source: src/linux/universal/linux-4.9/mm/memcontrol.c @ 31885

Last change on this file since 31885 was 31885, checked in by brainslayer, 2 months ago

update

File size: 154.7 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 struct workqueue_struct *memcg_kmem_cache_create_wq;
2156
2157static void memcg_kmem_cache_create_func(struct work_struct *w)
2158{
2159        struct memcg_kmem_cache_create_work *cw =
2160                container_of(w, struct memcg_kmem_cache_create_work, work);
2161        struct mem_cgroup *memcg = cw->memcg;
2162        struct kmem_cache *cachep = cw->cachep;
2163
2164        memcg_create_kmem_cache(memcg, cachep);
2165
2166        css_put(&memcg->css);
2167        kfree(cw);
2168}
2169
2170/*
2171 * Enqueue the creation of a per-memcg kmem_cache.
2172 */
2173static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2174                                               struct kmem_cache *cachep)
2175{
2176        struct memcg_kmem_cache_create_work *cw;
2177
2178        cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
2179        if (!cw)
2180                return;
2181
2182        css_get(&memcg->css);
2183
2184        cw->memcg = memcg;
2185        cw->cachep = cachep;
2186        INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
2187
2188        queue_work(memcg_kmem_cache_create_wq, &cw->work);
2189}
2190
2191static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2192                                             struct kmem_cache *cachep)
2193{
2194        /*
2195         * We need to stop accounting when we kmalloc, because if the
2196         * corresponding kmalloc cache is not yet created, the first allocation
2197         * in __memcg_schedule_kmem_cache_create will recurse.
2198         *
2199         * However, it is better to enclose the whole function. Depending on
2200         * the debugging options enabled, INIT_WORK(), for instance, can
2201         * trigger an allocation. This too, will make us recurse. Because at
2202         * this point we can't allow ourselves back into memcg_kmem_get_cache,
2203         * the safest choice is to do it like this, wrapping the whole function.
2204         */
2205        current->memcg_kmem_skip_account = 1;
2206        __memcg_schedule_kmem_cache_create(memcg, cachep);
2207        current->memcg_kmem_skip_account = 0;
2208}
2209
2210static inline bool memcg_kmem_bypass(void)
2211{
2212        if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
2213                return true;
2214        return false;
2215}
2216
2217/**
2218 * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2219 * @cachep: the original global kmem cache
2220 *
2221 * Return the kmem_cache we're supposed to use for a slab allocation.
2222 * We try to use the current memcg's version of the cache.
2223 *
2224 * If the cache does not exist yet, if we are the first user of it, we
2225 * create it asynchronously in a workqueue and let the current allocation
2226 * go through with the original cache.
2227 *
2228 * This function takes a reference to the cache it returns to assure it
2229 * won't get destroyed while we are working with it. Once the caller is
2230 * done with it, memcg_kmem_put_cache() must be called to release the
2231 * reference.
2232 */
2233struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep)
2234{
2235        struct mem_cgroup *memcg;
2236        struct kmem_cache *memcg_cachep;
2237        int kmemcg_id;
2238
2239        VM_BUG_ON(!is_root_cache(cachep));
2240
2241        if (memcg_kmem_bypass())
2242                return cachep;
2243
2244        if (current->memcg_kmem_skip_account)
2245                return cachep;
2246
2247        memcg = get_mem_cgroup_from_mm(current->mm);
2248        kmemcg_id = READ_ONCE(memcg->kmemcg_id);
2249        if (kmemcg_id < 0)
2250                goto out;
2251
2252        memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
2253        if (likely(memcg_cachep))
2254                return memcg_cachep;
2255
2256        /*
2257         * If we are in a safe context (can wait, and not in interrupt
2258         * context), we could be be predictable and return right away.
2259         * This would guarantee that the allocation being performed
2260         * already belongs in the new cache.
2261         *
2262         * However, there are some clashes that can arrive from locking.
2263         * For instance, because we acquire the slab_mutex while doing
2264         * memcg_create_kmem_cache, this means no further allocation
2265         * could happen with the slab_mutex held. So it's better to
2266         * defer everything.
2267         */
2268        memcg_schedule_kmem_cache_create(memcg, cachep);
2269out:
2270        css_put(&memcg->css);
2271        return cachep;
2272}
2273
2274/**
2275 * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2276 * @cachep: the cache returned by memcg_kmem_get_cache
2277 */
2278void memcg_kmem_put_cache(struct kmem_cache *cachep)
2279{
2280        if (!is_root_cache(cachep))
2281                css_put(&cachep->memcg_params.memcg->css);
2282}
2283
2284/**
2285 * memcg_kmem_charge: charge a kmem page
2286 * @page: page to charge
2287 * @gfp: reclaim mode
2288 * @order: allocation order
2289 * @memcg: memory cgroup to charge
2290 *
2291 * Returns 0 on success, an error code on failure.
2292 */
2293int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2294                            struct mem_cgroup *memcg)
2295{
2296        unsigned int nr_pages = 1 << order;
2297        struct page_counter *counter;
2298        int ret;
2299
2300        ret = try_charge(memcg, gfp, nr_pages);
2301        if (ret)
2302                return ret;
2303
2304        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2305            !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2306                cancel_charge(memcg, nr_pages);
2307                return -ENOMEM;
2308        }
2309
2310        page->mem_cgroup = memcg;
2311
2312        return 0;
2313}
2314
2315/**
2316 * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2317 * @page: page to charge
2318 * @gfp: reclaim mode
2319 * @order: allocation order
2320 *
2321 * Returns 0 on success, an error code on failure.
2322 */
2323int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
2324{
2325        struct mem_cgroup *memcg;
2326        int ret = 0;
2327
2328        if (memcg_kmem_bypass())
2329                return 0;
2330
2331        memcg = get_mem_cgroup_from_mm(current->mm);
2332        if (!mem_cgroup_is_root(memcg)) {
2333                ret = memcg_kmem_charge_memcg(page, gfp, order, memcg);
2334                if (!ret)
2335                        __SetPageKmemcg(page);
2336        }
2337        css_put(&memcg->css);
2338        return ret;
2339}
2340/**
2341 * memcg_kmem_uncharge: uncharge a kmem page
2342 * @page: page to uncharge
2343 * @order: allocation order
2344 */
2345void memcg_kmem_uncharge(struct page *page, int order)
2346{
2347        struct mem_cgroup *memcg = page->mem_cgroup;
2348        unsigned int nr_pages = 1 << order;
2349
2350        if (!memcg)
2351                return;
2352
2353        VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
2354
2355        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2356                page_counter_uncharge(&memcg->kmem, nr_pages);
2357
2358        page_counter_uncharge(&memcg->memory, nr_pages);
2359        if (do_memsw_account())
2360                page_counter_uncharge(&memcg->memsw, nr_pages);
2361
2362        page->mem_cgroup = NULL;
2363
2364        /* slab pages do not have PageKmemcg flag set */
2365        if (PageKmemcg(page))
2366                __ClearPageKmemcg(page);
2367
2368        css_put_many(&memcg->css, nr_pages);
2369}
2370#endif /* !CONFIG_SLOB */
2371
2372#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2373
2374/*
2375 * Because tail pages are not marked as "used", set it. We're under
2376 * zone_lru_lock and migration entries setup in all page mappings.
2377 */
2378void mem_cgroup_split_huge_fixup(struct page *head)
2379{
2380        int i;
2381
2382        if (mem_cgroup_disabled())
2383                return;
2384
2385        for (i = 1; i < HPAGE_PMD_NR; i++)
2386                head[i].mem_cgroup = head->mem_cgroup;
2387
2388        __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
2389                       HPAGE_PMD_NR);
2390}
2391#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2392
2393#ifdef CONFIG_MEMCG_SWAP
2394static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2395                                         bool charge)
2396{
2397        int val = (charge) ? 1 : -1;
2398        this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
2399}
2400
2401/**
2402 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2403 * @entry: swap entry to be moved
2404 * @from:  mem_cgroup which the entry is moved from
2405 * @to:  mem_cgroup which the entry is moved to
2406 *
2407 * It succeeds only when the swap_cgroup's record for this entry is the same
2408 * as the mem_cgroup's id of @from.
2409 *
2410 * Returns 0 on success, -EINVAL on failure.
2411 *
2412 * The caller must have charged to @to, IOW, called page_counter_charge() about
2413 * both res and memsw, and called css_get().
2414 */
2415static int mem_cgroup_move_swap_account(swp_entry_t entry,
2416                                struct mem_cgroup *from, struct mem_cgroup *to)
2417{
2418        unsigned short old_id, new_id;
2419
2420        old_id = mem_cgroup_id(from);
2421        new_id = mem_cgroup_id(to);
2422
2423        if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
2424                mem_cgroup_swap_statistics(from, false);
2425                mem_cgroup_swap_statistics(to, true);
2426                return 0;
2427        }
2428        return -EINVAL;
2429}
2430#else
2431static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
2432                                struct mem_cgroup *from, struct mem_cgroup *to)
2433{
2434        return -EINVAL;
2435}
2436#endif
2437
2438static DEFINE_MUTEX(memcg_limit_mutex);
2439
2440static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
2441                                   unsigned long limit)
2442{
2443        unsigned long curusage;
2444        unsigned long oldusage;
2445        bool enlarge = false;
2446        int retry_count;
2447        int ret;
2448
2449        /*
2450         * For keeping hierarchical_reclaim simple, how long we should retry
2451         * is depends on callers. We set our retry-count to be function
2452         * of # of children which we should visit in this loop.
2453         */
2454        retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2455                      mem_cgroup_count_children(memcg);
2456
2457        oldusage = page_counter_read(&memcg->memory);
2458
2459        do {
2460                if (signal_pending(current)) {
2461                        ret = -EINTR;
2462                        break;
2463                }
2464
2465                mutex_lock(&memcg_limit_mutex);
2466                if (limit > memcg->memsw.limit) {
2467                        mutex_unlock(&memcg_limit_mutex);
2468                        ret = -EINVAL;
2469                        break;
2470                }
2471                if (limit > memcg->memory.limit)
2472                        enlarge = true;
2473                ret = page_counter_limit(&memcg->memory, limit);
2474                mutex_unlock(&memcg_limit_mutex);
2475
2476                if (!ret)
2477                        break;
2478
2479                try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2480
2481                curusage = page_counter_read(&memcg->memory);
2482                /* Usage is reduced ? */
2483                if (curusage >= oldusage)
2484                        retry_count--;
2485                else
2486                        oldusage = curusage;
2487        } while (retry_count);
2488
2489        if (!ret && enlarge)
2490                memcg_oom_recover(memcg);
2491
2492        return ret;
2493}
2494
2495static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2496                                         unsigned long limit)
2497{
2498        unsigned long curusage;
2499        unsigned long oldusage;
2500        bool enlarge = false;
2501        int retry_count;
2502        int ret;
2503
2504        /* see mem_cgroup_resize_res_limit */
2505        retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2506                      mem_cgroup_count_children(memcg);
2507
2508        oldusage = page_counter_read(&memcg->memsw);
2509
2510        do {
2511                if (signal_pending(current)) {
2512                        ret = -EINTR;
2513                        break;
2514                }
2515
2516                mutex_lock(&memcg_limit_mutex);
2517                if (limit < memcg->memory.limit) {
2518                        mutex_unlock(&memcg_limit_mutex);
2519                        ret = -EINVAL;
2520                        break;
2521                }
2522                if (limit > memcg->memsw.limit)
2523                        enlarge = true;
2524                ret = page_counter_limit(&memcg->memsw, limit);
2525                mutex_unlock(&memcg_limit_mutex);
2526
2527                if (!ret)
2528                        break;
2529
2530                try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2531
2532                curusage = page_counter_read(&memcg->memsw);
2533                /* Usage is reduced ? */
2534                if (curusage >= oldusage)
2535                        retry_count--;
2536                else
2537                        oldusage = curusage;
2538        } while (retry_count);
2539
2540        if (!ret && enlarge)
2541                memcg_oom_recover(memcg);
2542
2543        return ret;
2544}
2545
2546unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
2547                                            gfp_t gfp_mask,
2548                                            unsigned long *total_scanned)
2549{
2550        unsigned long nr_reclaimed = 0;
2551        struct mem_cgroup_per_node *mz, *next_mz = NULL;
2552        unsigned long reclaimed;
2553        int loop = 0;
2554        struct mem_cgroup_tree_per_node *mctz;
2555        unsigned long excess;
2556        unsigned long nr_scanned;
2557
2558        if (order > 0)
2559                return 0;
2560
2561        mctz = soft_limit_tree_node(pgdat->node_id);
2562
2563        /*
2564         * Do not even bother to check the largest node if the root
2565         * is empty. Do it lockless to prevent lock bouncing. Races
2566         * are acceptable as soft limit is best effort anyway.
2567         */
2568        if (RB_EMPTY_ROOT(&mctz->rb_root))
2569                return 0;
2570
2571        /*
2572         * This loop can run a while, specially if mem_cgroup's continuously
2573         * keep exceeding their soft limit and putting the system under
2574         * pressure
2575         */
2576        do {
2577                if (next_mz)
2578                        mz = next_mz;
2579                else
2580                        mz = mem_cgroup_largest_soft_limit_node(mctz);
2581                if (!mz)
2582                        break;
2583
2584                nr_scanned = 0;
2585                reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
2586                                                    gfp_mask, &nr_scanned);
2587                nr_reclaimed += reclaimed;
2588                *total_scanned += nr_scanned;
2589                spin_lock_irq(&mctz->lock);
2590                __mem_cgroup_remove_exceeded(mz, mctz);
2591
2592                /*
2593                 * If we failed to reclaim anything from this memory cgroup
2594                 * it is time to move on to the next cgroup
2595                 */
2596                next_mz = NULL;
2597                if (!reclaimed)
2598                        next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2599
2600                excess = soft_limit_excess(mz->memcg);
2601                /*
2602                 * One school of thought says that we should not add
2603                 * back the node to the tree if reclaim returns 0.
2604                 * But our reclaim could return 0, simply because due
2605                 * to priority we are exposing a smaller subset of
2606                 * memory to reclaim from. Consider this as a longer
2607                 * term TODO.
2608                 */
2609                /* If excess == 0, no tree ops */
2610                __mem_cgroup_insert_exceeded(mz, mctz, excess);
2611                spin_unlock_irq(&mctz->lock);
2612                css_put(&mz->memcg->css);
2613                loop++;
2614                /*
2615                 * Could not reclaim anything and there are no more
2616                 * mem cgroups to try or we seem to be looping without
2617                 * reclaiming anything.
2618                 */
2619                if (!nr_reclaimed &&
2620                        (next_mz == NULL ||
2621                        loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2622                        break;
2623        } while (!nr_reclaimed);
2624        if (next_mz)
2625                css_put(&next_mz->memcg->css);
2626        return nr_reclaimed;
2627}
2628
2629/*
2630 * Test whether @memcg has children, dead or alive.  Note that this
2631 * function doesn't care whether @memcg has use_hierarchy enabled and
2632 * returns %true if there are child csses according to the cgroup
2633 * hierarchy.  Testing use_hierarchy is the caller's responsiblity.
2634 */
2635static inline bool memcg_has_children(struct mem_cgroup *memcg)
2636{
2637        bool ret;
2638
2639        rcu_read_lock();
2640        ret = css_next_child(NULL, &memcg->css);
2641        rcu_read_unlock();
2642        return ret;
2643}
2644
2645/*
2646 * Reclaims as many pages from the given memcg as possible.
2647 *
2648 * Caller is responsible for holding css reference for memcg.
2649 */
2650static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2651{
2652        int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
2653
2654        /* we call try-to-free pages for make this cgroup empty */
2655        lru_add_drain_all();
2656        /* try to free all pages in this cgroup */
2657        while (nr_retries && page_counter_read(&memcg->memory)) {
2658                int progress;
2659
2660                if (signal_pending(current))
2661                        return -EINTR;
2662
2663                progress = try_to_free_mem_cgroup_pages(memcg, 1,
2664                                                        GFP_KERNEL, true);
2665                if (!progress) {
2666                        nr_retries--;
2667                        /* maybe some writeback is necessary */
2668                        congestion_wait(BLK_RW_ASYNC, HZ/10);
2669                }
2670
2671        }
2672
2673        return 0;
2674}
2675
2676static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2677                                            char *buf, size_t nbytes,
2678                                            loff_t off)
2679{
2680        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
2681
2682        if (mem_cgroup_is_root(memcg))
2683                return -EINVAL;
2684        return mem_cgroup_force_empty(memcg) ?: nbytes;
2685}
2686
2687static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2688                                     struct cftype *cft)
2689{
2690        return mem_cgroup_from_css(css)->use_hierarchy;
2691}
2692
2693static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2694                                      struct cftype *cft, u64 val)
2695{
2696        int retval = 0;
2697        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
2698        struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
2699
2700        if (memcg->use_hierarchy == val)
2701                return 0;
2702
2703        /*
2704         * If parent's use_hierarchy is set, we can't make any modifications
2705         * in the child subtrees. If it is unset, then the change can
2706         * occur, provided the current cgroup has no children.
2707         *
2708         * For the root cgroup, parent_mem is NULL, we allow value to be
2709         * set if there are no children.
2710         */
2711        if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
2712                                (val == 1 || val == 0)) {
2713                if (!memcg_has_children(memcg))
2714                        memcg->use_hierarchy = val;
2715                else
2716                        retval = -EBUSY;
2717        } else
2718                retval = -EINVAL;
2719
2720        return retval;
2721}
2722
2723static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
2724{
2725        struct mem_cgroup *iter;
2726        int i;
2727
2728        memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
2729
2730        for_each_mem_cgroup_tree(iter, memcg) {
2731                for (i = 0; i < MEMCG_NR_STAT; i++)
2732                        stat[i] += mem_cgroup_read_stat(iter, i);
2733        }
2734}
2735
2736static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
2737{
2738        struct mem_cgroup *iter;
2739        int i;
2740
2741        memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
2742
2743        for_each_mem_cgroup_tree(iter, memcg) {
2744                for (i = 0; i < MEMCG_NR_EVENTS; i++)
2745                        events[i] += mem_cgroup_read_events(iter, i);
2746        }
2747}
2748
2749static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
2750{
2751        unsigned long val = 0;
2752
2753        if (mem_cgroup_is_root(memcg)) {
2754                struct mem_cgroup *iter;
2755
2756                for_each_mem_cgroup_tree(iter, memcg) {
2757                        val += mem_cgroup_read_stat(iter,
2758                                        MEM_CGROUP_STAT_CACHE);
2759                        val += mem_cgroup_read_stat(iter,
2760                                        MEM_CGROUP_STAT_RSS);
2761                        if (swap)
2762                                val += mem_cgroup_read_stat(iter,
2763                                                MEM_CGROUP_STAT_SWAP);
2764                }
2765        } else {
2766                if (!swap)
2767                        val = page_counter_read(&memcg->memory);
2768                else
2769                        val = page_counter_read(&memcg->memsw);
2770        }
2771        return val;
2772}
2773
2774enum {
2775        RES_USAGE,
2776        RES_LIMIT,
2777        RES_MAX_USAGE,
2778        RES_FAILCNT,
2779        RES_SOFT_LIMIT,
2780};
2781
2782static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
2783                               struct cftype *cft)
2784{
2785        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
2786        struct page_counter *counter;
2787
2788        switch (MEMFILE_TYPE(cft->private)) {
2789        case _MEM:
2790                counter = &memcg->memory;
2791                break;
2792        case _MEMSWAP:
2793                counter = &memcg->memsw;
2794                break;
2795        case _KMEM:
2796                counter = &memcg->kmem;
2797                break;
2798        case _TCP:
2799                counter = &memcg->tcpmem;
2800                break;
2801        default:
2802                BUG();
2803        }
2804
2805        switch (MEMFILE_ATTR(cft->private)) {
2806        case RES_USAGE:
2807                if (counter == &memcg->memory)
2808                        return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
2809                if (counter == &memcg->memsw)
2810                        return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
2811                return (u64)page_counter_read(counter) * PAGE_SIZE;
2812        case RES_LIMIT:
2813                return (u64)counter->limit * PAGE_SIZE;
2814        case RES_MAX_USAGE:
2815                return (u64)counter->watermark * PAGE_SIZE;
2816        case RES_FAILCNT:
2817                return counter->failcnt;
2818        case RES_SOFT_LIMIT:
2819                return (u64)memcg->soft_limit * PAGE_SIZE;
2820        default:
2821                BUG();
2822        }
2823}
2824
2825#ifndef CONFIG_SLOB
2826static int memcg_online_kmem(struct mem_cgroup *memcg)
2827{
2828        int memcg_id;
2829
2830        if (cgroup_memory_nokmem)
2831                return 0;
2832
2833        BUG_ON(memcg->kmemcg_id >= 0);
2834        BUG_ON(memcg->kmem_state);
2835
2836        memcg_id = memcg_alloc_cache_id();
2837        if (memcg_id < 0)
2838                return memcg_id;
2839
2840        static_branch_inc(&memcg_kmem_enabled_key);
2841        /*
2842         * A memory cgroup is considered kmem-online as soon as it gets
2843         * kmemcg_id. Setting the id after enabling static branching will
2844         * guarantee no one starts accounting before all call sites are
2845         * patched.
2846         */
2847        memcg->kmemcg_id = memcg_id;
2848        memcg->kmem_state = KMEM_ONLINE;
2849
2850        return 0;
2851}
2852
2853static void memcg_offline_kmem(struct mem_cgroup *memcg)
2854{
2855        struct cgroup_subsys_state *css;
2856        struct mem_cgroup *parent, *child;
2857        int kmemcg_id;
2858
2859        if (memcg->kmem_state != KMEM_ONLINE)
2860                return;
2861        /*
2862         * Clear the online state before clearing memcg_caches array
2863         * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2864         * guarantees that no cache will be created for this cgroup
2865         * after we are done (see memcg_create_kmem_cache()).
2866         */
2867        memcg->kmem_state = KMEM_ALLOCATED;
2868
2869        memcg_deactivate_kmem_caches(memcg);
2870
2871        kmemcg_id = memcg->kmemcg_id;
2872        BUG_ON(kmemcg_id < 0);
2873
2874        parent = parent_mem_cgroup(memcg);
2875        if (!parent)
2876                parent = root_mem_cgroup;
2877
2878        /*
2879         * Change kmemcg_id of this cgroup and all its descendants to the
2880         * parent's id, and then move all entries from this cgroup's list_lrus
2881         * to ones of the parent. After we have finished, all list_lrus
2882         * corresponding to this cgroup are guaranteed to remain empty. The
2883         * ordering is imposed by list_lru_node->lock taken by
2884         * memcg_drain_all_list_lrus().
2885         */
2886        rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
2887        css_for_each_descendant_pre(css, &memcg->css) {
2888                child = mem_cgroup_from_css(css);
2889                BUG_ON(child->kmemcg_id != kmemcg_id);
2890                child->kmemcg_id = parent->kmemcg_id;
2891                if (!memcg->use_hierarchy)
2892                        break;
2893        }
2894        rcu_read_unlock();
2895
2896        memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2897
2898        memcg_free_cache_id(kmemcg_id);
2899}
2900
2901static void memcg_free_kmem(struct mem_cgroup *memcg)
2902{
2903        /* css_alloc() failed, offlining didn't happen */
2904        if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2905                memcg_offline_kmem(memcg);
2906
2907        if (memcg->kmem_state == KMEM_ALLOCATED) {
2908                memcg_destroy_kmem_caches(memcg);
2909                static_branch_dec(&memcg_kmem_enabled_key);
2910                WARN_ON(page_counter_read(&memcg->kmem));
2911        }
2912}
2913#else
2914static int memcg_online_kmem(struct mem_cgroup *memcg)
2915{
2916        return 0;
2917}
2918static void memcg_offline_kmem(struct mem_cgroup *memcg)
2919{
2920}
2921static void memcg_free_kmem(struct mem_cgroup *memcg)
2922{
2923}
2924#endif /* !CONFIG_SLOB */
2925
2926static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2927                                   unsigned long limit)
2928{
2929        int ret;
2930
2931        mutex_lock(&memcg_limit_mutex);
2932        ret = page_counter_limit(&memcg->kmem, limit);
2933        mutex_unlock(&memcg_limit_mutex);
2934        return ret;
2935}
2936
2937static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2938{
2939        int ret;
2940
2941        mutex_lock(&memcg_limit_mutex);
2942
2943        ret = page_counter_limit(&memcg->tcpmem, limit);
2944        if (ret)
2945                goto out;
2946
2947        if (!memcg->tcpmem_active) {
2948                /*
2949                 * The active flag needs to be written after the static_key
2950                 * update. This is what guarantees that the socket activation
2951                 * function is the last one to run. See mem_cgroup_sk_alloc()
2952                 * for details, and note that we don't mark any socket as
2953                 * belonging to this memcg until that flag is up.
2954                 *
2955                 * We need to do this, because static_keys will span multiple
2956                 * sites, but we can't control their order. If we mark a socket
2957                 * as accounted, but the accounting functions are not patched in
2958                 * yet, we'll lose accounting.
2959                 *
2960                 * We never race with the readers in mem_cgroup_sk_alloc(),
2961                 * because when this value change, the code to process it is not
2962                 * patched in yet.
2963                 */
2964                static_branch_inc(&memcg_sockets_enabled_key);
2965                memcg->tcpmem_active = true;
2966        }
2967out:
2968        mutex_unlock(&memcg_limit_mutex);
2969        return ret;
2970}
2971
2972/*
2973 * The user of this function is...
2974 * RES_LIMIT.
2975 */
2976static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2977                                char *buf, size_t nbytes, loff_t off)
2978{
2979        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
2980        unsigned long nr_pages;
2981        int ret;
2982
2983        buf = strstrip(buf);
2984        ret = page_counter_memparse(buf, "-1", &nr_pages);
2985        if (ret)
2986                return ret;
2987
2988        switch (MEMFILE_ATTR(of_cft(of)->private)) {
2989        case RES_LIMIT:
2990                if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2991                        ret = -EINVAL;
2992                        break;
2993                }
2994                switch (MEMFILE_TYPE(of_cft(of)->private)) {
2995                case _MEM:
2996                        ret = mem_cgroup_resize_limit(memcg, nr_pages);
2997                        break;
2998                case _MEMSWAP:
2999                        ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3000                        break;
3001                case _KMEM:
3002                        ret = memcg_update_kmem_limit(memcg, nr_pages);
3003                        break;
3004                case _TCP:
3005                        ret = memcg_update_tcp_limit(memcg, nr_pages);
3006                        break;
3007                }
3008                break;
3009        case RES_SOFT_LIMIT:
3010                memcg->soft_limit = nr_pages;
3011                ret = 0;
3012                break;
3013        }
3014        return ret ?: nbytes;
3015}
3016
3017static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3018                                size_t nbytes, loff_t off)
3019{
3020        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
3021        struct page_counter *counter;
3022
3023        switch (MEMFILE_TYPE(of_cft(of)->private)) {
3024        case _MEM:
3025                counter = &memcg->memory;
3026                break;
3027        case _MEMSWAP:
3028                counter = &memcg->memsw;
3029                break;
3030        case _KMEM:
3031                counter = &memcg->kmem;
3032                break;
3033        case _TCP:
3034                counter = &memcg->tcpmem;
3035                break;
3036        default:
3037                BUG();
3038        }
3039
3040        switch (MEMFILE_ATTR(of_cft(of)->private)) {
3041        case RES_MAX_USAGE:
3042                page_counter_reset_watermark(counter);
3043                break;
3044        case RES_FAILCNT:
3045                counter->failcnt = 0;
3046                break;
3047        default:
3048                BUG();
3049        }
3050
3051        return nbytes;
3052}
3053
3054static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
3055                                        struct cftype *cft)
3056{
3057        return mem_cgroup_from_css(css)->move_charge_at_immigrate;
3058}
3059
3060#ifdef CONFIG_MMU
3061static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
3062                                        struct cftype *cft, u64 val)
3063{
3064        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3065
3066        if (val & ~MOVE_MASK)
3067                return -EINVAL;
3068
3069        /*
3070         * No kind of locking is needed in here, because ->can_attach() will
3071         * check this value once in the beginning of the process, and then carry
3072         * on with stale data. This means that changes to this value will only
3073         * affect task migrations starting after the change.
3074         */
3075        memcg->move_charge_at_immigrate = val;
3076        return 0;
3077}
3078#else
3079static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
3080                                        struct cftype *cft, u64 val)
3081{
3082        return -ENOSYS;
3083}
3084#endif
3085
3086#ifdef CONFIG_NUMA
3087static int memcg_numa_stat_show(struct seq_file *m, void *v)
3088{
3089        struct numa_stat {
3090                const char *name;
3091                unsigned int lru_mask;
3092        };
3093
3094        static const struct numa_stat stats[] = {
3095                { "total", LRU_ALL },
3096                { "file", LRU_ALL_FILE },
3097                { "anon", LRU_ALL_ANON },
3098                { "unevictable", BIT(LRU_UNEVICTABLE) },
3099        };
3100        const struct numa_stat *stat;
3101        int nid;
3102        unsigned long nr;
3103        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
3104
3105        for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3106                nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3107                seq_printf(m, "%s=%lu", stat->name, nr);
3108                for_each_node_state(nid, N_MEMORY) {
3109                        nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3110                                                          stat->lru_mask);
3111                        seq_printf(m, " N%d=%lu", nid, nr);
3112                }
3113                seq_putc(m, '\n');
3114        }
3115
3116        for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3117                struct mem_cgroup *iter;
3118
3119                nr = 0;
3120                for_each_mem_cgroup_tree(iter, memcg)
3121                        nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3122                seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3123                for_each_node_state(nid, N_MEMORY) {
3124                        nr = 0;
3125                        for_each_mem_cgroup_tree(iter, memcg)
3126                                nr += mem_cgroup_node_nr_lru_pages(
3127                                        iter, nid, stat->lru_mask);
3128                        seq_printf(m, " N%d=%lu", nid, nr);
3129                }
3130                seq_putc(m, '\n');
3131        }
3132
3133        return 0;
3134}
3135#endif /* CONFIG_NUMA */
3136
3137static int memcg_stat_show(struct seq_file *m, void *v)
3138{
3139        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
3140        unsigned long memory, memsw;
3141        struct mem_cgroup *mi;
3142        unsigned int i;
3143
3144        BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3145                     MEM_CGROUP_STAT_NSTATS);
3146        BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3147                     MEM_CGROUP_EVENTS_NSTATS);
3148        BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3149
3150        for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
3151                if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
3152                        continue;
3153                seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
3154                           mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
3155        }
3156
3157        for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3158                seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3159                           mem_cgroup_read_events(memcg, i));
3160
3161        for (i = 0; i < NR_LRU_LISTS; i++)
3162                seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3163                           mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3164
3165        /* Hierarchical information */
3166        memory = memsw = PAGE_COUNTER_MAX;
3167        for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3168                memory = min(memory, mi->memory.limit);
3169                memsw = min(memsw, mi->memsw.limit);
3170        }
3171        seq_printf(m, "hierarchical_memory_limit %llu\n",
3172                   (u64)memory * PAGE_SIZE);
3173        if (do_memsw_account())
3174                seq_printf(m, "hierarchical_memsw_limit %llu\n",
3175                           (u64)memsw * PAGE_SIZE);
3176
3177        for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
3178                unsigned long long val = 0;
3179
3180                if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
3181                        continue;
3182                for_each_mem_cgroup_tree(mi, memcg)
3183                        val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
3184                seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
3185        }
3186
3187        for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3188                unsigned long long val = 0;
3189
3190                for_each_mem_cgroup_tree(mi, memcg)
3191                        val += mem_cgroup_read_events(mi, i);
3192                seq_printf(m, "total_%s %llu\n",
3193                           mem_cgroup_events_names[i], val);
3194        }
3195
3196        for (i = 0; i < NR_LRU_LISTS; i++) {
3197                unsigned long long val = 0;
3198
3199                for_each_mem_cgroup_tree(mi, memcg)
3200                        val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3201                seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
3202        }
3203
3204#ifdef CONFIG_DEBUG_VM
3205        {
3206                pg_data_t *pgdat;
3207                struct mem_cgroup_per_node *mz;
3208                struct zone_reclaim_stat *rstat;
3209                unsigned long recent_rotated[2] = {0, 0};
3210                unsigned long recent_scanned[2] = {0, 0};
3211
3212                for_each_online_pgdat(pgdat) {
3213                        mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
3214                        rstat = &mz->lruvec.reclaim_stat;
3215
3216                        recent_rotated[0] += rstat->recent_rotated[0];
3217                        recent_rotated[1] += rstat->recent_rotated[1];
3218                        recent_scanned[0] += rstat->recent_scanned[0];
3219                        recent_scanned[1] += rstat->recent_scanned[1];
3220                }
3221                seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3222                seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3223                seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3224                seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
3225        }
3226#endif
3227
3228        return 0;
3229}
3230
3231static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3232                                      struct cftype *cft)
3233{
3234        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3235
3236        return mem_cgroup_swappiness(memcg);
3237}
3238
3239static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3240                                       struct cftype *cft, u64 val)
3241{
3242        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3243
3244        if (val > 100)
3245                return -EINVAL;
3246
3247        if (css->parent)
3248                memcg->swappiness = val;
3249        else
3250                vm_swappiness = val;
3251
3252        return 0;
3253}
3254
3255static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3256{
3257        struct mem_cgroup_threshold_ary *t;
3258        unsigned long usage;
3259        int i;
3260
3261        rcu_read_lock();
3262        if (!swap)
3263                t = rcu_dereference(memcg->thresholds.primary);
3264        else
3265                t = rcu_dereference(memcg->memsw_thresholds.primary);
3266
3267        if (!t)
3268                goto unlock;
3269
3270        usage = mem_cgroup_usage(memcg, swap);
3271
3272        /*
3273         * current_threshold points to threshold just below or equal to usage.
3274         * If it's not true, a threshold was crossed after last
3275         * call of __mem_cgroup_threshold().
3276         */
3277        i = t->current_threshold;
3278
3279        /*
3280         * Iterate backward over array of thresholds starting from
3281         * current_threshold and check if a threshold is crossed.
3282         * If none of thresholds below usage is crossed, we read
3283         * only one element of the array here.
3284         */
3285        for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3286                eventfd_signal(t->entries[i].eventfd, 1);
3287
3288        /* i = current_threshold + 1 */
3289        i++;
3290
3291        /*
3292         * Iterate forward over array of thresholds starting from
3293         * current_threshold+1 and check if a threshold is crossed.
3294         * If none of thresholds above usage is crossed, we read
3295         * only one element of the array here.
3296         */
3297        for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3298                eventfd_signal(t->entries[i].eventfd, 1);
3299
3300        /* Update current_threshold */
3301        t->current_threshold = i - 1;
3302unlock:
3303        rcu_read_unlock();
3304}
3305
3306static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3307{
3308        while (memcg) {
3309                __mem_cgroup_threshold(memcg, false);
3310                if (do_memsw_account())
3311                        __mem_cgroup_threshold(memcg, true);
3312
3313                memcg = parent_mem_cgroup(memcg);
3314        }
3315}
3316
3317static int compare_thresholds(const void *a, const void *b)
3318{
3319        const struct mem_cgroup_threshold *_a = a;
3320        const struct mem_cgroup_threshold *_b = b;
3321
3322        if (_a->threshold > _b->threshold)
3323                return 1;
3324
3325        if (_a->threshold < _b->threshold)
3326                return -1;
3327
3328        return 0;
3329}
3330
3331static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
3332{
3333        struct mem_cgroup_eventfd_list *ev;
3334
3335        spin_lock(&memcg_oom_lock);
3336
3337        list_for_each_entry(ev, &memcg->oom_notify, list)
3338                eventfd_signal(ev->eventfd, 1);
3339
3340        spin_unlock(&memcg_oom_lock);
3341        return 0;
3342}
3343
3344static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
3345{
3346        struct mem_cgroup *iter;
3347
3348        for_each_mem_cgroup_tree(iter, memcg)
3349                mem_cgroup_oom_notify_cb(iter);
3350}
3351
3352static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
3353        struct eventfd_ctx *eventfd, const char *args, enum res_type type)
3354{
3355        struct mem_cgroup_thresholds *thresholds;
3356        struct mem_cgroup_threshold_ary *new;
3357        unsigned long threshold;
3358        unsigned long usage;
3359        int i, size, ret;
3360
3361        ret = page_counter_memparse(args, "-1", &threshold);
3362        if (ret)
3363                return ret;
3364
3365        mutex_lock(&memcg->thresholds_lock);
3366
3367        if (type == _MEM) {
3368                thresholds = &memcg->thresholds;
3369                usage = mem_cgroup_usage(memcg, false);
3370        } else if (type == _MEMSWAP) {
3371                thresholds = &memcg->memsw_thresholds;
3372                usage = mem_cgroup_usage(memcg, true);
3373        } else
3374                BUG();
3375
3376        /* Check if a threshold crossed before adding a new one */
3377        if (thresholds->primary)
3378                __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3379
3380        size = thresholds->primary ? thresholds->primary->size + 1 : 1;
3381
3382        /* Allocate memory for new array of thresholds */
3383        new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
3384                        GFP_KERNEL);
3385        if (!new) {
3386                ret = -ENOMEM;
3387                goto unlock;
3388        }
3389        new->size = size;
3390
3391        /* Copy thresholds (if any) to new array */
3392        if (thresholds->primary) {
3393                memcpy(new->entries, thresholds->primary->entries, (size - 1) *
3394                                sizeof(struct mem_cgroup_threshold));
3395        }
3396
3397        /* Add new threshold */
3398        new->entries[size - 1].eventfd = eventfd;
3399        new->entries[size - 1].threshold = threshold;
3400
3401        /* Sort thresholds. Registering of new threshold isn't time-critical */
3402        sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
3403                        compare_thresholds, NULL);
3404
3405        /* Find current threshold */
3406        new->current_threshold = -1;
3407        for (i = 0; i < size; i++) {
3408                if (new->entries[i].threshold <= usage) {
3409                        /*
3410                         * new->current_threshold will not be used until
3411                         * rcu_assign_pointer(), so it's safe to increment
3412                         * it here.
3413                         */
3414                        ++new->current_threshold;
3415                } else
3416                        break;
3417        }
3418
3419        /* Free old spare buffer and save old primary buffer as spare */
3420        kfree(thresholds->spare);
3421        thresholds->spare = thresholds->primary;
3422
3423        rcu_assign_pointer(thresholds->primary, new);
3424
3425        /* To be sure that nobody uses thresholds */
3426        synchronize_rcu();
3427
3428unlock:
3429        mutex_unlock(&memcg->thresholds_lock);
3430
3431        return ret;
3432}
3433
3434static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
3435        struct eventfd_ctx *eventfd, const char *args)
3436{
3437        return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
3438}
3439
3440static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
3441        struct eventfd_ctx *eventfd, const char *args)
3442{
3443        return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
3444}
3445
3446static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3447        struct eventfd_ctx *eventfd, enum res_type type)
3448{
3449        struct mem_cgroup_thresholds *thresholds;
3450        struct mem_cgroup_threshold_ary *new;
3451        unsigned long usage;
3452        int i, j, size;
3453
3454        mutex_lock(&memcg->thresholds_lock);
3455
3456        if (type == _MEM) {
3457                thresholds = &memcg->thresholds;
3458                usage = mem_cgroup_usage(memcg, false);
3459        } else if (type == _MEMSWAP) {
3460                thresholds = &memcg->memsw_thresholds;
3461                usage = mem_cgroup_usage(memcg, true);
3462        } else
3463                BUG();
3464
3465        if (!thresholds->primary)
3466                goto unlock;
3467
3468        /* Check if a threshold crossed before removing */
3469        __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3470
3471        /* Calculate new number of threshold */
3472        size = 0;
3473        for (i = 0; i < thresholds->primary->size; i++) {
3474                if (thresholds->primary->entries[i].eventfd != eventfd)
3475                        size++;
3476        }
3477
3478        new = thresholds->spare;
3479
3480        /* Set thresholds array to NULL if we don't have thresholds */
3481        if (!size) {
3482                kfree(new);
3483                new = NULL;
3484                goto swap_buffers;
3485        }
3486
3487        new->size = size;
3488
3489        /* Copy thresholds and find current threshold */
3490        new->current_threshold = -1;
3491        for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3492                if (thresholds->primary->entries[i].eventfd == eventfd)
3493                        continue;
3494
3495                new->entries[j] = thresholds->primary->entries[i];
3496                if (new->entries[j].threshold <= usage) {
3497                        /*
3498                         * new->current_threshold will not be used
3499                         * until rcu_assign_pointer(), so it's safe to increment
3500                         * it here.
3501                         */
3502                        ++new->current_threshold;
3503                }
3504                j++;
3505        }
3506
3507swap_buffers:
3508        /* Swap primary and spare array */
3509        thresholds->spare = thresholds->primary;
3510
3511        rcu_assign_pointer(thresholds->primary, new);
3512
3513        /* To be sure that nobody uses thresholds */
3514        synchronize_rcu();
3515
3516        /* If all events are unregistered, free the spare array */
3517        if (!new) {
3518                kfree(thresholds->spare);
3519                thresholds->spare = NULL;
3520        }
3521unlock:
3522        mutex_unlock(&memcg->thresholds_lock);
3523}
3524
3525static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3526        struct eventfd_ctx *eventfd)
3527{
3528        return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
3529}
3530
3531static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3532        struct eventfd_ctx *eventfd)
3533{
3534        return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
3535}
3536
3537static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
3538        struct eventfd_ctx *eventfd, const char *args)
3539{
3540        struct mem_cgroup_eventfd_list *event;
3541
3542        event = kmalloc(sizeof(*event), GFP_KERNEL);
3543        if (!event)
3544                return -ENOMEM;
3545
3546        spin_lock(&memcg_oom_lock);
3547
3548        event->eventfd = eventfd;
3549        list_add(&event->list, &memcg->oom_notify);
3550
3551        /* already in OOM ? */
3552        if (memcg->under_oom)
3553                eventfd_signal(eventfd, 1);
3554        spin_unlock(&memcg_oom_lock);
3555
3556        return 0;
3557}
3558
3559static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
3560        struct eventfd_ctx *eventfd)
3561{
3562        struct mem_cgroup_eventfd_list *ev, *tmp;
3563
3564        spin_lock(&memcg_oom_lock);
3565
3566        list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
3567                if (ev->eventfd == eventfd) {
3568                        list_del(&ev->list);
3569                        kfree(ev);
3570                }
3571        }
3572
3573        spin_unlock(&memcg_oom_lock);
3574}
3575
3576static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
3577{
3578        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
3579
3580        seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
3581        seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
3582        return 0;
3583}
3584
3585static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
3586        struct cftype *cft, u64 val)
3587{
3588        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3589
3590        /* cannot set to root cgroup and only 0 and 1 are allowed */
3591        if (!css->parent || !((val == 0) || (val == 1)))
3592                return -EINVAL;
3593
3594        memcg->oom_kill_disable = val;
3595        if (!val)
3596                memcg_oom_recover(memcg);
3597
3598        return 0;
3599}
3600
3601#ifdef CONFIG_CGROUP_WRITEBACK
3602
3603struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3604{
3605        return &memcg->cgwb_list;
3606}
3607
3608static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3609{
3610        return wb_domain_init(&memcg->cgwb_domain, gfp);
3611}
3612
3613static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3614{
3615        wb_domain_exit(&memcg->cgwb_domain);
3616}
3617
3618static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3619{
3620        wb_domain_size_changed(&memcg->cgwb_domain);
3621}
3622
3623struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3624{
3625        struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3626
3627        if (!memcg->css.parent)
3628                return NULL;
3629
3630        return &memcg->cgwb_domain;
3631}
3632
3633/**
3634 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3635 * @wb: bdi_writeback in question
3636 * @pfilepages: out parameter for number of file pages
3637 * @pheadroom: out parameter for number of allocatable pages according to memcg
3638 * @pdirty: out parameter for number of dirty pages
3639 * @pwriteback: out parameter for number of pages under writeback
3640 *
3641 * Determine the numbers of file, headroom, dirty, and writeback pages in
3642 * @wb's memcg.  File, dirty and writeback are self-explanatory.  Headroom
3643 * is a bit more involved.
3644 *
3645 * A memcg's headroom is "min(max, high) - used".  In the hierarchy, the
3646 * headroom is calculated as the lowest headroom of itself and the
3647 * ancestors.  Note that this doesn't consider the actual amount of
3648 * available memory in the system.  The caller should further cap
3649 * *@pheadroom accordingly.
3650 */
3651void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3652                         unsigned long *pheadroom, unsigned long *pdirty,
3653                         unsigned long *pwriteback)
3654{
3655        struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3656        struct mem_cgroup *parent;
3657
3658        *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3659
3660        /* this should eventually include NR_UNSTABLE_NFS */
3661        *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
3662        *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3663                                                     (1 << LRU_ACTIVE_FILE));
3664        *pheadroom = PAGE_COUNTER_MAX;
3665
3666        while ((parent = parent_mem_cgroup(memcg))) {
3667                unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3668                unsigned long used = page_counter_read(&memcg->memory);
3669
3670                *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
3671                memcg = parent;
3672        }
3673}
3674
3675#else   /* CONFIG_CGROUP_WRITEBACK */
3676
3677static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3678{
3679        return 0;
3680}
3681
3682static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3683{
3684}
3685
3686static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3687{
3688}
3689
3690#endif  /* CONFIG_CGROUP_WRITEBACK */
3691
3692/*
3693 * DO NOT USE IN NEW FILES.
3694 *
3695 * "cgroup.event_control" implementation.
3696 *
3697 * This is way over-engineered.  It tries to support fully configurable
3698 * events for each user.  Such level of flexibility is completely
3699 * unnecessary especially in the light of the planned unified hierarchy.
3700 *
3701 * Please deprecate this and replace with something simpler if at all
3702 * possible.
3703 */
3704
3705/*
3706 * Unregister event and free resources.
3707 *
3708 * Gets called from workqueue.
3709 */
3710static void memcg_event_remove(struct work_struct *work)
3711{
3712        struct mem_cgroup_event *event =
3713                container_of(work, struct mem_cgroup_event, remove);
3714        struct mem_cgroup *memcg = event->memcg;
3715
3716        remove_wait_queue(event->wqh, &event->wait);
3717
3718        event->unregister_event(memcg, event->eventfd);
3719
3720        /* Notify userspace the event is going away. */
3721        eventfd_signal(event->eventfd, 1);
3722
3723        eventfd_ctx_put(event->eventfd);
3724        kfree(event);
3725        css_put(&memcg->css);
3726}
3727
3728/*
3729 * Gets called on POLLHUP on eventfd when user closes it.
3730 *
3731 * Called with wqh->lock held and interrupts disabled.
3732 */
3733static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3734                            int sync, void *key)
3735{
3736        struct mem_cgroup_event *event =
3737                container_of(wait, struct mem_cgroup_event, wait);
3738        struct mem_cgroup *memcg = event->memcg;
3739        unsigned long flags = (unsigned long)key;
3740
3741        if (flags & POLLHUP) {
3742                /*
3743                 * If the event has been detached at cgroup removal, we
3744                 * can simply return knowing the other side will cleanup
3745                 * for us.
3746                 *
3747                 * We can't race against event freeing since the other
3748                 * side will require wqh->lock via remove_wait_queue(),
3749                 * which we hold.
3750                 */
3751                spin_lock(&memcg->event_list_lock);
3752                if (!list_empty(&event->list)) {
3753                        list_del_init(&event->list);
3754                        /*
3755                         * We are in atomic context, but cgroup_event_remove()
3756                         * may sleep, so we have to call it in workqueue.
3757                         */
3758                        schedule_work(&event->remove);
3759                }
3760                spin_unlock(&memcg->event_list_lock);
3761        }
3762
3763        return 0;
3764}
3765
3766static void memcg_event_ptable_queue_proc(struct file *file,
3767                wait_queue_head_t *wqh, poll_table *pt)
3768{
3769        struct mem_cgroup_event *event =
3770                container_of(pt, struct mem_cgroup_event, pt);
3771
3772        event->wqh = wqh;
3773        add_wait_queue(wqh, &event->wait);
3774}
3775
3776/*
3777 * DO NOT USE IN NEW FILES.
3778 *
3779 * Parse input and register new cgroup event handler.
3780 *
3781 * Input must be in format '<event_fd> <control_fd> <args>'.
3782 * Interpretation of args is defined by control file implementation.
3783 */
3784static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3785                                         char *buf, size_t nbytes, loff_t off)
3786{
3787        struct cgroup_subsys_state *css = of_css(of);
3788        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3789        struct mem_cgroup_event *event;
3790        struct cgroup_subsys_state *cfile_css;
3791        unsigned int efd, cfd;
3792        struct fd efile;
3793        struct fd cfile;
3794        const char *name;
3795        char *endp;
3796        int ret;
3797
3798        buf = strstrip(buf);
3799
3800        efd = simple_strtoul(buf, &endp, 10);
3801        if (*endp != ' ')
3802                return -EINVAL;
3803        buf = endp + 1;
3804
3805        cfd = simple_strtoul(buf, &endp, 10);
3806        if ((*endp != ' ') && (*endp != '\0'))
3807                return -EINVAL;
3808        buf = endp + 1;
3809
3810        event = kzalloc(sizeof(*event), GFP_KERNEL);
3811        if (!event)
3812                return -ENOMEM;
3813
3814        event->memcg = memcg;
3815        INIT_LIST_HEAD(&event->list);
3816        init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3817        init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3818        INIT_WORK(&event->remove, memcg_event_remove);
3819
3820        efile = fdget(efd);
3821        if (!efile.file) {
3822                ret = -EBADF;
3823                goto out_kfree;
3824        }
3825
3826        event->eventfd = eventfd_ctx_fileget(efile.file);
3827        if (IS_ERR(event->eventfd)) {
3828                ret = PTR_ERR(event->eventfd);
3829                goto out_put_efile;
3830        }
3831
3832        cfile = fdget(cfd);
3833        if (!cfile.file) {
3834                ret = -EBADF;
3835                goto out_put_eventfd;
3836        }
3837
3838        /* the process need read permission on control file */
3839        /* AV: shouldn't we check that it's been opened for read instead? */
3840        ret = inode_permission(file_inode(cfile.file), MAY_READ);
3841        if (ret < 0)
3842                goto out_put_cfile;
3843
3844        /*
3845         * Determine the event callbacks and set them in @event.  This used
3846         * to be done via struct cftype but cgroup core no longer knows
3847         * about these events.  The following is crude but the whole thing
3848         * is for compatibility anyway.
3849         *
3850         * DO NOT ADD NEW FILES.
3851         */
3852        name = cfile.file->f_path.dentry->d_name.name;
3853
3854        if (!strcmp(name, "memory.usage_in_bytes")) {
3855                event->register_event = mem_cgroup_usage_register_event;
3856                event->unregister_event = mem_cgroup_usage_unregister_event;
3857        } else if (!strcmp(name, "memory.oom_control")) {
3858                event->register_event = mem_cgroup_oom_register_event;
3859                event->unregister_event = mem_cgroup_oom_unregister_event;
3860        } else if (!strcmp(name, "memory.pressure_level")) {
3861                event->register_event = vmpressure_register_event;
3862                event->unregister_event = vmpressure_unregister_event;
3863        } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
3864                event->register_event = memsw_cgroup_usage_register_event;
3865                event->unregister_event = memsw_cgroup_usage_unregister_event;
3866        } else {
3867                ret = -EINVAL;
3868                goto out_put_cfile;
3869        }
3870
3871        /*
3872         * Verify @cfile should belong to @css.  Also, remaining events are
3873         * automatically removed on cgroup destruction but the removal is
3874         * asynchronous, so take an extra ref on @css.
3875         */
3876        cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
3877                                               &memory_cgrp_subsys);
3878        ret = -EINVAL;
3879        if (IS_ERR(cfile_css))
3880                goto out_put_cfile;
3881        if (cfile_css != css) {
3882                css_put(cfile_css);
3883                goto out_put_cfile;
3884        }
3885
3886        ret = event->register_event(memcg, event->eventfd, buf);
3887        if (ret)
3888                goto out_put_css;
3889
3890        efile.file->f_op->poll(efile.file, &event->pt);
3891
3892        spin_lock(&memcg->event_list_lock);
3893        list_add(&event->list, &memcg->event_list);
3894        spin_unlock(&memcg->event_list_lock);
3895
3896        fdput(cfile);
3897        fdput(efile);
3898
3899        return nbytes;
3900
3901out_put_css:
3902        css_put(css);
3903out_put_cfile:
3904        fdput(cfile);
3905out_put_eventfd:
3906        eventfd_ctx_put(event->eventfd);
3907out_put_efile:
3908        fdput(efile);
3909out_kfree:
3910        kfree(event);
3911
3912        return ret;
3913}
3914
3915static struct cftype mem_cgroup_legacy_files[] = {
3916        {
3917                .name = "usage_in_bytes",
3918                .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
3919                .read_u64 = mem_cgroup_read_u64,
3920        },
3921        {
3922                .name = "max_usage_in_bytes",
3923                .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
3924                .write = mem_cgroup_reset,
3925                .read_u64 = mem_cgroup_read_u64,
3926        },
3927        {
3928                .name = "limit_in_bytes",
3929                .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
3930                .write = mem_cgroup_write,
3931                .read_u64 = mem_cgroup_read_u64,
3932        },
3933        {
3934                .name = "soft_limit_in_bytes",
3935                .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
3936                .write = mem_cgroup_write,
3937                .read_u64 = mem_cgroup_read_u64,
3938        },
3939        {
3940                .name = "failcnt",
3941                .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
3942                .write = mem_cgroup_reset,
3943                .read_u64 = mem_cgroup_read_u64,
3944        },
3945        {
3946                .name = "stat",
3947                .seq_show = memcg_stat_show,
3948        },
3949        {
3950                .name = "force_empty",
3951                .write = mem_cgroup_force_empty_write,
3952        },
3953        {
3954                .name = "use_hierarchy",
3955                .write_u64 = mem_cgroup_hierarchy_write,
3956                .read_u64 = mem_cgroup_hierarchy_read,
3957        },
3958        {
3959                .name = "cgroup.event_control",         /* XXX: for compat */
3960                .write = memcg_write_event_control,
3961                .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
3962        },
3963        {
3964                .name = "swappiness",
3965                .read_u64 = mem_cgroup_swappiness_read,
3966                .write_u64 = mem_cgroup_swappiness_write,
3967        },
3968        {
3969                .name = "move_charge_at_immigrate",
3970                .read_u64 = mem_cgroup_move_charge_read,
3971                .write_u64 = mem_cgroup_move_charge_write,
3972        },
3973        {
3974                .name = "oom_control",
3975                .seq_show = mem_cgroup_oom_control_read,
3976                .write_u64 = mem_cgroup_oom_control_write,
3977                .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3978        },
3979        {
3980                .name = "pressure_level",
3981        },
3982#ifdef CONFIG_NUMA
3983        {
3984                .name = "numa_stat",
3985                .seq_show = memcg_numa_stat_show,
3986        },
3987#endif
3988        {
3989                .name = "kmem.limit_in_bytes",
3990                .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
3991                .write = mem_cgroup_write,
3992                .read_u64 = mem_cgroup_read_u64,
3993        },
3994        {
3995                .name = "kmem.usage_in_bytes",
3996                .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
3997                .read_u64 = mem_cgroup_read_u64,
3998        },
3999        {
4000                .name = "kmem.failcnt",
4001                .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
4002                .write = mem_cgroup_reset,
4003                .read_u64 = mem_cgroup_read_u64,
4004        },
4005        {
4006                .name = "kmem.max_usage_in_bytes",
4007                .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
4008                .write = mem_cgroup_reset,
4009                .read_u64 = mem_cgroup_read_u64,
4010        },
4011#ifdef CONFIG_SLABINFO
4012        {
4013                .name = "kmem.slabinfo",
4014                .seq_start = slab_start,
4015                .seq_next = slab_next,
4016                .seq_stop = slab_stop,
4017                .seq_show = memcg_slab_show,
4018        },
4019#endif
4020        {
4021                .name = "kmem.tcp.limit_in_bytes",
4022                .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4023                .write = mem_cgroup_write,
4024                .read_u64 = mem_cgroup_read_u64,
4025        },
4026        {
4027                .name = "kmem.tcp.usage_in_bytes",
4028                .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4029                .read_u64 = mem_cgroup_read_u64,
4030        },
4031        {
4032                .name = "kmem.tcp.failcnt",
4033                .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4034                .write = mem_cgroup_reset,
4035                .read_u64 = mem_cgroup_read_u64,
4036        },
4037        {
4038                .name = "kmem.tcp.max_usage_in_bytes",
4039                .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4040                .write = mem_cgroup_reset,
4041                .read_u64 = mem_cgroup_read_u64,
4042        },
4043        { },    /* terminate */
4044};
4045
4046/*
4047 * Private memory cgroup IDR
4048 *
4049 * Swap-out records and page cache shadow entries need to store memcg
4050 * references in constrained space, so we maintain an ID space that is
4051 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4052 * memory-controlled cgroups to 64k.
4053 *
4054 * However, there usually are many references to the oflline CSS after
4055 * the cgroup has been destroyed, such as page cache or reclaimable
4056 * slab objects, that don't need to hang on to the ID. We want to keep
4057 * those dead CSS from occupying IDs, or we might quickly exhaust the
4058 * relatively small ID space and prevent the creation of new cgroups
4059 * even when there are much fewer than 64k cgroups - possibly none.
4060 *
4061 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4062 * be freed and recycled when it's no longer needed, which is usually
4063 * when the CSS is offlined.
4064 *
4065 * The only exception to that are records of swapped out tmpfs/shmem
4066 * pages that need to be attributed to live ancestors on swapin. But
4067 * those references are manageable from userspace.
4068 */
4069
4070static DEFINE_IDR(mem_cgroup_idr);
4071
4072static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n)
4073{
4074        VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0);
4075        atomic_add(n, &memcg->id.ref);
4076}
4077
4078static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
4079{
4080        VM_BUG_ON(atomic_read(&memcg->id.ref) < n);
4081        if (atomic_sub_and_test(n, &memcg->id.ref)) {
4082                idr_remove(&mem_cgroup_idr, memcg->id.id);
4083                memcg->id.id = 0;
4084
4085                /* Memcg ID pins CSS */
4086                css_put(&memcg->css);
4087        }
4088}
4089
4090static inline void mem_cgroup_id_get(struct mem_cgroup *memcg)
4091{
4092        mem_cgroup_id_get_many(memcg, 1);
4093}
4094
4095static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
4096{
4097        mem_cgroup_id_put_many(memcg, 1);
4098}
4099
4100/**
4101 * mem_cgroup_from_id - look up a memcg from a memcg id
4102 * @id: the memcg id to look up
4103 *
4104 * Caller must hold rcu_read_lock().
4105 */
4106struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
4107{
4108        WARN_ON_ONCE(!rcu_read_lock_held());
4109        return idr_find(&mem_cgroup_idr, id);
4110}
4111
4112static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
4113{
4114        struct mem_cgroup_per_node *pn;
4115        int tmp = node;
4116        /*
4117         * This routine is called against possible nodes.
4118         * But it's BUG to call kmalloc() against offline node.
4119         *
4120         * TODO: this routine can waste much memory for nodes which will
4121         *       never be onlined. It's better to use memory hotplug callback
4122         *       function.
4123         */
4124        if (!node_state(node, N_NORMAL_MEMORY))
4125                tmp = -1;
4126        pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
4127        if (!pn)
4128                return 1;
4129
4130        lruvec_init(&pn->lruvec);
4131        pn->usage_in_excess = 0;
4132        pn->on_tree = false;
4133        pn->memcg = memcg;
4134
4135        memcg->nodeinfo[node] = pn;
4136        return 0;
4137}
4138
4139static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
4140{
4141        kfree(memcg->nodeinfo[node]);
4142}
4143
4144static void __mem_cgroup_free(struct mem_cgroup *memcg)
4145{
4146        int node;
4147
4148        for_each_node(node)
4149                free_mem_cgroup_per_node_info(memcg, node);
4150        free_percpu(memcg->stat);
4151        kfree(memcg);
4152}
4153
4154static void mem_cgroup_free(struct mem_cgroup *memcg)
4155{
4156        memcg_wb_domain_exit(memcg);
4157        __mem_cgroup_free(memcg);
4158}
4159
4160static struct mem_cgroup *mem_cgroup_alloc(void)
4161{
4162        struct mem_cgroup *memcg;
4163        size_t size;
4164        int node;
4165
4166        size = sizeof(struct mem_cgroup);
4167        size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
4168
4169        memcg = kzalloc(size, GFP_KERNEL);
4170        if (!memcg)
4171                return NULL;
4172
4173        memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
4174                                 1, MEM_CGROUP_ID_MAX,
4175                                 GFP_KERNEL);
4176        if (memcg->id.id < 0)
4177                goto fail;
4178
4179        memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4180        if (!memcg->stat)
4181                goto fail;
4182
4183        for_each_node(node)
4184                if (alloc_mem_cgroup_per_node_info(memcg, node))
4185                        goto fail;
4186
4187        if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4188                goto fail;
4189
4190        INIT_WORK(&memcg->high_work, high_work_func);
4191        memcg->last_scanned_node = MAX_NUMNODES;
4192        INIT_LIST_HEAD(&memcg->oom_notify);
4193        mutex_init(&memcg->thresholds_lock);
4194        spin_lock_init(&memcg->move_lock);
4195        vmpressure_init(&memcg->vmpressure);
4196        INIT_LIST_HEAD(&memcg->event_list);
4197        spin_lock_init(&memcg->event_list_lock);
4198        memcg->socket_pressure = jiffies;
4199#ifndef CONFIG_SLOB
4200        memcg->kmemcg_id = -1;
4201#endif
4202#ifdef CONFIG_CGROUP_WRITEBACK
4203        INIT_LIST_HEAD(&memcg->cgwb_list);
4204#endif
4205        idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
4206        return memcg;
4207fail:
4208        if (memcg->id.id > 0)
4209                idr_remove(&mem_cgroup_idr, memcg->id.id);
4210        __mem_cgroup_free(memcg);
4211        return NULL;
4212}
4213
4214static struct cgroup_subsys_state * __ref
4215mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
4216{
4217        struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4218        struct mem_cgroup *memcg;
4219        long error = -ENOMEM;
4220
4221        memcg = mem_cgroup_alloc();
4222        if (!memcg)
4223                return ERR_PTR(error);
4224
4225        memcg->high = PAGE_COUNTER_MAX;
4226        memcg->soft_limit = PAGE_COUNTER_MAX;
4227        if (parent) {
4228                memcg->swappiness = mem_cgroup_swappiness(parent);
4229                memcg->oom_kill_disable = parent->oom_kill_disable;
4230        }
4231        if (parent && parent->use_hierarchy) {
4232                memcg->use_hierarchy = true;
4233                page_counter_init(&memcg->memory, &parent->memory);
4234                page_counter_init(&memcg->swap, &parent->swap);
4235                page_counter_init(&memcg->memsw, &parent->memsw);
4236                page_counter_init(&memcg->kmem, &parent->kmem);
4237                page_counter_init(&memcg->tcpmem, &parent->tcpmem);
4238        } else {
4239                page_counter_init(&memcg->memory, NULL);
4240                page_counter_init(&memcg->swap, NULL);
4241                page_counter_init(&memcg->memsw, NULL);
4242                page_counter_init(&memcg->kmem, NULL);
4243                page_counter_init(&memcg->tcpmem, NULL);
4244                /*
4245                 * Deeper hierachy with use_hierarchy == false doesn't make
4246                 * much sense so let cgroup subsystem know about this
4247                 * unfortunate state in our controller.
4248                 */
4249                if (parent != root_mem_cgroup)
4250                        memory_cgrp_subsys.broken_hierarchy = true;
4251        }
4252
4253        /* The following stuff does not apply to the root */
4254        if (!parent) {
4255                root_mem_cgroup = memcg;
4256                return &memcg->css;
4257        }
4258
4259        error = memcg_online_kmem(memcg);
4260        if (error)
4261                goto fail;
4262
4263        if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
4264                static_branch_inc(&memcg_sockets_enabled_key);
4265
4266        return &memcg->css;
4267fail:
4268        mem_cgroup_free(memcg);
4269        return ERR_PTR(-ENOMEM);
4270}
4271
4272static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
4273{
4274        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4275
4276        /* Online state pins memcg ID, memcg ID pins CSS */
4277        atomic_set(&memcg->id.ref, 1);
4278        css_get(css);
4279        return 0;
4280}
4281
4282static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
4283{
4284        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4285        struct mem_cgroup_event *event, *tmp;
4286
4287        /*
4288         * Unregister events and notify userspace.
4289         * Notify userspace about cgroup removing only after rmdir of cgroup
4290         * directory to avoid race between userspace and kernelspace.
4291         */
4292        spin_lock(&memcg->event_list_lock);
4293        list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
4294                list_del_init(&event->list);
4295                schedule_work(&event->remove);
4296        }
4297        spin_unlock(&memcg->event_list_lock);
4298
4299        memcg_offline_kmem(memcg);
4300        wb_memcg_offline(memcg);
4301
4302        mem_cgroup_id_put(memcg);
4303}
4304
4305static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4306{
4307        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4308
4309        invalidate_reclaim_iterators(memcg);
4310}
4311
4312static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
4313{
4314        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4315
4316        if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
4317                static_branch_dec(&memcg_sockets_enabled_key);
4318
4319        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
4320                static_branch_dec(&memcg_sockets_enabled_key);
4321
4322        vmpressure_cleanup(&memcg->vmpressure);
4323        cancel_work_sync(&memcg->high_work);
4324        mem_cgroup_remove_from_trees(memcg);
4325        memcg_free_kmem(memcg);
4326        mem_cgroup_free(memcg);
4327}
4328
4329/**
4330 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4331 * @css: the target css
4332 *
4333 * Reset the states of the mem_cgroup associated with @css.  This is
4334 * invoked when the userland requests disabling on the default hierarchy
4335 * but the memcg is pinned through dependency.  The memcg should stop
4336 * applying policies and should revert to the vanilla state as it may be
4337 * made visible again.
4338 *
4339 * The current implementation only resets the essential configurations.
4340 * This needs to be expanded to cover all the visible parts.
4341 */
4342static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4343{
4344        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4345
4346        page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4347        page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4348        page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4349        page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4350        page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
4351        memcg->low = 0;
4352        memcg->high = PAGE_COUNTER_MAX;
4353        memcg->soft_limit = PAGE_COUNTER_MAX;
4354        memcg_wb_domain_size_changed(memcg);
4355}
4356
4357#ifdef CONFIG_MMU
4358/* Handlers for move charge at task migration. */
4359static int mem_cgroup_do_precharge(unsigned long count)
4360{
4361        int ret;
4362
4363        /* Try a single bulk charge without reclaim first, kswapd may wake */
4364        ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
4365        if (!ret) {
4366                mc.precharge += count;
4367                return ret;
4368        }
4369
4370        /* Try charges one by one with reclaim, but do not retry */
4371        while (count--) {
4372                ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1);
4373                if (ret)
4374                        return ret;
4375                mc.precharge++;
4376                cond_resched();
4377        }
4378        return 0;
4379}
4380
4381union mc_target {
4382        struct page     *page;
4383        swp_entry_t     ent;
4384};
4385
4386enum mc_target_type {
4387        MC_TARGET_NONE = 0,
4388        MC_TARGET_PAGE,
4389        MC_TARGET_SWAP,
4390};
4391
4392static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4393                                                unsigned long addr, pte_t ptent)
4394{
4395        struct page *page = vm_normal_page(vma, addr, ptent);
4396
4397        if (!page || !page_mapped(page))
4398                return NULL;
4399        if (PageAnon(page)) {
4400                if (!(mc.flags & MOVE_ANON))
4401                        return NULL;
4402        } else {
4403                if (!(mc.flags & MOVE_FILE))
4404                        return NULL;
4405        }
4406        if (!get_page_unless_zero(page))
4407                return NULL;
4408
4409        return page;
4410}
4411
4412#ifdef CONFIG_SWAP
4413static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4414                        pte_t ptent, swp_entry_t *entry)
4415{
4416        struct page *page = NULL;
4417        swp_entry_t ent = pte_to_swp_entry(ptent);
4418
4419        if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
4420                return NULL;
4421        /*
4422         * Because lookup_swap_cache() updates some statistics counter,
4423         * we call find_get_page() with swapper_space directly.
4424         */
4425        page = find_get_page(swap_address_space(ent), swp_offset(ent));
4426        if (do_memsw_account())
4427                entry->val = ent.val;
4428
4429        return page;
4430}
4431#else
4432static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4433                        pte_t ptent, swp_entry_t *entry)
4434{
4435        return NULL;
4436}
4437#endif
4438
4439static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4440                        unsigned long addr, pte_t ptent, swp_entry_t *entry)
4441{
4442        struct page *page = NULL;
4443        struct address_space *mapping;
4444        pgoff_t pgoff;
4445
4446        if (!vma->vm_file) /* anonymous vma */
4447                return NULL;
4448        if (!(mc.flags & MOVE_FILE))
4449                return NULL;
4450
4451        mapping = vma->vm_file->f_mapping;
4452        pgoff = linear_page_index(vma, addr);
4453
4454        /* page is moved even if it's not RSS of this task(page-faulted). */
4455#ifdef CONFIG_SWAP
4456        /* shmem/tmpfs may report page out on swap: account for that too. */
4457        if (shmem_mapping(mapping)) {
4458                page = find_get_entry(mapping, pgoff);
4459                if (radix_tree_exceptional_entry(page)) {
4460                        swp_entry_t swp = radix_to_swp_entry(page);
4461                        if (do_memsw_account())
4462                                *entry = swp;
4463                        page = find_get_page(swap_address_space(swp),
4464                                             swp_offset(swp));
4465                }
4466        } else
4467                page = find_get_page(mapping, pgoff);
4468#else
4469        page = find_get_page(mapping, pgoff);
4470#endif
4471        return page;
4472}
4473
4474/**
4475 * mem_cgroup_move_account - move account of the page
4476 * @page: the page
4477 * @compound: charge the page as compound or small page
4478 * @from: mem_cgroup which the page is moved from.
4479 * @to: mem_cgroup which the page is moved to. @from != @to.
4480 *
4481 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
4482 *
4483 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4484 * from old cgroup.
4485 */
4486static int mem_cgroup_move_account(struct page *page,
4487                                   bool compound,
4488                                   struct mem_cgroup *from,
4489                                   struct mem_cgroup *to)
4490{
4491        unsigned long flags;
4492        unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
4493        int ret;
4494        bool anon;
4495
4496        VM_BUG_ON(from == to);
4497        VM_BUG_ON_PAGE(PageLRU(page), page);
4498        VM_BUG_ON(compound && !PageTransHuge(page));
4499
4500        /*
4501         * Prevent mem_cgroup_migrate() from looking at
4502         * page->mem_cgroup of its source page while we change it.
4503         */
4504        ret = -EBUSY;
4505        if (!trylock_page(page))
4506                goto out;
4507
4508        ret = -EINVAL;
4509        if (page->mem_cgroup != from)
4510                goto out_unlock;
4511
4512        anon = PageAnon(page);
4513
4514        spin_lock_irqsave(&from->move_lock, flags);
4515
4516        if (!anon && page_mapped(page)) {
4517                __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4518                               nr_pages);
4519                __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4520                               nr_pages);
4521        }
4522
4523        /*
4524         * move_lock grabbed above and caller set from->moving_account, so
4525         * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4526         * So mapping should be stable for dirty pages.
4527         */
4528        if (!anon && PageDirty(page)) {
4529                struct address_space *mapping = page_mapping(page);
4530
4531                if (mapping_cap_account_dirty(mapping)) {
4532                        __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4533                                       nr_pages);
4534                        __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4535                                       nr_pages);
4536                }
4537        }
4538
4539        if (PageWriteback(page)) {
4540                __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4541                               nr_pages);
4542                __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4543                               nr_pages);
4544        }
4545
4546        /*
4547         * It is safe to change page->mem_cgroup here because the page
4548         * is referenced, charged, and isolated - we can't race with
4549         * uncharging, charging, migration, or LRU putback.
4550         */
4551
4552        /* caller should have done css_get */
4553        page->mem_cgroup = to;
4554        spin_unlock_irqrestore(&from->move_lock, flags);
4555
4556        ret = 0;
4557
4558        local_irq_disable();
4559        mem_cgroup_charge_statistics(to, page, compound, nr_pages);
4560        memcg_check_events(to, page);
4561        mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
4562        memcg_check_events(from, page);
4563        local_irq_enable();
4564out_unlock:
4565        unlock_page(page);
4566out:
4567        return ret;
4568}
4569
4570/**
4571 * get_mctgt_type - get target type of moving charge
4572 * @vma: the vma the pte to be checked belongs
4573 * @addr: the address corresponding to the pte to be checked
4574 * @ptent: the pte to be checked
4575 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4576 *
4577 * Returns
4578 *   0(MC_TARGET_NONE): if the pte is not a target for move charge.
4579 *   1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4580 *     move charge. if @target is not NULL, the page is stored in target->page
4581 *     with extra refcnt got(Callers should handle it).
4582 *   2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4583 *     target for charge migration. if @target is not NULL, the entry is stored
4584 *     in target->ent.
4585 *
4586 * Called with pte lock held.
4587 */
4588
4589static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
4590                unsigned long addr, pte_t ptent, union mc_target *target)
4591{
4592        struct page *page = NULL;
4593        enum mc_target_type ret = MC_TARGET_NONE;
4594        swp_entry_t ent = { .val = 0 };
4595
4596        if (pte_present(ptent))
4597                page = mc_handle_present_pte(vma, addr, ptent);
4598        else if (is_swap_pte(ptent))
4599                page = mc_handle_swap_pte(vma, ptent, &ent);
4600        else if (pte_none(ptent))
4601                page = mc_handle_file_pte(vma, addr, ptent, &ent);
4602
4603        if (!page && !ent.val)
4604                return ret;
4605        if (page) {
4606                /*
4607                 * Do only loose check w/o serialization.
4608                 * mem_cgroup_move_account() checks the page is valid or
4609                 * not under LRU exclusion.
4610                 */
4611                if (page->mem_cgroup == mc.from) {
4612                        ret = MC_TARGET_PAGE;
4613                        if (target)
4614                                target->page = page;
4615                }
4616                if (!ret || !target)
4617                        put_page(page);
4618        }
4619        /* There is a swap entry and a page doesn't exist or isn't charged */
4620        if (ent.val && !ret &&
4621            mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
4622                ret = MC_TARGET_SWAP;
4623                if (target)
4624                        target->ent = ent;
4625        }
4626        return ret;
4627}
4628
4629#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4630/*
4631 * We don't consider swapping or file mapped pages because THP does not
4632 * support them for now.
4633 * Caller should make sure that pmd_trans_huge(pmd) is true.
4634 */
4635static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4636                unsigned long addr, pmd_t pmd, union mc_target *target)
4637{
4638        struct page *page = NULL;
4639        enum mc_target_type ret = MC_TARGET_NONE;
4640
4641        page = pmd_page(pmd);
4642        VM_BUG_ON_PAGE(!page || !PageHead(page), page);
4643        if (!(mc.flags & MOVE_ANON))
4644                return ret;
4645        if (page->mem_cgroup == mc.from) {
4646                ret = MC_TARGET_PAGE;
4647                if (target) {
4648                        get_page(page);
4649                        target->page = page;
4650                }
4651        }
4652        return ret;
4653}
4654#else
4655static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4656                unsigned long addr, pmd_t pmd, union mc_target *target)
4657{
4658        return MC_TARGET_NONE;
4659}
4660#endif
4661
4662static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4663                                        unsigned long addr, unsigned long end,
4664                                        struct mm_walk *walk)
4665{
4666        struct vm_area_struct *vma = walk->vma;
4667        pte_t *pte;
4668        spinlock_t *ptl;
4669
4670        ptl = pmd_trans_huge_lock(pmd, vma);
4671        if (ptl) {
4672                if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4673                        mc.precharge += HPAGE_PMD_NR;
4674                spin_unlock(ptl);
4675                return 0;
4676        }
4677
4678        if (pmd_trans_unstable(pmd))
4679                return 0;
4680        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4681        for (; addr != end; pte++, addr += PAGE_SIZE)
4682                if (get_mctgt_type(vma, addr, *pte, NULL))
4683                        mc.precharge++; /* increment precharge temporarily */
4684        pte_unmap_unlock(pte - 1, ptl);
4685        cond_resched();
4686
4687        return 0;
4688}
4689
4690static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4691{
4692        unsigned long precharge;
4693
4694        struct mm_walk mem_cgroup_count_precharge_walk = {
4695                .pmd_entry = mem_cgroup_count_precharge_pte_range,
4696                .mm = mm,
4697        };
4698        down_read(&mm->mmap_sem);
4699        walk_page_range(0, mm->highest_vm_end,
4700                        &mem_cgroup_count_precharge_walk);
4701        up_read(&mm->mmap_sem);
4702
4703        precharge = mc.precharge;
4704        mc.precharge = 0;
4705
4706        return precharge;
4707}
4708
4709static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4710{
4711        unsigned long precharge = mem_cgroup_count_precharge(mm);
4712
4713        VM_BUG_ON(mc.moving_task);
4714        mc.moving_task = current;
4715        return mem_cgroup_do_precharge(precharge);
4716}
4717
4718/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4719static void __mem_cgroup_clear_mc(void)
4720{
4721        struct mem_cgroup *from = mc.from;
4722        struct mem_cgroup *to = mc.to;
4723
4724        /* we must uncharge all the leftover precharges from mc.to */
4725        if (mc.precharge) {
4726                cancel_charge(mc.to, mc.precharge);
4727                mc.precharge = 0;
4728        }
4729        /*
4730         * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4731         * we must uncharge here.
4732         */
4733        if (mc.moved_charge) {
4734                cancel_charge(mc.from, mc.moved_charge);
4735                mc.moved_charge = 0;
4736        }
4737        /* we must fixup refcnts and charges */
4738        if (mc.moved_swap) {
4739                /* uncharge swap account from the old cgroup */
4740                if (!mem_cgroup_is_root(mc.from))
4741                        page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
4742
4743                mem_cgroup_id_put_many(mc.from, mc.moved_swap);
4744
4745                /*
4746                 * we charged both to->memory and to->memsw, so we
4747                 * should uncharge to->memory.
4748                 */
4749                if (!mem_cgroup_is_root(mc.to))
4750                        page_counter_uncharge(&mc.to->memory, mc.moved_swap);
4751
4752                mem_cgroup_id_get_many(mc.to, mc.moved_swap);
4753                css_put_many(&mc.to->css, mc.moved_swap);
4754
4755                mc.moved_swap = 0;
4756        }
4757        memcg_oom_recover(from);
4758        memcg_oom_recover(to);
4759        wake_up_all(&mc.waitq);
4760}
4761
4762static void mem_cgroup_clear_mc(void)
4763{
4764        struct mm_struct *mm = mc.mm;
4765
4766        /*
4767         * we must clear moving_task before waking up waiters at the end of
4768         * task migration.
4769         */
4770        mc.moving_task = NULL;
4771        __mem_cgroup_clear_mc();
4772        spin_lock(&mc.lock);
4773        mc.from = NULL;
4774        mc.to = NULL;
4775        mc.mm = NULL;
4776        spin_unlock(&mc.lock);
4777
4778        mmput(mm);
4779}
4780
4781static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
4782{
4783        struct cgroup_subsys_state *css;
4784        struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
4785        struct mem_cgroup *from;
4786        struct task_struct *leader, *p;
4787        struct mm_struct *mm;
4788        unsigned long move_flags;
4789        int ret = 0;
4790
4791        /* charge immigration isn't supported on the default hierarchy */
4792        if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
4793                return 0;
4794
4795        /*
4796         * Multi-process migrations only happen on the default hierarchy
4797         * where charge immigration is not used.  Perform charge
4798         * immigration if @tset contains a leader and whine if there are
4799         * multiple.
4800         */
4801        p = NULL;
4802        cgroup_taskset_for_each_leader(leader, css, tset) {
4803                WARN_ON_ONCE(p);
4804                p = leader;
4805                memcg = mem_cgroup_from_css(css);
4806        }
4807        if (!p)
4808                return 0;
4809
4810        /*
4811         * We are now commited to this value whatever it is. Changes in this
4812         * tunable will only affect upcoming migrations, not the current one.
4813         * So we need to save it, and keep it going.
4814         */
4815        move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4816        if (!move_flags)
4817                return 0;
4818
4819        from = mem_cgroup_from_task(p);
4820
4821        VM_BUG_ON(from == memcg);
4822
4823        mm = get_task_mm(p);
4824        if (!mm)
4825                return 0;
4826        /* We move charges only when we move a owner of the mm */
4827        if (mm->owner == p) {
4828                VM_BUG_ON(mc.from);
4829                VM_BUG_ON(mc.to);
4830                VM_BUG_ON(mc.precharge);
4831                VM_BUG_ON(mc.moved_charge);
4832                VM_BUG_ON(mc.moved_swap);
4833
4834                spin_lock(&mc.lock);
4835                mc.mm = mm;
4836                mc.from = from;
4837                mc.to = memcg;
4838                mc.flags = move_flags;
4839                spin_unlock(&mc.lock);
4840                /* We set mc.moving_task later */
4841
4842                ret = mem_cgroup_precharge_mc(mm);
4843                if (ret)
4844                        mem_cgroup_clear_mc();
4845        } else {
4846                mmput(mm);
4847        }
4848        return ret;
4849}
4850
4851static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
4852{
4853        if (mc.to)
4854                mem_cgroup_clear_mc();
4855}
4856
4857static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4858                                unsigned long addr, unsigned long end,
4859                                struct mm_walk *walk)
4860{
4861        int ret = 0;
4862        struct vm_area_struct *vma = walk->vma;
4863        pte_t *pte;
4864        spinlock_t *ptl;
4865        enum mc_target_type target_type;
4866        union mc_target target;
4867        struct page *page;
4868
4869        ptl = pmd_trans_huge_lock(pmd, vma);
4870        if (ptl) {
4871                if (mc.precharge < HPAGE_PMD_NR) {
4872                        spin_unlock(ptl);
4873                        return 0;
4874                }
4875                target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4876                if (target_type == MC_TARGET_PAGE) {
4877                        page = target.page;
4878                        if (!isolate_lru_page(page)) {
4879                                if (!mem_cgroup_move_account(page, true,
4880                                                             mc.from, mc.to)) {
4881                                        mc.precharge -= HPAGE_PMD_NR;
4882                                        mc.moved_charge += HPAGE_PMD_NR;
4883                                }
4884                                putback_lru_page(page);
4885                        }
4886                        put_page(page);
4887                }
4888                spin_unlock(ptl);
4889                return 0;
4890        }
4891
4892        if (pmd_trans_unstable(pmd))
4893                return 0;
4894retry:
4895        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4896        for (; addr != end; addr += PAGE_SIZE) {
4897                pte_t ptent = *(pte++);
4898                swp_entry_t ent;
4899
4900                if (!mc.precharge)
4901                        break;
4902
4903                switch (get_mctgt_type(vma, addr, ptent, &target)) {
4904                case MC_TARGET_PAGE:
4905                        page = target.page;
4906                        /*
4907                         * We can have a part of the split pmd here. Moving it
4908                         * can be done but it would be too convoluted so simply
4909                         * ignore such a partial THP and keep it in original
4910                         * memcg. There should be somebody mapping the head.
4911                         */
4912                        if (PageTransCompound(page))
4913                                goto put;
4914                        if (isolate_lru_page(page))
4915                                goto put;
4916                        if (!mem_cgroup_move_account(page, false,
4917                                                mc.from, mc.to)) {
4918                                mc.precharge--;
4919                                /* we uncharge from mc.from later. */
4920                                mc.moved_charge++;
4921                        }
4922                        putback_lru_page(page);
4923put:                    /* get_mctgt_type() gets the page */
4924                        put_page(page);
4925                        break;
4926                case MC_TARGET_SWAP:
4927                        ent = target.ent;
4928                        if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
4929                                mc.precharge--;
4930                                /* we fixup refcnts and charges later. */
4931                                mc.moved_swap++;
4932                        }
4933                        break;
4934                default:
4935                        break;
4936                }
4937        }
4938        pte_unmap_unlock(pte - 1, ptl);
4939        cond_resched();
4940
4941        if (addr != end) {
4942                /*
4943                 * We have consumed all precharges we got in can_attach().
4944                 * We try charge one by one, but don't do any additional
4945                 * charges to mc.to if we have failed in charge once in attach()
4946                 * phase.
4947                 */
4948                ret = mem_cgroup_do_precharge(1);
4949                if (!ret)
4950                        goto retry;
4951        }
4952
4953        return ret;
4954}
4955
4956static void mem_cgroup_move_charge(void)
4957{
4958        struct mm_walk mem_cgroup_move_charge_walk = {
4959                .pmd_entry = mem_cgroup_move_charge_pte_range,
4960                .mm = mc.mm,
4961        };
4962
4963        lru_add_drain_all();
4964        /*
4965         * Signal lock_page_memcg() to take the memcg's move_lock
4966         * while we're moving its pages to another memcg. Then wait
4967         * for already started RCU-only updates to finish.
4968         */
4969        atomic_inc(&mc.from->moving_account);
4970        synchronize_rcu();
4971retry:
4972        if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
4973                /*
4974                 * Someone who are holding the mmap_sem might be waiting in
4975                 * waitq. So we cancel all extra charges, wake up all waiters,
4976                 * and retry. Because we cancel precharges, we might not be able
4977                 * to move enough charges, but moving charge is a best-effort
4978                 * feature anyway, so it wouldn't be a big problem.
4979                 */
4980                __mem_cgroup_clear_mc();
4981                cond_resched();
4982                goto retry;
4983        }
4984        /*
4985         * When we have consumed all precharges and failed in doing
4986         * additional charge, the page walk just aborts.
4987         */
4988        walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk);
4989
4990        up_read(&mc.mm->mmap_sem);
4991        atomic_dec(&mc.from->moving_account);
4992}
4993
4994static void mem_cgroup_move_task(void)
4995{
4996        if (mc.to) {
4997                mem_cgroup_move_charge();
4998                mem_cgroup_clear_mc();
4999        }
5000}
5001#else   /* !CONFIG_MMU */
5002static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
5003{
5004        return 0;
5005}
5006static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
5007{
5008}
5009static void mem_cgroup_move_task(void)
5010{
5011}
5012#endif
5013
5014/*
5015 * Cgroup retains root cgroups across [un]mount cycles making it necessary
5016 * to verify whether we're attached to the default hierarchy on each mount
5017 * attempt.
5018 */
5019static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
5020{
5021        /*
5022         * use_hierarchy is forced on the default hierarchy.  cgroup core
5023         * guarantees that @root doesn't have any children, so turning it
5024         * on for the root memcg is enough.
5025         */
5026        if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5027                root_mem_cgroup->use_hierarchy = true;
5028        else
5029                root_mem_cgroup->use_hierarchy = false;
5030}
5031
5032static u64 memory_current_read(struct cgroup_subsys_state *css,
5033                               struct cftype *cft)
5034{
5035        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5036
5037        return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
5038}
5039
5040static int memory_low_show(struct seq_file *m, void *v)
5041{
5042        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5043        unsigned long low = READ_ONCE(memcg->low);
5044
5045        if (low == PAGE_COUNTER_MAX)
5046                seq_puts(m, "max\n");
5047        else
5048                seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5049
5050        return 0;
5051}
5052
5053static ssize_t memory_low_write(struct kernfs_open_file *of,
5054                                char *buf, size_t nbytes, loff_t off)
5055{
5056        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5057        unsigned long low;
5058        int err;
5059
5060        buf = strstrip(buf);
5061        err = page_counter_memparse(buf, "max", &low);
5062        if (err)
5063                return err;
5064
5065        memcg->low = low;
5066
5067        return nbytes;
5068}
5069
5070static int memory_high_show(struct seq_file *m, void *v)
5071{
5072        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5073        unsigned long high = READ_ONCE(memcg->high);
5074
5075        if (high == PAGE_COUNTER_MAX)
5076                seq_puts(m, "max\n");
5077        else
5078                seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5079
5080        return 0;
5081}
5082
5083static ssize_t memory_high_write(struct kernfs_open_file *of,
5084                                 char *buf, size_t nbytes, loff_t off)
5085{
5086        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5087        unsigned long nr_pages;
5088        unsigned long high;
5089        int err;
5090
5091        buf = strstrip(buf);
5092        err = page_counter_memparse(buf, "max", &high);
5093        if (err)
5094                return err;
5095
5096        memcg->high = high;
5097
5098        nr_pages = page_counter_read(&memcg->memory);
5099        if (nr_pages > high)
5100                try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5101                                             GFP_KERNEL, true);
5102
5103        memcg_wb_domain_size_changed(memcg);
5104        return nbytes;
5105}
5106
5107static int memory_max_show(struct seq_file *m, void *v)
5108{
5109        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5110        unsigned long max = READ_ONCE(memcg->memory.limit);
5111
5112        if (max == PAGE_COUNTER_MAX)
5113                seq_puts(m, "max\n");
5114        else
5115                seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5116
5117        return 0;
5118}
5119
5120static ssize_t memory_max_write(struct kernfs_open_file *of,
5121                                char *buf, size_t nbytes, loff_t off)
5122{
5123        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5124        unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5125        bool drained = false;
5126        unsigned long max;
5127        int err;
5128
5129        buf = strstrip(buf);
5130        err = page_counter_memparse(buf, "max", &max);
5131        if (err)
5132                return err;
5133
5134        xchg(&memcg->memory.limit, max);
5135
5136        for (;;) {
5137                unsigned long nr_pages = page_counter_read(&memcg->memory);
5138
5139                if (nr_pages <= max)
5140                        break;
5141
5142                if (signal_pending(current)) {
5143                        err = -EINTR;
5144                        break;
5145                }
5146
5147                if (!drained) {
5148                        drain_all_stock(memcg);
5149                        drained = true;
5150                        continue;
5151                }
5152
5153                if (nr_reclaims) {
5154                        if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5155                                                          GFP_KERNEL, true))
5156                                nr_reclaims--;
5157                        continue;
5158                }
5159
5160                mem_cgroup_events(memcg, MEMCG_OOM, 1);
5161                if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5162                        break;
5163        }
5164
5165        memcg_wb_domain_size_changed(memcg);
5166        return nbytes;
5167}
5168
5169static int memory_events_show(struct seq_file *m, void *v)
5170{
5171        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5172
5173        seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5174        seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5175        seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5176        seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5177
5178        return 0;
5179}
5180
5181static int memory_stat_show(struct seq_file *m, void *v)
5182{
5183        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5184        unsigned long stat[MEMCG_NR_STAT];
5185        unsigned long events[MEMCG_NR_EVENTS];
5186        int i;
5187
5188        /*
5189         * Provide statistics on the state of the memory subsystem as
5190         * well as cumulative event counters that show past behavior.
5191         *
5192         * This list is ordered following a combination of these gradients:
5193         * 1) generic big picture -> specifics and details
5194         * 2) reflecting userspace activity -> reflecting kernel heuristics
5195         *
5196         * Current memory state:
5197         */
5198
5199        tree_stat(memcg, stat);
5200        tree_events(memcg, events);
5201
5202        seq_printf(m, "anon %llu\n",
5203                   (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE);
5204        seq_printf(m, "file %llu\n",
5205                   (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE);
5206        seq_printf(m, "kernel_stack %llu\n",
5207                   (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024);
5208        seq_printf(m, "slab %llu\n",
5209                   (u64)(stat[MEMCG_SLAB_RECLAIMABLE] +
5210                         stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
5211        seq_printf(m, "sock %llu\n",
5212                   (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
5213
5214        seq_printf(m, "file_mapped %llu\n",
5215                   (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE);
5216        seq_printf(m, "file_dirty %llu\n",
5217                   (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE);
5218        seq_printf(m, "file_writeback %llu\n",
5219                   (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE);
5220
5221        for (i = 0; i < NR_LRU_LISTS; i++) {
5222                struct mem_cgroup *mi;
5223                unsigned long val = 0;
5224
5225                for_each_mem_cgroup_tree(mi, memcg)
5226                        val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5227                seq_printf(m, "%s %llu\n",
5228                           mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5229        }
5230
5231        seq_printf(m, "slab_reclaimable %llu\n",
5232                   (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE);
5233        seq_printf(m, "slab_unreclaimable %llu\n",
5234                   (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5235
5236        /* Accumulated memory events */
5237
5238        seq_printf(m, "pgfault %lu\n",
5239                   events[MEM_CGROUP_EVENTS_PGFAULT]);
5240        seq_printf(m, "pgmajfault %lu\n",
5241                   events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
5242
5243        return 0;
5244}
5245
5246static struct cftype memory_files[] = {
5247        {
5248                .name = "current",
5249                .flags = CFTYPE_NOT_ON_ROOT,
5250                .read_u64 = memory_current_read,
5251        },
5252        {
5253                .name = "low",
5254                .flags = CFTYPE_NOT_ON_ROOT,
5255                .seq_show = memory_low_show,
5256                .write = memory_low_write,
5257        },
5258        {
5259                .name = "high",
5260                .flags = CFTYPE_NOT_ON_ROOT,
5261                .seq_show = memory_high_show,
5262                .write = memory_high_write,
5263        },
5264        {
5265                .name = "max",
5266                .flags = CFTYPE_NOT_ON_ROOT,
5267                .seq_show = memory_max_show,
5268                .write = memory_max_write,
5269        },
5270        {
5271                .name = "events",
5272                .flags = CFTYPE_NOT_ON_ROOT,
5273                .file_offset = offsetof(struct mem_cgroup, events_file),
5274                .seq_show = memory_events_show,
5275        },
5276        {
5277                .name = "stat",
5278                .flags = CFTYPE_NOT_ON_ROOT,
5279                .seq_show = memory_stat_show,
5280        },
5281        { }     /* terminate */
5282};
5283
5284struct cgroup_subsys memory_cgrp_subsys = {
5285        .css_alloc = mem_cgroup_css_alloc,
5286        .css_online = mem_cgroup_css_online,
5287        .css_offline = mem_cgroup_css_offline,
5288        .css_released = mem_cgroup_css_released,
5289        .css_free = mem_cgroup_css_free,
5290        .css_reset = mem_cgroup_css_reset,
5291        .can_attach = mem_cgroup_can_attach,
5292        .cancel_attach = mem_cgroup_cancel_attach,
5293        .post_attach = mem_cgroup_move_task,
5294        .bind = mem_cgroup_bind,
5295        .dfl_cftypes = memory_files,
5296        .legacy_cftypes = mem_cgroup_legacy_files,
5297        .early_init = 0,
5298};
5299
5300/**
5301 * mem_cgroup_low - check if memory consumption is below the normal range
5302 * @root: the highest ancestor to consider
5303 * @memcg: the memory cgroup to check
5304 *
5305 * Returns %true if memory consumption of @memcg, and that of all
5306 * configurable ancestors up to @root, is below the normal range.
5307 */
5308bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5309{
5310        if (mem_cgroup_disabled())
5311                return false;
5312
5313        /*
5314         * The toplevel group doesn't have a configurable range, so
5315         * it's never low when looked at directly, and it is not
5316         * considered an ancestor when assessing the hierarchy.
5317         */
5318
5319        if (memcg == root_mem_cgroup)
5320                return false;
5321
5322        if (page_counter_read(&memcg->memory) >= memcg->low)
5323                return false;
5324
5325        while (memcg != root) {
5326                memcg = parent_mem_cgroup(memcg);
5327
5328                if (memcg == root_mem_cgroup)
5329                        break;
5330
5331                if (page_counter_read(&memcg->memory) >= memcg->low)
5332                        return false;
5333        }
5334        return true;
5335}
5336
5337/**
5338 * mem_cgroup_try_charge - try charging a page
5339 * @page: page to charge
5340 * @mm: mm context of the victim
5341 * @gfp_mask: reclaim mode
5342 * @memcgp: charged memcg return
5343 * @compound: charge the page as compound or small page
5344 *
5345 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5346 * pages according to @gfp_mask if necessary.
5347 *
5348 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5349 * Otherwise, an error code is returned.
5350 *
5351 * After page->mapping has been set up, the caller must finalize the
5352 * charge with mem_cgroup_commit_charge().  Or abort the transaction
5353 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5354 */
5355int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
5356                          gfp_t gfp_mask, struct mem_cgroup **memcgp,
5357                          bool compound)
5358{
5359        struct mem_cgroup *memcg = NULL;
5360        unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
5361        int ret = 0;
5362
5363        if (mem_cgroup_disabled())
5364                goto out;
5365
5366        if (PageSwapCache(page)) {
5367                /*
5368                 * Every swap fault against a single page tries to charge the
5369                 * page, bail as early as possible.  shmem_unuse() encounters
5370                 * already charged pages, too.  The USED bit is protected by
5371                 * the page lock, which serializes swap cache removal, which
5372                 * in turn serializes uncharging.
5373                 */
5374                VM_BUG_ON_PAGE(!PageLocked(page), page);
5375                if (page->mem_cgroup)
5376                        goto out;
5377
5378                if (do_swap_account) {
5379                        swp_entry_t ent = { .val = page_private(page), };
5380                        unsigned short id = lookup_swap_cgroup_id(ent);
5381
5382                        rcu_read_lock();
5383                        memcg = mem_cgroup_from_id(id);
5384                        if (memcg && !css_tryget_online(&memcg->css))
5385                                memcg = NULL;
5386                        rcu_read_unlock();
5387                }
5388        }
5389
5390        if (!memcg)
5391                memcg = get_mem_cgroup_from_mm(mm);
5392
5393        ret = try_charge(memcg, gfp_mask, nr_pages);
5394
5395        css_put(&memcg->css);
5396out:
5397        *memcgp = memcg;
5398        return ret;
5399}
5400
5401/**
5402 * mem_cgroup_commit_charge - commit a page charge
5403 * @page: page to charge
5404 * @memcg: memcg to charge the page to
5405 * @lrucare: page might be on LRU already
5406 * @compound: charge the page as compound or small page
5407 *
5408 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5409 * after page->mapping has been set up.  This must happen atomically
5410 * as part of the page instantiation, i.e. under the page table lock
5411 * for anonymous pages, under the page lock for page and swap cache.
5412 *
5413 * In addition, the page must not be on the LRU during the commit, to
5414 * prevent racing with task migration.  If it might be, use @lrucare.
5415 *
5416 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5417 */
5418void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
5419                              bool lrucare, bool compound)
5420{
5421        unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
5422
5423        VM_BUG_ON_PAGE(!page->mapping, page);
5424        VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5425
5426        if (mem_cgroup_disabled())
5427                return;
5428        /*
5429         * Swap faults will attempt to charge the same page multiple
5430         * times.  But reuse_swap_page() might have removed the page
5431         * from swapcache already, so we can't check PageSwapCache().
5432         */
5433        if (!memcg)
5434                return;
5435
5436        commit_charge(page, memcg, lrucare);
5437
5438        local_irq_disable();
5439        mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
5440        memcg_check_events(memcg, page);
5441        local_irq_enable();
5442
5443        if (do_memsw_account() && PageSwapCache(page)) {
5444                swp_entry_t entry = { .val = page_private(page) };
5445                /*
5446                 * The swap entry might not get freed for a long time,
5447                 * let's not wait for it.  The page already received a
5448                 * memory+swap charge, drop the swap entry duplicate.
5449                 */
5450                mem_cgroup_uncharge_swap(entry);
5451        }
5452}
5453
5454/**
5455 * mem_cgroup_cancel_charge - cancel a page charge
5456 * @page: page to charge
5457 * @memcg: memcg to charge the page to
5458 * @compound: charge the page as compound or small page
5459 *
5460 * Cancel a charge transaction started by mem_cgroup_try_charge().
5461 */
5462void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5463                bool compound)
5464{
5465        unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
5466
5467        if (mem_cgroup_disabled())
5468                return;
5469        /*
5470         * Swap faults will attempt to charge the same page multiple
5471         * times.  But reuse_swap_page() might have removed the page
5472         * from swapcache already, so we can't check PageSwapCache().
5473         */
5474        if (!memcg)
5475                return;
5476
5477        cancel_charge(memcg, nr_pages);
5478}
5479
5480static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
5481                           unsigned long nr_anon, unsigned long nr_file,
5482                           unsigned long nr_huge, unsigned long nr_kmem,
5483                           struct page *dummy_page)
5484{
5485        unsigned long nr_pages = nr_anon + nr_file + nr_kmem;
5486        unsigned long flags;
5487
5488        if (!mem_cgroup_is_root(memcg)) {
5489                page_counter_uncharge(&memcg->memory, nr_pages);
5490                if (do_memsw_account())
5491                        page_counter_uncharge(&memcg->memsw, nr_pages);
5492                if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem)
5493                        page_counter_uncharge(&memcg->kmem, nr_kmem);
5494                memcg_oom_recover(memcg);
5495        }
5496
5497        local_irq_save(flags);
5498        __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5499        __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5500        __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5501        __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
5502        __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
5503        memcg_check_events(memcg, dummy_page);
5504        local_irq_restore(flags);
5505
5506        if (!mem_cgroup_is_root(memcg))
5507                css_put_many(&memcg->css, nr_pages);
5508}
5509
5510static void uncharge_list(struct list_head *page_list)
5511{
5512        struct mem_cgroup *memcg = NULL;
5513        unsigned long nr_anon = 0;
5514        unsigned long nr_file = 0;
5515        unsigned long nr_huge = 0;
5516        unsigned long nr_kmem = 0;
5517        unsigned long pgpgout = 0;
5518        struct list_head *next;
5519        struct page *page;
5520
5521        /*
5522         * Note that the list can be a single page->lru; hence the
5523         * do-while loop instead of a simple list_for_each_entry().
5524         */
5525        next = page_list->next;
5526        do {
5527                page = list_entry(next, struct page, lru);
5528                next = page->lru.next;
5529
5530                VM_BUG_ON_PAGE(PageLRU(page), page);
5531                VM_BUG_ON_PAGE(page_count(page), page);
5532
5533                if (!page->mem_cgroup)
5534                        continue;
5535
5536                /*
5537                 * Nobody should be changing or seriously looking at
5538                 * page->mem_cgroup at this point, we have fully
5539                 * exclusive access to the page.
5540                 */
5541
5542                if (memcg != page->mem_cgroup) {
5543                        if (memcg) {
5544                                uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5545                                               nr_huge, nr_kmem, page);
5546                                pgpgout = nr_anon = nr_file =
5547                                        nr_huge = nr_kmem = 0;
5548                        }
5549                        memcg = page->mem_cgroup;
5550                }
5551
5552                if (!PageKmemcg(page)) {
5553                        unsigned int nr_pages = 1;
5554
5555                        if (PageTransHuge(page)) {
5556                                nr_pages <<= compound_order(page);
5557                                nr_huge += nr_pages;
5558                        }
5559                        if (PageAnon(page))
5560                                nr_anon += nr_pages;
5561                        else
5562                                nr_file += nr_pages;
5563                        pgpgout++;
5564                } else {
5565                        nr_kmem += 1 << compound_order(page);
5566                        __ClearPageKmemcg(page);
5567                }
5568
5569                page->mem_cgroup = NULL;
5570        } while (next != page_list);
5571
5572        if (memcg)
5573                uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5574                               nr_huge, nr_kmem, page);
5575}
5576
5577/**
5578 * mem_cgroup_uncharge - uncharge a page
5579 * @page: page to uncharge
5580 *
5581 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5582 * mem_cgroup_commit_charge().
5583 */
5584void mem_cgroup_uncharge(struct page *page)
5585{
5586        if (mem_cgroup_disabled())
5587                return;
5588
5589        /* Don't touch page->lru of any random page, pre-check: */
5590        if (!page->mem_cgroup)
5591                return;
5592
5593        INIT_LIST_HEAD(&page->lru);
5594        uncharge_list(&page->lru);
5595}
5596
5597/**
5598 * mem_cgroup_uncharge_list - uncharge a list of page
5599 * @page_list: list of pages to uncharge
5600 *
5601 * Uncharge a list of pages previously charged with
5602 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5603 */
5604void mem_cgroup_uncharge_list(struct list_head *page_list)
5605{
5606        if (mem_cgroup_disabled())
5607                return;
5608
5609        if (!list_empty(page_list))
5610                uncharge_list(page_list);
5611}
5612
5613/**
5614 * mem_cgroup_migrate - charge a page's replacement
5615 * @oldpage: currently circulating page
5616 * @newpage: replacement page
5617 *
5618 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5619 * be uncharged upon free.
5620 *
5621 * Both pages must be locked, @newpage->mapping must be set up.
5622 */
5623void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
5624{
5625        struct mem_cgroup *memcg;
5626        unsigned int nr_pages;
5627        bool compound;
5628        unsigned long flags;
5629
5630        VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5631        VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
5632        VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
5633        VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5634                       newpage);
5635
5636        if (mem_cgroup_disabled())
5637                return;
5638
5639        /* Page cache replacement: new page already charged? */
5640        if (newpage->mem_cgroup)
5641                return;
5642
5643        /* Swapcache readahead pages can get replaced before being charged */
5644        memcg = oldpage->mem_cgroup;
5645        if (!memcg)
5646                return;
5647
5648        /* Force-charge the new page. The old one will be freed soon */
5649        compound = PageTransHuge(newpage);
5650        nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5651
5652        page_counter_charge(&memcg->memory, nr_pages);
5653        if (do_memsw_account())
5654                page_counter_charge(&memcg->memsw, nr_pages);
5655        css_get_many(&memcg->css, nr_pages);
5656
5657        commit_charge(newpage, memcg, false);
5658
5659        local_irq_save(flags);
5660        mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5661        memcg_check_events(memcg, newpage);
5662        local_irq_restore(flags);
5663}
5664
5665DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
5666EXPORT_SYMBOL(memcg_sockets_enabled_key);
5667
5668void mem_cgroup_sk_alloc(struct sock *sk)
5669{
5670        struct mem_cgroup *memcg;
5671
5672        if (!mem_cgroup_sockets_enabled)
5673                return;
5674
5675        /*
5676         * Socket cloning can throw us here with sk_memcg already
5677         * filled. It won't however, necessarily happen from
5678         * process context. So the test for root memcg given
5679         * the current task's memcg won't help us in this case.
5680         *
5681         * Respecting the original socket's memcg is a better
5682         * decision in this case.
5683         */
5684        if (sk->sk_memcg) {
5685                BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5686                css_get(&sk->sk_memcg->css);
5687                return;
5688        }
5689
5690        rcu_read_lock();
5691        memcg = mem_cgroup_from_task(current);
5692        if (memcg == root_mem_cgroup)
5693                goto out;
5694        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
5695                goto out;
5696        if (css_tryget_online(&memcg->css))
5697                sk->sk_memcg = memcg;
5698out:
5699        rcu_read_unlock();
5700}
5701
5702void mem_cgroup_sk_free(struct sock *sk)
5703{
5704        if (sk->sk_memcg)
5705                css_put(&sk->sk_memcg->css);
5706}
5707
5708/**
5709 * mem_cgroup_charge_skmem - charge socket memory
5710 * @memcg: memcg to charge
5711 * @nr_pages: number of pages to charge
5712 *
5713 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5714 * @memcg's configured limit, %false if the charge had to be forced.
5715 */
5716bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5717{
5718        gfp_t gfp_mask = GFP_KERNEL;
5719
5720        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
5721                struct page_counter *fail;
5722
5723                if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5724                        memcg->tcpmem_pressure = 0;
5725                        return true;
5726                }
5727                page_counter_charge(&memcg->tcpmem, nr_pages);
5728                memcg->tcpmem_pressure = 1;
5729                return false;
5730        }
5731
5732        /* Don't block in the packet receive path */
5733        if (in_softirq())
5734                gfp_mask = GFP_NOWAIT;
5735
5736        this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5737
5738        if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5739                return true;
5740
5741        try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
5742        return false;
5743}
5744
5745/**
5746 * mem_cgroup_uncharge_skmem - uncharge socket memory
5747 * @memcg - memcg to uncharge
5748 * @nr_pages - number of pages to uncharge
5749 */
5750void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5751{
5752        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
5753                page_counter_uncharge(&memcg->tcpmem, nr_pages);
5754                return;
5755        }
5756
5757        this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5758
5759        page_counter_uncharge(&memcg->memory, nr_pages);
5760        css_put_many(&memcg->css, nr_pages);
5761}
5762
5763static int __init cgroup_memory(char *s)
5764{
5765        char *token;
5766
5767        while ((token = strsep(&s, ",")) != NULL) {
5768                if (!*token)
5769                        continue;
5770                if (!strcmp(token, "nosocket"))
5771                        cgroup_memory_nosocket = true;
5772                if (!strcmp(token, "nokmem"))
5773                        cgroup_memory_nokmem = true;
5774        }
5775        return 0;
5776}
5777__setup("cgroup.memory=", cgroup_memory);
5778
5779/*
5780 * subsys_initcall() for memory controller.
5781 *
5782 * Some parts like hotcpu_notifier() have to be initialized from this context
5783 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5784 * everything that doesn't depend on a specific mem_cgroup structure should
5785 * be initialized from here.
5786 */
5787static int __init mem_cgroup_init(void)
5788{
5789        int cpu, node;
5790
5791#ifndef CONFIG_SLOB
5792        /*
5793         * Kmem cache creation is mostly done with the slab_mutex held,
5794         * so use a special workqueue to avoid stalling all worker
5795         * threads in case lots of cgroups are created simultaneously.
5796         */
5797        memcg_kmem_cache_create_wq =
5798                alloc_ordered_workqueue("memcg_kmem_cache_create", 0);
5799        BUG_ON(!memcg_kmem_cache_create_wq);
5800#endif
5801
5802        hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
5803
5804        for_each_possible_cpu(cpu)
5805                INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5806                          drain_local_stock);
5807
5808        for_each_node(node) {
5809                struct mem_cgroup_tree_per_node *rtpn;
5810
5811                rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5812                                    node_online(node) ? node : NUMA_NO_NODE);
5813
5814                rtpn->rb_root = RB_ROOT;
5815                spin_lock_init(&rtpn->lock);
5816                soft_limit_tree.rb_tree_per_node[node] = rtpn;
5817        }
5818
5819        return 0;
5820}
5821subsys_initcall(mem_cgroup_init);
5822
5823#ifdef CONFIG_MEMCG_SWAP
5824static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
5825{
5826        while (!atomic_inc_not_zero(&memcg->id.ref)) {
5827                /*
5828                 * The root cgroup cannot be destroyed, so it's refcount must
5829                 * always be >= 1.
5830                 */
5831                if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
5832                        VM_BUG_ON(1);
5833                        break;
5834                }
5835                memcg = parent_mem_cgroup(memcg);
5836                if (!memcg)
5837                        memcg = root_mem_cgroup;
5838        }
5839        return memcg;
5840}
5841
5842/**
5843 * mem_cgroup_swapout - transfer a memsw charge to swap
5844 * @page: page whose memsw charge to transfer
5845 * @entry: swap entry to move the charge to
5846 *
5847 * Transfer the memsw charge of @page to @entry.
5848 */
5849void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5850{
5851        struct mem_cgroup *memcg, *swap_memcg;
5852        unsigned short oldid;
5853
5854        VM_BUG_ON_PAGE(PageLRU(page), page);
5855        VM_BUG_ON_PAGE(page_count(page), page);
5856
5857        if (!do_memsw_account())
5858                return;
5859
5860        memcg = page->mem_cgroup;
5861
5862        /* Readahead page, never charged */
5863        if (!memcg)
5864                return;
5865
5866        /*
5867         * In case the memcg owning these pages has been offlined and doesn't
5868         * have an ID allocated to it anymore, charge the closest online
5869         * ancestor for the swap instead and transfer the memory+swap charge.
5870         */
5871        swap_memcg = mem_cgroup_id_get_online(memcg);
5872        oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg));
5873        VM_BUG_ON_PAGE(oldid, page);
5874        mem_cgroup_swap_statistics(swap_memcg, true);
5875
5876        page->mem_cgroup = NULL;
5877
5878        if (!mem_cgroup_is_root(memcg))
5879                page_counter_uncharge(&memcg->memory, 1);
5880
5881        if (memcg != swap_memcg) {
5882                if (!mem_cgroup_is_root(swap_memcg))
5883                        page_counter_charge(&swap_memcg->memsw, 1);
5884                page_counter_uncharge(&memcg->memsw, 1);
5885        }
5886
5887        /*
5888         * Interrupts should be disabled here because the caller holds the
5889         * mapping->tree_lock lock which is taken with interrupts-off. It is
5890         * important here to have the interrupts disabled because it is the
5891         * only synchronisation we have for udpating the per-CPU variables.
5892         */
5893        VM_BUG_ON(!irqs_disabled());
5894        mem_cgroup_charge_statistics(memcg, page, false, -1);
5895        memcg_check_events(memcg, page);
5896
5897        if (!mem_cgroup_is_root(memcg))
5898                css_put(&memcg->css);
5899}
5900
5901/*
5902 * mem_cgroup_try_charge_swap - try charging a swap entry
5903 * @page: page being added to swap
5904 * @entry: swap entry to charge
5905 *
5906 * Try to charge @entry to the memcg that @page belongs to.
5907 *
5908 * Returns 0 on success, -ENOMEM on failure.
5909 */
5910int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5911{
5912        struct mem_cgroup *memcg;
5913        struct page_counter *counter;
5914        unsigned short oldid;
5915
5916        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5917                return 0;
5918
5919        memcg = page->mem_cgroup;
5920
5921        /* Readahead page, never charged */
5922        if (!memcg)
5923                return 0;
5924
5925        memcg = mem_cgroup_id_get_online(memcg);
5926
5927        if (!mem_cgroup_is_root(memcg) &&
5928            !page_counter_try_charge(&memcg->swap, 1, &counter)) {
5929                mem_cgroup_id_put(memcg);
5930                return -ENOMEM;
5931        }
5932
5933        oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5934        VM_BUG_ON_PAGE(oldid, page);
5935        mem_cgroup_swap_statistics(memcg, true);
5936
5937        return 0;
5938}
5939
5940/**
5941 * mem_cgroup_uncharge_swap - uncharge a swap entry
5942 * @entry: swap entry to uncharge
5943 *
5944 * Drop the swap charge associated with @entry.
5945 */
5946void mem_cgroup_uncharge_swap(swp_entry_t entry)
5947{
5948        struct mem_cgroup *memcg;
5949        unsigned short id;
5950
5951        if (!do_swap_account)
5952                return;
5953
5954        id = swap_cgroup_record(entry, 0);
5955        rcu_read_lock();
5956        memcg = mem_cgroup_from_id(id);
5957        if (memcg) {
5958                if (!mem_cgroup_is_root(memcg)) {
5959                        if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5960                                page_counter_uncharge(&memcg->swap, 1);
5961                        else
5962                                page_counter_uncharge(&memcg->memsw, 1);
5963                }
5964                mem_cgroup_swap_statistics(memcg, false);
5965                mem_cgroup_id_put(memcg);
5966        }
5967        rcu_read_unlock();
5968}
5969
5970long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
5971{
5972        long nr_swap_pages = get_nr_swap_pages();
5973
5974        if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5975                return nr_swap_pages;
5976        for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5977                nr_swap_pages = min_t(long, nr_swap_pages,
5978                                      READ_ONCE(memcg->swap.limit) -
5979                                      page_counter_read(&memcg->swap));
5980        return nr_swap_pages;
5981}
5982
5983bool mem_cgroup_swap_full(struct page *page)
5984{
5985        struct mem_cgroup *memcg;
5986
5987        VM_BUG_ON_PAGE(!PageLocked(page), page);
5988
5989        if (vm_swap_full())
5990                return true;
5991        if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5992                return false;
5993
5994        memcg = page->mem_cgroup;
5995        if (!memcg)
5996                return false;
5997
5998        for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5999                if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
6000                        return true;
6001
6002        return false;
6003}
6004
6005/* for remember boot option*/
6006#ifdef CONFIG_MEMCG_SWAP_ENABLED
6007static int really_do_swap_account __initdata = 1;
6008#else
6009static int really_do_swap_account __initdata;
6010#endif
6011
6012static int __init enable_swap_account(char *s)
6013{
6014        if (!strcmp(s, "1"))
6015                really_do_swap_account = 1;
6016        else if (!strcmp(s, "0"))
6017                really_do_swap_account = 0;
6018        return 1;
6019}
6020__setup("swapaccount=", enable_swap_account);
6021
6022static u64 swap_current_read(struct cgroup_subsys_state *css,
6023                             struct cftype *cft)
6024{
6025        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6026
6027        return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
6028}
6029
6030static int swap_max_show(struct seq_file *m, void *v)
6031{
6032        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
6033        unsigned long max = READ_ONCE(memcg->swap.limit);
6034
6035        if (max == PAGE_COUNTER_MAX)
6036                seq_puts(m, "max\n");
6037        else
6038                seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
6039
6040        return 0;
6041}
6042
6043static ssize_t swap_max_write(struct kernfs_open_file *of,
6044                              char *buf, size_t nbytes, loff_t off)
6045{
6046        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6047        unsigned long max;
6048        int err;
6049
6050        buf = strstrip(buf);
6051        err = page_counter_memparse(buf, "max", &max);
6052        if (err)
6053                return err;
6054
6055        mutex_lock(&memcg_limit_mutex);
6056        err = page_counter_limit(&memcg->swap, max);
6057        mutex_unlock(&memcg_limit_mutex);
6058        if (err)
6059                return err;
6060
6061        return nbytes;
6062}
6063
6064static struct cftype swap_files[] = {
6065        {
6066                .name = "swap.current",
6067                .flags = CFTYPE_NOT_ON_ROOT,
6068                .read_u64 = swap_current_read,
6069        },
6070        {
6071                .name = "swap.max",
6072                .flags = CFTYPE_NOT_ON_ROOT,
6073                .seq_show = swap_max_show,
6074                .write = swap_max_write,
6075        },
6076        { }     /* terminate */
6077};
6078
6079static struct cftype memsw_cgroup_files[] = {
6080        {
6081                .name = "memsw.usage_in_bytes",
6082                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6083                .read_u64 = mem_cgroup_read_u64,
6084        },
6085        {
6086                .name = "memsw.max_usage_in_bytes",
6087                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6088                .write = mem_cgroup_reset,
6089                .read_u64 = mem_cgroup_read_u64,
6090        },
6091        {
6092                .name = "memsw.limit_in_bytes",
6093                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6094                .write = mem_cgroup_write,
6095                .read_u64 = mem_cgroup_read_u64,
6096        },
6097        {
6098                .name = "memsw.failcnt",
6099                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6100                .write = mem_cgroup_reset,
6101                .read_u64 = mem_cgroup_read_u64,
6102        },
6103        { },    /* terminate */
6104};
6105
6106static int __init mem_cgroup_swap_init(void)
6107{
6108        if (!mem_cgroup_disabled() && really_do_swap_account) {
6109                do_swap_account = 1;
6110                WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
6111                                               swap_files));
6112                WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
6113                                                  memsw_cgroup_files));
6114        }
6115        return 0;
6116}
6117subsys_initcall(mem_cgroup_swap_init);
6118
6119#endif /* CONFIG_MEMCG_SWAP */
Note: See TracBrowser for help on using the repository browser.