node_info(memcg, node))
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005285 goto fail;
Balbir Singhf64c3f52009-09-23 15:56:37 -07005286
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005287 if (memcg_wb_domain_init(memcg, GFP_KERNEL))
5288 goto fail;
Balbir Singh28dbc4b2009-01-07 18:08:05 -08005289
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005290 INIT_WORK(&memcg->high_work, high_work_func);
Glauber Costad142e3e2013-02-22 16:34:52 -08005291 INIT_LIST_HEAD(&memcg->oom_notify);
Glauber Costad142e3e2013-02-22 16:34:52 -08005292 mutex_init(&memcg->thresholds_lock);
5293 spin_lock_init(&memcg->move_lock);
Anton Vorontsov70ddf632013-04-29 15:08:31 -07005294 vmpressure_init(&memcg->vmpressure);
Tejun Heofba94802013-11-22 18:20:43 -05005295 INIT_LIST_HEAD(&memcg->event_list);
5296 spin_lock_init(&memcg->event_list_lock);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005297 memcg->socket_pressure = jiffies;
Kirill Tkhai84c07d12018-08-17 15:47:25 -07005298#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydov900a38f2014-12-12 16:55:10 -08005299 memcg->kmemcg_id = -1;
Roman Gushchinbf4f0592020-08-06 23:20:49 -07005300 INIT_LIST_HEAD(&memcg->objcg_list);
Vladimir Davydov900a38f2014-12-12 16:55:10 -08005301#endif
Tejun Heo52ebea72015-05-22 17:13:37 -04005302#ifdef CONFIG_CGROUP_WRITEBACK
5303 INIT_LIST_HEAD(&memcg->cgwb_list);
Tejun Heo97b27822019-08-26 09:06:56 -07005304 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++)
5305 memcg->cgwb_frn[i].done =
5306 __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq);
Tejun Heo52ebea72015-05-22 17:13:37 -04005307#endif
Yang Shi87eaceb2019-09-23 15:38:15 -07005308#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5309 spin_lock_init(&memcg->deferred_split_queue.split_queue_lock);
5310 INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue);
5311 memcg->deferred_split_queue.split_queue_len = 0;
5312#endif
Johannes Weiner73f576c2016-07-20 15:44:57 -07005313 idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005314 return memcg;
5315fail:
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005316 mem_cgroup_id_remove(memcg);
Tahsin Erdogan40e952f2017-03-09 16:17:26 -08005317 __mem_cgroup_free(memcg);
Yafang Shao11d67612020-05-07 18:35:43 -07005318 return ERR_PTR(error);
Glauber Costad142e3e2013-02-22 16:34:52 -08005319}
5320
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005321static struct cgroup_subsys_state * __ref
5322mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
Glauber Costad142e3e2013-02-22 16:34:52 -08005323{
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005324 struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
Roman Gushchinb87d8ce2020-10-17 16:13:40 -07005325 struct mem_cgroup *memcg, *old_memcg;
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005326 long error = -ENOMEM;
Glauber Costad142e3e2013-02-22 16:34:52 -08005327
Roman Gushchinb87d8ce2020-10-17 16:13:40 -07005328 old_memcg = set_active_memcg(parent);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005329 memcg = mem_cgroup_alloc();
Roman Gushchinb87d8ce2020-10-17 16:13:40 -07005330 set_active_memcg(old_memcg);
Yafang Shao11d67612020-05-07 18:35:43 -07005331 if (IS_ERR(memcg))
5332 return ERR_CAST(memcg);
Li Zefan4219b2d2013-09-23 16:56:29 +08005333
Jakub Kicinskid1663a92020-06-01 21:49:49 -07005334 page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005335 memcg->soft_limit = PAGE_COUNTER_MAX;
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07005336 page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005337 if (parent) {
5338 memcg->swappiness = mem_cgroup_swappiness(parent);
5339 memcg->oom_kill_disable = parent->oom_kill_disable;
5340 }
Roman Gushchin8de15e92020-11-01 17:07:34 -08005341 if (!parent) {
5342 page_counter_init(&memcg->memory, NULL);
5343 page_counter_init(&memcg->swap, NULL);
5344 page_counter_init(&memcg->kmem, NULL);
5345 page_counter_init(&memcg->tcpmem, NULL);
5346 } else if (parent->use_hierarchy) {
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005347 memcg->use_hierarchy = true;
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005348 page_counter_init(&memcg->memory, &parent->memory);
Vladimir Davydov37e84352016-01-20 15:02:56 -08005349 page_counter_init(&memcg->swap, &parent->swap);
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005350 page_counter_init(&memcg->kmem, &parent->kmem);
Johannes Weiner0db15292016-01-20 15:02:50 -08005351 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
Balbir Singh18f59ea2009-01-07 18:08:07 -08005352 } else {
Roman Gushchin8de15e92020-11-01 17:07:34 -08005353 page_counter_init(&memcg->memory, &root_mem_cgroup->memory);
5354 page_counter_init(&memcg->swap, &root_mem_cgroup->swap);
5355 page_counter_init(&memcg->kmem, &root_mem_cgroup->kmem);
5356 page_counter_init(&memcg->tcpmem, &root_mem_cgroup->tcpmem);
Tejun Heo8c7f6ed2012-09-13 12:20:58 -07005357 /*
5358 * Deeper hierachy with use_hierarchy == false doesn't make
5359 * much sense so let cgroup subsystem know about this
5360 * unfortunate state in our controller.
5361 */
Glauber Costad142e3e2013-02-22 16:34:52 -08005362 if (parent != root_mem_cgroup)
Tejun Heo073219e2014-02-08 10:36:58 -05005363 memory_cgrp_subsys.broken_hierarchy = true;
Balbir Singh18f59ea2009-01-07 18:08:07 -08005364 }
Vladimir Davydovd6441632014-01-23 15:53:09 -08005365
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005366 /* The following stuff does not apply to the root */
5367 if (!parent) {
5368 root_mem_cgroup = memcg;
5369 return &memcg->css;
5370 }
5371
Vladimir Davydovb313aee2016-03-17 14:18:27 -07005372 error = memcg_online_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005373 if (error)
5374 goto fail;
Johannes Weiner127424c2016-01-20 15:02:32 -08005375
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005376 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08005377 static_branch_inc(&memcg_sockets_enabled_key);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005378
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005379 return &memcg->css;
5380fail:
Kirill Tkhai7e97de02018-08-02 15:36:01 -07005381 mem_cgroup_id_remove(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005382 mem_cgroup_free(memcg);
Yafang Shao11d67612020-05-07 18:35:43 -07005383 return ERR_PTR(error);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005384}
5385
Johannes Weiner73f576c2016-07-20 15:44:57 -07005386static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005387{
Vladimir Davydov58fa2a52016-10-07 16:57:29 -07005388 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5389
Kirill Tkhai0a4465d2018-08-17 15:47:37 -07005390 /*
5391 * A memcg must be visible for memcg_expand_shrinker_maps()
5392 * by the time the maps are allocated. So, we allocate maps
5393 * here, when for_each_mem_cgroup() can't skip it.
5394 */
5395 if (memcg_alloc_shrinker_maps(memcg)) {
5396 mem_cgroup_id_remove(memcg);
5397 return -ENOMEM;
5398 }
5399
Johannes Weiner73f576c2016-07-20 15:44:57 -07005400 /* Online state pins memcg ID, memcg ID pins CSS */
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07005401 refcount_set(&memcg->id.ref, 1);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005402 css_get(css);
Johannes Weiner2f7dd7a2014-10-02 16:16:57 -07005403 return 0;
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005404}
5405
Tejun Heoeb954192013-08-08 20:11:23 -04005406static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005407{
Tejun Heoeb954192013-08-08 20:11:23 -04005408 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo3bc942f2013-11-22 18:20:44 -05005409 struct mem_cgroup_event *event, *tmp;
Tejun Heo79bd9812013-11-22 18:20:42 -05005410
5411 /*
5412 * Unregister events and notify userspace.
5413 * Notify userspace about cgroup removing only after rmdir of cgroup
5414 * directory to avoid race between userspace and kernelspace.
5415 */
Tejun Heofba94802013-11-22 18:20:43 -05005416 spin_lock(&memcg->event_list_lock);
5417 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
Tejun Heo79bd9812013-11-22 18:20:42 -05005418 list_del_init(&event->list);
5419 schedule_work(&event->remove);
5420 }
Tejun Heofba94802013-11-22 18:20:43 -05005421 spin_unlock(&memcg->event_list_lock);
KAMEZAWA Hiroyukiec64f512009-04-02 16:57:26 -07005422
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07005423 page_counter_set_min(&memcg->memory, 0);
Roman Gushchin23067152018-06-07 17:06:22 -07005424 page_counter_set_low(&memcg->memory, 0);
Roman Gushchin63677c742017-09-06 16:21:47 -07005425
Johannes Weiner567e9ab2016-01-20 15:02:24 -08005426 memcg_offline_kmem(memcg);
Tejun Heo52ebea72015-05-22 17:13:37 -04005427 wb_memcg_offline(memcg);
Johannes Weiner73f576c2016-07-20 15:44:57 -07005428
Roman Gushchin591edfb2018-10-26 15:03:23 -07005429 drain_all_stock(memcg);
5430
Johannes Weiner73f576c2016-07-20 15:44:57 -07005431 mem_cgroup_id_put(memcg);
KAMEZAWA Hiroyukidf878fb2008-02-07 00:14:28 -08005432}
5433
Vladimir Davydov6df38682015-12-29 14:54:10 -08005434static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
5435{
5436 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5437
5438 invalidate_reclaim_iterators(memcg);
5439}
5440
Tejun Heoeb954192013-08-08 20:11:23 -04005441static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005442{
Tejun Heoeb954192013-08-08 20:11:23 -04005443 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Tejun Heo97b27822019-08-26 09:06:56 -07005444 int __maybe_unused i;
Daisuke Nishimurac268e992009-01-15 13:51:13 -08005445
Tejun Heo97b27822019-08-26 09:06:56 -07005446#ifdef CONFIG_CGROUP_WRITEBACK
5447 for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++)
5448 wb_wait_for_completion(&memcg->cgwb_frn[i].done);
5449#endif
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08005450 if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
Johannes Weineref129472016-01-14 15:21:34 -08005451 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08005452
Johannes Weiner0db15292016-01-20 15:02:50 -08005453 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
Vladimir Davydovd55f90b2016-01-20 15:02:44 -08005454 static_branch_dec(&memcg_sockets_enabled_key);
Johannes Weiner3893e302016-01-20 15:02:29 -08005455
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005456 vmpressure_cleanup(&memcg->vmpressure);
5457 cancel_work_sync(&memcg->high_work);
5458 mem_cgroup_remove_from_trees(memcg);
Kirill Tkhai0a4465d2018-08-17 15:47:37 -07005459 memcg_free_shrinker_maps(memcg);
Johannes Weinerd886f4e2016-01-20 15:02:47 -08005460 memcg_free_kmem(memcg);
Johannes Weiner0b8f73e2016-01-20 15:02:53 -08005461 mem_cgroup_free(memcg);
Balbir Singh8cdea7c2008-02-07 00:13:50 -08005462}
5463
Tejun Heo1ced9532014-07-08 18:02:57 -04005464/**
5465 * mem_cgroup_css_reset - reset the states of a mem_cgroup
5466 * @css: the target css
5467 *
5468 * Reset the states of the mem_cgroup associated with @css. This is
5469 * invoked when the userland requests disabling on the default hierarchy
5470 * but the memcg is pinned through dependency. The memcg should stop
5471 * applying policies and should revert to the vanilla state as it may be
5472 * made visible again.
5473 *
5474 * The current implementation only resets the essential configurations.
5475 * This needs to be expanded to cover all the visible parts.
5476 */
5477static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
5478{
5479 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5480
Roman Gushchinbbec2e12018-06-07 17:06:18 -07005481 page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX);
5482 page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX);
Roman Gushchinbbec2e12018-06-07 17:06:18 -07005483 page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX);
5484 page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX);
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07005485 page_counter_set_min(&memcg->memory, 0);
Roman Gushchin23067152018-06-07 17:06:22 -07005486 page_counter_set_low(&memcg->memory, 0);
Jakub Kicinskid1663a92020-06-01 21:49:49 -07005487 page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX);
Johannes Weiner24d404d2015-01-08 14:32:35 -08005488 memcg->soft_limit = PAGE_COUNTER_MAX;
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07005489 page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX);
Tejun Heo2529bb32015-05-22 18:23:34 -04005490 memcg_wb_domain_size_changed(memcg);
Tejun Heo1ced9532014-07-08 18:02:57 -04005491}
5492
Daisuke Nishimura02491442010-03-10 15:22:17 -08005493#ifdef CONFIG_MMU
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005494/* Handlers for move charge at task migration. */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005495static int mem_cgroup_do_precharge(unsigned long count)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005496{
Johannes Weiner05b84302014-08-06 16:05:59 -07005497 int ret;
Johannes Weiner9476db92014-08-06 16:05:55 -07005498
Mel Gormand0164ad2015-11-06 16:28:21 -08005499 /* Try a single bulk charge without reclaim first, kswapd may wake */
5500 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
Johannes Weiner9476db92014-08-06 16:05:55 -07005501 if (!ret) {
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005502 mc.precharge += count;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005503 return ret;
5504 }
Johannes Weiner9476db92014-08-06 16:05:55 -07005505
David Rientjes36745342017-01-24 15:18:10 -08005506 /* Try charges one by one with reclaim, but do not retry */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005507 while (count--) {
David Rientjes36745342017-01-24 15:18:10 -08005508 ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1);
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005509 if (ret)
KAMEZAWA Hiroyuki38c5d722012-01-12 17:19:01 -08005510 return ret;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005511 mc.precharge++;
Johannes Weiner9476db92014-08-06 16:05:55 -07005512 cond_resched();
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005513 }
Johannes Weiner9476db92014-08-06 16:05:55 -07005514 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005515}
5516
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005517union mc_target {
5518 struct page *page;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005519 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005520};
5521
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005522enum mc_target_type {
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005523 MC_TARGET_NONE = 0,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005524 MC_TARGET_PAGE,
Daisuke Nishimura02491442010-03-10 15:22:17 -08005525 MC_TARGET_SWAP,
Jérôme Glissec733a822017-09-08 16:11:54 -07005526 MC_TARGET_DEVICE,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005527};
5528
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005529static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5530 unsigned long addr, pte_t ptent)
5531{
Christoph Hellwig25b29952019-06-13 22:50:49 +02005532 struct page *page = vm_normal_page(vma, addr, ptent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005533
5534 if (!page || !page_mapped(page))
5535 return NULL;
5536 if (PageAnon(page)) {
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005537 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005538 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005539 } else {
5540 if (!(mc.flags & MOVE_FILE))
5541 return NULL;
5542 }
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005543 if (!get_page_unless_zero(page))
5544 return NULL;
5545
5546 return page;
5547}
5548
Jérôme Glissec733a822017-09-08 16:11:54 -07005549#if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005550static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07005551 pte_t ptent, swp_entry_t *entry)
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005552{
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005553 struct page *page = NULL;
5554 swp_entry_t ent = pte_to_swp_entry(ptent);
5555
Ralph Campbell9a137152020-10-13 16:53:13 -07005556 if (!(mc.flags & MOVE_ANON))
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005557 return NULL;
Jérôme Glissec733a822017-09-08 16:11:54 -07005558
5559 /*
5560 * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to
5561 * a device and because they are not accessible by CPU they are store
5562 * as special swap entry in the CPU page table.
5563 */
5564 if (is_device_private_entry(ent)) {
5565 page = device_private_entry_to_page(ent);
5566 /*
5567 * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have
5568 * a refcount of 1 when free (unlike normal page)
5569 */
5570 if (!page_ref_add_unless(page, 1, 1))
5571 return NULL;
5572 return page;
5573 }
5574
Ralph Campbell9a137152020-10-13 16:53:13 -07005575 if (non_swap_entry(ent))
5576 return NULL;
5577
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005578 /*
5579 * Because lookup_swap_cache() updates some statistics counter,
5580 * we call find_get_page() with swapper_space directly.
5581 */
Huang Yingf6ab1f72016-10-07 17:00:21 -07005582 page = find_get_page(swap_address_space(ent), swp_offset(ent));
Johannes Weiner2d1c4982020-06-03 16:02:14 -07005583 entry->val = ent.val;
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005584
5585 return page;
5586}
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005587#else
5588static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
Li RongQing48406ef2016-07-26 15:22:14 -07005589 pte_t ptent, swp_entry_t *entry)
KAMEZAWA Hiroyuki4b913552012-05-29 15:06:51 -07005590{
5591 return NULL;
5592}
5593#endif
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005594
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005595static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5596 unsigned long addr, pte_t ptent, swp_entry_t *entry)
5597{
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005598 if (!vma->vm_file) /* anonymous vma */
5599 return NULL;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005600 if (!(mc.flags & MOVE_FILE))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005601 return NULL;
5602
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005603 /* page is moved even if it's not RSS of this task(page-faulted). */
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07005604 /* shmem/tmpfs may report page out on swap: account for that too. */
Matthew Wilcox (Oracle)f5df8632020-10-13 16:51:21 -07005605 return find_get_incore_page(vma->vm_file->f_mapping,
5606 linear_page_index(vma, addr));
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005607}
5608
Chen Gangb1b0dea2015-04-14 15:47:35 -07005609/**
5610 * mem_cgroup_move_account - move account of the page
5611 * @page: the page
Li RongQing25843c22016-07-26 15:26:56 -07005612 * @compound: charge the page as compound or small page
Chen Gangb1b0dea2015-04-14 15:47:35 -07005613 * @from: mem_cgroup which the page is moved from.
5614 * @to: mem_cgroup which the page is moved to. @from != @to.
5615 *
Kirill A. Shutemov3ac808f2016-01-15 16:53:07 -08005616 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
Chen Gangb1b0dea2015-04-14 15:47:35 -07005617 *
5618 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
5619 * from old cgroup.
5620 */
5621static int mem_cgroup_move_account(struct page *page,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005622 bool compound,
Chen Gangb1b0dea2015-04-14 15:47:35 -07005623 struct mem_cgroup *from,
5624 struct mem_cgroup *to)
5625{
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005626 struct lruvec *from_vec, *to_vec;
5627 struct pglist_data *pgdat;
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07005628 unsigned int nr_pages = compound ? thp_nr_pages(page) : 1;
Chen Gangb1b0dea2015-04-14 15:47:35 -07005629 int ret;
5630
5631 VM_BUG_ON(from == to);
5632 VM_BUG_ON_PAGE(PageLRU(page), page);
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005633 VM_BUG_ON(compound && !PageTransHuge(page));
Chen Gangb1b0dea2015-04-14 15:47:35 -07005634
5635 /*
Johannes Weiner6a93ca82016-03-15 14:57:19 -07005636 * Prevent mem_cgroup_migrate() from looking at
Hugh Dickins45637ba2015-11-05 18:49:40 -08005637 * page->mem_cgroup of its source page while we change it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07005638 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08005639 ret = -EBUSY;
Chen Gangb1b0dea2015-04-14 15:47:35 -07005640 if (!trylock_page(page))
5641 goto out;
5642
5643 ret = -EINVAL;
5644 if (page->mem_cgroup != from)
5645 goto out_unlock;
5646
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005647 pgdat = page_pgdat(page);
Johannes Weiner867e5e12019-11-30 17:55:34 -08005648 from_vec = mem_cgroup_lruvec(from, pgdat);
5649 to_vec = mem_cgroup_lruvec(to, pgdat);
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005650
Johannes Weinerabb242f2020-06-03 16:01:28 -07005651 lock_page_memcg(page);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005652
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07005653 if (PageAnon(page)) {
5654 if (page_mapped(page)) {
5655 __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages);
5656 __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages);
Johannes Weiner468c3982020-06-03 16:02:01 -07005657 if (PageTransHuge(page)) {
5658 __mod_lruvec_state(from_vec, NR_ANON_THPS,
5659 -nr_pages);
5660 __mod_lruvec_state(to_vec, NR_ANON_THPS,
5661 nr_pages);
5662 }
5663
Johannes Weinerbe5d0a72020-06-03 16:01:57 -07005664 }
5665 } else {
Johannes Weiner0d1c2072020-06-03 16:01:54 -07005666 __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages);
5667 __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages);
5668
5669 if (PageSwapBacked(page)) {
5670 __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages);
5671 __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages);
5672 }
5673
Johannes Weiner49e50d22020-06-03 16:01:47 -07005674 if (page_mapped(page)) {
5675 __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages);
5676 __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages);
5677 }
Chen Gangb1b0dea2015-04-14 15:47:35 -07005678
Johannes Weiner49e50d22020-06-03 16:01:47 -07005679 if (PageDirty(page)) {
5680 struct address_space *mapping = page_mapping(page);
Greg Thelenc4843a72015-05-22 17:13:16 -04005681
Christoph Hellwigf56753a2020-09-24 08:51:40 +02005682 if (mapping_can_writeback(mapping)) {
Johannes Weiner49e50d22020-06-03 16:01:47 -07005683 __mod_lruvec_state(from_vec, NR_FILE_DIRTY,
5684 -nr_pages);
5685 __mod_lruvec_state(to_vec, NR_FILE_DIRTY,
5686 nr_pages);
5687 }
Greg Thelenc4843a72015-05-22 17:13:16 -04005688 }
5689 }
5690
Chen Gangb1b0dea2015-04-14 15:47:35 -07005691 if (PageWriteback(page)) {
Konstantin Khlebnikovae8af432019-10-18 20:20:11 -07005692 __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages);
5693 __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005694 }
5695
5696 /*
Johannes Weinerabb242f2020-06-03 16:01:28 -07005697 * All state has been migrated, let's switch to the new memcg.
5698 *
Chen Gangb1b0dea2015-04-14 15:47:35 -07005699 * It is safe to change page->mem_cgroup here because the page
Johannes Weinerabb242f2020-06-03 16:01:28 -07005700 * is referenced, charged, isolated, and locked: we can't race
5701 * with (un)charging, migration, LRU putback, or anything else
5702 * that would rely on a stable page->mem_cgroup.
5703 *
5704 * Note that lock_page_memcg is a memcg lock, not a page lock,
5705 * to save space. As soon as we switch page->mem_cgroup to a
5706 * new memcg that isn't locked, the above state can change
5707 * concurrently again. Make sure we're truly done with it.
Chen Gangb1b0dea2015-04-14 15:47:35 -07005708 */
Johannes Weinerabb242f2020-06-03 16:01:28 -07005709 smp_mb();
Chen Gangb1b0dea2015-04-14 15:47:35 -07005710
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07005711 css_get(&to->css);
5712 css_put(&from->css);
5713
5714 page->mem_cgroup = to;
Yang Shi87eaceb2019-09-23 15:38:15 -07005715
Johannes Weinerabb242f2020-06-03 16:01:28 -07005716 __unlock_page_memcg(from);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005717
5718 ret = 0;
5719
5720 local_irq_disable();
Johannes Weiner3fba69a2020-06-03 16:01:31 -07005721 mem_cgroup_charge_statistics(to, page, nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005722 memcg_check_events(to, page);
Johannes Weiner3fba69a2020-06-03 16:01:31 -07005723 mem_cgroup_charge_statistics(from, page, -nr_pages);
Chen Gangb1b0dea2015-04-14 15:47:35 -07005724 memcg_check_events(from, page);
5725 local_irq_enable();
5726out_unlock:
5727 unlock_page(page);
5728out:
5729 return ret;
5730}
5731
Li RongQing7cf78062016-05-27 14:27:46 -07005732/**
5733 * get_mctgt_type - get target type of moving charge
5734 * @vma: the vma the pte to be checked belongs
5735 * @addr: the address corresponding to the pte to be checked
5736 * @ptent: the pte to be checked
5737 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5738 *
5739 * Returns
5740 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5741 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5742 * move charge. if @target is not NULL, the page is stored in target->page
5743 * with extra refcnt got(Callers should handle it).
5744 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5745 * target for charge migration. if @target is not NULL, the entry is stored
5746 * in target->ent.
Christoph Hellwig25b29952019-06-13 22:50:49 +02005747 * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE
5748 * (so ZONE_DEVICE page and thus not on the lru).
Jérôme Glissedf6ad692017-09-08 16:12:24 -07005749 * For now we such page is charge like a regular page would be as for all
5750 * intent and purposes it is just special memory taking the place of a
5751 * regular page.
Jérôme Glissec733a822017-09-08 16:11:54 -07005752 *
5753 * See Documentations/vm/hmm.txt and include/linux/hmm.h
Li RongQing7cf78062016-05-27 14:27:46 -07005754 *
5755 * Called with pte lock held.
5756 */
5757
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005758static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005759 unsigned long addr, pte_t ptent, union mc_target *target)
5760{
Daisuke Nishimura02491442010-03-10 15:22:17 -08005761 struct page *page = NULL;
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005762 enum mc_target_type ret = MC_TARGET_NONE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005763 swp_entry_t ent = { .val = 0 };
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005764
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005765 if (pte_present(ptent))
5766 page = mc_handle_present_pte(vma, addr, ptent);
5767 else if (is_swap_pte(ptent))
Li RongQing48406ef2016-07-26 15:22:14 -07005768 page = mc_handle_swap_pte(vma, ptent, &ent);
Kirill A. Shutemov0661a332015-02-10 14:10:04 -08005769 else if (pte_none(ptent))
Daisuke Nishimura87946a72010-05-26 14:42:39 -07005770 page = mc_handle_file_pte(vma, addr, ptent, &ent);
Daisuke Nishimura90254a62010-05-26 14:42:38 -07005771
5772 if (!page && !ent.val)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005773 return ret;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005774 if (page) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005775 /*
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005776 * Do only loose check w/o serialization.
Johannes Weiner1306a852014-12-10 15:44:52 -08005777 * mem_cgroup_move_account() checks the page is valid or
Johannes Weiner0a31bc92014-08-08 14:19:22 -07005778 * not under LRU exclusion.
Daisuke Nishimura02491442010-03-10 15:22:17 -08005779 */
Johannes Weiner1306a852014-12-10 15:44:52 -08005780 if (page->mem_cgroup == mc.from) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08005781 ret = MC_TARGET_PAGE;
Christoph Hellwig25b29952019-06-13 22:50:49 +02005782 if (is_device_private_page(page))
Jérôme Glissec733a822017-09-08 16:11:54 -07005783 ret = MC_TARGET_DEVICE;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005784 if (target)
5785 target->page = page;
5786 }
5787 if (!ret || !target)
5788 put_page(page);
5789 }
Huang Ying3e14a572017-09-06 16:22:37 -07005790 /*
5791 * There is a swap entry and a page doesn't exist or isn't charged.
5792 * But we cannot move a tail-page in a THP.
5793 */
5794 if (ent.val && !ret && (!page || !PageTransCompound(page)) &&
Li Zefan34c00c32013-09-23 16:56:01 +08005795 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
KAMEZAWA Hiroyuki7f0f1542010-05-11 14:06:58 -07005796 ret = MC_TARGET_SWAP;
5797 if (target)
5798 target->ent = ent;
Daisuke Nishimura02491442010-03-10 15:22:17 -08005799 }
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005800 return ret;
5801}
5802
Naoya Horiguchi12724852012-03-21 16:34:28 -07005803#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5804/*
Huang Yingd6810d72017-09-06 16:22:45 -07005805 * We don't consider PMD mapped swapping or file mapped pages because THP does
5806 * not support them for now.
Naoya Horiguchi12724852012-03-21 16:34:28 -07005807 * Caller should make sure that pmd_trans_huge(pmd) is true.
5808 */
5809static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5810 unsigned long addr, pmd_t pmd, union mc_target *target)
5811{
5812 struct page *page = NULL;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005813 enum mc_target_type ret = MC_TARGET_NONE;
5814
Zi Yan84c3fc42017-09-08 16:11:01 -07005815 if (unlikely(is_swap_pmd(pmd))) {
5816 VM_BUG_ON(thp_migration_supported() &&
5817 !is_pmd_migration_entry(pmd));
5818 return ret;
5819 }
Naoya Horiguchi12724852012-03-21 16:34:28 -07005820 page = pmd_page(pmd);
Sasha Levin309381fea2014-01-23 15:52:54 -08005821 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005822 if (!(mc.flags & MOVE_ANON))
Naoya Horiguchi12724852012-03-21 16:34:28 -07005823 return ret;
Johannes Weiner1306a852014-12-10 15:44:52 -08005824 if (page->mem_cgroup == mc.from) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07005825 ret = MC_TARGET_PAGE;
5826 if (target) {
5827 get_page(page);
5828 target->page = page;
5829 }
5830 }
5831 return ret;
5832}
5833#else
5834static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5835 unsigned long addr, pmd_t pmd, union mc_target *target)
5836{
5837 return MC_TARGET_NONE;
5838}
5839#endif
5840
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005841static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5842 unsigned long addr, unsigned long end,
5843 struct mm_walk *walk)
5844{
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08005845 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005846 pte_t *pte;
5847 spinlock_t *ptl;
5848
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08005849 ptl = pmd_trans_huge_lock(pmd, vma);
5850 if (ptl) {
Jérôme Glissec733a822017-09-08 16:11:54 -07005851 /*
5852 * Note their can not be MC_TARGET_DEVICE for now as we do not
Christoph Hellwig25b29952019-06-13 22:50:49 +02005853 * support transparent huge page with MEMORY_DEVICE_PRIVATE but
5854 * this might change.
Jérôme Glissec733a822017-09-08 16:11:54 -07005855 */
Naoya Horiguchi12724852012-03-21 16:34:28 -07005856 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
5857 mc.precharge += HPAGE_PMD_NR;
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08005858 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07005859 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07005860 }
Dave Hansen03319322011-03-22 16:32:56 -07005861
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07005862 if (pmd_trans_unstable(pmd))
5863 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005864 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5865 for (; addr != end; pte++, addr += PAGE_SIZE)
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07005866 if (get_mctgt_type(vma, addr, *pte, NULL))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005867 mc.precharge++; /* increment precharge temporarily */
5868 pte_unmap_unlock(pte - 1, ptl);
5869 cond_resched();
5870
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005871 return 0;
5872}
5873
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02005874static const struct mm_walk_ops precharge_walk_ops = {
5875 .pmd_entry = mem_cgroup_count_precharge_pte_range,
5876};
5877
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005878static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5879{
5880 unsigned long precharge;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005881
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07005882 mmap_read_lock(mm);
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02005883 walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL);
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07005884 mmap_read_unlock(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005885
5886 precharge = mc.precharge;
5887 mc.precharge = 0;
5888
5889 return precharge;
5890}
5891
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005892static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5893{
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005894 unsigned long precharge = mem_cgroup_count_precharge(mm);
5895
5896 VM_BUG_ON(mc.moving_task);
5897 mc.moving_task = current;
5898 return mem_cgroup_do_precharge(precharge);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005899}
5900
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005901/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5902static void __mem_cgroup_clear_mc(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005903{
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005904 struct mem_cgroup *from = mc.from;
5905 struct mem_cgroup *to = mc.to;
5906
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005907 /* we must uncharge all the leftover precharges from mc.to */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005908 if (mc.precharge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005909 cancel_charge(mc.to, mc.precharge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005910 mc.precharge = 0;
5911 }
5912 /*
5913 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5914 * we must uncharge here.
5915 */
5916 if (mc.moved_charge) {
Johannes Weiner00501b52014-08-08 14:19:20 -07005917 cancel_charge(mc.from, mc.moved_charge);
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08005918 mc.moved_charge = 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005919 }
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005920 /* we must fixup refcnts and charges */
5921 if (mc.moved_swap) {
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005922 /* uncharge swap account from the old cgroup */
Johannes Weinerce00a962014-09-05 08:43:57 -04005923 if (!mem_cgroup_is_root(mc.from))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005924 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005925
Vladimir Davydov615d66c2016-08-11 15:33:03 -07005926 mem_cgroup_id_put_many(mc.from, mc.moved_swap);
5927
Johannes Weiner05b84302014-08-06 16:05:59 -07005928 /*
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005929 * we charged both to->memory and to->memsw, so we
5930 * should uncharge to->memory.
Johannes Weiner05b84302014-08-06 16:05:59 -07005931 */
Johannes Weinerce00a962014-09-05 08:43:57 -04005932 if (!mem_cgroup_is_root(mc.to))
Johannes Weiner3e32cb22014-12-10 15:42:31 -08005933 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005934
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08005935 mc.moved_swap = 0;
5936 }
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005937 memcg_oom_recover(from);
5938 memcg_oom_recover(to);
5939 wake_up_all(&mc.waitq);
5940}
5941
5942static void mem_cgroup_clear_mc(void)
5943{
Tejun Heo264a0ae2016-04-21 19:09:02 -04005944 struct mm_struct *mm = mc.mm;
5945
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08005946 /*
5947 * we must clear moving_task before waking up waiters at the end of
5948 * task migration.
5949 */
5950 mc.moving_task = NULL;
5951 __mem_cgroup_clear_mc();
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005952 spin_lock(&mc.lock);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005953 mc.from = NULL;
5954 mc.to = NULL;
Tejun Heo264a0ae2016-04-21 19:09:02 -04005955 mc.mm = NULL;
KAMEZAWA Hiroyuki2bd9bb22010-08-10 18:02:58 -07005956 spin_unlock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04005957
5958 mmput(mm);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08005959}
5960
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005961static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005962{
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005963 struct cgroup_subsys_state *css;
Ross Zwislereed67d72015-12-23 14:53:27 -07005964 struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
Tejun Heo9f2115f2015-09-08 15:01:10 -07005965 struct mem_cgroup *from;
Tejun Heo4530edd2015-09-11 15:00:19 -04005966 struct task_struct *leader, *p;
Tejun Heo9f2115f2015-09-08 15:01:10 -07005967 struct mm_struct *mm;
Johannes Weiner1dfab5a2015-02-11 15:26:09 -08005968 unsigned long move_flags;
Tejun Heo9f2115f2015-09-08 15:01:10 -07005969 int ret = 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005970
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005971 /* charge immigration isn't supported on the default hierarchy */
5972 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Tejun Heo9f2115f2015-09-08 15:01:10 -07005973 return 0;
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08005974
Tejun Heo4530edd2015-09-11 15:00:19 -04005975 /*
5976 * Multi-process migrations only happen on the default hierarchy
5977 * where charge immigration is not used. Perform charge
5978 * immigration if @tset contains a leader and whine if there are
5979 * multiple.
5980 */
5981 p = NULL;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005982 cgroup_taskset_for_each_leader(leader, css, tset) {
Tejun Heo4530edd2015-09-11 15:00:19 -04005983 WARN_ON_ONCE(p);
5984 p = leader;
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005985 memcg = mem_cgroup_from_css(css);
Tejun Heo4530edd2015-09-11 15:00:19 -04005986 }
5987 if (!p)
5988 return 0;
5989
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05005990 /*
5991 * We are now commited to this value whatever it is. Changes in this
5992 * tunable will only affect upcoming migrations, not the current one.
5993 * So we need to save it, and keep it going.
5994 */
5995 move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
5996 if (!move_flags)
5997 return 0;
5998
Tejun Heo9f2115f2015-09-08 15:01:10 -07005999 from = mem_cgroup_from_task(p);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006000
Tejun Heo9f2115f2015-09-08 15:01:10 -07006001 VM_BUG_ON(from == memcg);
Johannes Weiner247b1442014-12-10 15:44:11 -08006002
Tejun Heo9f2115f2015-09-08 15:01:10 -07006003 mm = get_task_mm(p);
6004 if (!mm)
6005 return 0;
6006 /* We move charges only when we move a owner of the mm */
6007 if (mm->owner == p) {
6008 VM_BUG_ON(mc.from);
6009 VM_BUG_ON(mc.to);
6010 VM_BUG_ON(mc.precharge);
6011 VM_BUG_ON(mc.moved_charge);
6012 VM_BUG_ON(mc.moved_swap);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006013
Tejun Heo9f2115f2015-09-08 15:01:10 -07006014 spin_lock(&mc.lock);
Tejun Heo264a0ae2016-04-21 19:09:02 -04006015 mc.mm = mm;
Tejun Heo9f2115f2015-09-08 15:01:10 -07006016 mc.from = from;
6017 mc.to = memcg;
6018 mc.flags = move_flags;
6019 spin_unlock(&mc.lock);
6020 /* We set mc.moving_task later */
6021
6022 ret = mem_cgroup_precharge_mc(mm);
6023 if (ret)
6024 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04006025 } else {
6026 mmput(mm);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006027 }
6028 return ret;
6029}
6030
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006031static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006032{
Johannes Weiner4e2f2452014-12-10 15:44:08 -08006033 if (mc.to)
6034 mem_cgroup_clear_mc();
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006035}
6036
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006037static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
6038 unsigned long addr, unsigned long end,
6039 struct mm_walk *walk)
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006040{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006041 int ret = 0;
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08006042 struct vm_area_struct *vma = walk->vma;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006043 pte_t *pte;
6044 spinlock_t *ptl;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006045 enum mc_target_type target_type;
6046 union mc_target target;
6047 struct page *page;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006048
Kirill A. Shutemovb6ec57f2016-01-21 16:40:25 -08006049 ptl = pmd_trans_huge_lock(pmd, vma);
6050 if (ptl) {
Hugh Dickins62ade862012-05-18 11:28:34 -07006051 if (mc.precharge < HPAGE_PMD_NR) {
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006052 spin_unlock(ptl);
Naoya Horiguchi12724852012-03-21 16:34:28 -07006053 return 0;
6054 }
6055 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
6056 if (target_type == MC_TARGET_PAGE) {
6057 page = target.page;
6058 if (!isolate_lru_page(page)) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08006059 if (!mem_cgroup_move_account(page, true,
Johannes Weiner1306a852014-12-10 15:44:52 -08006060 mc.from, mc.to)) {
Naoya Horiguchi12724852012-03-21 16:34:28 -07006061 mc.precharge -= HPAGE_PMD_NR;
6062 mc.moved_charge += HPAGE_PMD_NR;
6063 }
6064 putback_lru_page(page);
6065 }
6066 put_page(page);
Jérôme Glissec733a822017-09-08 16:11:54 -07006067 } else if (target_type == MC_TARGET_DEVICE) {
6068 page = target.page;
6069 if (!mem_cgroup_move_account(page, true,
6070 mc.from, mc.to)) {
6071 mc.precharge -= HPAGE_PMD_NR;
6072 mc.moved_charge += HPAGE_PMD_NR;
6073 }
6074 put_page(page);
Naoya Horiguchi12724852012-03-21 16:34:28 -07006075 }
Kirill A. Shutemovbf929152013-11-14 14:30:54 -08006076 spin_unlock(ptl);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07006077 return 0;
Naoya Horiguchi12724852012-03-21 16:34:28 -07006078 }
6079
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -07006080 if (pmd_trans_unstable(pmd))
6081 return 0;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006082retry:
6083 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
6084 for (; addr != end; addr += PAGE_SIZE) {
6085 pte_t ptent = *(pte++);
Jérôme Glissec733a822017-09-08 16:11:54 -07006086 bool device = false;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006087 swp_entry_t ent;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006088
6089 if (!mc.precharge)
6090 break;
6091
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006092 switch (get_mctgt_type(vma, addr, ptent, &target)) {
Jérôme Glissec733a822017-09-08 16:11:54 -07006093 case MC_TARGET_DEVICE:
6094 device = true;
Joe Perchese4a9bc52020-04-06 20:08:39 -07006095 fallthrough;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006096 case MC_TARGET_PAGE:
6097 page = target.page;
Kirill A. Shutemov53f92632016-01-15 16:53:42 -08006098 /*
6099 * We can have a part of the split pmd here. Moving it
6100 * can be done but it would be too convoluted so simply
6101 * ignore such a partial THP and keep it in original
6102 * memcg. There should be somebody mapping the head.
6103 */
6104 if (PageTransCompound(page))
6105 goto put;
Jérôme Glissec733a822017-09-08 16:11:54 -07006106 if (!device && isolate_lru_page(page))
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006107 goto put;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08006108 if (!mem_cgroup_move_account(page, false,
6109 mc.from, mc.to)) {
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006110 mc.precharge--;
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006111 /* we uncharge from mc.from later. */
6112 mc.moved_charge++;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006113 }
Jérôme Glissec733a822017-09-08 16:11:54 -07006114 if (!device)
6115 putback_lru_page(page);
Naoya Horiguchi8d32ff82012-03-21 16:34:27 -07006116put: /* get_mctgt_type() gets the page */
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006117 put_page(page);
6118 break;
Daisuke Nishimura02491442010-03-10 15:22:17 -08006119 case MC_TARGET_SWAP:
6120 ent = target.ent;
Hugh Dickinse91cbb42012-05-29 15:06:51 -07006121 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
Daisuke Nishimura02491442010-03-10 15:22:17 -08006122 mc.precharge--;
Hugh Dickins8d22a932020-07-23 21:15:24 -07006123 mem_cgroup_id_get_many(mc.to, 1);
6124 /* we fixup other refcnts and charges later. */
Daisuke Nishimura483c30b2010-03-10 15:22:18 -08006125 mc.moved_swap++;
6126 }
Daisuke Nishimura02491442010-03-10 15:22:17 -08006127 break;
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006128 default:
6129 break;
6130 }
6131 }
6132 pte_unmap_unlock(pte - 1, ptl);
6133 cond_resched();
6134
6135 if (addr != end) {
6136 /*
6137 * We have consumed all precharges we got in can_attach().
6138 * We try charge one by one, but don't do any additional
6139 * charges to mc.to if we have failed in charge once in attach()
6140 * phase.
6141 */
Daisuke Nishimura854ffa82010-03-10 15:22:15 -08006142 ret = mem_cgroup_do_precharge(1);
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006143 if (!ret)
6144 goto retry;
6145 }
6146
6147 return ret;
6148}
6149
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02006150static const struct mm_walk_ops charge_walk_ops = {
6151 .pmd_entry = mem_cgroup_move_charge_pte_range,
6152};
6153
Tejun Heo264a0ae2016-04-21 19:09:02 -04006154static void mem_cgroup_move_charge(void)
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006155{
Daisuke Nishimura4ffef5f2010-03-10 15:22:14 -08006156 lru_add_drain_all();
Johannes Weiner312722c2014-12-10 15:44:25 -08006157 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07006158 * Signal lock_page_memcg() to take the memcg's move_lock
6159 * while we're moving its pages to another memcg. Then wait
6160 * for already started RCU-only updates to finish.
Johannes Weiner312722c2014-12-10 15:44:25 -08006161 */
6162 atomic_inc(&mc.from->moving_account);
6163 synchronize_rcu();
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006164retry:
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07006165 if (unlikely(!mmap_read_trylock(mc.mm))) {
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006166 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07006167 * Someone who are holding the mmap_lock might be waiting in
Daisuke Nishimuradfe076b2011-01-13 15:47:41 -08006168 * waitq. So we cancel all extra charges, wake up all waiters,
6169 * and retry. Because we cancel precharges, we might not be able
6170 * to move enough charges, but moving charge is a best-effort
6171 * feature anyway, so it wouldn't be a big problem.
6172 */
6173 __mem_cgroup_clear_mc();
6174 cond_resched();
6175 goto retry;
6176 }
Naoya Horiguchi26bcd642015-02-11 15:27:57 -08006177 /*
6178 * When we have consumed all precharges and failed in doing
6179 * additional charge, the page walk just aborts.
6180 */
Christoph Hellwig7b86ac32019-08-28 16:19:54 +02006181 walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops,
6182 NULL);
James Morse0247f3f2016-10-07 17:00:12 -07006183
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07006184 mmap_read_unlock(mc.mm);
Johannes Weiner312722c2014-12-10 15:44:25 -08006185 atomic_dec(&mc.from->moving_account);
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006186}
6187
Tejun Heo264a0ae2016-04-21 19:09:02 -04006188static void mem_cgroup_move_task(void)
Balbir Singh67e465a2008-02-07 00:13:54 -08006189{
Tejun Heo264a0ae2016-04-21 19:09:02 -04006190 if (mc.to) {
6191 mem_cgroup_move_charge();
KOSAKI Motohiroa4336582011-06-15 15:08:13 -07006192 mem_cgroup_clear_mc();
Tejun Heo264a0ae2016-04-21 19:09:02 -04006193 }
Balbir Singh67e465a2008-02-07 00:13:54 -08006194}
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006195#else /* !CONFIG_MMU */
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006196static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006197{
6198 return 0;
6199}
Tejun Heo1f7dd3e52015-12-03 10:18:21 -05006200static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006201{
6202}
Tejun Heo264a0ae2016-04-21 19:09:02 -04006203static void mem_cgroup_move_task(void)
Daisuke Nishimura5cfb80a2010-03-23 13:35:11 -07006204{
6205}
6206#endif
Balbir Singh67e465a2008-02-07 00:13:54 -08006207
Tejun Heof00baae2013-04-15 13:41:15 -07006208/*
6209 * Cgroup retains root cgroups across [un]mount cycles making it necessary
Tejun Heoaa6ec292014-07-09 10:08:08 -04006210 * to verify whether we're attached to the default hierarchy on each mount
6211 * attempt.
Tejun Heof00baae2013-04-15 13:41:15 -07006212 */
Tejun Heoeb954192013-08-08 20:11:23 -04006213static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
Tejun Heof00baae2013-04-15 13:41:15 -07006214{
6215 /*
Tejun Heoaa6ec292014-07-09 10:08:08 -04006216 * use_hierarchy is forced on the default hierarchy. cgroup core
Tejun Heof00baae2013-04-15 13:41:15 -07006217 * guarantees that @root doesn't have any children, so turning it
6218 * on for the root memcg is enough.
6219 */
Tejun Heo9e10a132015-09-18 11:56:28 -04006220 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov7feee5902015-03-12 16:26:19 -07006221 root_mem_cgroup->use_hierarchy = true;
6222 else
6223 root_mem_cgroup->use_hierarchy = false;
Tejun Heof00baae2013-04-15 13:41:15 -07006224}
6225
Chris Down677dc972019-03-05 15:45:55 -08006226static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value)
6227{
6228 if (value == PAGE_COUNTER_MAX)
6229 seq_puts(m, "max\n");
6230 else
6231 seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE);
6232
6233 return 0;
6234}
6235
Johannes Weiner241994ed2015-02-11 15:26:06 -08006236static u64 memory_current_read(struct cgroup_subsys_state *css,
6237 struct cftype *cft)
6238{
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08006239 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6240
6241 return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006242}
6243
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006244static int memory_min_show(struct seq_file *m, void *v)
6245{
Chris Down677dc972019-03-05 15:45:55 -08006246 return seq_puts_memcg_tunable(m,
6247 READ_ONCE(mem_cgroup_from_seq(m)->memory.min));
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006248}
6249
6250static ssize_t memory_min_write(struct kernfs_open_file *of,
6251 char *buf, size_t nbytes, loff_t off)
6252{
6253 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6254 unsigned long min;
6255 int err;
6256
6257 buf = strstrip(buf);
6258 err = page_counter_memparse(buf, "max", &min);
6259 if (err)
6260 return err;
6261
6262 page_counter_set_min(&memcg->memory, min);
6263
6264 return nbytes;
6265}
6266
Johannes Weiner241994ed2015-02-11 15:26:06 -08006267static int memory_low_show(struct seq_file *m, void *v)
6268{
Chris Down677dc972019-03-05 15:45:55 -08006269 return seq_puts_memcg_tunable(m,
6270 READ_ONCE(mem_cgroup_from_seq(m)->memory.low));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006271}
6272
6273static ssize_t memory_low_write(struct kernfs_open_file *of,
6274 char *buf, size_t nbytes, loff_t off)
6275{
6276 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6277 unsigned long low;
6278 int err;
6279
6280 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08006281 err = page_counter_memparse(buf, "max", &low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006282 if (err)
6283 return err;
6284
Roman Gushchin23067152018-06-07 17:06:22 -07006285 page_counter_set_low(&memcg->memory, low);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006286
6287 return nbytes;
6288}
6289
6290static int memory_high_show(struct seq_file *m, void *v)
6291{
Jakub Kicinskid1663a92020-06-01 21:49:49 -07006292 return seq_puts_memcg_tunable(m,
6293 READ_ONCE(mem_cgroup_from_seq(m)->memory.high));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006294}
6295
6296static ssize_t memory_high_write(struct kernfs_open_file *of,
6297 char *buf, size_t nbytes, loff_t off)
6298{
6299 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Chris Downd977aa92020-08-06 23:21:58 -07006300 unsigned int nr_retries = MAX_RECLAIM_RETRIES;
Johannes Weiner8c8c3832019-11-30 17:50:09 -08006301 bool drained = false;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006302 unsigned long high;
6303 int err;
6304
6305 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08006306 err = page_counter_memparse(buf, "max", &high);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006307 if (err)
6308 return err;
6309
Johannes Weiner8c8c3832019-11-30 17:50:09 -08006310 for (;;) {
6311 unsigned long nr_pages = page_counter_read(&memcg->memory);
6312 unsigned long reclaimed;
Johannes Weiner588083b2016-03-17 14:20:25 -07006313
Johannes Weiner8c8c3832019-11-30 17:50:09 -08006314 if (nr_pages <= high)
6315 break;
6316
6317 if (signal_pending(current))
6318 break;
6319
6320 if (!drained) {
6321 drain_all_stock(memcg);
6322 drained = true;
6323 continue;
6324 }
6325
6326 reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
6327 GFP_KERNEL, true);
6328
6329 if (!reclaimed && !nr_retries--)
6330 break;
6331 }
6332
Roman Gushchin536d3bf22020-08-06 23:21:51 -07006333 page_counter_set_high(&memcg->memory, high);
6334
Johannes Weiner19ce33a2020-08-06 23:22:12 -07006335 memcg_wb_domain_size_changed(memcg);
6336
Johannes Weiner241994ed2015-02-11 15:26:06 -08006337 return nbytes;
6338}
6339
6340static int memory_max_show(struct seq_file *m, void *v)
6341{
Chris Down677dc972019-03-05 15:45:55 -08006342 return seq_puts_memcg_tunable(m,
6343 READ_ONCE(mem_cgroup_from_seq(m)->memory.max));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006344}
6345
6346static ssize_t memory_max_write(struct kernfs_open_file *of,
6347 char *buf, size_t nbytes, loff_t off)
6348{
6349 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
Chris Downd977aa92020-08-06 23:21:58 -07006350 unsigned int nr_reclaims = MAX_RECLAIM_RETRIES;
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006351 bool drained = false;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006352 unsigned long max;
6353 int err;
6354
6355 buf = strstrip(buf);
Johannes Weinerd2973692015-02-27 15:52:04 -08006356 err = page_counter_memparse(buf, "max", &max);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006357 if (err)
6358 return err;
6359
Roman Gushchinbbec2e12018-06-07 17:06:18 -07006360 xchg(&memcg->memory.max, max);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006361
6362 for (;;) {
6363 unsigned long nr_pages = page_counter_read(&memcg->memory);
6364
6365 if (nr_pages <= max)
6366 break;
6367
Johannes Weiner7249c9f2019-11-30 17:50:06 -08006368 if (signal_pending(current))
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006369 break;
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006370
6371 if (!drained) {
6372 drain_all_stock(memcg);
6373 drained = true;
6374 continue;
6375 }
6376
6377 if (nr_reclaims) {
6378 if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
6379 GFP_KERNEL, true))
6380 nr_reclaims--;
6381 continue;
6382 }
6383
Johannes Weinere27be242018-04-10 16:29:45 -07006384 memcg_memory_event(memcg, MEMCG_OOM);
Johannes Weinerb6e6edc2016-03-17 14:20:28 -07006385 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
6386 break;
6387 }
Johannes Weiner241994ed2015-02-11 15:26:06 -08006388
Tejun Heo2529bb32015-05-22 18:23:34 -04006389 memcg_wb_domain_size_changed(memcg);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006390 return nbytes;
6391}
6392
Shakeel Butt1e577f92019-07-11 20:55:55 -07006393static void __memory_events_show(struct seq_file *m, atomic_long_t *events)
6394{
6395 seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW]));
6396 seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH]));
6397 seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX]));
6398 seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM]));
6399 seq_printf(m, "oom_kill %lu\n",
6400 atomic_long_read(&events[MEMCG_OOM_KILL]));
6401}
6402
Johannes Weiner241994ed2015-02-11 15:26:06 -08006403static int memory_events_show(struct seq_file *m, void *v)
6404{
Chris Downaa9694b2019-03-05 15:45:52 -08006405 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006406
Shakeel Butt1e577f92019-07-11 20:55:55 -07006407 __memory_events_show(m, memcg->memory_events);
6408 return 0;
6409}
Johannes Weiner241994ed2015-02-11 15:26:06 -08006410
Shakeel Butt1e577f92019-07-11 20:55:55 -07006411static int memory_events_local_show(struct seq_file *m, void *v)
6412{
6413 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
6414
6415 __memory_events_show(m, memcg->memory_events_local);
Johannes Weiner241994ed2015-02-11 15:26:06 -08006416 return 0;
6417}
6418
Johannes Weiner587d9f72016-01-20 15:03:19 -08006419static int memory_stat_show(struct seq_file *m, void *v)
6420{
Chris Downaa9694b2019-03-05 15:45:52 -08006421 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Johannes Weinerc8713d02019-07-11 20:55:59 -07006422 char *buf;
Johannes Weiner587d9f72016-01-20 15:03:19 -08006423
Johannes Weinerc8713d02019-07-11 20:55:59 -07006424 buf = memory_stat_format(memcg);
6425 if (!buf)
6426 return -ENOMEM;
6427 seq_puts(m, buf);
6428 kfree(buf);
Johannes Weiner587d9f72016-01-20 15:03:19 -08006429 return 0;
6430}
6431
Muchun Song5f9a4f42020-10-13 16:52:59 -07006432#ifdef CONFIG_NUMA
6433static int memory_numa_stat_show(struct seq_file *m, void *v)
6434{
6435 int i;
6436 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
6437
6438 for (i = 0; i < ARRAY_SIZE(memory_stats); i++) {
6439 int nid;
6440
6441 if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS)
6442 continue;
6443
6444 seq_printf(m, "%s", memory_stats[i].name);
6445 for_each_node_state(nid, N_MEMORY) {
6446 u64 size;
6447 struct lruvec *lruvec;
6448
6449 lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid));
6450 size = lruvec_page_state(lruvec, memory_stats[i].idx);
6451 size *= memory_stats[i].ratio;
6452 seq_printf(m, " N%d=%llu", nid, size);
6453 }
6454 seq_putc(m, '\n');
6455 }
6456
6457 return 0;
6458}
6459#endif
6460
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006461static int memory_oom_group_show(struct seq_file *m, void *v)
6462{
Chris Downaa9694b2019-03-05 15:45:52 -08006463 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006464
6465 seq_printf(m, "%d\n", memcg->oom_group);
6466
6467 return 0;
6468}
6469
6470static ssize_t memory_oom_group_write(struct kernfs_open_file *of,
6471 char *buf, size_t nbytes, loff_t off)
6472{
6473 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6474 int ret, oom_group;
6475
6476 buf = strstrip(buf);
6477 if (!buf)
6478 return -EINVAL;
6479
6480 ret = kstrtoint(buf, 0, &oom_group);
6481 if (ret)
6482 return ret;
6483
6484 if (oom_group != 0 && oom_group != 1)
6485 return -EINVAL;
6486
6487 memcg->oom_group = oom_group;
6488
6489 return nbytes;
6490}
6491
Johannes Weiner241994ed2015-02-11 15:26:06 -08006492static struct cftype memory_files[] = {
6493 {
6494 .name = "current",
Johannes Weinerf5fc3c5d2015-11-05 18:50:23 -08006495 .flags = CFTYPE_NOT_ON_ROOT,
Johannes Weiner241994ed2015-02-11 15:26:06 -08006496 .read_u64 = memory_current_read,
6497 },
6498 {
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006499 .name = "min",
6500 .flags = CFTYPE_NOT_ON_ROOT,
6501 .seq_show = memory_min_show,
6502 .write = memory_min_write,
6503 },
6504 {
Johannes Weiner241994ed2015-02-11 15:26:06 -08006505 .name = "low",
6506 .flags = CFTYPE_NOT_ON_ROOT,
6507 .seq_show = memory_low_show,
6508 .write = memory_low_write,
6509 },
6510 {
6511 .name = "high",
6512 .flags = CFTYPE_NOT_ON_ROOT,
6513 .seq_show = memory_high_show,
6514 .write = memory_high_write,
6515 },
6516 {
6517 .name = "max",
6518 .flags = CFTYPE_NOT_ON_ROOT,
6519 .seq_show = memory_max_show,
6520 .write = memory_max_write,
6521 },
6522 {
6523 .name = "events",
6524 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo472912a2015-09-18 18:01:59 -04006525 .file_offset = offsetof(struct mem_cgroup, events_file),
Johannes Weiner241994ed2015-02-11 15:26:06 -08006526 .seq_show = memory_events_show,
6527 },
Johannes Weiner587d9f72016-01-20 15:03:19 -08006528 {
Shakeel Butt1e577f92019-07-11 20:55:55 -07006529 .name = "events.local",
6530 .flags = CFTYPE_NOT_ON_ROOT,
6531 .file_offset = offsetof(struct mem_cgroup, events_local_file),
6532 .seq_show = memory_events_local_show,
6533 },
6534 {
Johannes Weiner587d9f72016-01-20 15:03:19 -08006535 .name = "stat",
Johannes Weiner587d9f72016-01-20 15:03:19 -08006536 .seq_show = memory_stat_show,
6537 },
Muchun Song5f9a4f42020-10-13 16:52:59 -07006538#ifdef CONFIG_NUMA
6539 {
6540 .name = "numa_stat",
6541 .seq_show = memory_numa_stat_show,
6542 },
6543#endif
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07006544 {
6545 .name = "oom.group",
6546 .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE,
6547 .seq_show = memory_oom_group_show,
6548 .write = memory_oom_group_write,
6549 },
Johannes Weiner241994ed2015-02-11 15:26:06 -08006550 { } /* terminate */
6551};
6552
Tejun Heo073219e2014-02-08 10:36:58 -05006553struct cgroup_subsys memory_cgrp_subsys = {
Tejun Heo92fb9742012-11-19 08:13:38 -08006554 .css_alloc = mem_cgroup_css_alloc,
Glauber Costad142e3e2013-02-22 16:34:52 -08006555 .css_online = mem_cgroup_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -08006556 .css_offline = mem_cgroup_css_offline,
Vladimir Davydov6df38682015-12-29 14:54:10 -08006557 .css_released = mem_cgroup_css_released,
Tejun Heo92fb9742012-11-19 08:13:38 -08006558 .css_free = mem_cgroup_css_free,
Tejun Heo1ced9532014-07-08 18:02:57 -04006559 .css_reset = mem_cgroup_css_reset,
Daisuke Nishimura7dc74be2010-03-10 15:22:13 -08006560 .can_attach = mem_cgroup_can_attach,
6561 .cancel_attach = mem_cgroup_cancel_attach,
Tejun Heo264a0ae2016-04-21 19:09:02 -04006562 .post_attach = mem_cgroup_move_task,
Tejun Heof00baae2013-04-15 13:41:15 -07006563 .bind = mem_cgroup_bind,
Johannes Weiner241994ed2015-02-11 15:26:06 -08006564 .dfl_cftypes = memory_files,
6565 .legacy_cftypes = mem_cgroup_legacy_files,
KAMEZAWA Hiroyuki6d12e2d2008-02-07 00:14:31 -08006566 .early_init = 0,
Balbir Singh8cdea7c2008-02-07 00:13:50 -08006567};
KAMEZAWA Hiroyukic0777192009-01-07 18:07:57 -08006568
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006569/*
6570 * This function calculates an individual cgroup's effective
6571 * protection which is derived from its own memory.min/low, its
6572 * parent's and siblings' settings, as well as the actual memory
6573 * distribution in the tree.
6574 *
6575 * The following rules apply to the effective protection values:
6576 *
6577 * 1. At the first level of reclaim, effective protection is equal to
6578 * the declared protection in memory.min and memory.low.
6579 *
6580 * 2. To enable safe delegation of the protection configuration, at
6581 * subsequent levels the effective protection is capped to the
6582 * parent's effective protection.
6583 *
6584 * 3. To make complex and dynamic subtrees easier to configure, the
6585 * user is allowed to overcommit the declared protection at a given
6586 * level. If that is the case, the parent's effective protection is
6587 * distributed to the children in proportion to how much protection
6588 * they have declared and how much of it they are utilizing.
6589 *
6590 * This makes distribution proportional, but also work-conserving:
6591 * if one cgroup claims much more protection than it uses memory,
6592 * the unused remainder is available to its siblings.
6593 *
6594 * 4. Conversely, when the declared protection is undercommitted at a
6595 * given level, the distribution of the larger parental protection
6596 * budget is NOT proportional. A cgroup's protection from a sibling
6597 * is capped to its own memory.min/low setting.
6598 *
Johannes Weiner8a931f82020-04-01 21:07:07 -07006599 * 5. However, to allow protecting recursive subtrees from each other
6600 * without having to declare each individual cgroup's fixed share
6601 * of the ancestor's claim to protection, any unutilized -
6602 * "floating" - protection from up the tree is distributed in
6603 * proportion to each cgroup's *usage*. This makes the protection
6604 * neutral wrt sibling cgroups and lets them compete freely over
6605 * the shared parental protection budget, but it protects the
6606 * subtree as a whole from neighboring subtrees.
6607 *
6608 * Note that 4. and 5. are not in conflict: 4. is about protecting
6609 * against immediate siblings whereas 5. is about protecting against
6610 * neighboring subtrees.
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006611 */
6612static unsigned long effective_protection(unsigned long usage,
Johannes Weiner8a931f82020-04-01 21:07:07 -07006613 unsigned long parent_usage,
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006614 unsigned long setting,
6615 unsigned long parent_effective,
6616 unsigned long siblings_protected)
6617{
6618 unsigned long protected;
Johannes Weiner8a931f82020-04-01 21:07:07 -07006619 unsigned long ep;
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006620
6621 protected = min(usage, setting);
6622 /*
6623 * If all cgroups at this level combined claim and use more
6624 * protection then what the parent affords them, distribute
6625 * shares in proportion to utilization.
6626 *
6627 * We are using actual utilization rather than the statically
6628 * claimed protection in order to be work-conserving: claimed
6629 * but unused protection is available to siblings that would
6630 * otherwise get a smaller chunk than what they claimed.
6631 */
6632 if (siblings_protected > parent_effective)
6633 return protected * parent_effective / siblings_protected;
6634
6635 /*
6636 * Ok, utilized protection of all children is within what the
6637 * parent affords them, so we know whatever this child claims
6638 * and utilizes is effectively protected.
6639 *
6640 * If there is unprotected usage beyond this value, reclaim
6641 * will apply pressure in proportion to that amount.
6642 *
6643 * If there is unutilized protection, the cgroup will be fully
6644 * shielded from reclaim, but we do return a smaller value for
6645 * protection than what the group could enjoy in theory. This
6646 * is okay. With the overcommit distribution above, effective
6647 * protection is always dependent on how memory is actually
6648 * consumed among the siblings anyway.
6649 */
Johannes Weiner8a931f82020-04-01 21:07:07 -07006650 ep = protected;
6651
6652 /*
6653 * If the children aren't claiming (all of) the protection
6654 * afforded to them by the parent, distribute the remainder in
6655 * proportion to the (unprotected) memory of each cgroup. That
6656 * way, cgroups that aren't explicitly prioritized wrt each
6657 * other compete freely over the allowance, but they are
6658 * collectively protected from neighboring trees.
6659 *
6660 * We're using unprotected memory for the weight so that if
6661 * some cgroups DO claim explicit protection, we don't protect
6662 * the same bytes twice.
Johannes Weinercd324ed2020-06-25 20:30:16 -07006663 *
6664 * Check both usage and parent_usage against the respective
6665 * protected values. One should imply the other, but they
6666 * aren't read atomically - make sure the division is sane.
Johannes Weiner8a931f82020-04-01 21:07:07 -07006667 */
6668 if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT))
6669 return ep;
Johannes Weinercd324ed2020-06-25 20:30:16 -07006670 if (parent_effective > siblings_protected &&
6671 parent_usage > siblings_protected &&
6672 usage > protected) {
Johannes Weiner8a931f82020-04-01 21:07:07 -07006673 unsigned long unclaimed;
6674
6675 unclaimed = parent_effective - siblings_protected;
6676 unclaimed *= usage - protected;
6677 unclaimed /= parent_usage - siblings_protected;
6678
6679 ep += unclaimed;
6680 }
6681
6682 return ep;
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006683}
6684
Johannes Weiner241994ed2015-02-11 15:26:06 -08006685/**
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006686 * mem_cgroup_protected - check if memory consumption is in the normal range
Sean Christopherson34c81052017-07-10 15:48:05 -07006687 * @root: the top ancestor of the sub-tree being checked
Johannes Weiner241994ed2015-02-11 15:26:06 -08006688 * @memcg: the memory cgroup to check
6689 *
Roman Gushchin23067152018-06-07 17:06:22 -07006690 * WARNING: This function is not stateless! It can only be used as part
6691 * of a top-down tree iteration, not for isolated queries.
Johannes Weiner241994ed2015-02-11 15:26:06 -08006692 */
Chris Down45c7f7e2020-08-06 23:22:05 -07006693void mem_cgroup_calculate_protection(struct mem_cgroup *root,
6694 struct mem_cgroup *memcg)
Johannes Weiner241994ed2015-02-11 15:26:06 -08006695{
Johannes Weiner8a931f82020-04-01 21:07:07 -07006696 unsigned long usage, parent_usage;
Roman Gushchin23067152018-06-07 17:06:22 -07006697 struct mem_cgroup *parent;
6698
Johannes Weiner241994ed2015-02-11 15:26:06 -08006699 if (mem_cgroup_disabled())
Chris Down45c7f7e2020-08-06 23:22:05 -07006700 return;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006701
Sean Christopherson34c81052017-07-10 15:48:05 -07006702 if (!root)
6703 root = root_mem_cgroup;
Yafang Shao22f74962020-08-06 23:22:01 -07006704
6705 /*
6706 * Effective values of the reclaim targets are ignored so they
6707 * can be stale. Have a look at mem_cgroup_protection for more
6708 * details.
6709 * TODO: calculation should be more robust so that we do not need
6710 * that special casing.
6711 */
Sean Christopherson34c81052017-07-10 15:48:05 -07006712 if (memcg == root)
Chris Down45c7f7e2020-08-06 23:22:05 -07006713 return;
Johannes Weiner241994ed2015-02-11 15:26:06 -08006714
Roman Gushchin23067152018-06-07 17:06:22 -07006715 usage = page_counter_read(&memcg->memory);
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006716 if (!usage)
Chris Down45c7f7e2020-08-06 23:22:05 -07006717 return;
Sean Christopherson34c81052017-07-10 15:48:05 -07006718
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006719 parent = parent_mem_cgroup(memcg);
Roman Gushchindf2a4192018-06-14 15:26:17 -07006720 /* No parent means a non-hierarchical mode on v1 memcg */
6721 if (!parent)
Chris Down45c7f7e2020-08-06 23:22:05 -07006722 return;
Roman Gushchindf2a4192018-06-14 15:26:17 -07006723
Johannes Weinerbc50bcc2020-04-01 21:07:03 -07006724 if (parent == root) {
Chris Downc3d53202020-04-01 21:07:27 -07006725 memcg->memory.emin = READ_ONCE(memcg->memory.min);
Chris Down03960e32020-06-25 20:30:22 -07006726 memcg->memory.elow = READ_ONCE(memcg->memory.low);
Chris Down45c7f7e2020-08-06 23:22:05 -07006727 return;
Roman Gushchinbf8d5d52018-06-07 17:07:46 -07006728 }
6729
Johannes Weiner8a931f82020-04-01 21:07:07 -07006730 parent_usage = page_counter_read(&parent->memory);
6731
Chris Downb3a78222020-04-01 21:07:33 -07006732 WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage,
Chris Downc3d53202020-04-01 21:07:27 -07006733 READ_ONCE(memcg->memory.min),
6734 READ_ONCE(parent->memory.emin),
Chris Downb3a78222020-04-01 21:07:33 -07006735 atomic_long_read(&parent->memory.children_min_usage)));
Roman Gushchin23067152018-06-07 17:06:22 -07006736
Chris Downb3a78222020-04-01 21:07:33 -07006737 WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage,
Chris Down03960e32020-06-25 20:30:22 -07006738 READ_ONCE(memcg->memory.low),
6739 READ_ONCE(parent->memory.elow),
Chris Downb3a78222020-04-01 21:07:33 -07006740 atomic_long_read(&parent->memory.children_low_usage)));
Johannes Weiner241994ed2015-02-11 15:26:06 -08006741}
6742
Johannes Weiner00501b52014-08-08 14:19:20 -07006743/**
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07006744 * mem_cgroup_charge - charge a newly allocated page to a cgroup
Johannes Weiner00501b52014-08-08 14:19:20 -07006745 * @page: page to charge
6746 * @mm: mm context of the victim
6747 * @gfp_mask: reclaim mode
Johannes Weiner00501b52014-08-08 14:19:20 -07006748 *
6749 * Try to charge @page to the memcg that @mm belongs to, reclaiming
6750 * pages according to @gfp_mask if necessary.
6751 *
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07006752 * Returns 0 on success. Otherwise, an error code is returned.
Johannes Weiner00501b52014-08-08 14:19:20 -07006753 */
Johannes Weinerd9eb1ea2020-06-03 16:02:24 -07006754int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask)
Johannes Weiner00501b52014-08-08 14:19:20 -07006755{
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07006756 unsigned int nr_pages = thp_nr_pages(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07006757 struct mem_cgroup *memcg = NULL;
Johannes Weiner00501b52014-08-08 14:19:20 -07006758 int ret = 0;
6759
6760 if (mem_cgroup_disabled())
6761 goto out;
6762
6763 if (PageSwapCache(page)) {
Johannes Weiner2d1c4982020-06-03 16:02:14 -07006764 swp_entry_t ent = { .val = page_private(page), };
6765 unsigned short id;
6766
Johannes Weiner00501b52014-08-08 14:19:20 -07006767 /*
6768 * Every swap fault against a single page tries to charge the
6769 * page, bail as early as possible. shmem_unuse() encounters
Johannes Weinereccb52e2020-06-03 16:02:11 -07006770 * already charged pages, too. page->mem_cgroup is protected
6771 * by the page lock, which serializes swap cache removal, which
Johannes Weiner00501b52014-08-08 14:19:20 -07006772 * in turn serializes uncharging.
6773 */
Vladimir Davydove993d902015-09-09 15:35:35 -07006774 VM_BUG_ON_PAGE(!PageLocked(page), page);
Huang Yingabe28952017-09-06 16:22:41 -07006775 if (compound_head(page)->mem_cgroup)
Johannes Weiner00501b52014-08-08 14:19:20 -07006776 goto out;
Vladimir Davydove993d902015-09-09 15:35:35 -07006777
Johannes Weiner2d1c4982020-06-03 16:02:14 -07006778 id = lookup_swap_cgroup_id(ent);
6779 rcu_read_lock();
6780 memcg = mem_cgroup_from_id(id);
6781 if (memcg && !css_tryget_online(&memcg->css))
6782 memcg = NULL;
6783 rcu_read_unlock();
Johannes Weiner00501b52014-08-08 14:19:20 -07006784 }
6785
Johannes Weiner00501b52014-08-08 14:19:20 -07006786 if (!memcg)
6787 memcg = get_mem_cgroup_from_mm(mm);
6788
6789 ret = try_charge(memcg, gfp_mask, nr_pages);
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07006790 if (ret)
6791 goto out_put;
Johannes Weiner00501b52014-08-08 14:19:20 -07006792
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07006793 css_get(&memcg->css);
Johannes Weinerd9eb1ea2020-06-03 16:02:24 -07006794 commit_charge(page, memcg);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006795
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006796 local_irq_disable();
Johannes Weiner3fba69a2020-06-03 16:01:31 -07006797 mem_cgroup_charge_statistics(memcg, page, nr_pages);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006798 memcg_check_events(memcg, page);
6799 local_irq_enable();
Johannes Weiner00501b52014-08-08 14:19:20 -07006800
Johannes Weiner2d1c4982020-06-03 16:02:14 -07006801 if (PageSwapCache(page)) {
Johannes Weiner00501b52014-08-08 14:19:20 -07006802 swp_entry_t entry = { .val = page_private(page) };
6803 /*
6804 * The swap entry might not get freed for a long time,
6805 * let's not wait for it. The page already received a
6806 * memory+swap charge, drop the swap entry duplicate.
6807 */
Huang Ying38d8b4e2017-07-06 15:37:18 -07006808 mem_cgroup_uncharge_swap(entry, nr_pages);
Johannes Weiner00501b52014-08-08 14:19:20 -07006809 }
Johannes Weiner00501b52014-08-08 14:19:20 -07006810
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07006811out_put:
6812 css_put(&memcg->css);
6813out:
6814 return ret;
Johannes Weiner3fea5a42020-06-03 16:01:41 -07006815}
6816
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006817struct uncharge_gather {
6818 struct mem_cgroup *memcg;
Johannes Weiner9f762db2020-06-03 16:01:44 -07006819 unsigned long nr_pages;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006820 unsigned long pgpgout;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006821 unsigned long nr_kmem;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006822 struct page *dummy_page;
6823};
6824
6825static inline void uncharge_gather_clear(struct uncharge_gather *ug)
Johannes Weiner747db952014-08-08 14:19:24 -07006826{
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006827 memset(ug, 0, sizeof(*ug));
6828}
6829
6830static void uncharge_batch(const struct uncharge_gather *ug)
6831{
Johannes Weiner747db952014-08-08 14:19:24 -07006832 unsigned long flags;
6833
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006834 if (!mem_cgroup_is_root(ug->memcg)) {
Johannes Weiner9f762db2020-06-03 16:01:44 -07006835 page_counter_uncharge(&ug->memcg->memory, ug->nr_pages);
Johannes Weiner7941d212016-01-14 15:21:23 -08006836 if (do_memsw_account())
Johannes Weiner9f762db2020-06-03 16:01:44 -07006837 page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006838 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem)
6839 page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem);
6840 memcg_oom_recover(ug->memcg);
Johannes Weinerce00a962014-09-05 08:43:57 -04006841 }
Johannes Weiner747db952014-08-08 14:19:24 -07006842
6843 local_irq_save(flags);
Johannes Weinerc9019e92018-01-31 16:16:37 -08006844 __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout);
Johannes Weiner9f762db2020-06-03 16:01:44 -07006845 __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006846 memcg_check_events(ug->memcg, ug->dummy_page);
Johannes Weiner747db952014-08-08 14:19:24 -07006847 local_irq_restore(flags);
Michal Hockof1796542020-09-04 16:35:24 -07006848
6849 /* drop reference from uncharge_page */
6850 css_put(&ug->memcg->css);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006851}
6852
6853static void uncharge_page(struct page *page, struct uncharge_gather *ug)
6854{
Johannes Weiner9f762db2020-06-03 16:01:44 -07006855 unsigned long nr_pages;
6856
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006857 VM_BUG_ON_PAGE(PageLRU(page), page);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006858
6859 if (!page->mem_cgroup)
6860 return;
6861
6862 /*
6863 * Nobody should be changing or seriously looking at
6864 * page->mem_cgroup at this point, we have fully
6865 * exclusive access to the page.
6866 */
6867
6868 if (ug->memcg != page->mem_cgroup) {
6869 if (ug->memcg) {
6870 uncharge_batch(ug);
6871 uncharge_gather_clear(ug);
6872 }
6873 ug->memcg = page->mem_cgroup;
Michal Hockof1796542020-09-04 16:35:24 -07006874
6875 /* pairs with css_put in uncharge_batch */
6876 css_get(&ug->memcg->css);
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006877 }
6878
Johannes Weiner9f762db2020-06-03 16:01:44 -07006879 nr_pages = compound_nr(page);
6880 ug->nr_pages += nr_pages;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006881
Johannes Weiner9f762db2020-06-03 16:01:44 -07006882 if (!PageKmemcg(page)) {
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006883 ug->pgpgout++;
6884 } else {
Johannes Weiner9f762db2020-06-03 16:01:44 -07006885 ug->nr_kmem += nr_pages;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006886 __ClearPageKmemcg(page);
6887 }
6888
6889 ug->dummy_page = page;
6890 page->mem_cgroup = NULL;
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07006891 css_put(&ug->memcg->css);
Johannes Weiner747db952014-08-08 14:19:24 -07006892}
6893
6894static void uncharge_list(struct list_head *page_list)
6895{
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006896 struct uncharge_gather ug;
Johannes Weiner747db952014-08-08 14:19:24 -07006897 struct list_head *next;
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006898
6899 uncharge_gather_clear(&ug);
Johannes Weiner747db952014-08-08 14:19:24 -07006900
Johannes Weiner8b592652016-03-17 14:20:31 -07006901 /*
6902 * Note that the list can be a single page->lru; hence the
6903 * do-while loop instead of a simple list_for_each_entry().
6904 */
Johannes Weiner747db952014-08-08 14:19:24 -07006905 next = page_list->next;
6906 do {
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006907 struct page *page;
6908
Johannes Weiner747db952014-08-08 14:19:24 -07006909 page = list_entry(next, struct page, lru);
6910 next = page->lru.next;
6911
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006912 uncharge_page(page, &ug);
Johannes Weiner747db952014-08-08 14:19:24 -07006913 } while (next != page_list);
6914
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006915 if (ug.memcg)
6916 uncharge_batch(&ug);
Johannes Weiner747db952014-08-08 14:19:24 -07006917}
6918
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006919/**
6920 * mem_cgroup_uncharge - uncharge a page
6921 * @page: page to uncharge
6922 *
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07006923 * Uncharge a page previously charged with mem_cgroup_charge().
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006924 */
6925void mem_cgroup_uncharge(struct page *page)
6926{
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006927 struct uncharge_gather ug;
6928
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006929 if (mem_cgroup_disabled())
6930 return;
6931
Johannes Weiner747db952014-08-08 14:19:24 -07006932 /* Don't touch page->lru of any random page, pre-check: */
Johannes Weiner1306a852014-12-10 15:44:52 -08006933 if (!page->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006934 return;
6935
Jérôme Glissea9d5ade2017-09-08 16:11:50 -07006936 uncharge_gather_clear(&ug);
6937 uncharge_page(page, &ug);
6938 uncharge_batch(&ug);
Johannes Weiner747db952014-08-08 14:19:24 -07006939}
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006940
Johannes Weiner747db952014-08-08 14:19:24 -07006941/**
6942 * mem_cgroup_uncharge_list - uncharge a list of page
6943 * @page_list: list of pages to uncharge
6944 *
6945 * Uncharge a list of pages previously charged with
Johannes Weinerf0e45fb2020-06-03 16:02:07 -07006946 * mem_cgroup_charge().
Johannes Weiner747db952014-08-08 14:19:24 -07006947 */
6948void mem_cgroup_uncharge_list(struct list_head *page_list)
6949{
6950 if (mem_cgroup_disabled())
6951 return;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006952
Johannes Weiner747db952014-08-08 14:19:24 -07006953 if (!list_empty(page_list))
6954 uncharge_list(page_list);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006955}
6956
6957/**
Johannes Weiner6a93ca82016-03-15 14:57:19 -07006958 * mem_cgroup_migrate - charge a page's replacement
6959 * @oldpage: currently circulating page
6960 * @newpage: replacement page
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006961 *
Johannes Weiner6a93ca82016-03-15 14:57:19 -07006962 * Charge @newpage as a replacement page for @oldpage. @oldpage will
6963 * be uncharged upon free.
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006964 *
6965 * Both pages must be locked, @newpage->mapping must be set up.
6966 */
Johannes Weiner6a93ca82016-03-15 14:57:19 -07006967void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006968{
Johannes Weiner29833312014-12-10 15:44:02 -08006969 struct mem_cgroup *memcg;
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08006970 unsigned int nr_pages;
Tejun Heod93c4132016-06-24 14:49:54 -07006971 unsigned long flags;
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006972
6973 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
6974 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006975 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
Johannes Weiner6abb5a82014-08-08 14:19:33 -07006976 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
6977 newpage);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006978
6979 if (mem_cgroup_disabled())
6980 return;
6981
6982 /* Page cache replacement: new page already charged? */
Johannes Weiner1306a852014-12-10 15:44:52 -08006983 if (newpage->mem_cgroup)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006984 return;
6985
Hugh Dickins45637ba2015-11-05 18:49:40 -08006986 /* Swapcache readahead pages can get replaced before being charged */
Johannes Weiner1306a852014-12-10 15:44:52 -08006987 memcg = oldpage->mem_cgroup;
Johannes Weiner29833312014-12-10 15:44:02 -08006988 if (!memcg)
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006989 return;
6990
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08006991 /* Force-charge the new page. The old one will be freed soon */
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07006992 nr_pages = thp_nr_pages(newpage);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08006993
6994 page_counter_charge(&memcg->memory, nr_pages);
6995 if (do_memsw_account())
6996 page_counter_charge(&memcg->memsw, nr_pages);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07006997
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07006998 css_get(&memcg->css);
Johannes Weinerd9eb1ea2020-06-03 16:02:24 -07006999 commit_charge(newpage, memcg);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08007000
Tejun Heod93c4132016-06-24 14:49:54 -07007001 local_irq_save(flags);
Johannes Weiner3fba69a2020-06-03 16:01:31 -07007002 mem_cgroup_charge_statistics(memcg, newpage, nr_pages);
Johannes Weiner44b7a8d2016-01-20 15:03:16 -08007003 memcg_check_events(memcg, newpage);
Tejun Heod93c4132016-06-24 14:49:54 -07007004 local_irq_restore(flags);
Johannes Weiner0a31bc92014-08-08 14:19:22 -07007005}
7006
Johannes Weineref129472016-01-14 15:21:34 -08007007DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
Johannes Weiner11092082016-01-14 15:21:26 -08007008EXPORT_SYMBOL(memcg_sockets_enabled_key);
7009
Johannes Weiner2d758072016-10-07 17:00:58 -07007010void mem_cgroup_sk_alloc(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08007011{
7012 struct mem_cgroup *memcg;
7013
Johannes Weiner2d758072016-10-07 17:00:58 -07007014 if (!mem_cgroup_sockets_enabled)
7015 return;
7016
Shakeel Butte876ecc2020-03-09 22:16:05 -07007017 /* Do not associate the sock with unrelated interrupted task's memcg. */
7018 if (in_interrupt())
7019 return;
7020
Johannes Weiner11092082016-01-14 15:21:26 -08007021 rcu_read_lock();
7022 memcg = mem_cgroup_from_task(current);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007023 if (memcg == root_mem_cgroup)
7024 goto out;
Johannes Weiner0db15292016-01-20 15:02:50 -08007025 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007026 goto out;
Shakeel Butt8965aa22020-04-01 21:07:10 -07007027 if (css_tryget(&memcg->css))
Johannes Weiner11092082016-01-14 15:21:26 -08007028 sk->sk_memcg = memcg;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007029out:
Johannes Weiner11092082016-01-14 15:21:26 -08007030 rcu_read_unlock();
7031}
Johannes Weiner11092082016-01-14 15:21:26 -08007032
Johannes Weiner2d758072016-10-07 17:00:58 -07007033void mem_cgroup_sk_free(struct sock *sk)
Johannes Weiner11092082016-01-14 15:21:26 -08007034{
Johannes Weiner2d758072016-10-07 17:00:58 -07007035 if (sk->sk_memcg)
7036 css_put(&sk->sk_memcg->css);
Johannes Weiner11092082016-01-14 15:21:26 -08007037}
7038
7039/**
7040 * mem_cgroup_charge_skmem - charge socket memory
7041 * @memcg: memcg to charge
7042 * @nr_pages: number of pages to charge
7043 *
7044 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
7045 * @memcg's configured limit, %false if the charge had to be forced.
7046 */
7047bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
7048{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007049 gfp_t gfp_mask = GFP_KERNEL;
Johannes Weiner11092082016-01-14 15:21:26 -08007050
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007051 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08007052 struct page_counter *fail;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007053
Johannes Weiner0db15292016-01-20 15:02:50 -08007054 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
7055 memcg->tcpmem_pressure = 0;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007056 return true;
7057 }
Johannes Weiner0db15292016-01-20 15:02:50 -08007058 page_counter_charge(&memcg->tcpmem, nr_pages);
7059 memcg->tcpmem_pressure = 1;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007060 return false;
Johannes Weiner11092082016-01-14 15:21:26 -08007061 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08007062
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007063 /* Don't block in the packet receive path */
7064 if (in_softirq())
7065 gfp_mask = GFP_NOWAIT;
7066
Johannes Weinerc9019e92018-01-31 16:16:37 -08007067 mod_memcg_state(memcg, MEMCG_SOCK, nr_pages);
Johannes Weinerb2807f02016-01-20 15:03:22 -08007068
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007069 if (try_charge(memcg, gfp_mask, nr_pages) == 0)
7070 return true;
7071
7072 try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08007073 return false;
7074}
7075
7076/**
7077 * mem_cgroup_uncharge_skmem - uncharge socket memory
Mike Rapoportb7701a52018-02-06 15:42:13 -08007078 * @memcg: memcg to uncharge
7079 * @nr_pages: number of pages to uncharge
Johannes Weiner11092082016-01-14 15:21:26 -08007080 */
7081void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
7082{
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007083 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
Johannes Weiner0db15292016-01-20 15:02:50 -08007084 page_counter_uncharge(&memcg->tcpmem, nr_pages);
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007085 return;
7086 }
Johannes Weinerd886f4e2016-01-20 15:02:47 -08007087
Johannes Weinerc9019e92018-01-31 16:16:37 -08007088 mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages);
Johannes Weinerb2807f02016-01-20 15:03:22 -08007089
Roman Gushchin475d0482017-09-08 16:13:09 -07007090 refill_stock(memcg, nr_pages);
Johannes Weiner11092082016-01-14 15:21:26 -08007091}
7092
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007093static int __init cgroup_memory(char *s)
7094{
7095 char *token;
7096
7097 while ((token = strsep(&s, ",")) != NULL) {
7098 if (!*token)
7099 continue;
7100 if (!strcmp(token, "nosocket"))
7101 cgroup_memory_nosocket = true;
Vladimir Davydov04823c82016-01-20 15:02:38 -08007102 if (!strcmp(token, "nokmem"))
7103 cgroup_memory_nokmem = true;
Johannes Weinerf7e1cb62016-01-14 15:21:29 -08007104 }
7105 return 0;
7106}
7107__setup("cgroup.memory=", cgroup_memory);
Johannes Weiner11092082016-01-14 15:21:26 -08007108
Michal Hocko2d110852013-02-22 16:34:43 -08007109/*
Michal Hocko10813122013-02-22 16:35:41 -08007110 * subsys_initcall() for memory controller.
7111 *
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01007112 * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this
7113 * context because of lock dependencies (cgroup_lock -> cpu hotplug) but
7114 * basically everything that doesn't depend on a specific mem_cgroup structure
7115 * should be initialized from here.
Michal Hocko2d110852013-02-22 16:34:43 -08007116 */
7117static int __init mem_cgroup_init(void)
7118{
Johannes Weiner95a045f2015-02-11 15:26:33 -08007119 int cpu, node;
7120
Sebastian Andrzej Siewior308167f2016-11-03 15:49:59 +01007121 cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL,
7122 memcg_hotplug_cpu_dead);
Johannes Weiner95a045f2015-02-11 15:26:33 -08007123
7124 for_each_possible_cpu(cpu)
7125 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
7126 drain_local_stock);
7127
7128 for_each_node(node) {
7129 struct mem_cgroup_tree_per_node *rtpn;
Johannes Weiner95a045f2015-02-11 15:26:33 -08007130
7131 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
7132 node_online(node) ? node : NUMA_NO_NODE);
7133
Mel Gormanef8f2322016-07-28 15:46:05 -07007134 rtpn->rb_root = RB_ROOT;
Davidlohr Buesofa90b2f2017-09-08 16:15:21 -07007135 rtpn->rb_rightmost = NULL;
Mel Gormanef8f2322016-07-28 15:46:05 -07007136 spin_lock_init(&rtpn->lock);
Johannes Weiner95a045f2015-02-11 15:26:33 -08007137 soft_limit_tree.rb_tree_per_node[node] = rtpn;
7138 }
7139
Michal Hocko2d110852013-02-22 16:34:43 -08007140 return 0;
7141}
7142subsys_initcall(mem_cgroup_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08007143
7144#ifdef CONFIG_MEMCG_SWAP
Arnd Bergmann358c07f2016-08-25 15:17:08 -07007145static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
7146{
Kirill Tkhai1c2d4792018-10-26 15:09:28 -07007147 while (!refcount_inc_not_zero(&memcg->id.ref)) {
Arnd Bergmann358c07f2016-08-25 15:17:08 -07007148 /*
7149 * The root cgroup cannot be destroyed, so it's refcount must
7150 * always be >= 1.
7151 */
7152 if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
7153 VM_BUG_ON(1);
7154 break;
7155 }
7156 memcg = parent_mem_cgroup(memcg);
7157 if (!memcg)
7158 memcg = root_mem_cgroup;
7159 }
7160 return memcg;
7161}
7162
Johannes Weiner21afa382015-02-11 15:26:36 -08007163/**
7164 * mem_cgroup_swapout - transfer a memsw charge to swap
7165 * @page: page whose memsw charge to transfer
7166 * @entry: swap entry to move the charge to
7167 *
7168 * Transfer the memsw charge of @page to @entry.
7169 */
7170void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
7171{
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007172 struct mem_cgroup *memcg, *swap_memcg;
Huang Yingd6810d72017-09-06 16:22:45 -07007173 unsigned int nr_entries;
Johannes Weiner21afa382015-02-11 15:26:36 -08007174 unsigned short oldid;
7175
7176 VM_BUG_ON_PAGE(PageLRU(page), page);
7177 VM_BUG_ON_PAGE(page_count(page), page);
7178
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007179 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Johannes Weiner21afa382015-02-11 15:26:36 -08007180 return;
7181
7182 memcg = page->mem_cgroup;
7183
7184 /* Readahead page, never charged */
7185 if (!memcg)
7186 return;
7187
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007188 /*
7189 * In case the memcg owning these pages has been offlined and doesn't
7190 * have an ID allocated to it anymore, charge the closest online
7191 * ancestor for the swap instead and transfer the memory+swap charge.
7192 */
7193 swap_memcg = mem_cgroup_id_get_online(memcg);
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07007194 nr_entries = thp_nr_pages(page);
Huang Yingd6810d72017-09-06 16:22:45 -07007195 /* Get references for the tail pages, too */
7196 if (nr_entries > 1)
7197 mem_cgroup_id_get_many(swap_memcg, nr_entries - 1);
7198 oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg),
7199 nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007200 VM_BUG_ON_PAGE(oldid, page);
Johannes Weinerc9019e92018-01-31 16:16:37 -08007201 mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007202
7203 page->mem_cgroup = NULL;
7204
7205 if (!mem_cgroup_is_root(memcg))
Huang Yingd6810d72017-09-06 16:22:45 -07007206 page_counter_uncharge(&memcg->memory, nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007207
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007208 if (!cgroup_memory_noswap && memcg != swap_memcg) {
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007209 if (!mem_cgroup_is_root(swap_memcg))
Huang Yingd6810d72017-09-06 16:22:45 -07007210 page_counter_charge(&swap_memcg->memsw, nr_entries);
7211 page_counter_uncharge(&memcg->memsw, nr_entries);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007212 }
7213
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07007214 /*
7215 * Interrupts should be disabled here because the caller holds the
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07007216 * i_pages lock which is taken with interrupts-off. It is
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07007217 * important here to have the interrupts disabled because it is the
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07007218 * only synchronisation we have for updating the per-CPU variables.
Sebastian Andrzej Siewiorce9ce662015-09-04 15:47:50 -07007219 */
7220 VM_BUG_ON(!irqs_disabled());
Johannes Weiner3fba69a2020-06-03 16:01:31 -07007221 mem_cgroup_charge_statistics(memcg, page, -nr_entries);
Johannes Weiner21afa382015-02-11 15:26:36 -08007222 memcg_check_events(memcg, page);
Johannes Weiner73f576c2016-07-20 15:44:57 -07007223
Johannes Weiner1a3e1f42020-08-06 23:20:45 -07007224 css_put(&memcg->css);
Johannes Weiner21afa382015-02-11 15:26:36 -08007225}
7226
Huang Ying38d8b4e2017-07-06 15:37:18 -07007227/**
7228 * mem_cgroup_try_charge_swap - try charging swap space for a page
Vladimir Davydov37e84352016-01-20 15:02:56 -08007229 * @page: page being added to swap
7230 * @entry: swap entry to charge
7231 *
Huang Ying38d8b4e2017-07-06 15:37:18 -07007232 * Try to charge @page's memcg for the swap space at @entry.
Vladimir Davydov37e84352016-01-20 15:02:56 -08007233 *
7234 * Returns 0 on success, -ENOMEM on failure.
7235 */
7236int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
7237{
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -07007238 unsigned int nr_pages = thp_nr_pages(page);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007239 struct page_counter *counter;
Huang Ying38d8b4e2017-07-06 15:37:18 -07007240 struct mem_cgroup *memcg;
Vladimir Davydov37e84352016-01-20 15:02:56 -08007241 unsigned short oldid;
7242
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007243 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov37e84352016-01-20 15:02:56 -08007244 return 0;
7245
7246 memcg = page->mem_cgroup;
7247
7248 /* Readahead page, never charged */
7249 if (!memcg)
7250 return 0;
7251
Tejun Heof3a53a32018-06-07 17:05:35 -07007252 if (!entry.val) {
7253 memcg_memory_event(memcg, MEMCG_SWAP_FAIL);
Tejun Heobb98f2c2018-06-07 17:05:31 -07007254 return 0;
Tejun Heof3a53a32018-06-07 17:05:35 -07007255 }
Tejun Heobb98f2c2018-06-07 17:05:31 -07007256
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007257 memcg = mem_cgroup_id_get_online(memcg);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007258
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007259 if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) &&
Huang Ying38d8b4e2017-07-06 15:37:18 -07007260 !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) {
Tejun Heof3a53a32018-06-07 17:05:35 -07007261 memcg_memory_event(memcg, MEMCG_SWAP_MAX);
7262 memcg_memory_event(memcg, MEMCG_SWAP_FAIL);
Vladimir Davydov1f47b612016-08-11 15:33:00 -07007263 mem_cgroup_id_put(memcg);
7264 return -ENOMEM;
7265 }
7266
Huang Ying38d8b4e2017-07-06 15:37:18 -07007267 /* Get references for the tail pages, too */
7268 if (nr_pages > 1)
7269 mem_cgroup_id_get_many(memcg, nr_pages - 1);
7270 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007271 VM_BUG_ON_PAGE(oldid, page);
Johannes Weinerc9019e92018-01-31 16:16:37 -08007272 mod_memcg_state(memcg, MEMCG_SWAP, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007273
Vladimir Davydov37e84352016-01-20 15:02:56 -08007274 return 0;
7275}
7276
Johannes Weiner21afa382015-02-11 15:26:36 -08007277/**
Huang Ying38d8b4e2017-07-06 15:37:18 -07007278 * mem_cgroup_uncharge_swap - uncharge swap space
Johannes Weiner21afa382015-02-11 15:26:36 -08007279 * @entry: swap entry to uncharge
Huang Ying38d8b4e2017-07-06 15:37:18 -07007280 * @nr_pages: the amount of swap space to uncharge
Johannes Weiner21afa382015-02-11 15:26:36 -08007281 */
Huang Ying38d8b4e2017-07-06 15:37:18 -07007282void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages)
Johannes Weiner21afa382015-02-11 15:26:36 -08007283{
7284 struct mem_cgroup *memcg;
7285 unsigned short id;
7286
Huang Ying38d8b4e2017-07-06 15:37:18 -07007287 id = swap_cgroup_record(entry, 0, nr_pages);
Johannes Weiner21afa382015-02-11 15:26:36 -08007288 rcu_read_lock();
Vladimir Davydovadbe4272015-04-15 16:13:00 -07007289 memcg = mem_cgroup_from_id(id);
Johannes Weiner21afa382015-02-11 15:26:36 -08007290 if (memcg) {
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007291 if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) {
Vladimir Davydov37e84352016-01-20 15:02:56 -08007292 if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
Huang Ying38d8b4e2017-07-06 15:37:18 -07007293 page_counter_uncharge(&memcg->swap, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007294 else
Huang Ying38d8b4e2017-07-06 15:37:18 -07007295 page_counter_uncharge(&memcg->memsw, nr_pages);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007296 }
Johannes Weinerc9019e92018-01-31 16:16:37 -08007297 mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages);
Huang Ying38d8b4e2017-07-06 15:37:18 -07007298 mem_cgroup_id_put_many(memcg, nr_pages);
Johannes Weiner21afa382015-02-11 15:26:36 -08007299 }
7300 rcu_read_unlock();
7301}
7302
Vladimir Davydovd8b38432016-01-20 15:03:07 -08007303long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
7304{
7305 long nr_swap_pages = get_nr_swap_pages();
7306
Johannes Weinereccb52e2020-06-03 16:02:11 -07007307 if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydovd8b38432016-01-20 15:03:07 -08007308 return nr_swap_pages;
7309 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
7310 nr_swap_pages = min_t(long, nr_swap_pages,
Roman Gushchinbbec2e12018-06-07 17:06:18 -07007311 READ_ONCE(memcg->swap.max) -
Vladimir Davydovd8b38432016-01-20 15:03:07 -08007312 page_counter_read(&memcg->swap));
7313 return nr_swap_pages;
7314}
7315
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007316bool mem_cgroup_swap_full(struct page *page)
7317{
7318 struct mem_cgroup *memcg;
7319
7320 VM_BUG_ON_PAGE(!PageLocked(page), page);
7321
7322 if (vm_swap_full())
7323 return true;
Johannes Weinereccb52e2020-06-03 16:02:11 -07007324 if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007325 return false;
7326
7327 memcg = page->mem_cgroup;
7328 if (!memcg)
7329 return false;
7330
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007331 for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) {
7332 unsigned long usage = page_counter_read(&memcg->swap);
7333
7334 if (usage * 2 >= READ_ONCE(memcg->swap.high) ||
7335 usage * 2 >= READ_ONCE(memcg->swap.max))
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007336 return true;
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007337 }
Vladimir Davydov5ccc5ab2016-01-20 15:03:10 -08007338
7339 return false;
7340}
7341
Johannes Weinereccb52e2020-06-03 16:02:11 -07007342static int __init setup_swap_account(char *s)
Johannes Weiner21afa382015-02-11 15:26:36 -08007343{
7344 if (!strcmp(s, "1"))
Johannes Weinereccb52e2020-06-03 16:02:11 -07007345 cgroup_memory_noswap = 0;
Johannes Weiner21afa382015-02-11 15:26:36 -08007346 else if (!strcmp(s, "0"))
Johannes Weinereccb52e2020-06-03 16:02:11 -07007347 cgroup_memory_noswap = 1;
Johannes Weiner21afa382015-02-11 15:26:36 -08007348 return 1;
7349}
Johannes Weinereccb52e2020-06-03 16:02:11 -07007350__setup("swapaccount=", setup_swap_account);
Johannes Weiner21afa382015-02-11 15:26:36 -08007351
Vladimir Davydov37e84352016-01-20 15:02:56 -08007352static u64 swap_current_read(struct cgroup_subsys_state *css,
7353 struct cftype *cft)
7354{
7355 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
7356
7357 return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
7358}
7359
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007360static int swap_high_show(struct seq_file *m, void *v)
7361{
7362 return seq_puts_memcg_tunable(m,
7363 READ_ONCE(mem_cgroup_from_seq(m)->swap.high));
7364}
7365
7366static ssize_t swap_high_write(struct kernfs_open_file *of,
7367 char *buf, size_t nbytes, loff_t off)
7368{
7369 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
7370 unsigned long high;
7371 int err;
7372
7373 buf = strstrip(buf);
7374 err = page_counter_memparse(buf, "max", &high);
7375 if (err)
7376 return err;
7377
7378 page_counter_set_high(&memcg->swap, high);
7379
7380 return nbytes;
7381}
7382
Vladimir Davydov37e84352016-01-20 15:02:56 -08007383static int swap_max_show(struct seq_file *m, void *v)
7384{
Chris Down677dc972019-03-05 15:45:55 -08007385 return seq_puts_memcg_tunable(m,
7386 READ_ONCE(mem_cgroup_from_seq(m)->swap.max));
Vladimir Davydov37e84352016-01-20 15:02:56 -08007387}
7388
7389static ssize_t swap_max_write(struct kernfs_open_file *of,
7390 char *buf, size_t nbytes, loff_t off)
7391{
7392 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
7393 unsigned long max;
7394 int err;
7395
7396 buf = strstrip(buf);
7397 err = page_counter_memparse(buf, "max", &max);
7398 if (err)
7399 return err;
7400
Tejun Heobe091022018-06-07 17:09:21 -07007401 xchg(&memcg->swap.max, max);
Vladimir Davydov37e84352016-01-20 15:02:56 -08007402
7403 return nbytes;
7404}
7405
Tejun Heof3a53a32018-06-07 17:05:35 -07007406static int swap_events_show(struct seq_file *m, void *v)
7407{
Chris Downaa9694b2019-03-05 15:45:52 -08007408 struct mem_cgroup *memcg = mem_cgroup_from_seq(m);
Tejun Heof3a53a32018-06-07 17:05:35 -07007409
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007410 seq_printf(m, "high %lu\n",
7411 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH]));
Tejun Heof3a53a32018-06-07 17:05:35 -07007412 seq_printf(m, "max %lu\n",
7413 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX]));
7414 seq_printf(m, "fail %lu\n",
7415 atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL]));
7416
7417 return 0;
7418}
7419
Vladimir Davydov37e84352016-01-20 15:02:56 -08007420static struct cftype swap_files[] = {
7421 {
7422 .name = "swap.current",
7423 .flags = CFTYPE_NOT_ON_ROOT,
7424 .read_u64 = swap_current_read,
7425 },
7426 {
Jakub Kicinski4b82ab42020-06-01 21:49:52 -07007427 .name = "swap.high",
7428 .flags = CFTYPE_NOT_ON_ROOT,
7429 .seq_show = swap_high_show,
7430 .write = swap_high_write,
7431 },
7432 {
Vladimir Davydov37e84352016-01-20 15:02:56 -08007433 .name = "swap.max",
7434 .flags = CFTYPE_NOT_ON_ROOT,
7435 .seq_show = swap_max_show,
7436 .write = swap_max_write,
7437 },
Tejun Heof3a53a32018-06-07 17:05:35 -07007438 {
7439 .name = "swap.events",
7440 .flags = CFTYPE_NOT_ON_ROOT,
7441 .file_offset = offsetof(struct mem_cgroup, swap_events_file),
7442 .seq_show = swap_events_show,
7443 },
Vladimir Davydov37e84352016-01-20 15:02:56 -08007444 { } /* terminate */
7445};
7446
Johannes Weinereccb52e2020-06-03 16:02:11 -07007447static struct cftype memsw_files[] = {
Johannes Weiner21afa382015-02-11 15:26:36 -08007448 {
7449 .name = "memsw.usage_in_bytes",
7450 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
7451 .read_u64 = mem_cgroup_read_u64,
7452 },
7453 {
7454 .name = "memsw.max_usage_in_bytes",
7455 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
7456 .write = mem_cgroup_reset,
7457 .read_u64 = mem_cgroup_read_u64,
7458 },
7459 {
7460 .name = "memsw.limit_in_bytes",
7461 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
7462 .write = mem_cgroup_write,
7463 .read_u64 = mem_cgroup_read_u64,
7464 },
7465 {
7466 .name = "memsw.failcnt",
7467 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
7468 .write = mem_cgroup_reset,
7469 .read_u64 = mem_cgroup_read_u64,
7470 },
7471 { }, /* terminate */
7472};
7473
Bhupesh Sharma82ff1652020-07-23 21:15:21 -07007474/*
7475 * If mem_cgroup_swap_init() is implemented as a subsys_initcall()
7476 * instead of a core_initcall(), this could mean cgroup_memory_noswap still
7477 * remains set to false even when memcg is disabled via "cgroup_disable=memory"
7478 * boot parameter. This may result in premature OOPS inside
7479 * mem_cgroup_get_nr_swap_pages() function in corner cases.
7480 */
Johannes Weiner21afa382015-02-11 15:26:36 -08007481static int __init mem_cgroup_swap_init(void)
7482{
Johannes Weiner2d1c4982020-06-03 16:02:14 -07007483 /* No memory control -> no swap control */
7484 if (mem_cgroup_disabled())
7485 cgroup_memory_noswap = true;
7486
7487 if (cgroup_memory_noswap)
Johannes Weinereccb52e2020-06-03 16:02:11 -07007488 return 0;
7489
7490 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files));
7491 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files));
7492
Johannes Weiner21afa382015-02-11 15:26:36 -08007493 return 0;
7494}
Bhupesh Sharma82ff1652020-07-23 21:15:21 -07007495core_initcall(mem_cgroup_swap_init);
Johannes Weiner21afa382015-02-11 15:26:36 -08007496
7497#endif /* CONFIG_MEMCG_SWAP */