Root/mm/memcontrol.c

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 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
17 *
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
22 */
23
24#include <linux/res_counter.h>
25#include <linux/memcontrol.h>
26#include <linux/cgroup.h>
27#include <linux/mm.h>
28#include <linux/hugetlb.h>
29#include <linux/pagemap.h>
30#include <linux/smp.h>
31#include <linux/page-flags.h>
32#include <linux/backing-dev.h>
33#include <linux/bit_spinlock.h>
34#include <linux/rcupdate.h>
35#include <linux/limits.h>
36#include <linux/mutex.h>
37#include <linux/rbtree.h>
38#include <linux/shmem_fs.h>
39#include <linux/slab.h>
40#include <linux/swap.h>
41#include <linux/swapops.h>
42#include <linux/spinlock.h>
43#include <linux/eventfd.h>
44#include <linux/sort.h>
45#include <linux/fs.h>
46#include <linux/seq_file.h>
47#include <linux/vmalloc.h>
48#include <linux/mm_inline.h>
49#include <linux/page_cgroup.h>
50#include <linux/cpu.h>
51#include <linux/oom.h>
52#include "internal.h"
53
54#include <asm/uaccess.h>
55
56#include <trace/events/vmscan.h>
57
58struct cgroup_subsys mem_cgroup_subsys __read_mostly;
59#define MEM_CGROUP_RECLAIM_RETRIES 5
60struct mem_cgroup *root_mem_cgroup __read_mostly;
61
62#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
63/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
64int do_swap_account __read_mostly;
65
66/* for remember boot option*/
67#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
68static int really_do_swap_account __initdata = 1;
69#else
70static int really_do_swap_account __initdata = 0;
71#endif
72
73#else
74#define do_swap_account (0)
75#endif
76
77
78/*
79 * Statistics for memory cgroup.
80 */
81enum mem_cgroup_stat_index {
82    /*
83     * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
84     */
85    MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
86    MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
87    MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
88    MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
89    MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */
90    MEM_CGROUP_ON_MOVE, /* someone is moving account between groups */
91    MEM_CGROUP_STAT_NSTATS,
92};
93
94enum mem_cgroup_events_index {
95    MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */
96    MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */
97    MEM_CGROUP_EVENTS_COUNT, /* # of pages paged in/out */
98    MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */
99    MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */
100    MEM_CGROUP_EVENTS_NSTATS,
101};
102/*
103 * Per memcg event counter is incremented at every pagein/pageout. With THP,
104 * it will be incremated by the number of pages. This counter is used for
105 * for trigger some periodic events. This is straightforward and better
106 * than using jiffies etc. to handle periodic memcg event.
107 */
108enum mem_cgroup_events_target {
109    MEM_CGROUP_TARGET_THRESH,
110    MEM_CGROUP_TARGET_SOFTLIMIT,
111    MEM_CGROUP_TARGET_NUMAINFO,
112    MEM_CGROUP_NTARGETS,
113};
114#define THRESHOLDS_EVENTS_TARGET (128)
115#define SOFTLIMIT_EVENTS_TARGET (1024)
116#define NUMAINFO_EVENTS_TARGET (1024)
117
118struct mem_cgroup_stat_cpu {
119    long count[MEM_CGROUP_STAT_NSTATS];
120    unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
121    unsigned long targets[MEM_CGROUP_NTARGETS];
122};
123
124/*
125 * per-zone information in memory controller.
126 */
127struct mem_cgroup_per_zone {
128    /*
129     * spin_lock to protect the per cgroup LRU
130     */
131    struct list_head lists[NR_LRU_LISTS];
132    unsigned long count[NR_LRU_LISTS];
133
134    struct zone_reclaim_stat reclaim_stat;
135    struct rb_node tree_node; /* RB tree node */
136    unsigned long long usage_in_excess;/* Set to the value by which */
137                        /* the soft limit is exceeded*/
138    bool on_tree;
139    struct mem_cgroup *mem; /* Back pointer, we cannot */
140                        /* use container_of */
141};
142/* Macro for accessing counter */
143#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
144
145struct mem_cgroup_per_node {
146    struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
147};
148
149struct mem_cgroup_lru_info {
150    struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
151};
152
153/*
154 * Cgroups above their limits are maintained in a RB-Tree, independent of
155 * their hierarchy representation
156 */
157
158struct mem_cgroup_tree_per_zone {
159    struct rb_root rb_root;
160    spinlock_t lock;
161};
162
163struct mem_cgroup_tree_per_node {
164    struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
165};
166
167struct mem_cgroup_tree {
168    struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
169};
170
171static struct mem_cgroup_tree soft_limit_tree __read_mostly;
172
173struct mem_cgroup_threshold {
174    struct eventfd_ctx *eventfd;
175    u64 threshold;
176};
177
178/* For threshold */
179struct mem_cgroup_threshold_ary {
180    /* An array index points to threshold just below usage. */
181    int current_threshold;
182    /* Size of entries[] */
183    unsigned int size;
184    /* Array of thresholds */
185    struct mem_cgroup_threshold entries[0];
186};
187
188struct mem_cgroup_thresholds {
189    /* Primary thresholds array */
190    struct mem_cgroup_threshold_ary *primary;
191    /*
192     * Spare threshold array.
193     * This is needed to make mem_cgroup_unregister_event() "never fail".
194     * It must be able to store at least primary->size - 1 entries.
195     */
196    struct mem_cgroup_threshold_ary *spare;
197};
198
199/* for OOM */
200struct mem_cgroup_eventfd_list {
201    struct list_head list;
202    struct eventfd_ctx *eventfd;
203};
204
205static void mem_cgroup_threshold(struct mem_cgroup *mem);
206static void mem_cgroup_oom_notify(struct mem_cgroup *mem);
207
208/*
209 * The memory controller data structure. The memory controller controls both
210 * page cache and RSS per cgroup. We would eventually like to provide
211 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
212 * to help the administrator determine what knobs to tune.
213 *
214 * TODO: Add a water mark for the memory controller. Reclaim will begin when
215 * we hit the water mark. May be even add a low water mark, such that
216 * no reclaim occurs from a cgroup at it's low water mark, this is
217 * a feature that will be implemented much later in the future.
218 */
219struct mem_cgroup {
220    struct cgroup_subsys_state css;
221    /*
222     * the counter to account for memory usage
223     */
224    struct res_counter res;
225    /*
226     * the counter to account for mem+swap usage.
227     */
228    struct res_counter memsw;
229    /*
230     * Per cgroup active and inactive list, similar to the
231     * per zone LRU lists.
232     */
233    struct mem_cgroup_lru_info info;
234    /*
235     * While reclaiming in a hierarchy, we cache the last child we
236     * reclaimed from.
237     */
238    int last_scanned_child;
239    int last_scanned_node;
240#if MAX_NUMNODES > 1
241    nodemask_t scan_nodes;
242    atomic_t numainfo_events;
243    atomic_t numainfo_updating;
244#endif
245    /*
246     * Should the accounting and control be hierarchical, per subtree?
247     */
248    bool use_hierarchy;
249    atomic_t oom_lock;
250    atomic_t refcnt;
251
252    unsigned int swappiness;
253    /* OOM-Killer disable */
254    int oom_kill_disable;
255
256    /* set when res.limit == memsw.limit */
257    bool memsw_is_minimum;
258
259    /* protect arrays of thresholds */
260    struct mutex thresholds_lock;
261
262    /* thresholds for memory usage. RCU-protected */
263    struct mem_cgroup_thresholds thresholds;
264
265    /* thresholds for mem+swap usage. RCU-protected */
266    struct mem_cgroup_thresholds memsw_thresholds;
267
268    /* For oom notifier event fd */
269    struct list_head oom_notify;
270
271    /*
272     * Should we move charges of a task when a task is moved into this
273     * mem_cgroup ? And what type of charges should we move ?
274     */
275    unsigned long move_charge_at_immigrate;
276    /*
277     * percpu counter.
278     */
279    struct mem_cgroup_stat_cpu *stat;
280    /*
281     * used when a cpu is offlined or other synchronizations
282     * See mem_cgroup_read_stat().
283     */
284    struct mem_cgroup_stat_cpu nocpu_base;
285    spinlock_t pcp_counter_lock;
286};
287
288/* Stuffs for move charges at task migration. */
289/*
290 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
291 * left-shifted bitmap of these types.
292 */
293enum move_type {
294    MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
295    MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
296    NR_MOVE_TYPE,
297};
298
299/* "mc" and its members are protected by cgroup_mutex */
300static struct move_charge_struct {
301    spinlock_t lock; /* for from, to */
302    struct mem_cgroup *from;
303    struct mem_cgroup *to;
304    unsigned long precharge;
305    unsigned long moved_charge;
306    unsigned long moved_swap;
307    struct task_struct *moving_task; /* a task moving charges */
308    wait_queue_head_t waitq; /* a waitq for other context */
309} mc = {
310    .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
311    .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
312};
313
314static bool move_anon(void)
315{
316    return test_bit(MOVE_CHARGE_TYPE_ANON,
317                    &mc.to->move_charge_at_immigrate);
318}
319
320static bool move_file(void)
321{
322    return test_bit(MOVE_CHARGE_TYPE_FILE,
323                    &mc.to->move_charge_at_immigrate);
324}
325
326/*
327 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
328 * limit reclaim to prevent infinite loops, if they ever occur.
329 */
330#define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
331#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
332
333enum charge_type {
334    MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
335    MEM_CGROUP_CHARGE_TYPE_MAPPED,
336    MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
337    MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
338    MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
339    MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
340    NR_CHARGE_TYPE,
341};
342
343/* for encoding cft->private value on file */
344#define _MEM (0)
345#define _MEMSWAP (1)
346#define _OOM_TYPE (2)
347#define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
348#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
349#define MEMFILE_ATTR(val) ((val) & 0xffff)
350/* Used for OOM nofiier */
351#define OOM_CONTROL (0)
352
353/*
354 * Reclaim flags for mem_cgroup_hierarchical_reclaim
355 */
356#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
357#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
358#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
359#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
360#define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
361#define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
362
363static void mem_cgroup_get(struct mem_cgroup *mem);
364static void mem_cgroup_put(struct mem_cgroup *mem);
365static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
366static void drain_all_stock_async(struct mem_cgroup *mem);
367
368static struct mem_cgroup_per_zone *
369mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
370{
371    return &mem->info.nodeinfo[nid]->zoneinfo[zid];
372}
373
374struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem)
375{
376    return &mem->css;
377}
378
379static struct mem_cgroup_per_zone *
380page_cgroup_zoneinfo(struct mem_cgroup *mem, struct page *page)
381{
382    int nid = page_to_nid(page);
383    int zid = page_zonenum(page);
384
385    return mem_cgroup_zoneinfo(mem, nid, zid);
386}
387
388static struct mem_cgroup_tree_per_zone *
389soft_limit_tree_node_zone(int nid, int zid)
390{
391    return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
392}
393
394static struct mem_cgroup_tree_per_zone *
395soft_limit_tree_from_page(struct page *page)
396{
397    int nid = page_to_nid(page);
398    int zid = page_zonenum(page);
399
400    return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
401}
402
403static void
404__mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
405                struct mem_cgroup_per_zone *mz,
406                struct mem_cgroup_tree_per_zone *mctz,
407                unsigned long long new_usage_in_excess)
408{
409    struct rb_node **p = &mctz->rb_root.rb_node;
410    struct rb_node *parent = NULL;
411    struct mem_cgroup_per_zone *mz_node;
412
413    if (mz->on_tree)
414        return;
415
416    mz->usage_in_excess = new_usage_in_excess;
417    if (!mz->usage_in_excess)
418        return;
419    while (*p) {
420        parent = *p;
421        mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
422                    tree_node);
423        if (mz->usage_in_excess < mz_node->usage_in_excess)
424            p = &(*p)->rb_left;
425        /*
426         * We can't avoid mem cgroups that are over their soft
427         * limit by the same amount
428         */
429        else if (mz->usage_in_excess >= mz_node->usage_in_excess)
430            p = &(*p)->rb_right;
431    }
432    rb_link_node(&mz->tree_node, parent, p);
433    rb_insert_color(&mz->tree_node, &mctz->rb_root);
434    mz->on_tree = true;
435}
436
437static void
438__mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
439                struct mem_cgroup_per_zone *mz,
440                struct mem_cgroup_tree_per_zone *mctz)
441{
442    if (!mz->on_tree)
443        return;
444    rb_erase(&mz->tree_node, &mctz->rb_root);
445    mz->on_tree = false;
446}
447
448static void
449mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
450                struct mem_cgroup_per_zone *mz,
451                struct mem_cgroup_tree_per_zone *mctz)
452{
453    spin_lock(&mctz->lock);
454    __mem_cgroup_remove_exceeded(mem, mz, mctz);
455    spin_unlock(&mctz->lock);
456}
457
458
459static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page)
460{
461    unsigned long long excess;
462    struct mem_cgroup_per_zone *mz;
463    struct mem_cgroup_tree_per_zone *mctz;
464    int nid = page_to_nid(page);
465    int zid = page_zonenum(page);
466    mctz = soft_limit_tree_from_page(page);
467
468    /*
469     * Necessary to update all ancestors when hierarchy is used.
470     * because their event counter is not touched.
471     */
472    for (; mem; mem = parent_mem_cgroup(mem)) {
473        mz = mem_cgroup_zoneinfo(mem, nid, zid);
474        excess = res_counter_soft_limit_excess(&mem->res);
475        /*
476         * We have to update the tree if mz is on RB-tree or
477         * mem is over its softlimit.
478         */
479        if (excess || mz->on_tree) {
480            spin_lock(&mctz->lock);
481            /* if on-tree, remove it */
482            if (mz->on_tree)
483                __mem_cgroup_remove_exceeded(mem, mz, mctz);
484            /*
485             * Insert again. mz->usage_in_excess will be updated.
486             * If excess is 0, no tree ops.
487             */
488            __mem_cgroup_insert_exceeded(mem, mz, mctz, excess);
489            spin_unlock(&mctz->lock);
490        }
491    }
492}
493
494static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem)
495{
496    int node, zone;
497    struct mem_cgroup_per_zone *mz;
498    struct mem_cgroup_tree_per_zone *mctz;
499
500    for_each_node_state(node, N_POSSIBLE) {
501        for (zone = 0; zone < MAX_NR_ZONES; zone++) {
502            mz = mem_cgroup_zoneinfo(mem, node, zone);
503            mctz = soft_limit_tree_node_zone(node, zone);
504            mem_cgroup_remove_exceeded(mem, mz, mctz);
505        }
506    }
507}
508
509static struct mem_cgroup_per_zone *
510__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
511{
512    struct rb_node *rightmost = NULL;
513    struct mem_cgroup_per_zone *mz;
514
515retry:
516    mz = NULL;
517    rightmost = rb_last(&mctz->rb_root);
518    if (!rightmost)
519        goto done; /* Nothing to reclaim from */
520
521    mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
522    /*
523     * Remove the node now but someone else can add it back,
524     * we will to add it back at the end of reclaim to its correct
525     * position in the tree.
526     */
527    __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
528    if (!res_counter_soft_limit_excess(&mz->mem->res) ||
529        !css_tryget(&mz->mem->css))
530        goto retry;
531done:
532    return mz;
533}
534
535static struct mem_cgroup_per_zone *
536mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
537{
538    struct mem_cgroup_per_zone *mz;
539
540    spin_lock(&mctz->lock);
541    mz = __mem_cgroup_largest_soft_limit_node(mctz);
542    spin_unlock(&mctz->lock);
543    return mz;
544}
545
546/*
547 * Implementation Note: reading percpu statistics for memcg.
548 *
549 * Both of vmstat[] and percpu_counter has threshold and do periodic
550 * synchronization to implement "quick" read. There are trade-off between
551 * reading cost and precision of value. Then, we may have a chance to implement
552 * a periodic synchronizion of counter in memcg's counter.
553 *
554 * But this _read() function is used for user interface now. The user accounts
555 * memory usage by memory cgroup and he _always_ requires exact value because
556 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
557 * have to visit all online cpus and make sum. So, for now, unnecessary
558 * synchronization is not implemented. (just implemented for cpu hotplug)
559 *
560 * If there are kernel internal actions which can make use of some not-exact
561 * value, and reading all cpu value can be performance bottleneck in some
562 * common workload, threashold and synchonization as vmstat[] should be
563 * implemented.
564 */
565static long mem_cgroup_read_stat(struct mem_cgroup *mem,
566                 enum mem_cgroup_stat_index idx)
567{
568    long val = 0;
569    int cpu;
570
571    get_online_cpus();
572    for_each_online_cpu(cpu)
573        val += per_cpu(mem->stat->count[idx], cpu);
574#ifdef CONFIG_HOTPLUG_CPU
575    spin_lock(&mem->pcp_counter_lock);
576    val += mem->nocpu_base.count[idx];
577    spin_unlock(&mem->pcp_counter_lock);
578#endif
579    put_online_cpus();
580    return val;
581}
582
583static void mem_cgroup_swap_statistics(struct mem_cgroup *mem,
584                     bool charge)
585{
586    int val = (charge) ? 1 : -1;
587    this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
588}
589
590void mem_cgroup_pgfault(struct mem_cgroup *mem, int val)
591{
592    this_cpu_add(mem->stat->events[MEM_CGROUP_EVENTS_PGFAULT], val);
593}
594
595void mem_cgroup_pgmajfault(struct mem_cgroup *mem, int val)
596{
597    this_cpu_add(mem->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT], val);
598}
599
600static unsigned long mem_cgroup_read_events(struct mem_cgroup *mem,
601                        enum mem_cgroup_events_index idx)
602{
603    unsigned long val = 0;
604    int cpu;
605
606    for_each_online_cpu(cpu)
607        val += per_cpu(mem->stat->events[idx], cpu);
608#ifdef CONFIG_HOTPLUG_CPU
609    spin_lock(&mem->pcp_counter_lock);
610    val += mem->nocpu_base.events[idx];
611    spin_unlock(&mem->pcp_counter_lock);
612#endif
613    return val;
614}
615
616static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
617                     bool file, int nr_pages)
618{
619    preempt_disable();
620
621    if (file)
622        __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], nr_pages);
623    else
624        __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], nr_pages);
625
626    /* pagein of a big page is an event. So, ignore page size */
627    if (nr_pages > 0)
628        __this_cpu_inc(mem->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
629    else {
630        __this_cpu_inc(mem->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
631        nr_pages = -nr_pages; /* for event */
632    }
633
634    __this_cpu_add(mem->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages);
635
636    preempt_enable();
637}
638
639static unsigned long
640mem_cgroup_get_zonestat_node(struct mem_cgroup *mem, int nid, enum lru_list idx)
641{
642    struct mem_cgroup_per_zone *mz;
643    u64 total = 0;
644    int zid;
645
646    for (zid = 0; zid < MAX_NR_ZONES; zid++) {
647        mz = mem_cgroup_zoneinfo(mem, nid, zid);
648        total += MEM_CGROUP_ZSTAT(mz, idx);
649    }
650    return total;
651}
652static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
653                    enum lru_list idx)
654{
655    int nid;
656    u64 total = 0;
657
658    for_each_online_node(nid)
659        total += mem_cgroup_get_zonestat_node(mem, nid, idx);
660    return total;
661}
662
663static bool __memcg_event_check(struct mem_cgroup *mem, int target)
664{
665    unsigned long val, next;
666
667    val = this_cpu_read(mem->stat->events[MEM_CGROUP_EVENTS_COUNT]);
668    next = this_cpu_read(mem->stat->targets[target]);
669    /* from time_after() in jiffies.h */
670    return ((long)next - (long)val < 0);
671}
672
673static void __mem_cgroup_target_update(struct mem_cgroup *mem, int target)
674{
675    unsigned long val, next;
676
677    val = this_cpu_read(mem->stat->events[MEM_CGROUP_EVENTS_COUNT]);
678
679    switch (target) {
680    case MEM_CGROUP_TARGET_THRESH:
681        next = val + THRESHOLDS_EVENTS_TARGET;
682        break;
683    case MEM_CGROUP_TARGET_SOFTLIMIT:
684        next = val + SOFTLIMIT_EVENTS_TARGET;
685        break;
686    case MEM_CGROUP_TARGET_NUMAINFO:
687        next = val + NUMAINFO_EVENTS_TARGET;
688        break;
689    default:
690        return;
691    }
692
693    this_cpu_write(mem->stat->targets[target], next);
694}
695
696/*
697 * Check events in order.
698 *
699 */
700static void memcg_check_events(struct mem_cgroup *mem, struct page *page)
701{
702    /* threshold event is triggered in finer grain than soft limit */
703    if (unlikely(__memcg_event_check(mem, MEM_CGROUP_TARGET_THRESH))) {
704        mem_cgroup_threshold(mem);
705        __mem_cgroup_target_update(mem, MEM_CGROUP_TARGET_THRESH);
706        if (unlikely(__memcg_event_check(mem,
707                 MEM_CGROUP_TARGET_SOFTLIMIT))) {
708            mem_cgroup_update_tree(mem, page);
709            __mem_cgroup_target_update(mem,
710                           MEM_CGROUP_TARGET_SOFTLIMIT);
711        }
712#if MAX_NUMNODES > 1
713        if (unlikely(__memcg_event_check(mem,
714            MEM_CGROUP_TARGET_NUMAINFO))) {
715            atomic_inc(&mem->numainfo_events);
716            __mem_cgroup_target_update(mem,
717                MEM_CGROUP_TARGET_NUMAINFO);
718        }
719#endif
720    }
721}
722
723static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
724{
725    return container_of(cgroup_subsys_state(cont,
726                mem_cgroup_subsys_id), struct mem_cgroup,
727                css);
728}
729
730struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
731{
732    /*
733     * mm_update_next_owner() may clear mm->owner to NULL
734     * if it races with swapoff, page migration, etc.
735     * So this can be called with p == NULL.
736     */
737    if (unlikely(!p))
738        return NULL;
739
740    return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
741                struct mem_cgroup, css);
742}
743
744struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
745{
746    struct mem_cgroup *mem = NULL;
747
748    if (!mm)
749        return NULL;
750    /*
751     * Because we have no locks, mm->owner's may be being moved to other
752     * cgroup. We use css_tryget() here even if this looks
753     * pessimistic (rather than adding locks here).
754     */
755    rcu_read_lock();
756    do {
757        mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
758        if (unlikely(!mem))
759            break;
760    } while (!css_tryget(&mem->css));
761    rcu_read_unlock();
762    return mem;
763}
764
765/* The caller has to guarantee "mem" exists before calling this */
766static struct mem_cgroup *mem_cgroup_start_loop(struct mem_cgroup *mem)
767{
768    struct cgroup_subsys_state *css;
769    int found;
770
771    if (!mem) /* ROOT cgroup has the smallest ID */
772        return root_mem_cgroup; /*css_put/get against root is ignored*/
773    if (!mem->use_hierarchy) {
774        if (css_tryget(&mem->css))
775            return mem;
776        return NULL;
777    }
778    rcu_read_lock();
779    /*
780     * searching a memory cgroup which has the smallest ID under given
781     * ROOT cgroup. (ID >= 1)
782     */
783    css = css_get_next(&mem_cgroup_subsys, 1, &mem->css, &found);
784    if (css && css_tryget(css))
785        mem = container_of(css, struct mem_cgroup, css);
786    else
787        mem = NULL;
788    rcu_read_unlock();
789    return mem;
790}
791
792static struct mem_cgroup *mem_cgroup_get_next(struct mem_cgroup *iter,
793                    struct mem_cgroup *root,
794                    bool cond)
795{
796    int nextid = css_id(&iter->css) + 1;
797    int found;
798    int hierarchy_used;
799    struct cgroup_subsys_state *css;
800
801    hierarchy_used = iter->use_hierarchy;
802
803    css_put(&iter->css);
804    /* If no ROOT, walk all, ignore hierarchy */
805    if (!cond || (root && !hierarchy_used))
806        return NULL;
807
808    if (!root)
809        root = root_mem_cgroup;
810
811    do {
812        iter = NULL;
813        rcu_read_lock();
814
815        css = css_get_next(&mem_cgroup_subsys, nextid,
816                &root->css, &found);
817        if (css && css_tryget(css))
818            iter = container_of(css, struct mem_cgroup, css);
819        rcu_read_unlock();
820        /* If css is NULL, no more cgroups will be found */
821        nextid = found + 1;
822    } while (css && !iter);
823
824    return iter;
825}
826/*
827 * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please
828 * be careful that "break" loop is not allowed. We have reference count.
829 * Instead of that modify "cond" to be false and "continue" to exit the loop.
830 */
831#define for_each_mem_cgroup_tree_cond(iter, root, cond) \
832    for (iter = mem_cgroup_start_loop(root);\
833         iter != NULL;\
834         iter = mem_cgroup_get_next(iter, root, cond))
835
836#define for_each_mem_cgroup_tree(iter, root) \
837    for_each_mem_cgroup_tree_cond(iter, root, true)
838
839#define for_each_mem_cgroup_all(iter) \
840    for_each_mem_cgroup_tree_cond(iter, NULL, true)
841
842
843static inline bool mem_cgroup_is_root(struct mem_cgroup *mem)
844{
845    return (mem == root_mem_cgroup);
846}
847
848void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
849{
850    struct mem_cgroup *mem;
851
852    if (!mm)
853        return;
854
855    rcu_read_lock();
856    mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
857    if (unlikely(!mem))
858        goto out;
859
860    switch (idx) {
861    case PGMAJFAULT:
862        mem_cgroup_pgmajfault(mem, 1);
863        break;
864    case PGFAULT:
865        mem_cgroup_pgfault(mem, 1);
866        break;
867    default:
868        BUG();
869    }
870out:
871    rcu_read_unlock();
872}
873EXPORT_SYMBOL(mem_cgroup_count_vm_event);
874
875/*
876 * Following LRU functions are allowed to be used without PCG_LOCK.
877 * Operations are called by routine of global LRU independently from memcg.
878 * What we have to take care of here is validness of pc->mem_cgroup.
879 *
880 * Changes to pc->mem_cgroup happens when
881 * 1. charge
882 * 2. moving account
883 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
884 * It is added to LRU before charge.
885 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
886 * When moving account, the page is not on LRU. It's isolated.
887 */
888
889void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
890{
891    struct page_cgroup *pc;
892    struct mem_cgroup_per_zone *mz;
893
894    if (mem_cgroup_disabled())
895        return;
896    pc = lookup_page_cgroup(page);
897    /* can happen while we handle swapcache. */
898    if (!TestClearPageCgroupAcctLRU(pc))
899        return;
900    VM_BUG_ON(!pc->mem_cgroup);
901    /*
902     * We don't check PCG_USED bit. It's cleared when the "page" is finally
903     * removed from global LRU.
904     */
905    mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
906    /* huge page split is done under lru_lock. so, we have no races. */
907    MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page);
908    if (mem_cgroup_is_root(pc->mem_cgroup))
909        return;
910    VM_BUG_ON(list_empty(&pc->lru));
911    list_del_init(&pc->lru);
912}
913
914void mem_cgroup_del_lru(struct page *page)
915{
916    mem_cgroup_del_lru_list(page, page_lru(page));
917}
918
919/*
920 * Writeback is about to end against a page which has been marked for immediate
921 * reclaim. If it still appears to be reclaimable, move it to the tail of the
922 * inactive list.
923 */
924void mem_cgroup_rotate_reclaimable_page(struct page *page)
925{
926    struct mem_cgroup_per_zone *mz;
927    struct page_cgroup *pc;
928    enum lru_list lru = page_lru(page);
929
930    if (mem_cgroup_disabled())
931        return;
932
933    pc = lookup_page_cgroup(page);
934    /* unused or root page is not rotated. */
935    if (!PageCgroupUsed(pc))
936        return;
937    /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
938    smp_rmb();
939    if (mem_cgroup_is_root(pc->mem_cgroup))
940        return;
941    mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
942    list_move_tail(&pc->lru, &mz->lists[lru]);
943}
944
945void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
946{
947    struct mem_cgroup_per_zone *mz;
948    struct page_cgroup *pc;
949
950    if (mem_cgroup_disabled())
951        return;
952
953    pc = lookup_page_cgroup(page);
954    /* unused or root page is not rotated. */
955    if (!PageCgroupUsed(pc))
956        return;
957    /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
958    smp_rmb();
959    if (mem_cgroup_is_root(pc->mem_cgroup))
960        return;
961    mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
962    list_move(&pc->lru, &mz->lists[lru]);
963}
964
965void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
966{
967    struct page_cgroup *pc;
968    struct mem_cgroup_per_zone *mz;
969
970    if (mem_cgroup_disabled())
971        return;
972    pc = lookup_page_cgroup(page);
973    VM_BUG_ON(PageCgroupAcctLRU(pc));
974    if (!PageCgroupUsed(pc))
975        return;
976    /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
977    smp_rmb();
978    mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
979    /* huge page split is done under lru_lock. so, we have no races. */
980    MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page);
981    SetPageCgroupAcctLRU(pc);
982    if (mem_cgroup_is_root(pc->mem_cgroup))
983        return;
984    list_add(&pc->lru, &mz->lists[lru]);
985}
986
987/*
988 * At handling SwapCache and other FUSE stuff, pc->mem_cgroup may be changed
989 * while it's linked to lru because the page may be reused after it's fully
990 * uncharged. To handle that, unlink page_cgroup from LRU when charge it again.
991 * It's done under lock_page and expected that zone->lru_lock isnever held.
992 */
993static void mem_cgroup_lru_del_before_commit(struct page *page)
994{
995    unsigned long flags;
996    struct zone *zone = page_zone(page);
997    struct page_cgroup *pc = lookup_page_cgroup(page);
998
999    /*
1000     * Doing this check without taking ->lru_lock seems wrong but this
1001     * is safe. Because if page_cgroup's USED bit is unset, the page
1002     * will not be added to any memcg's LRU. If page_cgroup's USED bit is
1003     * set, the commit after this will fail, anyway.
1004     * This all charge/uncharge is done under some mutual execustion.
1005     * So, we don't need to taking care of changes in USED bit.
1006     */
1007    if (likely(!PageLRU(page)))
1008        return;
1009
1010    spin_lock_irqsave(&zone->lru_lock, flags);
1011    /*
1012     * Forget old LRU when this page_cgroup is *not* used. This Used bit
1013     * is guarded by lock_page() because the page is SwapCache.
1014     */
1015    if (!PageCgroupUsed(pc))
1016        mem_cgroup_del_lru_list(page, page_lru(page));
1017    spin_unlock_irqrestore(&zone->lru_lock, flags);
1018}
1019
1020static void mem_cgroup_lru_add_after_commit(struct page *page)
1021{
1022    unsigned long flags;
1023    struct zone *zone = page_zone(page);
1024    struct page_cgroup *pc = lookup_page_cgroup(page);
1025
1026    /* taking care of that the page is added to LRU while we commit it */
1027    if (likely(!PageLRU(page)))
1028        return;
1029    spin_lock_irqsave(&zone->lru_lock, flags);
1030    /* link when the page is linked to LRU but page_cgroup isn't */
1031    if (PageLRU(page) && !PageCgroupAcctLRU(pc))
1032        mem_cgroup_add_lru_list(page, page_lru(page));
1033    spin_unlock_irqrestore(&zone->lru_lock, flags);
1034}
1035
1036
1037void mem_cgroup_move_lists(struct page *page,
1038               enum lru_list from, enum lru_list to)
1039{
1040    if (mem_cgroup_disabled())
1041        return;
1042    mem_cgroup_del_lru_list(page, from);
1043    mem_cgroup_add_lru_list(page, to);
1044}
1045
1046int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
1047{
1048    int ret;
1049    struct mem_cgroup *curr = NULL;
1050    struct task_struct *p;
1051
1052    p = find_lock_task_mm(task);
1053    if (!p)
1054        return 0;
1055    curr = try_get_mem_cgroup_from_mm(p->mm);
1056    task_unlock(p);
1057    if (!curr)
1058        return 0;
1059    /*
1060     * We should check use_hierarchy of "mem" not "curr". Because checking
1061     * use_hierarchy of "curr" here make this function true if hierarchy is
1062     * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
1063     * hierarchy(even if use_hierarchy is disabled in "mem").
1064     */
1065    if (mem->use_hierarchy)
1066        ret = css_is_ancestor(&curr->css, &mem->css);
1067    else
1068        ret = (curr == mem);
1069    css_put(&curr->css);
1070    return ret;
1071}
1072
1073static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
1074{
1075    unsigned long active;
1076    unsigned long inactive;
1077    unsigned long gb;
1078    unsigned long inactive_ratio;
1079
1080    inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
1081    active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
1082
1083    gb = (inactive + active) >> (30 - PAGE_SHIFT);
1084    if (gb)
1085        inactive_ratio = int_sqrt(10 * gb);
1086    else
1087        inactive_ratio = 1;
1088
1089    if (present_pages) {
1090        present_pages[0] = inactive;
1091        present_pages[1] = active;
1092    }
1093
1094    return inactive_ratio;
1095}
1096
1097int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
1098{
1099    unsigned long active;
1100    unsigned long inactive;
1101    unsigned long present_pages[2];
1102    unsigned long inactive_ratio;
1103
1104    inactive_ratio = calc_inactive_ratio(memcg, present_pages);
1105
1106    inactive = present_pages[0];
1107    active = present_pages[1];
1108
1109    if (inactive * inactive_ratio < active)
1110        return 1;
1111
1112    return 0;
1113}
1114
1115int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
1116{
1117    unsigned long active;
1118    unsigned long inactive;
1119
1120    inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
1121    active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
1122
1123    return (active > inactive);
1124}
1125
1126unsigned long mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg,
1127                        struct zone *zone,
1128                        enum lru_list lru)
1129{
1130    int nid = zone_to_nid(zone);
1131    int zid = zone_idx(zone);
1132    struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
1133
1134    return MEM_CGROUP_ZSTAT(mz, lru);
1135}
1136
1137static unsigned long mem_cgroup_node_nr_file_lru_pages(struct mem_cgroup *memcg,
1138                            int nid)
1139{
1140    unsigned long ret;
1141
1142    ret = mem_cgroup_get_zonestat_node(memcg, nid, LRU_INACTIVE_FILE) +
1143        mem_cgroup_get_zonestat_node(memcg, nid, LRU_ACTIVE_FILE);
1144
1145    return ret;
1146}
1147
1148static unsigned long mem_cgroup_node_nr_anon_lru_pages(struct mem_cgroup *memcg,
1149                            int nid)
1150{
1151    unsigned long ret;
1152
1153    ret = mem_cgroup_get_zonestat_node(memcg, nid, LRU_INACTIVE_ANON) +
1154        mem_cgroup_get_zonestat_node(memcg, nid, LRU_ACTIVE_ANON);
1155    return ret;
1156}
1157
1158#if MAX_NUMNODES > 1
1159static unsigned long mem_cgroup_nr_file_lru_pages(struct mem_cgroup *memcg)
1160{
1161    u64 total = 0;
1162    int nid;
1163
1164    for_each_node_state(nid, N_HIGH_MEMORY)
1165        total += mem_cgroup_node_nr_file_lru_pages(memcg, nid);
1166
1167    return total;
1168}
1169
1170static unsigned long mem_cgroup_nr_anon_lru_pages(struct mem_cgroup *memcg)
1171{
1172    u64 total = 0;
1173    int nid;
1174
1175    for_each_node_state(nid, N_HIGH_MEMORY)
1176        total += mem_cgroup_node_nr_anon_lru_pages(memcg, nid);
1177
1178    return total;
1179}
1180
1181static unsigned long
1182mem_cgroup_node_nr_unevictable_lru_pages(struct mem_cgroup *memcg, int nid)
1183{
1184    return mem_cgroup_get_zonestat_node(memcg, nid, LRU_UNEVICTABLE);
1185}
1186
1187static unsigned long
1188mem_cgroup_nr_unevictable_lru_pages(struct mem_cgroup *memcg)
1189{
1190    u64 total = 0;
1191    int nid;
1192
1193    for_each_node_state(nid, N_HIGH_MEMORY)
1194        total += mem_cgroup_node_nr_unevictable_lru_pages(memcg, nid);
1195
1196    return total;
1197}
1198
1199static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
1200                            int nid)
1201{
1202    enum lru_list l;
1203    u64 total = 0;
1204
1205    for_each_lru(l)
1206        total += mem_cgroup_get_zonestat_node(memcg, nid, l);
1207
1208    return total;
1209}
1210
1211static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg)
1212{
1213    u64 total = 0;
1214    int nid;
1215
1216    for_each_node_state(nid, N_HIGH_MEMORY)
1217        total += mem_cgroup_node_nr_lru_pages(memcg, nid);
1218
1219    return total;
1220}
1221#endif /* CONFIG_NUMA */
1222
1223struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
1224                              struct zone *zone)
1225{
1226    int nid = zone_to_nid(zone);
1227    int zid = zone_idx(zone);
1228    struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
1229
1230    return &mz->reclaim_stat;
1231}
1232
1233struct zone_reclaim_stat *
1234mem_cgroup_get_reclaim_stat_from_page(struct page *page)
1235{
1236    struct page_cgroup *pc;
1237    struct mem_cgroup_per_zone *mz;
1238
1239    if (mem_cgroup_disabled())
1240        return NULL;
1241
1242    pc = lookup_page_cgroup(page);
1243    if (!PageCgroupUsed(pc))
1244        return NULL;
1245    /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1246    smp_rmb();
1247    mz = page_cgroup_zoneinfo(pc->mem_cgroup, page);
1248    return &mz->reclaim_stat;
1249}
1250
1251unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
1252                    struct list_head *dst,
1253                    unsigned long *scanned, int order,
1254                    int mode, struct zone *z,
1255                    struct mem_cgroup *mem_cont,
1256                    int active, int file)
1257{
1258    unsigned long nr_taken = 0;
1259    struct page *page;
1260    unsigned long scan;
1261    LIST_HEAD(pc_list);
1262    struct list_head *src;
1263    struct page_cgroup *pc, *tmp;
1264    int nid = zone_to_nid(z);
1265    int zid = zone_idx(z);
1266    struct mem_cgroup_per_zone *mz;
1267    int lru = LRU_FILE * file + active;
1268    int ret;
1269
1270    BUG_ON(!mem_cont);
1271    mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
1272    src = &mz->lists[lru];
1273
1274    scan = 0;
1275    list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
1276        if (scan >= nr_to_scan)
1277            break;
1278
1279        if (unlikely(!PageCgroupUsed(pc)))
1280            continue;
1281
1282        page = lookup_cgroup_page(pc);
1283
1284        if (unlikely(!PageLRU(page)))
1285            continue;
1286
1287        scan++;
1288        ret = __isolate_lru_page(page, mode, file);
1289        switch (ret) {
1290        case 0:
1291            list_move(&page->lru, dst);
1292            mem_cgroup_del_lru(page);
1293            nr_taken += hpage_nr_pages(page);
1294            break;
1295        case -EBUSY:
1296            /* we don't affect global LRU but rotate in our LRU */
1297            mem_cgroup_rotate_lru_list(page, page_lru(page));
1298            break;
1299        default:
1300            break;
1301        }
1302    }
1303
1304    *scanned = scan;
1305
1306    trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken,
1307                      0, 0, 0, mode);
1308
1309    return nr_taken;
1310}
1311
1312#define mem_cgroup_from_res_counter(counter, member) \
1313    container_of(counter, struct mem_cgroup, member)
1314
1315/**
1316 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1317 * @mem: the memory cgroup
1318 *
1319 * Returns the maximum amount of memory @mem can be charged with, in
1320 * pages.
1321 */
1322static unsigned long mem_cgroup_margin(struct mem_cgroup *mem)
1323{
1324    unsigned long long margin;
1325
1326    margin = res_counter_margin(&mem->res);
1327    if (do_swap_account)
1328        margin = min(margin, res_counter_margin(&mem->memsw));
1329    return margin >> PAGE_SHIFT;
1330}
1331
1332static unsigned int get_swappiness(struct mem_cgroup *memcg)
1333{
1334    struct cgroup *cgrp = memcg->css.cgroup;
1335
1336    /* root ? */
1337    if (cgrp->parent == NULL)
1338        return vm_swappiness;
1339
1340    return memcg->swappiness;
1341}
1342
1343static void mem_cgroup_start_move(struct mem_cgroup *mem)
1344{
1345    int cpu;
1346
1347    get_online_cpus();
1348    spin_lock(&mem->pcp_counter_lock);
1349    for_each_online_cpu(cpu)
1350        per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1;
1351    mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1;
1352    spin_unlock(&mem->pcp_counter_lock);
1353    put_online_cpus();
1354
1355    synchronize_rcu();
1356}
1357
1358static void mem_cgroup_end_move(struct mem_cgroup *mem)
1359{
1360    int cpu;
1361
1362    if (!mem)
1363        return;
1364    get_online_cpus();
1365    spin_lock(&mem->pcp_counter_lock);
1366    for_each_online_cpu(cpu)
1367        per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1;
1368    mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1;
1369    spin_unlock(&mem->pcp_counter_lock);
1370    put_online_cpus();
1371}
1372/*
1373 * 2 routines for checking "mem" is under move_account() or not.
1374 *
1375 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1376 * for avoiding race in accounting. If true,
1377 * pc->mem_cgroup may be overwritten.
1378 *
1379 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1380 * under hierarchy of moving cgroups. This is for
1381 * waiting at hith-memory prressure caused by "move".
1382 */
1383
1384static bool mem_cgroup_stealed(struct mem_cgroup *mem)
1385{
1386    VM_BUG_ON(!rcu_read_lock_held());
1387    return this_cpu_read(mem->stat->count[MEM_CGROUP_ON_MOVE]) > 0;
1388}
1389
1390static bool mem_cgroup_under_move(struct mem_cgroup *mem)
1391{
1392    struct mem_cgroup *from;
1393    struct mem_cgroup *to;
1394    bool ret = false;
1395    /*
1396     * Unlike task_move routines, we access mc.to, mc.from not under
1397     * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1398     */
1399    spin_lock(&mc.lock);
1400    from = mc.from;
1401    to = mc.to;
1402    if (!from)
1403        goto unlock;
1404    if (from == mem || to == mem
1405        || (mem->use_hierarchy && css_is_ancestor(&from->css, &mem->css))
1406        || (mem->use_hierarchy && css_is_ancestor(&to->css, &mem->css)))
1407        ret = true;
1408unlock:
1409    spin_unlock(&mc.lock);
1410    return ret;
1411}
1412
1413static bool mem_cgroup_wait_acct_move(struct mem_cgroup *mem)
1414{
1415    if (mc.moving_task && current != mc.moving_task) {
1416        if (mem_cgroup_under_move(mem)) {
1417            DEFINE_WAIT(wait);
1418            prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1419            /* moving charge context might have finished. */
1420            if (mc.moving_task)
1421                schedule();
1422            finish_wait(&mc.waitq, &wait);
1423            return true;
1424        }
1425    }
1426    return false;
1427}
1428
1429/**
1430 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1431 * @memcg: The memory cgroup that went over limit
1432 * @p: Task that is going to be killed
1433 *
1434 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1435 * enabled
1436 */
1437void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1438{
1439    struct cgroup *task_cgrp;
1440    struct cgroup *mem_cgrp;
1441    /*
1442     * Need a buffer in BSS, can't rely on allocations. The code relies
1443     * on the assumption that OOM is serialized for memory controller.
1444     * If this assumption is broken, revisit this code.
1445     */
1446    static char memcg_name[PATH_MAX];
1447    int ret;
1448
1449    if (!memcg || !p)
1450        return;
1451
1452
1453    rcu_read_lock();
1454
1455    mem_cgrp = memcg->css.cgroup;
1456    task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1457
1458    ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1459    if (ret < 0) {
1460        /*
1461         * Unfortunately, we are unable to convert to a useful name
1462         * But we'll still print out the usage information
1463         */
1464        rcu_read_unlock();
1465        goto done;
1466    }
1467    rcu_read_unlock();
1468
1469    printk(KERN_INFO "Task in %s killed", memcg_name);
1470
1471    rcu_read_lock();
1472    ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1473    if (ret < 0) {
1474        rcu_read_unlock();
1475        goto done;
1476    }
1477    rcu_read_unlock();
1478
1479    /*
1480     * Continues from above, so we don't need an KERN_ level
1481     */
1482    printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1483done:
1484
1485    printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1486        res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1487        res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1488        res_counter_read_u64(&memcg->res, RES_FAILCNT));
1489    printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1490        "failcnt %llu\n",
1491        res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1492        res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1493        res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1494}
1495
1496/*
1497 * This function returns the number of memcg under hierarchy tree. Returns
1498 * 1(self count) if no children.
1499 */
1500static int mem_cgroup_count_children(struct mem_cgroup *mem)
1501{
1502    int num = 0;
1503    struct mem_cgroup *iter;
1504
1505    for_each_mem_cgroup_tree(iter, mem)
1506        num++;
1507    return num;
1508}
1509
1510/*
1511 * Return the memory (and swap, if configured) limit for a memcg.
1512 */
1513u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1514{
1515    u64 limit;
1516    u64 memsw;
1517
1518    limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1519    limit += total_swap_pages << PAGE_SHIFT;
1520
1521    memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1522    /*
1523     * If memsw is finite and limits the amount of swap space available
1524     * to this memcg, return that limit.
1525     */
1526    return min(limit, memsw);
1527}
1528
1529/*
1530 * Visit the first child (need not be the first child as per the ordering
1531 * of the cgroup list, since we track last_scanned_child) of @mem and use
1532 * that to reclaim free pages from.
1533 */
1534static struct mem_cgroup *
1535mem_cgroup_select_victim(struct mem_cgroup *root_mem)
1536{
1537    struct mem_cgroup *ret = NULL;
1538    struct cgroup_subsys_state *css;
1539    int nextid, found;
1540
1541    if (!root_mem->use_hierarchy) {
1542        css_get(&root_mem->css);
1543        ret = root_mem;
1544    }
1545
1546    while (!ret) {
1547        rcu_read_lock();
1548        nextid = root_mem->last_scanned_child + 1;
1549        css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
1550                   &found);
1551        if (css && css_tryget(css))
1552            ret = container_of(css, struct mem_cgroup, css);
1553
1554        rcu_read_unlock();
1555        /* Updates scanning parameter */
1556        if (!css) {
1557            /* this means start scan from ID:1 */
1558            root_mem->last_scanned_child = 0;
1559        } else
1560            root_mem->last_scanned_child = found;
1561    }
1562
1563    return ret;
1564}
1565
1566/**
1567 * test_mem_cgroup_node_reclaimable
1568 * @mem: the target memcg
1569 * @nid: the node ID to be checked.
1570 * @noswap : specify true here if the user wants flle only information.
1571 *
1572 * This function returns whether the specified memcg contains any
1573 * reclaimable pages on a node. Returns true if there are any reclaimable
1574 * pages in the node.
1575 */
1576static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *mem,
1577        int nid, bool noswap)
1578{
1579    if (mem_cgroup_node_nr_file_lru_pages(mem, nid))
1580        return true;
1581    if (noswap || !total_swap_pages)
1582        return false;
1583    if (mem_cgroup_node_nr_anon_lru_pages(mem, nid))
1584        return true;
1585    return false;
1586
1587}
1588#if MAX_NUMNODES > 1
1589
1590/*
1591 * Always updating the nodemask is not very good - even if we have an empty
1592 * list or the wrong list here, we can start from some node and traverse all
1593 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1594 *
1595 */
1596static void mem_cgroup_may_update_nodemask(struct mem_cgroup *mem)
1597{
1598    int nid;
1599    /*
1600     * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1601     * pagein/pageout changes since the last update.
1602     */
1603    if (!atomic_read(&mem->numainfo_events))
1604        return;
1605    if (atomic_inc_return(&mem->numainfo_updating) > 1)
1606        return;
1607
1608    /* make a nodemask where this memcg uses memory from */
1609    mem->scan_nodes = node_states[N_HIGH_MEMORY];
1610
1611    for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) {
1612
1613        if (!test_mem_cgroup_node_reclaimable(mem, nid, false))
1614            node_clear(nid, mem->scan_nodes);
1615    }
1616
1617    atomic_set(&mem->numainfo_events, 0);
1618    atomic_set(&mem->numainfo_updating, 0);
1619}
1620
1621/*
1622 * Selecting a node where we start reclaim from. Because what we need is just
1623 * reducing usage counter, start from anywhere is O,K. Considering
1624 * memory reclaim from current node, there are pros. and cons.
1625 *
1626 * Freeing memory from current node means freeing memory from a node which
1627 * we'll use or we've used. So, it may make LRU bad. And if several threads
1628 * hit limits, it will see a contention on a node. But freeing from remote
1629 * node means more costs for memory reclaim because of memory latency.
1630 *
1631 * Now, we use round-robin. Better algorithm is welcomed.
1632 */
1633int mem_cgroup_select_victim_node(struct mem_cgroup *mem)
1634{
1635    int node;
1636
1637    mem_cgroup_may_update_nodemask(mem);
1638    node = mem->last_scanned_node;
1639
1640    node = next_node(node, mem->scan_nodes);
1641    if (node == MAX_NUMNODES)
1642        node = first_node(mem->scan_nodes);
1643    /*
1644     * We call this when we hit limit, not when pages are added to LRU.
1645     * No LRU may hold pages because all pages are UNEVICTABLE or
1646     * memcg is too small and all pages are not on LRU. In that case,
1647     * we use curret node.
1648     */
1649    if (unlikely(node == MAX_NUMNODES))
1650        node = numa_node_id();
1651
1652    mem->last_scanned_node = node;
1653    return node;
1654}
1655
1656/*
1657 * Check all nodes whether it contains reclaimable pages or not.
1658 * For quick scan, we make use of scan_nodes. This will allow us to skip
1659 * unused nodes. But scan_nodes is lazily updated and may not cotain
1660 * enough new information. We need to do double check.
1661 */
1662bool mem_cgroup_reclaimable(struct mem_cgroup *mem, bool noswap)
1663{
1664    int nid;
1665
1666    /*
1667     * quick check...making use of scan_node.
1668     * We can skip unused nodes.
1669     */
1670    if (!nodes_empty(mem->scan_nodes)) {
1671        for (nid = first_node(mem->scan_nodes);
1672             nid < MAX_NUMNODES;
1673             nid = next_node(nid, mem->scan_nodes)) {
1674
1675            if (test_mem_cgroup_node_reclaimable(mem, nid, noswap))
1676                return true;
1677        }
1678    }
1679    /*
1680     * Check rest of nodes.
1681     */
1682    for_each_node_state(nid, N_HIGH_MEMORY) {
1683        if (node_isset(nid, mem->scan_nodes))
1684            continue;
1685        if (test_mem_cgroup_node_reclaimable(mem, nid, noswap))
1686            return true;
1687    }
1688    return false;
1689}
1690
1691#else
1692int mem_cgroup_select_victim_node(struct mem_cgroup *mem)
1693{
1694    return 0;
1695}
1696
1697bool mem_cgroup_reclaimable(struct mem_cgroup *mem, bool noswap)
1698{
1699    return test_mem_cgroup_node_reclaimable(mem, 0, noswap);
1700}
1701#endif
1702
1703/*
1704 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1705 * we reclaimed from, so that we don't end up penalizing one child extensively
1706 * based on its position in the children list.
1707 *
1708 * root_mem is the original ancestor that we've been reclaim from.
1709 *
1710 * We give up and return to the caller when we visit root_mem twice.
1711 * (other groups can be removed while we're walking....)
1712 *
1713 * If shrink==true, for avoiding to free too much, this returns immedieately.
1714 */
1715static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
1716                        struct zone *zone,
1717                        gfp_t gfp_mask,
1718                        unsigned long reclaim_options,
1719                        unsigned long *total_scanned)
1720{
1721    struct mem_cgroup *victim;
1722    int ret, total = 0;
1723    int loop = 0;
1724    bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
1725    bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
1726    bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT;
1727    unsigned long excess;
1728    unsigned long nr_scanned;
1729
1730    excess = res_counter_soft_limit_excess(&root_mem->res) >> PAGE_SHIFT;
1731
1732    /* If memsw_is_minimum==1, swap-out is of-no-use. */
1733    if (!check_soft && root_mem->memsw_is_minimum)
1734        noswap = true;
1735
1736    while (1) {
1737        victim = mem_cgroup_select_victim(root_mem);
1738        if (victim == root_mem) {
1739            loop++;
1740            /*
1741             * We are not draining per cpu cached charges during
1742             * soft limit reclaim because global reclaim doesn't
1743             * care about charges. It tries to free some memory and
1744             * charges will not give any.
1745             */
1746            if (!check_soft && loop >= 1)
1747                drain_all_stock_async(root_mem);
1748            if (loop >= 2) {
1749                /*
1750                 * If we have not been able to reclaim
1751                 * anything, it might because there are
1752                 * no reclaimable pages under this hierarchy
1753                 */
1754                if (!check_soft || !total) {
1755                    css_put(&victim->css);
1756                    break;
1757                }
1758                /*
1759                 * We want to do more targeted reclaim.
1760                 * excess >> 2 is not to excessive so as to
1761                 * reclaim too much, nor too less that we keep
1762                 * coming back to reclaim from this cgroup
1763                 */
1764                if (total >= (excess >> 2) ||
1765                    (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) {
1766                    css_put(&victim->css);
1767                    break;
1768                }
1769            }
1770        }
1771        if (!mem_cgroup_reclaimable(victim, noswap)) {
1772            /* this cgroup's local usage == 0 */
1773            css_put(&victim->css);
1774            continue;
1775        }
1776        /* we use swappiness of local cgroup */
1777        if (check_soft) {
1778            ret = mem_cgroup_shrink_node_zone(victim, gfp_mask,
1779                noswap, get_swappiness(victim), zone,
1780                &nr_scanned);
1781            *total_scanned += nr_scanned;
1782        } else
1783            ret = try_to_free_mem_cgroup_pages(victim, gfp_mask,
1784                        noswap, get_swappiness(victim));
1785        css_put(&victim->css);
1786        /*
1787         * At shrinking usage, we can't check we should stop here or
1788         * reclaim more. It's depends on callers. last_scanned_child
1789         * will work enough for keeping fairness under tree.
1790         */
1791        if (shrink)
1792            return ret;
1793        total += ret;
1794        if (check_soft) {
1795            if (!res_counter_soft_limit_excess(&root_mem->res))
1796                return total;
1797        } else if (mem_cgroup_margin(root_mem))
1798            return total;
1799    }
1800    return total;
1801}
1802
1803/*
1804 * Check OOM-Killer is already running under our hierarchy.
1805 * If someone is running, return false.
1806 */
1807static bool mem_cgroup_oom_lock(struct mem_cgroup *mem)
1808{
1809    int x, lock_count = 0;
1810    struct mem_cgroup *iter;
1811
1812    for_each_mem_cgroup_tree(iter, mem) {
1813        x = atomic_inc_return(&iter->oom_lock);
1814        lock_count = max(x, lock_count);
1815    }
1816
1817    if (lock_count == 1)
1818        return true;
1819    return false;
1820}
1821
1822static int mem_cgroup_oom_unlock(struct mem_cgroup *mem)
1823{
1824    struct mem_cgroup *iter;
1825
1826    /*
1827     * When a new child is created while the hierarchy is under oom,
1828     * mem_cgroup_oom_lock() may not be called. We have to use
1829     * atomic_add_unless() here.
1830     */
1831    for_each_mem_cgroup_tree(iter, mem)
1832        atomic_add_unless(&iter->oom_lock, -1, 0);
1833    return 0;
1834}
1835
1836
1837static DEFINE_MUTEX(memcg_oom_mutex);
1838static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1839
1840struct oom_wait_info {
1841    struct mem_cgroup *mem;
1842    wait_queue_t wait;
1843};
1844
1845static int memcg_oom_wake_function(wait_queue_t *wait,
1846    unsigned mode, int sync, void *arg)
1847{
1848    struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg;
1849    struct oom_wait_info *oom_wait_info;
1850
1851    oom_wait_info = container_of(wait, struct oom_wait_info, wait);
1852
1853    if (oom_wait_info->mem == wake_mem)
1854        goto wakeup;
1855    /* if no hierarchy, no match */
1856    if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy)
1857        return 0;
1858    /*
1859     * Both of oom_wait_info->mem and wake_mem are stable under us.
1860     * Then we can use css_is_ancestor without taking care of RCU.
1861     */
1862    if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) &&
1863        !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css))
1864        return 0;
1865
1866wakeup:
1867    return autoremove_wake_function(wait, mode, sync, arg);
1868}
1869
1870static void memcg_wakeup_oom(struct mem_cgroup *mem)
1871{
1872    /* for filtering, pass "mem" as argument. */
1873    __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem);
1874}
1875
1876static void memcg_oom_recover(struct mem_cgroup *mem)
1877{
1878    if (mem && atomic_read(&mem->oom_lock))
1879        memcg_wakeup_oom(mem);
1880}
1881
1882/*
1883 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1884 */
1885bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask)
1886{
1887    struct oom_wait_info owait;
1888    bool locked, need_to_kill;
1889
1890    owait.mem = mem;
1891    owait.wait.flags = 0;
1892    owait.wait.func = memcg_oom_wake_function;
1893    owait.wait.private = current;
1894    INIT_LIST_HEAD(&owait.wait.task_list);
1895    need_to_kill = true;
1896    /* At first, try to OOM lock hierarchy under mem.*/
1897    mutex_lock(&memcg_oom_mutex);
1898    locked = mem_cgroup_oom_lock(mem);
1899    /*
1900     * Even if signal_pending(), we can't quit charge() loop without
1901     * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1902     * under OOM is always welcomed, use TASK_KILLABLE here.
1903     */
1904    prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
1905    if (!locked || mem->oom_kill_disable)
1906        need_to_kill = false;
1907    if (locked)
1908        mem_cgroup_oom_notify(mem);
1909    mutex_unlock(&memcg_oom_mutex);
1910
1911    if (need_to_kill) {
1912        finish_wait(&memcg_oom_waitq, &owait.wait);
1913        mem_cgroup_out_of_memory(mem, mask);
1914    } else {
1915        schedule();
1916        finish_wait(&memcg_oom_waitq, &owait.wait);
1917    }
1918    mutex_lock(&memcg_oom_mutex);
1919    mem_cgroup_oom_unlock(mem);
1920    memcg_wakeup_oom(mem);
1921    mutex_unlock(&memcg_oom_mutex);
1922
1923    if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1924        return false;
1925    /* Give chance to dying process */
1926    schedule_timeout(1);
1927    return true;
1928}
1929
1930/*
1931 * Currently used to update mapped file statistics, but the routine can be
1932 * generalized to update other statistics as well.
1933 *
1934 * Notes: Race condition
1935 *
1936 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1937 * it tends to be costly. But considering some conditions, we doesn't need
1938 * to do so _always_.
1939 *
1940 * Considering "charge", lock_page_cgroup() is not required because all
1941 * file-stat operations happen after a page is attached to radix-tree. There
1942 * are no race with "charge".
1943 *
1944 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1945 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1946 * if there are race with "uncharge". Statistics itself is properly handled
1947 * by flags.
1948 *
1949 * Considering "move", this is an only case we see a race. To make the race
1950 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1951 * possibility of race condition. If there is, we take a lock.
1952 */
1953
1954void mem_cgroup_update_page_stat(struct page *page,
1955                 enum mem_cgroup_page_stat_item idx, int val)
1956{
1957    struct mem_cgroup *mem;
1958    struct page_cgroup *pc = lookup_page_cgroup(page);
1959    bool need_unlock = false;
1960    unsigned long uninitialized_var(flags);
1961
1962    if (unlikely(!pc))
1963        return;
1964
1965    rcu_read_lock();
1966    mem = pc->mem_cgroup;
1967    if (unlikely(!mem || !PageCgroupUsed(pc)))
1968        goto out;
1969    /* pc->mem_cgroup is unstable ? */
1970    if (unlikely(mem_cgroup_stealed(mem)) || PageTransHuge(page)) {
1971        /* take a lock against to access pc->mem_cgroup */
1972        move_lock_page_cgroup(pc, &flags);
1973        need_unlock = true;
1974        mem = pc->mem_cgroup;
1975        if (!mem || !PageCgroupUsed(pc))
1976            goto out;
1977    }
1978
1979    switch (idx) {
1980    case MEMCG_NR_FILE_MAPPED:
1981        if (val > 0)
1982            SetPageCgroupFileMapped(pc);
1983        else if (!page_mapped(page))
1984            ClearPageCgroupFileMapped(pc);
1985        idx = MEM_CGROUP_STAT_FILE_MAPPED;
1986        break;
1987    default:
1988        BUG();
1989    }
1990
1991    this_cpu_add(mem->stat->count[idx], val);
1992
1993out:
1994    if (unlikely(need_unlock))
1995        move_unlock_page_cgroup(pc, &flags);
1996    rcu_read_unlock();
1997    return;
1998}
1999EXPORT_SYMBOL(mem_cgroup_update_page_stat);
2000
2001/*
2002 * size of first charge trial. "32" comes from vmscan.c's magic value.
2003 * TODO: maybe necessary to use big numbers in big irons.
2004 */
2005#define CHARGE_BATCH 32U
2006struct memcg_stock_pcp {
2007    struct mem_cgroup *cached; /* this never be root cgroup */
2008    unsigned int nr_pages;
2009    struct work_struct work;
2010    unsigned long flags;
2011#define FLUSHING_CACHED_CHARGE (0)
2012};
2013static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
2014static DEFINE_MUTEX(percpu_charge_mutex);
2015
2016/*
2017 * Try to consume stocked charge on this cpu. If success, one page is consumed
2018 * from local stock and true is returned. If the stock is 0 or charges from a
2019 * cgroup which is not current target, returns false. This stock will be
2020 * refilled.
2021 */
2022static bool consume_stock(struct mem_cgroup *mem)
2023{
2024    struct memcg_stock_pcp *stock;
2025    bool ret = true;
2026
2027    stock = &get_cpu_var(memcg_stock);
2028    if (mem == stock->cached && stock->nr_pages)
2029        stock->nr_pages--;
2030    else /* need to call res_counter_charge */
2031        ret = false;
2032    put_cpu_var(memcg_stock);
2033    return ret;
2034}
2035
2036/*
2037 * Returns stocks cached in percpu to res_counter and reset cached information.
2038 */
2039static void drain_stock(struct memcg_stock_pcp *stock)
2040{
2041    struct mem_cgroup *old = stock->cached;
2042
2043    if (stock->nr_pages) {
2044        unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2045
2046        res_counter_uncharge(&old->res, bytes);
2047        if (do_swap_account)
2048            res_counter_uncharge(&old->memsw, bytes);
2049        stock->nr_pages = 0;
2050    }
2051    stock->cached = NULL;
2052}
2053
2054/*
2055 * This must be called under preempt disabled or must be called by
2056 * a thread which is pinned to local cpu.
2057 */
2058static void drain_local_stock(struct work_struct *dummy)
2059{
2060    struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2061    drain_stock(stock);
2062    clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
2063}
2064
2065/*
2066 * Cache charges(val) which is from res_counter, to local per_cpu area.
2067 * This will be consumed by consume_stock() function, later.
2068 */
2069static void refill_stock(struct mem_cgroup *mem, unsigned int nr_pages)
2070{
2071    struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2072
2073    if (stock->cached != mem) { /* reset if necessary */
2074        drain_stock(stock);
2075        stock->cached = mem;
2076    }
2077    stock->nr_pages += nr_pages;
2078    put_cpu_var(memcg_stock);
2079}
2080
2081/*
2082 * Tries to drain stocked charges in other cpus. This function is asynchronous
2083 * and just put a work per cpu for draining localy on each cpu. Caller can
2084 * expects some charges will be back to res_counter later but cannot wait for
2085 * it.
2086 */
2087static void drain_all_stock_async(struct mem_cgroup *root_mem)
2088{
2089    int cpu, curcpu;
2090    /*
2091     * If someone calls draining, avoid adding more kworker runs.
2092     */
2093    if (!mutex_trylock(&percpu_charge_mutex))
2094        return;
2095    /* Notify other cpus that system-wide "drain" is running */
2096    get_online_cpus();
2097    /*
2098     * Get a hint for avoiding draining charges on the current cpu,
2099     * which must be exhausted by our charging. It is not required that
2100     * this be a precise check, so we use raw_smp_processor_id() instead of
2101     * getcpu()/putcpu().
2102     */
2103    curcpu = raw_smp_processor_id();
2104    for_each_online_cpu(cpu) {
2105        struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
2106        struct mem_cgroup *mem;
2107
2108        if (cpu == curcpu)
2109            continue;
2110
2111        mem = stock->cached;
2112        if (!mem)
2113            continue;
2114        if (mem != root_mem) {
2115            if (!root_mem->use_hierarchy)
2116                continue;
2117            /* check whether "mem" is under tree of "root_mem" */
2118            if (!css_is_ancestor(&mem->css, &root_mem->css))
2119                continue;
2120        }
2121        if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
2122            schedule_work_on(cpu, &stock->work);
2123    }
2124     put_online_cpus();
2125    mutex_unlock(&percpu_charge_mutex);
2126    /* We don't wait for flush_work */
2127}
2128
2129/* This is a synchronous drain interface. */
2130static void drain_all_stock_sync(void)
2131{
2132    /* called when force_empty is called */
2133    mutex_lock(&percpu_charge_mutex);
2134    schedule_on_each_cpu(drain_local_stock);
2135    mutex_unlock(&percpu_charge_mutex);
2136}
2137
2138/*
2139 * This function drains percpu counter value from DEAD cpu and
2140 * move it to local cpu. Note that this function can be preempted.
2141 */
2142static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *mem, int cpu)
2143{
2144    int i;
2145
2146    spin_lock(&mem->pcp_counter_lock);
2147    for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) {
2148        long x = per_cpu(mem->stat->count[i], cpu);
2149
2150        per_cpu(mem->stat->count[i], cpu) = 0;
2151        mem->nocpu_base.count[i] += x;
2152    }
2153    for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
2154        unsigned long x = per_cpu(mem->stat->events[i], cpu);
2155
2156        per_cpu(mem->stat->events[i], cpu) = 0;
2157        mem->nocpu_base.events[i] += x;
2158    }
2159    /* need to clear ON_MOVE value, works as a kind of lock. */
2160    per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0;
2161    spin_unlock(&mem->pcp_counter_lock);
2162}
2163
2164static void synchronize_mem_cgroup_on_move(struct mem_cgroup *mem, int cpu)
2165{
2166    int idx = MEM_CGROUP_ON_MOVE;
2167
2168    spin_lock(&mem->pcp_counter_lock);
2169    per_cpu(mem->stat->count[idx], cpu) = mem->nocpu_base.count[idx];
2170    spin_unlock(&mem->pcp_counter_lock);
2171}
2172
2173static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
2174                    unsigned long action,
2175                    void *hcpu)
2176{
2177    int cpu = (unsigned long)hcpu;
2178    struct memcg_stock_pcp *stock;
2179    struct mem_cgroup *iter;
2180
2181    if ((action == CPU_ONLINE)) {
2182        for_each_mem_cgroup_all(iter)
2183            synchronize_mem_cgroup_on_move(iter, cpu);
2184        return NOTIFY_OK;
2185    }
2186
2187    if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN)
2188        return NOTIFY_OK;
2189
2190    for_each_mem_cgroup_all(iter)
2191        mem_cgroup_drain_pcp_counter(iter, cpu);
2192
2193    stock = &per_cpu(memcg_stock, cpu);
2194    drain_stock(stock);
2195    return NOTIFY_OK;
2196}
2197
2198
2199/* See __mem_cgroup_try_charge() for details */
2200enum {
2201    CHARGE_OK, /* success */
2202    CHARGE_RETRY, /* need to retry but retry is not bad */
2203    CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
2204    CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
2205    CHARGE_OOM_DIE, /* the current is killed because of OOM */
2206};
2207
2208static int mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask,
2209                unsigned int nr_pages, bool oom_check)
2210{
2211    unsigned long csize = nr_pages * PAGE_SIZE;
2212    struct mem_cgroup *mem_over_limit;
2213    struct res_counter *fail_res;
2214    unsigned long flags = 0;
2215    int ret;
2216
2217    ret = res_counter_charge(&mem->res, csize, &fail_res);
2218
2219    if (likely(!ret)) {
2220        if (!do_swap_account)
2221            return CHARGE_OK;
2222        ret = res_counter_charge(&mem->memsw, csize, &fail_res);
2223        if (likely(!ret))
2224            return CHARGE_OK;
2225
2226        res_counter_uncharge(&mem->res, csize);
2227        mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2228        flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2229    } else
2230        mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
2231    /*
2232     * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2233     * of regular pages (CHARGE_BATCH), or a single regular page (1).
2234     *
2235     * Never reclaim on behalf of optional batching, retry with a
2236     * single page instead.
2237     */
2238    if (nr_pages == CHARGE_BATCH)
2239        return CHARGE_RETRY;
2240
2241    if (!(gfp_mask & __GFP_WAIT))
2242        return CHARGE_WOULDBLOCK;
2243
2244    ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
2245                          gfp_mask, flags, NULL);
2246    if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
2247        return CHARGE_RETRY;
2248    /*
2249     * Even though the limit is exceeded at this point, reclaim
2250     * may have been able to free some pages. Retry the charge
2251     * before killing the task.
2252     *
2253     * Only for regular pages, though: huge pages are rather
2254     * unlikely to succeed so close to the limit, and we fall back
2255     * to regular pages anyway in case of failure.
2256     */
2257    if (nr_pages == 1 && ret)
2258        return CHARGE_RETRY;
2259
2260    /*
2261     * At task move, charge accounts can be doubly counted. So, it's
2262     * better to wait until the end of task_move if something is going on.
2263     */
2264    if (mem_cgroup_wait_acct_move(mem_over_limit))
2265        return CHARGE_RETRY;
2266
2267    /* If we don't need to call oom-killer at el, return immediately */
2268    if (!oom_check)
2269        return CHARGE_NOMEM;
2270    /* check OOM */
2271    if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask))
2272        return CHARGE_OOM_DIE;
2273
2274    return CHARGE_RETRY;
2275}
2276
2277/*
2278 * Unlike exported interface, "oom" parameter is added. if oom==true,
2279 * oom-killer can be invoked.
2280 */
2281static int __mem_cgroup_try_charge(struct mm_struct *mm,
2282                   gfp_t gfp_mask,
2283                   unsigned int nr_pages,
2284                   struct mem_cgroup **memcg,
2285                   bool oom)
2286{
2287    unsigned int batch = max(CHARGE_BATCH, nr_pages);
2288    int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2289    struct mem_cgroup *mem = NULL;
2290    int ret;
2291
2292    /*
2293     * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2294     * in system level. So, allow to go ahead dying process in addition to
2295     * MEMDIE process.
2296     */
2297    if (unlikely(test_thread_flag(TIF_MEMDIE)
2298             || fatal_signal_pending(current)))
2299        goto bypass;
2300
2301    /*
2302     * We always charge the cgroup the mm_struct belongs to.
2303     * The mm_struct's mem_cgroup changes on task migration if the
2304     * thread group leader migrates. It's possible that mm is not
2305     * set, if so charge the init_mm (happens for pagecache usage).
2306     */
2307    if (!*memcg && !mm)
2308        goto bypass;
2309again:
2310    if (*memcg) { /* css should be a valid one */
2311        mem = *memcg;
2312        VM_BUG_ON(css_is_removed(&mem->css));
2313        if (mem_cgroup_is_root(mem))
2314            goto done;
2315        if (nr_pages == 1 && consume_stock(mem))
2316            goto done;
2317        css_get(&mem->css);
2318    } else {
2319        struct task_struct *p;
2320
2321        rcu_read_lock();
2322        p = rcu_dereference(mm->owner);
2323        /*
2324         * Because we don't have task_lock(), "p" can exit.
2325         * In that case, "mem" can point to root or p can be NULL with
2326         * race with swapoff. Then, we have small risk of mis-accouning.
2327         * But such kind of mis-account by race always happens because
2328         * we don't have cgroup_mutex(). It's overkill and we allo that
2329         * small race, here.
2330         * (*) swapoff at el will charge against mm-struct not against
2331         * task-struct. So, mm->owner can be NULL.
2332         */
2333        mem = mem_cgroup_from_task(p);
2334        if (!mem || mem_cgroup_is_root(mem)) {
2335            rcu_read_unlock();
2336            goto done;
2337        }
2338        if (nr_pages == 1 && consume_stock(mem)) {
2339            /*
2340             * It seems dagerous to access memcg without css_get().
2341             * But considering how consume_stok works, it's not
2342             * necessary. If consume_stock success, some charges
2343             * from this memcg are cached on this cpu. So, we
2344             * don't need to call css_get()/css_tryget() before
2345             * calling consume_stock().
2346             */
2347            rcu_read_unlock();
2348            goto done;
2349        }
2350        /* after here, we may be blocked. we need to get refcnt */
2351        if (!css_tryget(&mem->css)) {
2352            rcu_read_unlock();
2353            goto again;
2354        }
2355        rcu_read_unlock();
2356    }
2357
2358    do {
2359        bool oom_check;
2360
2361        /* If killed, bypass charge */
2362        if (fatal_signal_pending(current)) {
2363            css_put(&mem->css);
2364            goto bypass;
2365        }
2366
2367        oom_check = false;
2368        if (oom && !nr_oom_retries) {
2369            oom_check = true;
2370            nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2371        }
2372
2373        ret = mem_cgroup_do_charge(mem, gfp_mask, batch, oom_check);
2374        switch (ret) {
2375        case CHARGE_OK:
2376            break;
2377        case CHARGE_RETRY: /* not in OOM situation but retry */
2378            batch = nr_pages;
2379            css_put(&mem->css);
2380            mem = NULL;
2381            goto again;
2382        case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
2383            css_put(&mem->css);
2384            goto nomem;
2385        case CHARGE_NOMEM: /* OOM routine works */
2386            if (!oom) {
2387                css_put(&mem->css);
2388                goto nomem;
2389            }
2390            /* If oom, we never return -ENOMEM */
2391            nr_oom_retries--;
2392            break;
2393        case CHARGE_OOM_DIE: /* Killed by OOM Killer */
2394            css_put(&mem->css);
2395            goto bypass;
2396        }
2397    } while (ret != CHARGE_OK);
2398
2399    if (batch > nr_pages)
2400        refill_stock(mem, batch - nr_pages);
2401    css_put(&mem->css);
2402done:
2403    *memcg = mem;
2404    return 0;
2405nomem:
2406    *memcg = NULL;
2407    return -ENOMEM;
2408bypass:
2409    *memcg = NULL;
2410    return 0;
2411}
2412
2413/*
2414 * Somemtimes we have to undo a charge we got by try_charge().
2415 * This function is for that and do uncharge, put css's refcnt.
2416 * gotten by try_charge().
2417 */
2418static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem,
2419                       unsigned int nr_pages)
2420{
2421    if (!mem_cgroup_is_root(mem)) {
2422        unsigned long bytes = nr_pages * PAGE_SIZE;
2423
2424        res_counter_uncharge(&mem->res, bytes);
2425        if (do_swap_account)
2426            res_counter_uncharge(&mem->memsw, bytes);
2427    }
2428}
2429
2430/*
2431 * A helper function to get mem_cgroup from ID. must be called under
2432 * rcu_read_lock(). The caller must check css_is_removed() or some if
2433 * it's concern. (dropping refcnt from swap can be called against removed
2434 * memcg.)
2435 */
2436static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2437{
2438    struct cgroup_subsys_state *css;
2439
2440    /* ID 0 is unused ID */
2441    if (!id)
2442        return NULL;
2443    css = css_lookup(&mem_cgroup_subsys, id);
2444    if (!css)
2445        return NULL;
2446    return container_of(css, struct mem_cgroup, css);
2447}
2448
2449struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
2450{
2451    struct mem_cgroup *mem = NULL;
2452    struct page_cgroup *pc;
2453    unsigned short id;
2454    swp_entry_t ent;
2455
2456    VM_BUG_ON(!PageLocked(page));
2457
2458    pc = lookup_page_cgroup(page);
2459    lock_page_cgroup(pc);
2460    if (PageCgroupUsed(pc)) {
2461        mem = pc->mem_cgroup;
2462        if (mem && !css_tryget(&mem->css))
2463            mem = NULL;
2464    } else if (PageSwapCache(page)) {
2465        ent.val = page_private(page);
2466        id = lookup_swap_cgroup(ent);
2467        rcu_read_lock();
2468        mem = mem_cgroup_lookup(id);
2469        if (mem && !css_tryget(&mem->css))
2470            mem = NULL;
2471        rcu_read_unlock();
2472    }
2473    unlock_page_cgroup(pc);
2474    return mem;
2475}
2476
2477static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
2478                       struct page *page,
2479                       unsigned int nr_pages,
2480                       struct page_cgroup *pc,
2481                       enum charge_type ctype)
2482{
2483    lock_page_cgroup(pc);
2484    if (unlikely(PageCgroupUsed(pc))) {
2485        unlock_page_cgroup(pc);
2486        __mem_cgroup_cancel_charge(mem, nr_pages);
2487        return;
2488    }
2489    /*
2490     * we don't need page_cgroup_lock about tail pages, becase they are not
2491     * accessed by any other context at this point.
2492     */
2493    pc->mem_cgroup = mem;
2494    /*
2495     * We access a page_cgroup asynchronously without lock_page_cgroup().
2496     * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2497     * is accessed after testing USED bit. To make pc->mem_cgroup visible
2498     * before USED bit, we need memory barrier here.
2499     * See mem_cgroup_add_lru_list(), etc.
2500      */
2501    smp_wmb();
2502    switch (ctype) {
2503    case MEM_CGROUP_CHARGE_TYPE_CACHE:
2504    case MEM_CGROUP_CHARGE_TYPE_SHMEM:
2505        SetPageCgroupCache(pc);
2506        SetPageCgroupUsed(pc);
2507        break;
2508    case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2509        ClearPageCgroupCache(pc);
2510        SetPageCgroupUsed(pc);
2511        break;
2512    default:
2513        break;
2514    }
2515
2516    mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), nr_pages);
2517    unlock_page_cgroup(pc);
2518    /*
2519     * "charge_statistics" updated event counter. Then, check it.
2520     * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2521     * if they exceeds softlimit.
2522     */
2523    memcg_check_events(mem, page);
2524}
2525
2526#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2527
2528#define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2529            (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION))
2530/*
2531 * Because tail pages are not marked as "used", set it. We're under
2532 * zone->lru_lock, 'splitting on pmd' and compund_lock.
2533 */
2534void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail)
2535{
2536    struct page_cgroup *head_pc = lookup_page_cgroup(head);
2537    struct page_cgroup *tail_pc = lookup_page_cgroup(tail);
2538    unsigned long flags;
2539
2540    if (mem_cgroup_disabled())
2541        return;
2542    /*
2543     * We have no races with charge/uncharge but will have races with
2544     * page state accounting.
2545     */
2546    move_lock_page_cgroup(head_pc, &flags);
2547
2548    tail_pc->mem_cgroup = head_pc->mem_cgroup;
2549    smp_wmb(); /* see __commit_charge() */
2550    if (PageCgroupAcctLRU(head_pc)) {
2551        enum lru_list lru;
2552        struct mem_cgroup_per_zone *mz;
2553
2554        /*
2555         * LRU flags cannot be copied because we need to add tail
2556         *.page to LRU by generic call and our hook will be called.
2557         * We hold lru_lock, then, reduce counter directly.
2558         */
2559        lru = page_lru(head);
2560        mz = page_cgroup_zoneinfo(head_pc->mem_cgroup, head);
2561        MEM_CGROUP_ZSTAT(mz, lru) -= 1;
2562    }
2563    tail_pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2564    move_unlock_page_cgroup(head_pc, &flags);
2565}
2566#endif
2567
2568/**
2569 * mem_cgroup_move_account - move account of the page
2570 * @page: the page
2571 * @nr_pages: number of regular pages (>1 for huge pages)
2572 * @pc: page_cgroup of the page.
2573 * @from: mem_cgroup which the page is moved from.
2574 * @to: mem_cgroup which the page is moved to. @from != @to.
2575 * @uncharge: whether we should call uncharge and css_put against @from.
2576 *
2577 * The caller must confirm following.
2578 * - page is not on LRU (isolate_page() is useful.)
2579 * - compound_lock is held when nr_pages > 1
2580 *
2581 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2582 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
2583 * true, this function does "uncharge" from old cgroup, but it doesn't if
2584 * @uncharge is false, so a caller should do "uncharge".
2585 */
2586static int mem_cgroup_move_account(struct page *page,
2587                   unsigned int nr_pages,
2588                   struct page_cgroup *pc,
2589                   struct mem_cgroup *from,
2590                   struct mem_cgroup *to,
2591                   bool uncharge)
2592{
2593    unsigned long flags;
2594    int ret;
2595
2596    VM_BUG_ON(from == to);
2597    VM_BUG_ON(PageLRU(page));
2598    /*
2599     * The page is isolated from LRU. So, collapse function
2600     * will not handle this page. But page splitting can happen.
2601     * Do this check under compound_page_lock(). The caller should
2602     * hold it.
2603     */
2604    ret = -EBUSY;
2605    if (nr_pages > 1 && !PageTransHuge(page))
2606        goto out;
2607
2608    lock_page_cgroup(pc);
2609
2610    ret = -EINVAL;
2611    if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
2612        goto unlock;
2613
2614    move_lock_page_cgroup(pc, &flags);
2615
2616    if (PageCgroupFileMapped(pc)) {
2617        /* Update mapped_file data for mem_cgroup */
2618        preempt_disable();
2619        __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2620        __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2621        preempt_enable();
2622    }
2623    mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages);
2624    if (uncharge)
2625        /* This is not "cancel", but cancel_charge does all we need. */
2626        __mem_cgroup_cancel_charge(from, nr_pages);
2627
2628    /* caller should have done css_get */
2629    pc->mem_cgroup = to;
2630    mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages);
2631    /*
2632     * We charges against "to" which may not have any tasks. Then, "to"
2633     * can be under rmdir(). But in current implementation, caller of
2634     * this function is just force_empty() and move charge, so it's
2635     * guaranteed that "to" is never removed. So, we don't check rmdir
2636     * status here.
2637     */
2638    move_unlock_page_cgroup(pc, &flags);
2639    ret = 0;
2640unlock:
2641    unlock_page_cgroup(pc);
2642    /*
2643     * check events
2644     */
2645    memcg_check_events(to, page);
2646    memcg_check_events(from, page);
2647out:
2648    return ret;
2649}
2650
2651/*
2652 * move charges to its parent.
2653 */
2654
2655static int mem_cgroup_move_parent(struct page *page,
2656                  struct page_cgroup *pc,
2657                  struct mem_cgroup *child,
2658                  gfp_t gfp_mask)
2659{
2660    struct cgroup *cg = child->css.cgroup;
2661    struct cgroup *pcg = cg->parent;
2662    struct mem_cgroup *parent;
2663    unsigned int nr_pages;
2664    unsigned long uninitialized_var(flags);
2665    int ret;
2666
2667    /* Is ROOT ? */
2668    if (!pcg)
2669        return -EINVAL;
2670
2671    ret = -EBUSY;
2672    if (!get_page_unless_zero(page))
2673        goto out;
2674    if (isolate_lru_page(page))
2675        goto put;
2676
2677    nr_pages = hpage_nr_pages(page);
2678
2679    parent = mem_cgroup_from_cont(pcg);
2680    ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false);
2681    if (ret || !parent)
2682        goto put_back;
2683
2684    if (nr_pages > 1)
2685        flags = compound_lock_irqsave(page);
2686
2687    ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true);
2688    if (ret)
2689        __mem_cgroup_cancel_charge(parent, nr_pages);
2690
2691    if (nr_pages > 1)
2692        compound_unlock_irqrestore(page, flags);
2693put_back:
2694    putback_lru_page(page);
2695put:
2696    put_page(page);
2697out:
2698    return ret;
2699}
2700
2701/*
2702 * Charge the memory controller for page usage.
2703 * Return
2704 * 0 if the charge was successful
2705 * < 0 if the cgroup is over its limit
2706 */
2707static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
2708                gfp_t gfp_mask, enum charge_type ctype)
2709{
2710    struct mem_cgroup *mem = NULL;
2711    unsigned int nr_pages = 1;
2712    struct page_cgroup *pc;
2713    bool oom = true;
2714    int ret;
2715
2716    if (PageTransHuge(page)) {
2717        nr_pages <<= compound_order(page);
2718        VM_BUG_ON(!PageTransHuge(page));
2719        /*
2720         * Never OOM-kill a process for a huge page. The
2721         * fault handler will fall back to regular pages.
2722         */
2723        oom = false;
2724    }
2725
2726    pc = lookup_page_cgroup(page);
2727    BUG_ON(!pc); /* XXX: remove this and move pc lookup into commit */
2728
2729    ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &mem, oom);
2730    if (ret || !mem)
2731        return ret;
2732
2733    __mem_cgroup_commit_charge(mem, page, nr_pages, pc, ctype);
2734    return 0;
2735}
2736
2737int mem_cgroup_newpage_charge(struct page *page,
2738                  struct mm_struct *mm, gfp_t gfp_mask)
2739{
2740    if (mem_cgroup_disabled())
2741        return 0;
2742    /*
2743     * If already mapped, we don't have to account.
2744     * If page cache, page->mapping has address_space.
2745     * But page->mapping may have out-of-use anon_vma pointer,
2746     * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2747     * is NULL.
2748       */
2749    if (page_mapped(page) || (page->mapping && !PageAnon(page)))
2750        return 0;
2751    if (unlikely(!mm))
2752        mm = &init_mm;
2753    return mem_cgroup_charge_common(page, mm, gfp_mask,
2754                MEM_CGROUP_CHARGE_TYPE_MAPPED);
2755}
2756
2757static void
2758__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2759                    enum charge_type ctype);
2760
2761static void
2762__mem_cgroup_commit_charge_lrucare(struct page *page, struct mem_cgroup *mem,
2763                    enum charge_type ctype)
2764{
2765    struct page_cgroup *pc = lookup_page_cgroup(page);
2766    /*
2767     * In some case, SwapCache, FUSE(splice_buf->radixtree), the page
2768     * is already on LRU. It means the page may on some other page_cgroup's
2769     * LRU. Take care of it.
2770     */
2771    mem_cgroup_lru_del_before_commit(page);
2772    __mem_cgroup_commit_charge(mem, page, 1, pc, ctype);
2773    mem_cgroup_lru_add_after_commit(page);
2774    return;
2775}
2776
2777int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2778                gfp_t gfp_mask)
2779{
2780    struct mem_cgroup *mem = NULL;
2781    int ret;
2782
2783    if (mem_cgroup_disabled())
2784        return 0;
2785    if (PageCompound(page))
2786        return 0;
2787    /*
2788     * Corner case handling. This is called from add_to_page_cache()
2789     * in usual. But some FS (shmem) precharges this page before calling it
2790     * and call add_to_page_cache() with GFP_NOWAIT.
2791     *
2792     * For GFP_NOWAIT case, the page may be pre-charged before calling
2793     * add_to_page_cache(). (See shmem.c) check it here and avoid to call
2794     * charge twice. (It works but has to pay a bit larger cost.)
2795     * And when the page is SwapCache, it should take swap information
2796     * into account. This is under lock_page() now.
2797     */
2798    if (!(gfp_mask & __GFP_WAIT)) {
2799        struct page_cgroup *pc;
2800
2801        pc = lookup_page_cgroup(page);
2802        if (!pc)
2803            return 0;
2804        lock_page_cgroup(pc);
2805        if (PageCgroupUsed(pc)) {
2806            unlock_page_cgroup(pc);
2807            return 0;
2808        }
2809        unlock_page_cgroup(pc);
2810    }
2811
2812    if (unlikely(!mm))
2813        mm = &init_mm;
2814
2815    if (page_is_file_cache(page)) {
2816        ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, &mem, true);
2817        if (ret || !mem)
2818            return ret;
2819
2820        /*
2821         * FUSE reuses pages without going through the final
2822         * put that would remove them from the LRU list, make
2823         * sure that they get relinked properly.
2824         */
2825        __mem_cgroup_commit_charge_lrucare(page, mem,
2826                    MEM_CGROUP_CHARGE_TYPE_CACHE);
2827        return ret;
2828    }
2829    /* shmem */
2830    if (PageSwapCache(page)) {
2831        ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2832        if (!ret)
2833            __mem_cgroup_commit_charge_swapin(page, mem,
2834                    MEM_CGROUP_CHARGE_TYPE_SHMEM);
2835    } else
2836        ret = mem_cgroup_charge_common(page, mm, gfp_mask,
2837                    MEM_CGROUP_CHARGE_TYPE_SHMEM);
2838
2839    return ret;
2840}
2841
2842/*
2843 * While swap-in, try_charge -> commit or cancel, the page is locked.
2844 * And when try_charge() successfully returns, one refcnt to memcg without
2845 * struct page_cgroup is acquired. This refcnt will be consumed by
2846 * "commit()" or removed by "cancel()"
2847 */
2848int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2849                 struct page *page,
2850                 gfp_t mask, struct mem_cgroup **ptr)
2851{
2852    struct mem_cgroup *mem;
2853    int ret;
2854
2855    *ptr = NULL;
2856
2857    if (mem_cgroup_disabled())
2858        return 0;
2859
2860    if (!do_swap_account)
2861        goto charge_cur_mm;
2862    /*
2863     * A racing thread's fault, or swapoff, may have already updated
2864     * the pte, and even removed page from swap cache: in those cases
2865     * do_swap_page()'s pte_same() test will fail; but there's also a
2866     * KSM case which does need to charge the page.
2867     */
2868    if (!PageSwapCache(page))
2869        goto charge_cur_mm;
2870    mem = try_get_mem_cgroup_from_page(page);
2871    if (!mem)
2872        goto charge_cur_mm;
2873    *ptr = mem;
2874    ret = __mem_cgroup_try_charge(NULL, mask, 1, ptr, true);
2875    css_put(&mem->css);
2876    return ret;
2877charge_cur_mm:
2878    if (unlikely(!mm))
2879        mm = &init_mm;
2880    return __mem_cgroup_try_charge(mm, mask, 1, ptr, true);
2881}
2882
2883static void
2884__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2885                    enum charge_type ctype)
2886{
2887    if (mem_cgroup_disabled())
2888        return;
2889    if (!ptr)
2890        return;
2891    cgroup_exclude_rmdir(&ptr->css);
2892
2893    __mem_cgroup_commit_charge_lrucare(page, ptr, ctype);
2894    /*
2895     * Now swap is on-memory. This means this page may be
2896     * counted both as mem and swap....double count.
2897     * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2898     * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2899     * may call delete_from_swap_cache() before reach here.
2900     */
2901    if (do_swap_account && PageSwapCache(page)) {
2902        swp_entry_t ent = {.val = page_private(page)};
2903        unsigned short id;
2904        struct mem_cgroup *memcg;
2905
2906        id = swap_cgroup_record(ent, 0);
2907        rcu_read_lock();
2908        memcg = mem_cgroup_lookup(id);
2909        if (memcg) {
2910            /*
2911             * This recorded memcg can be obsolete one. So, avoid
2912             * calling css_tryget
2913             */
2914            if (!mem_cgroup_is_root(memcg))
2915                res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
2916            mem_cgroup_swap_statistics(memcg, false);
2917            mem_cgroup_put(memcg);
2918        }
2919        rcu_read_unlock();
2920    }
2921    /*
2922     * At swapin, we may charge account against cgroup which has no tasks.
2923     * So, rmdir()->pre_destroy() can be called while we do this charge.
2924     * In that case, we need to call pre_destroy() again. check it here.
2925     */
2926    cgroup_release_and_wakeup_rmdir(&ptr->css);
2927}
2928
2929void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
2930{
2931    __mem_cgroup_commit_charge_swapin(page, ptr,
2932                    MEM_CGROUP_CHARGE_TYPE_MAPPED);
2933}
2934
2935void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
2936{
2937    if (mem_cgroup_disabled())
2938        return;
2939    if (!mem)
2940        return;
2941    __mem_cgroup_cancel_charge(mem, 1);
2942}
2943
2944static void mem_cgroup_do_uncharge(struct mem_cgroup *mem,
2945                   unsigned int nr_pages,
2946                   const enum charge_type ctype)
2947{
2948    struct memcg_batch_info *batch = NULL;
2949    bool uncharge_memsw = true;
2950
2951    /* If swapout, usage of swap doesn't decrease */
2952    if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2953        uncharge_memsw = false;
2954
2955    batch = &current->memcg_batch;
2956    /*
2957     * In usual, we do css_get() when we remember memcg pointer.
2958     * But in this case, we keep res->usage until end of a series of
2959     * uncharges. Then, it's ok to ignore memcg's refcnt.
2960     */
2961    if (!batch->memcg)
2962        batch->memcg = mem;
2963    /*
2964     * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2965     * In those cases, all pages freed continuously can be expected to be in
2966     * the same cgroup and we have chance to coalesce uncharges.
2967     * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2968     * because we want to do uncharge as soon as possible.
2969     */
2970
2971    if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2972        goto direct_uncharge;
2973
2974    if (nr_pages > 1)
2975        goto direct_uncharge;
2976
2977    /*
2978     * In typical case, batch->memcg == mem. This means we can
2979     * merge a series of uncharges to an uncharge of res_counter.
2980     * If not, we uncharge res_counter ony by one.
2981     */
2982    if (batch->memcg != mem)
2983        goto direct_uncharge;
2984    /* remember freed charge and uncharge it later */
2985    batch->nr_pages++;
2986    if (uncharge_memsw)
2987        batch->memsw_nr_pages++;
2988    return;
2989direct_uncharge:
2990    res_counter_uncharge(&mem->res, nr_pages * PAGE_SIZE);
2991    if (uncharge_memsw)
2992        res_counter_uncharge(&mem->memsw, nr_pages * PAGE_SIZE);
2993    if (unlikely(batch->memcg != mem))
2994        memcg_oom_recover(mem);
2995    return;
2996}
2997
2998/*
2999 * uncharge if !page_mapped(page)
3000 */
3001static struct mem_cgroup *
3002__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
3003{
3004    struct mem_cgroup *mem = NULL;
3005    unsigned int nr_pages = 1;
3006    struct page_cgroup *pc;
3007
3008    if (mem_cgroup_disabled())
3009        return NULL;
3010
3011    if (PageSwapCache(page))
3012        return NULL;
3013
3014    if (PageTransHuge(page)) {
3015        nr_pages <<= compound_order(page);
3016        VM_BUG_ON(!PageTransHuge(page));
3017    }
3018    /*
3019     * Check if our page_cgroup is valid
3020     */
3021    pc = lookup_page_cgroup(page);
3022    if (unlikely(!pc || !PageCgroupUsed(pc)))
3023        return NULL;
3024
3025    lock_page_cgroup(pc);
3026
3027    mem = pc->mem_cgroup;
3028
3029    if (!PageCgroupUsed(pc))
3030        goto unlock_out;
3031
3032    switch (ctype) {
3033    case MEM_CGROUP_CHARGE_TYPE_MAPPED:
3034    case MEM_CGROUP_CHARGE_TYPE_DROP:
3035        /* See mem_cgroup_prepare_migration() */
3036        if (page_mapped(page) || PageCgroupMigration(pc))
3037            goto unlock_out;
3038        break;
3039    case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
3040        if (!PageAnon(page)) { /* Shared memory */
3041            if (page->mapping && !page_is_file_cache(page))
3042                goto unlock_out;
3043        } else if (page_mapped(page)) /* Anon */
3044                goto unlock_out;
3045        break;
3046    default:
3047        break;
3048    }
3049
3050    mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), -nr_pages);
3051
3052    ClearPageCgroupUsed(pc);
3053    /*
3054     * pc->mem_cgroup is not cleared here. It will be accessed when it's
3055     * freed from LRU. This is safe because uncharged page is expected not
3056     * to be reused (freed soon). Exception is SwapCache, it's handled by
3057     * special functions.
3058     */
3059
3060    unlock_page_cgroup(pc);
3061    /*
3062     * even after unlock, we have mem->res.usage here and this memcg
3063     * will never be freed.
3064     */
3065    memcg_check_events(mem, page);
3066    if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
3067        mem_cgroup_swap_statistics(mem, true);
3068        mem_cgroup_get(mem);
3069    }
3070    if (!mem_cgroup_is_root(mem))
3071        mem_cgroup_do_uncharge(mem, nr_pages, ctype);
3072
3073    return mem;
3074
3075unlock_out:
3076    unlock_page_cgroup(pc);
3077    return NULL;
3078}
3079
3080void mem_cgroup_uncharge_page(struct page *page)
3081{
3082    /* early check. */
3083    if (page_mapped(page))
3084        return;
3085    if (page->mapping && !PageAnon(page))
3086        return;
3087    __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
3088}
3089
3090void mem_cgroup_uncharge_cache_page(struct page *page)
3091{
3092    VM_BUG_ON(page_mapped(page));
3093    VM_BUG_ON(page->mapping);
3094    __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
3095}
3096
3097/*
3098 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3099 * In that cases, pages are freed continuously and we can expect pages
3100 * are in the same memcg. All these calls itself limits the number of
3101 * pages freed at once, then uncharge_start/end() is called properly.
3102 * This may be called prural(2) times in a context,
3103 */
3104
3105void mem_cgroup_uncharge_start(void)
3106{
3107    current->memcg_batch.do_batch++;
3108    /* We can do nest. */
3109    if (current->memcg_batch.do_batch == 1) {
3110        current->memcg_batch.memcg = NULL;
3111        current->memcg_batch.nr_pages = 0;
3112        current->memcg_batch.memsw_nr_pages = 0;
3113    }
3114}
3115
3116void mem_cgroup_uncharge_end(void)
3117{
3118    struct memcg_batch_info *batch = &current->memcg_batch;
3119
3120    if (!batch->do_batch)
3121        return;
3122
3123    batch->do_batch--;
3124    if (batch->do_batch) /* If stacked, do nothing. */
3125        return;
3126
3127    if (!batch->memcg)
3128        return;
3129    /*
3130     * This "batch->memcg" is valid without any css_get/put etc...
3131     * bacause we hide charges behind us.
3132     */
3133    if (batch->nr_pages)
3134        res_counter_uncharge(&batch->memcg->res,
3135                     batch->nr_pages * PAGE_SIZE);
3136    if (batch->memsw_nr_pages)
3137        res_counter_uncharge(&batch->memcg->memsw,
3138                     batch->memsw_nr_pages * PAGE_SIZE);
3139    memcg_oom_recover(batch->memcg);
3140    /* forget this pointer (for sanity check) */
3141    batch->memcg = NULL;
3142}
3143
3144#ifdef CONFIG_SWAP
3145/*
3146 * called after __delete_from_swap_cache() and drop "page" account.
3147 * memcg information is recorded to swap_cgroup of "ent"
3148 */
3149void
3150mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
3151{
3152    struct mem_cgroup *memcg;
3153    int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
3154
3155    if (!swapout) /* this was a swap cache but the swap is unused ! */
3156        ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
3157
3158    memcg = __mem_cgroup_uncharge_common(page, ctype);
3159
3160    /*
3161     * record memcg information, if swapout && memcg != NULL,
3162     * mem_cgroup_get() was called in uncharge().
3163     */
3164    if (do_swap_account && swapout && memcg)
3165        swap_cgroup_record(ent, css_id(&memcg->css));
3166}
3167#endif
3168
3169#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3170/*
3171 * called from swap_entry_free(). remove record in swap_cgroup and
3172 * uncharge "memsw" account.
3173 */
3174void mem_cgroup_uncharge_swap(swp_entry_t ent)
3175{
3176    struct mem_cgroup *memcg;
3177    unsigned short id;
3178
3179    if (!do_swap_account)
3180        return;
3181
3182    id = swap_cgroup_record(ent, 0);
3183    rcu_read_lock();
3184    memcg = mem_cgroup_lookup(id);
3185    if (memcg) {
3186        /*
3187         * We uncharge this because swap is freed.
3188         * This memcg can be obsolete one. We avoid calling css_tryget
3189         */
3190        if (!mem_cgroup_is_root(memcg))
3191            res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
3192        mem_cgroup_swap_statistics(memcg, false);
3193        mem_cgroup_put(memcg);
3194    }
3195    rcu_read_unlock();
3196}
3197
3198/**
3199 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3200 * @entry: swap entry to be moved
3201 * @from: mem_cgroup which the entry is moved from
3202 * @to: mem_cgroup which the entry is moved to
3203 * @need_fixup: whether we should fixup res_counters and refcounts.
3204 *
3205 * It succeeds only when the swap_cgroup's record for this entry is the same
3206 * as the mem_cgroup's id of @from.
3207 *
3208 * Returns 0 on success, -EINVAL on failure.
3209 *
3210 * The caller must have charged to @to, IOW, called res_counter_charge() about
3211 * both res and memsw, and called css_get().
3212 */
3213static int mem_cgroup_move_swap_account(swp_entry_t entry,
3214        struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
3215{
3216    unsigned short old_id, new_id;
3217
3218    old_id = css_id(&from->css);
3219    new_id = css_id(&to->css);
3220
3221    if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
3222        mem_cgroup_swap_statistics(from, false);
3223        mem_cgroup_swap_statistics(to, true);
3224        /*
3225         * This function is only called from task migration context now.
3226         * It postpones res_counter and refcount handling till the end
3227         * of task migration(mem_cgroup_clear_mc()) for performance
3228         * improvement. But we cannot postpone mem_cgroup_get(to)
3229         * because if the process that has been moved to @to does
3230         * swap-in, the refcount of @to might be decreased to 0.
3231         */
3232        mem_cgroup_get(to);
3233        if (need_fixup) {
3234            if (!mem_cgroup_is_root(from))
3235                res_counter_uncharge(&from->memsw, PAGE_SIZE);
3236            mem_cgroup_put(from);
3237            /*
3238             * we charged both to->res and to->memsw, so we should
3239             * uncharge to->res.
3240             */
3241            if (!mem_cgroup_is_root(to))
3242                res_counter_uncharge(&to->res, PAGE_SIZE);
3243        }
3244        return 0;
3245    }
3246    return -EINVAL;
3247}
3248#else
3249static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
3250        struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
3251{
3252    return -EINVAL;
3253}
3254#endif
3255
3256/*
3257 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3258 * page belongs to.
3259 */
3260int mem_cgroup_prepare_migration(struct page *page,
3261    struct page *newpage, struct mem_cgroup **ptr, gfp_t gfp_mask)
3262{
3263    struct mem_cgroup *mem = NULL;
3264    struct page_cgroup *pc;
3265    enum charge_type ctype;
3266    int ret = 0;
3267
3268    *ptr = NULL;
3269
3270    VM_BUG_ON(PageTransHuge(page));
3271    if (mem_cgroup_disabled())
3272        return 0;
3273
3274    pc = lookup_page_cgroup(page);
3275    lock_page_cgroup(pc);
3276    if (PageCgroupUsed(pc)) {
3277        mem = pc->mem_cgroup;
3278        css_get(&mem->css);
3279        /*
3280         * At migrating an anonymous page, its mapcount goes down
3281         * to 0 and uncharge() will be called. But, even if it's fully
3282         * unmapped, migration may fail and this page has to be
3283         * charged again. We set MIGRATION flag here and delay uncharge
3284         * until end_migration() is called
3285         *
3286         * Corner Case Thinking
3287         * A)
3288         * When the old page was mapped as Anon and it's unmap-and-freed
3289         * while migration was ongoing.
3290         * If unmap finds the old page, uncharge() of it will be delayed
3291         * until end_migration(). If unmap finds a new page, it's
3292         * uncharged when it make mapcount to be 1->0. If unmap code
3293         * finds swap_migration_entry, the new page will not be mapped
3294         * and end_migration() will find it(mapcount==0).
3295         *
3296         * B)
3297         * When the old page was mapped but migraion fails, the kernel
3298         * remaps it. A charge for it is kept by MIGRATION flag even
3299         * if mapcount goes down to 0. We can do remap successfully
3300         * without charging it again.
3301         *
3302         * C)
3303         * The "old" page is under lock_page() until the end of
3304         * migration, so, the old page itself will not be swapped-out.
3305         * If the new page is swapped out before end_migraton, our
3306         * hook to usual swap-out path will catch the event.
3307         */
3308        if (PageAnon(page))
3309            SetPageCgroupMigration(pc);
3310    }
3311    unlock_page_cgroup(pc);
3312    /*
3313     * If the page is not charged at this point,
3314     * we return here.
3315     */
3316    if (!mem)
3317        return 0;
3318
3319    *ptr = mem;
3320    ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, ptr, false);
3321    css_put(&mem->css);/* drop extra refcnt */
3322    if (ret || *ptr == NULL) {
3323        if (PageAnon(page)) {
3324            lock_page_cgroup(pc);
3325            ClearPageCgroupMigration(pc);
3326            unlock_page_cgroup(pc);
3327            /*
3328             * The old page may be fully unmapped while we kept it.
3329             */
3330            mem_cgroup_uncharge_page(page);
3331        }
3332        return -ENOMEM;
3333    }
3334    /*
3335     * We charge new page before it's used/mapped. So, even if unlock_page()
3336     * is called before end_migration, we can catch all events on this new
3337     * page. In the case new page is migrated but not remapped, new page's
3338     * mapcount will be finally 0 and we call uncharge in end_migration().
3339     */
3340    pc = lookup_page_cgroup(newpage);
3341    if (PageAnon(page))
3342        ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
3343    else if (page_is_file_cache(page))
3344        ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
3345    else
3346        ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
3347    __mem_cgroup_commit_charge(mem, page, 1, pc, ctype);
3348    return ret;
3349}
3350
3351/* remove redundant charge if migration failed*/
3352void mem_cgroup_end_migration(struct mem_cgroup *mem,
3353    struct page *oldpage, struct page *newpage, bool migration_ok)
3354{
3355    struct page *used, *unused;
3356    struct page_cgroup *pc;
3357
3358    if (!mem)
3359        return;
3360    /* blocks rmdir() */
3361    cgroup_exclude_rmdir(&mem->css);
3362    if (!migration_ok) {
3363        used = oldpage;
3364        unused = newpage;
3365    } else {
3366        used = newpage;
3367        unused = oldpage;
3368    }
3369    /*
3370     * We disallowed uncharge of pages under migration because mapcount
3371     * of the page goes down to zero, temporarly.
3372     * Clear the flag and check the page should be charged.
3373     */
3374    pc = lookup_page_cgroup(oldpage);
3375    lock_page_cgroup(pc);
3376    ClearPageCgroupMigration(pc);
3377    unlock_page_cgroup(pc);
3378
3379    __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE);
3380
3381    /*
3382     * If a page is a file cache, radix-tree replacement is very atomic
3383     * and we can skip this check. When it was an Anon page, its mapcount
3384     * goes down to 0. But because we added MIGRATION flage, it's not
3385     * uncharged yet. There are several case but page->mapcount check
3386     * and USED bit check in mem_cgroup_uncharge_page() will do enough
3387     * check. (see prepare_charge() also)
3388     */
3389    if (PageAnon(used))
3390        mem_cgroup_uncharge_page(used);
3391    /*
3392     * At migration, we may charge account against cgroup which has no
3393     * tasks.
3394     * So, rmdir()->pre_destroy() can be called while we do this charge.
3395     * In that case, we need to call pre_destroy() again. check it here.
3396     */
3397    cgroup_release_and_wakeup_rmdir(&mem->css);
3398}
3399
3400/*
3401 * A call to try to shrink memory usage on charge failure at shmem's swapin.
3402 * Calling hierarchical_reclaim is not enough because we should update
3403 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
3404 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
3405 * not from the memcg which this page would be charged to.
3406 * try_charge_swapin does all of these works properly.
3407 */
3408int mem_cgroup_shmem_charge_fallback(struct page *page,
3409                struct mm_struct *mm,
3410                gfp_t gfp_mask)
3411{
3412    struct mem_cgroup *mem;
3413    int ret;
3414
3415    if (mem_cgroup_disabled())
3416        return 0;
3417
3418    ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
3419    if (!ret)
3420        mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
3421
3422    return ret;
3423}
3424
3425#ifdef CONFIG_DEBUG_VM
3426static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
3427{
3428    struct page_cgroup *pc;
3429
3430    pc = lookup_page_cgroup(page);
3431    if (likely(pc) && PageCgroupUsed(pc))
3432        return pc;
3433    return NULL;
3434}
3435
3436bool mem_cgroup_bad_page_check(struct page *page)
3437{
3438    if (mem_cgroup_disabled())
3439        return false;
3440
3441    return lookup_page_cgroup_used(page) != NULL;
3442}
3443
3444void mem_cgroup_print_bad_page(struct page *page)
3445{
3446    struct page_cgroup *pc;
3447
3448    pc = lookup_page_cgroup_used(page);
3449    if (pc) {
3450        int ret = -1;
3451        char *path;
3452
3453        printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p",
3454               pc, pc->flags, pc->mem_cgroup);
3455
3456        path = kmalloc(PATH_MAX, GFP_KERNEL);
3457        if (path) {
3458            rcu_read_lock();
3459            ret = cgroup_path(pc->mem_cgroup->css.cgroup,
3460                            path, PATH_MAX);
3461            rcu_read_unlock();
3462        }
3463
3464        printk(KERN_CONT "(%s)\n",
3465                (ret < 0) ? "cannot get the path" : path);
3466        kfree(path);
3467    }
3468}
3469#endif
3470
3471static DEFINE_MUTEX(set_limit_mutex);
3472
3473static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
3474                unsigned long long val)
3475{
3476    int retry_count;
3477    u64 memswlimit, memlimit;
3478    int ret = 0;
3479    int children = mem_cgroup_count_children(memcg);
3480    u64 curusage, oldusage;
3481    int enlarge;
3482
3483    /*
3484     * For keeping hierarchical_reclaim simple, how long we should retry
3485     * is depends on callers. We set our retry-count to be function
3486     * of # of children which we should visit in this loop.
3487     */
3488    retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3489
3490    oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3491
3492    enlarge = 0;
3493    while (retry_count) {
3494        if (signal_pending(current)) {
3495            ret = -EINTR;
3496            break;
3497        }
3498        /*
3499         * Rather than hide all in some function, I do this in
3500         * open coded manner. You see what this really does.
3501         * We have to guarantee mem->res.limit < mem->memsw.limit.
3502         */
3503        mutex_lock(&set_limit_mutex);
3504        memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3505        if (memswlimit < val) {
3506            ret = -EINVAL;
3507            mutex_unlock(&set_limit_mutex);
3508            break;
3509        }
3510
3511        memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3512        if (memlimit < val)
3513            enlarge = 1;
3514
3515        ret = res_counter_set_limit(&memcg->res, val);
3516        if (!ret) {
3517            if (memswlimit == val)
3518                memcg->memsw_is_minimum = true;
3519            else
3520                memcg->memsw_is_minimum = false;
3521        }
3522        mutex_unlock(&set_limit_mutex);
3523
3524        if (!ret)
3525            break;
3526
3527        mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
3528                        MEM_CGROUP_RECLAIM_SHRINK,
3529                        NULL);
3530        curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3531        /* Usage is reduced ? */
3532          if (curusage >= oldusage)
3533            retry_count--;
3534        else
3535            oldusage = curusage;
3536    }
3537    if (!ret && enlarge)
3538        memcg_oom_recover(memcg);
3539
3540    return ret;
3541}
3542
3543static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3544                    unsigned long long val)
3545{
3546    int retry_count;
3547    u64 memlimit, memswlimit, oldusage, curusage;
3548    int children = mem_cgroup_count_children(memcg);
3549    int ret = -EBUSY;
3550    int enlarge = 0;
3551
3552    /* see mem_cgroup_resize_res_limit */
3553     retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3554    oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
3555    while (retry_count) {
3556        if (signal_pending(current)) {
3557            ret = -EINTR;
3558            break;
3559        }
3560        /*
3561         * Rather than hide all in some function, I do this in
3562         * open coded manner. You see what this really does.
3563         * We have to guarantee mem->res.limit < mem->memsw.limit.
3564         */
3565        mutex_lock(&set_limit_mutex);
3566        memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3567        if (memlimit > val) {
3568            ret = -EINVAL;
3569            mutex_unlock(&set_limit_mutex);
3570            break;
3571        }
3572        memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3573        if (memswlimit < val)
3574            enlarge = 1;
3575        ret = res_counter_set_limit(&memcg->memsw, val);
3576        if (!ret) {
3577            if (memlimit == val)
3578                memcg->memsw_is_minimum = true;
3579            else
3580                memcg->memsw_is_minimum = false;
3581        }
3582        mutex_unlock(&set_limit_mutex);
3583
3584        if (!ret)
3585            break;
3586
3587        mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
3588                        MEM_CGROUP_RECLAIM_NOSWAP |
3589                        MEM_CGROUP_RECLAIM_SHRINK,
3590                        NULL);
3591        curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
3592        /* Usage is reduced ? */
3593        if (curusage >= oldusage)
3594            retry_count--;
3595        else
3596            oldusage = curusage;
3597    }
3598    if (!ret && enlarge)
3599        memcg_oom_recover(memcg);
3600    return ret;
3601}
3602
3603unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
3604                        gfp_t gfp_mask,
3605                        unsigned long *total_scanned)
3606{
3607    unsigned long nr_reclaimed = 0;
3608    struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3609    unsigned long reclaimed;
3610    int loop = 0;
3611    struct mem_cgroup_tree_per_zone *mctz;
3612    unsigned long long excess;
3613    unsigned long nr_scanned;
3614
3615    if (order > 0)
3616        return 0;
3617
3618    mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
3619    /*
3620     * This loop can run a while, specially if mem_cgroup's continuously
3621     * keep exceeding their soft limit and putting the system under
3622     * pressure
3623     */
3624    do {
3625        if (next_mz)
3626            mz = next_mz;
3627        else
3628            mz = mem_cgroup_largest_soft_limit_node(mctz);
3629        if (!mz)
3630            break;
3631
3632        nr_scanned = 0;
3633        reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone,
3634                        gfp_mask,
3635                        MEM_CGROUP_RECLAIM_SOFT,
3636                        &nr_scanned);
3637        nr_reclaimed += reclaimed;
3638        *total_scanned += nr_scanned;
3639        spin_lock(&mctz->lock);
3640
3641        /*
3642         * If we failed to reclaim anything from this memory cgroup
3643         * it is time to move on to the next cgroup
3644         */
3645        next_mz = NULL;
3646        if (!reclaimed) {
3647            do {
3648                /*
3649                 * Loop until we find yet another one.
3650                 *
3651                 * By the time we get the soft_limit lock
3652                 * again, someone might have aded the
3653                 * group back on the RB tree. Iterate to
3654                 * make sure we get a different mem.
3655                 * mem_cgroup_largest_soft_limit_node returns
3656                 * NULL if no other cgroup is present on
3657                 * the tree
3658                 */
3659                next_mz =
3660                __mem_cgroup_largest_soft_limit_node(mctz);
3661                if (next_mz == mz)
3662                    css_put(&next_mz->mem->css);
3663                else /* next_mz == NULL or other memcg */
3664                    break;
3665            } while (1);
3666        }
3667        __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
3668        excess = res_counter_soft_limit_excess(&mz->mem->res);
3669        /*
3670         * One school of thought says that we should not add
3671         * back the node to the tree if reclaim returns 0.
3672         * But our reclaim could return 0, simply because due
3673         * to priority we are exposing a smaller subset of
3674         * memory to reclaim from. Consider this as a longer
3675         * term TODO.
3676         */
3677        /* If excess == 0, no tree ops */
3678        __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
3679        spin_unlock(&mctz->lock);
3680        css_put(&mz->mem->css);
3681        loop++;
3682        /*
3683         * Could not reclaim anything and there are no more
3684         * mem cgroups to try or we seem to be looping without
3685         * reclaiming anything.
3686         */
3687        if (!nr_reclaimed &&
3688            (next_mz == NULL ||
3689            loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3690            break;
3691    } while (!nr_reclaimed);
3692    if (next_mz)
3693        css_put(&next_mz->mem->css);
3694    return nr_reclaimed;
3695}
3696
3697/*
3698 * This routine traverse page_cgroup in given list and drop them all.
3699 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3700 */
3701static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
3702                int node, int zid, enum lru_list lru)
3703{
3704    struct zone *zone;
3705    struct mem_cgroup_per_zone *mz;
3706    struct page_cgroup *pc, *busy;
3707    unsigned long flags, loop;
3708    struct list_head *list;
3709    int ret = 0;
3710
3711    zone = &NODE_DATA(node)->node_zones[zid];
3712    mz = mem_cgroup_zoneinfo(mem, node, zid);
3713    list = &mz->lists[lru];
3714
3715    loop = MEM_CGROUP_ZSTAT(mz, lru);
3716    /* give some margin against EBUSY etc...*/
3717    loop += 256;
3718    busy = NULL;
3719    while (loop--) {
3720        struct page *page;
3721
3722        ret = 0;
3723        spin_lock_irqsave(&zone->lru_lock, flags);
3724        if (list_empty(list)) {
3725            spin_unlock_irqrestore(&zone->lru_lock, flags);
3726            break;
3727        }
3728        pc = list_entry(list->prev, struct page_cgroup, lru);
3729        if (busy == pc) {
3730            list_move(&pc->lru, list);
3731            busy = NULL;
3732            spin_unlock_irqrestore(&zone->lru_lock, flags);
3733            continue;
3734        }
3735        spin_unlock_irqrestore(&zone->lru_lock, flags);
3736
3737        page = lookup_cgroup_page(pc);
3738
3739        ret = mem_cgroup_move_parent(page, pc, mem, GFP_KERNEL);
3740        if (ret == -ENOMEM)
3741            break;
3742
3743        if (ret == -EBUSY || ret == -EINVAL) {
3744            /* found lock contention or "pc" is obsolete. */
3745            busy = pc;
3746            cond_resched();
3747        } else
3748            busy = NULL;
3749    }
3750
3751    if (!ret && !list_empty(list))
3752        return -EBUSY;
3753    return ret;
3754}
3755
3756/*
3757 * make mem_cgroup's charge to be 0 if there is no task.
3758 * This enables deleting this mem_cgroup.
3759 */
3760static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
3761{
3762    int ret;
3763    int node, zid, shrink;
3764    int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
3765    struct cgroup *cgrp = mem->css.cgroup;
3766
3767    css_get(&mem->css);
3768
3769    shrink = 0;
3770    /* should free all ? */
3771    if (free_all)
3772        goto try_to_free;
3773move_account:
3774    do {
3775        ret = -EBUSY;
3776        if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
3777            goto out;
3778        ret = -EINTR;
3779        if (signal_pending(current))
3780            goto out;
3781        /* This is for making all *used* pages to be on LRU. */
3782        lru_add_drain_all();
3783        drain_all_stock_sync();
3784        ret = 0;
3785        mem_cgroup_start_move(mem);
3786        for_each_node_state(node, N_HIGH_MEMORY) {
3787            for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
3788                enum lru_list l;
3789                for_each_lru(l) {
3790                    ret = mem_cgroup_force_empty_list(mem,
3791                            node, zid, l);
3792                    if (ret)
3793                        break;
3794                }
3795            }
3796            if (ret)
3797                break;
3798        }
3799        mem_cgroup_end_move(mem);
3800        memcg_oom_recover(mem);
3801        /* it seems parent cgroup doesn't have enough mem */
3802        if (ret == -ENOMEM)
3803            goto try_to_free;
3804        cond_resched();
3805    /* "ret" should also be checked to ensure all lists are empty. */
3806    } while (mem->res.usage > 0 || ret);
3807out:
3808    css_put(&mem->css);
3809    return ret;
3810
3811try_to_free:
3812    /* returns EBUSY if there is a task or if we come here twice. */
3813    if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
3814        ret = -EBUSY;
3815        goto out;
3816    }
3817    /* we call try-to-free pages for make this cgroup empty */
3818    lru_add_drain_all();
3819    /* try to free all pages in this cgroup */
3820    shrink = 1;
3821    while (nr_retries && mem->res.usage > 0) {
3822        int progress;
3823
3824        if (signal_pending(current)) {
3825            ret = -EINTR;
3826            goto out;
3827        }
3828        progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
3829                        false, get_swappiness(mem));
3830        if (!progress) {
3831            nr_retries--;
3832            /* maybe some writeback is necessary */
3833            congestion_wait(BLK_RW_ASYNC, HZ/10);
3834        }
3835
3836    }
3837    lru_add_drain();
3838    /* try move_account...there may be some *locked* pages. */
3839    goto move_account;
3840}
3841
3842int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3843{
3844    return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3845}
3846
3847
3848static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3849{
3850    return mem_cgroup_from_cont(cont)->use_hierarchy;
3851}
3852
3853static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3854                    u64 val)
3855{
3856    int retval = 0;
3857    struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3858    struct cgroup *parent = cont->parent;
3859    struct mem_cgroup *parent_mem = NULL;
3860
3861    if (parent)
3862        parent_mem = mem_cgroup_from_cont(parent);
3863
3864    cgroup_lock();
3865    /*
3866     * If parent's use_hierarchy is set, we can't make any modifications
3867     * in the child subtrees. If it is unset, then the change can
3868     * occur, provided the current cgroup has no children.
3869     *
3870     * For the root cgroup, parent_mem is NULL, we allow value to be
3871     * set if there are no children.
3872     */
3873    if ((!parent_mem || !parent_mem->use_hierarchy) &&
3874                (val == 1 || val == 0)) {
3875        if (list_empty(&cont->children))
3876            mem->use_hierarchy = val;
3877        else
3878            retval = -EBUSY;
3879    } else
3880        retval = -EINVAL;
3881    cgroup_unlock();
3882
3883    return retval;
3884}
3885
3886
3887static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *mem,
3888                           enum mem_cgroup_stat_index idx)
3889{
3890    struct mem_cgroup *iter;
3891    long val = 0;
3892
3893    /* Per-cpu values can be negative, use a signed accumulator */
3894    for_each_mem_cgroup_tree(iter, mem)
3895        val += mem_cgroup_read_stat(iter, idx);
3896
3897    if (val < 0) /* race ? */
3898        val = 0;
3899    return val;
3900}
3901
3902static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap)
3903{
3904    u64 val;
3905
3906    if (!mem_cgroup_is_root(mem)) {
3907        if (!swap)
3908            return res_counter_read_u64(&mem->res, RES_USAGE);
3909        else
3910            return res_counter_read_u64(&mem->memsw, RES_USAGE);
3911    }
3912
3913    val = mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_CACHE);
3914    val += mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_RSS);
3915
3916    if (swap)
3917        val += mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_SWAPOUT);
3918
3919    return val << PAGE_SHIFT;
3920}
3921
3922static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
3923{
3924    struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3925    u64 val;
3926    int type, name;
3927
3928    type = MEMFILE_TYPE(cft->private);
3929    name = MEMFILE_ATTR(cft->private);
3930    switch (type) {
3931    case _MEM:
3932        if (name == RES_USAGE)
3933            val = mem_cgroup_usage(mem, false);
3934        else
3935            val = res_counter_read_u64(&mem->res, name);
3936        break;
3937    case _MEMSWAP:
3938        if (name == RES_USAGE)
3939            val = mem_cgroup_usage(mem, true);
3940        else
3941            val = res_counter_read_u64(&mem->memsw, name);
3942        break;
3943    default:
3944        BUG();
3945        break;
3946    }
3947    return val;
3948}
3949/*
3950 * The user of this function is...
3951 * RES_LIMIT.
3952 */
3953static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3954                const char *buffer)
3955{
3956    struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3957    int type, name;
3958    unsigned long long val;
3959    int ret;
3960
3961    type = MEMFILE_TYPE(cft->private);
3962    name = MEMFILE_ATTR(cft->private);
3963    switch (name) {
3964    case RES_LIMIT:
3965        if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3966            ret = -EINVAL;
3967            break;
3968        }
3969        /* This function does all necessary parse...reuse it */
3970        ret = res_counter_memparse_write_strategy(buffer, &val);
3971        if (ret)
3972            break;
3973        if (type == _MEM)
3974            ret = mem_cgroup_resize_limit(memcg, val);
3975        else
3976            ret = mem_cgroup_resize_memsw_limit(memcg, val);
3977        break;
3978    case RES_SOFT_LIMIT:
3979        ret = res_counter_memparse_write_strategy(buffer, &val);
3980        if (ret)
3981            break;
3982        /*
3983         * For memsw, soft limits are hard to implement in terms
3984         * of semantics, for now, we support soft limits for
3985         * control without swap
3986         */
3987        if (type == _MEM)
3988            ret = res_counter_set_soft_limit(&memcg->res, val);
3989        else
3990            ret = -EINVAL;
3991        break;
3992    default:
3993        ret = -EINVAL; /* should be BUG() ? */
3994        break;
3995    }
3996    return ret;
3997}
3998
3999static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
4000        unsigned long long *mem_limit, unsigned long long *memsw_limit)
4001{
4002    struct cgroup *cgroup;
4003    unsigned long long min_limit, min_memsw_limit, tmp;
4004
4005    min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4006    min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4007    cgroup = memcg->css.cgroup;
4008    if (!memcg->use_hierarchy)
4009        goto out;
4010
4011    while (cgroup->parent) {
4012        cgroup = cgroup->parent;
4013        memcg = mem_cgroup_from_cont(cgroup);
4014        if (!memcg->use_hierarchy)
4015            break;
4016        tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
4017        min_limit = min(min_limit, tmp);
4018        tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4019        min_memsw_limit = min(min_memsw_limit, tmp);
4020    }
4021out:
4022    *mem_limit = min_limit;
4023    *memsw_limit = min_memsw_limit;
4024    return;
4025}
4026
4027static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
4028{
4029    struct mem_cgroup *mem;
4030    int type, name;
4031
4032    mem = mem_cgroup_from_cont(cont);
4033    type = MEMFILE_TYPE(event);
4034    name = MEMFILE_ATTR(event);
4035    switch (name) {
4036    case RES_MAX_USAGE:
4037        if (type == _MEM)
4038            res_counter_reset_max(&mem->res);
4039        else
4040            res_counter_reset_max(&mem->memsw);
4041        break;
4042    case RES_FAILCNT:
4043        if (type == _MEM)
4044            res_counter_reset_failcnt(&mem->res);
4045        else
4046            res_counter_reset_failcnt(&mem->memsw);
4047        break;
4048    }
4049
4050    return 0;
4051}
4052
4053static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
4054                    struct cftype *cft)
4055{
4056    return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
4057}
4058
4059#ifdef CONFIG_MMU
4060static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4061                    struct cftype *cft, u64 val)
4062{
4063    struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
4064
4065    if (val >= (1 << NR_MOVE_TYPE))
4066        return -EINVAL;
4067    /*
4068     * We check this value several times in both in can_attach() and
4069     * attach(), so we need cgroup lock to prevent this value from being
4070     * inconsistent.
4071     */
4072    cgroup_lock();
4073    mem->move_charge_at_immigrate = val;
4074    cgroup_unlock();
4075
4076    return 0;
4077}
4078#else
4079static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4080                    struct cftype *cft, u64 val)
4081{
4082    return -ENOSYS;
4083}
4084#endif
4085
4086
4087/* For read statistics */
4088enum {
4089    MCS_CACHE,
4090    MCS_RSS,
4091    MCS_FILE_MAPPED,
4092    MCS_PGPGIN,
4093    MCS_PGPGOUT,
4094    MCS_SWAP,
4095    MCS_PGFAULT,
4096    MCS_PGMAJFAULT,
4097    MCS_INACTIVE_ANON,
4098    MCS_ACTIVE_ANON,
4099    MCS_INACTIVE_FILE,
4100    MCS_ACTIVE_FILE,
4101    MCS_UNEVICTABLE,
4102    NR_MCS_STAT,
4103};
4104
4105struct mcs_total_stat {
4106    s64 stat[NR_MCS_STAT];
4107};
4108
4109struct {
4110    char *local_name;
4111    char *total_name;
4112} memcg_stat_strings[NR_MCS_STAT] = {
4113    {"cache", "total_cache"},
4114    {"rss", "total_rss"},
4115    {"mapped_file", "total_mapped_file"},
4116    {"pgpgin", "total_pgpgin"},
4117    {"pgpgout", "total_pgpgout"},
4118    {"swap", "total_swap"},
4119    {"pgfault", "total_pgfault"},
4120    {"pgmajfault", "total_pgmajfault"},
4121    {"inactive_anon", "total_inactive_anon"},
4122    {"active_anon", "total_active_anon"},
4123    {"inactive_file", "total_inactive_file"},
4124    {"active_file", "total_active_file"},
4125    {"unevictable", "total_unevictable"}
4126};
4127
4128
4129static void
4130mem_cgroup_get_local_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
4131{
4132    s64 val;
4133
4134    /* per cpu stat */
4135    val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
4136    s->stat[MCS_CACHE] += val * PAGE_SIZE;
4137    val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
4138    s->stat[MCS_RSS] += val * PAGE_SIZE;
4139    val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED);
4140    s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
4141    val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGPGIN);
4142    s->stat[MCS_PGPGIN] += val;
4143    val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGPGOUT);
4144    s->stat[MCS_PGPGOUT] += val;
4145    if (do_swap_account) {
4146        val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT);
4147        s->stat[MCS_SWAP] += val * PAGE_SIZE;
4148    }
4149    val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGFAULT);
4150    s->stat[MCS_PGFAULT] += val;
4151    val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGMAJFAULT);
4152    s->stat[MCS_PGMAJFAULT] += val;
4153
4154    /* per zone stat */
4155    val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
4156    s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
4157    val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
4158    s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
4159    val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
4160    s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
4161    val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
4162    s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
4163    val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
4164    s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
4165}
4166
4167static void
4168mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
4169{
4170    struct mem_cgroup *iter;
4171
4172    for_each_mem_cgroup_tree(iter, mem)
4173        mem_cgroup_get_local_stat(iter, s);
4174}
4175
4176#ifdef CONFIG_NUMA
4177static int mem_control_numa_stat_show(struct seq_file *m, void *arg)
4178{
4179    int nid;
4180    unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
4181    unsigned long node_nr;
4182    struct cgroup *cont = m->private;
4183    struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
4184
4185    total_nr = mem_cgroup_nr_lru_pages(mem_cont);
4186    seq_printf(m, "total=%lu", total_nr);
4187    for_each_node_state(nid, N_HIGH_MEMORY) {
4188        node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid);
4189        seq_printf(m, " N%d=%lu", nid, node_nr);
4190    }
4191    seq_putc(m, '\n');
4192
4193    file_nr = mem_cgroup_nr_file_lru_pages(mem_cont);
4194    seq_printf(m, "file=%lu", file_nr);
4195    for_each_node_state(nid, N_HIGH_MEMORY) {
4196        node_nr = mem_cgroup_node_nr_file_lru_pages(mem_cont, nid);
4197        seq_printf(m, " N%d=%lu", nid, node_nr);
4198    }
4199    seq_putc(m, '\n');
4200
4201    anon_nr = mem_cgroup_nr_anon_lru_pages(mem_cont);
4202    seq_printf(m, "anon=%lu", anon_nr);
4203    for_each_node_state(nid, N_HIGH_MEMORY) {
4204        node_nr = mem_cgroup_node_nr_anon_lru_pages(mem_cont, nid);
4205        seq_printf(m, " N%d=%lu", nid, node_nr);
4206    }
4207    seq_putc(m, '\n');
4208
4209    unevictable_nr = mem_cgroup_nr_unevictable_lru_pages(mem_cont);
4210    seq_printf(m, "unevictable=%lu", unevictable_nr);
4211    for_each_node_state(nid, N_HIGH_MEMORY) {
4212        node_nr = mem_cgroup_node_nr_unevictable_lru_pages(mem_cont,
4213                                    nid);
4214        seq_printf(m, " N%d=%lu", nid, node_nr);
4215    }
4216    seq_putc(m, '\n');
4217    return 0;
4218}
4219#endif /* CONFIG_NUMA */
4220
4221static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
4222                 struct cgroup_map_cb *cb)
4223{
4224    struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
4225    struct mcs_total_stat mystat;
4226    int i;
4227
4228    memset(&mystat, 0, sizeof(mystat));
4229    mem_cgroup_get_local_stat(mem_cont, &mystat);
4230
4231
4232    for (i = 0; i < NR_MCS_STAT; i++) {
4233        if (i == MCS_SWAP && !do_swap_account)
4234            continue;
4235        cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
4236    }
4237
4238    /* Hierarchical information */
4239    {
4240        unsigned long long limit, memsw_limit;
4241        memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
4242        cb->fill(cb, "hierarchical_memory_limit", limit);
4243        if (do_swap_account)
4244            cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
4245    }
4246
4247    memset(&mystat, 0, sizeof(mystat));
4248    mem_cgroup_get_total_stat(mem_cont, &mystat);
4249    for (i = 0; i < NR_MCS_STAT; i++) {
4250        if (i == MCS_SWAP && !do_swap_account)
4251            continue;
4252        cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
4253    }
4254
4255#ifdef CONFIG_DEBUG_VM
4256    cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
4257
4258    {
4259        int nid, zid;
4260        struct mem_cgroup_per_zone *mz;
4261        unsigned long recent_rotated[2] = {0, 0};
4262        unsigned long recent_scanned[2] = {0, 0};
4263
4264        for_each_online_node(nid)
4265            for (zid = 0; zid < MAX_NR_ZONES; zid++) {
4266                mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
4267
4268                recent_rotated[0] +=
4269                    mz->reclaim_stat.recent_rotated[0];
4270                recent_rotated[1] +=
4271                    mz->reclaim_stat.recent_rotated[1];
4272                recent_scanned[0] +=
4273                    mz->reclaim_stat.recent_scanned[0];
4274                recent_scanned[1] +=
4275                    mz->reclaim_stat.recent_scanned[1];
4276            }
4277        cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
4278        cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
4279        cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
4280        cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
4281    }
4282#endif
4283
4284    return 0;
4285}
4286
4287static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
4288{
4289    struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4290
4291    return get_swappiness(memcg);
4292}
4293
4294static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
4295                       u64 val)
4296{
4297    struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4298    struct mem_cgroup *parent;
4299
4300    if (val > 100)
4301        return -EINVAL;
4302
4303    if (cgrp->parent == NULL)
4304        return -EINVAL;
4305
4306    parent = mem_cgroup_from_cont(cgrp->parent);
4307
4308    cgroup_lock();
4309
4310    /* If under hierarchy, only empty-root can set this value */
4311    if ((parent->use_hierarchy) ||
4312        (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4313        cgroup_unlock();
4314        return -EINVAL;
4315    }
4316
4317    memcg->swappiness = val;
4318
4319    cgroup_unlock();
4320
4321    return 0;
4322}
4323
4324static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4325{
4326    struct mem_cgroup_threshold_ary *t;
4327    u64 usage;
4328    int i;
4329
4330    rcu_read_lock();
4331    if (!swap)
4332        t = rcu_dereference(memcg->thresholds.primary);
4333    else
4334        t = rcu_dereference(memcg->memsw_thresholds.primary);
4335
4336    if (!t)
4337        goto unlock;
4338
4339    usage = mem_cgroup_usage(memcg, swap);
4340
4341    /*
4342     * current_threshold points to threshold just below usage.
4343     * If it's not true, a threshold was crossed after last
4344     * call of __mem_cgroup_threshold().
4345     */
4346    i = t->current_threshold;
4347
4348    /*
4349     * Iterate backward over array of thresholds starting from
4350     * current_threshold and check if a threshold is crossed.
4351     * If none of thresholds below usage is crossed, we read
4352     * only one element of the array here.
4353     */
4354    for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4355        eventfd_signal(t->entries[i].eventfd, 1);
4356
4357    /* i = current_threshold + 1 */
4358    i++;
4359
4360    /*
4361     * Iterate forward over array of thresholds starting from
4362     * current_threshold+1 and check if a threshold is crossed.
4363     * If none of thresholds above usage is crossed, we read
4364     * only one element of the array here.
4365     */
4366    for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4367        eventfd_signal(t->entries[i].eventfd, 1);
4368
4369    /* Update current_threshold */
4370    t->current_threshold = i - 1;
4371unlock:
4372    rcu_read_unlock();
4373}
4374
4375static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4376{
4377    while (memcg) {
4378        __mem_cgroup_threshold(memcg, false);
4379        if (do_swap_account)
4380            __mem_cgroup_threshold(memcg, true);
4381
4382        memcg = parent_mem_cgroup(memcg);
4383    }
4384}
4385
4386static int compare_thresholds(const void *a, const void *b)
4387{
4388    const struct mem_cgroup_threshold *_a = a;
4389    const struct mem_cgroup_threshold *_b = b;
4390
4391    return _a->threshold - _b->threshold;
4392}
4393
4394static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem)
4395{
4396    struct mem_cgroup_eventfd_list *ev;
4397
4398    list_for_each_entry(ev, &mem->oom_notify, list)
4399        eventfd_signal(ev->eventfd, 1);
4400    return 0;
4401}
4402
4403static void mem_cgroup_oom_notify(struct mem_cgroup *mem)
4404{
4405    struct mem_cgroup *iter;
4406
4407    for_each_mem_cgroup_tree(iter, mem)
4408        mem_cgroup_oom_notify_cb(iter);
4409}
4410
4411static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
4412    struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4413{
4414    struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4415    struct mem_cgroup_thresholds *thresholds;
4416    struct mem_cgroup_threshold_ary *new;
4417    int type = MEMFILE_TYPE(cft->private);
4418    u64 threshold, usage;
4419    int i, size, ret;
4420
4421    ret = res_counter_memparse_write_strategy(args, &threshold);
4422    if (ret)
4423        return ret;
4424
4425    mutex_lock(&memcg->thresholds_lock);
4426
4427    if (type == _MEM)
4428        thresholds = &memcg->thresholds;
4429    else if (type == _MEMSWAP)
4430        thresholds = &memcg->memsw_thresholds;
4431    else
4432        BUG();
4433
4434    usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4435
4436    /* Check if a threshold crossed before adding a new one */
4437    if (thresholds->primary)
4438        __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4439
4440    size = thresholds->primary ? thresholds->primary->size + 1 : 1;
4441
4442    /* Allocate memory for new array of thresholds */
4443    new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
4444            GFP_KERNEL);
4445    if (!new) {
4446        ret = -ENOMEM;
4447        goto unlock;
4448    }
4449    new->size = size;
4450
4451    /* Copy thresholds (if any) to new array */
4452    if (thresholds->primary) {
4453        memcpy(new->entries, thresholds->primary->entries, (size - 1) *
4454                sizeof(struct mem_cgroup_threshold));
4455    }
4456
4457    /* Add new threshold */
4458    new->entries[size - 1].eventfd = eventfd;
4459    new->entries[size - 1].threshold = threshold;
4460
4461    /* Sort thresholds. Registering of new threshold isn't time-critical */
4462    sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
4463            compare_thresholds, NULL);
4464
4465    /* Find current threshold */
4466    new->current_threshold = -1;
4467    for (i = 0; i < size; i++) {
4468        if (new->entries[i].threshold < usage) {
4469            /*
4470             * new->current_threshold will not be used until
4471             * rcu_assign_pointer(), so it's safe to increment
4472             * it here.
4473             */
4474            ++new->current_threshold;
4475        }
4476    }
4477
4478    /* Free old spare buffer and save old primary buffer as spare */
4479    kfree(thresholds->spare);
4480    thresholds->spare = thresholds->primary;
4481
4482    rcu_assign_pointer(thresholds->primary, new);
4483
4484    /* To be sure that nobody uses thresholds */
4485    synchronize_rcu();
4486
4487unlock:
4488    mutex_unlock(&memcg->thresholds_lock);
4489
4490    return ret;
4491}
4492
4493static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
4494    struct cftype *cft, struct eventfd_ctx *eventfd)
4495{
4496    struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4497    struct mem_cgroup_thresholds *thresholds;
4498    struct mem_cgroup_threshold_ary *new;
4499    int type = MEMFILE_TYPE(cft->private);
4500    u64 usage;
4501    int i, j, size;
4502
4503    mutex_lock(&memcg->thresholds_lock);
4504    if (type == _MEM)
4505        thresholds = &memcg->thresholds;
4506    else if (type == _MEMSWAP)
4507        thresholds = &memcg->memsw_thresholds;
4508    else
4509        BUG();
4510
4511    /*
4512     * Something went wrong if we trying to unregister a threshold
4513     * if we don't have thresholds
4514     */
4515    BUG_ON(!thresholds);
4516
4517    usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4518
4519    /* Check if a threshold crossed before removing */
4520    __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4521
4522    /* Calculate new number of threshold */
4523    size = 0;
4524    for (i = 0; i < thresholds->primary->size; i++) {
4525        if (thresholds->primary->entries[i].eventfd != eventfd)
4526            size++;
4527    }
4528
4529    new = thresholds->spare;
4530
4531    /* Set thresholds array to NULL if we don't have thresholds */
4532    if (!size) {
4533        kfree(new);
4534        new = NULL;
4535        goto swap_buffers;
4536    }
4537
4538    new->size = size;
4539
4540    /* Copy thresholds and find current threshold */
4541    new->current_threshold = -1;
4542    for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4543        if (thresholds->primary->entries[i].eventfd == eventfd)
4544            continue;
4545
4546        new->entries[j] = thresholds->primary->entries[i];
4547        if (new->entries[j].threshold < usage) {
4548            /*
4549             * new->current_threshold will not be used
4550             * until rcu_assign_pointer(), so it's safe to increment
4551             * it here.
4552             */
4553            ++new->current_threshold;
4554        }
4555        j++;
4556    }
4557
4558swap_buffers:
4559    /* Swap primary and spare array */
4560    thresholds->spare = thresholds->primary;
4561    rcu_assign_pointer(thresholds->primary, new);
4562
4563    /* To be sure that nobody uses thresholds */
4564    synchronize_rcu();
4565
4566    mutex_unlock(&memcg->thresholds_lock);
4567}
4568
4569static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4570    struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4571{
4572    struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4573    struct mem_cgroup_eventfd_list *event;
4574    int type = MEMFILE_TYPE(cft->private);
4575
4576    BUG_ON(type != _OOM_TYPE);
4577    event = kmalloc(sizeof(*event), GFP_KERNEL);
4578    if (!event)
4579        return -ENOMEM;
4580
4581    mutex_lock(&memcg_oom_mutex);
4582
4583    event->eventfd = eventfd;
4584    list_add(&event->list, &memcg->oom_notify);
4585
4586    /* already in OOM ? */
4587    if (atomic_read(&memcg->oom_lock))
4588        eventfd_signal(eventfd, 1);
4589    mutex_unlock(&memcg_oom_mutex);
4590
4591    return 0;
4592}
4593
4594static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
4595    struct cftype *cft, struct eventfd_ctx *eventfd)
4596{
4597    struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
4598    struct mem_cgroup_eventfd_list *ev, *tmp;
4599    int type = MEMFILE_TYPE(cft->private);
4600
4601    BUG_ON(type != _OOM_TYPE);
4602
4603    mutex_lock(&memcg_oom_mutex);
4604
4605    list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) {
4606        if (ev->eventfd == eventfd) {
4607            list_del(&ev->list);
4608            kfree(ev);
4609        }
4610    }
4611
4612    mutex_unlock(&memcg_oom_mutex);
4613}
4614
4615static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4616    struct cftype *cft, struct cgroup_map_cb *cb)
4617{
4618    struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
4619
4620    cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable);
4621
4622    if (atomic_read(&mem->oom_lock))
4623        cb->fill(cb, "under_oom", 1);
4624    else
4625        cb->fill(cb, "under_oom", 0);
4626    return 0;
4627}
4628
4629static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4630    struct cftype *cft, u64 val)
4631{
4632    struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
4633    struct mem_cgroup *parent;
4634
4635    /* cannot set to root cgroup and only 0 and 1 are allowed */
4636    if (!cgrp->parent || !((val == 0) || (val == 1)))
4637        return -EINVAL;
4638
4639    parent = mem_cgroup_from_cont(cgrp->parent);
4640
4641    cgroup_lock();
4642    /* oom-kill-disable is a flag for subhierarchy. */
4643    if ((parent->use_hierarchy) ||
4644        (mem->use_hierarchy && !list_empty(&cgrp->children))) {
4645        cgroup_unlock();
4646        return -EINVAL;
4647    }
4648    mem->oom_kill_disable = val;
4649    if (!val)
4650        memcg_oom_recover(mem);
4651    cgroup_unlock();
4652    return 0;
4653}
4654
4655#ifdef CONFIG_NUMA
4656static const struct file_operations mem_control_numa_stat_file_operations = {
4657    .read = seq_read,
4658    .llseek = seq_lseek,
4659    .release = single_release,
4660};
4661
4662static int mem_control_numa_stat_open(struct inode *unused, struct file *file)
4663{
4664    struct cgroup *cont = file->f_dentry->d_parent->d_fsdata;
4665
4666    file->f_op = &mem_control_numa_stat_file_operations;
4667    return single_open(file, mem_control_numa_stat_show, cont);
4668}
4669#endif /* CONFIG_NUMA */
4670
4671static struct cftype mem_cgroup_files[] = {
4672    {
4673        .name = "usage_in_bytes",
4674        .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
4675        .read_u64 = mem_cgroup_read,
4676        .register_event = mem_cgroup_usage_register_event,
4677        .unregister_event = mem_cgroup_usage_unregister_event,
4678    },
4679    {
4680        .name = "max_usage_in_bytes",
4681        .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
4682        .trigger = mem_cgroup_reset,
4683        .read_u64 = mem_cgroup_read,
4684    },
4685    {
4686        .name = "limit_in_bytes",
4687        .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
4688        .write_string = mem_cgroup_write,
4689        .read_u64 = mem_cgroup_read,
4690    },
4691    {
4692        .name = "soft_limit_in_bytes",
4693        .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4694        .write_string = mem_cgroup_write,
4695        .read_u64 = mem_cgroup_read,
4696    },
4697    {
4698        .name = "failcnt",
4699        .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
4700        .trigger = mem_cgroup_reset,
4701        .read_u64 = mem_cgroup_read,
4702    },
4703    {
4704        .name = "stat",
4705        .read_map = mem_control_stat_show,
4706    },
4707    {
4708        .name = "force_empty",
4709        .trigger = mem_cgroup_force_empty_write,
4710    },
4711    {
4712        .name = "use_hierarchy",
4713        .write_u64 = mem_cgroup_hierarchy_write,
4714        .read_u64 = mem_cgroup_hierarchy_read,
4715    },
4716    {
4717        .name = "swappiness",
4718        .read_u64 = mem_cgroup_swappiness_read,
4719        .write_u64 = mem_cgroup_swappiness_write,
4720    },
4721    {
4722        .name = "move_charge_at_immigrate",
4723        .read_u64 = mem_cgroup_move_charge_read,
4724        .write_u64 = mem_cgroup_move_charge_write,
4725    },
4726    {
4727        .name = "oom_control",
4728        .read_map = mem_cgroup_oom_control_read,
4729        .write_u64 = mem_cgroup_oom_control_write,
4730        .register_event = mem_cgroup_oom_register_event,
4731        .unregister_event = mem_cgroup_oom_unregister_event,
4732        .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4733    },
4734#ifdef CONFIG_NUMA
4735    {
4736        .name = "numa_stat",
4737        .open = mem_control_numa_stat_open,
4738        .mode = S_IRUGO,
4739    },
4740#endif
4741};
4742
4743#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4744static struct cftype memsw_cgroup_files[] = {
4745    {
4746        .name = "memsw.usage_in_bytes",
4747        .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
4748        .read_u64 = mem_cgroup_read,
4749        .register_event = mem_cgroup_usage_register_event,
4750        .unregister_event = mem_cgroup_usage_unregister_event,
4751    },
4752    {
4753        .name = "memsw.max_usage_in_bytes",
4754        .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4755        .trigger = mem_cgroup_reset,
4756        .read_u64 = mem_cgroup_read,
4757    },
4758    {
4759        .name = "memsw.limit_in_bytes",
4760        .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4761        .write_string = mem_cgroup_write,
4762        .read_u64 = mem_cgroup_read,
4763    },
4764    {
4765        .name = "memsw.failcnt",
4766        .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4767        .trigger = mem_cgroup_reset,
4768        .read_u64 = mem_cgroup_read,
4769    },
4770};
4771
4772static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4773{
4774    if (!do_swap_account)
4775        return 0;
4776    return cgroup_add_files(cont, ss, memsw_cgroup_files,
4777                ARRAY_SIZE(memsw_cgroup_files));
4778};
4779#else
4780static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
4781{
4782    return 0;
4783}
4784#endif
4785
4786static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
4787{
4788    struct mem_cgroup_per_node *pn;
4789    struct mem_cgroup_per_zone *mz;
4790    enum lru_list l;
4791    int zone, tmp = node;
4792    /*
4793     * This routine is called against possible nodes.
4794     * But it's BUG to call kmalloc() against offline node.
4795     *
4796     * TODO: this routine can waste much memory for nodes which will
4797     * never be onlined. It's better to use memory hotplug callback
4798     * function.
4799     */
4800    if (!node_state(node, N_NORMAL_MEMORY))
4801        tmp = -1;
4802    pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
4803    if (!pn)
4804        return 1;
4805
4806    mem->info.nodeinfo[node] = pn;
4807    for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4808        mz = &pn->zoneinfo[zone];
4809        for_each_lru(l)
4810            INIT_LIST_HEAD(&mz->lists[l]);
4811        mz->usage_in_excess = 0;
4812        mz->on_tree = false;
4813        mz->mem = mem;
4814    }
4815    return 0;
4816}
4817
4818static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
4819{
4820    kfree(mem->info.nodeinfo[node]);
4821}
4822
4823static struct mem_cgroup *mem_cgroup_alloc(void)
4824{
4825    struct mem_cgroup *mem;
4826    int size = sizeof(struct mem_cgroup);
4827
4828    /* Can be very big if MAX_NUMNODES is very big */
4829    if (size < PAGE_SIZE)
4830        mem = kzalloc(size, GFP_KERNEL);
4831    else
4832        mem = vzalloc(size);
4833
4834    if (!mem)
4835        return NULL;
4836
4837    mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4838    if (!mem->stat)
4839        goto out_free;
4840    spin_lock_init(&mem->pcp_counter_lock);
4841    return mem;
4842
4843out_free:
4844    if (size < PAGE_SIZE)
4845        kfree(mem);
4846    else
4847        vfree(mem);
4848    return NULL;
4849}
4850
4851/*
4852 * At destroying mem_cgroup, references from swap_cgroup can remain.
4853 * (scanning all at force_empty is too costly...)
4854 *
4855 * Instead of clearing all references at force_empty, we remember
4856 * the number of reference from swap_cgroup and free mem_cgroup when
4857 * it goes down to 0.
4858 *
4859 * Removal of cgroup itself succeeds regardless of refs from swap.
4860 */
4861
4862static void __mem_cgroup_free(struct mem_cgroup *mem)
4863{
4864    int node;
4865
4866    mem_cgroup_remove_from_trees(mem);
4867    free_css_id(&mem_cgroup_subsys, &mem->css);
4868
4869    for_each_node_state(node, N_POSSIBLE)
4870        free_mem_cgroup_per_zone_info(mem, node);
4871
4872    free_percpu(mem->stat);
4873    if (sizeof(struct mem_cgroup) < PAGE_SIZE)
4874        kfree(mem);
4875    else
4876        vfree(mem);
4877}
4878
4879static void mem_cgroup_get(struct mem_cgroup *mem)
4880{
4881    atomic_inc(&mem->refcnt);
4882}
4883
4884static void __mem_cgroup_put(struct mem_cgroup *mem, int count)
4885{
4886    if (atomic_sub_and_test(count, &mem->refcnt)) {
4887        struct mem_cgroup *parent = parent_mem_cgroup(mem);
4888        __mem_cgroup_free(mem);
4889        if (parent)
4890            mem_cgroup_put(parent);
4891    }
4892}
4893
4894static void mem_cgroup_put(struct mem_cgroup *mem)
4895{
4896    __mem_cgroup_put(mem, 1);
4897}
4898
4899/*
4900 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4901 */
4902static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
4903{
4904    if (!mem->res.parent)
4905        return NULL;
4906    return mem_cgroup_from_res_counter(mem->res.parent, res);
4907}
4908
4909#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4910static void __init enable_swap_cgroup(void)
4911{
4912    if (!mem_cgroup_disabled() && really_do_swap_account)
4913        do_swap_account = 1;
4914}
4915#else
4916static void __init enable_swap_cgroup(void)
4917{
4918}
4919#endif
4920
4921static int mem_cgroup_soft_limit_tree_init(void)
4922{
4923    struct mem_cgroup_tree_per_node *rtpn;
4924    struct mem_cgroup_tree_per_zone *rtpz;
4925    int tmp, node, zone;
4926
4927    for_each_node_state(node, N_POSSIBLE) {
4928        tmp = node;
4929        if (!node_state(node, N_NORMAL_MEMORY))
4930            tmp = -1;
4931        rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4932        if (!rtpn)
4933            return 1;
4934
4935        soft_limit_tree.rb_tree_per_node[node] = rtpn;
4936
4937        for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4938            rtpz = &rtpn->rb_tree_per_zone[zone];
4939            rtpz->rb_root = RB_ROOT;
4940            spin_lock_init(&rtpz->lock);
4941        }
4942    }
4943    return 0;
4944}
4945
4946static struct cgroup_subsys_state * __ref
4947mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
4948{
4949    struct mem_cgroup *mem, *parent;
4950    long error = -ENOMEM;
4951    int node;
4952
4953    mem = mem_cgroup_alloc();
4954    if (!mem)
4955        return ERR_PTR(error);
4956
4957    for_each_node_state(node, N_POSSIBLE)
4958        if (alloc_mem_cgroup_per_zone_info(mem, node))
4959            goto free_out;
4960
4961    /* root ? */
4962    if (cont->parent == NULL) {
4963        int cpu;
4964        enable_swap_cgroup();
4965        parent = NULL;
4966        root_mem_cgroup = mem;
4967        if (mem_cgroup_soft_limit_tree_init())
4968            goto free_out;
4969        for_each_possible_cpu(cpu) {
4970            struct memcg_stock_pcp *stock =
4971                        &per_cpu(memcg_stock, cpu);
4972            INIT_WORK(&stock->work, drain_local_stock);
4973        }
4974        hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
4975    } else {
4976        parent = mem_cgroup_from_cont(cont->parent);
4977        mem->use_hierarchy = parent->use_hierarchy;
4978        mem->oom_kill_disable = parent->oom_kill_disable;
4979    }
4980
4981    if (parent && parent->use_hierarchy) {
4982        res_counter_init(&mem->res, &parent->res);
4983        res_counter_init(&mem->memsw, &parent->memsw);
4984        /*
4985         * We increment refcnt of the parent to ensure that we can
4986         * safely access it on res_counter_charge/uncharge.
4987         * This refcnt will be decremented when freeing this
4988         * mem_cgroup(see mem_cgroup_put).
4989         */
4990        mem_cgroup_get(parent);
4991    } else {
4992        res_counter_init(&mem->res, NULL);
4993        res_counter_init(&mem->memsw, NULL);
4994    }
4995    mem->last_scanned_child = 0;
4996    mem->last_scanned_node = MAX_NUMNODES;
4997    INIT_LIST_HEAD(&mem->oom_notify);
4998
4999    if (parent)
5000        mem->swappiness = get_swappiness(parent);
5001    atomic_set(&mem->refcnt, 1);
5002    mem->move_charge_at_immigrate = 0;
5003    mutex_init(&mem->thresholds_lock);
5004    return &mem->css;
5005free_out:
5006    __mem_cgroup_free(mem);
5007    root_mem_cgroup = NULL;
5008    return ERR_PTR(error);
5009}
5010
5011static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
5012                    struct cgroup *cont)
5013{
5014    struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
5015
5016    return mem_cgroup_force_empty(mem, false);
5017}
5018
5019static void mem_cgroup_destroy(struct cgroup_subsys *ss,
5020                struct cgroup *cont)
5021{
5022    struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
5023
5024    mem_cgroup_put(mem);
5025}
5026
5027static int mem_cgroup_populate(struct cgroup_subsys *ss,
5028                struct cgroup *cont)
5029{
5030    int ret;
5031
5032    ret = cgroup_add_files(cont, ss, mem_cgroup_files,
5033                ARRAY_SIZE(mem_cgroup_files));
5034
5035    if (!ret)
5036        ret = register_memsw_files(cont, ss);
5037    return ret;
5038}
5039
5040#ifdef CONFIG_MMU
5041/* Handlers for move charge at task migration. */
5042#define PRECHARGE_COUNT_AT_ONCE 256
5043static int mem_cgroup_do_precharge(unsigned long count)
5044{
5045    int ret = 0;
5046    int batch_count = PRECHARGE_COUNT_AT_ONCE;
5047    struct mem_cgroup *mem = mc.to;
5048
5049    if (mem_cgroup_is_root(mem)) {
5050        mc.precharge += count;
5051        /* we don't need css_get for root */
5052        return ret;
5053    }
5054    /* try to charge at once */
5055    if (count > 1) {
5056        struct res_counter *dummy;
5057        /*
5058         * "mem" cannot be under rmdir() because we've already checked
5059         * by cgroup_lock_live_cgroup() that it is not removed and we
5060         * are still under the same cgroup_mutex. So we can postpone
5061         * css_get().
5062         */
5063        if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy))
5064            goto one_by_one;
5065        if (do_swap_account && res_counter_charge(&mem->memsw,
5066                        PAGE_SIZE * count, &dummy)) {
5067            res_counter_uncharge(&mem->res, PAGE_SIZE * count);
5068            goto one_by_one;
5069        }
5070        mc.precharge += count;
5071        return ret;
5072    }
5073one_by_one:
5074    /* fall back to one by one charge */
5075    while (count--) {
5076        if (signal_pending(current)) {
5077            ret = -EINTR;
5078            break;
5079        }
5080        if (!batch_count--) {
5081            batch_count = PRECHARGE_COUNT_AT_ONCE;
5082            cond_resched();
5083        }
5084        ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, 1, &mem, false);
5085        if (ret || !mem)
5086            /* mem_cgroup_clear_mc() will do uncharge later */
5087            return -ENOMEM;
5088        mc.precharge++;
5089    }
5090    return ret;
5091}
5092
5093/**
5094 * is_target_pte_for_mc - check a pte whether it is valid for move charge
5095 * @vma: the vma the pte to be checked belongs
5096 * @addr: the address corresponding to the pte to be checked
5097 * @ptent: the pte to be checked
5098 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5099 *
5100 * Returns
5101 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5102 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5103 * move charge. if @target is not NULL, the page is stored in target->page
5104 * with extra refcnt got(Callers should handle it).
5105 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5106 * target for charge migration. if @target is not NULL, the entry is stored
5107 * in target->ent.
5108 *
5109 * Called with pte lock held.
5110 */
5111union mc_target {
5112    struct page *page;
5113    swp_entry_t ent;
5114};
5115
5116enum mc_target_type {
5117    MC_TARGET_NONE, /* not used */
5118    MC_TARGET_PAGE,
5119    MC_TARGET_SWAP,
5120};
5121
5122static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5123                        unsigned long addr, pte_t ptent)
5124{
5125    struct page *page = vm_normal_page(vma, addr, ptent);
5126
5127    if (!page || !page_mapped(page))
5128        return NULL;
5129    if (PageAnon(page)) {
5130        /* we don't move shared anon */
5131        if (!move_anon() || page_mapcount(page) > 2)
5132            return NULL;
5133    } else if (!move_file())
5134        /* we ignore mapcount for file pages */
5135        return NULL;
5136    if (!get_page_unless_zero(page))
5137        return NULL;
5138
5139    return page;
5140}
5141
5142static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5143            unsigned long addr, pte_t ptent, swp_entry_t *entry)
5144{
5145    int usage_count;
5146    struct page *page = NULL;
5147    swp_entry_t ent = pte_to_swp_entry(ptent);
5148
5149    if (!move_anon() || non_swap_entry(ent))
5150        return NULL;
5151    usage_count = mem_cgroup_count_swap_user(ent, &page);
5152    if (usage_count > 1) { /* we don't move shared anon */
5153        if (page)
5154            put_page(page);
5155        return NULL;
5156    }
5157    if (do_swap_account)
5158        entry->val = ent.val;
5159
5160    return page;
5161}
5162
5163static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5164            unsigned long addr, pte_t ptent, swp_entry_t *entry)
5165{
5166    struct page *page = NULL;
5167    struct inode *inode;
5168    struct address_space *mapping;
5169    pgoff_t pgoff;
5170
5171    if (!vma->vm_file) /* anonymous vma */
5172        return NULL;
5173    if (!move_file())
5174        return NULL;
5175
5176    inode = vma->vm_file->f_path.dentry->d_inode;
5177    mapping = vma->vm_file->f_mapping;
5178    if (pte_none(ptent))
5179        pgoff = linear_page_index(vma, addr);
5180    else /* pte_file(ptent) is true */
5181        pgoff = pte_to_pgoff(ptent);
5182
5183    /* page is moved even if it's not RSS of this task(page-faulted). */
5184    if (!mapping_cap_swap_backed(mapping)) { /* normal file */
5185        page = find_get_page(mapping, pgoff);
5186    } else { /* shmem/tmpfs file. we should take account of swap too. */
5187        swp_entry_t ent;
5188        mem_cgroup_get_shmem_target(inode, pgoff, &page, &ent);
5189        if (do_swap_account)
5190            entry->val = ent.val;
5191    }
5192
5193    return page;
5194}
5195
5196static int is_target_pte_for_mc(struct vm_area_struct *vma,
5197        unsigned long addr, pte_t ptent, union mc_target *target)
5198{
5199    struct page *page = NULL;
5200    struct page_cgroup *pc;
5201    int ret = 0;
5202    swp_entry_t ent = { .val = 0 };
5203
5204    if (pte_present(ptent))
5205        page = mc_handle_present_pte(vma, addr, ptent);
5206    else if (is_swap_pte(ptent))
5207        page = mc_handle_swap_pte(vma, addr, ptent, &ent);
5208    else if (pte_none(ptent) || pte_file(ptent))
5209        page = mc_handle_file_pte(vma, addr, ptent, &ent);
5210
5211    if (!page && !ent.val)
5212        return 0;
5213    if (page) {
5214        pc = lookup_page_cgroup(page);
5215        /*
5216         * Do only loose check w/o page_cgroup lock.
5217         * mem_cgroup_move_account() checks the pc is valid or not under
5218         * the lock.
5219         */
5220        if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5221            ret = MC_TARGET_PAGE;
5222            if (target)
5223                target->page = page;
5224        }
5225        if (!ret || !target)
5226            put_page(page);
5227    }
5228    /* There is a swap entry and a page doesn't exist or isn't charged */
5229    if (ent.val && !ret &&
5230            css_id(&mc.from->css) == lookup_swap_cgroup(ent)) {
5231        ret = MC_TARGET_SWAP;
5232        if (target)
5233            target->ent = ent;
5234    }
5235    return ret;
5236}
5237
5238static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5239                    unsigned long addr, unsigned long end,
5240                    struct mm_walk *walk)
5241{
5242    struct vm_area_struct *vma = walk->private;
5243    pte_t *pte;
5244    spinlock_t *ptl;
5245
5246    split_huge_page_pmd(walk->mm, pmd);
5247
5248    pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5249    for (; addr != end; pte++, addr += PAGE_SIZE)
5250        if (is_target_pte_for_mc(vma, addr, *pte, NULL))
5251            mc.precharge++; /* increment precharge temporarily */
5252    pte_unmap_unlock(pte - 1, ptl);
5253    cond_resched();
5254
5255    return 0;
5256}
5257
5258static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5259{
5260    unsigned long precharge;
5261    struct vm_area_struct *vma;
5262
5263    down_read(&mm->mmap_sem);
5264    for (vma = mm->mmap; vma; vma = vma->vm_next) {
5265        struct mm_walk mem_cgroup_count_precharge_walk = {
5266            .pmd_entry = mem_cgroup_count_precharge_pte_range,
5267            .mm = mm,
5268            .private = vma,
5269        };
5270        if (is_vm_hugetlb_page(vma))
5271            continue;
5272        walk_page_range(vma->vm_start, vma->vm_end,
5273                    &mem_cgroup_count_precharge_walk);
5274    }
5275    up_read(&mm->mmap_sem);
5276
5277    precharge = mc.precharge;
5278    mc.precharge = 0;
5279
5280    return precharge;
5281}
5282
5283static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5284{
5285    unsigned long precharge = mem_cgroup_count_precharge(mm);
5286
5287    VM_BUG_ON(mc.moving_task);
5288    mc.moving_task = current;
5289    return mem_cgroup_do_precharge(precharge);
5290}
5291
5292/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5293static void __mem_cgroup_clear_mc(void)
5294{
5295    struct mem_cgroup *from = mc.from;
5296    struct mem_cgroup *to = mc.to;
5297
5298    /* we must uncharge all the leftover precharges from mc.to */
5299    if (mc.precharge) {
5300        __mem_cgroup_cancel_charge(mc.to, mc.precharge);
5301        mc.precharge = 0;
5302    }
5303    /*
5304     * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5305     * we must uncharge here.
5306     */
5307    if (mc.moved_charge) {
5308        __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
5309        mc.moved_charge = 0;
5310    }
5311    /* we must fixup refcnts and charges */
5312    if (mc.moved_swap) {
5313        /* uncharge swap account from the old cgroup */
5314        if (!mem_cgroup_is_root(mc.from))
5315            res_counter_uncharge(&mc.from->memsw,
5316                        PAGE_SIZE * mc.moved_swap);
5317        __mem_cgroup_put(mc.from, mc.moved_swap);
5318
5319        if (!mem_cgroup_is_root(mc.to)) {
5320            /*
5321             * we charged both to->res and to->memsw, so we should
5322             * uncharge to->res.
5323             */
5324            res_counter_uncharge(&mc.to->res,
5325                        PAGE_SIZE * mc.moved_swap);
5326        }
5327        /* we've already done mem_cgroup_get(mc.to) */
5328        mc.moved_swap = 0;
5329    }
5330    memcg_oom_recover(from);
5331    memcg_oom_recover(to);
5332    wake_up_all(&mc.waitq);
5333}
5334
5335static void mem_cgroup_clear_mc(void)
5336{
5337    struct mem_cgroup *from = mc.from;
5338
5339    /*
5340     * we must clear moving_task before waking up waiters at the end of
5341     * task migration.
5342     */
5343    mc.moving_task = NULL;
5344    __mem_cgroup_clear_mc();
5345    spin_lock(&mc.lock);
5346    mc.from = NULL;
5347    mc.to = NULL;
5348    spin_unlock(&mc.lock);
5349    mem_cgroup_end_move(from);
5350}
5351
5352static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
5353                struct cgroup *cgroup,
5354                struct task_struct *p)
5355{
5356    int ret = 0;
5357    struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup);
5358
5359    if (mem->move_charge_at_immigrate) {
5360        struct mm_struct *mm;
5361        struct mem_cgroup *from = mem_cgroup_from_task(p);
5362
5363        VM_BUG_ON(from == mem);
5364
5365        mm = get_task_mm(p);
5366        if (!mm)
5367            return 0;
5368        /* We move charges only when we move a owner of the mm */
5369        if (mm->owner == p) {
5370            VM_BUG_ON(mc.from);
5371            VM_BUG_ON(mc.to);
5372            VM_BUG_ON(mc.precharge);
5373            VM_BUG_ON(mc.moved_charge);
5374            VM_BUG_ON(mc.moved_swap);
5375            mem_cgroup_start_move(from);
5376            spin_lock(&mc.lock);
5377            mc.from = from;
5378            mc.to = mem;
5379            spin_unlock(&mc.lock);
5380            /* We set mc.moving_task later */
5381
5382            ret = mem_cgroup_precharge_mc(mm);
5383            if (ret)
5384                mem_cgroup_clear_mc();
5385        }
5386        mmput(mm);
5387    }
5388    return ret;
5389}
5390
5391static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
5392                struct cgroup *cgroup,
5393                struct task_struct *p)
5394{
5395    mem_cgroup_clear_mc();
5396}
5397
5398static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5399                unsigned long addr, unsigned long end,
5400                struct mm_walk *walk)
5401{
5402    int ret = 0;
5403    struct vm_area_struct *vma = walk->private;
5404    pte_t *pte;
5405    spinlock_t *ptl;
5406
5407    split_huge_page_pmd(walk->mm, pmd);
5408retry:
5409    pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5410    for (; addr != end; addr += PAGE_SIZE) {
5411        pte_t ptent = *(pte++);
5412        union mc_target target;
5413        int type;
5414        struct page *page;
5415        struct page_cgroup *pc;
5416        swp_entry_t ent;
5417
5418        if (!mc.precharge)
5419            break;
5420
5421        type = is_target_pte_for_mc(vma, addr, ptent, &target);
5422        switch (type) {
5423        case MC_TARGET_PAGE:
5424            page = target.page;
5425            if (isolate_lru_page(page))
5426                goto put;
5427            pc = lookup_page_cgroup(page);
5428            if (!mem_cgroup_move_account(page, 1, pc,
5429                             mc.from, mc.to, false)) {
5430                mc.precharge--;
5431                /* we uncharge from mc.from later. */
5432                mc.moved_charge++;
5433            }
5434            putback_lru_page(page);
5435put: /* is_target_pte_for_mc() gets the page */
5436            put_page(page);
5437            break;
5438        case MC_TARGET_SWAP:
5439            ent = target.ent;
5440            if (!mem_cgroup_move_swap_account(ent,
5441                        mc.from, mc.to, false)) {
5442                mc.precharge--;
5443                /* we fixup refcnts and charges later. */
5444                mc.moved_swap++;
5445            }
5446            break;
5447        default:
5448            break;
5449        }
5450    }
5451    pte_unmap_unlock(pte - 1, ptl);
5452    cond_resched();
5453
5454    if (addr != end) {
5455        /*
5456         * We have consumed all precharges we got in can_attach().
5457         * We try charge one by one, but don't do any additional
5458         * charges to mc.to if we have failed in charge once in attach()
5459         * phase.
5460         */
5461        ret = mem_cgroup_do_precharge(1);
5462        if (!ret)
5463            goto retry;
5464    }
5465
5466    return ret;
5467}
5468
5469static void mem_cgroup_move_charge(struct mm_struct *mm)
5470{
5471    struct vm_area_struct *vma;
5472
5473    lru_add_drain_all();
5474retry:
5475    if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5476        /*
5477         * Someone who are holding the mmap_sem might be waiting in
5478         * waitq. So we cancel all extra charges, wake up all waiters,
5479         * and retry. Because we cancel precharges, we might not be able
5480         * to move enough charges, but moving charge is a best-effort
5481         * feature anyway, so it wouldn't be a big problem.
5482         */
5483        __mem_cgroup_clear_mc();
5484        cond_resched();
5485        goto retry;
5486    }
5487    for (vma = mm->mmap; vma; vma = vma->vm_next) {
5488        int ret;
5489        struct mm_walk mem_cgroup_move_charge_walk = {
5490            .pmd_entry = mem_cgroup_move_charge_pte_range,
5491            .mm = mm,
5492            .private = vma,
5493        };
5494        if (is_vm_hugetlb_page(vma))
5495            continue;
5496        ret = walk_page_range(vma->vm_start, vma->vm_end,
5497                        &mem_cgroup_move_charge_walk);
5498        if (ret)
5499            /*
5500             * means we have consumed all precharges and failed in
5501             * doing additional charge. Just abandon here.
5502             */
5503            break;
5504    }
5505    up_read(&mm->mmap_sem);
5506}
5507
5508static void mem_cgroup_move_task(struct cgroup_subsys *ss,
5509                struct cgroup *cont,
5510                struct cgroup *old_cont,
5511                struct task_struct *p)
5512{
5513    struct mm_struct *mm = get_task_mm(p);
5514
5515    if (mm) {
5516        if (mc.to)
5517            mem_cgroup_move_charge(mm);
5518        put_swap_token(mm);
5519        mmput(mm);
5520    }
5521    if (mc.to)
5522        mem_cgroup_clear_mc();
5523}
5524#else /* !CONFIG_MMU */
5525static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
5526                struct cgroup *cgroup,
5527                struct task_struct *p)
5528{
5529    return 0;
5530}
5531static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
5532                struct cgroup *cgroup,
5533                struct task_struct *p)
5534{
5535}
5536static void mem_cgroup_move_task(struct cgroup_subsys *ss,
5537                struct cgroup *cont,
5538                struct cgroup *old_cont,
5539                struct task_struct *p)
5540{
5541}
5542#endif
5543
5544struct cgroup_subsys mem_cgroup_subsys = {
5545    .name = "memory",
5546    .subsys_id = mem_cgroup_subsys_id,
5547    .create = mem_cgroup_create,
5548    .pre_destroy = mem_cgroup_pre_destroy,
5549    .destroy = mem_cgroup_destroy,
5550    .populate = mem_cgroup_populate,
5551    .can_attach = mem_cgroup_can_attach,
5552    .cancel_attach = mem_cgroup_cancel_attach,
5553    .attach = mem_cgroup_move_task,
5554    .early_init = 0,
5555    .use_id = 1,
5556};
5557
5558#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5559static int __init enable_swap_account(char *s)
5560{
5561    /* consider enabled if no parameter or 1 is given */
5562    if (!strcmp(s, "1"))
5563        really_do_swap_account = 1;
5564    else if (!strcmp(s, "0"))
5565        really_do_swap_account = 0;
5566    return 1;
5567}
5568__setup("swapaccount=", enable_swap_account);
5569
5570#endif
5571

Archive Download this file



interactive