source: src/linux/universal/linux-4.9/mm/memcontrol.c @ 31662

Last change on this file since 31662 was 31662, checked in by brainslayer, 4 months ago

use new squashfs in all kernels

File size: 154.3 KB
Line 
1/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
9 * Memory thresholds
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
12 *
13 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
16 *
17 * Native page reclaim
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
22 *
23 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
27 *
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31 * GNU General Public License for more details.
32 */
33
34#include <linux/page_counter.h>
35#include <linux/memcontrol.h>
36#include <linux/cgroup.h>
37#include <linux/mm.h>
38#include <linux/hugetlb.h>
39#include <linux/pagemap.h>
40#include <linux/smp.h>
41#include <linux/page-flags.h>
42#include <linux/backing-dev.h>
43#include <linux/bit_spinlock.h>
44#include <linux/rcupdate.h>
45#include <linux/limits.h>
46#include <linux/export.h>
47#include <linux/mutex.h>
48#include <linux/rbtree.h>
49#include <linux/slab.h>
50#include <linux/swap.h>
51#include <linux/swapops.h>
52#include <linux/spinlock.h>
53#include <linux/eventfd.h>
54#include <linux/poll.h>
55#include <linux/sort.h>
56#include <linux/fs.h>
57#include <linux/seq_file.h>
58#include <linux/vmpressure.h>
59#include <linux/mm_inline.h>
60#include <linux/swap_cgroup.h>
61#include <linux/cpu.h>
62#include <linux/oom.h>
63#include <linux/lockdep.h>
64#include <linux/file.h>
65#include <linux/tracehook.h>
66#include "internal.h"
67#include <net/sock.h>
68#include <net/ip.h>
69#include "slab.h"
70
71#include <asm/uaccess.h>
72
73#include <trace/events/vmscan.h>
74
75struct cgroup_subsys memory_cgrp_subsys __read_mostly;
76EXPORT_SYMBOL(memory_cgrp_subsys);
77
78struct mem_cgroup *root_mem_cgroup __read_mostly;
79
80#define MEM_CGROUP_RECLAIM_RETRIES      5
81
82/* Socket memory accounting disabled? */
83static bool cgroup_memory_nosocket;
84
85/* Kernel memory accounting disabled? */
86static bool cgroup_memory_nokmem;
87
88/* Whether the swap controller is active */
89#ifdef CONFIG_MEMCG_SWAP
90int do_swap_account __read_mostly;
91#else
92#define do_swap_account         0
93#endif
94
95/* Whether legacy memory+swap accounting is active */
96static bool do_memsw_account(void)
97{
98        return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account;
99}
100
101static const char * const mem_cgroup_stat_names[] = {
102        "cache",
103        "rss",
104        "rss_huge",
105        "mapped_file",
106        "dirty",
107        "writeback",
108        "swap",
109};
110
111static const char * const mem_cgroup_events_names[] = {
112        "pgpgin",
113        "pgpgout",
114        "pgfault",
115        "pgmajfault",
116};
117
118static const char * const mem_cgroup_lru_names[] = {
119        "inactive_anon",
120        "active_anon",
121        "inactive_file",
122        "active_file",
123        "unevictable",
124};
125
126#define THRESHOLDS_EVENTS_TARGET 128
127#define SOFTLIMIT_EVENTS_TARGET 1024
128#define NUMAINFO_EVENTS_TARGET  1024
129
130/*
131 * Cgroups above their limits are maintained in a RB-Tree, independent of
132 * their hierarchy representation
133 */
134
135struct mem_cgroup_tree_per_node {
136        struct rb_root rb_root;
137        spinlock_t lock;
138};
139
140struct mem_cgroup_tree {
141        struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
142};
143
144static struct mem_cgroup_tree soft_limit_tree __read_mostly;
145
146/* for OOM */
147struct mem_cgroup_eventfd_list {
148        struct list_head list;
149        struct eventfd_ctx *eventfd;
150};
151
152/*
153 * cgroup_event represents events which userspace want to receive.
154 */
155struct mem_cgroup_event {
156        /*
157         * memcg which the event belongs to.
158         */
159        struct mem_cgroup *memcg;
160        /*
161         * eventfd to signal userspace about the event.
162         */
163        struct eventfd_ctx *eventfd;
164        /*
165         * Each of these stored in a list by the cgroup.
166         */
167        struct list_head list;
168        /*
169         * register_event() callback will be used to add new userspace
170         * waiter for changes related to this event.  Use eventfd_signal()
171         * on eventfd to send notification to userspace.
172         */
173        int (*register_event)(struct mem_cgroup *memcg,
174                              struct eventfd_ctx *eventfd, const char *args);
175        /*
176         * unregister_event() callback will be called when userspace closes
177         * the eventfd or on cgroup removing.  This callback must be set,
178         * if you want provide notification functionality.
179         */
180        void (*unregister_event)(struct mem_cgroup *memcg,
181                                 struct eventfd_ctx *eventfd);
182        /*
183         * All fields below needed to unregister event when
184         * userspace closes eventfd.
185         */
186        poll_table pt;
187        wait_queue_head_t *wqh;
188        wait_queue_t wait;
189        struct work_struct remove;
190};
191
192static void mem_cgroup_threshold(struct mem_cgroup *memcg);
193static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
194
195/* Stuffs for move charges at task migration. */
196/*
197 * Types of charges to be moved.
198 */
199#define MOVE_ANON       0x1U
200#define MOVE_FILE       0x2U
201#define MOVE_MASK       (MOVE_ANON | MOVE_FILE)
202
203/* "mc" and its members are protected by cgroup_mutex */
204static struct move_charge_struct {
205        spinlock_t        lock; /* for from, to */
206        struct mm_struct  *mm;
207        struct mem_cgroup *from;
208        struct mem_cgroup *to;
209        unsigned long flags;
210        unsigned long precharge;
211        unsigned long moved_charge;
212        unsigned long moved_swap;
213        struct task_struct *moving_task;        /* a task moving charges */
214        wait_queue_head_t waitq;                /* a waitq for other context */
215} mc = {
216        .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
217        .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
218};
219
220/*
221 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
222 * limit reclaim to prevent infinite loops, if they ever occur.
223 */
224#define MEM_CGROUP_MAX_RECLAIM_LOOPS            100
225#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
226
227enum charge_type {
228        MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
229        MEM_CGROUP_CHARGE_TYPE_ANON,
230        MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
231        MEM_CGROUP_CHARGE_TYPE_DROP,    /* a page was unused swap cache */
232        NR_CHARGE_TYPE,
233};
234
235/* for encoding cft->private value on file */
236enum res_type {
237        _MEM,
238        _MEMSWAP,
239        _OOM_TYPE,
240        _KMEM,
241        _TCP,
242};
243
244#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
245#define MEMFILE_TYPE(val)       ((val) >> 16 & 0xffff)
246#define MEMFILE_ATTR(val)       ((val) & 0xffff)
247/* Used for OOM nofiier */
248#define OOM_CONTROL             (0)
249
250/* Some nice accessors for the vmpressure. */
251struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
252{
253        if (!memcg)
254                memcg = root_mem_cgroup;
255        return &memcg->vmpressure;
256}
257
258struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
259{
260        return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
261}
262
263static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
264{
265        return (memcg == root_mem_cgroup);
266}
267
268#ifndef CONFIG_SLOB
269/*
270 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
271 * The main reason for not using cgroup id for this:
272 *  this works better in sparse environments, where we have a lot of memcgs,
273 *  but only a few kmem-limited. Or also, if we have, for instance, 200
274 *  memcgs, and none but the 200th is kmem-limited, we'd have to have a
275 *  200 entry array for that.
276 *
277 * The current size of the caches array is stored in memcg_nr_cache_ids. It
278 * will double each time we have to increase it.
279 */
280static DEFINE_IDA(memcg_cache_ida);
281int memcg_nr_cache_ids;
282
283/* Protects memcg_nr_cache_ids */
284static DECLARE_RWSEM(memcg_cache_ids_sem);
285
286void memcg_get_cache_ids(void)
287{
288        down_read(&memcg_cache_ids_sem);
289}
290
291void memcg_put_cache_ids(void)
292{
293        up_read(&memcg_cache_ids_sem);
294}
295
296/*
297 * MIN_SIZE is different than 1, because we would like to avoid going through
298 * the alloc/free process all the time. In a small machine, 4 kmem-limited
299 * cgroups is a reasonable guess. In the future, it could be a parameter or
300 * tunable, but that is strictly not necessary.
301 *
302 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
303 * this constant directly from cgroup, but it is understandable that this is
304 * better kept as an internal representation in cgroup.c. In any case, the
305 * cgrp_id space is not getting any smaller, and we don't have to necessarily
306 * increase ours as well if it increases.
307 */
308#define MEMCG_CACHES_MIN_SIZE 4
309#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
310
311/*
312 * A lot of the calls to the cache allocation functions are expected to be
313 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
314 * conditional to this static branch, we'll have to allow modules that does
315 * kmem_cache_alloc and the such to see this symbol as well
316 */
317DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
318EXPORT_SYMBOL(memcg_kmem_enabled_key);
319
320#endif /* !CONFIG_SLOB */
321
322/**
323 * mem_cgroup_css_from_page - css of the memcg associated with a page
324 * @page: page of interest
325 *
326 * If memcg is bound to the default hierarchy, css of the memcg associated
327 * with @page is returned.  The returned css remains associated with @page
328 * until it is released.
329 *
330 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
331 * is returned.
332 */
333struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
334{
335        struct mem_cgroup *memcg;
336
337        memcg = page->mem_cgroup;
338
339        if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
340                memcg = root_mem_cgroup;
341
342        return &memcg->css;
343}
344
345/**
346 * page_cgroup_ino - return inode number of the memcg a page is charged to
347 * @page: the page
348 *
349 * Look up the closest online ancestor of the memory cgroup @page is charged to
350 * and return its inode number or 0 if @page is not charged to any cgroup. It
351 * is safe to call this function without holding a reference to @page.
352 *
353 * Note, this function is inherently racy, because there is nothing to prevent
354 * the cgroup inode from getting torn down and potentially reallocated a moment
355 * after page_cgroup_ino() returns, so it only should be used by callers that
356 * do not care (such as procfs interfaces).
357 */
358ino_t page_cgroup_ino(struct page *page)
359{
360        struct mem_cgroup *memcg;
361        unsigned long ino = 0;
362
363        rcu_read_lock();
364        memcg = READ_ONCE(page->mem_cgroup);
365        while (memcg && !(memcg->css.flags & CSS_ONLINE))
366                memcg = parent_mem_cgroup(memcg);
367        if (memcg)
368                ino = cgroup_ino(memcg->css.cgroup);
369        rcu_read_unlock();
370        return ino;
371}
372
373static struct mem_cgroup_per_node *
374mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page)
375{
376        int nid = page_to_nid(page);
377
378        return memcg->nodeinfo[nid];
379}
380
381static struct mem_cgroup_tree_per_node *
382soft_limit_tree_node(int nid)
383{
384        return soft_limit_tree.rb_tree_per_node[nid];
385}
386
387static struct mem_cgroup_tree_per_node *
388soft_limit_tree_from_page(struct page *page)
389{
390        int nid = page_to_nid(page);
391
392        return soft_limit_tree.rb_tree_per_node[nid];
393}
394
395static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz,
396                                         struct mem_cgroup_tree_per_node *mctz,
397                                         unsigned long new_usage_in_excess)
398{
399        struct rb_node **p = &mctz->rb_root.rb_node;
400        struct rb_node *parent = NULL;
401        struct mem_cgroup_per_node *mz_node;
402
403        if (mz->on_tree)
404                return;
405
406        mz->usage_in_excess = new_usage_in_excess;
407        if (!mz->usage_in_excess)
408                return;
409        while (*p) {
410                parent = *p;
411                mz_node = rb_entry(parent, struct mem_cgroup_per_node,
412                                        tree_node);
413                if (mz->usage_in_excess < mz_node->usage_in_excess)
414                        p = &(*p)->rb_left;
415                /*
416                 * We can't avoid mem cgroups that are over their soft
417                 * limit by the same amount
418                 */
419                else if (mz->usage_in_excess >= mz_node->usage_in_excess)
420                        p = &(*p)->rb_right;
421        }
422        rb_link_node(&mz->tree_node, parent, p);
423        rb_insert_color(&mz->tree_node, &mctz->rb_root);
424        mz->on_tree = true;
425}
426
427static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
428                                         struct mem_cgroup_tree_per_node *mctz)
429{
430        if (!mz->on_tree)
431                return;
432        rb_erase(&mz->tree_node, &mctz->rb_root);
433        mz->on_tree = false;
434}
435
436static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
437                                       struct mem_cgroup_tree_per_node *mctz)
438{
439        unsigned long flags;
440
441        spin_lock_irqsave(&mctz->lock, flags);
442        __mem_cgroup_remove_exceeded(mz, mctz);
443        spin_unlock_irqrestore(&mctz->lock, flags);
444}
445
446static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
447{
448        unsigned long nr_pages = page_counter_read(&memcg->memory);
449        unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
450        unsigned long excess = 0;
451
452        if (nr_pages > soft_limit)
453                excess = nr_pages - soft_limit;
454
455        return excess;
456}
457
458static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
459{
460        unsigned long excess;
461        struct mem_cgroup_per_node *mz;
462        struct mem_cgroup_tree_per_node *mctz;
463
464        mctz = soft_limit_tree_from_page(page);
465        /*
466         * Necessary to update all ancestors when hierarchy is used.
467         * because their event counter is not touched.
468         */
469        for (; memcg; memcg = parent_mem_cgroup(memcg)) {
470                mz = mem_cgroup_page_nodeinfo(memcg, page);
471                excess = soft_limit_excess(memcg);
472                /*
473                 * We have to update the tree if mz is on RB-tree or
474                 * mem is over its softlimit.
475                 */
476                if (excess || mz->on_tree) {
477                        unsigned long flags;
478
479                        spin_lock_irqsave(&mctz->lock, flags);
480                        /* if on-tree, remove it */
481                        if (mz->on_tree)
482                                __mem_cgroup_remove_exceeded(mz, mctz);
483                        /*
484                         * Insert again. mz->usage_in_excess will be updated.
485                         * If excess is 0, no tree ops.
486                         */
487                        __mem_cgroup_insert_exceeded(mz, mctz, excess);
488                        spin_unlock_irqrestore(&mctz->lock, flags);
489                }
490        }
491}
492
493static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
494{
495        struct mem_cgroup_tree_per_node *mctz;
496        struct mem_cgroup_per_node *mz;
497        int nid;
498
499        for_each_node(nid) {
500                mz = mem_cgroup_nodeinfo(memcg, nid);
501                mctz = soft_limit_tree_node(nid);
502                mem_cgroup_remove_exceeded(mz, mctz);
503        }
504}
505
506static struct mem_cgroup_per_node *
507__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
508{
509        struct rb_node *rightmost = NULL;
510        struct mem_cgroup_per_node *mz;
511
512retry:
513        mz = NULL;
514        rightmost = rb_last(&mctz->rb_root);
515        if (!rightmost)
516                goto done;              /* Nothing to reclaim from */
517
518        mz = rb_entry(rightmost, struct mem_cgroup_per_node, tree_node);
519        /*
520         * Remove the node now but someone else can add it back,
521         * we will to add it back at the end of reclaim to its correct
522         * position in the tree.
523         */
524        __mem_cgroup_remove_exceeded(mz, mctz);
525        if (!soft_limit_excess(mz->memcg) ||
526            !css_tryget_online(&mz->memcg->css))
527                goto retry;
528done:
529        return mz;
530}
531
532static struct mem_cgroup_per_node *
533mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
534{
535        struct mem_cgroup_per_node *mz;
536
537        spin_lock_irq(&mctz->lock);
538        mz = __mem_cgroup_largest_soft_limit_node(mctz);
539        spin_unlock_irq(&mctz->lock);
540        return mz;
541}
542
543/*
544 * Return page count for single (non recursive) @memcg.
545 *
546 * Implementation Note: reading percpu statistics for memcg.
547 *
548 * Both of vmstat[] and percpu_counter has threshold and do periodic
549 * synchronization to implement "quick" read. There are trade-off between
550 * reading cost and precision of value. Then, we may have a chance to implement
551 * a periodic synchronization of counter in memcg's counter.
552 *
553 * But this _read() function is used for user interface now. The user accounts
554 * memory usage by memory cgroup and he _always_ requires exact value because
555 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
556 * have to visit all online cpus and make sum. So, for now, unnecessary
557 * synchronization is not implemented. (just implemented for cpu hotplug)
558 *
559 * If there are kernel internal actions which can make use of some not-exact
560 * value, and reading all cpu value can be performance bottleneck in some
561 * common workload, threshold and synchronization as vmstat[] should be
562 * implemented.
563 */
564static unsigned long
565mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
566{
567        long val = 0;
568        int cpu;
569
570        /* Per-cpu values can be negative, use a signed accumulator */
571        for_each_possible_cpu(cpu)
572                val += per_cpu(memcg->stat->count[idx], cpu);
573        /*
574         * Summing races with updates, so val may be negative.  Avoid exposing
575         * transient negative values.
576         */
577        if (val < 0)
578                val = 0;
579        return val;
580}
581
582static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
583                                            enum mem_cgroup_events_index idx)
584{
585        unsigned long val = 0;
586        int cpu;
587
588        for_each_possible_cpu(cpu)
589                val += per_cpu(memcg->stat->events[idx], cpu);
590        return val;
591}
592
593static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
594                                         struct page *page,
595                                         bool compound, int nr_pages)
596{
597        /*
598         * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
599         * counted as CACHE even if it's on ANON LRU.
600         */
601        if (PageAnon(page))
602                __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
603                                nr_pages);
604        else
605                __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
606                                nr_pages);
607
608        if (compound) {
609                VM_BUG_ON_PAGE(!PageTransHuge(page), page);
610                __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
611                                nr_pages);
612        }
613
614        /* pagein of a big page is an event. So, ignore page size */
615        if (nr_pages > 0)
616                __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
617        else {
618                __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
619                nr_pages = -nr_pages; /* for event */
620        }
621
622        __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
623}
624
625unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
626                                           int nid, unsigned int lru_mask)
627{
628        struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg);
629        unsigned long nr = 0;
630        enum lru_list lru;
631
632        VM_BUG_ON((unsigned)nid >= nr_node_ids);
633
634        for_each_lru(lru) {
635                if (!(BIT(lru) & lru_mask))
636                        continue;
637                nr += mem_cgroup_get_lru_size(lruvec, lru);
638        }
639        return nr;
640}
641
642static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
643                        unsigned int lru_mask)
644{
645        unsigned long nr = 0;
646        int nid;
647
648        for_each_node_state(nid, N_MEMORY)
649                nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
650        return nr;
651}
652
653static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
654                                       enum mem_cgroup_events_target target)
655{
656        unsigned long val, next;
657
658        val = __this_cpu_read(memcg->stat->nr_page_events);
659        next = __this_cpu_read(memcg->stat->targets[target]);
660        /* from time_after() in jiffies.h */
661        if ((long)next - (long)val < 0) {
662                switch (target) {
663                case MEM_CGROUP_TARGET_THRESH:
664                        next = val + THRESHOLDS_EVENTS_TARGET;
665                        break;
666                case MEM_CGROUP_TARGET_SOFTLIMIT:
667                        next = val + SOFTLIMIT_EVENTS_TARGET;
668                        break;
669                case MEM_CGROUP_TARGET_NUMAINFO:
670                        next = val + NUMAINFO_EVENTS_TARGET;
671                        break;
672                default:
673                        break;
674                }
675                __this_cpu_write(memcg->stat->targets[target], next);
676                return true;
677        }
678        return false;
679}
680
681/*
682 * Check events in order.
683 *
684 */
685static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
686{
687        /* threshold event is triggered in finer grain than soft limit */
688        if (unlikely(mem_cgroup_event_ratelimit(memcg,
689                                                MEM_CGROUP_TARGET_THRESH))) {
690                bool do_softlimit;
691                bool do_numainfo __maybe_unused;
692
693                do_softlimit = mem_cgroup_event_ratelimit(memcg,
694                                                MEM_CGROUP_TARGET_SOFTLIMIT);
695#if MAX_NUMNODES > 1
696                do_numainfo = mem_cgroup_event_ratelimit(memcg,
697                                                MEM_CGROUP_TARGET_NUMAINFO);
698#endif
699                mem_cgroup_threshold(memcg);
700                if (unlikely(do_softlimit))
701                        mem_cgroup_update_tree(memcg, page);
702#if MAX_NUMNODES > 1
703                if (unlikely(do_numainfo))
704                        atomic_inc(&memcg->numainfo_events);
705#endif
706        }
707}
708
709struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
710{
711        /*
712         * mm_update_next_owner() may clear mm->owner to NULL
713         * if it races with swapoff, page migration, etc.
714         * So this can be called with p == NULL.
715         */
716        if (unlikely(!p))
717                return NULL;
718
719        return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
720}
721EXPORT_SYMBOL(mem_cgroup_from_task);
722
723static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
724{
725        struct mem_cgroup *memcg = NULL;
726
727        rcu_read_lock();
728        do {
729                /*
730                 * Page cache insertions can happen withou an
731                 * actual mm context, e.g. during disk probing
732                 * on boot, loopback IO, acct() writes etc.
733                 */
734                if (unlikely(!mm))
735                        memcg = root_mem_cgroup;
736                else {
737                        memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
738                        if (unlikely(!memcg))
739                                memcg = root_mem_cgroup;
740                }
741        } while (!css_tryget_online(&memcg->css));
742        rcu_read_unlock();
743        return memcg;
744}
745
746/**
747 * mem_cgroup_iter - iterate over memory cgroup hierarchy
748 * @root: hierarchy root
749 * @prev: previously returned memcg, NULL on first invocation
750 * @reclaim: cookie for shared reclaim walks, NULL for full walks
751 *
752 * Returns references to children of the hierarchy below @root, or
753 * @root itself, or %NULL after a full round-trip.
754 *
755 * Caller must pass the return value in @prev on subsequent
756 * invocations for reference counting, or use mem_cgroup_iter_break()
757 * to cancel a hierarchy walk before the round-trip is complete.
758 *
759 * Reclaimers can specify a zone and a priority level in @reclaim to
760 * divide up the memcgs in the hierarchy among all concurrent
761 * reclaimers operating on the same zone and priority.
762 */
763struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
764                                   struct mem_cgroup *prev,
765                                   struct mem_cgroup_reclaim_cookie *reclaim)
766{
767        struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
768        struct cgroup_subsys_state *css = NULL;
769        struct mem_cgroup *memcg = NULL;
770        struct mem_cgroup *pos = NULL;
771
772        if (mem_cgroup_disabled())
773                return NULL;
774
775        if (!root)
776                root = root_mem_cgroup;
777
778        if (prev && !reclaim)
779                pos = prev;
780
781        if (!root->use_hierarchy && root != root_mem_cgroup) {
782                if (prev)
783                        goto out;
784                return root;
785        }
786
787        rcu_read_lock();
788
789        if (reclaim) {
790                struct mem_cgroup_per_node *mz;
791
792                mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id);
793                iter = &mz->iter[reclaim->priority];
794
795                if (prev && reclaim->generation != iter->generation)
796                        goto out_unlock;
797
798                while (1) {
799                        pos = READ_ONCE(iter->position);
800                        if (!pos || css_tryget(&pos->css))
801                                break;
802                        /*
803                         * css reference reached zero, so iter->position will
804                         * be cleared by ->css_released. However, we should not
805                         * rely on this happening soon, because ->css_released
806                         * is called from a work queue, and by busy-waiting we
807                         * might block it. So we clear iter->position right
808                         * away.
809                         */
810                        (void)cmpxchg(&iter->position, pos, NULL);
811                }
812        }
813
814        if (pos)
815                css = &pos->css;
816
817        for (;;) {
818                css = css_next_descendant_pre(css, &root->css);
819                if (!css) {
820                        /*
821                         * Reclaimers share the hierarchy walk, and a
822                         * new one might jump in right at the end of
823                         * the hierarchy - make sure they see at least
824                         * one group and restart from the beginning.
825                         */
826                        if (!prev)
827                                continue;
828                        break;
829                }
830
831                /*
832                 * Verify the css and acquire a reference.  The root
833                 * is provided by the caller, so we know it's alive
834                 * and kicking, and don't take an extra reference.
835                 */
836                memcg = mem_cgroup_from_css(css);
837
838                if (css == &root->css)
839                        break;
840
841                if (css_tryget(css))
842                        break;
843
844                memcg = NULL;
845        }
846
847        if (reclaim) {
848                /*
849                 * The position could have already been updated by a competing
850                 * thread, so check that the value hasn't changed since we read
851                 * it to avoid reclaiming from the same cgroup twice.
852                 */
853                (void)cmpxchg(&iter->position, pos, memcg);
854
855                if (pos)
856                        css_put(&pos->css);
857
858                if (!memcg)
859                        iter->generation++;
860                else if (!prev)
861                        reclaim->generation = iter->generation;
862        }
863
864out_unlock:
865        rcu_read_unlock();
866out:
867        if (prev && prev != root)
868                css_put(&prev->css);
869
870        return memcg;
871}
872
873/**
874 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
875 * @root: hierarchy root
876 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
877 */
878void mem_cgroup_iter_break(struct mem_cgroup *root,
879                           struct mem_cgroup *prev)
880{
881        if (!root)
882                root = root_mem_cgroup;
883        if (prev && prev != root)
884                css_put(&prev->css);
885}
886
887static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
888{
889        struct mem_cgroup *memcg = dead_memcg;
890        struct mem_cgroup_reclaim_iter *iter;
891        struct mem_cgroup_per_node *mz;
892        int nid;
893        int i;
894
895        while ((memcg = parent_mem_cgroup(memcg))) {
896                for_each_node(nid) {
897                        mz = mem_cgroup_nodeinfo(memcg, nid);
898                        for (i = 0; i <= DEF_PRIORITY; i++) {
899                                iter = &mz->iter[i];
900                                cmpxchg(&iter->position,
901                                        dead_memcg, NULL);
902                        }
903                }
904        }
905}
906
907/*
908 * Iteration constructs for visiting all cgroups (under a tree).  If
909 * loops are exited prematurely (break), mem_cgroup_iter_break() must
910 * be used for reference counting.
911 */
912#define for_each_mem_cgroup_tree(iter, root)            \
913        for (iter = mem_cgroup_iter(root, NULL, NULL);  \
914             iter != NULL;                              \
915             iter = mem_cgroup_iter(root, iter, NULL))
916
917#define for_each_mem_cgroup(iter)                       \
918        for (iter = mem_cgroup_iter(NULL, NULL, NULL);  \
919             iter != NULL;                              \
920             iter = mem_cgroup_iter(NULL, iter, NULL))
921
922/**
923 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
924 * @memcg: hierarchy root
925 * @fn: function to call for each task
926 * @arg: argument passed to @fn
927 *
928 * This function iterates over tasks attached to @memcg or to any of its
929 * descendants and calls @fn for each task. If @fn returns a non-zero
930 * value, the function breaks the iteration loop and returns the value.
931 * Otherwise, it will iterate over all tasks and return 0.
932 *
933 * This function must not be called for the root memory cgroup.
934 */
935int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
936                          int (*fn)(struct task_struct *, void *), void *arg)
937{
938        struct mem_cgroup *iter;
939        int ret = 0;
940
941        BUG_ON(memcg == root_mem_cgroup);
942
943        for_each_mem_cgroup_tree(iter, memcg) {
944                struct css_task_iter it;
945                struct task_struct *task;
946
947                css_task_iter_start(&iter->css, &it);
948                while (!ret && (task = css_task_iter_next(&it)))
949                        ret = fn(task, arg);
950                css_task_iter_end(&it);
951                if (ret) {
952                        mem_cgroup_iter_break(memcg, iter);
953                        break;
954                }
955        }
956        return ret;
957}
958
959/**
960 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
961 * @page: the page
962 * @zone: zone of the page
963 *
964 * This function is only safe when following the LRU page isolation
965 * and putback protocol: the LRU lock must be held, and the page must
966 * either be PageLRU() or the caller must have isolated/allocated it.
967 */
968struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat)
969{
970        struct mem_cgroup_per_node *mz;
971        struct mem_cgroup *memcg;
972        struct lruvec *lruvec;
973
974        if (mem_cgroup_disabled()) {
975                lruvec = &pgdat->lruvec;
976                goto out;
977        }
978
979        memcg = page->mem_cgroup;
980        /*
981         * Swapcache readahead pages are added to the LRU - and
982         * possibly migrated - before they are charged.
983         */
984        if (!memcg)
985                memcg = root_mem_cgroup;
986
987        mz = mem_cgroup_page_nodeinfo(memcg, page);
988        lruvec = &mz->lruvec;
989out:
990        /*
991         * Since a node can be onlined after the mem_cgroup was created,
992         * we have to be prepared to initialize lruvec->zone here;
993         * and if offlined then reonlined, we need to reinitialize it.
994         */
995        if (unlikely(lruvec->pgdat != pgdat))
996                lruvec->pgdat = pgdat;
997        return lruvec;
998}
999
1000/**
1001 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1002 * @lruvec: mem_cgroup per zone lru vector
1003 * @lru: index of lru list the page is sitting on
1004 * @zid: zone id of the accounted pages
1005 * @nr_pages: positive when adding or negative when removing
1006 *
1007 * This function must be called under lru_lock, just before a page is added
1008 * to or just after a page is removed from an lru list (that ordering being
1009 * so as to allow it to check that lru_size 0 is consistent with list_empty).
1010 */
1011void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1012                                int zid, int nr_pages)
1013{
1014        struct mem_cgroup_per_node *mz;
1015        unsigned long *lru_size;
1016        long size;
1017
1018        if (mem_cgroup_disabled())
1019                return;
1020
1021        mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
1022        lru_size = &mz->lru_zone_size[zid][lru];
1023
1024        if (nr_pages < 0)
1025                *lru_size += nr_pages;
1026
1027        size = *lru_size;
1028        if (WARN_ONCE(size < 0,
1029                "%s(%p, %d, %d): lru_size %ld\n",
1030                __func__, lruvec, lru, nr_pages, size)) {
1031                VM_BUG_ON(1);
1032                *lru_size = 0;
1033        }
1034
1035        if (nr_pages > 0)
1036                *lru_size += nr_pages;
1037}
1038
1039bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
1040{
1041        struct mem_cgroup *task_memcg;
1042        struct task_struct *p;
1043        bool ret;
1044
1045        p = find_lock_task_mm(task);
1046        if (p) {
1047                task_memcg = get_mem_cgroup_from_mm(p->mm);
1048                task_unlock(p);
1049        } else {
1050                /*
1051                 * All threads may have already detached their mm's, but the oom
1052                 * killer still needs to detect if they have already been oom
1053                 * killed to prevent needlessly killing additional tasks.
1054                 */
1055                rcu_read_lock();
1056                task_memcg = mem_cgroup_from_task(task);
1057                css_get(&task_memcg->css);
1058                rcu_read_unlock();
1059        }
1060        ret = mem_cgroup_is_descendant(task_memcg, memcg);
1061        css_put(&task_memcg->css);
1062        return ret;
1063}
1064
1065/**
1066 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1067 * @memcg: the memory cgroup
1068 *
1069 * Returns the maximum amount of memory @mem can be charged with, in
1070 * pages.
1071 */
1072static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
1073{
1074        unsigned long margin = 0;
1075        unsigned long count;
1076        unsigned long limit;
1077
1078        count = page_counter_read(&memcg->memory);
1079        limit = READ_ONCE(memcg->memory.limit);
1080        if (count < limit)
1081                margin = limit - count;
1082
1083        if (do_memsw_account()) {
1084                count = page_counter_read(&memcg->memsw);
1085                limit = READ_ONCE(memcg->memsw.limit);
1086                if (count <= limit)
1087                        margin = min(margin, limit - count);
1088                else
1089                        margin = 0;
1090        }
1091
1092        return margin;
1093}
1094
1095/*
1096 * A routine for checking "mem" is under move_account() or not.
1097 *
1098 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1099 * moving cgroups. This is for waiting at high-memory pressure
1100 * caused by "move".
1101 */
1102static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
1103{
1104        struct mem_cgroup *from;
1105        struct mem_cgroup *to;
1106        bool ret = false;
1107        /*
1108         * Unlike task_move routines, we access mc.to, mc.from not under
1109         * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1110         */
1111        spin_lock(&mc.lock);
1112        from = mc.from;
1113        to = mc.to;
1114        if (!from)
1115                goto unlock;
1116
1117        ret = mem_cgroup_is_descendant(from, memcg) ||
1118                mem_cgroup_is_descendant(to, memcg);
1119unlock:
1120        spin_unlock(&mc.lock);
1121        return ret;
1122}
1123
1124static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
1125{
1126        if (mc.moving_task && current != mc.moving_task) {
1127                if (mem_cgroup_under_move(memcg)) {
1128                        DEFINE_WAIT(wait);
1129                        prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1130                        /* moving charge context might have finished. */
1131                        if (mc.moving_task)
1132                                schedule();
1133                        finish_wait(&mc.waitq, &wait);
1134                        return true;
1135                }
1136        }
1137        return false;
1138}
1139
1140#define K(x) ((x) << (PAGE_SHIFT-10))
1141/**
1142 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
1143 * @memcg: The memory cgroup that went over limit
1144 * @p: Task that is going to be killed
1145 *
1146 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1147 * enabled
1148 */
1149void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1150{
1151        struct mem_cgroup *iter;
1152        unsigned int i;
1153
1154        rcu_read_lock();
1155
1156        if (p) {
1157                pr_info("Task in ");
1158                pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1159                pr_cont(" killed as a result of limit of ");
1160        } else {
1161                pr_info("Memory limit reached of cgroup ");
1162        }
1163
1164        pr_cont_cgroup_path(memcg->css.cgroup);
1165        pr_cont("\n");
1166
1167        rcu_read_unlock();
1168
1169        pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1170                K((u64)page_counter_read(&memcg->memory)),
1171                K((u64)memcg->memory.limit), memcg->memory.failcnt);
1172        pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1173                K((u64)page_counter_read(&memcg->memsw)),
1174                K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1175        pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1176                K((u64)page_counter_read(&memcg->kmem)),
1177                K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
1178
1179        for_each_mem_cgroup_tree(iter, memcg) {
1180                pr_info("Memory cgroup stats for ");
1181                pr_cont_cgroup_path(iter->css.cgroup);
1182                pr_cont(":");
1183
1184                for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1185                        if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1186                                continue;
1187                        pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
1188                                K(mem_cgroup_read_stat(iter, i)));
1189                }
1190
1191                for (i = 0; i < NR_LRU_LISTS; i++)
1192                        pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1193                                K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1194
1195                pr_cont("\n");
1196        }
1197}
1198
1199/*
1200 * This function returns the number of memcg under hierarchy tree. Returns
1201 * 1(self count) if no children.
1202 */
1203static int mem_cgroup_count_children(struct mem_cgroup *memcg)
1204{
1205        int num = 0;
1206        struct mem_cgroup *iter;
1207
1208        for_each_mem_cgroup_tree(iter, memcg)
1209                num++;
1210        return num;
1211}
1212
1213/*
1214 * Return the memory (and swap, if configured) limit for a memcg.
1215 */
1216unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
1217{
1218        unsigned long limit;
1219
1220        limit = memcg->memory.limit;
1221        if (mem_cgroup_swappiness(memcg)) {
1222                unsigned long memsw_limit;
1223                unsigned long swap_limit;
1224
1225                memsw_limit = memcg->memsw.limit;
1226                swap_limit = memcg->swap.limit;
1227                swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1228                limit = min(limit + swap_limit, memsw_limit);
1229        }
1230        return limit;
1231}
1232
1233static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1234                                     int order)
1235{
1236        struct oom_control oc = {
1237                .zonelist = NULL,
1238                .nodemask = NULL,
1239                .memcg = memcg,
1240                .gfp_mask = gfp_mask,
1241                .order = order,
1242        };
1243        bool ret;
1244
1245        mutex_lock(&oom_lock);
1246        ret = out_of_memory(&oc);
1247        mutex_unlock(&oom_lock);
1248        return ret;
1249}
1250
1251#if MAX_NUMNODES > 1
1252
1253/**
1254 * test_mem_cgroup_node_reclaimable
1255 * @memcg: the target memcg
1256 * @nid: the node ID to be checked.
1257 * @noswap : specify true here if the user wants flle only information.
1258 *
1259 * This function returns whether the specified memcg contains any
1260 * reclaimable pages on a node. Returns true if there are any reclaimable
1261 * pages in the node.
1262 */
1263static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
1264                int nid, bool noswap)
1265{
1266        if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
1267                return true;
1268        if (noswap || !total_swap_pages)
1269                return false;
1270        if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
1271                return true;
1272        return false;
1273
1274}
1275
1276/*
1277 * Always updating the nodemask is not very good - even if we have an empty
1278 * list or the wrong list here, we can start from some node and traverse all
1279 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1280 *
1281 */
1282static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
1283{
1284        int nid;
1285        /*
1286         * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1287         * pagein/pageout changes since the last update.
1288         */
1289        if (!atomic_read(&memcg->numainfo_events))
1290                return;
1291        if (atomic_inc_return(&memcg->numainfo_updating) > 1)
1292                return;
1293
1294        /* make a nodemask where this memcg uses memory from */
1295        memcg->scan_nodes = node_states[N_MEMORY];
1296
1297        for_each_node_mask(nid, node_states[N_MEMORY]) {
1298
1299                if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1300                        node_clear(nid, memcg->scan_nodes);
1301        }
1302
1303        atomic_set(&memcg->numainfo_events, 0);
1304        atomic_set(&memcg->numainfo_updating, 0);
1305}
1306
1307/*
1308 * Selecting a node where we start reclaim from. Because what we need is just
1309 * reducing usage counter, start from anywhere is O,K. Considering
1310 * memory reclaim from current node, there are pros. and cons.
1311 *
1312 * Freeing memory from current node means freeing memory from a node which
1313 * we'll use or we've used. So, it may make LRU bad. And if several threads
1314 * hit limits, it will see a contention on a node. But freeing from remote
1315 * node means more costs for memory reclaim because of memory latency.
1316 *
1317 * Now, we use round-robin. Better algorithm is welcomed.
1318 */
1319int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1320{
1321        int node;
1322
1323        mem_cgroup_may_update_nodemask(memcg);
1324        node = memcg->last_scanned_node;
1325
1326        node = next_node_in(node, memcg->scan_nodes);
1327        /*
1328         * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1329         * last time it really checked all the LRUs due to rate limiting.
1330         * Fallback to the current node in that case for simplicity.
1331         */
1332        if (unlikely(node == MAX_NUMNODES))
1333                node = numa_node_id();
1334
1335        memcg->last_scanned_node = node;
1336        return node;
1337}
1338#else
1339int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1340{
1341        return 0;
1342}
1343#endif
1344
1345static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1346                                   pg_data_t *pgdat,
1347                                   gfp_t gfp_mask,
1348                                   unsigned long *total_scanned)
1349{
1350        struct mem_cgroup *victim = NULL;
1351        int total = 0;
1352        int loop = 0;
1353        unsigned long excess;
1354        unsigned long nr_scanned;
1355        struct mem_cgroup_reclaim_cookie reclaim = {
1356                .pgdat = pgdat,
1357                .priority = 0,
1358        };
1359
1360        excess = soft_limit_excess(root_memcg);
1361
1362        while (1) {
1363                victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1364                if (!victim) {
1365                        loop++;
1366                        if (loop >= 2) {
1367                                /*
1368                                 * If we have not been able to reclaim
1369                                 * anything, it might because there are
1370                                 * no reclaimable pages under this hierarchy
1371                                 */
1372                                if (!total)
1373                                        break;
1374                                /*
1375                                 * We want to do more targeted reclaim.
1376                                 * excess >> 2 is not to excessive so as to
1377                                 * reclaim too much, nor too less that we keep
1378                                 * coming back to reclaim from this cgroup
1379                                 */
1380                                if (total >= (excess >> 2) ||
1381                                        (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1382                                        break;
1383                        }
1384                        continue;
1385                }
1386                total += mem_cgroup_shrink_node(victim, gfp_mask, false,
1387                                        pgdat, &nr_scanned);
1388                *total_scanned += nr_scanned;
1389                if (!soft_limit_excess(root_memcg))
1390                        break;
1391        }
1392        mem_cgroup_iter_break(root_memcg, victim);
1393        return total;
1394}
1395
1396#ifdef CONFIG_LOCKDEP
1397static struct lockdep_map memcg_oom_lock_dep_map = {
1398        .name = "memcg_oom_lock",
1399};
1400#endif
1401
1402static DEFINE_SPINLOCK(memcg_oom_lock);
1403
1404/*
1405 * Check OOM-Killer is already running under our hierarchy.
1406 * If someone is running, return false.
1407 */
1408static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
1409{
1410        struct mem_cgroup *iter, *failed = NULL;
1411
1412        spin_lock(&memcg_oom_lock);
1413
1414        for_each_mem_cgroup_tree(iter, memcg) {
1415                if (iter->oom_lock) {
1416                        /*
1417                         * this subtree of our hierarchy is already locked
1418                         * so we cannot give a lock.
1419                         */
1420                        failed = iter;
1421                        mem_cgroup_iter_break(memcg, iter);
1422                        break;
1423                } else
1424                        iter->oom_lock = true;
1425        }
1426
1427        if (failed) {
1428                /*
1429                 * OK, we failed to lock the whole subtree so we have
1430                 * to clean up what we set up to the failing subtree
1431                 */
1432                for_each_mem_cgroup_tree(iter, memcg) {
1433                        if (iter == failed) {
1434                                mem_cgroup_iter_break(memcg, iter);
1435                                break;
1436                        }
1437                        iter->oom_lock = false;
1438                }
1439        } else
1440                mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
1441
1442        spin_unlock(&memcg_oom_lock);
1443
1444        return !failed;
1445}
1446
1447static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
1448{
1449        struct mem_cgroup *iter;
1450
1451        spin_lock(&memcg_oom_lock);
1452        mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
1453        for_each_mem_cgroup_tree(iter, memcg)
1454                iter->oom_lock = false;
1455        spin_unlock(&memcg_oom_lock);
1456}
1457
1458static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
1459{
1460        struct mem_cgroup *iter;
1461
1462        spin_lock(&memcg_oom_lock);
1463        for_each_mem_cgroup_tree(iter, memcg)
1464                iter->under_oom++;
1465        spin_unlock(&memcg_oom_lock);
1466}
1467
1468static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
1469{
1470        struct mem_cgroup *iter;
1471
1472        /*
1473         * When a new child is created while the hierarchy is under oom,
1474         * mem_cgroup_oom_lock() may not be called. Watch for underflow.
1475         */
1476        spin_lock(&memcg_oom_lock);
1477        for_each_mem_cgroup_tree(iter, memcg)
1478                if (iter->under_oom > 0)
1479                        iter->under_oom--;
1480        spin_unlock(&memcg_oom_lock);
1481}
1482
1483static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1484
1485struct oom_wait_info {
1486        struct mem_cgroup *memcg;
1487        wait_queue_t    wait;
1488};
1489
1490static int memcg_oom_wake_function(wait_queue_t *wait,
1491        unsigned mode, int sync, void *arg)
1492{
1493        struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1494        struct mem_cgroup *oom_wait_memcg;
1495        struct oom_wait_info *oom_wait_info;
1496
1497        oom_wait_info = container_of(wait, struct oom_wait_info, wait);
1498        oom_wait_memcg = oom_wait_info->memcg;
1499
1500        if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1501            !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
1502                return 0;
1503        return autoremove_wake_function(wait, mode, sync, arg);
1504}
1505
1506static void memcg_oom_recover(struct mem_cgroup *memcg)
1507{
1508        /*
1509         * For the following lockless ->under_oom test, the only required
1510         * guarantee is that it must see the state asserted by an OOM when
1511         * this function is called as a result of userland actions
1512         * triggered by the notification of the OOM.  This is trivially
1513         * achieved by invoking mem_cgroup_mark_under_oom() before
1514         * triggering notification.
1515         */
1516        if (memcg && memcg->under_oom)
1517                __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
1518}
1519
1520static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
1521{
1522        if (!current->memcg_may_oom)
1523                return;
1524        /*
1525         * We are in the middle of the charge context here, so we
1526         * don't want to block when potentially sitting on a callstack
1527         * that holds all kinds of filesystem and mm locks.
1528         *
1529         * Also, the caller may handle a failed allocation gracefully
1530         * (like optional page cache readahead) and so an OOM killer
1531         * invocation might not even be necessary.
1532         *
1533         * That's why we don't do anything here except remember the
1534         * OOM context and then deal with it at the end of the page
1535         * fault when the stack is unwound, the locks are released,
1536         * and when we know whether the fault was overall successful.
1537         */
1538        css_get(&memcg->css);
1539        current->memcg_in_oom = memcg;
1540        current->memcg_oom_gfp_mask = mask;
1541        current->memcg_oom_order = order;
1542}
1543
1544/**
1545 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1546 * @handle: actually kill/wait or just clean up the OOM state
1547 *
1548 * This has to be called at the end of a page fault if the memcg OOM
1549 * handler was enabled.
1550 *
1551 * Memcg supports userspace OOM handling where failed allocations must
1552 * sleep on a waitqueue until the userspace task resolves the
1553 * situation.  Sleeping directly in the charge context with all kinds
1554 * of locks held is not a good idea, instead we remember an OOM state
1555 * in the task and mem_cgroup_oom_synchronize() has to be called at
1556 * the end of the page fault to complete the OOM handling.
1557 *
1558 * Returns %true if an ongoing memcg OOM situation was detected and
1559 * completed, %false otherwise.
1560 */
1561bool mem_cgroup_oom_synchronize(bool handle)
1562{
1563        struct mem_cgroup *memcg = current->memcg_in_oom;
1564        struct oom_wait_info owait;
1565        bool locked;
1566
1567        /* OOM is global, do not handle */
1568        if (!memcg)
1569                return false;
1570
1571        if (!handle)
1572                goto cleanup;
1573
1574        owait.memcg = memcg;
1575        owait.wait.flags = 0;
1576        owait.wait.func = memcg_oom_wake_function;
1577        owait.wait.private = current;
1578        INIT_LIST_HEAD(&owait.wait.task_list);
1579
1580        prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
1581        mem_cgroup_mark_under_oom(memcg);
1582
1583        locked = mem_cgroup_oom_trylock(memcg);
1584
1585        if (locked)
1586                mem_cgroup_oom_notify(memcg);
1587
1588        if (locked && !memcg->oom_kill_disable) {
1589                mem_cgroup_unmark_under_oom(memcg);
1590                finish_wait(&memcg_oom_waitq, &owait.wait);
1591                mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1592                                         current->memcg_oom_order);
1593        } else {
1594                schedule();
1595                mem_cgroup_unmark_under_oom(memcg);
1596                finish_wait(&memcg_oom_waitq, &owait.wait);
1597        }
1598
1599        if (locked) {
1600                mem_cgroup_oom_unlock(memcg);
1601                /*
1602                 * There is no guarantee that an OOM-lock contender
1603                 * sees the wakeups triggered by the OOM kill
1604                 * uncharges.  Wake any sleepers explicitely.
1605                 */
1606                memcg_oom_recover(memcg);
1607        }
1608cleanup:
1609        current->memcg_in_oom = NULL;
1610        css_put(&memcg->css);
1611        return true;
1612}
1613
1614/**
1615 * lock_page_memcg - lock a page->mem_cgroup binding
1616 * @page: the page
1617 *
1618 * This function protects unlocked LRU pages from being moved to
1619 * another cgroup and stabilizes their page->mem_cgroup binding.
1620 */
1621void lock_page_memcg(struct page *page)
1622{
1623        struct mem_cgroup *memcg;
1624        unsigned long flags;
1625
1626        /*
1627         * The RCU lock is held throughout the transaction.  The fast
1628         * path can get away without acquiring the memcg->move_lock
1629         * because page moving starts with an RCU grace period.
1630         */
1631        rcu_read_lock();
1632
1633        if (mem_cgroup_disabled())
1634                return;
1635again:
1636        memcg = page->mem_cgroup;
1637        if (unlikely(!memcg))
1638                return;
1639
1640        if (atomic_read(&memcg->moving_account) <= 0)
1641                return;
1642
1643        spin_lock_irqsave(&memcg->move_lock, flags);
1644        if (memcg != page->mem_cgroup) {
1645                spin_unlock_irqrestore(&memcg->move_lock, flags);
1646                goto again;
1647        }
1648
1649        /*
1650         * When charge migration first begins, we can have locked and
1651         * unlocked page stat updates happening concurrently.  Track
1652         * the task who has the lock for unlock_page_memcg().
1653         */
1654        memcg->move_lock_task = current;
1655        memcg->move_lock_flags = flags;
1656
1657        return;
1658}
1659EXPORT_SYMBOL(lock_page_memcg);
1660
1661/**
1662 * unlock_page_memcg - unlock a page->mem_cgroup binding
1663 * @page: the page
1664 */
1665void unlock_page_memcg(struct page *page)
1666{
1667        struct mem_cgroup *memcg = page->mem_cgroup;
1668
1669        if (memcg && memcg->move_lock_task == current) {
1670                unsigned long flags = memcg->move_lock_flags;
1671
1672                memcg->move_lock_task = NULL;
1673                memcg->move_lock_flags = 0;
1674
1675                spin_unlock_irqrestore(&memcg->move_lock, flags);
1676        }
1677
1678        rcu_read_unlock();
1679}
1680EXPORT_SYMBOL(unlock_page_memcg);
1681
1682/*
1683 * size of first charge trial. "32" comes from vmscan.c's magic value.
1684 * TODO: maybe necessary to use big numbers in big irons.
1685 */
1686#define CHARGE_BATCH    32U
1687struct memcg_stock_pcp {
1688        struct mem_cgroup *cached; /* this never be root cgroup */
1689        unsigned int nr_pages;
1690        struct work_struct work;
1691        unsigned long flags;
1692#define FLUSHING_CACHED_CHARGE  0
1693};
1694static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
1695static DEFINE_MUTEX(percpu_charge_mutex);
1696
1697/**
1698 * consume_stock: Try to consume stocked charge on this cpu.
1699 * @memcg: memcg to consume from.
1700 * @nr_pages: how many pages to charge.
1701 *
1702 * The charges will only happen if @memcg matches the current cpu's memcg
1703 * stock, and at least @nr_pages are available in that stock.  Failure to
1704 * service an allocation will refill the stock.
1705 *
1706 * returns true if successful, false otherwise.
1707 */
1708static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
1709{
1710        struct memcg_stock_pcp *stock;
1711        unsigned long flags;
1712        bool ret = false;
1713
1714        if (nr_pages > CHARGE_BATCH)
1715                return ret;
1716
1717        local_irq_save(flags);
1718
1719        stock = this_cpu_ptr(&memcg_stock);
1720        if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
1721                stock->nr_pages -= nr_pages;
1722                ret = true;
1723        }
1724
1725        local_irq_restore(flags);
1726
1727        return ret;
1728}
1729
1730/*
1731 * Returns stocks cached in percpu and reset cached information.
1732 */
1733static void drain_stock(struct memcg_stock_pcp *stock)
1734{
1735        struct mem_cgroup *old = stock->cached;
1736
1737        if (stock->nr_pages) {
1738                page_counter_uncharge(&old->memory, stock->nr_pages);
1739                if (do_memsw_account())
1740                        page_counter_uncharge(&old->memsw, stock->nr_pages);
1741                css_put_many(&old->css, stock->nr_pages);
1742                stock->nr_pages = 0;
1743        }
1744        stock->cached = NULL;
1745}
1746
1747static void drain_local_stock(struct work_struct *dummy)
1748{
1749        struct memcg_stock_pcp *stock;
1750        unsigned long flags;
1751
1752        local_irq_save(flags);
1753
1754        stock = this_cpu_ptr(&memcg_stock);
1755        drain_stock(stock);
1756        clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
1757
1758        local_irq_restore(flags);
1759}
1760
1761/*
1762 * Cache charges(val) to local per_cpu area.
1763 * This will be consumed by consume_stock() function, later.
1764 */
1765static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
1766{
1767        struct memcg_stock_pcp *stock;
1768        unsigned long flags;
1769
1770        local_irq_save(flags);
1771
1772        stock = this_cpu_ptr(&memcg_stock);
1773        if (stock->cached != memcg) { /* reset if necessary */
1774                drain_stock(stock);
1775                stock->cached = memcg;
1776        }
1777        stock->nr_pages += nr_pages;
1778
1779        local_irq_restore(flags);
1780}
1781
1782/*
1783 * Drains all per-CPU charge caches for given root_memcg resp. subtree
1784 * of the hierarchy under it.
1785 */
1786static void drain_all_stock(struct mem_cgroup *root_memcg)
1787{
1788        int cpu, curcpu;
1789
1790        /* If someone's already draining, avoid adding running more workers. */
1791        if (!mutex_trylock(&percpu_charge_mutex))
1792                return;
1793        /* Notify other cpus that system-wide "drain" is running */
1794        get_online_cpus();
1795        curcpu = get_cpu();
1796        for_each_online_cpu(cpu) {
1797                struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
1798                struct mem_cgroup *memcg;
1799
1800                memcg = stock->cached;
1801                if (!memcg || !stock->nr_pages)
1802                        continue;
1803                if (!mem_cgroup_is_descendant(memcg, root_memcg))
1804                        continue;
1805                if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1806                        if (cpu == curcpu)
1807                                drain_local_stock(&stock->work);
1808                        else
1809                                schedule_work_on(cpu, &stock->work);
1810                }
1811        }
1812        put_cpu();
1813        put_online_cpus();
1814        mutex_unlock(&percpu_charge_mutex);
1815}
1816
1817static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
1818                                        unsigned long action,
1819                                        void *hcpu)
1820{
1821        int cpu = (unsigned long)hcpu;
1822        struct memcg_stock_pcp *stock;
1823
1824        if (action == CPU_ONLINE)
1825                return NOTIFY_OK;
1826
1827        if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
1828                return NOTIFY_OK;
1829
1830        stock = &per_cpu(memcg_stock, cpu);
1831        drain_stock(stock);
1832        return NOTIFY_OK;
1833}
1834
1835static void reclaim_high(struct mem_cgroup *memcg,
1836                         unsigned int nr_pages,
1837                         gfp_t gfp_mask)
1838{
1839        do {
1840                if (page_counter_read(&memcg->memory) <= memcg->high)
1841                        continue;
1842                mem_cgroup_events(memcg, MEMCG_HIGH, 1);
1843                try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1844        } while ((memcg = parent_mem_cgroup(memcg)));
1845}
1846
1847static void high_work_func(struct work_struct *work)
1848{
1849        struct mem_cgroup *memcg;
1850
1851        memcg = container_of(work, struct mem_cgroup, high_work);
1852        reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1853}
1854
1855/*
1856 * Scheduled by try_charge() to be executed from the userland return path
1857 * and reclaims memory over the high limit.
1858 */
1859void mem_cgroup_handle_over_high(void)
1860{
1861        unsigned int nr_pages = current->memcg_nr_pages_over_high;
1862        struct mem_cgroup *memcg;
1863
1864        if (likely(!nr_pages))
1865                return;
1866
1867        memcg = get_mem_cgroup_from_mm(current->mm);
1868        reclaim_high(memcg, nr_pages, GFP_KERNEL);
1869        css_put(&memcg->css);
1870        current->memcg_nr_pages_over_high = 0;
1871}
1872
1873static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1874                      unsigned int nr_pages)
1875{
1876        unsigned int batch = max(CHARGE_BATCH, nr_pages);
1877        int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1878        struct mem_cgroup *mem_over_limit;
1879        struct page_counter *counter;
1880        unsigned long nr_reclaimed;
1881        bool may_swap = true;
1882        bool drained = false;
1883
1884        if (mem_cgroup_is_root(memcg))
1885                return 0;
1886retry:
1887        if (consume_stock(memcg, nr_pages))
1888                return 0;
1889
1890        if (!do_memsw_account() ||
1891            page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1892                if (page_counter_try_charge(&memcg->memory, batch, &counter))
1893                        goto done_restock;
1894                if (do_memsw_account())
1895                        page_counter_uncharge(&memcg->memsw, batch);
1896                mem_over_limit = mem_cgroup_from_counter(counter, memory);
1897        } else {
1898                mem_over_limit = mem_cgroup_from_counter(counter, memsw);
1899                may_swap = false;
1900        }
1901
1902        if (batch > nr_pages) {
1903                batch = nr_pages;
1904                goto retry;
1905        }
1906
1907        /*
1908         * Unlike in global OOM situations, memcg is not in a physical
1909         * memory shortage.  Allow dying and OOM-killed tasks to
1910         * bypass the last charges so that they can exit quickly and
1911         * free their memory.
1912         */
1913        if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1914                     fatal_signal_pending(current) ||
1915                     current->flags & PF_EXITING))
1916                goto force;
1917
1918        /*
1919         * Prevent unbounded recursion when reclaim operations need to
1920         * allocate memory. This might exceed the limits temporarily,
1921         * but we prefer facilitating memory reclaim and getting back
1922         * under the limit over triggering OOM kills in these cases.
1923         */
1924        if (unlikely(current->flags & PF_MEMALLOC))
1925                goto force;
1926
1927        if (unlikely(task_in_memcg_oom(current)))
1928                goto nomem;
1929
1930        if (!gfpflags_allow_blocking(gfp_mask))
1931                goto nomem;
1932
1933        mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
1934
1935        nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1936                                                    gfp_mask, may_swap);
1937
1938        if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
1939                goto retry;
1940
1941        if (!drained) {
1942                drain_all_stock(mem_over_limit);
1943                drained = true;
1944                goto retry;
1945        }
1946
1947        if (gfp_mask & __GFP_NORETRY)
1948                goto nomem;
1949        /*
1950         * Even though the limit is exceeded at this point, reclaim
1951         * may have been able to free some pages.  Retry the charge
1952         * before killing the task.
1953         *
1954         * Only for regular pages, though: huge pages are rather
1955         * unlikely to succeed so close to the limit, and we fall back
1956         * to regular pages anyway in case of failure.
1957         */
1958        if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
1959                goto retry;
1960        /*
1961         * At task move, charge accounts can be doubly counted. So, it's
1962         * better to wait until the end of task_move if something is going on.
1963         */
1964        if (mem_cgroup_wait_acct_move(mem_over_limit))
1965                goto retry;
1966
1967        if (nr_retries--)
1968                goto retry;
1969
1970        if (gfp_mask & __GFP_NOFAIL)
1971                goto force;
1972
1973        if (fatal_signal_pending(current))
1974                goto force;
1975
1976        mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
1977
1978        mem_cgroup_oom(mem_over_limit, gfp_mask,
1979                       get_order(nr_pages * PAGE_SIZE));
1980nomem:
1981        if (!(gfp_mask & __GFP_NOFAIL))
1982                return -ENOMEM;
1983force:
1984        /*
1985         * The allocation either can't fail or will lead to more memory
1986         * being freed very soon.  Allow memory usage go over the limit
1987         * temporarily by force charging it.
1988         */
1989        page_counter_charge(&memcg->memory, nr_pages);
1990        if (do_memsw_account())
1991                page_counter_charge(&memcg->memsw, nr_pages);
1992        css_get_many(&memcg->css, nr_pages);
1993
1994        return 0;
1995
1996done_restock:
1997        css_get_many(&memcg->css, batch);
1998        if (batch > nr_pages)
1999                refill_stock(memcg, batch - nr_pages);
2000
2001        /*
2002         * If the hierarchy is above the normal consumption range, schedule
2003         * reclaim on returning to userland.  We can perform reclaim here
2004         * if __GFP_RECLAIM but let's always punt for simplicity and so that
2005         * GFP_KERNEL can consistently be used during reclaim.  @memcg is
2006         * not recorded as it most likely matches current's and won't
2007         * change in the meantime.  As high limit is checked again before
2008         * reclaim, the cost of mismatch is negligible.
2009         */
2010        do {
2011                if (page_counter_read(&memcg->memory) > memcg->high) {
2012                        /* Don't bother a random interrupted task */
2013                        if (in_interrupt()) {
2014                                schedule_work(&memcg->high_work);
2015                                break;
2016                        }
2017                        current->memcg_nr_pages_over_high += batch;
2018                        set_notify_resume(current);
2019                        break;
2020                }
2021        } while ((memcg = parent_mem_cgroup(memcg)));
2022
2023        return 0;
2024}
2025
2026static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
2027{
2028        if (mem_cgroup_is_root(memcg))
2029                return;
2030
2031        page_counter_uncharge(&memcg->memory, nr_pages);
2032        if (do_memsw_account())
2033                page_counter_uncharge(&memcg->memsw, nr_pages);
2034
2035        css_put_many(&memcg->css, nr_pages);
2036}
2037
2038static void lock_page_lru(struct page *page, int *isolated)
2039{
2040        struct zone *zone = page_zone(page);
2041
2042        spin_lock_irq(zone_lru_lock(zone));
2043        if (PageLRU(page)) {
2044                struct lruvec *lruvec;
2045
2046                lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
2047                ClearPageLRU(page);
2048                del_page_from_lru_list(page, lruvec, page_lru(page));
2049                *isolated = 1;
2050        } else
2051                *isolated = 0;
2052}
2053
2054static void unlock_page_lru(struct page *page, int isolated)
2055{
2056        struct zone *zone = page_zone(page);
2057
2058        if (isolated) {
2059                struct lruvec *lruvec;
2060
2061                lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
2062                VM_BUG_ON_PAGE(PageLRU(page), page);
2063                SetPageLRU(page);
2064                add_page_to_lru_list(page, lruvec, page_lru(page));
2065        }
2066        spin_unlock_irq(zone_lru_lock(zone));
2067}
2068
2069static void commit_charge(struct page *page, struct mem_cgroup *memcg,
2070                          bool lrucare)
2071{
2072        int isolated;
2073
2074        VM_BUG_ON_PAGE(page->mem_cgroup, page);
2075
2076        /*
2077         * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2078         * may already be on some other mem_cgroup's LRU.  Take care of it.
2079         */
2080        if (lrucare)
2081                lock_page_lru(page, &isolated);
2082
2083        /*
2084         * Nobody should be changing or seriously looking at
2085         * page->mem_cgroup at this point:
2086         *
2087         * - the page is uncharged
2088         *
2089         * - the page is off-LRU
2090         *
2091         * - an anonymous fault has exclusive page access, except for
2092         *   a locked page table
2093         *
2094         * - a page cache insertion, a swapin fault, or a migration
2095         *   have the page locked
2096         */
2097        page->mem_cgroup = memcg;
2098
2099        if (lrucare)
2100                unlock_page_lru(page, isolated);
2101}
2102
2103#ifndef CONFIG_SLOB
2104static int memcg_alloc_cache_id(void)
2105{
2106        int id, size;
2107        int err;
2108
2109        id = ida_simple_get(&memcg_cache_ida,
2110                            0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2111        if (id < 0)
2112                return id;
2113
2114        if (id < memcg_nr_cache_ids)
2115                return id;
2116
2117        /*
2118         * There's no space for the new id in memcg_caches arrays,
2119         * so we have to grow them.
2120         */
2121        down_write(&memcg_cache_ids_sem);
2122
2123        size = 2 * (id + 1);
2124        if (size < MEMCG_CACHES_MIN_SIZE)
2125                size = MEMCG_CACHES_MIN_SIZE;
2126        else if (size > MEMCG_CACHES_MAX_SIZE)
2127                size = MEMCG_CACHES_MAX_SIZE;
2128
2129        err = memcg_update_all_caches(size);
2130        if (!err)
2131                err = memcg_update_all_list_lrus(size);
2132        if (!err)
2133                memcg_nr_cache_ids = size;
2134
2135        up_write(&memcg_cache_ids_sem);
2136
2137        if (err) {
2138                ida_simple_remove(&memcg_cache_ida, id);
2139                return err;
2140        }
2141        return id;
2142}
2143
2144static void memcg_free_cache_id(int id)
2145{
2146        ida_simple_remove(&memcg_cache_ida, id);
2147}
2148
2149struct memcg_kmem_cache_create_work {
2150        struct mem_cgroup *memcg;
2151        struct kmem_cache *cachep;
2152        struct work_struct work;
2153};
2154
2155static void memcg_kmem_cache_create_func(struct work_struct *w)
2156{
2157        struct memcg_kmem_cache_create_work *cw =
2158                container_of(w, struct memcg_kmem_cache_create_work, work);
2159        struct mem_cgroup *memcg = cw->memcg;
2160        struct kmem_cache *cachep = cw->cachep;
2161
2162        memcg_create_kmem_cache(memcg, cachep);
2163
2164        css_put(&memcg->css);
2165        kfree(cw);
2166}
2167
2168/*
2169 * Enqueue the creation of a per-memcg kmem_cache.
2170 */
2171static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2172                                               struct kmem_cache *cachep)
2173{
2174        struct memcg_kmem_cache_create_work *cw;
2175
2176        cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
2177        if (!cw)
2178                return;
2179
2180        css_get(&memcg->css);
2181
2182        cw->memcg = memcg;
2183        cw->cachep = cachep;
2184        INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
2185
2186        schedule_work(&cw->work);
2187}
2188
2189static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2190                                             struct kmem_cache *cachep)
2191{
2192        /*
2193         * We need to stop accounting when we kmalloc, because if the
2194         * corresponding kmalloc cache is not yet created, the first allocation
2195         * in __memcg_schedule_kmem_cache_create will recurse.
2196         *
2197         * However, it is better to enclose the whole function. Depending on
2198         * the debugging options enabled, INIT_WORK(), for instance, can
2199         * trigger an allocation. This too, will make us recurse. Because at
2200         * this point we can't allow ourselves back into memcg_kmem_get_cache,
2201         * the safest choice is to do it like this, wrapping the whole function.
2202         */
2203        current->memcg_kmem_skip_account = 1;
2204        __memcg_schedule_kmem_cache_create(memcg, cachep);
2205        current->memcg_kmem_skip_account = 0;
2206}
2207
2208static inline bool memcg_kmem_bypass(void)
2209{
2210        if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
2211                return true;
2212        return false;
2213}
2214
2215/**
2216 * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2217 * @cachep: the original global kmem cache
2218 *
2219 * Return the kmem_cache we're supposed to use for a slab allocation.
2220 * We try to use the current memcg's version of the cache.
2221 *
2222 * If the cache does not exist yet, if we are the first user of it, we
2223 * create it asynchronously in a workqueue and let the current allocation
2224 * go through with the original cache.
2225 *
2226 * This function takes a reference to the cache it returns to assure it
2227 * won't get destroyed while we are working with it. Once the caller is
2228 * done with it, memcg_kmem_put_cache() must be called to release the
2229 * reference.
2230 */
2231struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep)
2232{
2233        struct mem_cgroup *memcg;
2234        struct kmem_cache *memcg_cachep;
2235        int kmemcg_id;
2236
2237        VM_BUG_ON(!is_root_cache(cachep));
2238
2239        if (memcg_kmem_bypass())
2240                return cachep;
2241
2242        if (current->memcg_kmem_skip_account)
2243                return cachep;
2244
2245        memcg = get_mem_cgroup_from_mm(current->mm);
2246        kmemcg_id = READ_ONCE(memcg->kmemcg_id);
2247        if (kmemcg_id < 0)
2248                goto out;
2249
2250        memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
2251        if (likely(memcg_cachep))
2252                return memcg_cachep;
2253
2254        /*
2255         * If we are in a safe context (can wait, and not in interrupt
2256         * context), we could be be predictable and return right away.
2257         * This would guarantee that the allocation being performed
2258         * already belongs in the new cache.
2259         *
2260         * However, there are some clashes that can arrive from locking.
2261         * For instance, because we acquire the slab_mutex while doing
2262         * memcg_create_kmem_cache, this means no further allocation
2263         * could happen with the slab_mutex held. So it's better to
2264         * defer everything.
2265         */
2266        memcg_schedule_kmem_cache_create(memcg, cachep);
2267out:
2268        css_put(&memcg->css);
2269        return cachep;
2270}
2271
2272/**
2273 * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2274 * @cachep: the cache returned by memcg_kmem_get_cache
2275 */
2276void memcg_kmem_put_cache(struct kmem_cache *cachep)
2277{
2278        if (!is_root_cache(cachep))
2279                css_put(&cachep->memcg_params.memcg->css);
2280}
2281
2282/**
2283 * memcg_kmem_charge: charge a kmem page
2284 * @page: page to charge
2285 * @gfp: reclaim mode
2286 * @order: allocation order
2287 * @memcg: memory cgroup to charge
2288 *
2289 * Returns 0 on success, an error code on failure.
2290 */
2291int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2292                            struct mem_cgroup *memcg)
2293{
2294        unsigned int nr_pages = 1 << order;
2295        struct page_counter *counter;
2296        int ret;
2297
2298        ret = try_charge(memcg, gfp, nr_pages);
2299        if (ret)
2300                return ret;
2301
2302        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2303            !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2304                cancel_charge(memcg, nr_pages);
2305                return -ENOMEM;
2306        }
2307
2308        page->mem_cgroup = memcg;
2309
2310        return 0;
2311}
2312
2313/**
2314 * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2315 * @page: page to charge
2316 * @gfp: reclaim mode
2317 * @order: allocation order
2318 *
2319 * Returns 0 on success, an error code on failure.
2320 */
2321int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
2322{
2323        struct mem_cgroup *memcg;
2324        int ret = 0;
2325
2326        if (memcg_kmem_bypass())
2327                return 0;
2328
2329        memcg = get_mem_cgroup_from_mm(current->mm);
2330        if (!mem_cgroup_is_root(memcg)) {
2331                ret = memcg_kmem_charge_memcg(page, gfp, order, memcg);
2332                if (!ret)
2333                        __SetPageKmemcg(page);
2334        }
2335        css_put(&memcg->css);
2336        return ret;
2337}
2338/**
2339 * memcg_kmem_uncharge: uncharge a kmem page
2340 * @page: page to uncharge
2341 * @order: allocation order
2342 */
2343void memcg_kmem_uncharge(struct page *page, int order)
2344{
2345        struct mem_cgroup *memcg = page->mem_cgroup;
2346        unsigned int nr_pages = 1 << order;
2347
2348        if (!memcg)
2349                return;
2350
2351        VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
2352
2353        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2354                page_counter_uncharge(&memcg->kmem, nr_pages);
2355
2356        page_counter_uncharge(&memcg->memory, nr_pages);
2357        if (do_memsw_account())
2358                page_counter_uncharge(&memcg->memsw, nr_pages);
2359
2360        page->mem_cgroup = NULL;
2361
2362        /* slab pages do not have PageKmemcg flag set */
2363        if (PageKmemcg(page))
2364                __ClearPageKmemcg(page);
2365
2366        css_put_many(&memcg->css, nr_pages);
2367}
2368#endif /* !CONFIG_SLOB */
2369
2370#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2371
2372/*
2373 * Because tail pages are not marked as "used", set it. We're under
2374 * zone_lru_lock and migration entries setup in all page mappings.
2375 */
2376void mem_cgroup_split_huge_fixup(struct page *head)
2377{
2378        int i;
2379
2380        if (mem_cgroup_disabled())
2381                return;
2382
2383        for (i = 1; i < HPAGE_PMD_NR; i++)
2384                head[i].mem_cgroup = head->mem_cgroup;
2385
2386        __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
2387                       HPAGE_PMD_NR);
2388}
2389#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2390
2391#ifdef CONFIG_MEMCG_SWAP
2392static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2393                                         bool charge)
2394{
2395        int val = (charge) ? 1 : -1;
2396        this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
2397}
2398
2399/**
2400 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2401 * @entry: swap entry to be moved
2402 * @from:  mem_cgroup which the entry is moved from
2403 * @to:  mem_cgroup which the entry is moved to
2404 *
2405 * It succeeds only when the swap_cgroup's record for this entry is the same
2406 * as the mem_cgroup's id of @from.
2407 *
2408 * Returns 0 on success, -EINVAL on failure.
2409 *
2410 * The caller must have charged to @to, IOW, called page_counter_charge() about
2411 * both res and memsw, and called css_get().
2412 */
2413static int mem_cgroup_move_swap_account(swp_entry_t entry,
2414                                struct mem_cgroup *from, struct mem_cgroup *to)
2415{
2416        unsigned short old_id, new_id;
2417
2418        old_id = mem_cgroup_id(from);
2419        new_id = mem_cgroup_id(to);
2420
2421        if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
2422                mem_cgroup_swap_statistics(from, false);
2423                mem_cgroup_swap_statistics(to, true);
2424                return 0;
2425        }
2426        return -EINVAL;
2427}
2428#else
2429static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
2430                                struct mem_cgroup *from, struct mem_cgroup *to)
2431{
2432        return -EINVAL;
2433}
2434#endif
2435
2436static DEFINE_MUTEX(memcg_limit_mutex);
2437
2438static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
2439                                   unsigned long limit)
2440{
2441        unsigned long curusage;
2442        unsigned long oldusage;
2443        bool enlarge = false;
2444        int retry_count;
2445        int ret;
2446
2447        /*
2448         * For keeping hierarchical_reclaim simple, how long we should retry
2449         * is depends on callers. We set our retry-count to be function
2450         * of # of children which we should visit in this loop.
2451         */
2452        retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2453                      mem_cgroup_count_children(memcg);
2454
2455        oldusage = page_counter_read(&memcg->memory);
2456
2457        do {
2458                if (signal_pending(current)) {
2459                        ret = -EINTR;
2460                        break;
2461                }
2462
2463                mutex_lock(&memcg_limit_mutex);
2464                if (limit > memcg->memsw.limit) {
2465                        mutex_unlock(&memcg_limit_mutex);
2466                        ret = -EINVAL;
2467                        break;
2468                }
2469                if (limit > memcg->memory.limit)
2470                        enlarge = true;
2471                ret = page_counter_limit(&memcg->memory, limit);
2472                mutex_unlock(&memcg_limit_mutex);
2473
2474                if (!ret)
2475                        break;
2476
2477                try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2478
2479                curusage = page_counter_read(&memcg->memory);
2480                /* Usage is reduced ? */
2481                if (curusage >= oldusage)
2482                        retry_count--;
2483                else
2484                        oldusage = curusage;
2485        } while (retry_count);
2486
2487        if (!ret && enlarge)
2488                memcg_oom_recover(memcg);
2489
2490        return ret;
2491}
2492
2493static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2494                                         unsigned long limit)
2495{
2496        unsigned long curusage;
2497        unsigned long oldusage;
2498        bool enlarge = false;
2499        int retry_count;
2500        int ret;
2501
2502        /* see mem_cgroup_resize_res_limit */
2503        retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2504                      mem_cgroup_count_children(memcg);
2505
2506        oldusage = page_counter_read(&memcg->memsw);
2507
2508        do {
2509                if (signal_pending(current)) {
2510                        ret = -EINTR;
2511                        break;
2512                }
2513
2514                mutex_lock(&memcg_limit_mutex);
2515                if (limit < memcg->memory.limit) {
2516                        mutex_unlock(&memcg_limit_mutex);
2517                        ret = -EINVAL;
2518                        break;
2519                }
2520                if (limit > memcg->memsw.limit)
2521                        enlarge = true;
2522                ret = page_counter_limit(&memcg->memsw, limit);
2523                mutex_unlock(&memcg_limit_mutex);
2524
2525                if (!ret)
2526                        break;
2527
2528                try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2529
2530                curusage = page_counter_read(&memcg->memsw);
2531                /* Usage is reduced ? */
2532                if (curusage >= oldusage)
2533                        retry_count--;
2534                else
2535                        oldusage = curusage;
2536        } while (retry_count);
2537
2538        if (!ret && enlarge)
2539                memcg_oom_recover(memcg);
2540
2541        return ret;
2542}
2543
2544unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
2545                                            gfp_t gfp_mask,
2546                                            unsigned long *total_scanned)
2547{
2548        unsigned long nr_reclaimed = 0;
2549        struct mem_cgroup_per_node *mz, *next_mz = NULL;
2550        unsigned long reclaimed;
2551        int loop = 0;
2552        struct mem_cgroup_tree_per_node *mctz;
2553        unsigned long excess;
2554        unsigned long nr_scanned;
2555
2556        if (order > 0)
2557                return 0;
2558
2559        mctz = soft_limit_tree_node(pgdat->node_id);
2560
2561        /*
2562         * Do not even bother to check the largest node if the root
2563         * is empty. Do it lockless to prevent lock bouncing. Races
2564         * are acceptable as soft limit is best effort anyway.
2565         */
2566        if (RB_EMPTY_ROOT(&mctz->rb_root))
2567                return 0;
2568
2569        /*
2570         * This loop can run a while, specially if mem_cgroup's continuously
2571         * keep exceeding their soft limit and putting the system under
2572         * pressure
2573         */
2574        do {
2575                if (next_mz)
2576                        mz = next_mz;
2577                else
2578                        mz = mem_cgroup_largest_soft_limit_node(mctz);
2579                if (!mz)
2580                        break;
2581
2582                nr_scanned = 0;
2583                reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
2584                                                    gfp_mask, &nr_scanned);
2585                nr_reclaimed += reclaimed;
2586                *total_scanned += nr_scanned;
2587                spin_lock_irq(&mctz->lock);
2588                __mem_cgroup_remove_exceeded(mz, mctz);
2589
2590                /*
2591                 * If we failed to reclaim anything from this memory cgroup
2592                 * it is time to move on to the next cgroup
2593                 */
2594                next_mz = NULL;
2595                if (!reclaimed)
2596                        next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2597
2598                excess = soft_limit_excess(mz->memcg);
2599                /*
2600                 * One school of thought says that we should not add
2601                 * back the node to the tree if reclaim returns 0.
2602                 * But our reclaim could return 0, simply because due
2603                 * to priority we are exposing a smaller subset of
2604                 * memory to reclaim from. Consider this as a longer
2605                 * term TODO.
2606                 */
2607                /* If excess == 0, no tree ops */
2608                __mem_cgroup_insert_exceeded(mz, mctz, excess);
2609                spin_unlock_irq(&mctz->lock);
2610                css_put(&mz->memcg->css);
2611                loop++;
2612                /*
2613                 * Could not reclaim anything and there are no more
2614                 * mem cgroups to try or we seem to be looping without
2615                 * reclaiming anything.
2616                 */
2617                if (!nr_reclaimed &&
2618                        (next_mz == NULL ||
2619                        loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2620                        break;
2621        } while (!nr_reclaimed);
2622        if (next_mz)
2623                css_put(&next_mz->memcg->css);
2624        return nr_reclaimed;
2625}
2626
2627/*
2628 * Test whether @memcg has children, dead or alive.  Note that this
2629 * function doesn't care whether @memcg has use_hierarchy enabled and
2630 * returns %true if there are child csses according to the cgroup
2631 * hierarchy.  Testing use_hierarchy is the caller's responsiblity.
2632 */
2633static inline bool memcg_has_children(struct mem_cgroup *memcg)
2634{
2635        bool ret;
2636
2637        rcu_read_lock();
2638        ret = css_next_child(NULL, &memcg->css);
2639        rcu_read_unlock();
2640        return ret;
2641}
2642
2643/*
2644 * Reclaims as many pages from the given memcg as possible.
2645 *
2646 * Caller is responsible for holding css reference for memcg.
2647 */
2648static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2649{
2650        int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
2651
2652        /* we call try-to-free pages for make this cgroup empty */
2653        lru_add_drain_all();
2654        /* try to free all pages in this cgroup */
2655        while (nr_retries && page_counter_read(&memcg->memory)) {
2656                int progress;
2657
2658                if (signal_pending(current))
2659                        return -EINTR;
2660
2661                progress = try_to_free_mem_cgroup_pages(memcg, 1,
2662                                                        GFP_KERNEL, true);
2663                if (!progress) {
2664                        nr_retries--;
2665                        /* maybe some writeback is necessary */
2666                        congestion_wait(BLK_RW_ASYNC, HZ/10);
2667                }
2668
2669        }
2670
2671        return 0;
2672}
2673
2674static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2675                                            char *buf, size_t nbytes,
2676                                            loff_t off)
2677{
2678        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
2679
2680        if (mem_cgroup_is_root(memcg))
2681                return -EINVAL;
2682        return mem_cgroup_force_empty(memcg) ?: nbytes;
2683}
2684
2685static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2686                                     struct cftype *cft)
2687{
2688        return mem_cgroup_from_css(css)->use_hierarchy;
2689}
2690
2691static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2692                                      struct cftype *cft, u64 val)
2693{
2694        int retval = 0;
2695        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
2696        struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
2697
2698        if (memcg->use_hierarchy == val)
2699                return 0;
2700
2701        /*
2702         * If parent's use_hierarchy is set, we can't make any modifications
2703         * in the child subtrees. If it is unset, then the change can
2704         * occur, provided the current cgroup has no children.
2705         *
2706         * For the root cgroup, parent_mem is NULL, we allow value to be
2707         * set if there are no children.
2708         */
2709        if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
2710                                (val == 1 || val == 0)) {
2711                if (!memcg_has_children(memcg))
2712                        memcg->use_hierarchy = val;
2713                else
2714                        retval = -EBUSY;
2715        } else
2716                retval = -EINVAL;
2717
2718        return retval;
2719}
2720
2721static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
2722{
2723        struct mem_cgroup *iter;
2724        int i;
2725
2726        memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
2727
2728        for_each_mem_cgroup_tree(iter, memcg) {
2729                for (i = 0; i < MEMCG_NR_STAT; i++)
2730                        stat[i] += mem_cgroup_read_stat(iter, i);
2731        }
2732}
2733
2734static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
2735{
2736        struct mem_cgroup *iter;
2737        int i;
2738
2739        memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
2740
2741        for_each_mem_cgroup_tree(iter, memcg) {
2742                for (i = 0; i < MEMCG_NR_EVENTS; i++)
2743                        events[i] += mem_cgroup_read_events(iter, i);
2744        }
2745}
2746
2747static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
2748{
2749        unsigned long val = 0;
2750
2751        if (mem_cgroup_is_root(memcg)) {
2752                struct mem_cgroup *iter;
2753
2754                for_each_mem_cgroup_tree(iter, memcg) {
2755                        val += mem_cgroup_read_stat(iter,
2756                                        MEM_CGROUP_STAT_CACHE);
2757                        val += mem_cgroup_read_stat(iter,
2758                                        MEM_CGROUP_STAT_RSS);
2759                        if (swap)
2760                                val += mem_cgroup_read_stat(iter,
2761                                                MEM_CGROUP_STAT_SWAP);
2762                }
2763        } else {
2764                if (!swap)
2765                        val = page_counter_read(&memcg->memory);
2766                else
2767                        val = page_counter_read(&memcg->memsw);
2768        }
2769        return val;
2770}
2771
2772enum {
2773        RES_USAGE,
2774        RES_LIMIT,
2775        RES_MAX_USAGE,
2776        RES_FAILCNT,
2777        RES_SOFT_LIMIT,
2778};
2779
2780static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
2781                               struct cftype *cft)
2782{
2783        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
2784        struct page_counter *counter;
2785
2786        switch (MEMFILE_TYPE(cft->private)) {
2787        case _MEM:
2788                counter = &memcg->memory;
2789                break;
2790        case _MEMSWAP:
2791                counter = &memcg->memsw;
2792                break;
2793        case _KMEM:
2794                counter = &memcg->kmem;
2795                break;
2796        case _TCP:
2797                counter = &memcg->tcpmem;
2798                break;
2799        default:
2800                BUG();
2801        }
2802
2803        switch (MEMFILE_ATTR(cft->private)) {
2804        case RES_USAGE:
2805                if (counter == &memcg->memory)
2806                        return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
2807                if (counter == &memcg->memsw)
2808                        return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
2809                return (u64)page_counter_read(counter) * PAGE_SIZE;
2810        case RES_LIMIT:
2811                return (u64)counter->limit * PAGE_SIZE;
2812        case RES_MAX_USAGE:
2813                return (u64)counter->watermark * PAGE_SIZE;
2814        case RES_FAILCNT:
2815                return counter->failcnt;
2816        case RES_SOFT_LIMIT:
2817                return (u64)memcg->soft_limit * PAGE_SIZE;
2818        default:
2819                BUG();
2820        }
2821}
2822
2823#ifndef CONFIG_SLOB
2824static int memcg_online_kmem(struct mem_cgroup *memcg)
2825{
2826        int memcg_id;
2827
2828        if (cgroup_memory_nokmem)
2829                return 0;
2830
2831        BUG_ON(memcg->kmemcg_id >= 0);
2832        BUG_ON(memcg->kmem_state);
2833
2834        memcg_id = memcg_alloc_cache_id();
2835        if (memcg_id < 0)
2836                return memcg_id;
2837
2838        static_branch_inc(&memcg_kmem_enabled_key);
2839        /*
2840         * A memory cgroup is considered kmem-online as soon as it gets
2841         * kmemcg_id. Setting the id after enabling static branching will
2842         * guarantee no one starts accounting before all call sites are
2843         * patched.
2844         */
2845        memcg->kmemcg_id = memcg_id;
2846        memcg->kmem_state = KMEM_ONLINE;
2847
2848        return 0;
2849}
2850
2851static void memcg_offline_kmem(struct mem_cgroup *memcg)
2852{
2853        struct cgroup_subsys_state *css;
2854        struct mem_cgroup *parent, *child;
2855        int kmemcg_id;
2856
2857        if (memcg->kmem_state != KMEM_ONLINE)
2858                return;
2859        /*
2860         * Clear the online state before clearing memcg_caches array
2861         * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2862         * guarantees that no cache will be created for this cgroup
2863         * after we are done (see memcg_create_kmem_cache()).
2864         */
2865        memcg->kmem_state = KMEM_ALLOCATED;
2866
2867        memcg_deactivate_kmem_caches(memcg);
2868
2869        kmemcg_id = memcg->kmemcg_id;
2870        BUG_ON(kmemcg_id < 0);
2871
2872        parent = parent_mem_cgroup(memcg);
2873        if (!parent)
2874                parent = root_mem_cgroup;
2875
2876        /*
2877         * Change kmemcg_id of this cgroup and all its descendants to the
2878         * parent's id, and then move all entries from this cgroup's list_lrus
2879         * to ones of the parent. After we have finished, all list_lrus
2880         * corresponding to this cgroup are guaranteed to remain empty. The
2881         * ordering is imposed by list_lru_node->lock taken by
2882         * memcg_drain_all_list_lrus().
2883         */
2884        rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
2885        css_for_each_descendant_pre(css, &memcg->css) {
2886                child = mem_cgroup_from_css(css);
2887                BUG_ON(child->kmemcg_id != kmemcg_id);
2888                child->kmemcg_id = parent->kmemcg_id;
2889                if (!memcg->use_hierarchy)
2890                        break;
2891        }
2892        rcu_read_unlock();
2893
2894        memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2895
2896        memcg_free_cache_id(kmemcg_id);
2897}
2898
2899static void memcg_free_kmem(struct mem_cgroup *memcg)
2900{
2901        /* css_alloc() failed, offlining didn't happen */
2902        if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2903                memcg_offline_kmem(memcg);
2904
2905        if (memcg->kmem_state == KMEM_ALLOCATED) {
2906                memcg_destroy_kmem_caches(memcg);
2907                static_branch_dec(&memcg_kmem_enabled_key);
2908                WARN_ON(page_counter_read(&memcg->kmem));
2909        }
2910}
2911#else
2912static int memcg_online_kmem(struct mem_cgroup *memcg)
2913{
2914        return 0;
2915}
2916static void memcg_offline_kmem(struct mem_cgroup *memcg)
2917{
2918}
2919static void memcg_free_kmem(struct mem_cgroup *memcg)
2920{
2921}
2922#endif /* !CONFIG_SLOB */
2923
2924static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2925                                   unsigned long limit)
2926{
2927        int ret;
2928
2929        mutex_lock(&memcg_limit_mutex);
2930        ret = page_counter_limit(&memcg->kmem, limit);
2931        mutex_unlock(&memcg_limit_mutex);
2932        return ret;
2933}
2934
2935static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2936{
2937        int ret;
2938
2939        mutex_lock(&memcg_limit_mutex);
2940
2941        ret = page_counter_limit(&memcg->tcpmem, limit);
2942        if (ret)
2943                goto out;
2944
2945        if (!memcg->tcpmem_active) {
2946                /*
2947                 * The active flag needs to be written after the static_key
2948                 * update. This is what guarantees that the socket activation
2949                 * function is the last one to run. See mem_cgroup_sk_alloc()
2950                 * for details, and note that we don't mark any socket as
2951                 * belonging to this memcg until that flag is up.
2952                 *
2953                 * We need to do this, because static_keys will span multiple
2954                 * sites, but we can't control their order. If we mark a socket
2955                 * as accounted, but the accounting functions are not patched in
2956                 * yet, we'll lose accounting.
2957                 *
2958                 * We never race with the readers in mem_cgroup_sk_alloc(),
2959                 * because when this value change, the code to process it is not
2960                 * patched in yet.
2961                 */
2962                static_branch_inc(&memcg_sockets_enabled_key);
2963                memcg->tcpmem_active = true;
2964        }
2965out:
2966        mutex_unlock(&memcg_limit_mutex);
2967        return ret;
2968}
2969
2970/*
2971 * The user of this function is...
2972 * RES_LIMIT.
2973 */
2974static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2975                                char *buf, size_t nbytes, loff_t off)
2976{
2977        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
2978        unsigned long nr_pages;
2979        int ret;
2980
2981        buf = strstrip(buf);
2982        ret = page_counter_memparse(buf, "-1", &nr_pages);
2983        if (ret)
2984                return ret;
2985
2986        switch (MEMFILE_ATTR(of_cft(of)->private)) {
2987        case RES_LIMIT:
2988                if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2989                        ret = -EINVAL;
2990                        break;
2991                }
2992                switch (MEMFILE_TYPE(of_cft(of)->private)) {
2993                case _MEM:
2994                        ret = mem_cgroup_resize_limit(memcg, nr_pages);
2995                        break;
2996                case _MEMSWAP:
2997                        ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
2998                        break;
2999                case _KMEM:
3000                        ret = memcg_update_kmem_limit(memcg, nr_pages);
3001                        break;
3002                case _TCP:
3003                        ret = memcg_update_tcp_limit(memcg, nr_pages);
3004                        break;
3005                }
3006                break;
3007        case RES_SOFT_LIMIT:
3008                memcg->soft_limit = nr_pages;
3009                ret = 0;
3010                break;
3011        }
3012        return ret ?: nbytes;
3013}
3014
3015static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3016                                size_t nbytes, loff_t off)
3017{
3018        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
3019        struct page_counter *counter;
3020
3021        switch (MEMFILE_TYPE(of_cft(of)->private)) {
3022        case _MEM:
3023                counter = &memcg->memory;
3024                break;
3025        case _MEMSWAP:
3026                counter = &memcg->memsw;
3027                break;
3028        case _KMEM:
3029                counter = &memcg->kmem;
3030                break;
3031        case _TCP:
3032                counter = &memcg->tcpmem;
3033                break;
3034        default:
3035                BUG();
3036        }
3037
3038        switch (MEMFILE_ATTR(of_cft(of)->private)) {
3039        case RES_MAX_USAGE:
3040                page_counter_reset_watermark(counter);
3041                break;
3042        case RES_FAILCNT:
3043                counter->failcnt = 0;
3044                break;
3045        default:
3046                BUG();
3047        }
3048
3049        return nbytes;
3050}
3051
3052static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
3053                                        struct cftype *cft)
3054{
3055        return mem_cgroup_from_css(css)->move_charge_at_immigrate;
3056}
3057
3058#ifdef CONFIG_MMU
3059static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
3060                                        struct cftype *cft, u64 val)
3061{
3062        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3063
3064        if (val & ~MOVE_MASK)
3065                return -EINVAL;
3066
3067        /*
3068         * No kind of locking is needed in here, because ->can_attach() will
3069         * check this value once in the beginning of the process, and then carry
3070         * on with stale data. This means that changes to this value will only
3071         * affect task migrations starting after the change.
3072         */
3073        memcg->move_charge_at_immigrate = val;
3074        return 0;
3075}
3076#else
3077static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
3078                                        struct cftype *cft, u64 val)
3079{
3080        return -ENOSYS;
3081}
3082#endif
3083
3084#ifdef CONFIG_NUMA
3085static int memcg_numa_stat_show(struct seq_file *m, void *v)
3086{
3087        struct numa_stat {
3088                const char *name;
3089                unsigned int lru_mask;
3090        };
3091
3092        static const struct numa_stat stats[] = {
3093                { "total", LRU_ALL },
3094                { "file", LRU_ALL_FILE },
3095                { "anon", LRU_ALL_ANON },
3096                { "unevictable", BIT(LRU_UNEVICTABLE) },
3097        };
3098        const struct numa_stat *stat;
3099        int nid;
3100        unsigned long nr;
3101        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
3102
3103        for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3104                nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3105                seq_printf(m, "%s=%lu", stat->name, nr);
3106                for_each_node_state(nid, N_MEMORY) {
3107                        nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3108                                                          stat->lru_mask);
3109                        seq_printf(m, " N%d=%lu", nid, nr);
3110                }
3111                seq_putc(m, '\n');
3112        }
3113
3114        for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3115                struct mem_cgroup *iter;
3116
3117                nr = 0;
3118                for_each_mem_cgroup_tree(iter, memcg)
3119                        nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3120                seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3121                for_each_node_state(nid, N_MEMORY) {
3122                        nr = 0;
3123                        for_each_mem_cgroup_tree(iter, memcg)
3124                                nr += mem_cgroup_node_nr_lru_pages(
3125                                        iter, nid, stat->lru_mask);
3126                        seq_printf(m, " N%d=%lu", nid, nr);
3127                }
3128                seq_putc(m, '\n');
3129        }
3130
3131        return 0;
3132}
3133#endif /* CONFIG_NUMA */
3134
3135static int memcg_stat_show(struct seq_file *m, void *v)
3136{
3137        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
3138        unsigned long memory, memsw;
3139        struct mem_cgroup *mi;
3140        unsigned int i;
3141
3142        BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3143                     MEM_CGROUP_STAT_NSTATS);
3144        BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3145                     MEM_CGROUP_EVENTS_NSTATS);
3146        BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3147
3148        for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
3149                if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
3150                        continue;
3151                seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
3152                           mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
3153        }
3154
3155        for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3156                seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3157                           mem_cgroup_read_events(memcg, i));
3158
3159        for (i = 0; i < NR_LRU_LISTS; i++)
3160                seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3161                           mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3162
3163        /* Hierarchical information */
3164        memory = memsw = PAGE_COUNTER_MAX;
3165        for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3166                memory = min(memory, mi->memory.limit);
3167                memsw = min(memsw, mi->memsw.limit);
3168        }
3169        seq_printf(m, "hierarchical_memory_limit %llu\n",
3170                   (u64)memory * PAGE_SIZE);
3171        if (do_memsw_account())
3172                seq_printf(m, "hierarchical_memsw_limit %llu\n",
3173                           (u64)memsw * PAGE_SIZE);
3174
3175        for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
3176                unsigned long long val = 0;
3177
3178                if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
3179                        continue;
3180                for_each_mem_cgroup_tree(mi, memcg)
3181                        val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
3182                seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
3183        }
3184
3185        for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3186                unsigned long long val = 0;
3187
3188                for_each_mem_cgroup_tree(mi, memcg)
3189                        val += mem_cgroup_read_events(mi, i);
3190                seq_printf(m, "total_%s %llu\n",
3191                           mem_cgroup_events_names[i], val);
3192        }
3193
3194        for (i = 0; i < NR_LRU_LISTS; i++) {
3195                unsigned long long val = 0;
3196
3197                for_each_mem_cgroup_tree(mi, memcg)
3198                        val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3199                seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
3200        }
3201
3202#ifdef CONFIG_DEBUG_VM
3203        {
3204                pg_data_t *pgdat;
3205                struct mem_cgroup_per_node *mz;
3206                struct zone_reclaim_stat *rstat;
3207                unsigned long recent_rotated[2] = {0, 0};
3208                unsigned long recent_scanned[2] = {0, 0};
3209
3210                for_each_online_pgdat(pgdat) {
3211                        mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
3212                        rstat = &mz->lruvec.reclaim_stat;
3213
3214                        recent_rotated[0] += rstat->recent_rotated[0];
3215                        recent_rotated[1] += rstat->recent_rotated[1];
3216                        recent_scanned[0] += rstat->recent_scanned[0];
3217                        recent_scanned[1] += rstat->recent_scanned[1];
3218                }
3219                seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3220                seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3221                seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3222                seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
3223        }
3224#endif
3225
3226        return 0;
3227}
3228
3229static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3230                                      struct cftype *cft)
3231{
3232        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3233
3234        return mem_cgroup_swappiness(memcg);
3235}
3236
3237static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3238                                       struct cftype *cft, u64 val)
3239{
3240        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3241
3242        if (val > 100)
3243                return -EINVAL;
3244
3245        if (css->parent)
3246                memcg->swappiness = val;
3247        else
3248                vm_swappiness = val;
3249
3250        return 0;
3251}
3252
3253static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3254{
3255        struct mem_cgroup_threshold_ary *t;
3256        unsigned long usage;
3257        int i;
3258
3259        rcu_read_lock();
3260        if (!swap)
3261                t = rcu_dereference(memcg->thresholds.primary);
3262        else
3263                t = rcu_dereference(memcg->memsw_thresholds.primary);
3264
3265        if (!t)
3266                goto unlock;
3267
3268        usage = mem_cgroup_usage(memcg, swap);
3269
3270        /*
3271         * current_threshold points to threshold just below or equal to usage.
3272         * If it's not true, a threshold was crossed after last
3273         * call of __mem_cgroup_threshold().
3274         */
3275        i = t->current_threshold;
3276
3277        /*
3278         * Iterate backward over array of thresholds starting from
3279         * current_threshold and check if a threshold is crossed.
3280         * If none of thresholds below usage is crossed, we read
3281         * only one element of the array here.
3282         */
3283        for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3284                eventfd_signal(t->entries[i].eventfd, 1);
3285
3286        /* i = current_threshold + 1 */
3287        i++;
3288
3289        /*
3290         * Iterate forward over array of thresholds starting from
3291         * current_threshold+1 and check if a threshold is crossed.
3292         * If none of thresholds above usage is crossed, we read
3293         * only one element of the array here.
3294         */
3295        for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3296                eventfd_signal(t->entries[i].eventfd, 1);
3297
3298        /* Update current_threshold */
3299        t->current_threshold = i - 1;
3300unlock:
3301        rcu_read_unlock();
3302}
3303
3304static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3305{
3306        while (memcg) {
3307                __mem_cgroup_threshold(memcg, false);
3308                if (do_memsw_account())
3309                        __mem_cgroup_threshold(memcg, true);
3310
3311                memcg = parent_mem_cgroup(memcg);
3312        }
3313}
3314
3315static int compare_thresholds(const void *a, const void *b)
3316{
3317        const struct mem_cgroup_threshold *_a = a;
3318        const struct mem_cgroup_threshold *_b = b;
3319
3320        if (_a->threshold > _b->threshold)
3321                return 1;
3322
3323        if (_a->threshold < _b->threshold)
3324                return -1;
3325
3326        return 0;
3327}
3328
3329static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
3330{
3331        struct mem_cgroup_eventfd_list *ev;
3332
3333        spin_lock(&memcg_oom_lock);
3334
3335        list_for_each_entry(ev, &memcg->oom_notify, list)
3336                eventfd_signal(ev->eventfd, 1);
3337
3338        spin_unlock(&memcg_oom_lock);
3339        return 0;
3340}
3341
3342static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
3343{
3344        struct mem_cgroup *iter;
3345
3346        for_each_mem_cgroup_tree(iter, memcg)
3347                mem_cgroup_oom_notify_cb(iter);
3348}
3349
3350static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
3351        struct eventfd_ctx *eventfd, const char *args, enum res_type type)
3352{
3353        struct mem_cgroup_thresholds *thresholds;
3354        struct mem_cgroup_threshold_ary *new;
3355        unsigned long threshold;
3356        unsigned long usage;
3357        int i, size, ret;
3358
3359        ret = page_counter_memparse(args, "-1", &threshold);
3360        if (ret)
3361                return ret;
3362
3363        mutex_lock(&memcg->thresholds_lock);
3364
3365        if (type == _MEM) {
3366                thresholds = &memcg->thresholds;
3367                usage = mem_cgroup_usage(memcg, false);
3368        } else if (type == _MEMSWAP) {
3369                thresholds = &memcg->memsw_thresholds;
3370                usage = mem_cgroup_usage(memcg, true);
3371        } else
3372                BUG();
3373
3374        /* Check if a threshold crossed before adding a new one */
3375        if (thresholds->primary)
3376                __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3377
3378        size = thresholds->primary ? thresholds->primary->size + 1 : 1;
3379
3380        /* Allocate memory for new array of thresholds */
3381        new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
3382                        GFP_KERNEL);
3383        if (!new) {
3384                ret = -ENOMEM;
3385                goto unlock;
3386        }
3387        new->size = size;
3388
3389        /* Copy thresholds (if any) to new array */
3390        if (thresholds->primary) {
3391                memcpy(new->entries, thresholds->primary->entries, (size - 1) *
3392                                sizeof(struct mem_cgroup_threshold));
3393        }
3394
3395        /* Add new threshold */
3396        new->entries[size - 1].eventfd = eventfd;
3397        new->entries[size - 1].threshold = threshold;
3398
3399        /* Sort thresholds. Registering of new threshold isn't time-critical */
3400        sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
3401                        compare_thresholds, NULL);
3402
3403        /* Find current threshold */
3404        new->current_threshold = -1;
3405        for (i = 0; i < size; i++) {
3406                if (new->entries[i].threshold <= usage) {
3407                        /*
3408                         * new->current_threshold will not be used until
3409                         * rcu_assign_pointer(), so it's safe to increment
3410                         * it here.
3411                         */
3412                        ++new->current_threshold;
3413                } else
3414                        break;
3415        }
3416
3417        /* Free old spare buffer and save old primary buffer as spare */
3418        kfree(thresholds->spare);
3419        thresholds->spare = thresholds->primary;
3420
3421        rcu_assign_pointer(thresholds->primary, new);
3422
3423        /* To be sure that nobody uses thresholds */
3424        synchronize_rcu();
3425
3426unlock:
3427        mutex_unlock(&memcg->thresholds_lock);
3428
3429        return ret;
3430}
3431
3432static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
3433        struct eventfd_ctx *eventfd, const char *args)
3434{
3435        return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
3436}
3437
3438static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
3439        struct eventfd_ctx *eventfd, const char *args)
3440{
3441        return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
3442}
3443
3444static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3445        struct eventfd_ctx *eventfd, enum res_type type)
3446{
3447        struct mem_cgroup_thresholds *thresholds;
3448        struct mem_cgroup_threshold_ary *new;
3449        unsigned long usage;
3450        int i, j, size;
3451
3452        mutex_lock(&memcg->thresholds_lock);
3453
3454        if (type == _MEM) {
3455                thresholds = &memcg->thresholds;
3456                usage = mem_cgroup_usage(memcg, false);
3457        } else if (type == _MEMSWAP) {
3458                thresholds = &memcg->memsw_thresholds;
3459                usage = mem_cgroup_usage(memcg, true);
3460        } else
3461                BUG();
3462
3463        if (!thresholds->primary)
3464                goto unlock;
3465
3466        /* Check if a threshold crossed before removing */
3467        __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3468
3469        /* Calculate new number of threshold */
3470        size = 0;
3471        for (i = 0; i < thresholds->primary->size; i++) {
3472                if (thresholds->primary->entries[i].eventfd != eventfd)
3473                        size++;
3474        }
3475
3476        new = thresholds->spare;
3477
3478        /* Set thresholds array to NULL if we don't have thresholds */
3479        if (!size) {
3480                kfree(new);
3481                new = NULL;
3482                goto swap_buffers;
3483        }
3484
3485        new->size = size;
3486
3487        /* Copy thresholds and find current threshold */
3488        new->current_threshold = -1;
3489        for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3490                if (thresholds->primary->entries[i].eventfd == eventfd)
3491                        continue;
3492
3493                new->entries[j] = thresholds->primary->entries[i];
3494                if (new->entries[j].threshold <= usage) {
3495                        /*
3496                         * new->current_threshold will not be used
3497                         * until rcu_assign_pointer(), so it's safe to increment
3498                         * it here.
3499                         */
3500                        ++new->current_threshold;
3501                }
3502                j++;
3503        }
3504
3505swap_buffers:
3506        /* Swap primary and spare array */
3507        thresholds->spare = thresholds->primary;
3508
3509        rcu_assign_pointer(thresholds->primary, new);
3510
3511        /* To be sure that nobody uses thresholds */
3512        synchronize_rcu();
3513
3514        /* If all events are unregistered, free the spare array */
3515        if (!new) {
3516                kfree(thresholds->spare);
3517                thresholds->spare = NULL;
3518        }
3519unlock:
3520        mutex_unlock(&memcg->thresholds_lock);
3521}
3522
3523static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3524        struct eventfd_ctx *eventfd)
3525{
3526        return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
3527}
3528
3529static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3530        struct eventfd_ctx *eventfd)
3531{
3532        return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
3533}
3534
3535static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
3536        struct eventfd_ctx *eventfd, const char *args)
3537{
3538        struct mem_cgroup_eventfd_list *event;
3539
3540        event = kmalloc(sizeof(*event), GFP_KERNEL);
3541        if (!event)
3542                return -ENOMEM;
3543
3544        spin_lock(&memcg_oom_lock);
3545
3546        event->eventfd = eventfd;
3547        list_add(&event->list, &memcg->oom_notify);
3548
3549        /* already in OOM ? */
3550        if (memcg->under_oom)
3551                eventfd_signal(eventfd, 1);
3552        spin_unlock(&memcg_oom_lock);
3553
3554        return 0;
3555}
3556
3557static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
3558        struct eventfd_ctx *eventfd)
3559{
3560        struct mem_cgroup_eventfd_list *ev, *tmp;
3561
3562        spin_lock(&memcg_oom_lock);
3563
3564        list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
3565                if (ev->eventfd == eventfd) {
3566                        list_del(&ev->list);
3567                        kfree(ev);
3568                }
3569        }
3570
3571        spin_unlock(&memcg_oom_lock);
3572}
3573
3574static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
3575{
3576        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
3577
3578        seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
3579        seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
3580        return 0;
3581}
3582
3583static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
3584        struct cftype *cft, u64 val)
3585{
3586        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3587
3588        /* cannot set to root cgroup and only 0 and 1 are allowed */
3589        if (!css->parent || !((val == 0) || (val == 1)))
3590                return -EINVAL;
3591
3592        memcg->oom_kill_disable = val;
3593        if (!val)
3594                memcg_oom_recover(memcg);
3595
3596        return 0;
3597}
3598
3599#ifdef CONFIG_CGROUP_WRITEBACK
3600
3601struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3602{
3603        return &memcg->cgwb_list;
3604}
3605
3606static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3607{
3608        return wb_domain_init(&memcg->cgwb_domain, gfp);
3609}
3610
3611static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3612{
3613        wb_domain_exit(&memcg->cgwb_domain);
3614}
3615
3616static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3617{
3618        wb_domain_size_changed(&memcg->cgwb_domain);
3619}
3620
3621struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3622{
3623        struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3624
3625        if (!memcg->css.parent)
3626                return NULL;
3627
3628        return &memcg->cgwb_domain;
3629}
3630
3631/**
3632 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3633 * @wb: bdi_writeback in question
3634 * @pfilepages: out parameter for number of file pages
3635 * @pheadroom: out parameter for number of allocatable pages according to memcg
3636 * @pdirty: out parameter for number of dirty pages
3637 * @pwriteback: out parameter for number of pages under writeback
3638 *
3639 * Determine the numbers of file, headroom, dirty, and writeback pages in
3640 * @wb's memcg.  File, dirty and writeback are self-explanatory.  Headroom
3641 * is a bit more involved.
3642 *
3643 * A memcg's headroom is "min(max, high) - used".  In the hierarchy, the
3644 * headroom is calculated as the lowest headroom of itself and the
3645 * ancestors.  Note that this doesn't consider the actual amount of
3646 * available memory in the system.  The caller should further cap
3647 * *@pheadroom accordingly.
3648 */
3649void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3650                         unsigned long *pheadroom, unsigned long *pdirty,
3651                         unsigned long *pwriteback)
3652{
3653        struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3654        struct mem_cgroup *parent;
3655
3656        *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3657
3658        /* this should eventually include NR_UNSTABLE_NFS */
3659        *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
3660        *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3661                                                     (1 << LRU_ACTIVE_FILE));
3662        *pheadroom = PAGE_COUNTER_MAX;
3663
3664        while ((parent = parent_mem_cgroup(memcg))) {
3665                unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3666                unsigned long used = page_counter_read(&memcg->memory);
3667
3668                *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
3669                memcg = parent;
3670        }
3671}
3672
3673#else   /* CONFIG_CGROUP_WRITEBACK */
3674
3675static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3676{
3677        return 0;
3678}
3679
3680static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3681{
3682}
3683
3684static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3685{
3686}
3687
3688#endif  /* CONFIG_CGROUP_WRITEBACK */
3689
3690/*
3691 * DO NOT USE IN NEW FILES.
3692 *
3693 * "cgroup.event_control" implementation.
3694 *
3695 * This is way over-engineered.  It tries to support fully configurable
3696 * events for each user.  Such level of flexibility is completely
3697 * unnecessary especially in the light of the planned unified hierarchy.
3698 *
3699 * Please deprecate this and replace with something simpler if at all
3700 * possible.
3701 */
3702
3703/*
3704 * Unregister event and free resources.
3705 *
3706 * Gets called from workqueue.
3707 */
3708static void memcg_event_remove(struct work_struct *work)
3709{
3710        struct mem_cgroup_event *event =
3711                container_of(work, struct mem_cgroup_event, remove);
3712        struct mem_cgroup *memcg = event->memcg;
3713
3714        remove_wait_queue(event->wqh, &event->wait);
3715
3716        event->unregister_event(memcg, event->eventfd);
3717
3718        /* Notify userspace the event is going away. */
3719        eventfd_signal(event->eventfd, 1);
3720
3721        eventfd_ctx_put(event->eventfd);
3722        kfree(event);
3723        css_put(&memcg->css);
3724}
3725
3726/*
3727 * Gets called on POLLHUP on eventfd when user closes it.
3728 *
3729 * Called with wqh->lock held and interrupts disabled.
3730 */
3731static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3732                            int sync, void *key)
3733{
3734        struct mem_cgroup_event *event =
3735                container_of(wait, struct mem_cgroup_event, wait);
3736        struct mem_cgroup *memcg = event->memcg;
3737        unsigned long flags = (unsigned long)key;
3738
3739        if (flags & POLLHUP) {
3740                /*
3741                 * If the event has been detached at cgroup removal, we
3742                 * can simply return knowing the other side will cleanup
3743                 * for us.
3744                 *
3745                 * We can't race against event freeing since the other
3746                 * side will require wqh->lock via remove_wait_queue(),
3747                 * which we hold.
3748                 */
3749                spin_lock(&memcg->event_list_lock);
3750                if (!list_empty(&event->list)) {
3751                        list_del_init(&event->list);
3752                        /*
3753                         * We are in atomic context, but cgroup_event_remove()
3754                         * may sleep, so we have to call it in workqueue.
3755                         */
3756                        schedule_work(&event->remove);
3757                }
3758                spin_unlock(&memcg->event_list_lock);
3759        }
3760
3761        return 0;
3762}
3763
3764static void memcg_event_ptable_queue_proc(struct file *file,
3765                wait_queue_head_t *wqh, poll_table *pt)
3766{
3767        struct mem_cgroup_event *event =
3768                container_of(pt, struct mem_cgroup_event, pt);
3769
3770        event->wqh = wqh;
3771        add_wait_queue(wqh, &event->wait);
3772}
3773
3774/*
3775 * DO NOT USE IN NEW FILES.
3776 *
3777 * Parse input and register new cgroup event handler.
3778 *
3779 * Input must be in format '<event_fd> <control_fd> <args>'.
3780 * Interpretation of args is defined by control file implementation.
3781 */
3782static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3783                                         char *buf, size_t nbytes, loff_t off)
3784{
3785        struct cgroup_subsys_state *css = of_css(of);
3786        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3787        struct mem_cgroup_event *event;
3788        struct cgroup_subsys_state *cfile_css;
3789        unsigned int efd, cfd;
3790        struct fd efile;
3791        struct fd cfile;
3792        const char *name;
3793        char *endp;
3794        int ret;
3795
3796        buf = strstrip(buf);
3797
3798        efd = simple_strtoul(buf, &endp, 10);
3799        if (*endp != ' ')
3800                return -EINVAL;
3801        buf = endp + 1;
3802
3803        cfd = simple_strtoul(buf, &endp, 10);
3804        if ((*endp != ' ') && (*endp != '\0'))
3805                return -EINVAL;
3806        buf = endp + 1;
3807
3808        event = kzalloc(sizeof(*event), GFP_KERNEL);
3809        if (!event)
3810                return -ENOMEM;
3811
3812        event->memcg = memcg;
3813        INIT_LIST_HEAD(&event->list);
3814        init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3815        init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3816        INIT_WORK(&event->remove, memcg_event_remove);
3817
3818        efile = fdget(efd);
3819        if (!efile.file) {
3820                ret = -EBADF;
3821                goto out_kfree;
3822        }
3823
3824        event->eventfd = eventfd_ctx_fileget(efile.file);
3825        if (IS_ERR(event->eventfd)) {
3826                ret = PTR_ERR(event->eventfd);
3827                goto out_put_efile;
3828        }
3829
3830        cfile = fdget(cfd);
3831        if (!cfile.file) {
3832                ret = -EBADF;
3833                goto out_put_eventfd;
3834        }
3835
3836        /* the process need read permission on control file */
3837        /* AV: shouldn't we check that it's been opened for read instead? */
3838        ret = inode_permission(file_inode(cfile.file), MAY_READ);
3839        if (ret < 0)
3840                goto out_put_cfile;
3841
3842        /*
3843         * Determine the event callbacks and set them in @event.  This used
3844         * to be done via struct cftype but cgroup core no longer knows
3845         * about these events.  The following is crude but the whole thing
3846         * is for compatibility anyway.
3847         *
3848         * DO NOT ADD NEW FILES.
3849         */
3850        name = cfile.file->f_path.dentry->d_name.name;
3851
3852        if (!strcmp(name, "memory.usage_in_bytes")) {
3853                event->register_event = mem_cgroup_usage_register_event;
3854                event->unregister_event = mem_cgroup_usage_unregister_event;
3855        } else if (!strcmp(name, "memory.oom_control")) {
3856                event->register_event = mem_cgroup_oom_register_event;
3857                event->unregister_event = mem_cgroup_oom_unregister_event;
3858        } else if (!strcmp(name, "memory.pressure_level")) {
3859                event->register_event = vmpressure_register_event;
3860                event->unregister_event = vmpressure_unregister_event;
3861        } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
3862                event->register_event = memsw_cgroup_usage_register_event;
3863                event->unregister_event = memsw_cgroup_usage_unregister_event;
3864        } else {
3865                ret = -EINVAL;
3866                goto out_put_cfile;
3867        }
3868
3869        /*
3870         * Verify @cfile should belong to @css.  Also, remaining events are
3871         * automatically removed on cgroup destruction but the removal is
3872         * asynchronous, so take an extra ref on @css.
3873         */
3874        cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
3875                                               &memory_cgrp_subsys);
3876        ret = -EINVAL;
3877        if (IS_ERR(cfile_css))
3878                goto out_put_cfile;
3879        if (cfile_css != css) {
3880                css_put(cfile_css);
3881                goto out_put_cfile;
3882        }
3883
3884        ret = event->register_event(memcg, event->eventfd, buf);
3885        if (ret)
3886                goto out_put_css;
3887
3888        efile.file->f_op->poll(efile.file, &event->pt);
3889
3890        spin_lock(&memcg->event_list_lock);
3891        list_add(&event->list, &memcg->event_list);
3892        spin_unlock(&memcg->event_list_lock);
3893
3894        fdput(cfile);
3895        fdput(efile);
3896
3897        return nbytes;
3898
3899out_put_css:
3900        css_put(css);
3901out_put_cfile:
3902        fdput(cfile);
3903out_put_eventfd:
3904        eventfd_ctx_put(event->eventfd);
3905out_put_efile:
3906        fdput(efile);
3907out_kfree:
3908        kfree(event);
3909
3910        return ret;
3911}
3912
3913static struct cftype mem_cgroup_legacy_files[] = {
3914        {
3915                .name = "usage_in_bytes",
3916                .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
3917                .read_u64 = mem_cgroup_read_u64,
3918        },
3919        {
3920                .name = "max_usage_in_bytes",
3921                .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
3922                .write = mem_cgroup_reset,
3923                .read_u64 = mem_cgroup_read_u64,
3924        },
3925        {
3926                .name = "limit_in_bytes",
3927                .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
3928                .write = mem_cgroup_write,
3929                .read_u64 = mem_cgroup_read_u64,
3930        },
3931        {
3932                .name = "soft_limit_in_bytes",
3933                .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
3934                .write = mem_cgroup_write,
3935                .read_u64 = mem_cgroup_read_u64,
3936        },
3937        {
3938                .name = "failcnt",
3939                .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
3940                .write = mem_cgroup_reset,
3941                .read_u64 = mem_cgroup_read_u64,
3942        },
3943        {
3944                .name = "stat",
3945                .seq_show = memcg_stat_show,
3946        },
3947        {
3948                .name = "force_empty",
3949                .write = mem_cgroup_force_empty_write,
3950        },
3951        {
3952                .name = "use_hierarchy",
3953                .write_u64 = mem_cgroup_hierarchy_write,
3954                .read_u64 = mem_cgroup_hierarchy_read,
3955        },
3956        {
3957                .name = "cgroup.event_control",         /* XXX: for compat */
3958                .write = memcg_write_event_control,
3959                .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
3960        },
3961        {
3962                .name = "swappiness",
3963                .read_u64 = mem_cgroup_swappiness_read,
3964                .write_u64 = mem_cgroup_swappiness_write,
3965        },
3966        {
3967                .name = "move_charge_at_immigrate",
3968                .read_u64 = mem_cgroup_move_charge_read,
3969                .write_u64 = mem_cgroup_move_charge_write,
3970        },
3971        {
3972                .name = "oom_control",
3973                .seq_show = mem_cgroup_oom_control_read,
3974                .write_u64 = mem_cgroup_oom_control_write,
3975                .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3976        },
3977        {
3978                .name = "pressure_level",
3979        },
3980#ifdef CONFIG_NUMA
3981        {
3982                .name = "numa_stat",
3983                .seq_show = memcg_numa_stat_show,
3984        },
3985#endif
3986        {
3987                .name = "kmem.limit_in_bytes",
3988                .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
3989                .write = mem_cgroup_write,
3990                .read_u64 = mem_cgroup_read_u64,
3991        },
3992        {
3993                .name = "kmem.usage_in_bytes",
3994                .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
3995                .read_u64 = mem_cgroup_read_u64,
3996        },
3997        {
3998                .name = "kmem.failcnt",
3999                .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
4000                .write = mem_cgroup_reset,
4001                .read_u64 = mem_cgroup_read_u64,
4002        },
4003        {
4004                .name = "kmem.max_usage_in_bytes",
4005                .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
4006                .write = mem_cgroup_reset,
4007                .read_u64 = mem_cgroup_read_u64,
4008        },
4009#ifdef CONFIG_SLABINFO
4010        {
4011                .name = "kmem.slabinfo",
4012                .seq_start = slab_start,
4013                .seq_next = slab_next,
4014                .seq_stop = slab_stop,
4015                .seq_show = memcg_slab_show,
4016        },
4017#endif
4018        {
4019                .name = "kmem.tcp.limit_in_bytes",
4020                .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4021                .write = mem_cgroup_write,
4022                .read_u64 = mem_cgroup_read_u64,
4023        },
4024        {
4025                .name = "kmem.tcp.usage_in_bytes",
4026                .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4027                .read_u64 = mem_cgroup_read_u64,
4028        },
4029        {
4030                .name = "kmem.tcp.failcnt",
4031                .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4032                .write = mem_cgroup_reset,
4033                .read_u64 = mem_cgroup_read_u64,
4034        },
4035        {
4036                .name = "kmem.tcp.max_usage_in_bytes",
4037                .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4038                .write = mem_cgroup_reset,
4039                .read_u64 = mem_cgroup_read_u64,
4040        },
4041        { },    /* terminate */
4042};
4043
4044/*
4045 * Private memory cgroup IDR
4046 *
4047 * Swap-out records and page cache shadow entries need to store memcg
4048 * references in constrained space, so we maintain an ID space that is
4049 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4050 * memory-controlled cgroups to 64k.
4051 *
4052 * However, there usually are many references to the oflline CSS after
4053 * the cgroup has been destroyed, such as page cache or reclaimable
4054 * slab objects, that don't need to hang on to the ID. We want to keep
4055 * those dead CSS from occupying IDs, or we might quickly exhaust the
4056 * relatively small ID space and prevent the creation of new cgroups
4057 * even when there are much fewer than 64k cgroups - possibly none.
4058 *
4059 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4060 * be freed and recycled when it's no longer needed, which is usually
4061 * when the CSS is offlined.
4062 *
4063 * The only exception to that are records of swapped out tmpfs/shmem
4064 * pages that need to be attributed to live ancestors on swapin. But
4065 * those references are manageable from userspace.
4066 */
4067
4068static DEFINE_IDR(mem_cgroup_idr);
4069
4070static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n)
4071{
4072        VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0);
4073        atomic_add(n, &memcg->id.ref);
4074}
4075
4076static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
4077{
4078        VM_BUG_ON(atomic_read(&memcg->id.ref) < n);
4079        if (atomic_sub_and_test(n, &memcg->id.ref)) {
4080                idr_remove(&mem_cgroup_idr, memcg->id.id);
4081                memcg->id.id = 0;
4082
4083                /* Memcg ID pins CSS */
4084                css_put(&memcg->css);
4085        }
4086}
4087
4088static inline void mem_cgroup_id_get(struct mem_cgroup *memcg)
4089{
4090        mem_cgroup_id_get_many(memcg, 1);
4091}
4092
4093static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
4094{
4095        mem_cgroup_id_put_many(memcg, 1);
4096}
4097
4098/**
4099 * mem_cgroup_from_id - look up a memcg from a memcg id
4100 * @id: the memcg id to look up
4101 *
4102 * Caller must hold rcu_read_lock().
4103 */
4104struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
4105{
4106        WARN_ON_ONCE(!rcu_read_lock_held());
4107        return idr_find(&mem_cgroup_idr, id);
4108}
4109
4110static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
4111{
4112        struct mem_cgroup_per_node *pn;
4113        int tmp = node;
4114        /*
4115         * This routine is called against possible nodes.
4116         * But it's BUG to call kmalloc() against offline node.
4117         *
4118         * TODO: this routine can waste much memory for nodes which will
4119         *       never be onlined. It's better to use memory hotplug callback
4120         *       function.
4121         */
4122        if (!node_state(node, N_NORMAL_MEMORY))
4123                tmp = -1;
4124        pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
4125        if (!pn)
4126                return 1;
4127
4128        lruvec_init(&pn->lruvec);
4129        pn->usage_in_excess = 0;
4130        pn->on_tree = false;
4131        pn->memcg = memcg;
4132
4133        memcg->nodeinfo[node] = pn;
4134        return 0;
4135}
4136
4137static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
4138{
4139        kfree(memcg->nodeinfo[node]);
4140}
4141
4142static void __mem_cgroup_free(struct mem_cgroup *memcg)
4143{
4144        int node;
4145
4146        for_each_node(node)
4147                free_mem_cgroup_per_node_info(memcg, node);
4148        free_percpu(memcg->stat);
4149        kfree(memcg);
4150}
4151
4152static void mem_cgroup_free(struct mem_cgroup *memcg)
4153{
4154        memcg_wb_domain_exit(memcg);
4155        __mem_cgroup_free(memcg);
4156}
4157
4158static struct mem_cgroup *mem_cgroup_alloc(void)
4159{
4160        struct mem_cgroup *memcg;
4161        size_t size;
4162        int node;
4163
4164        size = sizeof(struct mem_cgroup);
4165        size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
4166
4167        memcg = kzalloc(size, GFP_KERNEL);
4168        if (!memcg)
4169                return NULL;
4170
4171        memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
4172                                 1, MEM_CGROUP_ID_MAX,
4173                                 GFP_KERNEL);
4174        if (memcg->id.id < 0)
4175                goto fail;
4176
4177        memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4178        if (!memcg->stat)
4179                goto fail;
4180
4181        for_each_node(node)
4182                if (alloc_mem_cgroup_per_node_info(memcg, node))
4183                        goto fail;
4184
4185        if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4186                goto fail;
4187
4188        INIT_WORK(&memcg->high_work, high_work_func);
4189        memcg->last_scanned_node = MAX_NUMNODES;
4190        INIT_LIST_HEAD(&memcg->oom_notify);
4191        mutex_init(&memcg->thresholds_lock);
4192        spin_lock_init(&memcg->move_lock);
4193        vmpressure_init(&memcg->vmpressure);
4194        INIT_LIST_HEAD(&memcg->event_list);
4195        spin_lock_init(&memcg->event_list_lock);
4196        memcg->socket_pressure = jiffies;
4197#ifndef CONFIG_SLOB
4198        memcg->kmemcg_id = -1;
4199#endif
4200#ifdef CONFIG_CGROUP_WRITEBACK
4201        INIT_LIST_HEAD(&memcg->cgwb_list);
4202#endif
4203        idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
4204        return memcg;
4205fail:
4206        if (memcg->id.id > 0)
4207                idr_remove(&mem_cgroup_idr, memcg->id.id);
4208        __mem_cgroup_free(memcg);
4209        return NULL;
4210}
4211
4212static struct cgroup_subsys_state * __ref
4213mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
4214{
4215        struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4216        struct mem_cgroup *memcg;
4217        long error = -ENOMEM;
4218
4219        memcg = mem_cgroup_alloc();
4220        if (!memcg)
4221                return ERR_PTR(error);
4222
4223        memcg->high = PAGE_COUNTER_MAX;
4224        memcg->soft_limit = PAGE_COUNTER_MAX;
4225        if (parent) {
4226                memcg->swappiness = mem_cgroup_swappiness(parent);
4227                memcg->oom_kill_disable = parent->oom_kill_disable;
4228        }
4229        if (parent && parent->use_hierarchy) {
4230                memcg->use_hierarchy = true;
4231                page_counter_init(&memcg->memory, &parent->memory);
4232                page_counter_init(&memcg->swap, &parent->swap);
4233                page_counter_init(&memcg->memsw, &parent->memsw);
4234                page_counter_init(&memcg->kmem, &parent->kmem);
4235                page_counter_init(&memcg->tcpmem, &parent->tcpmem);
4236        } else {
4237                page_counter_init(&memcg->memory, NULL);
4238                page_counter_init(&memcg->swap, NULL);
4239                page_counter_init(&memcg->memsw, NULL);
4240                page_counter_init(&memcg->kmem, NULL);
4241                page_counter_init(&memcg->tcpmem, NULL);
4242                /*
4243                 * Deeper hierachy with use_hierarchy == false doesn't make
4244                 * much sense so let cgroup subsystem know about this
4245                 * unfortunate state in our controller.
4246                 */
4247                if (parent != root_mem_cgroup)
4248                        memory_cgrp_subsys.broken_hierarchy = true;
4249        }
4250
4251        /* The following stuff does not apply to the root */
4252        if (!parent) {
4253                root_mem_cgroup = memcg;
4254                return &memcg->css;
4255        }
4256
4257        error = memcg_online_kmem(memcg);
4258        if (error)
4259                goto fail;
4260
4261        if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
4262                static_branch_inc(&memcg_sockets_enabled_key);
4263
4264        return &memcg->css;
4265fail:
4266        mem_cgroup_free(memcg);
4267        return ERR_PTR(-ENOMEM);
4268}
4269
4270static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
4271{
4272        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4273
4274        /* Online state pins memcg ID, memcg ID pins CSS */
4275        atomic_set(&memcg->id.ref, 1);
4276        css_get(css);
4277        return 0;
4278}
4279
4280static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
4281{
4282        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4283        struct mem_cgroup_event *event, *tmp;
4284
4285        /*
4286         * Unregister events and notify userspace.
4287         * Notify userspace about cgroup removing only after rmdir of cgroup
4288         * directory to avoid race between userspace and kernelspace.
4289         */
4290        spin_lock(&memcg->event_list_lock);
4291        list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
4292                list_del_init(&event->list);
4293                schedule_work(&event->remove);
4294        }
4295        spin_unlock(&memcg->event_list_lock);
4296
4297        memcg_offline_kmem(memcg);
4298        wb_memcg_offline(memcg);
4299
4300        mem_cgroup_id_put(memcg);
4301}
4302
4303static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4304{
4305        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4306
4307        invalidate_reclaim_iterators(memcg);
4308}
4309
4310static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
4311{
4312        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4313
4314        if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
4315                static_branch_dec(&memcg_sockets_enabled_key);
4316
4317        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
4318                static_branch_dec(&memcg_sockets_enabled_key);
4319
4320        vmpressure_cleanup(&memcg->vmpressure);
4321        cancel_work_sync(&memcg->high_work);
4322        mem_cgroup_remove_from_trees(memcg);
4323        memcg_free_kmem(memcg);
4324        mem_cgroup_free(memcg);
4325}
4326
4327/**
4328 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4329 * @css: the target css
4330 *
4331 * Reset the states of the mem_cgroup associated with @css.  This is
4332 * invoked when the userland requests disabling on the default hierarchy
4333 * but the memcg is pinned through dependency.  The memcg should stop
4334 * applying policies and should revert to the vanilla state as it may be
4335 * made visible again.
4336 *
4337 * The current implementation only resets the essential configurations.
4338 * This needs to be expanded to cover all the visible parts.
4339 */
4340static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4341{
4342        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4343
4344        page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4345        page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4346        page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4347        page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4348        page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
4349        memcg->low = 0;
4350        memcg->high = PAGE_COUNTER_MAX;
4351        memcg->soft_limit = PAGE_COUNTER_MAX;
4352        memcg_wb_domain_size_changed(memcg);
4353}
4354
4355#ifdef CONFIG_MMU
4356/* Handlers for move charge at task migration. */
4357static int mem_cgroup_do_precharge(unsigned long count)
4358{
4359        int ret;
4360
4361        /* Try a single bulk charge without reclaim first, kswapd may wake */
4362        ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
4363        if (!ret) {
4364                mc.precharge += count;
4365                return ret;
4366        }
4367
4368        /* Try charges one by one with reclaim, but do not retry */
4369        while (count--) {
4370                ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1);
4371                if (ret)
4372                        return ret;
4373                mc.precharge++;
4374                cond_resched();
4375        }
4376        return 0;
4377}
4378
4379union mc_target {
4380        struct page     *page;
4381        swp_entry_t     ent;
4382};
4383
4384enum mc_target_type {
4385        MC_TARGET_NONE = 0,
4386        MC_TARGET_PAGE,
4387        MC_TARGET_SWAP,
4388};
4389
4390static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4391                                                unsigned long addr, pte_t ptent)
4392{
4393        struct page *page = vm_normal_page(vma, addr, ptent);
4394
4395        if (!page || !page_mapped(page))
4396                return NULL;
4397        if (PageAnon(page)) {
4398                if (!(mc.flags & MOVE_ANON))
4399                        return NULL;
4400        } else {
4401                if (!(mc.flags & MOVE_FILE))
4402                        return NULL;
4403        }
4404        if (!get_page_unless_zero(page))
4405                return NULL;
4406
4407        return page;
4408}
4409
4410#ifdef CONFIG_SWAP
4411static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4412                        pte_t ptent, swp_entry_t *entry)
4413{
4414        struct page *page = NULL;
4415        swp_entry_t ent = pte_to_swp_entry(ptent);
4416
4417        if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
4418                return NULL;
4419        /*
4420         * Because lookup_swap_cache() updates some statistics counter,
4421         * we call find_get_page() with swapper_space directly.
4422         */
4423        page = find_get_page(swap_address_space(ent), swp_offset(ent));
4424        if (do_memsw_account())
4425                entry->val = ent.val;
4426
4427        return page;
4428}
4429#else
4430static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4431                        pte_t ptent, swp_entry_t *entry)
4432{
4433        return NULL;
4434}
4435#endif
4436
4437static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4438                        unsigned long addr, pte_t ptent, swp_entry_t *entry)
4439{
4440        struct page *page = NULL;
4441        struct address_space *mapping;
4442        pgoff_t pgoff;
4443
4444        if (!vma->vm_file) /* anonymous vma */
4445                return NULL;
4446        if (!(mc.flags & MOVE_FILE))
4447                return NULL;
4448
4449        mapping = vma->vm_file->f_mapping;
4450        pgoff = linear_page_index(vma, addr);
4451
4452        /* page is moved even if it's not RSS of this task(page-faulted). */
4453#ifdef CONFIG_SWAP
4454        /* shmem/tmpfs may report page out on swap: account for that too. */
4455        if (shmem_mapping(mapping)) {
4456                page = find_get_entry(mapping, pgoff);
4457                if (radix_tree_exceptional_entry(page)) {
4458                        swp_entry_t swp = radix_to_swp_entry(page);
4459                        if (do_memsw_account())
4460                                *entry = swp;
4461                        page = find_get_page(swap_address_space(swp),
4462                                             swp_offset(swp));
4463                }
4464        } else
4465                page = find_get_page(mapping, pgoff);
4466#else
4467        page = find_get_page(mapping, pgoff);
4468#endif
4469        return page;
4470}
4471
4472/**
4473 * mem_cgroup_move_account - move account of the page
4474 * @page: the page
4475 * @compound: charge the page as compound or small page
4476 * @from: mem_cgroup which the page is moved from.
4477 * @to: mem_cgroup which the page is moved to. @from != @to.
4478 *
4479 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
4480 *
4481 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4482 * from old cgroup.
4483 */
4484static int mem_cgroup_move_account(struct page *page,
4485                                   bool compound,
4486                                   struct mem_cgroup *from,
4487                                   struct mem_cgroup *to)
4488{
4489        unsigned long flags;
4490        unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
4491        int ret;
4492        bool anon;
4493
4494        VM_BUG_ON(from == to);
4495        VM_BUG_ON_PAGE(PageLRU(page), page);
4496        VM_BUG_ON(compound && !PageTransHuge(page));
4497
4498        /*
4499         * Prevent mem_cgroup_migrate() from looking at
4500         * page->mem_cgroup of its source page while we change it.
4501         */
4502        ret = -EBUSY;
4503        if (!trylock_page(page))
4504                goto out;
4505
4506        ret = -EINVAL;
4507        if (page->mem_cgroup != from)
4508                goto out_unlock;
4509
4510        anon = PageAnon(page);
4511
4512        spin_lock_irqsave(&from->move_lock, flags);
4513
4514        if (!anon && page_mapped(page)) {
4515                __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4516                               nr_pages);
4517                __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4518                               nr_pages);
4519        }
4520
4521        /*
4522         * move_lock grabbed above and caller set from->moving_account, so
4523         * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4524         * So mapping should be stable for dirty pages.
4525         */
4526        if (!anon && PageDirty(page)) {
4527                struct address_space *mapping = page_mapping(page);
4528
4529                if (mapping_cap_account_dirty(mapping)) {
4530                        __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4531                                       nr_pages);
4532                        __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4533                                       nr_pages);
4534                }
4535        }
4536
4537        if (PageWriteback(page)) {
4538                __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4539                               nr_pages);
4540                __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4541                               nr_pages);
4542        }
4543
4544        /*
4545         * It is safe to change page->mem_cgroup here because the page
4546         * is referenced, charged, and isolated - we can't race with
4547         * uncharging, charging, migration, or LRU putback.
4548         */
4549
4550        /* caller should have done css_get */
4551        page->mem_cgroup = to;
4552        spin_unlock_irqrestore(&from->move_lock, flags);
4553
4554        ret = 0;
4555
4556        local_irq_disable();
4557        mem_cgroup_charge_statistics(to, page, compound, nr_pages);
4558        memcg_check_events(to, page);
4559        mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
4560        memcg_check_events(from, page);
4561        local_irq_enable();
4562out_unlock:
4563        unlock_page(page);
4564out:
4565        return ret;
4566}
4567
4568/**
4569 * get_mctgt_type - get target type of moving charge
4570 * @vma: the vma the pte to be checked belongs
4571 * @addr: the address corresponding to the pte to be checked
4572 * @ptent: the pte to be checked
4573 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4574 *
4575 * Returns
4576 *   0(MC_TARGET_NONE): if the pte is not a target for move charge.
4577 *   1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4578 *     move charge. if @target is not NULL, the page is stored in target->page
4579 *     with extra refcnt got(Callers should handle it).
4580 *   2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4581 *     target for charge migration. if @target is not NULL, the entry is stored
4582 *     in target->ent.
4583 *
4584 * Called with pte lock held.
4585 */
4586
4587static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
4588                unsigned long addr, pte_t ptent, union mc_target *target)
4589{
4590        struct page *page = NULL;
4591        enum mc_target_type ret = MC_TARGET_NONE;
4592        swp_entry_t ent = { .val = 0 };
4593
4594        if (pte_present(ptent))
4595                page = mc_handle_present_pte(vma, addr, ptent);
4596        else if (is_swap_pte(ptent))
4597                page = mc_handle_swap_pte(vma, ptent, &ent);
4598        else if (pte_none(ptent))
4599                page = mc_handle_file_pte(vma, addr, ptent, &ent);
4600
4601        if (!page && !ent.val)
4602                return ret;
4603        if (page) {
4604                /*
4605                 * Do only loose check w/o serialization.
4606                 * mem_cgroup_move_account() checks the page is valid or
4607                 * not under LRU exclusion.
4608                 */
4609                if (page->mem_cgroup == mc.from) {
4610                        ret = MC_TARGET_PAGE;
4611                        if (target)
4612                                target->page = page;
4613                }
4614                if (!ret || !target)
4615                        put_page(page);
4616        }
4617        /* There is a swap entry and a page doesn't exist or isn't charged */
4618        if (ent.val && !ret &&
4619            mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
4620                ret = MC_TARGET_SWAP;
4621                if (target)
4622                        target->ent = ent;
4623        }
4624        return ret;
4625}
4626
4627#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4628/*
4629 * We don't consider swapping or file mapped pages because THP does not
4630 * support them for now.
4631 * Caller should make sure that pmd_trans_huge(pmd) is true.
4632 */
4633static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4634                unsigned long addr, pmd_t pmd, union mc_target *target)
4635{
4636        struct page *page = NULL;
4637        enum mc_target_type ret = MC_TARGET_NONE;
4638
4639        page = pmd_page(pmd);
4640        VM_BUG_ON_PAGE(!page || !PageHead(page), page);
4641        if (!(mc.flags & MOVE_ANON))
4642                return ret;
4643        if (page->mem_cgroup == mc.from) {
4644                ret = MC_TARGET_PAGE;
4645                if (target) {
4646                        get_page(page);
4647                        target->page = page;
4648                }
4649        }
4650        return ret;
4651}
4652#else
4653static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4654                unsigned long addr, pmd_t pmd, union mc_target *target)
4655{
4656        return MC_TARGET_NONE;
4657}
4658#endif
4659
4660static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4661                                        unsigned long addr, unsigned long end,
4662                                        struct mm_walk *walk)
4663{
4664        struct vm_area_struct *vma = walk->vma;
4665        pte_t *pte;
4666        spinlock_t *ptl;
4667
4668        ptl = pmd_trans_huge_lock(pmd, vma);
4669        if (ptl) {
4670                if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4671                        mc.precharge += HPAGE_PMD_NR;
4672                spin_unlock(ptl);
4673                return 0;
4674        }
4675
4676        if (pmd_trans_unstable(pmd))
4677                return 0;
4678        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4679        for (; addr != end; pte++, addr += PAGE_SIZE)
4680                if (get_mctgt_type(vma, addr, *pte, NULL))
4681                        mc.precharge++; /* increment precharge temporarily */
4682        pte_unmap_unlock(pte - 1, ptl);
4683        cond_resched();
4684
4685        return 0;
4686}
4687
4688static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4689{
4690        unsigned long precharge;
4691
4692        struct mm_walk mem_cgroup_count_precharge_walk = {
4693                .pmd_entry = mem_cgroup_count_precharge_pte_range,
4694                .mm = mm,
4695        };
4696        down_read(&mm->mmap_sem);
4697        walk_page_range(0, mm->highest_vm_end,
4698                        &mem_cgroup_count_precharge_walk);
4699        up_read(&mm->mmap_sem);
4700
4701        precharge = mc.precharge;
4702        mc.precharge = 0;
4703
4704        return precharge;
4705}
4706
4707static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4708{
4709        unsigned long precharge = mem_cgroup_count_precharge(mm);
4710
4711        VM_BUG_ON(mc.moving_task);
4712        mc.moving_task = current;
4713        return mem_cgroup_do_precharge(precharge);
4714}
4715
4716/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4717static void __mem_cgroup_clear_mc(void)
4718{
4719        struct mem_cgroup *from = mc.from;
4720        struct mem_cgroup *to = mc.to;
4721
4722        /* we must uncharge all the leftover precharges from mc.to */
4723        if (mc.precharge) {
4724                cancel_charge(mc.to, mc.precharge);
4725                mc.precharge = 0;
4726        }
4727        /*
4728         * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4729         * we must uncharge here.
4730         */
4731        if (mc.moved_charge) {
4732                cancel_charge(mc.from, mc.moved_charge);
4733                mc.moved_charge = 0;
4734        }
4735        /* we must fixup refcnts and charges */
4736        if (mc.moved_swap) {
4737                /* uncharge swap account from the old cgroup */
4738                if (!mem_cgroup_is_root(mc.from))
4739                        page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
4740
4741                mem_cgroup_id_put_many(mc.from, mc.moved_swap);
4742
4743                /*
4744                 * we charged both to->memory and to->memsw, so we
4745                 * should uncharge to->memory.
4746                 */
4747                if (!mem_cgroup_is_root(mc.to))
4748                        page_counter_uncharge(&mc.to->memory, mc.moved_swap);
4749
4750                mem_cgroup_id_get_many(mc.to, mc.moved_swap);
4751                css_put_many(&mc.to->css, mc.moved_swap);
4752
4753                mc.moved_swap = 0;
4754        }
4755        memcg_oom_recover(from);
4756        memcg_oom_recover(to);
4757        wake_up_all(&mc.waitq);
4758}
4759
4760static void mem_cgroup_clear_mc(void)
4761{
4762        struct mm_struct *mm = mc.mm;
4763
4764        /*
4765         * we must clear moving_task before waking up waiters at the end of
4766         * task migration.
4767         */
4768        mc.moving_task = NULL;
4769        __mem_cgroup_clear_mc();
4770        spin_lock(&mc.lock);
4771        mc.from = NULL;
4772        mc.to = NULL;
4773        mc.mm = NULL;
4774        spin_unlock(&mc.lock);
4775
4776        mmput(mm);
4777}
4778
4779static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
4780{
4781        struct cgroup_subsys_state *css;
4782        struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
4783        struct mem_cgroup *from;
4784        struct task_struct *leader, *p;
4785        struct mm_struct *mm;
4786        unsigned long move_flags;
4787        int ret = 0;
4788
4789        /* charge immigration isn't supported on the default hierarchy */
4790        if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
4791                return 0;
4792
4793        /*
4794         * Multi-process migrations only happen on the default hierarchy
4795         * where charge immigration is not used.  Perform charge
4796         * immigration if @tset contains a leader and whine if there are
4797         * multiple.
4798         */
4799        p = NULL;
4800        cgroup_taskset_for_each_leader(leader, css, tset) {
4801                WARN_ON_ONCE(p);
4802                p = leader;
4803                memcg = mem_cgroup_from_css(css);
4804        }
4805        if (!p)
4806                return 0;
4807
4808        /*
4809         * We are now commited to this value whatever it is. Changes in this
4810         * tunable will only affect upcoming migrations, not the current one.
4811         * So we need to save it, and keep it going.
4812         */
4813        move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4814        if (!move_flags)
4815                return 0;
4816
4817        from = mem_cgroup_from_task(p);
4818
4819        VM_BUG_ON(from == memcg);
4820
4821        mm = get_task_mm(p);
4822        if (!mm)
4823                return 0;
4824        /* We move charges only when we move a owner of the mm */
4825        if (mm->owner == p) {
4826                VM_BUG_ON(mc.from);
4827                VM_BUG_ON(mc.to);
4828                VM_BUG_ON(mc.precharge);
4829                VM_BUG_ON(mc.moved_charge);
4830                VM_BUG_ON(mc.moved_swap);
4831
4832                spin_lock(&mc.lock);
4833                mc.mm = mm;
4834                mc.from = from;
4835                mc.to = memcg;
4836                mc.flags = move_flags;
4837                spin_unlock(&mc.lock);
4838                /* We set mc.moving_task later */
4839
4840                ret = mem_cgroup_precharge_mc(mm);
4841                if (ret)
4842                        mem_cgroup_clear_mc();
4843        } else {
4844                mmput(mm);
4845        }
4846        return ret;
4847}
4848
4849static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
4850{
4851        if (mc.to)
4852                mem_cgroup_clear_mc();
4853}
4854
4855static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4856                                unsigned long addr, unsigned long end,
4857                                struct mm_walk *walk)
4858{
4859        int ret = 0;
4860        struct vm_area_struct *vma = walk->vma;
4861        pte_t *pte;
4862        spinlock_t *ptl;
4863        enum mc_target_type target_type;
4864        union mc_target target;
4865        struct page *page;
4866
4867        ptl = pmd_trans_huge_lock(pmd, vma);
4868        if (ptl) {
4869                if (mc.precharge < HPAGE_PMD_NR) {
4870                        spin_unlock(ptl);
4871                        return 0;
4872                }
4873                target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4874                if (target_type == MC_TARGET_PAGE) {
4875                        page = target.page;
4876                        if (!isolate_lru_page(page)) {
4877                                if (!mem_cgroup_move_account(page, true,
4878                                                             mc.from, mc.to)) {
4879                                        mc.precharge -= HPAGE_PMD_NR;
4880                                        mc.moved_charge += HPAGE_PMD_NR;
4881                                }
4882                                putback_lru_page(page);
4883                        }
4884                        put_page(page);
4885                }
4886                spin_unlock(ptl);
4887                return 0;
4888        }
4889
4890        if (pmd_trans_unstable(pmd))
4891                return 0;
4892retry:
4893        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4894        for (; addr != end; addr += PAGE_SIZE) {
4895                pte_t ptent = *(pte++);
4896                swp_entry_t ent;
4897
4898                if (!mc.precharge)
4899                        break;
4900
4901                switch (get_mctgt_type(vma, addr, ptent, &target)) {
4902                case MC_TARGET_PAGE:
4903                        page = target.page;
4904                        /*
4905                         * We can have a part of the split pmd here. Moving it
4906                         * can be done but it would be too convoluted so simply
4907                         * ignore such a partial THP and keep it in original
4908                         * memcg. There should be somebody mapping the head.
4909                         */
4910                        if (PageTransCompound(page))
4911                                goto put;
4912                        if (isolate_lru_page(page))
4913                                goto put;
4914                        if (!mem_cgroup_move_account(page, false,
4915                                                mc.from, mc.to)) {
4916                                mc.precharge--;
4917                                /* we uncharge from mc.from later. */
4918                                mc.moved_charge++;
4919                        }
4920                        putback_lru_page(page);
4921put:                    /* get_mctgt_type() gets the page */
4922                        put_page(page);
4923                        break;
4924                case MC_TARGET_SWAP:
4925                        ent = target.ent;
4926                        if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
4927                                mc.precharge--;
4928                                /* we fixup refcnts and charges later. */
4929                                mc.moved_swap++;
4930                        }
4931                        break;
4932                default:
4933                        break;
4934                }
4935        }
4936        pte_unmap_unlock(pte - 1, ptl);
4937        cond_resched();
4938
4939        if (addr != end) {
4940                /*
4941                 * We have consumed all precharges we got in can_attach().
4942                 * We try charge one by one, but don't do any additional
4943                 * charges to mc.to if we have failed in charge once in attach()
4944                 * phase.
4945                 */
4946                ret = mem_cgroup_do_precharge(1);
4947                if (!ret)
4948                        goto retry;
4949        }
4950
4951        return ret;
4952}
4953
4954static void mem_cgroup_move_charge(void)
4955{
4956        struct mm_walk mem_cgroup_move_charge_walk = {
4957                .pmd_entry = mem_cgroup_move_charge_pte_range,
4958                .mm = mc.mm,
4959        };
4960
4961        lru_add_drain_all();
4962        /*
4963         * Signal lock_page_memcg() to take the memcg's move_lock
4964         * while we're moving its pages to another memcg. Then wait
4965         * for already started RCU-only updates to finish.
4966         */
4967        atomic_inc(&mc.from->moving_account);
4968        synchronize_rcu();
4969retry:
4970        if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
4971                /*
4972                 * Someone who are holding the mmap_sem might be waiting in
4973                 * waitq. So we cancel all extra charges, wake up all waiters,
4974                 * and retry. Because we cancel precharges, we might not be able
4975                 * to move enough charges, but moving charge is a best-effort
4976                 * feature anyway, so it wouldn't be a big problem.
4977                 */
4978                __mem_cgroup_clear_mc();
4979                cond_resched();
4980                goto retry;
4981        }
4982        /*
4983         * When we have consumed all precharges and failed in doing
4984         * additional charge, the page walk just aborts.
4985         */
4986        walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk);
4987
4988        up_read(&mc.mm->mmap_sem);
4989        atomic_dec(&mc.from->moving_account);
4990}
4991
4992static void mem_cgroup_move_task(void)
4993{
4994        if (mc.to) {
4995                mem_cgroup_move_charge();
4996                mem_cgroup_clear_mc();
4997        }
4998}
4999#else   /* !CONFIG_MMU */
5000static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
5001{
5002        return 0;
5003}
5004static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
5005{
5006}
5007static void mem_cgroup_move_task(void)
5008{
5009}
5010#endif
5011
5012/*
5013 * Cgroup retains root cgroups across [un]mount cycles making it necessary
5014 * to verify whether we're attached to the default hierarchy on each mount
5015 * attempt.
5016 */
5017static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
5018{
5019        /*
5020         * use_hierarchy is forced on the default hierarchy.  cgroup core
5021         * guarantees that @root doesn't have any children, so turning it
5022         * on for the root memcg is enough.
5023         */
5024        if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5025                root_mem_cgroup->use_hierarchy = true;
5026        else
5027                root_mem_cgroup->use_hierarchy = false;
5028}
5029
5030static u64 memory_current_read(struct cgroup_subsys_state *css,
5031                               struct cftype *cft)
5032{
5033        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5034
5035        return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
5036}
5037
5038static int memory_low_show(struct seq_file *m, void *v)
5039{
5040        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5041        unsigned long low = READ_ONCE(memcg->low);
5042
5043        if (low == PAGE_COUNTER_MAX)
5044                seq_puts(m, "max\n");
5045        else
5046                seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5047
5048        return 0;
5049}
5050
5051static ssize_t memory_low_write(struct kernfs_open_file *of,
5052                                char *buf, size_t nbytes, loff_t off)
5053{
5054        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5055        unsigned long low;
5056        int err;
5057
5058        buf = strstrip(buf);
5059        err = page_counter_memparse(buf, "max", &low);
5060        if (err)
5061                return err;
5062
5063        memcg->low = low;
5064
5065        return nbytes;
5066}
5067
5068static int memory_high_show(struct seq_file *m, void *v)
5069{
5070        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5071        unsigned long high = READ_ONCE(memcg->high);
5072
5073        if (high == PAGE_COUNTER_MAX)
5074                seq_puts(m, "max\n");
5075        else
5076                seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5077
5078        return 0;
5079}
5080
5081static ssize_t memory_high_write(struct kernfs_open_file *of,
5082                                 char *buf, size_t nbytes, loff_t off)
5083{
5084        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5085        unsigned long nr_pages;
5086        unsigned long high;
5087        int err;
5088
5089        buf = strstrip(buf);
5090        err = page_counter_memparse(buf, "max", &high);
5091        if (err)
5092                return err;
5093
5094        memcg->high = high;
5095
5096        nr_pages = page_counter_read(&memcg->memory);
5097        if (nr_pages > high)
5098                try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5099                                             GFP_KERNEL, true);
5100
5101        memcg_wb_domain_size_changed(memcg);
5102        return nbytes;
5103}
5104
5105static int memory_max_show(struct seq_file *m, void *v)
5106{
5107        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5108        unsigned long max = READ_ONCE(memcg->memory.limit);
5109
5110        if (max == PAGE_COUNTER_MAX)
5111                seq_puts(m, "max\n");
5112        else
5113                seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5114
5115        return 0;
5116}
5117
5118static ssize_t memory_max_write(struct kernfs_open_file *of,
5119                                char *buf, size_t nbytes, loff_t off)
5120{
5121        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5122        unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5123        bool drained = false;
5124        unsigned long max;
5125        int err;
5126
5127        buf = strstrip(buf);
5128        err = page_counter_memparse(buf, "max", &max);
5129        if (err)
5130                return err;
5131
5132        xchg(&memcg->memory.limit, max);
5133
5134        for (;;) {
5135                unsigned long nr_pages = page_counter_read(&memcg->memory);
5136
5137                if (nr_pages <= max)
5138                        break;
5139
5140                if (signal_pending(current)) {
5141                        err = -EINTR;
5142                        break;
5143                }
5144
5145                if (!drained) {
5146                        drain_all_stock(memcg);
5147                        drained = true;
5148                        continue;
5149                }
5150
5151                if (nr_reclaims) {
5152                        if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5153                                                          GFP_KERNEL, true))
5154                                nr_reclaims--;
5155                        continue;
5156                }
5157
5158                mem_cgroup_events(memcg, MEMCG_OOM, 1);
5159                if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5160                        break;
5161        }
5162
5163        memcg_wb_domain_size_changed(memcg);
5164        return nbytes;
5165}
5166
5167static int memory_events_show(struct seq_file *m, void *v)
5168{
5169        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5170
5171        seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5172        seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5173        seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5174        seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5175
5176        return 0;
5177}
5178
5179static int memory_stat_show(struct seq_file *m, void *v)
5180{
5181        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5182        unsigned long stat[MEMCG_NR_STAT];
5183        unsigned long events[MEMCG_NR_EVENTS];
5184        int i;
5185
5186        /*
5187         * Provide statistics on the state of the memory subsystem as
5188         * well as cumulative event counters that show past behavior.
5189         *
5190         * This list is ordered following a combination of these gradients:
5191         * 1) generic big picture -> specifics and details
5192         * 2) reflecting userspace activity -> reflecting kernel heuristics
5193         *
5194         * Current memory state:
5195         */
5196
5197        tree_stat(memcg, stat);
5198        tree_events(memcg, events);
5199
5200        seq_printf(m, "anon %llu\n",
5201                   (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE);
5202        seq_printf(m, "file %llu\n",
5203                   (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE);
5204        seq_printf(m, "kernel_stack %llu\n",
5205                   (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024);
5206        seq_printf(m, "slab %llu\n",
5207                   (u64)(stat[MEMCG_SLAB_RECLAIMABLE] +
5208                         stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
5209        seq_printf(m, "sock %llu\n",
5210                   (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
5211
5212        seq_printf(m, "file_mapped %llu\n",
5213                   (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE);
5214        seq_printf(m, "file_dirty %llu\n",
5215                   (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE);
5216        seq_printf(m, "file_writeback %llu\n",
5217                   (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE);
5218
5219        for (i = 0; i < NR_LRU_LISTS; i++) {
5220                struct mem_cgroup *mi;
5221                unsigned long val = 0;
5222
5223                for_each_mem_cgroup_tree(mi, memcg)
5224                        val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5225                seq_printf(m, "%s %llu\n",
5226                           mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5227        }
5228
5229        seq_printf(m, "slab_reclaimable %llu\n",
5230                   (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE);
5231        seq_printf(m, "slab_unreclaimable %llu\n",
5232                   (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5233
5234        /* Accumulated memory events */
5235
5236        seq_printf(m, "pgfault %lu\n",
5237                   events[MEM_CGROUP_EVENTS_PGFAULT]);
5238        seq_printf(m, "pgmajfault %lu\n",
5239                   events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
5240
5241        return 0;
5242}
5243
5244static struct cftype memory_files[] = {
5245        {
5246                .name = "current",
5247                .flags = CFTYPE_NOT_ON_ROOT,
5248                .read_u64 = memory_current_read,
5249        },
5250        {
5251                .name = "low",
5252                .flags = CFTYPE_NOT_ON_ROOT,
5253                .seq_show = memory_low_show,
5254                .write = memory_low_write,
5255        },
5256        {
5257                .name = "high",
5258                .flags = CFTYPE_NOT_ON_ROOT,
5259                .seq_show = memory_high_show,
5260                .write = memory_high_write,
5261        },
5262        {
5263                .name = "max",
5264                .flags = CFTYPE_NOT_ON_ROOT,
5265                .seq_show = memory_max_show,
5266                .write = memory_max_write,
5267        },
5268        {
5269                .name = "events",
5270                .flags = CFTYPE_NOT_ON_ROOT,
5271                .file_offset = offsetof(struct mem_cgroup, events_file),
5272                .seq_show = memory_events_show,
5273        },
5274        {
5275                .name = "stat",
5276                .flags = CFTYPE_NOT_ON_ROOT,
5277                .seq_show = memory_stat_show,
5278        },
5279        { }     /* terminate */
5280};
5281
5282struct cgroup_subsys memory_cgrp_subsys = {
5283        .css_alloc = mem_cgroup_css_alloc,
5284        .css_online = mem_cgroup_css_online,
5285        .css_offline = mem_cgroup_css_offline,
5286        .css_released = mem_cgroup_css_released,
5287        .css_free = mem_cgroup_css_free,
5288        .css_reset = mem_cgroup_css_reset,
5289        .can_attach = mem_cgroup_can_attach,
5290        .cancel_attach = mem_cgroup_cancel_attach,
5291        .post_attach = mem_cgroup_move_task,
5292        .bind = mem_cgroup_bind,
5293        .dfl_cftypes = memory_files,
5294        .legacy_cftypes = mem_cgroup_legacy_files,
5295        .early_init = 0,
5296};
5297
5298/**
5299 * mem_cgroup_low - check if memory consumption is below the normal range
5300 * @root: the highest ancestor to consider
5301 * @memcg: the memory cgroup to check
5302 *
5303 * Returns %true if memory consumption of @memcg, and that of all
5304 * configurable ancestors up to @root, is below the normal range.
5305 */
5306bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5307{
5308        if (mem_cgroup_disabled())
5309                return false;
5310
5311        /*
5312         * The toplevel group doesn't have a configurable range, so
5313         * it's never low when looked at directly, and it is not
5314         * considered an ancestor when assessing the hierarchy.
5315         */
5316
5317        if (memcg == root_mem_cgroup)
5318                return false;
5319
5320        if (page_counter_read(&memcg->memory) >= memcg->low)
5321                return false;
5322
5323        while (memcg != root) {
5324                memcg = parent_mem_cgroup(memcg);
5325
5326                if (memcg == root_mem_cgroup)
5327                        break;
5328
5329                if (page_counter_read(&memcg->memory) >= memcg->low)
5330                        return false;
5331        }
5332        return true;
5333}
5334
5335/**
5336 * mem_cgroup_try_charge - try charging a page
5337 * @page: page to charge
5338 * @mm: mm context of the victim
5339 * @gfp_mask: reclaim mode
5340 * @memcgp: charged memcg return
5341 * @compound: charge the page as compound or small page
5342 *
5343 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5344 * pages according to @gfp_mask if necessary.
5345 *
5346 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5347 * Otherwise, an error code is returned.
5348 *
5349 * After page->mapping has been set up, the caller must finalize the
5350 * charge with mem_cgroup_commit_charge().  Or abort the transaction
5351 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5352 */
5353int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
5354                          gfp_t gfp_mask, struct mem_cgroup **memcgp,
5355                          bool compound)
5356{
5357        struct mem_cgroup *memcg = NULL;
5358        unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
5359        int ret = 0;
5360
5361        if (mem_cgroup_disabled())
5362                goto out;
5363
5364        if (PageSwapCache(page)) {
5365                /*
5366                 * Every swap fault against a single page tries to charge the
5367                 * page, bail as early as possible.  shmem_unuse() encounters
5368                 * already charged pages, too.  The USED bit is protected by
5369                 * the page lock, which serializes swap cache removal, which
5370                 * in turn serializes uncharging.
5371                 */
5372                VM_BUG_ON_PAGE(!PageLocked(page), page);
5373                if (page->mem_cgroup)
5374                        goto out;
5375
5376                if (do_swap_account) {
5377                        swp_entry_t ent = { .val = page_private(page), };
5378                        unsigned short id = lookup_swap_cgroup_id(ent);
5379
5380                        rcu_read_lock();
5381                        memcg = mem_cgroup_from_id(id);
5382                        if (memcg && !css_tryget_online(&memcg->css))
5383                                memcg = NULL;
5384                        rcu_read_unlock();
5385                }
5386        }
5387
5388        if (!memcg)
5389                memcg = get_mem_cgroup_from_mm(mm);
5390
5391        ret = try_charge(memcg, gfp_mask, nr_pages);
5392
5393        css_put(&memcg->css);
5394out:
5395        *memcgp = memcg;
5396        return ret;
5397}
5398
5399/**
5400 * mem_cgroup_commit_charge - commit a page charge
5401 * @page: page to charge
5402 * @memcg: memcg to charge the page to
5403 * @lrucare: page might be on LRU already
5404 * @compound: charge the page as compound or small page
5405 *
5406 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5407 * after page->mapping has been set up.  This must happen atomically
5408 * as part of the page instantiation, i.e. under the page table lock
5409 * for anonymous pages, under the page lock for page and swap cache.
5410 *
5411 * In addition, the page must not be on the LRU during the commit, to
5412 * prevent racing with task migration.  If it might be, use @lrucare.
5413 *
5414 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5415 */
5416void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
5417                              bool lrucare, bool compound)
5418{
5419        unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
5420
5421        VM_BUG_ON_PAGE(!page->mapping, page);
5422        VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5423
5424        if (mem_cgroup_disabled())
5425                return;
5426        /*
5427         * Swap faults will attempt to charge the same page multiple
5428         * times.  But reuse_swap_page() might have removed the page
5429         * from swapcache already, so we can't check PageSwapCache().
5430         */
5431        if (!memcg)
5432                return;
5433
5434        commit_charge(page, memcg, lrucare);
5435
5436        local_irq_disable();
5437        mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
5438        memcg_check_events(memcg, page);
5439        local_irq_enable();
5440
5441        if (do_memsw_account() && PageSwapCache(page)) {
5442                swp_entry_t entry = { .val = page_private(page) };
5443                /*
5444                 * The swap entry might not get freed for a long time,
5445                 * let's not wait for it.  The page already received a
5446                 * memory+swap charge, drop the swap entry duplicate.
5447                 */
5448                mem_cgroup_uncharge_swap(entry);
5449        }
5450}
5451
5452/**
5453 * mem_cgroup_cancel_charge - cancel a page charge
5454 * @page: page to charge
5455 * @memcg: memcg to charge the page to
5456 * @compound: charge the page as compound or small page
5457 *
5458 * Cancel a charge transaction started by mem_cgroup_try_charge().
5459 */
5460void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5461                bool compound)
5462{
5463        unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
5464
5465        if (mem_cgroup_disabled())
5466                return;
5467        /*
5468         * Swap faults will attempt to charge the same page multiple
5469         * times.  But reuse_swap_page() might have removed the page
5470         * from swapcache already, so we can't check PageSwapCache().
5471         */
5472        if (!memcg)
5473                return;
5474
5475        cancel_charge(memcg, nr_pages);
5476}
5477
5478static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
5479                           unsigned long nr_anon, unsigned long nr_file,
5480                           unsigned long nr_huge, unsigned long nr_kmem,
5481                           struct page *dummy_page)
5482{
5483        unsigned long nr_pages = nr_anon + nr_file + nr_kmem;
5484        unsigned long flags;
5485
5486        if (!mem_cgroup_is_root(memcg)) {
5487                page_counter_uncharge(&memcg->memory, nr_pages);
5488                if (do_memsw_account())
5489                        page_counter_uncharge(&memcg->memsw, nr_pages);
5490                if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem)
5491                        page_counter_uncharge(&memcg->kmem, nr_kmem);
5492                memcg_oom_recover(memcg);
5493        }
5494
5495        local_irq_save(flags);
5496        __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5497        __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5498        __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5499        __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
5500        __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
5501        memcg_check_events(memcg, dummy_page);
5502        local_irq_restore(flags);
5503
5504        if (!mem_cgroup_is_root(memcg))
5505                css_put_many(&memcg->css, nr_pages);
5506}
5507
5508static void uncharge_list(struct list_head *page_list)
5509{
5510        struct mem_cgroup *memcg = NULL;
5511        unsigned long nr_anon = 0;
5512        unsigned long nr_file = 0;
5513        unsigned long nr_huge = 0;
5514        unsigned long nr_kmem = 0;
5515        unsigned long pgpgout = 0;
5516        struct list_head *next;
5517        struct page *page;
5518
5519        /*
5520         * Note that the list can be a single page->lru; hence the
5521         * do-while loop instead of a simple list_for_each_entry().
5522         */
5523        next = page_list->next;
5524        do {
5525                page = list_entry(next, struct page, lru);
5526                next = page->lru.next;
5527
5528                VM_BUG_ON_PAGE(PageLRU(page), page);
5529                VM_BUG_ON_PAGE(page_count(page), page);
5530
5531                if (!page->mem_cgroup)
5532                        continue;
5533
5534                /*
5535                 * Nobody should be changing or seriously looking at
5536                 * page->mem_cgroup at this point, we have fully
5537                 * exclusive access to the page.
5538                 */
5539
5540                if (memcg != page->mem_cgroup) {
5541                        if (memcg) {
5542                                uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5543                                               nr_huge, nr_kmem, page);
5544                                pgpgout = nr_anon = nr_file =
5545                                        nr_huge = nr_kmem = 0;
5546                        }
5547                        memcg = page->mem_cgroup;
5548                }
5549
5550                if (!PageKmemcg(page)) {
5551                        unsigned int nr_pages = 1;
5552
5553                        if (PageTransHuge(page)) {
5554                                nr_pages <<= compound_order(page);
5555                                nr_huge += nr_pages;
5556                        }
5557                        if (PageAnon(page))
5558                                nr_anon += nr_pages;
5559                        else
5560                                nr_file += nr_pages;
5561                        pgpgout++;
5562                } else {
5563                        nr_kmem += 1 << compound_order(page);
5564                        __ClearPageKmemcg(page);
5565                }
5566
5567                page->mem_cgroup = NULL;
5568        } while (next != page_list);
5569
5570        if (memcg)
5571                uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5572                               nr_huge, nr_kmem, page);
5573}
5574
5575/**
5576 * mem_cgroup_uncharge - uncharge a page
5577 * @page: page to uncharge
5578 *
5579 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5580 * mem_cgroup_commit_charge().
5581 */
5582void mem_cgroup_uncharge(struct page *page)
5583{
5584        if (mem_cgroup_disabled())
5585                return;
5586
5587        /* Don't touch page->lru of any random page, pre-check: */
5588        if (!page->mem_cgroup)
5589                return;
5590
5591        INIT_LIST_HEAD(&page->lru);
5592        uncharge_list(&page->lru);
5593}
5594
5595/**
5596 * mem_cgroup_uncharge_list - uncharge a list of page
5597 * @page_list: list of pages to uncharge
5598 *
5599 * Uncharge a list of pages previously charged with
5600 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5601 */
5602void mem_cgroup_uncharge_list(struct list_head *page_list)
5603{
5604        if (mem_cgroup_disabled())
5605                return;
5606
5607        if (!list_empty(page_list))
5608                uncharge_list(page_list);
5609}
5610
5611/**
5612 * mem_cgroup_migrate - charge a page's replacement
5613 * @oldpage: currently circulating page
5614 * @newpage: replacement page
5615 *
5616 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5617 * be uncharged upon free.
5618 *
5619 * Both pages must be locked, @newpage->mapping must be set up.
5620 */
5621void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
5622{
5623        struct mem_cgroup *memcg;
5624        unsigned int nr_pages;
5625        bool compound;
5626        unsigned long flags;
5627
5628        VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5629        VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
5630        VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
5631        VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5632                       newpage);
5633
5634        if (mem_cgroup_disabled())
5635                return;
5636
5637        /* Page cache replacement: new page already charged? */
5638        if (newpage->mem_cgroup)
5639                return;
5640
5641        /* Swapcache readahead pages can get replaced before being charged */
5642        memcg = oldpage->mem_cgroup;
5643        if (!memcg)
5644                return;
5645
5646        /* Force-charge the new page. The old one will be freed soon */
5647        compound = PageTransHuge(newpage);
5648        nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5649
5650        page_counter_charge(&memcg->memory, nr_pages);
5651        if (do_memsw_account())
5652                page_counter_charge(&memcg->memsw, nr_pages);
5653        css_get_many(&memcg->css, nr_pages);
5654
5655        commit_charge(newpage, memcg, false);
5656
5657        local_irq_save(flags);
5658        mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5659        memcg_check_events(memcg, newpage);
5660        local_irq_restore(flags);
5661}
5662
5663DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
5664EXPORT_SYMBOL(memcg_sockets_enabled_key);
5665
5666void mem_cgroup_sk_alloc(struct sock *sk)
5667{
5668        struct mem_cgroup *memcg;
5669
5670        if (!mem_cgroup_sockets_enabled)
5671                return;
5672
5673        /*
5674         * Socket cloning can throw us here with sk_memcg already
5675         * filled. It won't however, necessarily happen from
5676         * process context. So the test for root memcg given
5677         * the current task's memcg won't help us in this case.
5678         *
5679         * Respecting the original socket's memcg is a better
5680         * decision in this case.
5681         */
5682        if (sk->sk_memcg) {
5683                BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5684                css_get(&sk->sk_memcg->css);
5685                return;
5686        }
5687
5688        rcu_read_lock();
5689        memcg = mem_cgroup_from_task(current);
5690        if (memcg == root_mem_cgroup)
5691                goto out;
5692        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
5693                goto out;
5694        if (css_tryget_online(&memcg->css))
5695                sk->sk_memcg = memcg;
5696out:
5697        rcu_read_unlock();
5698}
5699
5700void mem_cgroup_sk_free(struct sock *sk)
5701{
5702        if (sk->sk_memcg)
5703                css_put(&sk->sk_memcg->css);
5704}
5705
5706/**
5707 * mem_cgroup_charge_skmem - charge socket memory
5708 * @memcg: memcg to charge
5709 * @nr_pages: number of pages to charge
5710 *
5711 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5712 * @memcg's configured limit, %false if the charge had to be forced.
5713 */
5714bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5715{
5716        gfp_t gfp_mask = GFP_KERNEL;
5717
5718        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
5719                struct page_counter *fail;
5720
5721                if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5722                        memcg->tcpmem_pressure = 0;
5723                        return true;
5724                }
5725                page_counter_charge(&memcg->tcpmem, nr_pages);
5726                memcg->tcpmem_pressure = 1;
5727                return false;
5728        }
5729
5730        /* Don't block in the packet receive path */
5731        if (in_softirq())
5732                gfp_mask = GFP_NOWAIT;
5733
5734        this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5735
5736        if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5737                return true;
5738
5739        try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
5740        return false;
5741}
5742
5743/**
5744 * mem_cgroup_uncharge_skmem - uncharge socket memory
5745 * @memcg - memcg to uncharge
5746 * @nr_pages - number of pages to uncharge
5747 */
5748void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5749{
5750        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
5751                page_counter_uncharge(&memcg->tcpmem, nr_pages);
5752                return;
5753        }
5754
5755        this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5756
5757        page_counter_uncharge(&memcg->memory, nr_pages);
5758        css_put_many(&memcg->css, nr_pages);
5759}
5760
5761static int __init cgroup_memory(char *s)
5762{
5763        char *token;
5764
5765        while ((token = strsep(&s, ",")) != NULL) {
5766                if (!*token)
5767                        continue;
5768                if (!strcmp(token, "nosocket"))
5769                        cgroup_memory_nosocket = true;
5770                if (!strcmp(token, "nokmem"))
5771                        cgroup_memory_nokmem = true;
5772        }
5773        return 0;
5774}
5775__setup("cgroup.memory=", cgroup_memory);
5776
5777/*
5778 * subsys_initcall() for memory controller.
5779 *
5780 * Some parts like hotcpu_notifier() have to be initialized from this context
5781 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5782 * everything that doesn't depend on a specific mem_cgroup structure should
5783 * be initialized from here.
5784 */
5785static int __init mem_cgroup_init(void)
5786{
5787        int cpu, node;
5788
5789        hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
5790
5791        for_each_possible_cpu(cpu)
5792                INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5793                          drain_local_stock);
5794
5795        for_each_node(node) {
5796                struct mem_cgroup_tree_per_node *rtpn;
5797
5798                rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5799                                    node_online(node) ? node : NUMA_NO_NODE);
5800
5801                rtpn->rb_root = RB_ROOT;
5802                spin_lock_init(&rtpn->lock);
5803                soft_limit_tree.rb_tree_per_node[node] = rtpn;
5804        }
5805
5806        return 0;
5807}
5808subsys_initcall(mem_cgroup_init);
5809
5810#ifdef CONFIG_MEMCG_SWAP
5811static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
5812{
5813        while (!atomic_inc_not_zero(&memcg->id.ref)) {
5814                /*
5815                 * The root cgroup cannot be destroyed, so it's refcount must
5816                 * always be >= 1.
5817                 */
5818                if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
5819                        VM_BUG_ON(1);
5820                        break;
5821                }
5822                memcg = parent_mem_cgroup(memcg);
5823                if (!memcg)
5824                        memcg = root_mem_cgroup;
5825        }
5826        return memcg;
5827}
5828
5829/**
5830 * mem_cgroup_swapout - transfer a memsw charge to swap
5831 * @page: page whose memsw charge to transfer
5832 * @entry: swap entry to move the charge to
5833 *
5834 * Transfer the memsw charge of @page to @entry.
5835 */
5836void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5837{
5838        struct mem_cgroup *memcg, *swap_memcg;
5839        unsigned short oldid;
5840
5841        VM_BUG_ON_PAGE(PageLRU(page), page);
5842        VM_BUG_ON_PAGE(page_count(page), page);
5843
5844        if (!do_memsw_account())
5845                return;
5846
5847        memcg = page->mem_cgroup;
5848
5849        /* Readahead page, never charged */
5850        if (!memcg)
5851                return;
5852
5853        /*
5854         * In case the memcg owning these pages has been offlined and doesn't
5855         * have an ID allocated to it anymore, charge the closest online
5856         * ancestor for the swap instead and transfer the memory+swap charge.
5857         */
5858        swap_memcg = mem_cgroup_id_get_online(memcg);
5859        oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg));
5860        VM_BUG_ON_PAGE(oldid, page);
5861        mem_cgroup_swap_statistics(swap_memcg, true);
5862
5863        page->mem_cgroup = NULL;
5864
5865        if (!mem_cgroup_is_root(memcg))
5866                page_counter_uncharge(&memcg->memory, 1);
5867
5868        if (memcg != swap_memcg) {
5869                if (!mem_cgroup_is_root(swap_memcg))
5870                        page_counter_charge(&swap_memcg->memsw, 1);
5871                page_counter_uncharge(&memcg->memsw, 1);
5872        }
5873
5874        /*
5875         * Interrupts should be disabled here because the caller holds the
5876         * mapping->tree_lock lock which is taken with interrupts-off. It is
5877         * important here to have the interrupts disabled because it is the
5878         * only synchronisation we have for udpating the per-CPU variables.
5879         */
5880        VM_BUG_ON(!irqs_disabled());
5881        mem_cgroup_charge_statistics(memcg, page, false, -1);
5882        memcg_check_events(memcg, page);
5883
5884        if (!mem_cgroup_is_root(memcg))
5885                css_put(&memcg->css);
5886}
5887
5888/*
5889 * mem_cgroup_try_charge_swap - try charging a swap entry
5890 * @page: page being added to swap
5891 * @entry: swap entry to charge
5892 *
5893 * Try to charge @entry to the memcg that @page belongs to.
5894 *
5895 * Returns 0 on success, -ENOMEM on failure.
5896 */
5897int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5898{
5899        struct mem_cgroup *memcg;
5900        struct page_counter *counter;
5901        unsigned short oldid;
5902
5903        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5904                return 0;
5905
5906        memcg = page->mem_cgroup;
5907
5908        /* Readahead page, never charged */
5909        if (!memcg)
5910                return 0;
5911
5912        memcg = mem_cgroup_id_get_online(memcg);
5913
5914        if (!mem_cgroup_is_root(memcg) &&
5915            !page_counter_try_charge(&memcg->swap, 1, &counter)) {
5916                mem_cgroup_id_put(memcg);
5917                return -ENOMEM;
5918        }
5919
5920        oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5921        VM_BUG_ON_PAGE(oldid, page);
5922        mem_cgroup_swap_statistics(memcg, true);
5923
5924        return 0;
5925}
5926
5927/**
5928 * mem_cgroup_uncharge_swap - uncharge a swap entry
5929 * @entry: swap entry to uncharge
5930 *
5931 * Drop the swap charge associated with @entry.
5932 */
5933void mem_cgroup_uncharge_swap(swp_entry_t entry)
5934{
5935        struct mem_cgroup *memcg;
5936        unsigned short id;
5937
5938        if (!do_swap_account)
5939                return;
5940
5941        id = swap_cgroup_record(entry, 0);
5942        rcu_read_lock();
5943        memcg = mem_cgroup_from_id(id);
5944        if (memcg) {
5945                if (!mem_cgroup_is_root(memcg)) {
5946                        if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5947                                page_counter_uncharge(&memcg->swap, 1);
5948                        else
5949                                page_counter_uncharge(&memcg->memsw, 1);
5950                }
5951                mem_cgroup_swap_statistics(memcg, false);
5952                mem_cgroup_id_put(memcg);
5953        }
5954        rcu_read_unlock();
5955}
5956
5957long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
5958{
5959        long nr_swap_pages = get_nr_swap_pages();
5960
5961        if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5962                return nr_swap_pages;
5963        for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5964                nr_swap_pages = min_t(long, nr_swap_pages,
5965                                      READ_ONCE(memcg->swap.limit) -
5966                                      page_counter_read(&memcg->swap));
5967        return nr_swap_pages;
5968}
5969
5970bool mem_cgroup_swap_full(struct page *page)
5971{
5972        struct mem_cgroup *memcg;
5973
5974        VM_BUG_ON_PAGE(!PageLocked(page), page);
5975
5976        if (vm_swap_full())
5977                return true;
5978        if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5979                return false;
5980
5981        memcg = page->mem_cgroup;
5982        if (!memcg)
5983                return false;
5984
5985        for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5986                if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
5987                        return true;
5988
5989        return false;
5990}
5991
5992/* for remember boot option*/
5993#ifdef CONFIG_MEMCG_SWAP_ENABLED
5994static int really_do_swap_account __initdata = 1;
5995#else
5996static int really_do_swap_account __initdata;
5997#endif
5998
5999static int __init enable_swap_account(char *s)
6000{
6001        if (!strcmp(s, "1"))
6002                really_do_swap_account = 1;
6003        else if (!strcmp(s, "0"))
6004                really_do_swap_account = 0;
6005        return 1;
6006}
6007__setup("swapaccount=", enable_swap_account);
6008
6009static u64 swap_current_read(struct cgroup_subsys_state *css,
6010                             struct cftype *cft)
6011{
6012        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6013
6014        return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
6015}
6016
6017static int swap_max_show(struct seq_file *m, void *v)
6018{
6019        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
6020        unsigned long max = READ_ONCE(memcg->swap.limit);
6021
6022        if (max == PAGE_COUNTER_MAX)
6023                seq_puts(m, "max\n");
6024        else
6025                seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
6026
6027        return 0;
6028}
6029
6030static ssize_t swap_max_write(struct kernfs_open_file *of,
6031                              char *buf, size_t nbytes, loff_t off)
6032{
6033        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6034        unsigned long max;
6035        int err;
6036
6037        buf = strstrip(buf);
6038        err = page_counter_memparse(buf, "max", &max);
6039        if (err)
6040                return err;
6041
6042        mutex_lock(&memcg_limit_mutex);
6043        err = page_counter_limit(&memcg->swap, max);
6044        mutex_unlock(&memcg_limit_mutex);
6045        if (err)
6046                return err;
6047
6048        return nbytes;
6049}
6050
6051static struct cftype swap_files[] = {
6052        {
6053                .name = "swap.current",
6054                .flags = CFTYPE_NOT_ON_ROOT,
6055                .read_u64 = swap_current_read,
6056        },
6057        {
6058                .name = "swap.max",
6059                .flags = CFTYPE_NOT_ON_ROOT,
6060                .seq_show = swap_max_show,
6061                .write = swap_max_write,
6062        },
6063        { }     /* terminate */
6064};
6065
6066static struct cftype memsw_cgroup_files[] = {
6067        {
6068                .name = "memsw.usage_in_bytes",
6069                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6070                .read_u64 = mem_cgroup_read_u64,
6071        },
6072        {
6073                .name = "memsw.max_usage_in_bytes",
6074                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6075                .write = mem_cgroup_reset,
6076                .read_u64 = mem_cgroup_read_u64,
6077        },
6078        {
6079                .name = "memsw.limit_in_bytes",
6080                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6081                .write = mem_cgroup_write,
6082                .read_u64 = mem_cgroup_read_u64,
6083        },
6084        {
6085                .name = "memsw.failcnt",
6086                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6087                .write = mem_cgroup_reset,
6088                .read_u64 = mem_cgroup_read_u64,
6089        },
6090        { },    /* terminate */
6091};
6092
6093static int __init mem_cgroup_swap_init(void)
6094{
6095        if (!mem_cgroup_disabled() && really_do_swap_account) {
6096                do_swap_account = 1;
6097                WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
6098                                               swap_files));
6099                WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
6100                                                  memsw_cgroup_files));
6101        }
6102        return 0;
6103}
6104subsys_initcall(mem_cgroup_swap_init);
6105
6106#endif /* CONFIG_MEMCG_SWAP */
Note: See TracBrowser for help on using the repository browser.