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/slab.h>
39#include <linux/swap.h>
40#include <linux/swapops.h>
41#include <linux/spinlock.h>
42#include <linux/eventfd.h>
43#include <linux/sort.h>
44#include <linux/fs.h>
45#include <linux/seq_file.h>
46#include <linux/vmalloc.h>
47#include <linux/mm_inline.h>
48#include <linux/page_cgroup.h>
49#include <linux/cpu.h>
50#include "internal.h"
51
52#include <asm/uaccess.h>
53
54struct cgroup_subsys mem_cgroup_subsys __read_mostly;
55#define MEM_CGROUP_RECLAIM_RETRIES 5
56struct mem_cgroup *root_mem_cgroup __read_mostly;
57
58#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
59/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
60int do_swap_account __read_mostly;
61static int really_do_swap_account __initdata = 1; /* for remember boot option*/
62#else
63#define do_swap_account (0)
64#endif
65
66/*
67 * Per memcg event counter is incremented at every pagein/pageout. This counter
68 * is used for trigger some periodic events. This is straightforward and better
69 * than using jiffies etc. to handle periodic memcg event.
70 *
71 * These values will be used as !((event) & ((1 <<(thresh)) - 1))
72 */
73#define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */
74#define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */
75
76/*
77 * Statistics for memory cgroup.
78 */
79enum mem_cgroup_stat_index {
80    /*
81     * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
82     */
83    MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
84    MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
85    MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
86    MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
87    MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
88    MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
89    MEM_CGROUP_EVENTS, /* incremented at every pagein/pageout */
90
91    MEM_CGROUP_STAT_NSTATS,
92};
93
94struct mem_cgroup_stat_cpu {
95    s64 count[MEM_CGROUP_STAT_NSTATS];
96};
97
98/*
99 * per-zone information in memory controller.
100 */
101struct mem_cgroup_per_zone {
102    /*
103     * spin_lock to protect the per cgroup LRU
104     */
105    struct list_head lists[NR_LRU_LISTS];
106    unsigned long count[NR_LRU_LISTS];
107
108    struct zone_reclaim_stat reclaim_stat;
109    struct rb_node tree_node; /* RB tree node */
110    unsigned long long usage_in_excess;/* Set to the value by which */
111                        /* the soft limit is exceeded*/
112    bool on_tree;
113    struct mem_cgroup *mem; /* Back pointer, we cannot */
114                        /* use container_of */
115};
116/* Macro for accessing counter */
117#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
118
119struct mem_cgroup_per_node {
120    struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
121};
122
123struct mem_cgroup_lru_info {
124    struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
125};
126
127/*
128 * Cgroups above their limits are maintained in a RB-Tree, independent of
129 * their hierarchy representation
130 */
131
132struct mem_cgroup_tree_per_zone {
133    struct rb_root rb_root;
134    spinlock_t lock;
135};
136
137struct mem_cgroup_tree_per_node {
138    struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
139};
140
141struct mem_cgroup_tree {
142    struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
143};
144
145static struct mem_cgroup_tree soft_limit_tree __read_mostly;
146
147struct mem_cgroup_threshold {
148    struct eventfd_ctx *eventfd;
149    u64 threshold;
150};
151
152/* For threshold */
153struct mem_cgroup_threshold_ary {
154    /* An array index points to threshold just below usage. */
155    int current_threshold;
156    /* Size of entries[] */
157    unsigned int size;
158    /* Array of thresholds */
159    struct mem_cgroup_threshold entries[0];
160};
161
162struct mem_cgroup_thresholds {
163    /* Primary thresholds array */
164    struct mem_cgroup_threshold_ary *primary;
165    /*
166     * Spare threshold array.
167     * This is needed to make mem_cgroup_unregister_event() "never fail".
168     * It must be able to store at least primary->size - 1 entries.
169     */
170    struct mem_cgroup_threshold_ary *spare;
171};
172
173/* for OOM */
174struct mem_cgroup_eventfd_list {
175    struct list_head list;
176    struct eventfd_ctx *eventfd;
177};
178
179static void mem_cgroup_threshold(struct mem_cgroup *mem);
180static void mem_cgroup_oom_notify(struct mem_cgroup *mem);
181
182/*
183 * The memory controller data structure. The memory controller controls both
184 * page cache and RSS per cgroup. We would eventually like to provide
185 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
186 * to help the administrator determine what knobs to tune.
187 *
188 * TODO: Add a water mark for the memory controller. Reclaim will begin when
189 * we hit the water mark. May be even add a low water mark, such that
190 * no reclaim occurs from a cgroup at it's low water mark, this is
191 * a feature that will be implemented much later in the future.
192 */
193struct mem_cgroup {
194    struct cgroup_subsys_state css;
195    /*
196     * the counter to account for memory usage
197     */
198    struct res_counter res;
199    /*
200     * the counter to account for mem+swap usage.
201     */
202    struct res_counter memsw;
203    /*
204     * Per cgroup active and inactive list, similar to the
205     * per zone LRU lists.
206     */
207    struct mem_cgroup_lru_info info;
208
209    /*
210      protect against reclaim related member.
211    */
212    spinlock_t reclaim_param_lock;
213
214    int prev_priority; /* for recording reclaim priority */
215
216    /*
217     * While reclaiming in a hierarchy, we cache the last child we
218     * reclaimed from.
219     */
220    int last_scanned_child;
221    /*
222     * Should the accounting and control be hierarchical, per subtree?
223     */
224    bool use_hierarchy;
225    atomic_t oom_lock;
226    atomic_t refcnt;
227
228    unsigned int swappiness;
229    /* OOM-Killer disable */
230    int oom_kill_disable;
231
232    /* set when res.limit == memsw.limit */
233    bool memsw_is_minimum;
234
235    /* protect arrays of thresholds */
236    struct mutex thresholds_lock;
237
238    /* thresholds for memory usage. RCU-protected */
239    struct mem_cgroup_thresholds thresholds;
240
241    /* thresholds for mem+swap usage. RCU-protected */
242    struct mem_cgroup_thresholds memsw_thresholds;
243
244    /* For oom notifier event fd */
245    struct list_head oom_notify;
246
247    /*
248     * Should we move charges of a task when a task is moved into this
249     * mem_cgroup ? And what type of charges should we move ?
250     */
251    unsigned long move_charge_at_immigrate;
252    /*
253     * percpu counter.
254     */
255    struct mem_cgroup_stat_cpu *stat;
256};
257
258/* Stuffs for move charges at task migration. */
259/*
260 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
261 * left-shifted bitmap of these types.
262 */
263enum move_type {
264    MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
265    MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
266    NR_MOVE_TYPE,
267};
268
269/* "mc" and its members are protected by cgroup_mutex */
270static struct move_charge_struct {
271    struct mem_cgroup *from;
272    struct mem_cgroup *to;
273    unsigned long precharge;
274    unsigned long moved_charge;
275    unsigned long moved_swap;
276    struct task_struct *moving_task; /* a task moving charges */
277    wait_queue_head_t waitq; /* a waitq for other context */
278} mc = {
279    .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
280};
281
282static bool move_anon(void)
283{
284    return test_bit(MOVE_CHARGE_TYPE_ANON,
285                    &mc.to->move_charge_at_immigrate);
286}
287
288static bool move_file(void)
289{
290    return test_bit(MOVE_CHARGE_TYPE_FILE,
291                    &mc.to->move_charge_at_immigrate);
292}
293
294/*
295 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
296 * limit reclaim to prevent infinite loops, if they ever occur.
297 */
298#define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
299#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
300
301enum charge_type {
302    MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
303    MEM_CGROUP_CHARGE_TYPE_MAPPED,
304    MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
305    MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
306    MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
307    MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
308    NR_CHARGE_TYPE,
309};
310
311/* only for here (for easy reading.) */
312#define PCGF_CACHE (1UL << PCG_CACHE)
313#define PCGF_USED (1UL << PCG_USED)
314#define PCGF_LOCK (1UL << PCG_LOCK)
315/* Not used, but added here for completeness */
316#define PCGF_ACCT (1UL << PCG_ACCT)
317
318/* for encoding cft->private value on file */
319#define _MEM (0)
320#define _MEMSWAP (1)
321#define _OOM_TYPE (2)
322#define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
323#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
324#define MEMFILE_ATTR(val) ((val) & 0xffff)
325/* Used for OOM nofiier */
326#define OOM_CONTROL (0)
327
328/*
329 * Reclaim flags for mem_cgroup_hierarchical_reclaim
330 */
331#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
332#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
333#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
334#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
335#define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
336#define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
337
338static void mem_cgroup_get(struct mem_cgroup *mem);
339static void mem_cgroup_put(struct mem_cgroup *mem);
340static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
341static void drain_all_stock_async(void);
342
343static struct mem_cgroup_per_zone *
344mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
345{
346    return &mem->info.nodeinfo[nid]->zoneinfo[zid];
347}
348
349struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem)
350{
351    return &mem->css;
352}
353
354static struct mem_cgroup_per_zone *
355page_cgroup_zoneinfo(struct page_cgroup *pc)
356{
357    struct mem_cgroup *mem = pc->mem_cgroup;
358    int nid = page_cgroup_nid(pc);
359    int zid = page_cgroup_zid(pc);
360
361    if (!mem)
362        return NULL;
363
364    return mem_cgroup_zoneinfo(mem, nid, zid);
365}
366
367static struct mem_cgroup_tree_per_zone *
368soft_limit_tree_node_zone(int nid, int zid)
369{
370    return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
371}
372
373static struct mem_cgroup_tree_per_zone *
374soft_limit_tree_from_page(struct page *page)
375{
376    int nid = page_to_nid(page);
377    int zid = page_zonenum(page);
378
379    return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
380}
381
382static void
383__mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
384                struct mem_cgroup_per_zone *mz,
385                struct mem_cgroup_tree_per_zone *mctz,
386                unsigned long long new_usage_in_excess)
387{
388    struct rb_node **p = &mctz->rb_root.rb_node;
389    struct rb_node *parent = NULL;
390    struct mem_cgroup_per_zone *mz_node;
391
392    if (mz->on_tree)
393        return;
394
395    mz->usage_in_excess = new_usage_in_excess;
396    if (!mz->usage_in_excess)
397        return;
398    while (*p) {
399        parent = *p;
400        mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
401                    tree_node);
402        if (mz->usage_in_excess < mz_node->usage_in_excess)
403            p = &(*p)->rb_left;
404        /*
405         * We can't avoid mem cgroups that are over their soft
406         * limit by the same amount
407         */
408        else if (mz->usage_in_excess >= mz_node->usage_in_excess)
409            p = &(*p)->rb_right;
410    }
411    rb_link_node(&mz->tree_node, parent, p);
412    rb_insert_color(&mz->tree_node, &mctz->rb_root);
413    mz->on_tree = true;
414}
415
416static void
417__mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
418                struct mem_cgroup_per_zone *mz,
419                struct mem_cgroup_tree_per_zone *mctz)
420{
421    if (!mz->on_tree)
422        return;
423    rb_erase(&mz->tree_node, &mctz->rb_root);
424    mz->on_tree = false;
425}
426
427static void
428mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
429                struct mem_cgroup_per_zone *mz,
430                struct mem_cgroup_tree_per_zone *mctz)
431{
432    spin_lock(&mctz->lock);
433    __mem_cgroup_remove_exceeded(mem, mz, mctz);
434    spin_unlock(&mctz->lock);
435}
436
437
438static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page)
439{
440    unsigned long long excess;
441    struct mem_cgroup_per_zone *mz;
442    struct mem_cgroup_tree_per_zone *mctz;
443    int nid = page_to_nid(page);
444    int zid = page_zonenum(page);
445    mctz = soft_limit_tree_from_page(page);
446
447    /*
448     * Necessary to update all ancestors when hierarchy is used.
449     * because their event counter is not touched.
450     */
451    for (; mem; mem = parent_mem_cgroup(mem)) {
452        mz = mem_cgroup_zoneinfo(mem, nid, zid);
453        excess = res_counter_soft_limit_excess(&mem->res);
454        /*
455         * We have to update the tree if mz is on RB-tree or
456         * mem is over its softlimit.
457         */
458        if (excess || mz->on_tree) {
459            spin_lock(&mctz->lock);
460            /* if on-tree, remove it */
461            if (mz->on_tree)
462                __mem_cgroup_remove_exceeded(mem, mz, mctz);
463            /*
464             * Insert again. mz->usage_in_excess will be updated.
465             * If excess is 0, no tree ops.
466             */
467            __mem_cgroup_insert_exceeded(mem, mz, mctz, excess);
468            spin_unlock(&mctz->lock);
469        }
470    }
471}
472
473static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem)
474{
475    int node, zone;
476    struct mem_cgroup_per_zone *mz;
477    struct mem_cgroup_tree_per_zone *mctz;
478
479    for_each_node_state(node, N_POSSIBLE) {
480        for (zone = 0; zone < MAX_NR_ZONES; zone++) {
481            mz = mem_cgroup_zoneinfo(mem, node, zone);
482            mctz = soft_limit_tree_node_zone(node, zone);
483            mem_cgroup_remove_exceeded(mem, mz, mctz);
484        }
485    }
486}
487
488static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem)
489{
490    return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT;
491}
492
493static struct mem_cgroup_per_zone *
494__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
495{
496    struct rb_node *rightmost = NULL;
497    struct mem_cgroup_per_zone *mz;
498
499retry:
500    mz = NULL;
501    rightmost = rb_last(&mctz->rb_root);
502    if (!rightmost)
503        goto done; /* Nothing to reclaim from */
504
505    mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
506    /*
507     * Remove the node now but someone else can add it back,
508     * we will to add it back at the end of reclaim to its correct
509     * position in the tree.
510     */
511    __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
512    if (!res_counter_soft_limit_excess(&mz->mem->res) ||
513        !css_tryget(&mz->mem->css))
514        goto retry;
515done:
516    return mz;
517}
518
519static struct mem_cgroup_per_zone *
520mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
521{
522    struct mem_cgroup_per_zone *mz;
523
524    spin_lock(&mctz->lock);
525    mz = __mem_cgroup_largest_soft_limit_node(mctz);
526    spin_unlock(&mctz->lock);
527    return mz;
528}
529
530static s64 mem_cgroup_read_stat(struct mem_cgroup *mem,
531        enum mem_cgroup_stat_index idx)
532{
533    int cpu;
534    s64 val = 0;
535
536    for_each_possible_cpu(cpu)
537        val += per_cpu(mem->stat->count[idx], cpu);
538    return val;
539}
540
541static s64 mem_cgroup_local_usage(struct mem_cgroup *mem)
542{
543    s64 ret;
544
545    ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
546    ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
547    return ret;
548}
549
550static void mem_cgroup_swap_statistics(struct mem_cgroup *mem,
551                     bool charge)
552{
553    int val = (charge) ? 1 : -1;
554    this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
555}
556
557static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
558                     struct page_cgroup *pc,
559                     bool charge)
560{
561    int val = (charge) ? 1 : -1;
562
563    preempt_disable();
564
565    if (PageCgroupCache(pc))
566        __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], val);
567    else
568        __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], val);
569
570    if (charge)
571        __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]);
572    else
573        __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]);
574    __this_cpu_inc(mem->stat->count[MEM_CGROUP_EVENTS]);
575
576    preempt_enable();
577}
578
579static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
580                    enum lru_list idx)
581{
582    int nid, zid;
583    struct mem_cgroup_per_zone *mz;
584    u64 total = 0;
585
586    for_each_online_node(nid)
587        for (zid = 0; zid < MAX_NR_ZONES; zid++) {
588            mz = mem_cgroup_zoneinfo(mem, nid, zid);
589            total += MEM_CGROUP_ZSTAT(mz, idx);
590        }
591    return total;
592}
593
594static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift)
595{
596    s64 val;
597
598    val = this_cpu_read(mem->stat->count[MEM_CGROUP_EVENTS]);
599
600    return !(val & ((1 << event_mask_shift) - 1));
601}
602
603/*
604 * Check events in order.
605 *
606 */
607static void memcg_check_events(struct mem_cgroup *mem, struct page *page)
608{
609    /* threshold event is triggered in finer grain than soft limit */
610    if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) {
611        mem_cgroup_threshold(mem);
612        if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH)))
613            mem_cgroup_update_tree(mem, page);
614    }
615}
616
617static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
618{
619    return container_of(cgroup_subsys_state(cont,
620                mem_cgroup_subsys_id), struct mem_cgroup,
621                css);
622}
623
624struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
625{
626    /*
627     * mm_update_next_owner() may clear mm->owner to NULL
628     * if it races with swapoff, page migration, etc.
629     * So this can be called with p == NULL.
630     */
631    if (unlikely(!p))
632        return NULL;
633
634    return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
635                struct mem_cgroup, css);
636}
637
638static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
639{
640    struct mem_cgroup *mem = NULL;
641
642    if (!mm)
643        return NULL;
644    /*
645     * Because we have no locks, mm->owner's may be being moved to other
646     * cgroup. We use css_tryget() here even if this looks
647     * pessimistic (rather than adding locks here).
648     */
649    rcu_read_lock();
650    do {
651        mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
652        if (unlikely(!mem))
653            break;
654    } while (!css_tryget(&mem->css));
655    rcu_read_unlock();
656    return mem;
657}
658
659/*
660 * Call callback function against all cgroup under hierarchy tree.
661 */
662static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
663              int (*func)(struct mem_cgroup *, void *))
664{
665    int found, ret, nextid;
666    struct cgroup_subsys_state *css;
667    struct mem_cgroup *mem;
668
669    if (!root->use_hierarchy)
670        return (*func)(root, data);
671
672    nextid = 1;
673    do {
674        ret = 0;
675        mem = NULL;
676
677        rcu_read_lock();
678        css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
679                   &found);
680        if (css && css_tryget(css))
681            mem = container_of(css, struct mem_cgroup, css);
682        rcu_read_unlock();
683
684        if (mem) {
685            ret = (*func)(mem, data);
686            css_put(&mem->css);
687        }
688        nextid = found + 1;
689    } while (!ret && css);
690
691    return ret;
692}
693
694static inline bool mem_cgroup_is_root(struct mem_cgroup *mem)
695{
696    return (mem == root_mem_cgroup);
697}
698
699/*
700 * Following LRU functions are allowed to be used without PCG_LOCK.
701 * Operations are called by routine of global LRU independently from memcg.
702 * What we have to take care of here is validness of pc->mem_cgroup.
703 *
704 * Changes to pc->mem_cgroup happens when
705 * 1. charge
706 * 2. moving account
707 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
708 * It is added to LRU before charge.
709 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
710 * When moving account, the page is not on LRU. It's isolated.
711 */
712
713void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
714{
715    struct page_cgroup *pc;
716    struct mem_cgroup_per_zone *mz;
717
718    if (mem_cgroup_disabled())
719        return;
720    pc = lookup_page_cgroup(page);
721    /* can happen while we handle swapcache. */
722    if (!TestClearPageCgroupAcctLRU(pc))
723        return;
724    VM_BUG_ON(!pc->mem_cgroup);
725    /*
726     * We don't check PCG_USED bit. It's cleared when the "page" is finally
727     * removed from global LRU.
728     */
729    mz = page_cgroup_zoneinfo(pc);
730    MEM_CGROUP_ZSTAT(mz, lru) -= 1;
731    if (mem_cgroup_is_root(pc->mem_cgroup))
732        return;
733    VM_BUG_ON(list_empty(&pc->lru));
734    list_del_init(&pc->lru);
735    return;
736}
737
738void mem_cgroup_del_lru(struct page *page)
739{
740    mem_cgroup_del_lru_list(page, page_lru(page));
741}
742
743void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
744{
745    struct mem_cgroup_per_zone *mz;
746    struct page_cgroup *pc;
747
748    if (mem_cgroup_disabled())
749        return;
750
751    pc = lookup_page_cgroup(page);
752    /*
753     * Used bit is set without atomic ops but after smp_wmb().
754     * For making pc->mem_cgroup visible, insert smp_rmb() here.
755     */
756    smp_rmb();
757    /* unused or root page is not rotated. */
758    if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup))
759        return;
760    mz = page_cgroup_zoneinfo(pc);
761    list_move(&pc->lru, &mz->lists[lru]);
762}
763
764void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
765{
766    struct page_cgroup *pc;
767    struct mem_cgroup_per_zone *mz;
768
769    if (mem_cgroup_disabled())
770        return;
771    pc = lookup_page_cgroup(page);
772    VM_BUG_ON(PageCgroupAcctLRU(pc));
773    /*
774     * Used bit is set without atomic ops but after smp_wmb().
775     * For making pc->mem_cgroup visible, insert smp_rmb() here.
776     */
777    smp_rmb();
778    if (!PageCgroupUsed(pc))
779        return;
780
781    mz = page_cgroup_zoneinfo(pc);
782    MEM_CGROUP_ZSTAT(mz, lru) += 1;
783    SetPageCgroupAcctLRU(pc);
784    if (mem_cgroup_is_root(pc->mem_cgroup))
785        return;
786    list_add(&pc->lru, &mz->lists[lru]);
787}
788
789/*
790 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
791 * lru because the page may.be reused after it's fully uncharged (because of
792 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
793 * it again. This function is only used to charge SwapCache. It's done under
794 * lock_page and expected that zone->lru_lock is never held.
795 */
796static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
797{
798    unsigned long flags;
799    struct zone *zone = page_zone(page);
800    struct page_cgroup *pc = lookup_page_cgroup(page);
801
802    spin_lock_irqsave(&zone->lru_lock, flags);
803    /*
804     * Forget old LRU when this page_cgroup is *not* used. This Used bit
805     * is guarded by lock_page() because the page is SwapCache.
806     */
807    if (!PageCgroupUsed(pc))
808        mem_cgroup_del_lru_list(page, page_lru(page));
809    spin_unlock_irqrestore(&zone->lru_lock, flags);
810}
811
812static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
813{
814    unsigned long flags;
815    struct zone *zone = page_zone(page);
816    struct page_cgroup *pc = lookup_page_cgroup(page);
817
818    spin_lock_irqsave(&zone->lru_lock, flags);
819    /* link when the page is linked to LRU but page_cgroup isn't */
820    if (PageLRU(page) && !PageCgroupAcctLRU(pc))
821        mem_cgroup_add_lru_list(page, page_lru(page));
822    spin_unlock_irqrestore(&zone->lru_lock, flags);
823}
824
825
826void mem_cgroup_move_lists(struct page *page,
827               enum lru_list from, enum lru_list to)
828{
829    if (mem_cgroup_disabled())
830        return;
831    mem_cgroup_del_lru_list(page, from);
832    mem_cgroup_add_lru_list(page, to);
833}
834
835int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
836{
837    int ret;
838    struct mem_cgroup *curr = NULL;
839
840    task_lock(task);
841    rcu_read_lock();
842    curr = try_get_mem_cgroup_from_mm(task->mm);
843    rcu_read_unlock();
844    task_unlock(task);
845    if (!curr)
846        return 0;
847    /*
848     * We should check use_hierarchy of "mem" not "curr". Because checking
849     * use_hierarchy of "curr" here make this function true if hierarchy is
850     * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
851     * hierarchy(even if use_hierarchy is disabled in "mem").
852     */
853    if (mem->use_hierarchy)
854        ret = css_is_ancestor(&curr->css, &mem->css);
855    else
856        ret = (curr == mem);
857    css_put(&curr->css);
858    return ret;
859}
860
861/*
862 * prev_priority control...this will be used in memory reclaim path.
863 */
864int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
865{
866    int prev_priority;
867
868    spin_lock(&mem->reclaim_param_lock);
869    prev_priority = mem->prev_priority;
870    spin_unlock(&mem->reclaim_param_lock);
871
872    return prev_priority;
873}
874
875void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
876{
877    spin_lock(&mem->reclaim_param_lock);
878    if (priority < mem->prev_priority)
879        mem->prev_priority = priority;
880    spin_unlock(&mem->reclaim_param_lock);
881}
882
883void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
884{
885    spin_lock(&mem->reclaim_param_lock);
886    mem->prev_priority = priority;
887    spin_unlock(&mem->reclaim_param_lock);
888}
889
890static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
891{
892    unsigned long active;
893    unsigned long inactive;
894    unsigned long gb;
895    unsigned long inactive_ratio;
896
897    inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
898    active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
899
900    gb = (inactive + active) >> (30 - PAGE_SHIFT);
901    if (gb)
902        inactive_ratio = int_sqrt(10 * gb);
903    else
904        inactive_ratio = 1;
905
906    if (present_pages) {
907        present_pages[0] = inactive;
908        present_pages[1] = active;
909    }
910
911    return inactive_ratio;
912}
913
914int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
915{
916    unsigned long active;
917    unsigned long inactive;
918    unsigned long present_pages[2];
919    unsigned long inactive_ratio;
920
921    inactive_ratio = calc_inactive_ratio(memcg, present_pages);
922
923    inactive = present_pages[0];
924    active = present_pages[1];
925
926    if (inactive * inactive_ratio < active)
927        return 1;
928
929    return 0;
930}
931
932int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
933{
934    unsigned long active;
935    unsigned long inactive;
936
937    inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
938    active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
939
940    return (active > inactive);
941}
942
943unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
944                       struct zone *zone,
945                       enum lru_list lru)
946{
947    int nid = zone->zone_pgdat->node_id;
948    int zid = zone_idx(zone);
949    struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
950
951    return MEM_CGROUP_ZSTAT(mz, lru);
952}
953
954struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
955                              struct zone *zone)
956{
957    int nid = zone->zone_pgdat->node_id;
958    int zid = zone_idx(zone);
959    struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
960
961    return &mz->reclaim_stat;
962}
963
964struct zone_reclaim_stat *
965mem_cgroup_get_reclaim_stat_from_page(struct page *page)
966{
967    struct page_cgroup *pc;
968    struct mem_cgroup_per_zone *mz;
969
970    if (mem_cgroup_disabled())
971        return NULL;
972
973    pc = lookup_page_cgroup(page);
974    /*
975     * Used bit is set without atomic ops but after smp_wmb().
976     * For making pc->mem_cgroup visible, insert smp_rmb() here.
977     */
978    smp_rmb();
979    if (!PageCgroupUsed(pc))
980        return NULL;
981
982    mz = page_cgroup_zoneinfo(pc);
983    if (!mz)
984        return NULL;
985
986    return &mz->reclaim_stat;
987}
988
989unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
990                    struct list_head *dst,
991                    unsigned long *scanned, int order,
992                    int mode, struct zone *z,
993                    struct mem_cgroup *mem_cont,
994                    int active, int file)
995{
996    unsigned long nr_taken = 0;
997    struct page *page;
998    unsigned long scan;
999    LIST_HEAD(pc_list);
1000    struct list_head *src;
1001    struct page_cgroup *pc, *tmp;
1002    int nid = z->zone_pgdat->node_id;
1003    int zid = zone_idx(z);
1004    struct mem_cgroup_per_zone *mz;
1005    int lru = LRU_FILE * file + active;
1006    int ret;
1007
1008    BUG_ON(!mem_cont);
1009    mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
1010    src = &mz->lists[lru];
1011
1012    scan = 0;
1013    list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
1014        if (scan >= nr_to_scan)
1015            break;
1016
1017        page = pc->page;
1018        if (unlikely(!PageCgroupUsed(pc)))
1019            continue;
1020        if (unlikely(!PageLRU(page)))
1021            continue;
1022
1023        scan++;
1024        ret = __isolate_lru_page(page, mode, file);
1025        switch (ret) {
1026        case 0:
1027            list_move(&page->lru, dst);
1028            mem_cgroup_del_lru(page);
1029            nr_taken++;
1030            break;
1031        case -EBUSY:
1032            /* we don't affect global LRU but rotate in our LRU */
1033            mem_cgroup_rotate_lru_list(page, page_lru(page));
1034            break;
1035        default:
1036            break;
1037        }
1038    }
1039
1040    *scanned = scan;
1041    return nr_taken;
1042}
1043
1044#define mem_cgroup_from_res_counter(counter, member) \
1045    container_of(counter, struct mem_cgroup, member)
1046
1047static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
1048{
1049    if (do_swap_account) {
1050        if (res_counter_check_under_limit(&mem->res) &&
1051            res_counter_check_under_limit(&mem->memsw))
1052            return true;
1053    } else
1054        if (res_counter_check_under_limit(&mem->res))
1055            return true;
1056    return false;
1057}
1058
1059static unsigned int get_swappiness(struct mem_cgroup *memcg)
1060{
1061    struct cgroup *cgrp = memcg->css.cgroup;
1062    unsigned int swappiness;
1063
1064    /* root ? */
1065    if (cgrp->parent == NULL)
1066        return vm_swappiness;
1067
1068    spin_lock(&memcg->reclaim_param_lock);
1069    swappiness = memcg->swappiness;
1070    spin_unlock(&memcg->reclaim_param_lock);
1071
1072    return swappiness;
1073}
1074
1075static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
1076{
1077    int *val = data;
1078    (*val)++;
1079    return 0;
1080}
1081
1082/**
1083 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1084 * @memcg: The memory cgroup that went over limit
1085 * @p: Task that is going to be killed
1086 *
1087 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1088 * enabled
1089 */
1090void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1091{
1092    struct cgroup *task_cgrp;
1093    struct cgroup *mem_cgrp;
1094    /*
1095     * Need a buffer in BSS, can't rely on allocations. The code relies
1096     * on the assumption that OOM is serialized for memory controller.
1097     * If this assumption is broken, revisit this code.
1098     */
1099    static char memcg_name[PATH_MAX];
1100    int ret;
1101
1102    if (!memcg || !p)
1103        return;
1104
1105
1106    rcu_read_lock();
1107
1108    mem_cgrp = memcg->css.cgroup;
1109    task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1110
1111    ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1112    if (ret < 0) {
1113        /*
1114         * Unfortunately, we are unable to convert to a useful name
1115         * But we'll still print out the usage information
1116         */
1117        rcu_read_unlock();
1118        goto done;
1119    }
1120    rcu_read_unlock();
1121
1122    printk(KERN_INFO "Task in %s killed", memcg_name);
1123
1124    rcu_read_lock();
1125    ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1126    if (ret < 0) {
1127        rcu_read_unlock();
1128        goto done;
1129    }
1130    rcu_read_unlock();
1131
1132    /*
1133     * Continues from above, so we don't need an KERN_ level
1134     */
1135    printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1136done:
1137
1138    printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1139        res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1140        res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1141        res_counter_read_u64(&memcg->res, RES_FAILCNT));
1142    printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1143        "failcnt %llu\n",
1144        res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1145        res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1146        res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1147}
1148
1149/*
1150 * This function returns the number of memcg under hierarchy tree. Returns
1151 * 1(self count) if no children.
1152 */
1153static int mem_cgroup_count_children(struct mem_cgroup *mem)
1154{
1155    int num = 0;
1156     mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
1157    return num;
1158}
1159
1160/*
1161 * Visit the first child (need not be the first child as per the ordering
1162 * of the cgroup list, since we track last_scanned_child) of @mem and use
1163 * that to reclaim free pages from.
1164 */
1165static struct mem_cgroup *
1166mem_cgroup_select_victim(struct mem_cgroup *root_mem)
1167{
1168    struct mem_cgroup *ret = NULL;
1169    struct cgroup_subsys_state *css;
1170    int nextid, found;
1171
1172    if (!root_mem->use_hierarchy) {
1173        css_get(&root_mem->css);
1174        ret = root_mem;
1175    }
1176
1177    while (!ret) {
1178        rcu_read_lock();
1179        nextid = root_mem->last_scanned_child + 1;
1180        css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
1181                   &found);
1182        if (css && css_tryget(css))
1183            ret = container_of(css, struct mem_cgroup, css);
1184
1185        rcu_read_unlock();
1186        /* Updates scanning parameter */
1187        spin_lock(&root_mem->reclaim_param_lock);
1188        if (!css) {
1189            /* this means start scan from ID:1 */
1190            root_mem->last_scanned_child = 0;
1191        } else
1192            root_mem->last_scanned_child = found;
1193        spin_unlock(&root_mem->reclaim_param_lock);
1194    }
1195
1196    return ret;
1197}
1198
1199/*
1200 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1201 * we reclaimed from, so that we don't end up penalizing one child extensively
1202 * based on its position in the children list.
1203 *
1204 * root_mem is the original ancestor that we've been reclaim from.
1205 *
1206 * We give up and return to the caller when we visit root_mem twice.
1207 * (other groups can be removed while we're walking....)
1208 *
1209 * If shrink==true, for avoiding to free too much, this returns immedieately.
1210 */
1211static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
1212                        struct zone *zone,
1213                        gfp_t gfp_mask,
1214                        unsigned long reclaim_options)
1215{
1216    struct mem_cgroup *victim;
1217    int ret, total = 0;
1218    int loop = 0;
1219    bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
1220    bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
1221    bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT;
1222    unsigned long excess = mem_cgroup_get_excess(root_mem);
1223
1224    /* If memsw_is_minimum==1, swap-out is of-no-use. */
1225    if (root_mem->memsw_is_minimum)
1226        noswap = true;
1227
1228    while (1) {
1229        victim = mem_cgroup_select_victim(root_mem);
1230        if (victim == root_mem) {
1231            loop++;
1232            if (loop >= 1)
1233                drain_all_stock_async();
1234            if (loop >= 2) {
1235                /*
1236                 * If we have not been able to reclaim
1237                 * anything, it might because there are
1238                 * no reclaimable pages under this hierarchy
1239                 */
1240                if (!check_soft || !total) {
1241                    css_put(&victim->css);
1242                    break;
1243                }
1244                /*
1245                 * We want to do more targetted reclaim.
1246                 * excess >> 2 is not to excessive so as to
1247                 * reclaim too much, nor too less that we keep
1248                 * coming back to reclaim from this cgroup
1249                 */
1250                if (total >= (excess >> 2) ||
1251                    (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) {
1252                    css_put(&victim->css);
1253                    break;
1254                }
1255            }
1256        }
1257        if (!mem_cgroup_local_usage(victim)) {
1258            /* this cgroup's local usage == 0 */
1259            css_put(&victim->css);
1260            continue;
1261        }
1262        /* we use swappiness of local cgroup */
1263        if (check_soft)
1264            ret = mem_cgroup_shrink_node_zone(victim, gfp_mask,
1265                noswap, get_swappiness(victim), zone,
1266                zone->zone_pgdat->node_id);
1267        else
1268            ret = try_to_free_mem_cgroup_pages(victim, gfp_mask,
1269                        noswap, get_swappiness(victim));
1270        css_put(&victim->css);
1271        /*
1272         * At shrinking usage, we can't check we should stop here or
1273         * reclaim more. It's depends on callers. last_scanned_child
1274         * will work enough for keeping fairness under tree.
1275         */
1276        if (shrink)
1277            return ret;
1278        total += ret;
1279        if (check_soft) {
1280            if (res_counter_check_under_soft_limit(&root_mem->res))
1281                return total;
1282        } else if (mem_cgroup_check_under_limit(root_mem))
1283            return 1 + total;
1284    }
1285    return total;
1286}
1287
1288static int mem_cgroup_oom_lock_cb(struct mem_cgroup *mem, void *data)
1289{
1290    int *val = (int *)data;
1291    int x;
1292    /*
1293     * Logically, we can stop scanning immediately when we find
1294     * a memcg is already locked. But condidering unlock ops and
1295     * creation/removal of memcg, scan-all is simple operation.
1296     */
1297    x = atomic_inc_return(&mem->oom_lock);
1298    *val = max(x, *val);
1299    return 0;
1300}
1301/*
1302 * Check OOM-Killer is already running under our hierarchy.
1303 * If someone is running, return false.
1304 */
1305static bool mem_cgroup_oom_lock(struct mem_cgroup *mem)
1306{
1307    int lock_count = 0;
1308
1309    mem_cgroup_walk_tree(mem, &lock_count, mem_cgroup_oom_lock_cb);
1310
1311    if (lock_count == 1)
1312        return true;
1313    return false;
1314}
1315
1316static int mem_cgroup_oom_unlock_cb(struct mem_cgroup *mem, void *data)
1317{
1318    /*
1319     * When a new child is created while the hierarchy is under oom,
1320     * mem_cgroup_oom_lock() may not be called. We have to use
1321     * atomic_add_unless() here.
1322     */
1323    atomic_add_unless(&mem->oom_lock, -1, 0);
1324    return 0;
1325}
1326
1327static void mem_cgroup_oom_unlock(struct mem_cgroup *mem)
1328{
1329    mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_unlock_cb);
1330}
1331
1332static DEFINE_MUTEX(memcg_oom_mutex);
1333static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1334
1335struct oom_wait_info {
1336    struct mem_cgroup *mem;
1337    wait_queue_t wait;
1338};
1339
1340static int memcg_oom_wake_function(wait_queue_t *wait,
1341    unsigned mode, int sync, void *arg)
1342{
1343    struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg;
1344    struct oom_wait_info *oom_wait_info;
1345
1346    oom_wait_info = container_of(wait, struct oom_wait_info, wait);
1347
1348    if (oom_wait_info->mem == wake_mem)
1349        goto wakeup;
1350    /* if no hierarchy, no match */
1351    if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy)
1352        return 0;
1353    /*
1354     * Both of oom_wait_info->mem and wake_mem are stable under us.
1355     * Then we can use css_is_ancestor without taking care of RCU.
1356     */
1357    if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) &&
1358        !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css))
1359        return 0;
1360
1361wakeup:
1362    return autoremove_wake_function(wait, mode, sync, arg);
1363}
1364
1365static void memcg_wakeup_oom(struct mem_cgroup *mem)
1366{
1367    /* for filtering, pass "mem" as argument. */
1368    __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem);
1369}
1370
1371static void memcg_oom_recover(struct mem_cgroup *mem)
1372{
1373    if (atomic_read(&mem->oom_lock))
1374        memcg_wakeup_oom(mem);
1375}
1376
1377/*
1378 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1379 */
1380bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask)
1381{
1382    struct oom_wait_info owait;
1383    bool locked, need_to_kill;
1384
1385    owait.mem = mem;
1386    owait.wait.flags = 0;
1387    owait.wait.func = memcg_oom_wake_function;
1388    owait.wait.private = current;
1389    INIT_LIST_HEAD(&owait.wait.task_list);
1390    need_to_kill = true;
1391    /* At first, try to OOM lock hierarchy under mem.*/
1392    mutex_lock(&memcg_oom_mutex);
1393    locked = mem_cgroup_oom_lock(mem);
1394    /*
1395     * Even if signal_pending(), we can't quit charge() loop without
1396     * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1397     * under OOM is always welcomed, use TASK_KILLABLE here.
1398     */
1399    prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
1400    if (!locked || mem->oom_kill_disable)
1401        need_to_kill = false;
1402    if (locked)
1403        mem_cgroup_oom_notify(mem);
1404    mutex_unlock(&memcg_oom_mutex);
1405
1406    if (need_to_kill) {
1407        finish_wait(&memcg_oom_waitq, &owait.wait);
1408        mem_cgroup_out_of_memory(mem, mask);
1409    } else {
1410        schedule();
1411        finish_wait(&memcg_oom_waitq, &owait.wait);
1412    }
1413    mutex_lock(&memcg_oom_mutex);
1414    mem_cgroup_oom_unlock(mem);
1415    memcg_wakeup_oom(mem);
1416    mutex_unlock(&memcg_oom_mutex);
1417
1418    if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1419        return false;
1420    /* Give chance to dying process */
1421    schedule_timeout(1);
1422    return true;
1423}
1424
1425/*
1426 * Currently used to update mapped file statistics, but the routine can be
1427 * generalized to update other statistics as well.
1428 */
1429void mem_cgroup_update_file_mapped(struct page *page, int val)
1430{
1431    struct mem_cgroup *mem;
1432    struct page_cgroup *pc;
1433
1434    pc = lookup_page_cgroup(page);
1435    if (unlikely(!pc))
1436        return;
1437
1438    lock_page_cgroup(pc);
1439    mem = pc->mem_cgroup;
1440    if (!mem || !PageCgroupUsed(pc))
1441        goto done;
1442
1443    /*
1444     * Preemption is already disabled. We can use __this_cpu_xxx
1445     */
1446    if (val > 0) {
1447        __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
1448        SetPageCgroupFileMapped(pc);
1449    } else {
1450        __this_cpu_dec(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
1451        ClearPageCgroupFileMapped(pc);
1452    }
1453
1454done:
1455    unlock_page_cgroup(pc);
1456}
1457
1458/*
1459 * size of first charge trial. "32" comes from vmscan.c's magic value.
1460 * TODO: maybe necessary to use big numbers in big irons.
1461 */
1462#define CHARGE_SIZE (32 * PAGE_SIZE)
1463struct memcg_stock_pcp {
1464    struct mem_cgroup *cached; /* this never be root cgroup */
1465    int charge;
1466    struct work_struct work;
1467};
1468static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
1469static atomic_t memcg_drain_count;
1470
1471/*
1472 * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
1473 * from local stock and true is returned. If the stock is 0 or charges from a
1474 * cgroup which is not current target, returns false. This stock will be
1475 * refilled.
1476 */
1477static bool consume_stock(struct mem_cgroup *mem)
1478{
1479    struct memcg_stock_pcp *stock;
1480    bool ret = true;
1481
1482    stock = &get_cpu_var(memcg_stock);
1483    if (mem == stock->cached && stock->charge)
1484        stock->charge -= PAGE_SIZE;
1485    else /* need to call res_counter_charge */
1486        ret = false;
1487    put_cpu_var(memcg_stock);
1488    return ret;
1489}
1490
1491/*
1492 * Returns stocks cached in percpu to res_counter and reset cached information.
1493 */
1494static void drain_stock(struct memcg_stock_pcp *stock)
1495{
1496    struct mem_cgroup *old = stock->cached;
1497
1498    if (stock->charge) {
1499        res_counter_uncharge(&old->res, stock->charge);
1500        if (do_swap_account)
1501            res_counter_uncharge(&old->memsw, stock->charge);
1502    }
1503    stock->cached = NULL;
1504    stock->charge = 0;
1505}
1506
1507/*
1508 * This must be called under preempt disabled or must be called by
1509 * a thread which is pinned to local cpu.
1510 */
1511static void drain_local_stock(struct work_struct *dummy)
1512{
1513    struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
1514    drain_stock(stock);
1515}
1516
1517/*
1518 * Cache charges(val) which is from res_counter, to local per_cpu area.
1519 * This will be consumed by consume_stock() function, later.
1520 */
1521static void refill_stock(struct mem_cgroup *mem, int val)
1522{
1523    struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
1524
1525    if (stock->cached != mem) { /* reset if necessary */
1526        drain_stock(stock);
1527        stock->cached = mem;
1528    }
1529    stock->charge += val;
1530    put_cpu_var(memcg_stock);
1531}
1532
1533/*
1534 * Tries to drain stocked charges in other cpus. This function is asynchronous
1535 * and just put a work per cpu for draining localy on each cpu. Caller can
1536 * expects some charges will be back to res_counter later but cannot wait for
1537 * it.
1538 */
1539static void drain_all_stock_async(void)
1540{
1541    int cpu;
1542    /* This function is for scheduling "drain" in asynchronous way.
1543     * The result of "drain" is not directly handled by callers. Then,
1544     * if someone is calling drain, we don't have to call drain more.
1545     * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1546     * there is a race. We just do loose check here.
1547     */
1548    if (atomic_read(&memcg_drain_count))
1549        return;
1550    /* Notify other cpus that system-wide "drain" is running */
1551    atomic_inc(&memcg_drain_count);
1552    get_online_cpus();
1553    for_each_online_cpu(cpu) {
1554        struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
1555        schedule_work_on(cpu, &stock->work);
1556    }
1557     put_online_cpus();
1558    atomic_dec(&memcg_drain_count);
1559    /* We don't wait for flush_work */
1560}
1561
1562/* This is a synchronous drain interface. */
1563static void drain_all_stock_sync(void)
1564{
1565    /* called when force_empty is called */
1566    atomic_inc(&memcg_drain_count);
1567    schedule_on_each_cpu(drain_local_stock);
1568    atomic_dec(&memcg_drain_count);
1569}
1570
1571static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb,
1572                    unsigned long action,
1573                    void *hcpu)
1574{
1575    int cpu = (unsigned long)hcpu;
1576    struct memcg_stock_pcp *stock;
1577
1578    if (action != CPU_DEAD)
1579        return NOTIFY_OK;
1580    stock = &per_cpu(memcg_stock, cpu);
1581    drain_stock(stock);
1582    return NOTIFY_OK;
1583}
1584
1585/*
1586 * Unlike exported interface, "oom" parameter is added. if oom==true,
1587 * oom-killer can be invoked.
1588 */
1589static int __mem_cgroup_try_charge(struct mm_struct *mm,
1590            gfp_t gfp_mask, struct mem_cgroup **memcg, bool oom)
1591{
1592    struct mem_cgroup *mem, *mem_over_limit;
1593    int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1594    struct res_counter *fail_res;
1595    int csize = CHARGE_SIZE;
1596
1597    /*
1598     * Unlike gloval-vm's OOM-kill, we're not in memory shortage
1599     * in system level. So, allow to go ahead dying process in addition to
1600     * MEMDIE process.
1601     */
1602    if (unlikely(test_thread_flag(TIF_MEMDIE)
1603             || fatal_signal_pending(current)))
1604        goto bypass;
1605
1606    /*
1607     * We always charge the cgroup the mm_struct belongs to.
1608     * The mm_struct's mem_cgroup changes on task migration if the
1609     * thread group leader migrates. It's possible that mm is not
1610     * set, if so charge the init_mm (happens for pagecache usage).
1611     */
1612    mem = *memcg;
1613    if (likely(!mem)) {
1614        mem = try_get_mem_cgroup_from_mm(mm);
1615        *memcg = mem;
1616    } else {
1617        css_get(&mem->css);
1618    }
1619    if (unlikely(!mem))
1620        return 0;
1621
1622    VM_BUG_ON(css_is_removed(&mem->css));
1623    if (mem_cgroup_is_root(mem))
1624        goto done;
1625
1626    while (1) {
1627        int ret = 0;
1628        unsigned long flags = 0;
1629
1630        if (consume_stock(mem))
1631            goto done;
1632
1633        ret = res_counter_charge(&mem->res, csize, &fail_res);
1634        if (likely(!ret)) {
1635            if (!do_swap_account)
1636                break;
1637            ret = res_counter_charge(&mem->memsw, csize, &fail_res);
1638            if (likely(!ret))
1639                break;
1640            /* mem+swap counter fails */
1641            res_counter_uncharge(&mem->res, csize);
1642            flags |= MEM_CGROUP_RECLAIM_NOSWAP;
1643            mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1644                                    memsw);
1645        } else
1646            /* mem counter fails */
1647            mem_over_limit = mem_cgroup_from_res_counter(fail_res,
1648                                    res);
1649
1650        /* reduce request size and retry */
1651        if (csize > PAGE_SIZE) {
1652            csize = PAGE_SIZE;
1653            continue;
1654        }
1655        if (!(gfp_mask & __GFP_WAIT))
1656            goto nomem;
1657
1658        ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
1659                        gfp_mask, flags);
1660        if (ret)
1661            continue;
1662
1663        /*
1664         * try_to_free_mem_cgroup_pages() might not give us a full
1665         * picture of reclaim. Some pages are reclaimed and might be
1666         * moved to swap cache or just unmapped from the cgroup.
1667         * Check the limit again to see if the reclaim reduced the
1668         * current usage of the cgroup before giving up
1669         *
1670         */
1671        if (mem_cgroup_check_under_limit(mem_over_limit))
1672            continue;
1673
1674        /* try to avoid oom while someone is moving charge */
1675        if (mc.moving_task && current != mc.moving_task) {
1676            struct mem_cgroup *from, *to;
1677            bool do_continue = false;
1678            /*
1679             * There is a small race that "from" or "to" can be
1680             * freed by rmdir, so we use css_tryget().
1681             */
1682            from = mc.from;
1683            to = mc.to;
1684            if (from && css_tryget(&from->css)) {
1685                if (mem_over_limit->use_hierarchy)
1686                    do_continue = css_is_ancestor(
1687                            &from->css,
1688                            &mem_over_limit->css);
1689                else
1690                    do_continue = (from == mem_over_limit);
1691                css_put(&from->css);
1692            }
1693            if (!do_continue && to && css_tryget(&to->css)) {
1694                if (mem_over_limit->use_hierarchy)
1695                    do_continue = css_is_ancestor(
1696                            &to->css,
1697                            &mem_over_limit->css);
1698                else
1699                    do_continue = (to == mem_over_limit);
1700                css_put(&to->css);
1701            }
1702            if (do_continue) {
1703                DEFINE_WAIT(wait);
1704                prepare_to_wait(&mc.waitq, &wait,
1705                            TASK_INTERRUPTIBLE);
1706                /* moving charge context might have finished. */
1707                if (mc.moving_task)
1708                    schedule();
1709                finish_wait(&mc.waitq, &wait);
1710                continue;
1711            }
1712        }
1713
1714        if (!nr_retries--) {
1715            if (!oom)
1716                goto nomem;
1717            if (mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) {
1718                nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1719                continue;
1720            }
1721            /* When we reach here, current task is dying .*/
1722            css_put(&mem->css);
1723            goto bypass;
1724        }
1725    }
1726    if (csize > PAGE_SIZE)
1727        refill_stock(mem, csize - PAGE_SIZE);
1728done:
1729    return 0;
1730nomem:
1731    css_put(&mem->css);
1732    return -ENOMEM;
1733bypass:
1734    *memcg = NULL;
1735    return 0;
1736}
1737
1738/*
1739 * Somemtimes we have to undo a charge we got by try_charge().
1740 * This function is for that and do uncharge, put css's refcnt.
1741 * gotten by try_charge().
1742 */
1743static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem,
1744                            unsigned long count)
1745{
1746    if (!mem_cgroup_is_root(mem)) {
1747        res_counter_uncharge(&mem->res, PAGE_SIZE * count);
1748        if (do_swap_account)
1749            res_counter_uncharge(&mem->memsw, PAGE_SIZE * count);
1750        VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags));
1751        WARN_ON_ONCE(count > INT_MAX);
1752        __css_put(&mem->css, (int)count);
1753    }
1754    /* we don't need css_put for root */
1755}
1756
1757static void mem_cgroup_cancel_charge(struct mem_cgroup *mem)
1758{
1759    __mem_cgroup_cancel_charge(mem, 1);
1760}
1761
1762/*
1763 * A helper function to get mem_cgroup from ID. must be called under
1764 * rcu_read_lock(). The caller must check css_is_removed() or some if
1765 * it's concern. (dropping refcnt from swap can be called against removed
1766 * memcg.)
1767 */
1768static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
1769{
1770    struct cgroup_subsys_state *css;
1771
1772    /* ID 0 is unused ID */
1773    if (!id)
1774        return NULL;
1775    css = css_lookup(&mem_cgroup_subsys, id);
1776    if (!css)
1777        return NULL;
1778    return container_of(css, struct mem_cgroup, css);
1779}
1780
1781struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
1782{
1783    struct mem_cgroup *mem = NULL;
1784    struct page_cgroup *pc;
1785    unsigned short id;
1786    swp_entry_t ent;
1787
1788    VM_BUG_ON(!PageLocked(page));
1789
1790    pc = lookup_page_cgroup(page);
1791    lock_page_cgroup(pc);
1792    if (PageCgroupUsed(pc)) {
1793        mem = pc->mem_cgroup;
1794        if (mem && !css_tryget(&mem->css))
1795            mem = NULL;
1796    } else if (PageSwapCache(page)) {
1797        ent.val = page_private(page);
1798        id = lookup_swap_cgroup(ent);
1799        rcu_read_lock();
1800        mem = mem_cgroup_lookup(id);
1801        if (mem && !css_tryget(&mem->css))
1802            mem = NULL;
1803        rcu_read_unlock();
1804    }
1805    unlock_page_cgroup(pc);
1806    return mem;
1807}
1808
1809/*
1810 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
1811 * USED state. If already USED, uncharge and return.
1812 */
1813
1814static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
1815                     struct page_cgroup *pc,
1816                     enum charge_type ctype)
1817{
1818    /* try_charge() can return NULL to *memcg, taking care of it. */
1819    if (!mem)
1820        return;
1821
1822    lock_page_cgroup(pc);
1823    if (unlikely(PageCgroupUsed(pc))) {
1824        unlock_page_cgroup(pc);
1825        mem_cgroup_cancel_charge(mem);
1826        return;
1827    }
1828
1829    pc->mem_cgroup = mem;
1830    /*
1831     * We access a page_cgroup asynchronously without lock_page_cgroup().
1832     * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
1833     * is accessed after testing USED bit. To make pc->mem_cgroup visible
1834     * before USED bit, we need memory barrier here.
1835     * See mem_cgroup_add_lru_list(), etc.
1836      */
1837    smp_wmb();
1838    switch (ctype) {
1839    case MEM_CGROUP_CHARGE_TYPE_CACHE:
1840    case MEM_CGROUP_CHARGE_TYPE_SHMEM:
1841        SetPageCgroupCache(pc);
1842        SetPageCgroupUsed(pc);
1843        break;
1844    case MEM_CGROUP_CHARGE_TYPE_MAPPED:
1845        ClearPageCgroupCache(pc);
1846        SetPageCgroupUsed(pc);
1847        break;
1848    default:
1849        break;
1850    }
1851
1852    mem_cgroup_charge_statistics(mem, pc, true);
1853
1854    unlock_page_cgroup(pc);
1855    /*
1856     * "charge_statistics" updated event counter. Then, check it.
1857     * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
1858     * if they exceeds softlimit.
1859     */
1860    memcg_check_events(mem, pc->page);
1861}
1862
1863/**
1864 * __mem_cgroup_move_account - move account of the page
1865 * @pc: page_cgroup of the page.
1866 * @from: mem_cgroup which the page is moved from.
1867 * @to: mem_cgroup which the page is moved to. @from != @to.
1868 * @uncharge: whether we should call uncharge and css_put against @from.
1869 *
1870 * The caller must confirm following.
1871 * - page is not on LRU (isolate_page() is useful.)
1872 * - the pc is locked, used, and ->mem_cgroup points to @from.
1873 *
1874 * This function doesn't do "charge" nor css_get to new cgroup. It should be
1875 * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
1876 * true, this function does "uncharge" from old cgroup, but it doesn't if
1877 * @uncharge is false, so a caller should do "uncharge".
1878 */
1879
1880static void __mem_cgroup_move_account(struct page_cgroup *pc,
1881    struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge)
1882{
1883    VM_BUG_ON(from == to);
1884    VM_BUG_ON(PageLRU(pc->page));
1885    VM_BUG_ON(!PageCgroupLocked(pc));
1886    VM_BUG_ON(!PageCgroupUsed(pc));
1887    VM_BUG_ON(pc->mem_cgroup != from);
1888
1889    if (PageCgroupFileMapped(pc)) {
1890        /* Update mapped_file data for mem_cgroup */
1891        preempt_disable();
1892        __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
1893        __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
1894        preempt_enable();
1895    }
1896    mem_cgroup_charge_statistics(from, pc, false);
1897    if (uncharge)
1898        /* This is not "cancel", but cancel_charge does all we need. */
1899        mem_cgroup_cancel_charge(from);
1900
1901    /* caller should have done css_get */
1902    pc->mem_cgroup = to;
1903    mem_cgroup_charge_statistics(to, pc, true);
1904    /*
1905     * We charges against "to" which may not have any tasks. Then, "to"
1906     * can be under rmdir(). But in current implementation, caller of
1907     * this function is just force_empty() and move charge, so it's
1908     * garanteed that "to" is never removed. So, we don't check rmdir
1909     * status here.
1910     */
1911}
1912
1913/*
1914 * check whether the @pc is valid for moving account and call
1915 * __mem_cgroup_move_account()
1916 */
1917static int mem_cgroup_move_account(struct page_cgroup *pc,
1918        struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge)
1919{
1920    int ret = -EINVAL;
1921    lock_page_cgroup(pc);
1922    if (PageCgroupUsed(pc) && pc->mem_cgroup == from) {
1923        __mem_cgroup_move_account(pc, from, to, uncharge);
1924        ret = 0;
1925    }
1926    unlock_page_cgroup(pc);
1927    /*
1928     * check events
1929     */
1930    memcg_check_events(to, pc->page);
1931    memcg_check_events(from, pc->page);
1932    return ret;
1933}
1934
1935/*
1936 * move charges to its parent.
1937 */
1938
1939static int mem_cgroup_move_parent(struct page_cgroup *pc,
1940                  struct mem_cgroup *child,
1941                  gfp_t gfp_mask)
1942{
1943    struct page *page = pc->page;
1944    struct cgroup *cg = child->css.cgroup;
1945    struct cgroup *pcg = cg->parent;
1946    struct mem_cgroup *parent;
1947    int ret;
1948
1949    /* Is ROOT ? */
1950    if (!pcg)
1951        return -EINVAL;
1952
1953    ret = -EBUSY;
1954    if (!get_page_unless_zero(page))
1955        goto out;
1956    if (isolate_lru_page(page))
1957        goto put;
1958
1959    parent = mem_cgroup_from_cont(pcg);
1960    ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
1961    if (ret || !parent)
1962        goto put_back;
1963
1964    ret = mem_cgroup_move_account(pc, child, parent, true);
1965    if (ret)
1966        mem_cgroup_cancel_charge(parent);
1967put_back:
1968    putback_lru_page(page);
1969put:
1970    put_page(page);
1971out:
1972    return ret;
1973}
1974
1975/*
1976 * Charge the memory controller for page usage.
1977 * Return
1978 * 0 if the charge was successful
1979 * < 0 if the cgroup is over its limit
1980 */
1981static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
1982                gfp_t gfp_mask, enum charge_type ctype,
1983                struct mem_cgroup *memcg)
1984{
1985    struct mem_cgroup *mem;
1986    struct page_cgroup *pc;
1987    int ret;
1988
1989    pc = lookup_page_cgroup(page);
1990    /* can happen at boot */
1991    if (unlikely(!pc))
1992        return 0;
1993    prefetchw(pc);
1994
1995    mem = memcg;
1996    ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
1997    if (ret || !mem)
1998        return ret;
1999
2000    __mem_cgroup_commit_charge(mem, pc, ctype);
2001    return 0;
2002}
2003
2004int mem_cgroup_newpage_charge(struct page *page,
2005                  struct mm_struct *mm, gfp_t gfp_mask)
2006{
2007    if (mem_cgroup_disabled())
2008        return 0;
2009    if (PageCompound(page))
2010        return 0;
2011    /*
2012     * If already mapped, we don't have to account.
2013     * If page cache, page->mapping has address_space.
2014     * But page->mapping may have out-of-use anon_vma pointer,
2015     * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2016     * is NULL.
2017       */
2018    if (page_mapped(page) || (page->mapping && !PageAnon(page)))
2019        return 0;
2020    if (unlikely(!mm))
2021        mm = &init_mm;
2022    return mem_cgroup_charge_common(page, mm, gfp_mask,
2023                MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
2024}
2025
2026static void
2027__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2028                    enum charge_type ctype);
2029
2030int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2031                gfp_t gfp_mask)
2032{
2033    struct mem_cgroup *mem = NULL;
2034    int ret;
2035
2036    if (mem_cgroup_disabled())
2037        return 0;
2038    if (PageCompound(page))
2039        return 0;
2040    /*
2041     * Corner case handling. This is called from add_to_page_cache()
2042     * in usual. But some FS (shmem) precharges this page before calling it
2043     * and call add_to_page_cache() with GFP_NOWAIT.
2044     *
2045     * For GFP_NOWAIT case, the page may be pre-charged before calling
2046     * add_to_page_cache(). (See shmem.c) check it here and avoid to call
2047     * charge twice. (It works but has to pay a bit larger cost.)
2048     * And when the page is SwapCache, it should take swap information
2049     * into account. This is under lock_page() now.
2050     */
2051    if (!(gfp_mask & __GFP_WAIT)) {
2052        struct page_cgroup *pc;
2053
2054
2055        pc = lookup_page_cgroup(page);
2056        if (!pc)
2057            return 0;
2058        lock_page_cgroup(pc);
2059        if (PageCgroupUsed(pc)) {
2060            unlock_page_cgroup(pc);
2061            return 0;
2062        }
2063        unlock_page_cgroup(pc);
2064    }
2065
2066    if (unlikely(!mm && !mem))
2067        mm = &init_mm;
2068
2069    if (page_is_file_cache(page))
2070        return mem_cgroup_charge_common(page, mm, gfp_mask,
2071                MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
2072
2073    /* shmem */
2074    if (PageSwapCache(page)) {
2075        ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2076        if (!ret)
2077            __mem_cgroup_commit_charge_swapin(page, mem,
2078                    MEM_CGROUP_CHARGE_TYPE_SHMEM);
2079    } else
2080        ret = mem_cgroup_charge_common(page, mm, gfp_mask,
2081                    MEM_CGROUP_CHARGE_TYPE_SHMEM, mem);
2082
2083    return ret;
2084}
2085
2086/*
2087 * While swap-in, try_charge -> commit or cancel, the page is locked.
2088 * And when try_charge() successfully returns, one refcnt to memcg without
2089 * struct page_cgroup is acquired. This refcnt will be consumed by
2090 * "commit()" or removed by "cancel()"
2091 */
2092int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2093                 struct page *page,
2094                 gfp_t mask, struct mem_cgroup **ptr)
2095{
2096    struct mem_cgroup *mem;
2097    int ret;
2098
2099    if (mem_cgroup_disabled())
2100        return 0;
2101
2102    if (!do_swap_account)
2103        goto charge_cur_mm;
2104    /*
2105     * A racing thread's fault, or swapoff, may have already updated
2106     * the pte, and even removed page from swap cache: in those cases
2107     * do_swap_page()'s pte_same() test will fail; but there's also a
2108     * KSM case which does need to charge the page.
2109     */
2110    if (!PageSwapCache(page))
2111        goto charge_cur_mm;
2112    mem = try_get_mem_cgroup_from_page(page);
2113    if (!mem)
2114        goto charge_cur_mm;
2115    *ptr = mem;
2116    ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
2117    /* drop extra refcnt from tryget */
2118    css_put(&mem->css);
2119    return ret;
2120charge_cur_mm:
2121    if (unlikely(!mm))
2122        mm = &init_mm;
2123    return __mem_cgroup_try_charge(mm, mask, ptr, true);
2124}
2125
2126static void
2127__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
2128                    enum charge_type ctype)
2129{
2130    struct page_cgroup *pc;
2131
2132    if (mem_cgroup_disabled())
2133        return;
2134    if (!ptr)
2135        return;
2136    cgroup_exclude_rmdir(&ptr->css);
2137    pc = lookup_page_cgroup(page);
2138    mem_cgroup_lru_del_before_commit_swapcache(page);
2139    __mem_cgroup_commit_charge(ptr, pc, ctype);
2140    mem_cgroup_lru_add_after_commit_swapcache(page);
2141    /*
2142     * Now swap is on-memory. This means this page may be
2143     * counted both as mem and swap....double count.
2144     * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2145     * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2146     * may call delete_from_swap_cache() before reach here.
2147     */
2148    if (do_swap_account && PageSwapCache(page)) {
2149        swp_entry_t ent = {.val = page_private(page)};
2150        unsigned short id;
2151        struct mem_cgroup *memcg;
2152
2153        id = swap_cgroup_record(ent, 0);
2154        rcu_read_lock();
2155        memcg = mem_cgroup_lookup(id);
2156        if (memcg) {
2157            /*
2158             * This recorded memcg can be obsolete one. So, avoid
2159             * calling css_tryget
2160             */
2161            if (!mem_cgroup_is_root(memcg))
2162                res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
2163            mem_cgroup_swap_statistics(memcg, false);
2164            mem_cgroup_put(memcg);
2165        }
2166        rcu_read_unlock();
2167    }
2168    /*
2169     * At swapin, we may charge account against cgroup which has no tasks.
2170     * So, rmdir()->pre_destroy() can be called while we do this charge.
2171     * In that case, we need to call pre_destroy() again. check it here.
2172     */
2173    cgroup_release_and_wakeup_rmdir(&ptr->css);
2174}
2175
2176void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
2177{
2178    __mem_cgroup_commit_charge_swapin(page, ptr,
2179                    MEM_CGROUP_CHARGE_TYPE_MAPPED);
2180}
2181
2182void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
2183{
2184    if (mem_cgroup_disabled())
2185        return;
2186    if (!mem)
2187        return;
2188    mem_cgroup_cancel_charge(mem);
2189}
2190
2191static void
2192__do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype)
2193{
2194    struct memcg_batch_info *batch = NULL;
2195    bool uncharge_memsw = true;
2196    /* If swapout, usage of swap doesn't decrease */
2197    if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2198        uncharge_memsw = false;
2199
2200    batch = &current->memcg_batch;
2201    /*
2202     * In usual, we do css_get() when we remember memcg pointer.
2203     * But in this case, we keep res->usage until end of a series of
2204     * uncharges. Then, it's ok to ignore memcg's refcnt.
2205     */
2206    if (!batch->memcg)
2207        batch->memcg = mem;
2208    /*
2209     * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2210     * In those cases, all pages freed continously can be expected to be in
2211     * the same cgroup and we have chance to coalesce uncharges.
2212     * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2213     * because we want to do uncharge as soon as possible.
2214     */
2215
2216    if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2217        goto direct_uncharge;
2218
2219    /*
2220     * In typical case, batch->memcg == mem. This means we can
2221     * merge a series of uncharges to an uncharge of res_counter.
2222     * If not, we uncharge res_counter ony by one.
2223     */
2224    if (batch->memcg != mem)
2225        goto direct_uncharge;
2226    /* remember freed charge and uncharge it later */
2227    batch->bytes += PAGE_SIZE;
2228    if (uncharge_memsw)
2229        batch->memsw_bytes += PAGE_SIZE;
2230    return;
2231direct_uncharge:
2232    res_counter_uncharge(&mem->res, PAGE_SIZE);
2233    if (uncharge_memsw)
2234        res_counter_uncharge(&mem->memsw, PAGE_SIZE);
2235    if (unlikely(batch->memcg != mem))
2236        memcg_oom_recover(mem);
2237    return;
2238}
2239
2240/*
2241 * uncharge if !page_mapped(page)
2242 */
2243static struct mem_cgroup *
2244__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
2245{
2246    struct page_cgroup *pc;
2247    struct mem_cgroup *mem = NULL;
2248    struct mem_cgroup_per_zone *mz;
2249
2250    if (mem_cgroup_disabled())
2251        return NULL;
2252
2253    if (PageSwapCache(page))
2254        return NULL;
2255
2256    /*
2257     * Check if our page_cgroup is valid
2258     */
2259    pc = lookup_page_cgroup(page);
2260    if (unlikely(!pc || !PageCgroupUsed(pc)))
2261        return NULL;
2262
2263    lock_page_cgroup(pc);
2264
2265    mem = pc->mem_cgroup;
2266
2267    if (!PageCgroupUsed(pc))
2268        goto unlock_out;
2269
2270    switch (ctype) {
2271    case MEM_CGROUP_CHARGE_TYPE_MAPPED:
2272    case MEM_CGROUP_CHARGE_TYPE_DROP:
2273        /* See mem_cgroup_prepare_migration() */
2274        if (page_mapped(page) || PageCgroupMigration(pc))
2275            goto unlock_out;
2276        break;
2277    case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
2278        if (!PageAnon(page)) { /* Shared memory */
2279            if (page->mapping && !page_is_file_cache(page))
2280                goto unlock_out;
2281        } else if (page_mapped(page)) /* Anon */
2282                goto unlock_out;
2283        break;
2284    default:
2285        break;
2286    }
2287
2288    if (!mem_cgroup_is_root(mem))
2289        __do_uncharge(mem, ctype);
2290    if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2291        mem_cgroup_swap_statistics(mem, true);
2292    mem_cgroup_charge_statistics(mem, pc, false);
2293
2294    ClearPageCgroupUsed(pc);
2295    /*
2296     * pc->mem_cgroup is not cleared here. It will be accessed when it's
2297     * freed from LRU. This is safe because uncharged page is expected not
2298     * to be reused (freed soon). Exception is SwapCache, it's handled by
2299     * special functions.
2300     */
2301
2302    mz = page_cgroup_zoneinfo(pc);
2303    unlock_page_cgroup(pc);
2304
2305    memcg_check_events(mem, page);
2306    /* at swapout, this memcg will be accessed to record to swap */
2307    if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2308        css_put(&mem->css);
2309
2310    return mem;
2311
2312unlock_out:
2313    unlock_page_cgroup(pc);
2314    return NULL;
2315}
2316
2317void mem_cgroup_uncharge_page(struct page *page)
2318{
2319    /* early check. */
2320    if (page_mapped(page))
2321        return;
2322    if (page->mapping && !PageAnon(page))
2323        return;
2324    __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
2325}
2326
2327void mem_cgroup_uncharge_cache_page(struct page *page)
2328{
2329    VM_BUG_ON(page_mapped(page));
2330    VM_BUG_ON(page->mapping);
2331    __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
2332}
2333
2334/*
2335 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2336 * In that cases, pages are freed continuously and we can expect pages
2337 * are in the same memcg. All these calls itself limits the number of
2338 * pages freed at once, then uncharge_start/end() is called properly.
2339 * This may be called prural(2) times in a context,
2340 */
2341
2342void mem_cgroup_uncharge_start(void)
2343{
2344    current->memcg_batch.do_batch++;
2345    /* We can do nest. */
2346    if (current->memcg_batch.do_batch == 1) {
2347        current->memcg_batch.memcg = NULL;
2348        current->memcg_batch.bytes = 0;
2349        current->memcg_batch.memsw_bytes = 0;
2350    }
2351}
2352
2353void mem_cgroup_uncharge_end(void)
2354{
2355    struct memcg_batch_info *batch = &current->memcg_batch;
2356
2357    if (!batch->do_batch)
2358        return;
2359
2360    batch->do_batch--;
2361    if (batch->do_batch) /* If stacked, do nothing. */
2362        return;
2363
2364    if (!batch->memcg)
2365        return;
2366    /*
2367     * This "batch->memcg" is valid without any css_get/put etc...
2368     * bacause we hide charges behind us.
2369     */
2370    if (batch->bytes)
2371        res_counter_uncharge(&batch->memcg->res, batch->bytes);
2372    if (batch->memsw_bytes)
2373        res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes);
2374    memcg_oom_recover(batch->memcg);
2375    /* forget this pointer (for sanity check) */
2376    batch->memcg = NULL;
2377}
2378
2379#ifdef CONFIG_SWAP
2380/*
2381 * called after __delete_from_swap_cache() and drop "page" account.
2382 * memcg information is recorded to swap_cgroup of "ent"
2383 */
2384void
2385mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
2386{
2387    struct mem_cgroup *memcg;
2388    int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
2389
2390    if (!swapout) /* this was a swap cache but the swap is unused ! */
2391        ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
2392
2393    memcg = __mem_cgroup_uncharge_common(page, ctype);
2394
2395    /* record memcg information */
2396    if (do_swap_account && swapout && memcg) {
2397        swap_cgroup_record(ent, css_id(&memcg->css));
2398        mem_cgroup_get(memcg);
2399    }
2400    if (swapout && memcg)
2401        css_put(&memcg->css);
2402}
2403#endif
2404
2405#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2406/*
2407 * called from swap_entry_free(). remove record in swap_cgroup and
2408 * uncharge "memsw" account.
2409 */
2410void mem_cgroup_uncharge_swap(swp_entry_t ent)
2411{
2412    struct mem_cgroup *memcg;
2413    unsigned short id;
2414
2415    if (!do_swap_account)
2416        return;
2417
2418    id = swap_cgroup_record(ent, 0);
2419    rcu_read_lock();
2420    memcg = mem_cgroup_lookup(id);
2421    if (memcg) {
2422        /*
2423         * We uncharge this because swap is freed.
2424         * This memcg can be obsolete one. We avoid calling css_tryget
2425         */
2426        if (!mem_cgroup_is_root(memcg))
2427            res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
2428        mem_cgroup_swap_statistics(memcg, false);
2429        mem_cgroup_put(memcg);
2430    }
2431    rcu_read_unlock();
2432}
2433
2434/**
2435 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2436 * @entry: swap entry to be moved
2437 * @from: mem_cgroup which the entry is moved from
2438 * @to: mem_cgroup which the entry is moved to
2439 * @need_fixup: whether we should fixup res_counters and refcounts.
2440 *
2441 * It succeeds only when the swap_cgroup's record for this entry is the same
2442 * as the mem_cgroup's id of @from.
2443 *
2444 * Returns 0 on success, -EINVAL on failure.
2445 *
2446 * The caller must have charged to @to, IOW, called res_counter_charge() about
2447 * both res and memsw, and called css_get().
2448 */
2449static int mem_cgroup_move_swap_account(swp_entry_t entry,
2450        struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
2451{
2452    unsigned short old_id, new_id;
2453
2454    old_id = css_id(&from->css);
2455    new_id = css_id(&to->css);
2456
2457    if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
2458        mem_cgroup_swap_statistics(from, false);
2459        mem_cgroup_swap_statistics(to, true);
2460        /*
2461         * This function is only called from task migration context now.
2462         * It postpones res_counter and refcount handling till the end
2463         * of task migration(mem_cgroup_clear_mc()) for performance
2464         * improvement. But we cannot postpone mem_cgroup_get(to)
2465         * because if the process that has been moved to @to does
2466         * swap-in, the refcount of @to might be decreased to 0.
2467         */
2468        mem_cgroup_get(to);
2469        if (need_fixup) {
2470            if (!mem_cgroup_is_root(from))
2471                res_counter_uncharge(&from->memsw, PAGE_SIZE);
2472            mem_cgroup_put(from);
2473            /*
2474             * we charged both to->res and to->memsw, so we should
2475             * uncharge to->res.
2476             */
2477            if (!mem_cgroup_is_root(to))
2478                res_counter_uncharge(&to->res, PAGE_SIZE);
2479            css_put(&to->css);
2480        }
2481        return 0;
2482    }
2483    return -EINVAL;
2484}
2485#else
2486static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
2487        struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
2488{
2489    return -EINVAL;
2490}
2491#endif
2492
2493/*
2494 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2495 * page belongs to.
2496 */
2497int mem_cgroup_prepare_migration(struct page *page,
2498    struct page *newpage, struct mem_cgroup **ptr)
2499{
2500    struct page_cgroup *pc;
2501    struct mem_cgroup *mem = NULL;
2502    enum charge_type ctype;
2503    int ret = 0;
2504
2505    if (mem_cgroup_disabled())
2506        return 0;
2507
2508    pc = lookup_page_cgroup(page);
2509    lock_page_cgroup(pc);
2510    if (PageCgroupUsed(pc)) {
2511        mem = pc->mem_cgroup;
2512        css_get(&mem->css);
2513        /*
2514         * At migrating an anonymous page, its mapcount goes down
2515         * to 0 and uncharge() will be called. But, even if it's fully
2516         * unmapped, migration may fail and this page has to be
2517         * charged again. We set MIGRATION flag here and delay uncharge
2518         * until end_migration() is called
2519         *
2520         * Corner Case Thinking
2521         * A)
2522         * When the old page was mapped as Anon and it's unmap-and-freed
2523         * while migration was ongoing.
2524         * If unmap finds the old page, uncharge() of it will be delayed
2525         * until end_migration(). If unmap finds a new page, it's
2526         * uncharged when it make mapcount to be 1->0. If unmap code
2527         * finds swap_migration_entry, the new page will not be mapped
2528         * and end_migration() will find it(mapcount==0).
2529         *
2530         * B)
2531         * When the old page was mapped but migraion fails, the kernel
2532         * remaps it. A charge for it is kept by MIGRATION flag even
2533         * if mapcount goes down to 0. We can do remap successfully
2534         * without charging it again.
2535         *
2536         * C)
2537         * The "old" page is under lock_page() until the end of
2538         * migration, so, the old page itself will not be swapped-out.
2539         * If the new page is swapped out before end_migraton, our
2540         * hook to usual swap-out path will catch the event.
2541         */
2542        if (PageAnon(page))
2543            SetPageCgroupMigration(pc);
2544    }
2545    unlock_page_cgroup(pc);
2546    /*
2547     * If the page is not charged at this point,
2548     * we return here.
2549     */
2550    if (!mem)
2551        return 0;
2552
2553    *ptr = mem;
2554    ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, ptr, false);
2555    css_put(&mem->css);/* drop extra refcnt */
2556    if (ret || *ptr == NULL) {
2557        if (PageAnon(page)) {
2558            lock_page_cgroup(pc);
2559            ClearPageCgroupMigration(pc);
2560            unlock_page_cgroup(pc);
2561            /*
2562             * The old page may be fully unmapped while we kept it.
2563             */
2564            mem_cgroup_uncharge_page(page);
2565        }
2566        return -ENOMEM;
2567    }
2568    /*
2569     * We charge new page before it's used/mapped. So, even if unlock_page()
2570     * is called before end_migration, we can catch all events on this new
2571     * page. In the case new page is migrated but not remapped, new page's
2572     * mapcount will be finally 0 and we call uncharge in end_migration().
2573     */
2574    pc = lookup_page_cgroup(newpage);
2575    if (PageAnon(page))
2576        ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
2577    else if (page_is_file_cache(page))
2578        ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
2579    else
2580        ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
2581    __mem_cgroup_commit_charge(mem, pc, ctype);
2582    return ret;
2583}
2584
2585/* remove redundant charge if migration failed*/
2586void mem_cgroup_end_migration(struct mem_cgroup *mem,
2587    struct page *oldpage, struct page *newpage)
2588{
2589    struct page *used, *unused;
2590    struct page_cgroup *pc;
2591
2592    if (!mem)
2593        return;
2594    /* blocks rmdir() */
2595    cgroup_exclude_rmdir(&mem->css);
2596    /* at migration success, oldpage->mapping is NULL. */
2597    if (oldpage->mapping) {
2598        used = oldpage;
2599        unused = newpage;
2600    } else {
2601        used = newpage;
2602        unused = oldpage;
2603    }
2604    /*
2605     * We disallowed uncharge of pages under migration because mapcount
2606     * of the page goes down to zero, temporarly.
2607     * Clear the flag and check the page should be charged.
2608     */
2609    pc = lookup_page_cgroup(oldpage);
2610    lock_page_cgroup(pc);
2611    ClearPageCgroupMigration(pc);
2612    unlock_page_cgroup(pc);
2613
2614    if (unused != oldpage)
2615        pc = lookup_page_cgroup(unused);
2616    __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE);
2617
2618    pc = lookup_page_cgroup(used);
2619    /*
2620     * If a page is a file cache, radix-tree replacement is very atomic
2621     * and we can skip this check. When it was an Anon page, its mapcount
2622     * goes down to 0. But because we added MIGRATION flage, it's not
2623     * uncharged yet. There are several case but page->mapcount check
2624     * and USED bit check in mem_cgroup_uncharge_page() will do enough
2625     * check. (see prepare_charge() also)
2626     */
2627    if (PageAnon(used))
2628        mem_cgroup_uncharge_page(used);
2629    /*
2630     * At migration, we may charge account against cgroup which has no
2631     * tasks.
2632     * So, rmdir()->pre_destroy() can be called while we do this charge.
2633     * In that case, we need to call pre_destroy() again. check it here.
2634     */
2635    cgroup_release_and_wakeup_rmdir(&mem->css);
2636}
2637
2638/*
2639 * A call to try to shrink memory usage on charge failure at shmem's swapin.
2640 * Calling hierarchical_reclaim is not enough because we should update
2641 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
2642 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
2643 * not from the memcg which this page would be charged to.
2644 * try_charge_swapin does all of these works properly.
2645 */
2646int mem_cgroup_shmem_charge_fallback(struct page *page,
2647                struct mm_struct *mm,
2648                gfp_t gfp_mask)
2649{
2650    struct mem_cgroup *mem = NULL;
2651    int ret;
2652
2653    if (mem_cgroup_disabled())
2654        return 0;
2655
2656    ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
2657    if (!ret)
2658        mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
2659
2660    return ret;
2661}
2662
2663static DEFINE_MUTEX(set_limit_mutex);
2664
2665static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
2666                unsigned long long val)
2667{
2668    int retry_count;
2669    u64 memswlimit, memlimit;
2670    int ret = 0;
2671    int children = mem_cgroup_count_children(memcg);
2672    u64 curusage, oldusage;
2673    int enlarge;
2674
2675    /*
2676     * For keeping hierarchical_reclaim simple, how long we should retry
2677     * is depends on callers. We set our retry-count to be function
2678     * of # of children which we should visit in this loop.
2679     */
2680    retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
2681
2682    oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2683
2684    enlarge = 0;
2685    while (retry_count) {
2686        if (signal_pending(current)) {
2687            ret = -EINTR;
2688            break;
2689        }
2690        /*
2691         * Rather than hide all in some function, I do this in
2692         * open coded manner. You see what this really does.
2693         * We have to guarantee mem->res.limit < mem->memsw.limit.
2694         */
2695        mutex_lock(&set_limit_mutex);
2696        memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2697        if (memswlimit < val) {
2698            ret = -EINVAL;
2699            mutex_unlock(&set_limit_mutex);
2700            break;
2701        }
2702
2703        memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2704        if (memlimit < val)
2705            enlarge = 1;
2706
2707        ret = res_counter_set_limit(&memcg->res, val);
2708        if (!ret) {
2709            if (memswlimit == val)
2710                memcg->memsw_is_minimum = true;
2711            else
2712                memcg->memsw_is_minimum = false;
2713        }
2714        mutex_unlock(&set_limit_mutex);
2715
2716        if (!ret)
2717            break;
2718
2719        mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
2720                        MEM_CGROUP_RECLAIM_SHRINK);
2721        curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
2722        /* Usage is reduced ? */
2723          if (curusage >= oldusage)
2724            retry_count--;
2725        else
2726            oldusage = curusage;
2727    }
2728    if (!ret && enlarge)
2729        memcg_oom_recover(memcg);
2730
2731    return ret;
2732}
2733
2734static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2735                    unsigned long long val)
2736{
2737    int retry_count;
2738    u64 memlimit, memswlimit, oldusage, curusage;
2739    int children = mem_cgroup_count_children(memcg);
2740    int ret = -EBUSY;
2741    int enlarge = 0;
2742
2743    /* see mem_cgroup_resize_res_limit */
2744     retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
2745    oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2746    while (retry_count) {
2747        if (signal_pending(current)) {
2748            ret = -EINTR;
2749            break;
2750        }
2751        /*
2752         * Rather than hide all in some function, I do this in
2753         * open coded manner. You see what this really does.
2754         * We have to guarantee mem->res.limit < mem->memsw.limit.
2755         */
2756        mutex_lock(&set_limit_mutex);
2757        memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
2758        if (memlimit > val) {
2759            ret = -EINVAL;
2760            mutex_unlock(&set_limit_mutex);
2761            break;
2762        }
2763        memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
2764        if (memswlimit < val)
2765            enlarge = 1;
2766        ret = res_counter_set_limit(&memcg->memsw, val);
2767        if (!ret) {
2768            if (memlimit == val)
2769                memcg->memsw_is_minimum = true;
2770            else
2771                memcg->memsw_is_minimum = false;
2772        }
2773        mutex_unlock(&set_limit_mutex);
2774
2775        if (!ret)
2776            break;
2777
2778        mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
2779                        MEM_CGROUP_RECLAIM_NOSWAP |
2780                        MEM_CGROUP_RECLAIM_SHRINK);
2781        curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
2782        /* Usage is reduced ? */
2783        if (curusage >= oldusage)
2784            retry_count--;
2785        else
2786            oldusage = curusage;
2787    }
2788    if (!ret && enlarge)
2789        memcg_oom_recover(memcg);
2790    return ret;
2791}
2792
2793unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2794                        gfp_t gfp_mask, int nid,
2795                        int zid)
2796{
2797    unsigned long nr_reclaimed = 0;
2798    struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2799    unsigned long reclaimed;
2800    int loop = 0;
2801    struct mem_cgroup_tree_per_zone *mctz;
2802    unsigned long long excess;
2803
2804    if (order > 0)
2805        return 0;
2806
2807    mctz = soft_limit_tree_node_zone(nid, zid);
2808    /*
2809     * This loop can run a while, specially if mem_cgroup's continuously
2810     * keep exceeding their soft limit and putting the system under
2811     * pressure
2812     */
2813    do {
2814        if (next_mz)
2815            mz = next_mz;
2816        else
2817            mz = mem_cgroup_largest_soft_limit_node(mctz);
2818        if (!mz)
2819            break;
2820
2821        reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone,
2822                        gfp_mask,
2823                        MEM_CGROUP_RECLAIM_SOFT);
2824        nr_reclaimed += reclaimed;
2825        spin_lock(&mctz->lock);
2826
2827        /*
2828         * If we failed to reclaim anything from this memory cgroup
2829         * it is time to move on to the next cgroup
2830         */
2831        next_mz = NULL;
2832        if (!reclaimed) {
2833            do {
2834                /*
2835                 * Loop until we find yet another one.
2836                 *
2837                 * By the time we get the soft_limit lock
2838                 * again, someone might have aded the
2839                 * group back on the RB tree. Iterate to
2840                 * make sure we get a different mem.
2841                 * mem_cgroup_largest_soft_limit_node returns
2842                 * NULL if no other cgroup is present on
2843                 * the tree
2844                 */
2845                next_mz =
2846                __mem_cgroup_largest_soft_limit_node(mctz);
2847                if (next_mz == mz) {
2848                    css_put(&next_mz->mem->css);
2849                    next_mz = NULL;
2850                } else /* next_mz == NULL or other memcg */
2851                    break;
2852            } while (1);
2853        }
2854        __mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
2855        excess = res_counter_soft_limit_excess(&mz->mem->res);
2856        /*
2857         * One school of thought says that we should not add
2858         * back the node to the tree if reclaim returns 0.
2859         * But our reclaim could return 0, simply because due
2860         * to priority we are exposing a smaller subset of
2861         * memory to reclaim from. Consider this as a longer
2862         * term TODO.
2863         */
2864        /* If excess == 0, no tree ops */
2865        __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
2866        spin_unlock(&mctz->lock);
2867        css_put(&mz->mem->css);
2868        loop++;
2869        /*
2870         * Could not reclaim anything and there are no more
2871         * mem cgroups to try or we seem to be looping without
2872         * reclaiming anything.
2873         */
2874        if (!nr_reclaimed &&
2875            (next_mz == NULL ||
2876            loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2877            break;
2878    } while (!nr_reclaimed);
2879    if (next_mz)
2880        css_put(&next_mz->mem->css);
2881    return nr_reclaimed;
2882}
2883
2884/*
2885 * This routine traverse page_cgroup in given list and drop them all.
2886 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
2887 */
2888static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
2889                int node, int zid, enum lru_list lru)
2890{
2891    struct zone *zone;
2892    struct mem_cgroup_per_zone *mz;
2893    struct page_cgroup *pc, *busy;
2894    unsigned long flags, loop;
2895    struct list_head *list;
2896    int ret = 0;
2897
2898    zone = &NODE_DATA(node)->node_zones[zid];
2899    mz = mem_cgroup_zoneinfo(mem, node, zid);
2900    list = &mz->lists[lru];
2901
2902    loop = MEM_CGROUP_ZSTAT(mz, lru);
2903    /* give some margin against EBUSY etc...*/
2904    loop += 256;
2905    busy = NULL;
2906    while (loop--) {
2907        ret = 0;
2908        spin_lock_irqsave(&zone->lru_lock, flags);
2909        if (list_empty(list)) {
2910            spin_unlock_irqrestore(&zone->lru_lock, flags);
2911            break;
2912        }
2913        pc = list_entry(list->prev, struct page_cgroup, lru);
2914        if (busy == pc) {
2915            list_move(&pc->lru, list);
2916            busy = NULL;
2917            spin_unlock_irqrestore(&zone->lru_lock, flags);
2918            continue;
2919        }
2920        spin_unlock_irqrestore(&zone->lru_lock, flags);
2921
2922        ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
2923        if (ret == -ENOMEM)
2924            break;
2925
2926        if (ret == -EBUSY || ret == -EINVAL) {
2927            /* found lock contention or "pc" is obsolete. */
2928            busy = pc;
2929            cond_resched();
2930        } else
2931            busy = NULL;
2932    }
2933
2934    if (!ret && !list_empty(list))
2935        return -EBUSY;
2936    return ret;
2937}
2938
2939/*
2940 * make mem_cgroup's charge to be 0 if there is no task.
2941 * This enables deleting this mem_cgroup.
2942 */
2943static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
2944{
2945    int ret;
2946    int node, zid, shrink;
2947    int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
2948    struct cgroup *cgrp = mem->css.cgroup;
2949
2950    css_get(&mem->css);
2951
2952    shrink = 0;
2953    /* should free all ? */
2954    if (free_all)
2955        goto try_to_free;
2956move_account:
2957    do {
2958        ret = -EBUSY;
2959        if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
2960            goto out;
2961        ret = -EINTR;
2962        if (signal_pending(current))
2963            goto out;
2964        /* This is for making all *used* pages to be on LRU. */
2965        lru_add_drain_all();
2966        drain_all_stock_sync();
2967        ret = 0;
2968        for_each_node_state(node, N_HIGH_MEMORY) {
2969            for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
2970                enum lru_list l;
2971                for_each_lru(l) {
2972                    ret = mem_cgroup_force_empty_list(mem,
2973                            node, zid, l);
2974                    if (ret)
2975                        break;
2976                }
2977            }
2978            if (ret)
2979                break;
2980        }
2981        memcg_oom_recover(mem);
2982        /* it seems parent cgroup doesn't have enough mem */
2983        if (ret == -ENOMEM)
2984            goto try_to_free;
2985        cond_resched();
2986    /* "ret" should also be checked to ensure all lists are empty. */
2987    } while (mem->res.usage > 0 || ret);
2988out:
2989    css_put(&mem->css);
2990    return ret;
2991
2992try_to_free:
2993    /* returns EBUSY if there is a task or if we come here twice. */
2994    if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
2995        ret = -EBUSY;
2996        goto out;
2997    }
2998    /* we call try-to-free pages for make this cgroup empty */
2999    lru_add_drain_all();
3000    /* try to free all pages in this cgroup */
3001    shrink = 1;
3002    while (nr_retries && mem->res.usage > 0) {
3003        int progress;
3004
3005        if (signal_pending(current)) {
3006            ret = -EINTR;
3007            goto out;
3008        }
3009        progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
3010                        false, get_swappiness(mem));
3011        if (!progress) {
3012            nr_retries--;
3013            /* maybe some writeback is necessary */
3014            congestion_wait(BLK_RW_ASYNC, HZ/10);
3015        }
3016
3017    }
3018    lru_add_drain();
3019    /* try move_account...there may be some *locked* pages. */
3020    goto move_account;
3021}
3022
3023int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3024{
3025    return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3026}
3027
3028
3029static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3030{
3031    return mem_cgroup_from_cont(cont)->use_hierarchy;
3032}
3033
3034static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3035                    u64 val)
3036{
3037    int retval = 0;
3038    struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3039    struct cgroup *parent = cont->parent;
3040    struct mem_cgroup *parent_mem = NULL;
3041
3042    if (parent)
3043        parent_mem = mem_cgroup_from_cont(parent);
3044
3045    cgroup_lock();
3046    /*
3047     * If parent's use_hierarchy is set, we can't make any modifications
3048     * in the child subtrees. If it is unset, then the change can
3049     * occur, provided the current cgroup has no children.
3050     *
3051     * For the root cgroup, parent_mem is NULL, we allow value to be
3052     * set if there are no children.
3053     */
3054    if ((!parent_mem || !parent_mem->use_hierarchy) &&
3055                (val == 1 || val == 0)) {
3056        if (list_empty(&cont->children))
3057            mem->use_hierarchy = val;
3058        else
3059            retval = -EBUSY;
3060    } else
3061        retval = -EINVAL;
3062    cgroup_unlock();
3063
3064    return retval;
3065}
3066
3067struct mem_cgroup_idx_data {
3068    s64 val;
3069    enum mem_cgroup_stat_index idx;
3070};
3071
3072static int
3073mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data)
3074{
3075    struct mem_cgroup_idx_data *d = data;
3076    d->val += mem_cgroup_read_stat(mem, d->idx);
3077    return 0;
3078}
3079
3080static void
3081mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem,
3082                enum mem_cgroup_stat_index idx, s64 *val)
3083{
3084    struct mem_cgroup_idx_data d;
3085    d.idx = idx;
3086    d.val = 0;
3087    mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat);
3088    *val = d.val;
3089}
3090
3091static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap)
3092{
3093    u64 idx_val, val;
3094
3095    if (!mem_cgroup_is_root(mem)) {
3096        if (!swap)
3097            return res_counter_read_u64(&mem->res, RES_USAGE);
3098        else
3099            return res_counter_read_u64(&mem->memsw, RES_USAGE);
3100    }
3101
3102    mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE, &idx_val);
3103    val = idx_val;
3104    mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS, &idx_val);
3105    val += idx_val;
3106
3107    if (swap) {
3108        mem_cgroup_get_recursive_idx_stat(mem,
3109                MEM_CGROUP_STAT_SWAPOUT, &idx_val);
3110        val += idx_val;
3111    }
3112
3113    return val << PAGE_SHIFT;
3114}
3115
3116static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
3117{
3118    struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
3119    u64 val;
3120    int type, name;
3121
3122    type = MEMFILE_TYPE(cft->private);
3123    name = MEMFILE_ATTR(cft->private);
3124    switch (type) {
3125    case _MEM:
3126        if (name == RES_USAGE)
3127            val = mem_cgroup_usage(mem, false);
3128        else
3129            val = res_counter_read_u64(&mem->res, name);
3130        break;
3131    case _MEMSWAP:
3132        if (name == RES_USAGE)
3133            val = mem_cgroup_usage(mem, true);
3134        else
3135            val = res_counter_read_u64(&mem->memsw, name);
3136        break;
3137    default:
3138        BUG();
3139        break;
3140    }
3141    return val;
3142}
3143/*
3144 * The user of this function is...
3145 * RES_LIMIT.
3146 */
3147static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3148                const char *buffer)
3149{
3150    struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3151    int type, name;
3152    unsigned long long val;
3153    int ret;
3154
3155    type = MEMFILE_TYPE(cft->private);
3156    name = MEMFILE_ATTR(cft->private);
3157    switch (name) {
3158    case RES_LIMIT:
3159        if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3160            ret = -EINVAL;
3161            break;
3162        }
3163        /* This function does all necessary parse...reuse it */
3164        ret = res_counter_memparse_write_strategy(buffer, &val);
3165        if (ret)
3166            break;
3167        if (type == _MEM)
3168            ret = mem_cgroup_resize_limit(memcg, val);
3169        else
3170            ret = mem_cgroup_resize_memsw_limit(memcg, val);
3171        break;
3172    case RES_SOFT_LIMIT:
3173        ret = res_counter_memparse_write_strategy(buffer, &val);
3174        if (ret)
3175            break;
3176        /*
3177         * For memsw, soft limits are hard to implement in terms
3178         * of semantics, for now, we support soft limits for
3179         * control without swap
3180         */
3181        if (type == _MEM)
3182            ret = res_counter_set_soft_limit(&memcg->res, val);
3183        else
3184            ret = -EINVAL;
3185        break;
3186    default:
3187        ret = -EINVAL; /* should be BUG() ? */
3188        break;
3189    }
3190    return ret;
3191}
3192
3193static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
3194        unsigned long long *mem_limit, unsigned long long *memsw_limit)
3195{
3196    struct cgroup *cgroup;
3197    unsigned long long min_limit, min_memsw_limit, tmp;
3198
3199    min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3200    min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3201    cgroup = memcg->css.cgroup;
3202    if (!memcg->use_hierarchy)
3203        goto out;
3204
3205    while (cgroup->parent) {
3206        cgroup = cgroup->parent;
3207        memcg = mem_cgroup_from_cont(cgroup);
3208        if (!memcg->use_hierarchy)
3209            break;
3210        tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
3211        min_limit = min(min_limit, tmp);
3212        tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3213        min_memsw_limit = min(min_memsw_limit, tmp);
3214    }
3215out:
3216    *mem_limit = min_limit;
3217    *memsw_limit = min_memsw_limit;
3218    return;
3219}
3220
3221static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
3222{
3223    struct mem_cgroup *mem;
3224    int type, name;
3225
3226    mem = mem_cgroup_from_cont(cont);
3227    type = MEMFILE_TYPE(event);
3228    name = MEMFILE_ATTR(event);
3229    switch (name) {
3230    case RES_MAX_USAGE:
3231        if (type == _MEM)
3232            res_counter_reset_max(&mem->res);
3233        else
3234            res_counter_reset_max(&mem->memsw);
3235        break;
3236    case RES_FAILCNT:
3237        if (type == _MEM)
3238            res_counter_reset_failcnt(&mem->res);
3239        else
3240            res_counter_reset_failcnt(&mem->memsw);
3241        break;
3242    }
3243
3244    return 0;
3245}
3246
3247static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
3248                    struct cftype *cft)
3249{
3250    return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
3251}
3252
3253#ifdef CONFIG_MMU
3254static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3255                    struct cftype *cft, u64 val)
3256{
3257    struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
3258
3259    if (val >= (1 << NR_MOVE_TYPE))
3260        return -EINVAL;
3261    /*
3262     * We check this value several times in both in can_attach() and
3263     * attach(), so we need cgroup lock to prevent this value from being
3264     * inconsistent.
3265     */
3266    cgroup_lock();
3267    mem->move_charge_at_immigrate = val;
3268    cgroup_unlock();
3269
3270    return 0;
3271}
3272#else
3273static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
3274                    struct cftype *cft, u64 val)
3275{
3276    return -ENOSYS;
3277}
3278#endif
3279
3280
3281/* For read statistics */
3282enum {
3283    MCS_CACHE,
3284    MCS_RSS,
3285    MCS_FILE_MAPPED,
3286    MCS_PGPGIN,
3287    MCS_PGPGOUT,
3288    MCS_SWAP,
3289    MCS_INACTIVE_ANON,
3290    MCS_ACTIVE_ANON,
3291    MCS_INACTIVE_FILE,
3292    MCS_ACTIVE_FILE,
3293    MCS_UNEVICTABLE,
3294    NR_MCS_STAT,
3295};
3296
3297struct mcs_total_stat {
3298    s64 stat[NR_MCS_STAT];
3299};
3300
3301struct {
3302    char *local_name;
3303    char *total_name;
3304} memcg_stat_strings[NR_MCS_STAT] = {
3305    {"cache", "total_cache"},
3306    {"rss", "total_rss"},
3307    {"mapped_file", "total_mapped_file"},
3308    {"pgpgin", "total_pgpgin"},
3309    {"pgpgout", "total_pgpgout"},
3310    {"swap", "total_swap"},
3311    {"inactive_anon", "total_inactive_anon"},
3312    {"active_anon", "total_active_anon"},
3313    {"inactive_file", "total_inactive_file"},
3314    {"active_file", "total_active_file"},
3315    {"unevictable", "total_unevictable"}
3316};
3317
3318
3319static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
3320{
3321    struct mcs_total_stat *s = data;
3322    s64 val;
3323
3324    /* per cpu stat */
3325    val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
3326    s->stat[MCS_CACHE] += val * PAGE_SIZE;
3327    val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
3328    s->stat[MCS_RSS] += val * PAGE_SIZE;
3329    val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED);
3330    s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
3331    val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT);
3332    s->stat[MCS_PGPGIN] += val;
3333    val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT);
3334    s->stat[MCS_PGPGOUT] += val;
3335    if (do_swap_account) {
3336        val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT);
3337        s->stat[MCS_SWAP] += val * PAGE_SIZE;
3338    }
3339
3340    /* per zone stat */
3341    val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
3342    s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
3343    val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
3344    s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
3345    val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
3346    s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
3347    val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
3348    s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
3349    val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
3350    s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
3351    return 0;
3352}
3353
3354static void
3355mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
3356{
3357    mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
3358}
3359
3360static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
3361                 struct cgroup_map_cb *cb)
3362{
3363    struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
3364    struct mcs_total_stat mystat;
3365    int i;
3366
3367    memset(&mystat, 0, sizeof(mystat));
3368    mem_cgroup_get_local_stat(mem_cont, &mystat);
3369
3370    for (i = 0; i < NR_MCS_STAT; i++) {
3371        if (i == MCS_SWAP && !do_swap_account)
3372            continue;
3373        cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
3374    }
3375
3376    /* Hierarchical information */
3377    {
3378        unsigned long long limit, memsw_limit;
3379        memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
3380        cb->fill(cb, "hierarchical_memory_limit", limit);
3381        if (do_swap_account)
3382            cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
3383    }
3384
3385    memset(&mystat, 0, sizeof(mystat));
3386    mem_cgroup_get_total_stat(mem_cont, &mystat);
3387    for (i = 0; i < NR_MCS_STAT; i++) {
3388        if (i == MCS_SWAP && !do_swap_account)
3389            continue;
3390        cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
3391    }
3392
3393#ifdef CONFIG_DEBUG_VM
3394    cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
3395
3396    {
3397        int nid, zid;
3398        struct mem_cgroup_per_zone *mz;
3399        unsigned long recent_rotated[2] = {0, 0};
3400        unsigned long recent_scanned[2] = {0, 0};
3401
3402        for_each_online_node(nid)
3403            for (zid = 0; zid < MAX_NR_ZONES; zid++) {
3404                mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
3405
3406                recent_rotated[0] +=
3407                    mz->reclaim_stat.recent_rotated[0];
3408                recent_rotated[1] +=
3409                    mz->reclaim_stat.recent_rotated[1];
3410                recent_scanned[0] +=
3411                    mz->reclaim_stat.recent_scanned[0];
3412                recent_scanned[1] +=
3413                    mz->reclaim_stat.recent_scanned[1];
3414            }
3415        cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
3416        cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
3417        cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
3418        cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
3419    }
3420#endif
3421
3422    return 0;
3423}
3424
3425static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
3426{
3427    struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3428
3429    return get_swappiness(memcg);
3430}
3431
3432static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
3433                       u64 val)
3434{
3435    struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3436    struct mem_cgroup *parent;
3437
3438    if (val > 100)
3439        return -EINVAL;
3440
3441    if (cgrp->parent == NULL)
3442        return -EINVAL;
3443
3444    parent = mem_cgroup_from_cont(cgrp->parent);
3445
3446    cgroup_lock();
3447
3448    /* If under hierarchy, only empty-root can set this value */
3449    if ((parent->use_hierarchy) ||
3450        (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
3451        cgroup_unlock();
3452        return -EINVAL;
3453    }
3454
3455    spin_lock(&memcg->reclaim_param_lock);
3456    memcg->swappiness = val;
3457    spin_unlock(&memcg->reclaim_param_lock);
3458
3459    cgroup_unlock();
3460
3461    return 0;
3462}
3463
3464static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3465{
3466    struct mem_cgroup_threshold_ary *t;
3467    u64 usage;
3468    int i;
3469
3470    rcu_read_lock();
3471    if (!swap)
3472        t = rcu_dereference(memcg->thresholds.primary);
3473    else
3474        t = rcu_dereference(memcg->memsw_thresholds.primary);
3475
3476    if (!t)
3477        goto unlock;
3478
3479    usage = mem_cgroup_usage(memcg, swap);
3480
3481    /*
3482     * current_threshold points to threshold just below usage.
3483     * If it's not true, a threshold was crossed after last
3484     * call of __mem_cgroup_threshold().
3485     */
3486    i = t->current_threshold;
3487
3488    /*
3489     * Iterate backward over array of thresholds starting from
3490     * current_threshold and check if a threshold is crossed.
3491     * If none of thresholds below usage is crossed, we read
3492     * only one element of the array here.
3493     */
3494    for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3495        eventfd_signal(t->entries[i].eventfd, 1);
3496
3497    /* i = current_threshold + 1 */
3498    i++;
3499
3500    /*
3501     * Iterate forward over array of thresholds starting from
3502     * current_threshold+1 and check if a threshold is crossed.
3503     * If none of thresholds above usage is crossed, we read
3504     * only one element of the array here.
3505     */
3506    for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3507        eventfd_signal(t->entries[i].eventfd, 1);
3508
3509    /* Update current_threshold */
3510    t->current_threshold = i - 1;
3511unlock:
3512    rcu_read_unlock();
3513}
3514
3515static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3516{
3517    __mem_cgroup_threshold(memcg, false);
3518    if (do_swap_account)
3519        __mem_cgroup_threshold(memcg, true);
3520}
3521
3522static int compare_thresholds(const void *a, const void *b)
3523{
3524    const struct mem_cgroup_threshold *_a = a;
3525    const struct mem_cgroup_threshold *_b = b;
3526
3527    return _a->threshold - _b->threshold;
3528}
3529
3530static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem, void *data)
3531{
3532    struct mem_cgroup_eventfd_list *ev;
3533
3534    list_for_each_entry(ev, &mem->oom_notify, list)
3535        eventfd_signal(ev->eventfd, 1);
3536    return 0;
3537}
3538
3539static void mem_cgroup_oom_notify(struct mem_cgroup *mem)
3540{
3541    mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_notify_cb);
3542}
3543
3544static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
3545    struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
3546{
3547    struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3548    struct mem_cgroup_thresholds *thresholds;
3549    struct mem_cgroup_threshold_ary *new;
3550    int type = MEMFILE_TYPE(cft->private);
3551    u64 threshold, usage;
3552    int i, size, ret;
3553
3554    ret = res_counter_memparse_write_strategy(args, &threshold);
3555    if (ret)
3556        return ret;
3557
3558    mutex_lock(&memcg->thresholds_lock);
3559
3560    if (type == _MEM)
3561        thresholds = &memcg->thresholds;
3562    else if (type == _MEMSWAP)
3563        thresholds = &memcg->memsw_thresholds;
3564    else
3565        BUG();
3566
3567    usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
3568
3569    /* Check if a threshold crossed before adding a new one */
3570    if (thresholds->primary)
3571        __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3572
3573    size = thresholds->primary ? thresholds->primary->size + 1 : 1;
3574
3575    /* Allocate memory for new array of thresholds */
3576    new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
3577            GFP_KERNEL);
3578    if (!new) {
3579        ret = -ENOMEM;
3580        goto unlock;
3581    }
3582    new->size = size;
3583
3584    /* Copy thresholds (if any) to new array */
3585    if (thresholds->primary) {
3586        memcpy(new->entries, thresholds->primary->entries, (size - 1) *
3587                sizeof(struct mem_cgroup_threshold));
3588    }
3589
3590    /* Add new threshold */
3591    new->entries[size - 1].eventfd = eventfd;
3592    new->entries[size - 1].threshold = threshold;
3593
3594    /* Sort thresholds. Registering of new threshold isn't time-critical */
3595    sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
3596            compare_thresholds, NULL);
3597
3598    /* Find current threshold */
3599    new->current_threshold = -1;
3600    for (i = 0; i < size; i++) {
3601        if (new->entries[i].threshold < usage) {
3602            /*
3603             * new->current_threshold will not be used until
3604             * rcu_assign_pointer(), so it's safe to increment
3605             * it here.
3606             */
3607            ++new->current_threshold;
3608        }
3609    }
3610
3611    /* Free old spare buffer and save old primary buffer as spare */
3612    kfree(thresholds->spare);
3613    thresholds->spare = thresholds->primary;
3614
3615    rcu_assign_pointer(thresholds->primary, new);
3616
3617    /* To be sure that nobody uses thresholds */
3618    synchronize_rcu();
3619
3620unlock:
3621    mutex_unlock(&memcg->thresholds_lock);
3622
3623    return ret;
3624}
3625
3626static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
3627    struct cftype *cft, struct eventfd_ctx *eventfd)
3628{
3629    struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3630    struct mem_cgroup_thresholds *thresholds;
3631    struct mem_cgroup_threshold_ary *new;
3632    int type = MEMFILE_TYPE(cft->private);
3633    u64 usage;
3634    int i, j, size;
3635
3636    mutex_lock(&memcg->thresholds_lock);
3637    if (type == _MEM)
3638        thresholds = &memcg->thresholds;
3639    else if (type == _MEMSWAP)
3640        thresholds = &memcg->memsw_thresholds;
3641    else
3642        BUG();
3643
3644    /*
3645     * Something went wrong if we trying to unregister a threshold
3646     * if we don't have thresholds
3647     */
3648    BUG_ON(!thresholds);
3649
3650    usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
3651
3652    /* Check if a threshold crossed before removing */
3653    __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3654
3655    /* Calculate new number of threshold */
3656    size = 0;
3657    for (i = 0; i < thresholds->primary->size; i++) {
3658        if (thresholds->primary->entries[i].eventfd != eventfd)
3659            size++;
3660    }
3661
3662    new = thresholds->spare;
3663
3664    /* Set thresholds array to NULL if we don't have thresholds */
3665    if (!size) {
3666        kfree(new);
3667        new = NULL;
3668        goto swap_buffers;
3669    }
3670
3671    new->size = size;
3672
3673    /* Copy thresholds and find current threshold */
3674    new->current_threshold = -1;
3675    for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3676        if (thresholds->primary->entries[i].eventfd == eventfd)
3677            continue;
3678
3679        new->entries[j] = thresholds->primary->entries[i];
3680        if (new->entries[j].threshold < usage) {
3681            /*
3682             * new->current_threshold will not be used
3683             * until rcu_assign_pointer(), so it's safe to increment
3684             * it here.
3685             */
3686            ++new->current_threshold;
3687        }
3688        j++;
3689    }
3690
3691swap_buffers:
3692    /* Swap primary and spare array */
3693    thresholds->spare = thresholds->primary;
3694    rcu_assign_pointer(thresholds->primary, new);
3695
3696    /* To be sure that nobody uses thresholds */
3697    synchronize_rcu();
3698
3699    mutex_unlock(&memcg->thresholds_lock);
3700}
3701
3702static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
3703    struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
3704{
3705    struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
3706    struct mem_cgroup_eventfd_list *event;
3707    int type = MEMFILE_TYPE(cft->private);
3708
3709    BUG_ON(type != _OOM_TYPE);
3710    event = kmalloc(sizeof(*event), GFP_KERNEL);
3711    if (!event)
3712        return -ENOMEM;
3713
3714    mutex_lock(&memcg_oom_mutex);
3715
3716    event->eventfd = eventfd;
3717    list_add(&event->list, &memcg->oom_notify);
3718
3719    /* already in OOM ? */
3720    if (atomic_read(&memcg->oom_lock))
3721        eventfd_signal(eventfd, 1);
3722    mutex_unlock(&memcg_oom_mutex);
3723
3724    return 0;
3725}
3726
3727static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
3728    struct cftype *cft, struct eventfd_ctx *eventfd)
3729{
3730    struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
3731    struct mem_cgroup_eventfd_list *ev, *tmp;
3732    int type = MEMFILE_TYPE(cft->private);
3733
3734    BUG_ON(type != _OOM_TYPE);
3735
3736    mutex_lock(&memcg_oom_mutex);
3737
3738    list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) {
3739        if (ev->eventfd == eventfd) {
3740            list_del(&ev->list);
3741            kfree(ev);
3742        }
3743    }
3744
3745    mutex_unlock(&memcg_oom_mutex);
3746}
3747
3748static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
3749    struct cftype *cft, struct cgroup_map_cb *cb)
3750{
3751    struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
3752
3753    cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable);
3754
3755    if (atomic_read(&mem->oom_lock))
3756        cb->fill(cb, "under_oom", 1);
3757    else
3758        cb->fill(cb, "under_oom", 0);
3759    return 0;
3760}
3761
3762/*
3763 */
3764static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
3765    struct cftype *cft, u64 val)
3766{
3767    struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
3768    struct mem_cgroup *parent;
3769
3770    /* cannot set to root cgroup and only 0 and 1 are allowed */
3771    if (!cgrp->parent || !((val == 0) || (val == 1)))
3772        return -EINVAL;
3773
3774    parent = mem_cgroup_from_cont(cgrp->parent);
3775
3776    cgroup_lock();
3777    /* oom-kill-disable is a flag for subhierarchy. */
3778    if ((parent->use_hierarchy) ||
3779        (mem->use_hierarchy && !list_empty(&cgrp->children))) {
3780        cgroup_unlock();
3781        return -EINVAL;
3782    }
3783    mem->oom_kill_disable = val;
3784    if (!val)
3785        memcg_oom_recover(mem);
3786    cgroup_unlock();
3787    return 0;
3788}
3789
3790static struct cftype mem_cgroup_files[] = {
3791    {
3792        .name = "usage_in_bytes",
3793        .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
3794        .read_u64 = mem_cgroup_read,
3795        .register_event = mem_cgroup_usage_register_event,
3796        .unregister_event = mem_cgroup_usage_unregister_event,
3797    },
3798    {
3799        .name = "max_usage_in_bytes",
3800        .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
3801        .trigger = mem_cgroup_reset,
3802        .read_u64 = mem_cgroup_read,
3803    },
3804    {
3805        .name = "limit_in_bytes",
3806        .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
3807        .write_string = mem_cgroup_write,
3808        .read_u64 = mem_cgroup_read,
3809    },
3810    {
3811        .name = "soft_limit_in_bytes",
3812        .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
3813        .write_string = mem_cgroup_write,
3814        .read_u64 = mem_cgroup_read,
3815    },
3816    {
3817        .name = "failcnt",
3818        .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
3819        .trigger = mem_cgroup_reset,
3820        .read_u64 = mem_cgroup_read,
3821    },
3822    {
3823        .name = "stat",
3824        .read_map = mem_control_stat_show,
3825    },
3826    {
3827        .name = "force_empty",
3828        .trigger = mem_cgroup_force_empty_write,
3829    },
3830    {
3831        .name = "use_hierarchy",
3832        .write_u64 = mem_cgroup_hierarchy_write,
3833        .read_u64 = mem_cgroup_hierarchy_read,
3834    },
3835    {
3836        .name = "swappiness",
3837        .read_u64 = mem_cgroup_swappiness_read,
3838        .write_u64 = mem_cgroup_swappiness_write,
3839    },
3840    {
3841        .name = "move_charge_at_immigrate",
3842        .read_u64 = mem_cgroup_move_charge_read,
3843        .write_u64 = mem_cgroup_move_charge_write,
3844    },
3845    {
3846        .name = "oom_control",
3847        .read_map = mem_cgroup_oom_control_read,
3848        .write_u64 = mem_cgroup_oom_control_write,
3849        .register_event = mem_cgroup_oom_register_event,
3850        .unregister_event = mem_cgroup_oom_unregister_event,
3851        .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3852    },
3853};
3854
3855#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3856static struct cftype memsw_cgroup_files[] = {
3857    {
3858        .name = "memsw.usage_in_bytes",
3859        .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
3860        .read_u64 = mem_cgroup_read,
3861        .register_event = mem_cgroup_usage_register_event,
3862        .unregister_event = mem_cgroup_usage_unregister_event,
3863    },
3864    {
3865        .name = "memsw.max_usage_in_bytes",
3866        .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
3867        .trigger = mem_cgroup_reset,
3868        .read_u64 = mem_cgroup_read,
3869    },
3870    {
3871        .name = "memsw.limit_in_bytes",
3872        .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
3873        .write_string = mem_cgroup_write,
3874        .read_u64 = mem_cgroup_read,
3875    },
3876    {
3877        .name = "memsw.failcnt",
3878        .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
3879        .trigger = mem_cgroup_reset,
3880        .read_u64 = mem_cgroup_read,
3881    },
3882};
3883
3884static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
3885{
3886    if (!do_swap_account)
3887        return 0;
3888    return cgroup_add_files(cont, ss, memsw_cgroup_files,
3889                ARRAY_SIZE(memsw_cgroup_files));
3890};
3891#else
3892static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
3893{
3894    return 0;
3895}
3896#endif
3897
3898static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
3899{
3900    struct mem_cgroup_per_node *pn;
3901    struct mem_cgroup_per_zone *mz;
3902    enum lru_list l;
3903    int zone, tmp = node;
3904    /*
3905     * This routine is called against possible nodes.
3906     * But it's BUG to call kmalloc() against offline node.
3907     *
3908     * TODO: this routine can waste much memory for nodes which will
3909     * never be onlined. It's better to use memory hotplug callback
3910     * function.
3911     */
3912    if (!node_state(node, N_NORMAL_MEMORY))
3913        tmp = -1;
3914    pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
3915    if (!pn)
3916        return 1;
3917
3918    mem->info.nodeinfo[node] = pn;
3919    memset(pn, 0, sizeof(*pn));
3920
3921    for (zone = 0; zone < MAX_NR_ZONES; zone++) {
3922        mz = &pn->zoneinfo[zone];
3923        for_each_lru(l)
3924            INIT_LIST_HEAD(&mz->lists[l]);
3925        mz->usage_in_excess = 0;
3926        mz->on_tree = false;
3927        mz->mem = mem;
3928    }
3929    return 0;
3930}
3931
3932static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
3933{
3934    kfree(mem->info.nodeinfo[node]);
3935}
3936
3937static struct mem_cgroup *mem_cgroup_alloc(void)
3938{
3939    struct mem_cgroup *mem;
3940    int size = sizeof(struct mem_cgroup);
3941
3942    /* Can be very big if MAX_NUMNODES is very big */
3943    if (size < PAGE_SIZE)
3944        mem = kmalloc(size, GFP_KERNEL);
3945    else
3946        mem = vmalloc(size);
3947
3948    if (!mem)
3949        return NULL;
3950
3951    memset(mem, 0, size);
3952    mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
3953    if (!mem->stat) {
3954        if (size < PAGE_SIZE)
3955            kfree(mem);
3956        else
3957            vfree(mem);
3958        mem = NULL;
3959    }
3960    return mem;
3961}
3962
3963/*
3964 * At destroying mem_cgroup, references from swap_cgroup can remain.
3965 * (scanning all at force_empty is too costly...)
3966 *
3967 * Instead of clearing all references at force_empty, we remember
3968 * the number of reference from swap_cgroup and free mem_cgroup when
3969 * it goes down to 0.
3970 *
3971 * Removal of cgroup itself succeeds regardless of refs from swap.
3972 */
3973
3974static void __mem_cgroup_free(struct mem_cgroup *mem)
3975{
3976    int node;
3977
3978    mem_cgroup_remove_from_trees(mem);
3979    free_css_id(&mem_cgroup_subsys, &mem->css);
3980
3981    for_each_node_state(node, N_POSSIBLE)
3982        free_mem_cgroup_per_zone_info(mem, node);
3983
3984    free_percpu(mem->stat);
3985    if (sizeof(struct mem_cgroup) < PAGE_SIZE)
3986        kfree(mem);
3987    else
3988        vfree(mem);
3989}
3990
3991static void mem_cgroup_get(struct mem_cgroup *mem)
3992{
3993    atomic_inc(&mem->refcnt);
3994}
3995
3996static void __mem_cgroup_put(struct mem_cgroup *mem, int count)
3997{
3998    if (atomic_sub_and_test(count, &mem->refcnt)) {
3999        struct mem_cgroup *parent = parent_mem_cgroup(mem);
4000        __mem_cgroup_free(mem);
4001        if (parent)
4002            mem_cgroup_put(parent);
4003    }
4004}
4005
4006static void mem_cgroup_put(struct mem_cgroup *mem)
4007{
4008    __mem_cgroup_put(mem, 1);
4009}
4010
4011/*
4012 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4013 */
4014static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
4015{
4016    if (!mem->res.parent)
4017        return NULL;
4018    return mem_cgroup_from_res_counter(mem->res.parent, res);
4019}
4020
4021#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4022static void __init enable_swap_cgroup(void)
4023{
4024    if (!mem_cgroup_disabled() && really_do_swap_account)
4025        do_swap_account = 1;
4026}
4027#else
4028static void __init enable_swap_cgroup(void)
4029{
4030}
4031#endif
4032
4033static int mem_cgroup_soft_limit_tree_init(void)
4034{
4035    struct mem_cgroup_tree_per_node *rtpn;
4036    struct mem_cgroup_tree_per_zone *rtpz;
4037    int tmp, node, zone;
4038
4039    for_each_node_state(node, N_POSSIBLE) {
4040        tmp = node;
4041        if (!node_state(node, N_NORMAL_MEMORY))
4042            tmp = -1;
4043        rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4044        if (!rtpn)
4045            return 1;
4046
4047        soft_limit_tree.rb_tree_per_node[node] = rtpn;
4048
4049        for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4050            rtpz = &rtpn->rb_tree_per_zone[zone];
4051            rtpz->rb_root = RB_ROOT;
4052            spin_lock_init(&rtpz->lock);
4053        }
4054    }
4055    return 0;
4056}
4057
4058static struct cgroup_subsys_state * __ref
4059mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
4060{
4061    struct mem_cgroup *mem, *parent;
4062    long error = -ENOMEM;
4063    int node;
4064
4065    mem = mem_cgroup_alloc();
4066    if (!mem)
4067        return ERR_PTR(error);
4068
4069    for_each_node_state(node, N_POSSIBLE)
4070        if (alloc_mem_cgroup_per_zone_info(mem, node))
4071            goto free_out;
4072
4073    /* root ? */
4074    if (cont->parent == NULL) {
4075        int cpu;
4076        enable_swap_cgroup();
4077        parent = NULL;
4078        root_mem_cgroup = mem;
4079        if (mem_cgroup_soft_limit_tree_init())
4080            goto free_out;
4081        for_each_possible_cpu(cpu) {
4082            struct memcg_stock_pcp *stock =
4083                        &per_cpu(memcg_stock, cpu);
4084            INIT_WORK(&stock->work, drain_local_stock);
4085        }
4086        hotcpu_notifier(memcg_stock_cpu_callback, 0);
4087    } else {
4088        parent = mem_cgroup_from_cont(cont->parent);
4089        mem->use_hierarchy = parent->use_hierarchy;
4090        mem->oom_kill_disable = parent->oom_kill_disable;
4091    }
4092
4093    if (parent && parent->use_hierarchy) {
4094        res_counter_init(&mem->res, &parent->res);
4095        res_counter_init(&mem->memsw, &parent->memsw);
4096        /*
4097         * We increment refcnt of the parent to ensure that we can
4098         * safely access it on res_counter_charge/uncharge.
4099         * This refcnt will be decremented when freeing this
4100         * mem_cgroup(see mem_cgroup_put).
4101         */
4102        mem_cgroup_get(parent);
4103    } else {
4104        res_counter_init(&mem->res, NULL);
4105        res_counter_init(&mem->memsw, NULL);
4106    }
4107    mem->last_scanned_child = 0;
4108    spin_lock_init(&mem->reclaim_param_lock);
4109    INIT_LIST_HEAD(&mem->oom_notify);
4110
4111    if (parent)
4112        mem->swappiness = get_swappiness(parent);
4113    atomic_set(&mem->refcnt, 1);
4114    mem->move_charge_at_immigrate = 0;
4115    mutex_init(&mem->thresholds_lock);
4116    return &mem->css;
4117free_out:
4118    __mem_cgroup_free(mem);
4119    root_mem_cgroup = NULL;
4120    return ERR_PTR(error);
4121}
4122
4123static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
4124                    struct cgroup *cont)
4125{
4126    struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
4127
4128    return mem_cgroup_force_empty(mem, false);
4129}
4130
4131static void mem_cgroup_destroy(struct cgroup_subsys *ss,
4132                struct cgroup *cont)
4133{
4134    struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
4135
4136    mem_cgroup_put(mem);
4137}
4138
4139static int mem_cgroup_populate(struct cgroup_subsys *ss,
4140                struct cgroup *cont)
4141{
4142    int ret;
4143
4144    ret = cgroup_add_files(cont, ss, mem_cgroup_files,
4145                ARRAY_SIZE(mem_cgroup_files));
4146
4147    if (!ret)
4148        ret = register_memsw_files(cont, ss);
4149    return ret;
4150}
4151
4152#ifdef CONFIG_MMU
4153/* Handlers for move charge at task migration. */
4154#define PRECHARGE_COUNT_AT_ONCE 256
4155static int mem_cgroup_do_precharge(unsigned long count)
4156{
4157    int ret = 0;
4158    int batch_count = PRECHARGE_COUNT_AT_ONCE;
4159    struct mem_cgroup *mem = mc.to;
4160
4161    if (mem_cgroup_is_root(mem)) {
4162        mc.precharge += count;
4163        /* we don't need css_get for root */
4164        return ret;
4165    }
4166    /* try to charge at once */
4167    if (count > 1) {
4168        struct res_counter *dummy;
4169        /*
4170         * "mem" cannot be under rmdir() because we've already checked
4171         * by cgroup_lock_live_cgroup() that it is not removed and we
4172         * are still under the same cgroup_mutex. So we can postpone
4173         * css_get().
4174         */
4175        if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy))
4176            goto one_by_one;
4177        if (do_swap_account && res_counter_charge(&mem->memsw,
4178                        PAGE_SIZE * count, &dummy)) {
4179            res_counter_uncharge(&mem->res, PAGE_SIZE * count);
4180            goto one_by_one;
4181        }
4182        mc.precharge += count;
4183        VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags));
4184        WARN_ON_ONCE(count > INT_MAX);
4185        __css_get(&mem->css, (int)count);
4186        return ret;
4187    }
4188one_by_one:
4189    /* fall back to one by one charge */
4190    while (count--) {
4191        if (signal_pending(current)) {
4192            ret = -EINTR;
4193            break;
4194        }
4195        if (!batch_count--) {
4196            batch_count = PRECHARGE_COUNT_AT_ONCE;
4197            cond_resched();
4198        }
4199        ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
4200        if (ret || !mem)
4201            /* mem_cgroup_clear_mc() will do uncharge later */
4202            return -ENOMEM;
4203        mc.precharge++;
4204    }
4205    return ret;
4206}
4207
4208/**
4209 * is_target_pte_for_mc - check a pte whether it is valid for move charge
4210 * @vma: the vma the pte to be checked belongs
4211 * @addr: the address corresponding to the pte to be checked
4212 * @ptent: the pte to be checked
4213 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4214 *
4215 * Returns
4216 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4217 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4218 * move charge. if @target is not NULL, the page is stored in target->page
4219 * with extra refcnt got(Callers should handle it).
4220 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4221 * target for charge migration. if @target is not NULL, the entry is stored
4222 * in target->ent.
4223 *
4224 * Called with pte lock held.
4225 */
4226union mc_target {
4227    struct page *page;
4228    swp_entry_t ent;
4229};
4230
4231enum mc_target_type {
4232    MC_TARGET_NONE, /* not used */
4233    MC_TARGET_PAGE,
4234    MC_TARGET_SWAP,
4235};
4236
4237static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4238                        unsigned long addr, pte_t ptent)
4239{
4240    struct page *page = vm_normal_page(vma, addr, ptent);
4241
4242    if (!page || !page_mapped(page))
4243        return NULL;
4244    if (PageAnon(page)) {
4245        /* we don't move shared anon */
4246        if (!move_anon() || page_mapcount(page) > 2)
4247            return NULL;
4248    } else if (!move_file())
4249        /* we ignore mapcount for file pages */
4250        return NULL;
4251    if (!get_page_unless_zero(page))
4252        return NULL;
4253
4254    return page;
4255}
4256
4257static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4258            unsigned long addr, pte_t ptent, swp_entry_t *entry)
4259{
4260    int usage_count;
4261    struct page *page = NULL;
4262    swp_entry_t ent = pte_to_swp_entry(ptent);
4263
4264    if (!move_anon() || non_swap_entry(ent))
4265        return NULL;
4266    usage_count = mem_cgroup_count_swap_user(ent, &page);
4267    if (usage_count > 1) { /* we don't move shared anon */
4268        if (page)
4269            put_page(page);
4270        return NULL;
4271    }
4272    if (do_swap_account)
4273        entry->val = ent.val;
4274
4275    return page;
4276}
4277
4278static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4279            unsigned long addr, pte_t ptent, swp_entry_t *entry)
4280{
4281    struct page *page = NULL;
4282    struct inode *inode;
4283    struct address_space *mapping;
4284    pgoff_t pgoff;
4285
4286    if (!vma->vm_file) /* anonymous vma */
4287        return NULL;
4288    if (!move_file())
4289        return NULL;
4290
4291    inode = vma->vm_file->f_path.dentry->d_inode;
4292    mapping = vma->vm_file->f_mapping;
4293    if (pte_none(ptent))
4294        pgoff = linear_page_index(vma, addr);
4295    else /* pte_file(ptent) is true */
4296        pgoff = pte_to_pgoff(ptent);
4297
4298    /* page is moved even if it's not RSS of this task(page-faulted). */
4299    if (!mapping_cap_swap_backed(mapping)) { /* normal file */
4300        page = find_get_page(mapping, pgoff);
4301    } else { /* shmem/tmpfs file. we should take account of swap too. */
4302        swp_entry_t ent;
4303        mem_cgroup_get_shmem_target(inode, pgoff, &page, &ent);
4304        if (do_swap_account)
4305            entry->val = ent.val;
4306    }
4307
4308    return page;
4309}
4310
4311static int is_target_pte_for_mc(struct vm_area_struct *vma,
4312        unsigned long addr, pte_t ptent, union mc_target *target)
4313{
4314    struct page *page = NULL;
4315    struct page_cgroup *pc;
4316    int ret = 0;
4317    swp_entry_t ent = { .val = 0 };
4318
4319    if (pte_present(ptent))
4320        page = mc_handle_present_pte(vma, addr, ptent);
4321    else if (is_swap_pte(ptent))
4322        page = mc_handle_swap_pte(vma, addr, ptent, &ent);
4323    else if (pte_none(ptent) || pte_file(ptent))
4324        page = mc_handle_file_pte(vma, addr, ptent, &ent);
4325
4326    if (!page && !ent.val)
4327        return 0;
4328    if (page) {
4329        pc = lookup_page_cgroup(page);
4330        /*
4331         * Do only loose check w/o page_cgroup lock.
4332         * mem_cgroup_move_account() checks the pc is valid or not under
4333         * the lock.
4334         */
4335        if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
4336            ret = MC_TARGET_PAGE;
4337            if (target)
4338                target->page = page;
4339        }
4340        if (!ret || !target)
4341            put_page(page);
4342    }
4343    /* There is a swap entry and a page doesn't exist or isn't charged */
4344    if (ent.val && !ret &&
4345            css_id(&mc.from->css) == lookup_swap_cgroup(ent)) {
4346        ret = MC_TARGET_SWAP;
4347        if (target)
4348            target->ent = ent;
4349    }
4350    return ret;
4351}
4352
4353static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4354                    unsigned long addr, unsigned long end,
4355                    struct mm_walk *walk)
4356{
4357    struct vm_area_struct *vma = walk->private;
4358    pte_t *pte;
4359    spinlock_t *ptl;
4360
4361    pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4362    for (; addr != end; pte++, addr += PAGE_SIZE)
4363        if (is_target_pte_for_mc(vma, addr, *pte, NULL))
4364            mc.precharge++; /* increment precharge temporarily */
4365    pte_unmap_unlock(pte - 1, ptl);
4366    cond_resched();
4367
4368    return 0;
4369}
4370
4371static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4372{
4373    unsigned long precharge;
4374    struct vm_area_struct *vma;
4375
4376    down_read(&mm->mmap_sem);
4377    for (vma = mm->mmap; vma; vma = vma->vm_next) {
4378        struct mm_walk mem_cgroup_count_precharge_walk = {
4379            .pmd_entry = mem_cgroup_count_precharge_pte_range,
4380            .mm = mm,
4381            .private = vma,
4382        };
4383        if (is_vm_hugetlb_page(vma))
4384            continue;
4385        walk_page_range(vma->vm_start, vma->vm_end,
4386                    &mem_cgroup_count_precharge_walk);
4387    }
4388    up_read(&mm->mmap_sem);
4389
4390    precharge = mc.precharge;
4391    mc.precharge = 0;
4392
4393    return precharge;
4394}
4395
4396static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4397{
4398    return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm));
4399}
4400
4401static void mem_cgroup_clear_mc(void)
4402{
4403    /* we must uncharge all the leftover precharges from mc.to */
4404    if (mc.precharge) {
4405        __mem_cgroup_cancel_charge(mc.to, mc.precharge);
4406        mc.precharge = 0;
4407        memcg_oom_recover(mc.to);
4408    }
4409    /*
4410     * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4411     * we must uncharge here.
4412     */
4413    if (mc.moved_charge) {
4414        __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
4415        mc.moved_charge = 0;
4416        memcg_oom_recover(mc.from);
4417    }
4418    /* we must fixup refcnts and charges */
4419    if (mc.moved_swap) {
4420        WARN_ON_ONCE(mc.moved_swap > INT_MAX);
4421        /* uncharge swap account from the old cgroup */
4422        if (!mem_cgroup_is_root(mc.from))
4423            res_counter_uncharge(&mc.from->memsw,
4424                        PAGE_SIZE * mc.moved_swap);
4425        __mem_cgroup_put(mc.from, mc.moved_swap);
4426
4427        if (!mem_cgroup_is_root(mc.to)) {
4428            /*
4429             * we charged both to->res and to->memsw, so we should
4430             * uncharge to->res.
4431             */
4432            res_counter_uncharge(&mc.to->res,
4433                        PAGE_SIZE * mc.moved_swap);
4434            VM_BUG_ON(test_bit(CSS_ROOT, &mc.to->css.flags));
4435            __css_put(&mc.to->css, mc.moved_swap);
4436        }
4437        /* we've already done mem_cgroup_get(mc.to) */
4438
4439        mc.moved_swap = 0;
4440    }
4441    mc.from = NULL;
4442    mc.to = NULL;
4443    mc.moving_task = NULL;
4444    wake_up_all(&mc.waitq);
4445}
4446
4447static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
4448                struct cgroup *cgroup,
4449                struct task_struct *p,
4450                bool threadgroup)
4451{
4452    int ret = 0;
4453    struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup);
4454
4455    if (mem->move_charge_at_immigrate) {
4456        struct mm_struct *mm;
4457        struct mem_cgroup *from = mem_cgroup_from_task(p);
4458
4459        VM_BUG_ON(from == mem);
4460
4461        mm = get_task_mm(p);
4462        if (!mm)
4463            return 0;
4464        /* We move charges only when we move a owner of the mm */
4465        if (mm->owner == p) {
4466            VM_BUG_ON(mc.from);
4467            VM_BUG_ON(mc.to);
4468            VM_BUG_ON(mc.precharge);
4469            VM_BUG_ON(mc.moved_charge);
4470            VM_BUG_ON(mc.moved_swap);
4471            VM_BUG_ON(mc.moving_task);
4472            mc.from = from;
4473            mc.to = mem;
4474            mc.precharge = 0;
4475            mc.moved_charge = 0;
4476            mc.moved_swap = 0;
4477            mc.moving_task = current;
4478
4479            ret = mem_cgroup_precharge_mc(mm);
4480            if (ret)
4481                mem_cgroup_clear_mc();
4482        }
4483        mmput(mm);
4484    }
4485    return ret;
4486}
4487
4488static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
4489                struct cgroup *cgroup,
4490                struct task_struct *p,
4491                bool threadgroup)
4492{
4493    mem_cgroup_clear_mc();
4494}
4495
4496static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4497                unsigned long addr, unsigned long end,
4498                struct mm_walk *walk)
4499{
4500    int ret = 0;
4501    struct vm_area_struct *vma = walk->private;
4502    pte_t *pte;
4503    spinlock_t *ptl;
4504
4505retry:
4506    pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4507    for (; addr != end; addr += PAGE_SIZE) {
4508        pte_t ptent = *(pte++);
4509        union mc_target target;
4510        int type;
4511        struct page *page;
4512        struct page_cgroup *pc;
4513        swp_entry_t ent;
4514
4515        if (!mc.precharge)
4516            break;
4517
4518        type = is_target_pte_for_mc(vma, addr, ptent, &target);
4519        switch (type) {
4520        case MC_TARGET_PAGE:
4521            page = target.page;
4522            if (isolate_lru_page(page))
4523                goto put;
4524            pc = lookup_page_cgroup(page);
4525            if (!mem_cgroup_move_account(pc,
4526                        mc.from, mc.to, false)) {
4527                mc.precharge--;
4528                /* we uncharge from mc.from later. */
4529                mc.moved_charge++;
4530            }
4531            putback_lru_page(page);
4532put: /* is_target_pte_for_mc() gets the page */
4533            put_page(page);
4534            break;
4535        case MC_TARGET_SWAP:
4536            ent = target.ent;
4537            if (!mem_cgroup_move_swap_account(ent,
4538                        mc.from, mc.to, false)) {
4539                mc.precharge--;
4540                /* we fixup refcnts and charges later. */
4541                mc.moved_swap++;
4542            }
4543            break;
4544        default:
4545            break;
4546        }
4547    }
4548    pte_unmap_unlock(pte - 1, ptl);
4549    cond_resched();
4550
4551    if (addr != end) {
4552        /*
4553         * We have consumed all precharges we got in can_attach().
4554         * We try charge one by one, but don't do any additional
4555         * charges to mc.to if we have failed in charge once in attach()
4556         * phase.
4557         */
4558        ret = mem_cgroup_do_precharge(1);
4559        if (!ret)
4560            goto retry;
4561    }
4562
4563    return ret;
4564}
4565
4566static void mem_cgroup_move_charge(struct mm_struct *mm)
4567{
4568    struct vm_area_struct *vma;
4569
4570    lru_add_drain_all();
4571    down_read(&mm->mmap_sem);
4572    for (vma = mm->mmap; vma; vma = vma->vm_next) {
4573        int ret;
4574        struct mm_walk mem_cgroup_move_charge_walk = {
4575            .pmd_entry = mem_cgroup_move_charge_pte_range,
4576            .mm = mm,
4577            .private = vma,
4578        };
4579        if (is_vm_hugetlb_page(vma))
4580            continue;
4581        ret = walk_page_range(vma->vm_start, vma->vm_end,
4582                        &mem_cgroup_move_charge_walk);
4583        if (ret)
4584            /*
4585             * means we have consumed all precharges and failed in
4586             * doing additional charge. Just abandon here.
4587             */
4588            break;
4589    }
4590    up_read(&mm->mmap_sem);
4591}
4592
4593static void mem_cgroup_move_task(struct cgroup_subsys *ss,
4594                struct cgroup *cont,
4595                struct cgroup *old_cont,
4596                struct task_struct *p,
4597                bool threadgroup)
4598{
4599    struct mm_struct *mm;
4600
4601    if (!mc.to)
4602        /* no need to move charge */
4603        return;
4604
4605    mm = get_task_mm(p);
4606    if (mm) {
4607        mem_cgroup_move_charge(mm);
4608        mmput(mm);
4609    }
4610    mem_cgroup_clear_mc();
4611}
4612#else /* !CONFIG_MMU */
4613static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
4614                struct cgroup *cgroup,
4615                struct task_struct *p,
4616                bool threadgroup)
4617{
4618    return 0;
4619}
4620static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
4621                struct cgroup *cgroup,
4622                struct task_struct *p,
4623                bool threadgroup)
4624{
4625}
4626static void mem_cgroup_move_task(struct cgroup_subsys *ss,
4627                struct cgroup *cont,
4628                struct cgroup *old_cont,
4629                struct task_struct *p,
4630                bool threadgroup)
4631{
4632}
4633#endif
4634
4635struct cgroup_subsys mem_cgroup_subsys = {
4636    .name = "memory",
4637    .subsys_id = mem_cgroup_subsys_id,
4638    .create = mem_cgroup_create,
4639    .pre_destroy = mem_cgroup_pre_destroy,
4640    .destroy = mem_cgroup_destroy,
4641    .populate = mem_cgroup_populate,
4642    .can_attach = mem_cgroup_can_attach,
4643    .cancel_attach = mem_cgroup_cancel_attach,
4644    .attach = mem_cgroup_move_task,
4645    .early_init = 0,
4646    .use_id = 1,
4647};
4648
4649#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4650
4651static int __init disable_swap_account(char *s)
4652{
4653    really_do_swap_account = 0;
4654    return 1;
4655}
4656__setup("noswapaccount", disable_swap_account);
4657#endif
4658

Archive Download this file



interactive