linux/mm/memcontrol.c
<<
>>
Prefs
   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/export.h>
  37#include <linux/mutex.h>
  38#include <linux/rbtree.h>
  39#include <linux/slab.h>
  40#include <linux/swap.h>
  41#include <linux/swapops.h>
  42#include <linux/spinlock.h>
  43#include <linux/eventfd.h>
  44#include <linux/sort.h>
  45#include <linux/fs.h>
  46#include <linux/seq_file.h>
  47#include <linux/vmalloc.h>
  48#include <linux/mm_inline.h>
  49#include <linux/page_cgroup.h>
  50#include <linux/cpu.h>
  51#include <linux/oom.h>
  52#include "internal.h"
  53#include <net/sock.h>
  54#include <net/tcp_memcontrol.h>
  55
  56#include <asm/uaccess.h>
  57
  58#include <trace/events/vmscan.h>
  59
  60struct cgroup_subsys mem_cgroup_subsys __read_mostly;
  61#define MEM_CGROUP_RECLAIM_RETRIES      5
  62static struct mem_cgroup *root_mem_cgroup __read_mostly;
  63
  64#ifdef CONFIG_MEMCG_SWAP
  65/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
  66int do_swap_account __read_mostly;
  67
  68/* for remember boot option*/
  69#ifdef CONFIG_MEMCG_SWAP_ENABLED
  70static int really_do_swap_account __initdata = 1;
  71#else
  72static int really_do_swap_account __initdata = 0;
  73#endif
  74
  75#else
  76#define do_swap_account         0
  77#endif
  78
  79
  80/*
  81 * Statistics for memory cgroup.
  82 */
  83enum mem_cgroup_stat_index {
  84        /*
  85         * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
  86         */
  87        MEM_CGROUP_STAT_CACHE,     /* # of pages charged as cache */
  88        MEM_CGROUP_STAT_RSS,       /* # of pages charged as anon rss */
  89        MEM_CGROUP_STAT_FILE_MAPPED,  /* # of pages charged as file rss */
  90        MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */
  91        MEM_CGROUP_STAT_NSTATS,
  92};
  93
  94static const char * const mem_cgroup_stat_names[] = {
  95        "cache",
  96        "rss",
  97        "mapped_file",
  98        "swap",
  99};
 100
 101enum mem_cgroup_events_index {
 102        MEM_CGROUP_EVENTS_PGPGIN,       /* # of pages paged in */
 103        MEM_CGROUP_EVENTS_PGPGOUT,      /* # of pages paged out */
 104        MEM_CGROUP_EVENTS_PGFAULT,      /* # of page-faults */
 105        MEM_CGROUP_EVENTS_PGMAJFAULT,   /* # of major page-faults */
 106        MEM_CGROUP_EVENTS_NSTATS,
 107};
 108
 109static const char * const mem_cgroup_events_names[] = {
 110        "pgpgin",
 111        "pgpgout",
 112        "pgfault",
 113        "pgmajfault",
 114};
 115
 116/*
 117 * Per memcg event counter is incremented at every pagein/pageout. With THP,
 118 * it will be incremated by the number of pages. This counter is used for
 119 * for trigger some periodic events. This is straightforward and better
 120 * than using jiffies etc. to handle periodic memcg event.
 121 */
 122enum mem_cgroup_events_target {
 123        MEM_CGROUP_TARGET_THRESH,
 124        MEM_CGROUP_TARGET_SOFTLIMIT,
 125        MEM_CGROUP_TARGET_NUMAINFO,
 126        MEM_CGROUP_NTARGETS,
 127};
 128#define THRESHOLDS_EVENTS_TARGET 128
 129#define SOFTLIMIT_EVENTS_TARGET 1024
 130#define NUMAINFO_EVENTS_TARGET  1024
 131
 132struct mem_cgroup_stat_cpu {
 133        long count[MEM_CGROUP_STAT_NSTATS];
 134        unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
 135        unsigned long nr_page_events;
 136        unsigned long targets[MEM_CGROUP_NTARGETS];
 137};
 138
 139struct mem_cgroup_reclaim_iter {
 140        /* css_id of the last scanned hierarchy member */
 141        int position;
 142        /* scan generation, increased every round-trip */
 143        unsigned int generation;
 144};
 145
 146/*
 147 * per-zone information in memory controller.
 148 */
 149struct mem_cgroup_per_zone {
 150        struct lruvec           lruvec;
 151        unsigned long           lru_size[NR_LRU_LISTS];
 152
 153        struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
 154
 155        struct rb_node          tree_node;      /* RB tree node */
 156        unsigned long long      usage_in_excess;/* Set to the value by which */
 157                                                /* the soft limit is exceeded*/
 158        bool                    on_tree;
 159        struct mem_cgroup       *memcg;         /* Back pointer, we cannot */
 160                                                /* use container_of        */
 161};
 162
 163struct mem_cgroup_per_node {
 164        struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
 165};
 166
 167struct mem_cgroup_lru_info {
 168        struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
 169};
 170
 171/*
 172 * Cgroups above their limits are maintained in a RB-Tree, independent of
 173 * their hierarchy representation
 174 */
 175
 176struct mem_cgroup_tree_per_zone {
 177        struct rb_root rb_root;
 178        spinlock_t lock;
 179};
 180
 181struct mem_cgroup_tree_per_node {
 182        struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
 183};
 184
 185struct mem_cgroup_tree {
 186        struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
 187};
 188
 189static struct mem_cgroup_tree soft_limit_tree __read_mostly;
 190
 191struct mem_cgroup_threshold {
 192        struct eventfd_ctx *eventfd;
 193        u64 threshold;
 194};
 195
 196/* For threshold */
 197struct mem_cgroup_threshold_ary {
 198        /* An array index points to threshold just below or equal to usage. */
 199        int current_threshold;
 200        /* Size of entries[] */
 201        unsigned int size;
 202        /* Array of thresholds */
 203        struct mem_cgroup_threshold entries[0];
 204};
 205
 206struct mem_cgroup_thresholds {
 207        /* Primary thresholds array */
 208        struct mem_cgroup_threshold_ary *primary;
 209        /*
 210         * Spare threshold array.
 211         * This is needed to make mem_cgroup_unregister_event() "never fail".
 212         * It must be able to store at least primary->size - 1 entries.
 213         */
 214        struct mem_cgroup_threshold_ary *spare;
 215};
 216
 217/* for OOM */
 218struct mem_cgroup_eventfd_list {
 219        struct list_head list;
 220        struct eventfd_ctx *eventfd;
 221};
 222
 223static void mem_cgroup_threshold(struct mem_cgroup *memcg);
 224static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
 225
 226/*
 227 * The memory controller data structure. The memory controller controls both
 228 * page cache and RSS per cgroup. We would eventually like to provide
 229 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
 230 * to help the administrator determine what knobs to tune.
 231 *
 232 * TODO: Add a water mark for the memory controller. Reclaim will begin when
 233 * we hit the water mark. May be even add a low water mark, such that
 234 * no reclaim occurs from a cgroup at it's low water mark, this is
 235 * a feature that will be implemented much later in the future.
 236 */
 237struct mem_cgroup {
 238        struct cgroup_subsys_state css;
 239        /*
 240         * the counter to account for memory usage
 241         */
 242        struct res_counter res;
 243
 244        union {
 245                /*
 246                 * the counter to account for mem+swap usage.
 247                 */
 248                struct res_counter memsw;
 249
 250                /*
 251                 * rcu_freeing is used only when freeing struct mem_cgroup,
 252                 * so put it into a union to avoid wasting more memory.
 253                 * It must be disjoint from the css field.  It could be
 254                 * in a union with the res field, but res plays a much
 255                 * larger part in mem_cgroup life than memsw, and might
 256                 * be of interest, even at time of free, when debugging.
 257                 * So share rcu_head with the less interesting memsw.
 258                 */
 259                struct rcu_head rcu_freeing;
 260                /*
 261                 * We also need some space for a worker in deferred freeing.
 262                 * By the time we call it, rcu_freeing is no longer in use.
 263                 */
 264                struct work_struct work_freeing;
 265        };
 266
 267        /*
 268         * Per cgroup active and inactive list, similar to the
 269         * per zone LRU lists.
 270         */
 271        struct mem_cgroup_lru_info info;
 272        int last_scanned_node;
 273#if MAX_NUMNODES > 1
 274        nodemask_t      scan_nodes;
 275        atomic_t        numainfo_events;
 276        atomic_t        numainfo_updating;
 277#endif
 278        /*
 279         * Should the accounting and control be hierarchical, per subtree?
 280         */
 281        bool use_hierarchy;
 282
 283        bool            oom_lock;
 284        atomic_t        under_oom;
 285
 286        atomic_t        refcnt;
 287
 288        int     swappiness;
 289        /* OOM-Killer disable */
 290        int             oom_kill_disable;
 291
 292        /* set when res.limit == memsw.limit */
 293        bool            memsw_is_minimum;
 294
 295        /* protect arrays of thresholds */
 296        struct mutex thresholds_lock;
 297
 298        /* thresholds for memory usage. RCU-protected */
 299        struct mem_cgroup_thresholds thresholds;
 300
 301        /* thresholds for mem+swap usage. RCU-protected */
 302        struct mem_cgroup_thresholds memsw_thresholds;
 303
 304        /* For oom notifier event fd */
 305        struct list_head oom_notify;
 306
 307        /*
 308         * Should we move charges of a task when a task is moved into this
 309         * mem_cgroup ? And what type of charges should we move ?
 310         */
 311        unsigned long   move_charge_at_immigrate;
 312        /*
 313         * set > 0 if pages under this cgroup are moving to other cgroup.
 314         */
 315        atomic_t        moving_account;
 316        /* taken only while moving_account > 0 */
 317        spinlock_t      move_lock;
 318        /*
 319         * percpu counter.
 320         */
 321        struct mem_cgroup_stat_cpu __percpu *stat;
 322        /*
 323         * used when a cpu is offlined or other synchronizations
 324         * See mem_cgroup_read_stat().
 325         */
 326        struct mem_cgroup_stat_cpu nocpu_base;
 327        spinlock_t pcp_counter_lock;
 328
 329#ifdef CONFIG_INET
 330        struct tcp_memcontrol tcp_mem;
 331#endif
 332};
 333
 334/* Stuffs for move charges at task migration. */
 335/*
 336 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
 337 * left-shifted bitmap of these types.
 338 */
 339enum move_type {
 340        MOVE_CHARGE_TYPE_ANON,  /* private anonymous page and swap of it */
 341        MOVE_CHARGE_TYPE_FILE,  /* file page(including tmpfs) and swap of it */
 342        NR_MOVE_TYPE,
 343};
 344
 345/* "mc" and its members are protected by cgroup_mutex */
 346static struct move_charge_struct {
 347        spinlock_t        lock; /* for from, to */
 348        struct mem_cgroup *from;
 349        struct mem_cgroup *to;
 350        unsigned long precharge;
 351        unsigned long moved_charge;
 352        unsigned long moved_swap;
 353        struct task_struct *moving_task;        /* a task moving charges */
 354        wait_queue_head_t waitq;                /* a waitq for other context */
 355} mc = {
 356        .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
 357        .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
 358};
 359
 360static bool move_anon(void)
 361{
 362        return test_bit(MOVE_CHARGE_TYPE_ANON,
 363                                        &mc.to->move_charge_at_immigrate);
 364}
 365
 366static bool move_file(void)
 367{
 368        return test_bit(MOVE_CHARGE_TYPE_FILE,
 369                                        &mc.to->move_charge_at_immigrate);
 370}
 371
 372/*
 373 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
 374 * limit reclaim to prevent infinite loops, if they ever occur.
 375 */
 376#define MEM_CGROUP_MAX_RECLAIM_LOOPS            100
 377#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
 378
 379enum charge_type {
 380        MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
 381        MEM_CGROUP_CHARGE_TYPE_ANON,
 382        MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
 383        MEM_CGROUP_CHARGE_TYPE_DROP,    /* a page was unused swap cache */
 384        NR_CHARGE_TYPE,
 385};
 386
 387/* for encoding cft->private value on file */
 388#define _MEM                    (0)
 389#define _MEMSWAP                (1)
 390#define _OOM_TYPE               (2)
 391#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
 392#define MEMFILE_TYPE(val)       ((val) >> 16 & 0xffff)
 393#define MEMFILE_ATTR(val)       ((val) & 0xffff)
 394/* Used for OOM nofiier */
 395#define OOM_CONTROL             (0)
 396
 397/*
 398 * Reclaim flags for mem_cgroup_hierarchical_reclaim
 399 */
 400#define MEM_CGROUP_RECLAIM_NOSWAP_BIT   0x0
 401#define MEM_CGROUP_RECLAIM_NOSWAP       (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
 402#define MEM_CGROUP_RECLAIM_SHRINK_BIT   0x1
 403#define MEM_CGROUP_RECLAIM_SHRINK       (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
 404
 405static void mem_cgroup_get(struct mem_cgroup *memcg);
 406static void mem_cgroup_put(struct mem_cgroup *memcg);
 407
 408static inline
 409struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
 410{
 411        return container_of(s, struct mem_cgroup, css);
 412}
 413
 414/* Writing them here to avoid exposing memcg's inner layout */
 415#ifdef CONFIG_MEMCG_KMEM
 416#include <net/sock.h>
 417#include <net/ip.h>
 418
 419static bool mem_cgroup_is_root(struct mem_cgroup *memcg);
 420void sock_update_memcg(struct sock *sk)
 421{
 422        if (mem_cgroup_sockets_enabled) {
 423                struct mem_cgroup *memcg;
 424                struct cg_proto *cg_proto;
 425
 426                BUG_ON(!sk->sk_prot->proto_cgroup);
 427
 428                /* Socket cloning can throw us here with sk_cgrp already
 429                 * filled. It won't however, necessarily happen from
 430                 * process context. So the test for root memcg given
 431                 * the current task's memcg won't help us in this case.
 432                 *
 433                 * Respecting the original socket's memcg is a better
 434                 * decision in this case.
 435                 */
 436                if (sk->sk_cgrp) {
 437                        BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
 438                        mem_cgroup_get(sk->sk_cgrp->memcg);
 439                        return;
 440                }
 441
 442                rcu_read_lock();
 443                memcg = mem_cgroup_from_task(current);
 444                cg_proto = sk->sk_prot->proto_cgroup(memcg);
 445                if (!mem_cgroup_is_root(memcg) && memcg_proto_active(cg_proto)) {
 446                        mem_cgroup_get(memcg);
 447                        sk->sk_cgrp = cg_proto;
 448                }
 449                rcu_read_unlock();
 450        }
 451}
 452EXPORT_SYMBOL(sock_update_memcg);
 453
 454void sock_release_memcg(struct sock *sk)
 455{
 456        if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
 457                struct mem_cgroup *memcg;
 458                WARN_ON(!sk->sk_cgrp->memcg);
 459                memcg = sk->sk_cgrp->memcg;
 460                mem_cgroup_put(memcg);
 461        }
 462}
 463
 464#ifdef CONFIG_INET
 465struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
 466{
 467        if (!memcg || mem_cgroup_is_root(memcg))
 468                return NULL;
 469
 470        return &memcg->tcp_mem.cg_proto;
 471}
 472EXPORT_SYMBOL(tcp_proto_cgroup);
 473#endif /* CONFIG_INET */
 474#endif /* CONFIG_MEMCG_KMEM */
 475
 476#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
 477static void disarm_sock_keys(struct mem_cgroup *memcg)
 478{
 479        if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto))
 480                return;
 481        static_key_slow_dec(&memcg_socket_limit_enabled);
 482}
 483#else
 484static void disarm_sock_keys(struct mem_cgroup *memcg)
 485{
 486}
 487#endif
 488
 489static void drain_all_stock_async(struct mem_cgroup *memcg);
 490
 491static struct mem_cgroup_per_zone *
 492mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
 493{
 494        return &memcg->info.nodeinfo[nid]->zoneinfo[zid];
 495}
 496
 497struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
 498{
 499        return &memcg->css;
 500}
 501
 502static struct mem_cgroup_per_zone *
 503page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
 504{
 505        int nid = page_to_nid(page);
 506        int zid = page_zonenum(page);
 507
 508        return mem_cgroup_zoneinfo(memcg, nid, zid);
 509}
 510
 511static struct mem_cgroup_tree_per_zone *
 512soft_limit_tree_node_zone(int nid, int zid)
 513{
 514        return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
 515}
 516
 517static struct mem_cgroup_tree_per_zone *
 518soft_limit_tree_from_page(struct page *page)
 519{
 520        int nid = page_to_nid(page);
 521        int zid = page_zonenum(page);
 522
 523        return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
 524}
 525
 526static void
 527__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
 528                                struct mem_cgroup_per_zone *mz,
 529                                struct mem_cgroup_tree_per_zone *mctz,
 530                                unsigned long long new_usage_in_excess)
 531{
 532        struct rb_node **p = &mctz->rb_root.rb_node;
 533        struct rb_node *parent = NULL;
 534        struct mem_cgroup_per_zone *mz_node;
 535
 536        if (mz->on_tree)
 537                return;
 538
 539        mz->usage_in_excess = new_usage_in_excess;
 540        if (!mz->usage_in_excess)
 541                return;
 542        while (*p) {
 543                parent = *p;
 544                mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
 545                                        tree_node);
 546                if (mz->usage_in_excess < mz_node->usage_in_excess)
 547                        p = &(*p)->rb_left;
 548                /*
 549                 * We can't avoid mem cgroups that are over their soft
 550                 * limit by the same amount
 551                 */
 552                else if (mz->usage_in_excess >= mz_node->usage_in_excess)
 553                        p = &(*p)->rb_right;
 554        }
 555        rb_link_node(&mz->tree_node, parent, p);
 556        rb_insert_color(&mz->tree_node, &mctz->rb_root);
 557        mz->on_tree = true;
 558}
 559
 560static void
 561__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
 562                                struct mem_cgroup_per_zone *mz,
 563                                struct mem_cgroup_tree_per_zone *mctz)
 564{
 565        if (!mz->on_tree)
 566                return;
 567        rb_erase(&mz->tree_node, &mctz->rb_root);
 568        mz->on_tree = false;
 569}
 570
 571static void
 572mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
 573                                struct mem_cgroup_per_zone *mz,
 574                                struct mem_cgroup_tree_per_zone *mctz)
 575{
 576        spin_lock(&mctz->lock);
 577        __mem_cgroup_remove_exceeded(memcg, mz, mctz);
 578        spin_unlock(&mctz->lock);
 579}
 580
 581
 582static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
 583{
 584        unsigned long long excess;
 585        struct mem_cgroup_per_zone *mz;
 586        struct mem_cgroup_tree_per_zone *mctz;
 587        int nid = page_to_nid(page);
 588        int zid = page_zonenum(page);
 589        mctz = soft_limit_tree_from_page(page);
 590
 591        /*
 592         * Necessary to update all ancestors when hierarchy is used.
 593         * because their event counter is not touched.
 594         */
 595        for (; memcg; memcg = parent_mem_cgroup(memcg)) {
 596                mz = mem_cgroup_zoneinfo(memcg, nid, zid);
 597                excess = res_counter_soft_limit_excess(&memcg->res);
 598                /*
 599                 * We have to update the tree if mz is on RB-tree or
 600                 * mem is over its softlimit.
 601                 */
 602                if (excess || mz->on_tree) {
 603                        spin_lock(&mctz->lock);
 604                        /* if on-tree, remove it */
 605                        if (mz->on_tree)
 606                                __mem_cgroup_remove_exceeded(memcg, mz, mctz);
 607                        /*
 608                         * Insert again. mz->usage_in_excess will be updated.
 609                         * If excess is 0, no tree ops.
 610                         */
 611                        __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
 612                        spin_unlock(&mctz->lock);
 613                }
 614        }
 615}
 616
 617static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
 618{
 619        int node, zone;
 620        struct mem_cgroup_per_zone *mz;
 621        struct mem_cgroup_tree_per_zone *mctz;
 622
 623        for_each_node(node) {
 624                for (zone = 0; zone < MAX_NR_ZONES; zone++) {
 625                        mz = mem_cgroup_zoneinfo(memcg, node, zone);
 626                        mctz = soft_limit_tree_node_zone(node, zone);
 627                        mem_cgroup_remove_exceeded(memcg, mz, mctz);
 628                }
 629        }
 630}
 631
 632static struct mem_cgroup_per_zone *
 633__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
 634{
 635        struct rb_node *rightmost = NULL;
 636        struct mem_cgroup_per_zone *mz;
 637
 638retry:
 639        mz = NULL;
 640        rightmost = rb_last(&mctz->rb_root);
 641        if (!rightmost)
 642                goto done;              /* Nothing to reclaim from */
 643
 644        mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
 645        /*
 646         * Remove the node now but someone else can add it back,
 647         * we will to add it back at the end of reclaim to its correct
 648         * position in the tree.
 649         */
 650        __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
 651        if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
 652                !css_tryget(&mz->memcg->css))
 653                goto retry;
 654done:
 655        return mz;
 656}
 657
 658static struct mem_cgroup_per_zone *
 659mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
 660{
 661        struct mem_cgroup_per_zone *mz;
 662
 663        spin_lock(&mctz->lock);
 664        mz = __mem_cgroup_largest_soft_limit_node(mctz);
 665        spin_unlock(&mctz->lock);
 666        return mz;
 667}
 668
 669/*
 670 * Implementation Note: reading percpu statistics for memcg.
 671 *
 672 * Both of vmstat[] and percpu_counter has threshold and do periodic
 673 * synchronization to implement "quick" read. There are trade-off between
 674 * reading cost and precision of value. Then, we may have a chance to implement
 675 * a periodic synchronizion of counter in memcg's counter.
 676 *
 677 * But this _read() function is used for user interface now. The user accounts
 678 * memory usage by memory cgroup and he _always_ requires exact value because
 679 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
 680 * have to visit all online cpus and make sum. So, for now, unnecessary
 681 * synchronization is not implemented. (just implemented for cpu hotplug)
 682 *
 683 * If there are kernel internal actions which can make use of some not-exact
 684 * value, and reading all cpu value can be performance bottleneck in some
 685 * common workload, threashold and synchonization as vmstat[] should be
 686 * implemented.
 687 */
 688static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
 689                                 enum mem_cgroup_stat_index idx)
 690{
 691        long val = 0;
 692        int cpu;
 693
 694        get_online_cpus();
 695        for_each_online_cpu(cpu)
 696                val += per_cpu(memcg->stat->count[idx], cpu);
 697#ifdef CONFIG_HOTPLUG_CPU
 698        spin_lock(&memcg->pcp_counter_lock);
 699        val += memcg->nocpu_base.count[idx];
 700        spin_unlock(&memcg->pcp_counter_lock);
 701#endif
 702        put_online_cpus();
 703        return val;
 704}
 705
 706static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
 707                                         bool charge)
 708{
 709        int val = (charge) ? 1 : -1;
 710        this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
 711}
 712
 713static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
 714                                            enum mem_cgroup_events_index idx)
 715{
 716        unsigned long val = 0;
 717        int cpu;
 718
 719        for_each_online_cpu(cpu)
 720                val += per_cpu(memcg->stat->events[idx], cpu);
 721#ifdef CONFIG_HOTPLUG_CPU
 722        spin_lock(&memcg->pcp_counter_lock);
 723        val += memcg->nocpu_base.events[idx];
 724        spin_unlock(&memcg->pcp_counter_lock);
 725#endif
 726        return val;
 727}
 728
 729static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
 730                                         bool anon, int nr_pages)
 731{
 732        preempt_disable();
 733
 734        /*
 735         * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
 736         * counted as CACHE even if it's on ANON LRU.
 737         */
 738        if (anon)
 739                __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
 740                                nr_pages);
 741        else
 742                __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
 743                                nr_pages);
 744
 745        /* pagein of a big page is an event. So, ignore page size */
 746        if (nr_pages > 0)
 747                __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
 748        else {
 749                __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
 750                nr_pages = -nr_pages; /* for event */
 751        }
 752
 753        __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
 754
 755        preempt_enable();
 756}
 757
 758unsigned long
 759mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
 760{
 761        struct mem_cgroup_per_zone *mz;
 762
 763        mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
 764        return mz->lru_size[lru];
 765}
 766
 767static unsigned long
 768mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
 769                        unsigned int lru_mask)
 770{
 771        struct mem_cgroup_per_zone *mz;
 772        enum lru_list lru;
 773        unsigned long ret = 0;
 774
 775        mz = mem_cgroup_zoneinfo(memcg, nid, zid);
 776
 777        for_each_lru(lru) {
 778                if (BIT(lru) & lru_mask)
 779                        ret += mz->lru_size[lru];
 780        }
 781        return ret;
 782}
 783
 784static unsigned long
 785mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
 786                        int nid, unsigned int lru_mask)
 787{
 788        u64 total = 0;
 789        int zid;
 790
 791        for (zid = 0; zid < MAX_NR_ZONES; zid++)
 792                total += mem_cgroup_zone_nr_lru_pages(memcg,
 793                                                nid, zid, lru_mask);
 794
 795        return total;
 796}
 797
 798static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
 799                        unsigned int lru_mask)
 800{
 801        int nid;
 802        u64 total = 0;
 803
 804        for_each_node_state(nid, N_HIGH_MEMORY)
 805                total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
 806        return total;
 807}
 808
 809static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
 810                                       enum mem_cgroup_events_target target)
 811{
 812        unsigned long val, next;
 813
 814        val = __this_cpu_read(memcg->stat->nr_page_events);
 815        next = __this_cpu_read(memcg->stat->targets[target]);
 816        /* from time_after() in jiffies.h */
 817        if ((long)next - (long)val < 0) {
 818                switch (target) {
 819                case MEM_CGROUP_TARGET_THRESH:
 820                        next = val + THRESHOLDS_EVENTS_TARGET;
 821                        break;
 822                case MEM_CGROUP_TARGET_SOFTLIMIT:
 823                        next = val + SOFTLIMIT_EVENTS_TARGET;
 824                        break;
 825                case MEM_CGROUP_TARGET_NUMAINFO:
 826                        next = val + NUMAINFO_EVENTS_TARGET;
 827                        break;
 828                default:
 829                        break;
 830                }
 831                __this_cpu_write(memcg->stat->targets[target], next);
 832                return true;
 833        }
 834        return false;
 835}
 836
 837/*
 838 * Check events in order.
 839 *
 840 */
 841static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
 842{
 843        preempt_disable();
 844        /* threshold event is triggered in finer grain than soft limit */
 845        if (unlikely(mem_cgroup_event_ratelimit(memcg,
 846                                                MEM_CGROUP_TARGET_THRESH))) {
 847                bool do_softlimit;
 848                bool do_numainfo __maybe_unused;
 849
 850                do_softlimit = mem_cgroup_event_ratelimit(memcg,
 851                                                MEM_CGROUP_TARGET_SOFTLIMIT);
 852#if MAX_NUMNODES > 1
 853                do_numainfo = mem_cgroup_event_ratelimit(memcg,
 854                                                MEM_CGROUP_TARGET_NUMAINFO);
 855#endif
 856                preempt_enable();
 857
 858                mem_cgroup_threshold(memcg);
 859                if (unlikely(do_softlimit))
 860                        mem_cgroup_update_tree(memcg, page);
 861#if MAX_NUMNODES > 1
 862                if (unlikely(do_numainfo))
 863                        atomic_inc(&memcg->numainfo_events);
 864#endif
 865        } else
 866                preempt_enable();
 867}
 868
 869struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
 870{
 871        return mem_cgroup_from_css(
 872                cgroup_subsys_state(cont, mem_cgroup_subsys_id));
 873}
 874
 875struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
 876{
 877        /*
 878         * mm_update_next_owner() may clear mm->owner to NULL
 879         * if it races with swapoff, page migration, etc.
 880         * So this can be called with p == NULL.
 881         */
 882        if (unlikely(!p))
 883                return NULL;
 884
 885        return mem_cgroup_from_css(task_subsys_state(p, mem_cgroup_subsys_id));
 886}
 887
 888struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
 889{
 890        struct mem_cgroup *memcg = NULL;
 891
 892        if (!mm)
 893                return NULL;
 894        /*
 895         * Because we have no locks, mm->owner's may be being moved to other
 896         * cgroup. We use css_tryget() here even if this looks
 897         * pessimistic (rather than adding locks here).
 898         */
 899        rcu_read_lock();
 900        do {
 901                memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 902                if (unlikely(!memcg))
 903                        break;
 904        } while (!css_tryget(&memcg->css));
 905        rcu_read_unlock();
 906        return memcg;
 907}
 908
 909/**
 910 * mem_cgroup_iter - iterate over memory cgroup hierarchy
 911 * @root: hierarchy root
 912 * @prev: previously returned memcg, NULL on first invocation
 913 * @reclaim: cookie for shared reclaim walks, NULL for full walks
 914 *
 915 * Returns references to children of the hierarchy below @root, or
 916 * @root itself, or %NULL after a full round-trip.
 917 *
 918 * Caller must pass the return value in @prev on subsequent
 919 * invocations for reference counting, or use mem_cgroup_iter_break()
 920 * to cancel a hierarchy walk before the round-trip is complete.
 921 *
 922 * Reclaimers can specify a zone and a priority level in @reclaim to
 923 * divide up the memcgs in the hierarchy among all concurrent
 924 * reclaimers operating on the same zone and priority.
 925 */
 926struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
 927                                   struct mem_cgroup *prev,
 928                                   struct mem_cgroup_reclaim_cookie *reclaim)
 929{
 930        struct mem_cgroup *memcg = NULL;
 931        int id = 0;
 932
 933        if (mem_cgroup_disabled())
 934                return NULL;
 935
 936        if (!root)
 937                root = root_mem_cgroup;
 938
 939        if (prev && !reclaim)
 940                id = css_id(&prev->css);
 941
 942        if (prev && prev != root)
 943                css_put(&prev->css);
 944
 945        if (!root->use_hierarchy && root != root_mem_cgroup) {
 946                if (prev)
 947                        return NULL;
 948                return root;
 949        }
 950
 951        while (!memcg) {
 952                struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
 953                struct cgroup_subsys_state *css;
 954
 955                if (reclaim) {
 956                        int nid = zone_to_nid(reclaim->zone);
 957                        int zid = zone_idx(reclaim->zone);
 958                        struct mem_cgroup_per_zone *mz;
 959
 960                        mz = mem_cgroup_zoneinfo(root, nid, zid);
 961                        iter = &mz->reclaim_iter[reclaim->priority];
 962                        if (prev && reclaim->generation != iter->generation)
 963                                return NULL;
 964                        id = iter->position;
 965                }
 966
 967                rcu_read_lock();
 968                css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id);
 969                if (css) {
 970                        if (css == &root->css || css_tryget(css))
 971                                memcg = mem_cgroup_from_css(css);
 972                } else
 973                        id = 0;
 974                rcu_read_unlock();
 975
 976                if (reclaim) {
 977                        iter->position = id;
 978                        if (!css)
 979                                iter->generation++;
 980                        else if (!prev && memcg)
 981                                reclaim->generation = iter->generation;
 982                }
 983
 984                if (prev && !css)
 985                        return NULL;
 986        }
 987        return memcg;
 988}
 989
 990/**
 991 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
 992 * @root: hierarchy root
 993 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
 994 */
 995void mem_cgroup_iter_break(struct mem_cgroup *root,
 996                           struct mem_cgroup *prev)
 997{
 998        if (!root)
 999                root = root_mem_cgroup;
1000        if (prev && prev != root)
1001                css_put(&prev->css);
1002}
1003
1004/*
1005 * Iteration constructs for visiting all cgroups (under a tree).  If
1006 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1007 * be used for reference counting.
1008 */
1009#define for_each_mem_cgroup_tree(iter, root)            \
1010        for (iter = mem_cgroup_iter(root, NULL, NULL);  \
1011             iter != NULL;                              \
1012             iter = mem_cgroup_iter(root, iter, NULL))
1013
1014#define for_each_mem_cgroup(iter)                       \
1015        for (iter = mem_cgroup_iter(NULL, NULL, NULL);  \
1016             iter != NULL;                              \
1017             iter = mem_cgroup_iter(NULL, iter, NULL))
1018
1019static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
1020{
1021        return (memcg == root_mem_cgroup);
1022}
1023
1024void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
1025{
1026        struct mem_cgroup *memcg;
1027
1028        if (!mm)
1029                return;
1030
1031        rcu_read_lock();
1032        memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1033        if (unlikely(!memcg))
1034                goto out;
1035
1036        switch (idx) {
1037        case PGFAULT:
1038                this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1039                break;
1040        case PGMAJFAULT:
1041                this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
1042                break;
1043        default:
1044                BUG();
1045        }
1046out:
1047        rcu_read_unlock();
1048}
1049EXPORT_SYMBOL(mem_cgroup_count_vm_event);
1050
1051/**
1052 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1053 * @zone: zone of the wanted lruvec
1054 * @memcg: memcg of the wanted lruvec
1055 *
1056 * Returns the lru list vector holding pages for the given @zone and
1057 * @mem.  This can be the global zone lruvec, if the memory controller
1058 * is disabled.
1059 */
1060struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1061                                      struct mem_cgroup *memcg)
1062{
1063        struct mem_cgroup_per_zone *mz;
1064        struct lruvec *lruvec;
1065
1066        if (mem_cgroup_disabled()) {
1067                lruvec = &zone->lruvec;
1068                goto out;
1069        }
1070
1071        mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
1072        lruvec = &mz->lruvec;
1073out:
1074        /*
1075         * Since a node can be onlined after the mem_cgroup was created,
1076         * we have to be prepared to initialize lruvec->zone here;
1077         * and if offlined then reonlined, we need to reinitialize it.
1078         */
1079        if (unlikely(lruvec->zone != zone))
1080                lruvec->zone = zone;
1081        return lruvec;
1082}
1083
1084/*
1085 * Following LRU functions are allowed to be used without PCG_LOCK.
1086 * Operations are called by routine of global LRU independently from memcg.
1087 * What we have to take care of here is validness of pc->mem_cgroup.
1088 *
1089 * Changes to pc->mem_cgroup happens when
1090 * 1. charge
1091 * 2. moving account
1092 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1093 * It is added to LRU before charge.
1094 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1095 * When moving account, the page is not on LRU. It's isolated.
1096 */
1097
1098/**
1099 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
1100 * @page: the page
1101 * @zone: zone of the page
1102 */
1103struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
1104{
1105        struct mem_cgroup_per_zone *mz;
1106        struct mem_cgroup *memcg;
1107        struct page_cgroup *pc;
1108        struct lruvec *lruvec;
1109
1110        if (mem_cgroup_disabled()) {
1111                lruvec = &zone->lruvec;
1112                goto out;
1113        }
1114
1115        pc = lookup_page_cgroup(page);
1116        memcg = pc->mem_cgroup;
1117
1118        /*
1119         * Surreptitiously switch any uncharged offlist page to root:
1120         * an uncharged page off lru does nothing to secure
1121         * its former mem_cgroup from sudden removal.
1122         *
1123         * Our caller holds lru_lock, and PageCgroupUsed is updated
1124         * under page_cgroup lock: between them, they make all uses
1125         * of pc->mem_cgroup safe.
1126         */
1127        if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
1128                pc->mem_cgroup = memcg = root_mem_cgroup;
1129
1130        mz = page_cgroup_zoneinfo(memcg, page);
1131        lruvec = &mz->lruvec;
1132out:
1133        /*
1134         * Since a node can be onlined after the mem_cgroup was created,
1135         * we have to be prepared to initialize lruvec->zone here;
1136         * and if offlined then reonlined, we need to reinitialize it.
1137         */
1138        if (unlikely(lruvec->zone != zone))
1139                lruvec->zone = zone;
1140        return lruvec;
1141}
1142
1143/**
1144 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1145 * @lruvec: mem_cgroup per zone lru vector
1146 * @lru: index of lru list the page is sitting on
1147 * @nr_pages: positive when adding or negative when removing
1148 *
1149 * This function must be called when a page is added to or removed from an
1150 * lru list.
1151 */
1152void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1153                                int nr_pages)
1154{
1155        struct mem_cgroup_per_zone *mz;
1156        unsigned long *lru_size;
1157
1158        if (mem_cgroup_disabled())
1159                return;
1160
1161        mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1162        lru_size = mz->lru_size + lru;
1163        *lru_size += nr_pages;
1164        VM_BUG_ON((long)(*lru_size) < 0);
1165}
1166
1167/*
1168 * Checks whether given mem is same or in the root_mem_cgroup's
1169 * hierarchy subtree
1170 */
1171bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1172                                  struct mem_cgroup *memcg)
1173{
1174        if (root_memcg == memcg)
1175                return true;
1176        if (!root_memcg->use_hierarchy || !memcg)
1177                return false;
1178        return css_is_ancestor(&memcg->css, &root_memcg->css);
1179}
1180
1181static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1182                                       struct mem_cgroup *memcg)
1183{
1184        bool ret;
1185
1186        rcu_read_lock();
1187        ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
1188        rcu_read_unlock();
1189        return ret;
1190}
1191
1192int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg)
1193{
1194        int ret;
1195        struct mem_cgroup *curr = NULL;
1196        struct task_struct *p;
1197
1198        p = find_lock_task_mm(task);
1199        if (p) {
1200                curr = try_get_mem_cgroup_from_mm(p->mm);
1201                task_unlock(p);
1202        } else {
1203                /*
1204                 * All threads may have already detached their mm's, but the oom
1205                 * killer still needs to detect if they have already been oom
1206                 * killed to prevent needlessly killing additional tasks.
1207                 */
1208                task_lock(task);
1209                curr = mem_cgroup_from_task(task);
1210                if (curr)
1211                        css_get(&curr->css);
1212                task_unlock(task);
1213        }
1214        if (!curr)
1215                return 0;
1216        /*
1217         * We should check use_hierarchy of "memcg" not "curr". Because checking
1218         * use_hierarchy of "curr" here make this function true if hierarchy is
1219         * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1220         * hierarchy(even if use_hierarchy is disabled in "memcg").
1221         */
1222        ret = mem_cgroup_same_or_subtree(memcg, curr);
1223        css_put(&curr->css);
1224        return ret;
1225}
1226
1227int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
1228{
1229        unsigned long inactive_ratio;
1230        unsigned long inactive;
1231        unsigned long active;
1232        unsigned long gb;
1233
1234        inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1235        active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
1236
1237        gb = (inactive + active) >> (30 - PAGE_SHIFT);
1238        if (gb)
1239                inactive_ratio = int_sqrt(10 * gb);
1240        else
1241                inactive_ratio = 1;
1242
1243        return inactive * inactive_ratio < active;
1244}
1245
1246int mem_cgroup_inactive_file_is_low(struct lruvec *lruvec)
1247{
1248        unsigned long active;
1249        unsigned long inactive;
1250
1251        inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_FILE);
1252        active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_FILE);
1253
1254        return (active > inactive);
1255}
1256
1257#define mem_cgroup_from_res_counter(counter, member)    \
1258        container_of(counter, struct mem_cgroup, member)
1259
1260/**
1261 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1262 * @memcg: the memory cgroup
1263 *
1264 * Returns the maximum amount of memory @mem can be charged with, in
1265 * pages.
1266 */
1267static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
1268{
1269        unsigned long long margin;
1270
1271        margin = res_counter_margin(&memcg->res);
1272        if (do_swap_account)
1273                margin = min(margin, res_counter_margin(&memcg->memsw));
1274        return margin >> PAGE_SHIFT;
1275}
1276
1277int mem_cgroup_swappiness(struct mem_cgroup *memcg)
1278{
1279        struct cgroup *cgrp = memcg->css.cgroup;
1280
1281        /* root ? */
1282        if (cgrp->parent == NULL)
1283                return vm_swappiness;
1284
1285        return memcg->swappiness;
1286}
1287
1288/*
1289 * memcg->moving_account is used for checking possibility that some thread is
1290 * calling move_account(). When a thread on CPU-A starts moving pages under
1291 * a memcg, other threads should check memcg->moving_account under
1292 * rcu_read_lock(), like this:
1293 *
1294 *         CPU-A                                    CPU-B
1295 *                                              rcu_read_lock()
1296 *         memcg->moving_account+1              if (memcg->mocing_account)
1297 *                                                   take heavy locks.
1298 *         synchronize_rcu()                    update something.
1299 *                                              rcu_read_unlock()
1300 *         start move here.
1301 */
1302
1303/* for quick checking without looking up memcg */
1304atomic_t memcg_moving __read_mostly;
1305
1306static void mem_cgroup_start_move(struct mem_cgroup *memcg)
1307{
1308        atomic_inc(&memcg_moving);
1309        atomic_inc(&memcg->moving_account);
1310        synchronize_rcu();
1311}
1312
1313static void mem_cgroup_end_move(struct mem_cgroup *memcg)
1314{
1315        /*
1316         * Now, mem_cgroup_clear_mc() may call this function with NULL.
1317         * We check NULL in callee rather than caller.
1318         */
1319        if (memcg) {
1320                atomic_dec(&memcg_moving);
1321                atomic_dec(&memcg->moving_account);
1322        }
1323}
1324
1325/*
1326 * 2 routines for checking "mem" is under move_account() or not.
1327 *
1328 * mem_cgroup_stolen() -  checking whether a cgroup is mc.from or not. This
1329 *                        is used for avoiding races in accounting.  If true,
1330 *                        pc->mem_cgroup may be overwritten.
1331 *
1332 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1333 *                        under hierarchy of moving cgroups. This is for
1334 *                        waiting at hith-memory prressure caused by "move".
1335 */
1336
1337static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
1338{
1339        VM_BUG_ON(!rcu_read_lock_held());
1340        return atomic_read(&memcg->moving_account) > 0;
1341}
1342
1343static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
1344{
1345        struct mem_cgroup *from;
1346        struct mem_cgroup *to;
1347        bool ret = false;
1348        /*
1349         * Unlike task_move routines, we access mc.to, mc.from not under
1350         * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1351         */
1352        spin_lock(&mc.lock);
1353        from = mc.from;
1354        to = mc.to;
1355        if (!from)
1356                goto unlock;
1357
1358        ret = mem_cgroup_same_or_subtree(memcg, from)
1359                || mem_cgroup_same_or_subtree(memcg, to);
1360unlock:
1361        spin_unlock(&mc.lock);
1362        return ret;
1363}
1364
1365static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
1366{
1367        if (mc.moving_task && current != mc.moving_task) {
1368                if (mem_cgroup_under_move(memcg)) {
1369                        DEFINE_WAIT(wait);
1370                        prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1371                        /* moving charge context might have finished. */
1372                        if (mc.moving_task)
1373                                schedule();
1374                        finish_wait(&mc.waitq, &wait);
1375                        return true;
1376                }
1377        }
1378        return false;
1379}
1380
1381/*
1382 * Take this lock when
1383 * - a code tries to modify page's memcg while it's USED.
1384 * - a code tries to modify page state accounting in a memcg.
1385 * see mem_cgroup_stolen(), too.
1386 */
1387static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1388                                  unsigned long *flags)
1389{
1390        spin_lock_irqsave(&memcg->move_lock, *flags);
1391}
1392
1393static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1394                                unsigned long *flags)
1395{
1396        spin_unlock_irqrestore(&memcg->move_lock, *flags);
1397}
1398
1399/**
1400 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1401 * @memcg: The memory cgroup that went over limit
1402 * @p: Task that is going to be killed
1403 *
1404 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1405 * enabled
1406 */
1407void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1408{
1409        struct cgroup *task_cgrp;
1410        struct cgroup *mem_cgrp;
1411        /*
1412         * Need a buffer in BSS, can't rely on allocations. The code relies
1413         * on the assumption that OOM is serialized for memory controller.
1414         * If this assumption is broken, revisit this code.
1415         */
1416        static char memcg_name[PATH_MAX];
1417        int ret;
1418
1419        if (!memcg || !p)
1420                return;
1421
1422        rcu_read_lock();
1423
1424        mem_cgrp = memcg->css.cgroup;
1425        task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1426
1427        ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1428        if (ret < 0) {
1429                /*
1430                 * Unfortunately, we are unable to convert to a useful name
1431                 * But we'll still print out the usage information
1432                 */
1433                rcu_read_unlock();
1434                goto done;
1435        }
1436        rcu_read_unlock();
1437
1438        printk(KERN_INFO "Task in %s killed", memcg_name);
1439
1440        rcu_read_lock();
1441        ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1442        if (ret < 0) {
1443                rcu_read_unlock();
1444                goto done;
1445        }
1446        rcu_read_unlock();
1447
1448        /*
1449         * Continues from above, so we don't need an KERN_ level
1450         */
1451        printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1452done:
1453
1454        printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1455                res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1456                res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1457                res_counter_read_u64(&memcg->res, RES_FAILCNT));
1458        printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1459                "failcnt %llu\n",
1460                res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1461                res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1462                res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1463}
1464
1465/*
1466 * This function returns the number of memcg under hierarchy tree. Returns
1467 * 1(self count) if no children.
1468 */
1469static int mem_cgroup_count_children(struct mem_cgroup *memcg)
1470{
1471        int num = 0;
1472        struct mem_cgroup *iter;
1473
1474        for_each_mem_cgroup_tree(iter, memcg)
1475                num++;
1476        return num;
1477}
1478
1479/*
1480 * Return the memory (and swap, if configured) limit for a memcg.
1481 */
1482static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1483{
1484        u64 limit;
1485
1486        limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1487
1488        /*
1489         * Do not consider swap space if we cannot swap due to swappiness
1490         */
1491        if (mem_cgroup_swappiness(memcg)) {
1492                u64 memsw;
1493
1494                limit += total_swap_pages << PAGE_SHIFT;
1495                memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1496
1497                /*
1498                 * If memsw is finite and limits the amount of swap space
1499                 * available to this memcg, return that limit.
1500                 */
1501                limit = min(limit, memsw);
1502        }
1503
1504        return limit;
1505}
1506
1507void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1508                              int order)
1509{
1510        struct mem_cgroup *iter;
1511        unsigned long chosen_points = 0;
1512        unsigned long totalpages;
1513        unsigned int points = 0;
1514        struct task_struct *chosen = NULL;
1515
1516        /*
1517         * If current has a pending SIGKILL, then automatically select it.  The
1518         * goal is to allow it to allocate so that it may quickly exit and free
1519         * its memory.
1520         */
1521        if (fatal_signal_pending(current)) {
1522                set_thread_flag(TIF_MEMDIE);
1523                return;
1524        }
1525
1526        check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
1527        totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
1528        for_each_mem_cgroup_tree(iter, memcg) {
1529                struct cgroup *cgroup = iter->css.cgroup;
1530                struct cgroup_iter it;
1531                struct task_struct *task;
1532
1533                cgroup_iter_start(cgroup, &it);
1534                while ((task = cgroup_iter_next(cgroup, &it))) {
1535                        switch (oom_scan_process_thread(task, totalpages, NULL,
1536                                                        false)) {
1537                        case OOM_SCAN_SELECT:
1538                                if (chosen)
1539                                        put_task_struct(chosen);
1540                                chosen = task;
1541                                chosen_points = ULONG_MAX;
1542                                get_task_struct(chosen);
1543                                /* fall through */
1544                        case OOM_SCAN_CONTINUE:
1545                                continue;
1546                        case OOM_SCAN_ABORT:
1547                                cgroup_iter_end(cgroup, &it);
1548                                mem_cgroup_iter_break(memcg, iter);
1549                                if (chosen)
1550                                        put_task_struct(chosen);
1551                                return;
1552                        case OOM_SCAN_OK:
1553                                break;
1554                        };
1555                        points = oom_badness(task, memcg, NULL, totalpages);
1556                        if (points > chosen_points) {
1557                                if (chosen)
1558                                        put_task_struct(chosen);
1559                                chosen = task;
1560                                chosen_points = points;
1561                                get_task_struct(chosen);
1562                        }
1563                }
1564                cgroup_iter_end(cgroup, &it);
1565        }
1566
1567        if (!chosen)
1568                return;
1569        points = chosen_points * 1000 / totalpages;
1570        oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1571                         NULL, "Memory cgroup out of memory");
1572}
1573
1574static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1575                                        gfp_t gfp_mask,
1576                                        unsigned long flags)
1577{
1578        unsigned long total = 0;
1579        bool noswap = false;
1580        int loop;
1581
1582        if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1583                noswap = true;
1584        if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1585                noswap = true;
1586
1587        for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1588                if (loop)
1589                        drain_all_stock_async(memcg);
1590                total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1591                /*
1592                 * Allow limit shrinkers, which are triggered directly
1593                 * by userspace, to catch signals and stop reclaim
1594                 * after minimal progress, regardless of the margin.
1595                 */
1596                if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1597                        break;
1598                if (mem_cgroup_margin(memcg))
1599                        break;
1600                /*
1601                 * If nothing was reclaimed after two attempts, there
1602                 * may be no reclaimable pages in this hierarchy.
1603                 */
1604                if (loop && !total)
1605                        break;
1606        }
1607        return total;
1608}
1609
1610/**
1611 * test_mem_cgroup_node_reclaimable
1612 * @memcg: the target memcg
1613 * @nid: the node ID to be checked.
1614 * @noswap : specify true here if the user wants flle only information.
1615 *
1616 * This function returns whether the specified memcg contains any
1617 * reclaimable pages on a node. Returns true if there are any reclaimable
1618 * pages in the node.
1619 */
1620static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
1621                int nid, bool noswap)
1622{
1623        if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
1624                return true;
1625        if (noswap || !total_swap_pages)
1626                return false;
1627        if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
1628                return true;
1629        return false;
1630
1631}
1632#if MAX_NUMNODES > 1
1633
1634/*
1635 * Always updating the nodemask is not very good - even if we have an empty
1636 * list or the wrong list here, we can start from some node and traverse all
1637 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1638 *
1639 */
1640static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
1641{
1642        int nid;
1643        /*
1644         * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1645         * pagein/pageout changes since the last update.
1646         */
1647        if (!atomic_read(&memcg->numainfo_events))
1648                return;
1649        if (atomic_inc_return(&memcg->numainfo_updating) > 1)
1650                return;
1651
1652        /* make a nodemask where this memcg uses memory from */
1653        memcg->scan_nodes = node_states[N_HIGH_MEMORY];
1654
1655        for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) {
1656
1657                if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1658                        node_clear(nid, memcg->scan_nodes);
1659        }
1660
1661        atomic_set(&memcg->numainfo_events, 0);
1662        atomic_set(&memcg->numainfo_updating, 0);
1663}
1664
1665/*
1666 * Selecting a node where we start reclaim from. Because what we need is just
1667 * reducing usage counter, start from anywhere is O,K. Considering
1668 * memory reclaim from current node, there are pros. and cons.
1669 *
1670 * Freeing memory from current node means freeing memory from a node which
1671 * we'll use or we've used. So, it may make LRU bad. And if several threads
1672 * hit limits, it will see a contention on a node. But freeing from remote
1673 * node means more costs for memory reclaim because of memory latency.
1674 *
1675 * Now, we use round-robin. Better algorithm is welcomed.
1676 */
1677int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1678{
1679        int node;
1680
1681        mem_cgroup_may_update_nodemask(memcg);
1682        node = memcg->last_scanned_node;
1683
1684        node = next_node(node, memcg->scan_nodes);
1685        if (node == MAX_NUMNODES)
1686                node = first_node(memcg->scan_nodes);
1687        /*
1688         * We call this when we hit limit, not when pages are added to LRU.
1689         * No LRU may hold pages because all pages are UNEVICTABLE or
1690         * memcg is too small and all pages are not on LRU. In that case,
1691         * we use curret node.
1692         */
1693        if (unlikely(node == MAX_NUMNODES))
1694                node = numa_node_id();
1695
1696        memcg->last_scanned_node = node;
1697        return node;
1698}
1699
1700/*
1701 * Check all nodes whether it contains reclaimable pages or not.
1702 * For quick scan, we make use of scan_nodes. This will allow us to skip
1703 * unused nodes. But scan_nodes is lazily updated and may not cotain
1704 * enough new information. We need to do double check.
1705 */
1706static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
1707{
1708        int nid;
1709
1710        /*
1711         * quick check...making use of scan_node.
1712         * We can skip unused nodes.
1713         */
1714        if (!nodes_empty(memcg->scan_nodes)) {
1715                for (nid = first_node(memcg->scan_nodes);
1716                     nid < MAX_NUMNODES;
1717                     nid = next_node(nid, memcg->scan_nodes)) {
1718
1719                        if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1720                                return true;
1721                }
1722        }
1723        /*
1724         * Check rest of nodes.
1725         */
1726        for_each_node_state(nid, N_HIGH_MEMORY) {
1727                if (node_isset(nid, memcg->scan_nodes))
1728                        continue;
1729                if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1730                        return true;
1731        }
1732        return false;
1733}
1734
1735#else
1736int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1737{
1738        return 0;
1739}
1740
1741static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
1742{
1743        return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
1744}
1745#endif
1746
1747static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1748                                   struct zone *zone,
1749                                   gfp_t gfp_mask,
1750                                   unsigned long *total_scanned)
1751{
1752        struct mem_cgroup *victim = NULL;
1753        int total = 0;
1754        int loop = 0;
1755        unsigned long excess;
1756        unsigned long nr_scanned;
1757        struct mem_cgroup_reclaim_cookie reclaim = {
1758                .zone = zone,
1759                .priority = 0,
1760        };
1761
1762        excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
1763
1764        while (1) {
1765                victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1766                if (!victim) {
1767                        loop++;
1768                        if (loop >= 2) {
1769                                /*
1770                                 * If we have not been able to reclaim
1771                                 * anything, it might because there are
1772                                 * no reclaimable pages under this hierarchy
1773                                 */
1774                                if (!total)
1775                                        break;
1776                                /*
1777                                 * We want to do more targeted reclaim.
1778                                 * excess >> 2 is not to excessive so as to
1779                                 * reclaim too much, nor too less that we keep
1780                                 * coming back to reclaim from this cgroup
1781                                 */
1782                                if (total >= (excess >> 2) ||
1783                                        (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1784                                        break;
1785                        }
1786                        continue;
1787                }
1788                if (!mem_cgroup_reclaimable(victim, false))
1789                        continue;
1790                total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1791                                                     zone, &nr_scanned);
1792                *total_scanned += nr_scanned;
1793                if (!res_counter_soft_limit_excess(&root_memcg->res))
1794                        break;
1795        }
1796        mem_cgroup_iter_break(root_memcg, victim);
1797        return total;
1798}
1799
1800/*
1801 * Check OOM-Killer is already running under our hierarchy.
1802 * If someone is running, return false.
1803 * Has to be called with memcg_oom_lock
1804 */
1805static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
1806{
1807        struct mem_cgroup *iter, *failed = NULL;
1808
1809        for_each_mem_cgroup_tree(iter, memcg) {
1810                if (iter->oom_lock) {
1811                        /*
1812                         * this subtree of our hierarchy is already locked
1813                         * so we cannot give a lock.
1814                         */
1815                        failed = iter;
1816                        mem_cgroup_iter_break(memcg, iter);
1817                        break;
1818                } else
1819                        iter->oom_lock = true;
1820        }
1821
1822        if (!failed)
1823                return true;
1824
1825        /*
1826         * OK, we failed to lock the whole subtree so we have to clean up
1827         * what we set up to the failing subtree
1828         */
1829        for_each_mem_cgroup_tree(iter, memcg) {
1830                if (iter == failed) {
1831                        mem_cgroup_iter_break(memcg, iter);
1832                        break;
1833                }
1834                iter->oom_lock = false;
1835        }
1836        return false;
1837}
1838
1839/*
1840 * Has to be called with memcg_oom_lock
1841 */
1842static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
1843{
1844        struct mem_cgroup *iter;
1845
1846        for_each_mem_cgroup_tree(iter, memcg)
1847                iter->oom_lock = false;
1848        return 0;
1849}
1850
1851static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
1852{
1853        struct mem_cgroup *iter;
1854
1855        for_each_mem_cgroup_tree(iter, memcg)
1856                atomic_inc(&iter->under_oom);
1857}
1858
1859static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
1860{
1861        struct mem_cgroup *iter;
1862
1863        /*
1864         * When a new child is created while the hierarchy is under oom,
1865         * mem_cgroup_oom_lock() may not be called. We have to use
1866         * atomic_add_unless() here.
1867         */
1868        for_each_mem_cgroup_tree(iter, memcg)
1869                atomic_add_unless(&iter->under_oom, -1, 0);
1870}
1871
1872static DEFINE_SPINLOCK(memcg_oom_lock);
1873static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1874
1875struct oom_wait_info {
1876        struct mem_cgroup *memcg;
1877        wait_queue_t    wait;
1878};
1879
1880static int memcg_oom_wake_function(wait_queue_t *wait,
1881        unsigned mode, int sync, void *arg)
1882{
1883        struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1884        struct mem_cgroup *oom_wait_memcg;
1885        struct oom_wait_info *oom_wait_info;
1886
1887        oom_wait_info = container_of(wait, struct oom_wait_info, wait);
1888        oom_wait_memcg = oom_wait_info->memcg;
1889
1890        /*
1891         * Both of oom_wait_info->memcg and wake_memcg are stable under us.
1892         * Then we can use css_is_ancestor without taking care of RCU.
1893         */
1894        if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
1895                && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
1896                return 0;
1897        return autoremove_wake_function(wait, mode, sync, arg);
1898}
1899
1900static void memcg_wakeup_oom(struct mem_cgroup *memcg)
1901{
1902        /* for filtering, pass "memcg" as argument. */
1903        __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
1904}
1905
1906static void memcg_oom_recover(struct mem_cgroup *memcg)
1907{
1908        if (memcg && atomic_read(&memcg->under_oom))
1909                memcg_wakeup_oom(memcg);
1910}
1911
1912/*
1913 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1914 */
1915static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask,
1916                                  int order)
1917{
1918        struct oom_wait_info owait;
1919        bool locked, need_to_kill;
1920
1921        owait.memcg = memcg;
1922        owait.wait.flags = 0;
1923        owait.wait.func = memcg_oom_wake_function;
1924        owait.wait.private = current;
1925        INIT_LIST_HEAD(&owait.wait.task_list);
1926        need_to_kill = true;
1927        mem_cgroup_mark_under_oom(memcg);
1928
1929        /* At first, try to OOM lock hierarchy under memcg.*/
1930        spin_lock(&memcg_oom_lock);
1931        locked = mem_cgroup_oom_lock(memcg);
1932        /*
1933         * Even if signal_pending(), we can't quit charge() loop without
1934         * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1935         * under OOM is always welcomed, use TASK_KILLABLE here.
1936         */
1937        prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
1938        if (!locked || memcg->oom_kill_disable)
1939                need_to_kill = false;
1940        if (locked)
1941                mem_cgroup_oom_notify(memcg);
1942        spin_unlock(&memcg_oom_lock);
1943
1944        if (need_to_kill) {
1945                finish_wait(&memcg_oom_waitq, &owait.wait);
1946                mem_cgroup_out_of_memory(memcg, mask, order);
1947        } else {
1948                schedule();
1949                finish_wait(&memcg_oom_waitq, &owait.wait);
1950        }
1951        spin_lock(&memcg_oom_lock);
1952        if (locked)
1953                mem_cgroup_oom_unlock(memcg);
1954        memcg_wakeup_oom(memcg);
1955        spin_unlock(&memcg_oom_lock);
1956
1957        mem_cgroup_unmark_under_oom(memcg);
1958
1959        if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1960                return false;
1961        /* Give chance to dying process */
1962        schedule_timeout_uninterruptible(1);
1963        return true;
1964}
1965
1966/*
1967 * Currently used to update mapped file statistics, but the routine can be
1968 * generalized to update other statistics as well.
1969 *
1970 * Notes: Race condition
1971 *
1972 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1973 * it tends to be costly. But considering some conditions, we doesn't need
1974 * to do so _always_.
1975 *
1976 * Considering "charge", lock_page_cgroup() is not required because all
1977 * file-stat operations happen after a page is attached to radix-tree. There
1978 * are no race with "charge".
1979 *
1980 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1981 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1982 * if there are race with "uncharge". Statistics itself is properly handled
1983 * by flags.
1984 *
1985 * Considering "move", this is an only case we see a race. To make the race
1986 * small, we check mm->moving_account and detect there are possibility of race
1987 * If there is, we take a lock.
1988 */
1989
1990void __mem_cgroup_begin_update_page_stat(struct page *page,
1991                                bool *locked, unsigned long *flags)
1992{
1993        struct mem_cgroup *memcg;
1994        struct page_cgroup *pc;
1995
1996        pc = lookup_page_cgroup(page);
1997again:
1998        memcg = pc->mem_cgroup;
1999        if (unlikely(!memcg || !PageCgroupUsed(pc)))
2000                return;
2001        /*
2002         * If this memory cgroup is not under account moving, we don't
2003         * need to take move_lock_mem_cgroup(). Because we already hold
2004         * rcu_read_lock(), any calls to move_account will be delayed until
2005         * rcu_read_unlock() if mem_cgroup_stolen() == true.
2006         */
2007        if (!mem_cgroup_stolen(memcg))
2008                return;
2009
2010        move_lock_mem_cgroup(memcg, flags);
2011        if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
2012                move_unlock_mem_cgroup(memcg, flags);
2013                goto again;
2014        }
2015        *locked = true;
2016}
2017
2018void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
2019{
2020        struct page_cgroup *pc = lookup_page_cgroup(page);
2021
2022        /*
2023         * It's guaranteed that pc->mem_cgroup never changes while
2024         * lock is held because a routine modifies pc->mem_cgroup
2025         * should take move_lock_mem_cgroup().
2026         */
2027        move_unlock_mem_cgroup(pc->mem_cgroup, flags);
2028}
2029
2030void mem_cgroup_update_page_stat(struct page *page,
2031                                 enum mem_cgroup_page_stat_item idx, int val)
2032{
2033        struct mem_cgroup *memcg;
2034        struct page_cgroup *pc = lookup_page_cgroup(page);
2035        unsigned long uninitialized_var(flags);
2036
2037        if (mem_cgroup_disabled())
2038                return;
2039
2040        memcg = pc->mem_cgroup;
2041        if (unlikely(!memcg || !PageCgroupUsed(pc)))
2042                return;
2043
2044        switch (idx) {
2045        case MEMCG_NR_FILE_MAPPED:
2046                idx = MEM_CGROUP_STAT_FILE_MAPPED;
2047                break;
2048        default:
2049                BUG();
2050        }
2051
2052        this_cpu_add(memcg->stat->count[idx], val);
2053}
2054
2055/*
2056 * size of first charge trial. "32" comes from vmscan.c's magic value.
2057 * TODO: maybe necessary to use big numbers in big irons.
2058 */
2059#define CHARGE_BATCH    32U
2060struct memcg_stock_pcp {
2061        struct mem_cgroup *cached; /* this never be root cgroup */
2062        unsigned int nr_pages;
2063        struct work_struct work;
2064        unsigned long flags;
2065#define FLUSHING_CACHED_CHARGE  0
2066};
2067static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
2068static DEFINE_MUTEX(percpu_charge_mutex);
2069
2070/*
2071 * Try to consume stocked charge on this cpu. If success, one page is consumed
2072 * from local stock and true is returned. If the stock is 0 or charges from a
2073 * cgroup which is not current target, returns false. This stock will be
2074 * refilled.
2075 */
2076static bool consume_stock(struct mem_cgroup *memcg)
2077{
2078        struct memcg_stock_pcp *stock;
2079        bool ret = true;
2080
2081        stock = &get_cpu_var(memcg_stock);
2082        if (memcg == stock->cached && stock->nr_pages)
2083                stock->nr_pages--;
2084        else /* need to call res_counter_charge */
2085                ret = false;
2086        put_cpu_var(memcg_stock);
2087        return ret;
2088}
2089
2090/*
2091 * Returns stocks cached in percpu to res_counter and reset cached information.
2092 */
2093static void drain_stock(struct memcg_stock_pcp *stock)
2094{
2095        struct mem_cgroup *old = stock->cached;
2096
2097        if (stock->nr_pages) {
2098                unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2099
2100                res_counter_uncharge(&old->res, bytes);
2101                if (do_swap_account)
2102                        res_counter_uncharge(&old->memsw, bytes);
2103                stock->nr_pages = 0;
2104        }
2105        stock->cached = NULL;
2106}
2107
2108/*
2109 * This must be called under preempt disabled or must be called by
2110 * a thread which is pinned to local cpu.
2111 */
2112static void drain_local_stock(struct work_struct *dummy)
2113{
2114        struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2115        drain_stock(stock);
2116        clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
2117}
2118
2119/*
2120 * Cache charges(val) which is from res_counter, to local per_cpu area.
2121 * This will be consumed by consume_stock() function, later.
2122 */
2123static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
2124{
2125        struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2126
2127        if (stock->cached != memcg) { /* reset if necessary */
2128                drain_stock(stock);
2129                stock->cached = memcg;
2130        }
2131        stock->nr_pages += nr_pages;
2132        put_cpu_var(memcg_stock);
2133}
2134
2135/*
2136 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2137 * of the hierarchy under it. sync flag says whether we should block
2138 * until the work is done.
2139 */
2140static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
2141{
2142        int cpu, curcpu;
2143
2144        /* Notify other cpus that system-wide "drain" is running */
2145        get_online_cpus();
2146        curcpu = get_cpu();
2147        for_each_online_cpu(cpu) {
2148                struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
2149                struct mem_cgroup *memcg;
2150
2151                memcg = stock->cached;
2152                if (!memcg || !stock->nr_pages)
2153                        continue;
2154                if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
2155                        continue;
2156                if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2157                        if (cpu == curcpu)
2158                                drain_local_stock(&stock->work);
2159                        else
2160                                schedule_work_on(cpu, &stock->work);
2161                }
2162        }
2163        put_cpu();
2164
2165        if (!sync)
2166                goto out;
2167
2168        for_each_online_cpu(cpu) {
2169                struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
2170                if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
2171                        flush_work(&stock->work);
2172        }
2173out:
2174        put_online_cpus();
2175}
2176
2177/*
2178 * Tries to drain stocked charges in other cpus. This function is asynchronous
2179 * and just put a work per cpu for draining localy on each cpu. Caller can
2180 * expects some charges will be back to res_counter later but cannot wait for
2181 * it.
2182 */
2183static void drain_all_stock_async(struct mem_cgroup *root_memcg)
2184{
2185        /*
2186         * If someone calls draining, avoid adding more kworker runs.
2187         */
2188        if (!mutex_trylock(&percpu_charge_mutex))
2189                return;
2190        drain_all_stock(root_memcg, false);
2191        mutex_unlock(&percpu_charge_mutex);
2192}
2193
2194/* This is a synchronous drain interface. */
2195static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
2196{
2197        /* called when force_empty is called */
2198        mutex_lock(&percpu_charge_mutex);
2199        drain_all_stock(root_memcg, true);
2200        mutex_unlock(&percpu_charge_mutex);
2201}
2202
2203/*
2204 * This function drains percpu counter value from DEAD cpu and
2205 * move it to local cpu. Note that this function can be preempted.
2206 */
2207static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
2208{
2209        int i;
2210
2211        spin_lock(&memcg->pcp_counter_lock);
2212        for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
2213                long x = per_cpu(memcg->stat->count[i], cpu);
2214
2215                per_cpu(memcg->stat->count[i], cpu) = 0;
2216                memcg->nocpu_base.count[i] += x;
2217        }
2218        for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
2219                unsigned long x = per_cpu(memcg->stat->events[i], cpu);
2220
2221                per_cpu(memcg->stat->events[i], cpu) = 0;
2222                memcg->nocpu_base.events[i] += x;
2223        }
2224        spin_unlock(&memcg->pcp_counter_lock);
2225}
2226
2227static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
2228                                        unsigned long action,
2229                                        void *hcpu)
2230{
2231        int cpu = (unsigned long)hcpu;
2232        struct memcg_stock_pcp *stock;
2233        struct mem_cgroup *iter;
2234
2235        if (action == CPU_ONLINE)
2236                return NOTIFY_OK;
2237
2238        if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
2239                return NOTIFY_OK;
2240
2241        for_each_mem_cgroup(iter)
2242                mem_cgroup_drain_pcp_counter(iter, cpu);
2243
2244        stock = &per_cpu(memcg_stock, cpu);
2245        drain_stock(stock);
2246        return NOTIFY_OK;
2247}
2248
2249
2250/* See __mem_cgroup_try_charge() for details */
2251enum {
2252        CHARGE_OK,              /* success */
2253        CHARGE_RETRY,           /* need to retry but retry is not bad */
2254        CHARGE_NOMEM,           /* we can't do more. return -ENOMEM */
2255        CHARGE_WOULDBLOCK,      /* GFP_WAIT wasn't set and no enough res. */
2256        CHARGE_OOM_DIE,         /* the current is killed because of OOM */
2257};
2258
2259static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
2260                                unsigned int nr_pages, bool oom_check)
2261{
2262        unsigned long csize = nr_pages * PAGE_SIZE;
2263        struct mem_cgroup *mem_over_limit;
2264        struct res_counter *fail_res;
2265        unsigned long flags = 0;
2266        int ret;
2267
2268        ret = res_counter_charge(&memcg->res, csize, &fail_res);
2269
2270        if (likely(!ret)) {
2271                if (!do_swap_account)
2272                        return CHARGE_OK;
2273                ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
2274                if (likely(!ret))
2275                        return CHARGE_OK;
2276
2277                res_counter_uncharge(&memcg->res, csize);
2278                mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2279                flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2280        } else
2281                mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
2282        /*
2283         * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2284         * of regular pages (CHARGE_BATCH), or a single regular page (1).
2285         *
2286         * Never reclaim on behalf of optional batching, retry with a
2287         * single page instead.
2288         */
2289        if (nr_pages == CHARGE_BATCH)
2290                return CHARGE_RETRY;
2291
2292        if (!(gfp_mask & __GFP_WAIT))
2293                return CHARGE_WOULDBLOCK;
2294
2295        ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
2296        if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
2297                return CHARGE_RETRY;
2298        /*
2299         * Even though the limit is exceeded at this point, reclaim
2300         * may have been able to free some pages.  Retry the charge
2301         * before killing the task.
2302         *
2303         * Only for regular pages, though: huge pages are rather
2304         * unlikely to succeed so close to the limit, and we fall back
2305         * to regular pages anyway in case of failure.
2306         */
2307        if (nr_pages == 1 && ret)
2308                return CHARGE_RETRY;
2309
2310        /*
2311         * At task move, charge accounts can be doubly counted. So, it's
2312         * better to wait until the end of task_move if something is going on.
2313         */
2314        if (mem_cgroup_wait_acct_move(mem_over_limit))
2315                return CHARGE_RETRY;
2316
2317        /* If we don't need to call oom-killer at el, return immediately */
2318        if (!oom_check)
2319                return CHARGE_NOMEM;
2320        /* check OOM */
2321        if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize)))
2322                return CHARGE_OOM_DIE;
2323
2324        return CHARGE_RETRY;
2325}
2326
2327/*
2328 * __mem_cgroup_try_charge() does
2329 * 1. detect memcg to be charged against from passed *mm and *ptr,
2330 * 2. update res_counter
2331 * 3. call memory reclaim if necessary.
2332 *
2333 * In some special case, if the task is fatal, fatal_signal_pending() or
2334 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2335 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2336 * as possible without any hazards. 2: all pages should have a valid
2337 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2338 * pointer, that is treated as a charge to root_mem_cgroup.
2339 *
2340 * So __mem_cgroup_try_charge() will return
2341 *  0       ...  on success, filling *ptr with a valid memcg pointer.
2342 *  -ENOMEM ...  charge failure because of resource limits.
2343 *  -EINTR  ...  if thread is fatal. *ptr is filled with root_mem_cgroup.
2344 *
2345 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2346 * the oom-killer can be invoked.
2347 */
2348static int __mem_cgroup_try_charge(struct mm_struct *mm,
2349                                   gfp_t gfp_mask,
2350                                   unsigned int nr_pages,
2351                                   struct mem_cgroup **ptr,
2352                                   bool oom)
2353{
2354        unsigned int batch = max(CHARGE_BATCH, nr_pages);
2355        int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2356        struct mem_cgroup *memcg = NULL;
2357        int ret;
2358
2359        /*
2360         * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2361         * in system level. So, allow to go ahead dying process in addition to
2362         * MEMDIE process.
2363         */
2364        if (unlikely(test_thread_flag(TIF_MEMDIE)
2365                     || fatal_signal_pending(current)))
2366                goto bypass;
2367
2368        /*
2369         * We always charge the cgroup the mm_struct belongs to.
2370         * The mm_struct's mem_cgroup changes on task migration if the
2371         * thread group leader migrates. It's possible that mm is not
2372         * set, if so charge the root memcg (happens for pagecache usage).
2373         */
2374        if (!*ptr && !mm)
2375                *ptr = root_mem_cgroup;
2376again:
2377        if (*ptr) { /* css should be a valid one */
2378                memcg = *ptr;
2379                VM_BUG_ON(css_is_removed(&memcg->css));
2380                if (mem_cgroup_is_root(memcg))
2381                        goto done;
2382                if (nr_pages == 1 && consume_stock(memcg))
2383                        goto done;
2384                css_get(&memcg->css);
2385        } else {
2386                struct task_struct *p;
2387
2388                rcu_read_lock();
2389                p = rcu_dereference(mm->owner);
2390                /*
2391                 * Because we don't have task_lock(), "p" can exit.
2392                 * In that case, "memcg" can point to root or p can be NULL with
2393                 * race with swapoff. Then, we have small risk of mis-accouning.
2394                 * But such kind of mis-account by race always happens because
2395                 * we don't have cgroup_mutex(). It's overkill and we allo that
2396                 * small race, here.
2397                 * (*) swapoff at el will charge against mm-struct not against
2398                 * task-struct. So, mm->owner can be NULL.
2399                 */
2400                memcg = mem_cgroup_from_task(p);
2401                if (!memcg)
2402                        memcg = root_mem_cgroup;
2403                if (mem_cgroup_is_root(memcg)) {
2404                        rcu_read_unlock();
2405                        goto done;
2406                }
2407                if (nr_pages == 1 && consume_stock(memcg)) {
2408                        /*
2409                         * It seems dagerous to access memcg without css_get().
2410                         * But considering how consume_stok works, it's not
2411                         * necessary. If consume_stock success, some charges
2412                         * from this memcg are cached on this cpu. So, we
2413                         * don't need to call css_get()/css_tryget() before
2414                         * calling consume_stock().
2415                         */
2416                        rcu_read_unlock();
2417                        goto done;
2418                }
2419                /* after here, we may be blocked. we need to get refcnt */
2420                if (!css_tryget(&memcg->css)) {
2421                        rcu_read_unlock();
2422                        goto again;
2423                }
2424                rcu_read_unlock();
2425        }
2426
2427        do {
2428                bool oom_check;
2429
2430                /* If killed, bypass charge */
2431                if (fatal_signal_pending(current)) {
2432                        css_put(&memcg->css);
2433                        goto bypass;
2434                }
2435
2436                oom_check = false;
2437                if (oom && !nr_oom_retries) {
2438                        oom_check = true;
2439                        nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2440                }
2441
2442                ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check);
2443                switch (ret) {
2444                case CHARGE_OK:
2445                        break;
2446                case CHARGE_RETRY: /* not in OOM situation but retry */
2447                        batch = nr_pages;
2448                        css_put(&memcg->css);
2449                        memcg = NULL;
2450                        goto again;
2451                case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
2452                        css_put(&memcg->css);
2453                        goto nomem;
2454                case CHARGE_NOMEM: /* OOM routine works */
2455                        if (!oom) {
2456                                css_put(&memcg->css);
2457                                goto nomem;
2458                        }
2459                        /* If oom, we never return -ENOMEM */
2460                        nr_oom_retries--;
2461                        break;
2462                case CHARGE_OOM_DIE: /* Killed by OOM Killer */
2463                        css_put(&memcg->css);
2464                        goto bypass;
2465                }
2466        } while (ret != CHARGE_OK);
2467
2468        if (batch > nr_pages)
2469                refill_stock(memcg, batch - nr_pages);
2470        css_put(&memcg->css);
2471done:
2472        *ptr = memcg;
2473        return 0;
2474nomem:
2475        *ptr = NULL;
2476        return -ENOMEM;
2477bypass:
2478        *ptr = root_mem_cgroup;
2479        return -EINTR;
2480}
2481
2482/*
2483 * Somemtimes we have to undo a charge we got by try_charge().
2484 * This function is for that and do uncharge, put css's refcnt.
2485 * gotten by try_charge().
2486 */
2487static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
2488                                       unsigned int nr_pages)
2489{
2490        if (!mem_cgroup_is_root(memcg)) {
2491                unsigned long bytes = nr_pages * PAGE_SIZE;
2492
2493                res_counter_uncharge(&memcg->res, bytes);
2494                if (do_swap_account)
2495                        res_counter_uncharge(&memcg->memsw, bytes);
2496        }
2497}
2498
2499/*
2500 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2501 * This is useful when moving usage to parent cgroup.
2502 */
2503static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2504                                        unsigned int nr_pages)
2505{
2506        unsigned long bytes = nr_pages * PAGE_SIZE;
2507
2508        if (mem_cgroup_is_root(memcg))
2509                return;
2510
2511        res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2512        if (do_swap_account)
2513                res_counter_uncharge_until(&memcg->memsw,
2514                                                memcg->memsw.parent, bytes);
2515}
2516
2517/*
2518 * A helper function to get mem_cgroup from ID. must be called under
2519 * rcu_read_lock(). The caller must check css_is_removed() or some if
2520 * it's concern. (dropping refcnt from swap can be called against removed
2521 * memcg.)
2522 */
2523static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2524{
2525        struct cgroup_subsys_state *css;
2526
2527        /* ID 0 is unused ID */
2528        if (!id)
2529                return NULL;
2530        css = css_lookup(&mem_cgroup_subsys, id);
2531        if (!css)
2532                return NULL;
2533        return mem_cgroup_from_css(css);
2534}
2535
2536struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
2537{
2538        struct mem_cgroup *memcg = NULL;
2539        struct page_cgroup *pc;
2540        unsigned short id;
2541        swp_entry_t ent;
2542
2543        VM_BUG_ON(!PageLocked(page));
2544
2545        pc = lookup_page_cgroup(page);
2546        lock_page_cgroup(pc);
2547        if (PageCgroupUsed(pc)) {
2548                memcg = pc->mem_cgroup;
2549                if (memcg && !css_tryget(&memcg->css))
2550                        memcg = NULL;
2551        } else if (PageSwapCache(page)) {
2552                ent.val = page_private(page);
2553                id = lookup_swap_cgroup_id(ent);
2554                rcu_read_lock();
2555                memcg = mem_cgroup_lookup(id);
2556                if (memcg && !css_tryget(&memcg->css))
2557                        memcg = NULL;
2558                rcu_read_unlock();
2559        }
2560        unlock_page_cgroup(pc);
2561        return memcg;
2562}
2563
2564static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
2565                                       struct page *page,
2566                                       unsigned int nr_pages,
2567                                       enum charge_type ctype,
2568                                       bool lrucare)
2569{
2570        struct page_cgroup *pc = lookup_page_cgroup(page);
2571        struct zone *uninitialized_var(zone);
2572        struct lruvec *lruvec;
2573        bool was_on_lru = false;
2574        bool anon;
2575
2576        lock_page_cgroup(pc);
2577        VM_BUG_ON(PageCgroupUsed(pc));
2578        /*
2579         * we don't need page_cgroup_lock about tail pages, becase they are not
2580         * accessed by any other context at this point.
2581         */
2582
2583        /*
2584         * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2585         * may already be on some other mem_cgroup's LRU.  Take care of it.
2586         */
2587        if (lrucare) {
2588                zone = page_zone(page);
2589                spin_lock_irq(&zone->lru_lock);
2590                if (PageLRU(page)) {
2591                        lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
2592                        ClearPageLRU(page);
2593                        del_page_from_lru_list(page, lruvec, page_lru(page));
2594                        was_on_lru = true;
2595                }
2596        }
2597
2598        pc->mem_cgroup = memcg;
2599        /*
2600         * We access a page_cgroup asynchronously without lock_page_cgroup().
2601         * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2602         * is accessed after testing USED bit. To make pc->mem_cgroup visible
2603         * before USED bit, we need memory barrier here.
2604         * See mem_cgroup_add_lru_list(), etc.
2605         */
2606        smp_wmb();
2607        SetPageCgroupUsed(pc);
2608
2609        if (lrucare) {
2610                if (was_on_lru) {
2611                        lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
2612                        VM_BUG_ON(PageLRU(page));
2613                        SetPageLRU(page);
2614                        add_page_to_lru_list(page, lruvec, page_lru(page));
2615                }
2616                spin_unlock_irq(&zone->lru_lock);
2617        }
2618
2619        if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON)
2620                anon = true;
2621        else
2622                anon = false;
2623
2624        mem_cgroup_charge_statistics(memcg, anon, nr_pages);
2625        unlock_page_cgroup(pc);
2626
2627        /*
2628         * "charge_statistics" updated event counter. Then, check it.
2629         * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2630         * if they exceeds softlimit.
2631         */
2632        memcg_check_events(memcg, page);
2633}
2634
2635#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2636
2637#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
2638/*
2639 * Because tail pages are not marked as "used", set it. We're under
2640 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2641 * charge/uncharge will be never happen and move_account() is done under
2642 * compound_lock(), so we don't have to take care of races.
2643 */
2644void mem_cgroup_split_huge_fixup(struct page *head)
2645{
2646        struct page_cgroup *head_pc = lookup_page_cgroup(head);
2647        struct page_cgroup *pc;
2648        int i;
2649
2650        if (mem_cgroup_disabled())
2651                return;
2652        for (i = 1; i < HPAGE_PMD_NR; i++) {
2653                pc = head_pc + i;
2654                pc->mem_cgroup = head_pc->mem_cgroup;
2655                smp_wmb();/* see __commit_charge() */
2656                pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2657        }
2658}
2659#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2660
2661/**
2662 * mem_cgroup_move_account - move account of the page
2663 * @page: the page
2664 * @nr_pages: number of regular pages (>1 for huge pages)
2665 * @pc: page_cgroup of the page.
2666 * @from: mem_cgroup which the page is moved from.
2667 * @to: mem_cgroup which the page is moved to. @from != @to.
2668 *
2669 * The caller must confirm following.
2670 * - page is not on LRU (isolate_page() is useful.)
2671 * - compound_lock is held when nr_pages > 1
2672 *
2673 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
2674 * from old cgroup.
2675 */
2676static int mem_cgroup_move_account(struct page *page,
2677                                   unsigned int nr_pages,
2678                                   struct page_cgroup *pc,
2679                                   struct mem_cgroup *from,
2680                                   struct mem_cgroup *to)
2681{
2682        unsigned long flags;
2683        int ret;
2684        bool anon = PageAnon(page);
2685
2686        VM_BUG_ON(from == to);
2687        VM_BUG_ON(PageLRU(page));
2688        /*
2689         * The page is isolated from LRU. So, collapse function
2690         * will not handle this page. But page splitting can happen.
2691         * Do this check under compound_page_lock(). The caller should
2692         * hold it.
2693         */
2694        ret = -EBUSY;
2695        if (nr_pages > 1 && !PageTransHuge(page))
2696                goto out;
2697
2698        lock_page_cgroup(pc);
2699
2700        ret = -EINVAL;
2701        if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
2702                goto unlock;
2703
2704        move_lock_mem_cgroup(from, &flags);
2705
2706        if (!anon && page_mapped(page)) {
2707                /* Update mapped_file data for mem_cgroup */
2708                preempt_disable();
2709                __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2710                __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2711                preempt_enable();
2712        }
2713        mem_cgroup_charge_statistics(from, anon, -nr_pages);
2714
2715        /* caller should have done css_get */
2716        pc->mem_cgroup = to;
2717        mem_cgroup_charge_statistics(to, anon, nr_pages);
2718        /*
2719         * We charges against "to" which may not have any tasks. Then, "to"
2720         * can be under rmdir(). But in current implementation, caller of
2721         * this function is just force_empty() and move charge, so it's
2722         * guaranteed that "to" is never removed. So, we don't check rmdir
2723         * status here.
2724         */
2725        move_unlock_mem_cgroup(from, &flags);
2726        ret = 0;
2727unlock:
2728        unlock_page_cgroup(pc);
2729        /*
2730         * check events
2731         */
2732        memcg_check_events(to, page);
2733        memcg_check_events(from, page);
2734out:
2735        return ret;
2736}
2737
2738/*
2739 * move charges to its parent.
2740 */
2741
2742static int mem_cgroup_move_parent(struct page *page,
2743                                  struct page_cgroup *pc,
2744                                  struct mem_cgroup *child)
2745{
2746        struct mem_cgroup *parent;
2747        unsigned int nr_pages;
2748        unsigned long uninitialized_var(flags);
2749        int ret;
2750
2751        /* Is ROOT ? */
2752        if (mem_cgroup_is_root(child))
2753                return -EINVAL;
2754
2755        ret = -EBUSY;
2756        if (!get_page_unless_zero(page))
2757                goto out;
2758        if (isolate_lru_page(page))
2759                goto put;
2760
2761        nr_pages = hpage_nr_pages(page);
2762
2763        parent = parent_mem_cgroup(child);
2764        /*
2765         * If no parent, move charges to root cgroup.
2766         */
2767        if (!parent)
2768                parent = root_mem_cgroup;
2769
2770        if (nr_pages > 1)
2771                flags = compound_lock_irqsave(page);
2772
2773        ret = mem_cgroup_move_account(page, nr_pages,
2774                                pc, child, parent);
2775        if (!ret)
2776                __mem_cgroup_cancel_local_charge(child, nr_pages);
2777
2778        if (nr_pages > 1)
2779                compound_unlock_irqrestore(page, flags);
2780        putback_lru_page(page);
2781put:
2782        put_page(page);
2783out:
2784        return ret;
2785}
2786
2787/*
2788 * Charge the memory controller for page usage.
2789 * Return
2790 * 0 if the charge was successful
2791 * < 0 if the cgroup is over its limit
2792 */
2793static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
2794                                gfp_t gfp_mask, enum charge_type ctype)
2795{
2796        struct mem_cgroup *memcg = NULL;
2797        unsigned int nr_pages = 1;
2798        bool oom = true;
2799        int ret;
2800
2801        if (PageTransHuge(page)) {
2802                nr_pages <<= compound_order(page);
2803                VM_BUG_ON(!PageTransHuge(page));
2804                /*
2805                 * Never OOM-kill a process for a huge page.  The
2806                 * fault handler will fall back to regular pages.
2807                 */
2808                oom = false;
2809        }
2810
2811        ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
2812        if (ret == -ENOMEM)
2813                return ret;
2814        __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
2815        return 0;
2816}
2817
2818int mem_cgroup_newpage_charge(struct page *page,
2819                              struct mm_struct *mm, gfp_t gfp_mask)
2820{
2821        if (mem_cgroup_disabled())
2822                return 0;
2823        VM_BUG_ON(page_mapped(page));
2824        VM_BUG_ON(page->mapping && !PageAnon(page));
2825        VM_BUG_ON(!mm);
2826        return mem_cgroup_charge_common(page, mm, gfp_mask,
2827                                        MEM_CGROUP_CHARGE_TYPE_ANON);
2828}
2829
2830/*
2831 * While swap-in, try_charge -> commit or cancel, the page is locked.
2832 * And when try_charge() successfully returns, one refcnt to memcg without
2833 * struct page_cgroup is acquired. This refcnt will be consumed by
2834 * "commit()" or removed by "cancel()"
2835 */
2836static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2837                                          struct page *page,
2838                                          gfp_t mask,
2839                                          struct mem_cgroup **memcgp)
2840{
2841        struct mem_cgroup *memcg;
2842        struct page_cgroup *pc;
2843        int ret;
2844
2845        pc = lookup_page_cgroup(page);
2846        /*
2847         * Every swap fault against a single page tries to charge the
2848         * page, bail as early as possible.  shmem_unuse() encounters
2849         * already charged pages, too.  The USED bit is protected by
2850         * the page lock, which serializes swap cache removal, which
2851         * in turn serializes uncharging.
2852         */
2853        if (PageCgroupUsed(pc))
2854                return 0;
2855        if (!do_swap_account)
2856                goto charge_cur_mm;
2857        memcg = try_get_mem_cgroup_from_page(page);
2858        if (!memcg)
2859                goto charge_cur_mm;
2860        *memcgp = memcg;
2861        ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
2862        css_put(&memcg->css);
2863        if (ret == -EINTR)
2864                ret = 0;
2865        return ret;
2866charge_cur_mm:
2867        ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
2868        if (ret == -EINTR)
2869                ret = 0;
2870        return ret;
2871}
2872
2873int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page,
2874                                 gfp_t gfp_mask, struct mem_cgroup **memcgp)
2875{
2876        *memcgp = NULL;
2877        if (mem_cgroup_disabled())
2878                return 0;
2879        /*
2880         * A racing thread's fault, or swapoff, may have already
2881         * updated the pte, and even removed page from swap cache: in
2882         * those cases unuse_pte()'s pte_same() test will fail; but
2883         * there's also a KSM case which does need to charge the page.
2884         */
2885        if (!PageSwapCache(page)) {
2886                int ret;
2887
2888                ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true);
2889                if (ret == -EINTR)
2890                        ret = 0;
2891                return ret;
2892        }
2893        return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
2894}
2895
2896void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
2897{
2898        if (mem_cgroup_disabled())
2899                return;
2900        if (!memcg)
2901                return;
2902        __mem_cgroup_cancel_charge(memcg, 1);
2903}
2904
2905static void
2906__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
2907                                        enum charge_type ctype)
2908{
2909        if (mem_cgroup_disabled())
2910                return;
2911        if (!memcg)
2912                return;
2913        cgroup_exclude_rmdir(&memcg->css);
2914
2915        __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
2916        /*
2917         * Now swap is on-memory. This means this page may be
2918         * counted both as mem and swap....double count.
2919         * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2920         * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2921         * may call delete_from_swap_cache() before reach here.
2922         */
2923        if (do_swap_account && PageSwapCache(page)) {
2924                swp_entry_t ent = {.val = page_private(page)};
2925                mem_cgroup_uncharge_swap(ent);
2926        }
2927        /*
2928         * At swapin, we may charge account against cgroup which has no tasks.
2929         * So, rmdir()->pre_destroy() can be called while we do this charge.
2930         * In that case, we need to call pre_destroy() again. check it here.
2931         */
2932        cgroup_release_and_wakeup_rmdir(&memcg->css);
2933}
2934
2935void mem_cgroup_commit_charge_swapin(struct page *page,
2936                                     struct mem_cgroup *memcg)
2937{
2938        __mem_cgroup_commit_charge_swapin(page, memcg,
2939                                          MEM_CGROUP_CHARGE_TYPE_ANON);
2940}
2941
2942int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2943                                gfp_t gfp_mask)
2944{
2945        struct mem_cgroup *memcg = NULL;
2946        enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
2947        int ret;
2948
2949        if (mem_cgroup_disabled())
2950                return 0;
2951        if (PageCompound(page))
2952                return 0;
2953
2954        if (!PageSwapCache(page))
2955                ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
2956        else { /* page is swapcache/shmem */
2957                ret = __mem_cgroup_try_charge_swapin(mm, page,
2958                                                     gfp_mask, &memcg);
2959                if (!ret)
2960                        __mem_cgroup_commit_charge_swapin(page, memcg, type);
2961        }
2962        return ret;
2963}
2964
2965static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
2966                                   unsigned int nr_pages,
2967                                   const enum charge_type ctype)
2968{
2969        struct memcg_batch_info *batch = NULL;
2970        bool uncharge_memsw = true;
2971
2972        /* If swapout, usage of swap doesn't decrease */
2973        if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2974                uncharge_memsw = false;
2975
2976        batch = &current->memcg_batch;
2977        /*
2978         * In usual, we do css_get() when we remember memcg pointer.
2979         * But in this case, we keep res->usage until end of a series of
2980         * uncharges. Then, it's ok to ignore memcg's refcnt.
2981         */
2982        if (!batch->memcg)
2983                batch->memcg = memcg;
2984        /*
2985         * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2986         * In those cases, all pages freed continuously can be expected to be in
2987         * the same cgroup and we have chance to coalesce uncharges.
2988         * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2989         * because we want to do uncharge as soon as possible.
2990         */
2991
2992        if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2993                goto direct_uncharge;
2994
2995        if (nr_pages > 1)
2996                goto direct_uncharge;
2997
2998        /*
2999         * In typical case, batch->memcg == mem. This means we can
3000         * merge a series of uncharges to an uncharge of res_counter.
3001         * If not, we uncharge res_counter ony by one.
3002         */
3003        if (batch->memcg != memcg)
3004                goto direct_uncharge;
3005        /* remember freed charge and uncharge it later */
3006        batch->nr_pages++;
3007        if (uncharge_memsw)
3008                batch->memsw_nr_pages++;
3009        return;
3010direct_uncharge:
3011        res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
3012        if (uncharge_memsw)
3013                res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
3014        if (unlikely(batch->memcg != memcg))
3015                memcg_oom_recover(memcg);
3016}
3017
3018/*
3019 * uncharge if !page_mapped(page)
3020 */
3021static struct mem_cgroup *
3022__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype,
3023                             bool end_migration)
3024{
3025        struct mem_cgroup *memcg = NULL;
3026        unsigned int nr_pages = 1;
3027        struct page_cgroup *pc;
3028        bool anon;
3029
3030        if (mem_cgroup_disabled())
3031                return NULL;
3032
3033        VM_BUG_ON(PageSwapCache(page));
3034
3035        if (PageTransHuge(page)) {
3036                nr_pages <<= compound_order(page);
3037                VM_BUG_ON(!PageTransHuge(page));
3038        }
3039        /*
3040         * Check if our page_cgroup is valid
3041         */
3042        pc = lookup_page_cgroup(page);
3043        if (unlikely(!PageCgroupUsed(pc)))
3044                return NULL;
3045
3046        lock_page_cgroup(pc);
3047
3048        memcg = pc->mem_cgroup;
3049
3050        if (!PageCgroupUsed(pc))
3051                goto unlock_out;
3052
3053        anon = PageAnon(page);
3054
3055        switch (ctype) {
3056        case MEM_CGROUP_CHARGE_TYPE_ANON:
3057                /*
3058                 * Generally PageAnon tells if it's the anon statistics to be
3059                 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
3060                 * used before page reached the stage of being marked PageAnon.
3061                 */
3062                anon = true;
3063                /* fallthrough */
3064        case MEM_CGROUP_CHARGE_TYPE_DROP:
3065                /* See mem_cgroup_prepare_migration() */
3066                if (page_mapped(page))
3067                        goto unlock_out;
3068                /*
3069                 * Pages under migration may not be uncharged.  But
3070                 * end_migration() /must/ be the one uncharging the
3071                 * unused post-migration page and so it has to call
3072                 * here with the migration bit still set.  See the
3073                 * res_counter handling below.
3074                 */
3075                if (!end_migration && PageCgroupMigration(pc))
3076                        goto unlock_out;
3077                break;
3078        case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
3079                if (!PageAnon(page)) {  /* Shared memory */
3080                        if (page->mapping && !page_is_file_cache(page))
3081                                goto unlock_out;
3082                } else if (page_mapped(page)) /* Anon */
3083                                goto unlock_out;
3084                break;
3085        default:
3086                break;
3087        }
3088
3089        mem_cgroup_charge_statistics(memcg, anon, -nr_pages);
3090
3091        ClearPageCgroupUsed(pc);
3092        /*
3093         * pc->mem_cgroup is not cleared here. It will be accessed when it's
3094         * freed from LRU. This is safe because uncharged page is expected not
3095         * to be reused (freed soon). Exception is SwapCache, it's handled by
3096         * special functions.
3097         */
3098
3099        unlock_page_cgroup(pc);
3100        /*
3101         * even after unlock, we have memcg->res.usage here and this memcg
3102         * will never be freed.
3103         */
3104        memcg_check_events(memcg, page);
3105        if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
3106                mem_cgroup_swap_statistics(memcg, true);
3107                mem_cgroup_get(memcg);
3108        }
3109        /*
3110         * Migration does not charge the res_counter for the
3111         * replacement page, so leave it alone when phasing out the
3112         * page that is unused after the migration.
3113         */
3114        if (!end_migration && !mem_cgroup_is_root(memcg))
3115                mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
3116
3117        return memcg;
3118
3119unlock_out:
3120        unlock_page_cgroup(pc);
3121        return NULL;
3122}
3123
3124void mem_cgroup_uncharge_page(struct page *page)
3125{
3126        /* early check. */
3127        if (page_mapped(page))
3128                return;
3129        VM_BUG_ON(page->mapping && !PageAnon(page));
3130        if (PageSwapCache(page))
3131                return;
3132        __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false);
3133}
3134
3135void mem_cgroup_uncharge_cache_page(struct page *page)
3136{
3137        VM_BUG_ON(page_mapped(page));
3138        VM_BUG_ON(page->mapping);
3139        __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false);
3140}
3141
3142/*
3143 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3144 * In that cases, pages are freed continuously and we can expect pages
3145 * are in the same memcg. All these calls itself limits the number of
3146 * pages freed at once, then uncharge_start/end() is called properly.
3147 * This may be called prural(2) times in a context,
3148 */
3149
3150void mem_cgroup_uncharge_start(void)
3151{
3152        current->memcg_batch.do_batch++;
3153        /* We can do nest. */
3154        if (current->memcg_batch.do_batch == 1) {
3155                current->memcg_batch.memcg = NULL;
3156                current->memcg_batch.nr_pages = 0;
3157                current->memcg_batch.memsw_nr_pages = 0;
3158        }
3159}
3160
3161void mem_cgroup_uncharge_end(void)
3162{
3163        struct memcg_batch_info *batch = &current->memcg_batch;
3164
3165        if (!batch->do_batch)
3166                return;
3167
3168        batch->do_batch--;
3169        if (batch->do_batch) /* If stacked, do nothing. */
3170                return;
3171
3172        if (!batch->memcg)
3173                return;
3174        /*
3175         * This "batch->memcg" is valid without any css_get/put etc...
3176         * bacause we hide charges behind us.
3177         */
3178        if (batch->nr_pages)
3179                res_counter_uncharge(&batch->memcg->res,
3180                                     batch->nr_pages * PAGE_SIZE);
3181        if (batch->memsw_nr_pages)
3182                res_counter_uncharge(&batch->memcg->memsw,
3183                                     batch->memsw_nr_pages * PAGE_SIZE);
3184        memcg_oom_recover(batch->memcg);
3185        /* forget this pointer (for sanity check) */
3186        batch->memcg = NULL;
3187}
3188
3189#ifdef CONFIG_SWAP
3190/*
3191 * called after __delete_from_swap_cache() and drop "page" account.
3192 * memcg information is recorded to swap_cgroup of "ent"
3193 */
3194void
3195mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
3196{
3197        struct mem_cgroup *memcg;
3198        int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
3199
3200        if (!swapout) /* this was a swap cache but the swap is unused ! */
3201                ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
3202
3203        memcg = __mem_cgroup_uncharge_common(page, ctype, false);
3204
3205        /*
3206         * record memcg information,  if swapout && memcg != NULL,
3207         * mem_cgroup_get() was called in uncharge().
3208         */
3209        if (do_swap_account && swapout && memcg)
3210                swap_cgroup_record(ent, css_id(&memcg->css));
3211}
3212#endif
3213
3214#ifdef CONFIG_MEMCG_SWAP
3215/*
3216 * called from swap_entry_free(). remove record in swap_cgroup and
3217 * uncharge "memsw" account.
3218 */
3219void mem_cgroup_uncharge_swap(swp_entry_t ent)
3220{
3221        struct mem_cgroup *memcg;
3222        unsigned short id;
3223
3224        if (!do_swap_account)
3225                return;
3226
3227        id = swap_cgroup_record(ent, 0);
3228        rcu_read_lock();
3229        memcg = mem_cgroup_lookup(id);
3230        if (memcg) {
3231                /*
3232                 * We uncharge this because swap is freed.
3233                 * This memcg can be obsolete one. We avoid calling css_tryget
3234                 */
3235                if (!mem_cgroup_is_root(memcg))
3236                        res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
3237                mem_cgroup_swap_statistics(memcg, false);
3238                mem_cgroup_put(memcg);
3239        }
3240        rcu_read_unlock();
3241}
3242
3243/**
3244 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3245 * @entry: swap entry to be moved
3246 * @from:  mem_cgroup which the entry is moved from
3247 * @to:  mem_cgroup which the entry is moved to
3248 *
3249 * It succeeds only when the swap_cgroup's record for this entry is the same
3250 * as the mem_cgroup's id of @from.
3251 *
3252 * Returns 0 on success, -EINVAL on failure.
3253 *
3254 * The caller must have charged to @to, IOW, called res_counter_charge() about
3255 * both res and memsw, and called css_get().
3256 */
3257static int mem_cgroup_move_swap_account(swp_entry_t entry,
3258                                struct mem_cgroup *from, struct mem_cgroup *to)
3259{
3260        unsigned short old_id, new_id;
3261
3262        old_id = css_id(&from->css);
3263        new_id = css_id(&to->css);
3264
3265        if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
3266                mem_cgroup_swap_statistics(from, false);
3267                mem_cgroup_swap_statistics(to, true);
3268                /*
3269                 * This function is only called from task migration context now.
3270                 * It postpones res_counter and refcount handling till the end
3271                 * of task migration(mem_cgroup_clear_mc()) for performance
3272                 * improvement. But we cannot postpone mem_cgroup_get(to)
3273                 * because if the process that has been moved to @to does
3274                 * swap-in, the refcount of @to might be decreased to 0.
3275                 */
3276                mem_cgroup_get(to);
3277                return 0;
3278        }
3279        return -EINVAL;
3280}
3281#else
3282static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
3283                                struct mem_cgroup *from, struct mem_cgroup *to)
3284{
3285        return -EINVAL;
3286}
3287#endif
3288
3289/*
3290 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3291 * page belongs to.
3292 */
3293void mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
3294                                  struct mem_cgroup **memcgp)
3295{
3296        struct mem_cgroup *memcg = NULL;
3297        struct page_cgroup *pc;
3298        enum charge_type ctype;
3299
3300        *memcgp = NULL;
3301
3302        VM_BUG_ON(PageTransHuge(page));
3303        if (mem_cgroup_disabled())
3304                return;
3305
3306        pc = lookup_page_cgroup(page);
3307        lock_page_cgroup(pc);
3308        if (PageCgroupUsed(pc)) {
3309                memcg = pc->mem_cgroup;
3310                css_get(&memcg->css);
3311                /*
3312                 * At migrating an anonymous page, its mapcount goes down
3313                 * to 0 and uncharge() will be called. But, even if it's fully
3314                 * unmapped, migration may fail and this page has to be
3315                 * charged again. We set MIGRATION flag here and delay uncharge
3316                 * until end_migration() is called
3317                 *
3318                 * Corner Case Thinking
3319                 * A)
3320                 * When the old page was mapped as Anon and it's unmap-and-freed
3321                 * while migration was ongoing.
3322                 * If unmap finds the old page, uncharge() of it will be delayed
3323                 * until end_migration(). If unmap finds a new page, it's
3324                 * uncharged when it make mapcount to be 1->0. If unmap code
3325                 * finds swap_migration_entry, the new page will not be mapped
3326                 * and end_migration() will find it(mapcount==0).
3327                 *
3328                 * B)
3329                 * When the old page was mapped but migraion fails, the kernel
3330                 * remaps it. A charge for it is kept by MIGRATION flag even
3331                 * if mapcount goes down to 0. We can do remap successfully
3332                 * without charging it again.
3333                 *
3334                 * C)
3335                 * The "old" page is under lock_page() until the end of
3336                 * migration, so, the old page itself will not be swapped-out.
3337                 * If the new page is swapped out before end_migraton, our
3338                 * hook to usual swap-out path will catch the event.
3339                 */
3340                if (PageAnon(page))
3341                        SetPageCgroupMigration(pc);
3342        }
3343        unlock_page_cgroup(pc);
3344        /*
3345         * If the page is not charged at this point,
3346         * we return here.
3347         */
3348        if (!memcg)
3349                return;
3350
3351        *memcgp = memcg;
3352        /*
3353         * We charge new page before it's used/mapped. So, even if unlock_page()
3354         * is called before end_migration, we can catch all events on this new
3355         * page. In the case new page is migrated but not remapped, new page's
3356         * mapcount will be finally 0 and we call uncharge in end_migration().
3357         */
3358        if (PageAnon(page))
3359                ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
3360        else
3361                ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
3362        /*
3363         * The page is committed to the memcg, but it's not actually
3364         * charged to the res_counter since we plan on replacing the
3365         * old one and only one page is going to be left afterwards.
3366         */
3367        __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false);
3368}
3369
3370/* remove redundant charge if migration failed*/
3371void mem_cgroup_end_migration(struct mem_cgroup *memcg,
3372        struct page *oldpage, struct page *newpage, bool migration_ok)
3373{
3374        struct page *used, *unused;
3375        struct page_cgroup *pc;
3376        bool anon;
3377
3378        if (!memcg)
3379                return;
3380        /* blocks rmdir() */
3381        cgroup_exclude_rmdir(&memcg->css);
3382        if (!migration_ok) {
3383                used = oldpage;
3384                unused = newpage;
3385        } else {
3386                used = newpage;
3387                unused = oldpage;
3388        }
3389        anon = PageAnon(used);
3390        __mem_cgroup_uncharge_common(unused,
3391                                     anon ? MEM_CGROUP_CHARGE_TYPE_ANON
3392                                     : MEM_CGROUP_CHARGE_TYPE_CACHE,
3393                                     true);
3394        css_put(&memcg->css);
3395        /*
3396         * We disallowed uncharge of pages under migration because mapcount
3397         * of the page goes down to zero, temporarly.
3398         * Clear the flag and check the page should be charged.
3399         */
3400        pc = lookup_page_cgroup(oldpage);
3401        lock_page_cgroup(pc);
3402        ClearPageCgroupMigration(pc);
3403        unlock_page_cgroup(pc);
3404
3405        /*
3406         * If a page is a file cache, radix-tree replacement is very atomic
3407         * and we can skip this check. When it was an Anon page, its mapcount
3408         * goes down to 0. But because we added MIGRATION flage, it's not
3409         * uncharged yet. There are several case but page->mapcount check
3410         * and USED bit check in mem_cgroup_uncharge_page() will do enough
3411         * check. (see prepare_charge() also)
3412         */
3413        if (anon)
3414                mem_cgroup_uncharge_page(used);
3415        /*
3416         * At migration, we may charge account against cgroup which has no
3417         * tasks.
3418         * So, rmdir()->pre_destroy() can be called while we do this charge.
3419         * In that case, we need to call pre_destroy() again. check it here.
3420         */
3421        cgroup_release_and_wakeup_rmdir(&memcg->css);
3422}
3423
3424/*
3425 * At replace page cache, newpage is not under any memcg but it's on
3426 * LRU. So, this function doesn't touch res_counter but handles LRU
3427 * in correct way. Both pages are locked so we cannot race with uncharge.
3428 */
3429void mem_cgroup_replace_page_cache(struct page *oldpage,
3430                                  struct page *newpage)
3431{
3432        struct mem_cgroup *memcg = NULL;
3433        struct page_cgroup *pc;
3434        enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
3435
3436        if (mem_cgroup_disabled())
3437                return;
3438
3439        pc = lookup_page_cgroup(oldpage);
3440        /* fix accounting on old pages */
3441        lock_page_cgroup(pc);
3442        if (PageCgroupUsed(pc)) {
3443                memcg = pc->mem_cgroup;
3444                mem_cgroup_charge_statistics(memcg, false, -1);
3445                ClearPageCgroupUsed(pc);
3446        }
3447        unlock_page_cgroup(pc);
3448
3449        /*
3450         * When called from shmem_replace_page(), in some cases the
3451         * oldpage has already been charged, and in some cases not.
3452         */
3453        if (!memcg)
3454                return;
3455        /*
3456         * Even if newpage->mapping was NULL before starting replacement,
3457         * the newpage may be on LRU(or pagevec for LRU) already. We lock
3458         * LRU while we overwrite pc->mem_cgroup.
3459         */
3460        __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
3461}
3462
3463#ifdef CONFIG_DEBUG_VM
3464static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
3465{
3466        struct page_cgroup *pc;
3467
3468        pc = lookup_page_cgroup(page);
3469        /*
3470         * Can be NULL while feeding pages into the page allocator for
3471         * the first time, i.e. during boot or memory hotplug;
3472         * or when mem_cgroup_disabled().
3473         */
3474        if (likely(pc) && PageCgroupUsed(pc))
3475                return pc;
3476        return NULL;
3477}
3478
3479bool mem_cgroup_bad_page_check(struct page *page)
3480{
3481        if (mem_cgroup_disabled())
3482                return false;
3483
3484        return lookup_page_cgroup_used(page) != NULL;
3485}
3486
3487void mem_cgroup_print_bad_page(struct page *page)
3488{
3489        struct page_cgroup *pc;
3490
3491        pc = lookup_page_cgroup_used(page);
3492        if (pc) {
3493                printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3494                       pc, pc->flags, pc->mem_cgroup);
3495        }
3496}
3497#endif
3498
3499static DEFINE_MUTEX(set_limit_mutex);
3500
3501static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
3502                                unsigned long long val)
3503{
3504        int retry_count;
3505        u64 memswlimit, memlimit;
3506        int ret = 0;
3507        int children = mem_cgroup_count_children(memcg);
3508        u64 curusage, oldusage;
3509        int enlarge;
3510
3511        /*
3512         * For keeping hierarchical_reclaim simple, how long we should retry
3513         * is depends on callers. We set our retry-count to be function
3514         * of # of children which we should visit in this loop.
3515         */
3516        retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3517
3518        oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3519
3520        enlarge = 0;
3521        while (retry_count) {
3522                if (signal_pending(current)) {
3523                        ret = -EINTR;
3524                        break;
3525                }
3526                /*
3527                 * Rather than hide all in some function, I do this in
3528                 * open coded manner. You see what this really does.
3529                 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3530                 */
3531                mutex_lock(&set_limit_mutex);
3532                memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3533                if (memswlimit < val) {
3534                        ret = -EINVAL;
3535                        mutex_unlock(&set_limit_mutex);
3536                        break;
3537                }
3538
3539                memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3540                if (memlimit < val)
3541                        enlarge = 1;
3542
3543                ret = res_counter_set_limit(&memcg->res, val);
3544                if (!ret) {
3545                        if (memswlimit == val)
3546                                memcg->memsw_is_minimum = true;
3547                        else
3548                                memcg->memsw_is_minimum = false;
3549                }
3550                mutex_unlock(&set_limit_mutex);
3551
3552                if (!ret)
3553                        break;
3554
3555                mem_cgroup_reclaim(memcg, GFP_KERNEL,
3556                                   MEM_CGROUP_RECLAIM_SHRINK);
3557                curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3558                /* Usage is reduced ? */
3559                if (curusage >= oldusage)
3560                        retry_count--;
3561                else
3562                        oldusage = curusage;
3563        }
3564        if (!ret && enlarge)
3565                memcg_oom_recover(memcg);
3566
3567        return ret;
3568}
3569
3570static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3571                                        unsigned long long val)
3572{
3573        int retry_count;
3574        u64 memlimit, memswlimit, oldusage, curusage;
3575        int children = mem_cgroup_count_children(memcg);
3576        int ret = -EBUSY;
3577        int enlarge = 0;
3578
3579        /* see mem_cgroup_resize_res_limit */
3580        retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3581        oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
3582        while (retry_count) {
3583                if (signal_pending(current)) {
3584                        ret = -EINTR;
3585                        break;
3586                }
3587                /*
3588                 * Rather than hide all in some function, I do this in
3589                 * open coded manner. You see what this really does.
3590                 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3591                 */
3592                mutex_lock(&set_limit_mutex);
3593                memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3594                if (memlimit > val) {
3595                        ret = -EINVAL;
3596                        mutex_unlock(&set_limit_mutex);
3597                        break;
3598                }
3599                memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3600                if (memswlimit < val)
3601                        enlarge = 1;
3602                ret = res_counter_set_limit(&memcg->memsw, val);
3603                if (!ret) {
3604                        if (memlimit == val)
3605                                memcg->memsw_is_minimum = true;
3606                        else
3607                                memcg->memsw_is_minimum = false;
3608                }
3609                mutex_unlock(&set_limit_mutex);
3610
3611                if (!ret)
3612                        break;
3613
3614                mem_cgroup_reclaim(memcg, GFP_KERNEL,
3615                                   MEM_CGROUP_RECLAIM_NOSWAP |
3616                                   MEM_CGROUP_RECLAIM_SHRINK);
3617                curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
3618                /* Usage is reduced ? */
3619                if (curusage >= oldusage)
3620                        retry_count--;
3621                else
3622                        oldusage = curusage;
3623        }
3624        if (!ret && enlarge)
3625                memcg_oom_recover(memcg);
3626        return ret;
3627}
3628
3629unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
3630                                            gfp_t gfp_mask,
3631                                            unsigned long *total_scanned)
3632{
3633        unsigned long nr_reclaimed = 0;
3634        struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3635        unsigned long reclaimed;
3636        int loop = 0;
3637        struct mem_cgroup_tree_per_zone *mctz;
3638        unsigned long long excess;
3639        unsigned long nr_scanned;
3640
3641        if (order > 0)
3642                return 0;
3643
3644        mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
3645        /*
3646         * This loop can run a while, specially if mem_cgroup's continuously
3647         * keep exceeding their soft limit and putting the system under
3648         * pressure
3649         */
3650        do {
3651                if (next_mz)
3652                        mz = next_mz;
3653                else
3654                        mz = mem_cgroup_largest_soft_limit_node(mctz);
3655                if (!mz)
3656                        break;
3657
3658                nr_scanned = 0;
3659                reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
3660                                                    gfp_mask, &nr_scanned);
3661                nr_reclaimed += reclaimed;
3662                *total_scanned += nr_scanned;
3663                spin_lock(&mctz->lock);
3664
3665                /*
3666                 * If we failed to reclaim anything from this memory cgroup
3667                 * it is time to move on to the next cgroup
3668                 */
3669                next_mz = NULL;
3670                if (!reclaimed) {
3671                        do {
3672                                /*
3673                                 * Loop until we find yet another one.
3674                                 *
3675                                 * By the time we get the soft_limit lock
3676                                 * again, someone might have aded the
3677                                 * group back on the RB tree. Iterate to
3678                                 * make sure we get a different mem.
3679                                 * mem_cgroup_largest_soft_limit_node returns
3680                                 * NULL if no other cgroup is present on
3681                                 * the tree
3682                                 */
3683                                next_mz =
3684                                __mem_cgroup_largest_soft_limit_node(mctz);
3685                                if (next_mz == mz)
3686                                        css_put(&next_mz->memcg->css);
3687                                else /* next_mz == NULL or other memcg */
3688                                        break;
3689                        } while (1);
3690                }
3691                __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
3692                excess = res_counter_soft_limit_excess(&mz->memcg->res);
3693                /*
3694                 * One school of thought says that we should not add
3695                 * back the node to the tree if reclaim returns 0.
3696                 * But our reclaim could return 0, simply because due
3697                 * to priority we are exposing a smaller subset of
3698                 * memory to reclaim from. Consider this as a longer
3699                 * term TODO.
3700                 */
3701                /* If excess == 0, no tree ops */
3702                __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
3703                spin_unlock(&mctz->lock);
3704                css_put(&mz->memcg->css);
3705                loop++;
3706                /*
3707                 * Could not reclaim anything and there are no more
3708                 * mem cgroups to try or we seem to be looping without
3709                 * reclaiming anything.
3710                 */
3711                if (!nr_reclaimed &&
3712                        (next_mz == NULL ||
3713                        loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3714                        break;
3715        } while (!nr_reclaimed);
3716        if (next_mz)
3717                css_put(&next_mz->memcg->css);
3718        return nr_reclaimed;
3719}
3720
3721/*
3722 * Traverse a specified page_cgroup list and try to drop them all.  This doesn't
3723 * reclaim the pages page themselves - it just removes the page_cgroups.
3724 * Returns true if some page_cgroups were not freed, indicating that the caller
3725 * must retry this operation.
3726 */
3727static bool mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
3728                                int node, int zid, enum lru_list lru)
3729{
3730        struct lruvec *lruvec;
3731        unsigned long flags, loop;
3732        struct list_head *list;
3733        struct page *busy;
3734        struct zone *zone;
3735
3736        zone = &NODE_DATA(node)->node_zones[zid];
3737        lruvec = mem_cgroup_zone_lruvec(zone, memcg);
3738        list = &lruvec->lists[lru];
3739
3740        loop = mem_cgroup_get_lru_size(lruvec, lru);
3741        /* give some margin against EBUSY etc...*/
3742        loop += 256;
3743        busy = NULL;
3744        while (loop--) {
3745                struct page_cgroup *pc;
3746                struct page *page;
3747
3748                spin_lock_irqsave(&zone->lru_lock, flags);
3749                if (list_empty(list)) {
3750                        spin_unlock_irqrestore(&zone->lru_lock, flags);
3751                        break;
3752                }
3753                page = list_entry(list->prev, struct page, lru);
3754                if (busy == page) {
3755                        list_move(&page->lru, list);
3756                        busy = NULL;
3757                        spin_unlock_irqrestore(&zone->lru_lock, flags);
3758                        continue;
3759                }
3760                spin_unlock_irqrestore(&zone->lru_lock, flags);
3761
3762                pc = lookup_page_cgroup(page);
3763
3764                if (mem_cgroup_move_parent(page, pc, memcg)) {
3765                        /* found lock contention or "pc" is obsolete. */
3766                        busy = page;
3767                        cond_resched();
3768                } else
3769                        busy = NULL;
3770        }
3771        return !list_empty(list);
3772}
3773
3774/*
3775 * make mem_cgroup's charge to be 0 if there is no task.
3776 * This enables deleting this mem_cgroup.
3777 */
3778static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all)
3779{
3780        int ret;
3781        int node, zid, shrink;
3782        int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
3783        struct cgroup *cgrp = memcg->css.cgroup;
3784
3785        css_get(&memcg->css);
3786
3787        shrink = 0;
3788        /* should free all ? */
3789        if (free_all)
3790                goto try_to_free;
3791move_account:
3792        do {
3793                ret = -EBUSY;
3794                if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
3795                        goto out;
3796                /* This is for making all *used* pages to be on LRU. */
3797                lru_add_drain_all();
3798                drain_all_stock_sync(memcg);
3799                ret = 0;
3800                mem_cgroup_start_move(memcg);
3801                for_each_node_state(node, N_HIGH_MEMORY) {
3802                        for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
3803                                enum lru_list lru;
3804                                for_each_lru(lru) {
3805                                        ret = mem_cgroup_force_empty_list(memcg,
3806                                                        node, zid, lru);
3807                                        if (ret)
3808                                                break;
3809                                }
3810                        }
3811                        if (ret)
3812                                break;
3813                }
3814                mem_cgroup_end_move(memcg);
3815                memcg_oom_recover(memcg);
3816                cond_resched();
3817        /* "ret" should also be checked to ensure all lists are empty. */
3818        } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0 || ret);
3819out:
3820        css_put(&memcg->css);
3821        return ret;
3822
3823try_to_free:
3824        /* returns EBUSY if there is a task or if we come here twice. */
3825        if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
3826                ret = -EBUSY;
3827                goto out;
3828        }
3829        /* we call try-to-free pages for make this cgroup empty */
3830        lru_add_drain_all();
3831        /* try to free all pages in this cgroup */
3832        shrink = 1;
3833        while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
3834                int progress;
3835
3836                if (signal_pending(current)) {
3837                        ret = -EINTR;
3838                        goto out;
3839                }
3840                progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
3841                                                false);
3842                if (!progress) {
3843                        nr_retries--;
3844                        /* maybe some writeback is necessary */
3845                        congestion_wait(BLK_RW_ASYNC, HZ/10);
3846                }
3847
3848        }
3849        lru_add_drain();
3850        /* try move_account...there may be some *locked* pages. */
3851        goto move_account;
3852}
3853
3854static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3855{
3856        return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3857}
3858
3859
3860static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3861{
3862        return mem_cgroup_from_cont(cont)->use_hierarchy;
3863}
3864
3865static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3866                                        u64 val)
3867{
3868        int retval = 0;
3869        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3870        struct cgroup *parent = cont->parent;
3871        struct mem_cgroup *parent_memcg = NULL;
3872
3873        if (parent)
3874                parent_memcg = mem_cgroup_from_cont(parent);
3875
3876        cgroup_lock();
3877
3878        if (memcg->use_hierarchy == val)
3879                goto out;
3880
3881        /*
3882         * If parent's use_hierarchy is set, we can't make any modifications
3883         * in the child subtrees. If it is unset, then the change can
3884         * occur, provided the current cgroup has no children.
3885         *
3886         * For the root cgroup, parent_mem is NULL, we allow value to be
3887         * set if there are no children.
3888         */
3889        if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
3890                                (val == 1 || val == 0)) {
3891                if (list_empty(&cont->children))
3892                        memcg->use_hierarchy = val;
3893                else
3894                        retval = -EBUSY;
3895        } else
3896                retval = -EINVAL;
3897
3898out:
3899        cgroup_unlock();
3900
3901        return retval;
3902}
3903
3904
3905static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
3906                                               enum mem_cgroup_stat_index idx)
3907{
3908        struct mem_cgroup *iter;
3909        long val = 0;
3910
3911        /* Per-cpu values can be negative, use a signed accumulator */
3912        for_each_mem_cgroup_tree(iter, memcg)
3913                val += mem_cgroup_read_stat(iter, idx);
3914
3915        if (val < 0) /* race ? */
3916                val = 0;
3917        return val;
3918}
3919
3920static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
3921{
3922        u64 val;
3923
3924        if (!mem_cgroup_is_root(memcg)) {
3925                if (!swap)
3926                        return res_counter_read_u64(&memcg->res, RES_USAGE);
3927                else
3928                        return res_counter_read_u64(&memcg->memsw, RES_USAGE);
3929        }
3930
3931        val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
3932        val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
3933
3934        if (swap)
3935                val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
3936
3937        return val << PAGE_SHIFT;
3938}
3939
3940static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft,
3941                               struct file *file, char __user *buf,
3942                               size_t nbytes, loff_t *ppos)
3943{
3944        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3945        char str[64];
3946        u64 val;
3947        int type, name, len;
3948
3949        type = MEMFILE_TYPE(cft->private);
3950        name = MEMFILE_ATTR(cft->private);
3951
3952        if (!do_swap_account && type == _MEMSWAP)
3953                return -EOPNOTSUPP;
3954
3955        switch (type) {
3956        case _MEM:
3957                if (name == RES_USAGE)
3958                        val = mem_cgroup_usage(memcg, false);
3959                else
3960                        val = res_counter_read_u64(&memcg->res, name);
3961                break;
3962        case _MEMSWAP:
3963                if (name == RES_USAGE)
3964                        val = mem_cgroup_usage(memcg, true);
3965                else
3966                        val = res_counter_read_u64(&memcg->memsw, name);
3967                break;
3968        default:
3969                BUG();
3970        }
3971
3972        len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val);
3973        return simple_read_from_buffer(buf, nbytes, ppos, str, len);
3974}
3975/*
3976 * The user of this function is...
3977 * RES_LIMIT.
3978 */
3979static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3980                            const char *buffer)
3981{
3982        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3983        int type, name;
3984        unsigned long long val;
3985        int ret;
3986
3987        type = MEMFILE_TYPE(cft->private);
3988        name = MEMFILE_ATTR(cft->private);
3989
3990        if (!do_swap_account && type == _MEMSWAP)
3991                return -EOPNOTSUPP;
3992
3993        switch (name) {
3994        case RES_LIMIT:
3995                if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3996                        ret = -EINVAL;
3997                        break;
3998                }
3999                /* This function does all necessary parse...reuse it */
4000                ret = res_counter_memparse_write_strategy(buffer, &val);
4001                if (ret)
4002                        break;
4003                if (type == _MEM)
4004                        ret = mem_cgroup_resize_limit(memcg, val);
4005                else
4006                        ret = mem_cgroup_resize_memsw_limit(memcg, val);
4007                break;
4008        case RES_SOFT_LIMIT:
4009                ret = res_counter_memparse_write_strategy(buffer, &val);
4010                if (ret)
4011                        break;
4012                /*
4013                 * For memsw, soft limits are hard to implement in terms
4014                 * of semantics, for now, we support soft limits for
4015                 * control without swap
4016                 */
4017                if (type == _MEM)
4018                        ret = res_counter_set_soft_limit(&memcg->res, val);
4019                else
4020                        ret = -EINVAL;
4021                break;
4022        default:
4023                ret = -EINVAL; /* should be BUG() ? */
4024                break;
4025        }
4026        return ret;
4027}
4028
4029static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
4030                unsigned long long *mem_limit, unsigned long long *memsw_limit)
4031{
4032        struct cgroup *cgroup;
4033        unsigned long long min_limit, min_memsw_limit, tmp;
4034
4035        min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4036        min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4037        cgroup = memcg->css.cgroup;
4038        if (!memcg->use_hierarchy)
4039                goto out;
4040
4041        while (cgroup->parent) {
4042                cgroup = cgroup->parent;
4043                memcg = mem_cgroup_from_cont(cgroup);
4044                if (!memcg->use_hierarchy)
4045                        break;
4046                tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
4047                min_limit = min(min_limit, tmp);
4048                tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4049                min_memsw_limit = min(min_memsw_limit, tmp);
4050        }
4051out:
4052        *mem_limit = min_limit;
4053        *memsw_limit = min_memsw_limit;
4054}
4055
4056static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
4057{
4058        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
4059        int type, name;
4060
4061        type = MEMFILE_TYPE(event);
4062        name = MEMFILE_ATTR(event);
4063
4064        if (!do_swap_account && type == _MEMSWAP)
4065                return -EOPNOTSUPP;
4066
4067        switch (name) {
4068        case RES_MAX_USAGE:
4069                if (type == _MEM)
4070                        res_counter_reset_max(&memcg->res);
4071                else
4072                        res_counter_reset_max(&memcg->memsw);
4073                break;
4074        case RES_FAILCNT:
4075                if (type == _MEM)
4076                        res_counter_reset_failcnt(&memcg->res);
4077                else
4078                        res_counter_reset_failcnt(&memcg->memsw);
4079                break;
4080        }
4081
4082        return 0;
4083}
4084
4085static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
4086                                        struct cftype *cft)
4087{
4088        return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
4089}
4090
4091#ifdef CONFIG_MMU
4092static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4093                                        struct cftype *cft, u64 val)
4094{
4095        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4096
4097        if (val >= (1 << NR_MOVE_TYPE))
4098                return -EINVAL;
4099        /*
4100         * We check this value several times in both in can_attach() and
4101         * attach(), so we need cgroup lock to prevent this value from being
4102         * inconsistent.
4103         */
4104        cgroup_lock();
4105        memcg->move_charge_at_immigrate = val;
4106        cgroup_unlock();
4107
4108        return 0;
4109}
4110#else
4111static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4112                                        struct cftype *cft, u64 val)
4113{
4114        return -ENOSYS;
4115}
4116#endif
4117
4118#ifdef CONFIG_NUMA
4119static int memcg_numa_stat_show(struct cgroup *cont, struct cftype *cft,
4120                                      struct seq_file *m)
4121{
4122        int nid;
4123        unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
4124        unsigned long node_nr;
4125        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
4126
4127        total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL);
4128        seq_printf(m, "total=%lu", total_nr);
4129        for_each_node_state(nid, N_HIGH_MEMORY) {
4130                node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL);
4131                seq_printf(m, " N%d=%lu", nid, node_nr);
4132        }
4133        seq_putc(m, '\n');
4134
4135        file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE);
4136        seq_printf(m, "file=%lu", file_nr);
4137        for_each_node_state(nid, N_HIGH_MEMORY) {
4138                node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
4139                                LRU_ALL_FILE);
4140                seq_printf(m, " N%d=%lu", nid, node_nr);
4141        }
4142        seq_putc(m, '\n');
4143
4144        anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON);
4145        seq_printf(m, "anon=%lu", anon_nr);
4146        for_each_node_state(nid, N_HIGH_MEMORY) {
4147                node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
4148                                LRU_ALL_ANON);
4149                seq_printf(m, " N%d=%lu", nid, node_nr);
4150        }
4151        seq_putc(m, '\n');
4152
4153        unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
4154        seq_printf(m, "unevictable=%lu", unevictable_nr);
4155        for_each_node_state(nid, N_HIGH_MEMORY) {
4156                node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
4157                                BIT(LRU_UNEVICTABLE));
4158                seq_printf(m, " N%d=%lu", nid, node_nr);
4159        }
4160        seq_putc(m, '\n');
4161        return 0;
4162}
4163#endif /* CONFIG_NUMA */
4164
4165static const char * const mem_cgroup_lru_names[] = {
4166        "inactive_anon",
4167        "active_anon",
4168        "inactive_file",
4169        "active_file",
4170        "unevictable",
4171};
4172
4173static inline void mem_cgroup_lru_names_not_uptodate(void)
4174{
4175        BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
4176}
4177
4178static int memcg_stat_show(struct cgroup *cont, struct cftype *cft,
4179                                 struct seq_file *m)
4180{
4181        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
4182        struct mem_cgroup *mi;
4183        unsigned int i;
4184
4185        for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
4186                if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
4187                        continue;
4188                seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
4189                           mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
4190        }
4191
4192        for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
4193                seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
4194                           mem_cgroup_read_events(memcg, i));
4195
4196        for (i = 0; i < NR_LRU_LISTS; i++)
4197                seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
4198                           mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
4199
4200        /* Hierarchical information */
4201        {
4202                unsigned long long limit, memsw_limit;
4203                memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
4204                seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
4205                if (do_swap_account)
4206                        seq_printf(m, "hierarchical_memsw_limit %llu\n",
4207                                   memsw_limit);
4208        }
4209
4210        for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
4211                long long val = 0;
4212
4213                if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
4214                        continue;
4215                for_each_mem_cgroup_tree(mi, memcg)
4216                        val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
4217                seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
4218        }
4219
4220        for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
4221                unsigned long long val = 0;
4222
4223                for_each_mem_cgroup_tree(mi, memcg)
4224                        val += mem_cgroup_read_events(mi, i);
4225                seq_printf(m, "total_%s %llu\n",
4226                           mem_cgroup_events_names[i], val);
4227        }
4228
4229        for (i = 0; i < NR_LRU_LISTS; i++) {
4230                unsigned long long val = 0;
4231
4232                for_each_mem_cgroup_tree(mi, memcg)
4233                        val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
4234                seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
4235        }
4236
4237#ifdef CONFIG_DEBUG_VM
4238        {
4239                int nid, zid;
4240                struct mem_cgroup_per_zone *mz;
4241                struct zone_reclaim_stat *rstat;
4242                unsigned long recent_rotated[2] = {0, 0};
4243                unsigned long recent_scanned[2] = {0, 0};
4244
4245                for_each_online_node(nid)
4246                        for (zid = 0; zid < MAX_NR_ZONES; zid++) {
4247                                mz = mem_cgroup_zoneinfo(memcg, nid, zid);
4248                                rstat = &mz->lruvec.reclaim_stat;
4249
4250                                recent_rotated[0] += rstat->recent_rotated[0];
4251                                recent_rotated[1] += rstat->recent_rotated[1];
4252                                recent_scanned[0] += rstat->recent_scanned[0];
4253                                recent_scanned[1] += rstat->recent_scanned[1];
4254                        }
4255                seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
4256                seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
4257                seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
4258                seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
4259        }
4260#endif
4261
4262        return 0;
4263}
4264
4265static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
4266{
4267        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4268
4269        return mem_cgroup_swappiness(memcg);
4270}
4271
4272static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
4273                                       u64 val)
4274{
4275        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4276        struct mem_cgroup *parent;
4277
4278        if (val > 100)
4279                return -EINVAL;
4280
4281        if (cgrp->parent == NULL)
4282                return -EINVAL;
4283
4284        parent = mem_cgroup_from_cont(cgrp->parent);
4285
4286        cgroup_lock();
4287
4288        /* If under hierarchy, only empty-root can set this value */
4289        if ((parent->use_hierarchy) ||
4290            (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4291                cgroup_unlock();
4292                return -EINVAL;
4293        }
4294
4295        memcg->swappiness = val;
4296
4297        cgroup_unlock();
4298
4299        return 0;
4300}
4301
4302static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4303{
4304        struct mem_cgroup_threshold_ary *t;
4305        u64 usage;
4306        int i;
4307
4308        rcu_read_lock();
4309        if (!swap)
4310                t = rcu_dereference(memcg->thresholds.primary);
4311        else
4312                t = rcu_dereference(memcg->memsw_thresholds.primary);
4313
4314        if (!t)
4315                goto unlock;
4316
4317        usage = mem_cgroup_usage(memcg, swap);
4318
4319        /*
4320         * current_threshold points to threshold just below or equal to usage.
4321         * If it's not true, a threshold was crossed after last
4322         * call of __mem_cgroup_threshold().
4323         */
4324        i = t->current_threshold;
4325
4326        /*
4327         * Iterate backward over array of thresholds starting from
4328         * current_threshold and check if a threshold is crossed.
4329         * If none of thresholds below usage is crossed, we read
4330         * only one element of the array here.
4331         */
4332        for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4333                eventfd_signal(t->entries[i].eventfd, 1);
4334
4335        /* i = current_threshold + 1 */
4336        i++;
4337
4338        /*
4339         * Iterate forward over array of thresholds starting from
4340         * current_threshold+1 and check if a threshold is crossed.
4341         * If none of thresholds above usage is crossed, we read
4342         * only one element of the array here.
4343         */
4344        for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4345                eventfd_signal(t->entries[i].eventfd, 1);
4346
4347        /* Update current_threshold */
4348        t->current_threshold = i - 1;
4349unlock:
4350        rcu_read_unlock();
4351}
4352
4353static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4354{
4355        while (memcg) {
4356                __mem_cgroup_threshold(memcg, false);
4357                if (do_swap_account)
4358                        __mem_cgroup_threshold(memcg, true);
4359
4360                memcg = parent_mem_cgroup(memcg);
4361        }
4362}
4363
4364static int compare_thresholds(const void *a, const void *b)
4365{
4366        const struct mem_cgroup_threshold *_a = a;
4367        const struct mem_cgroup_threshold *_b = b;
4368
4369        return _a->threshold - _b->threshold;
4370}
4371
4372static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
4373{
4374        struct mem_cgroup_eventfd_list *ev;
4375
4376        list_for_each_entry(ev, &memcg->oom_notify, list)
4377                eventfd_signal(ev->eventfd, 1);
4378        return 0;
4379}
4380
4381static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
4382{
4383        struct mem_cgroup *iter;
4384
4385        for_each_mem_cgroup_tree(iter, memcg)
4386                mem_cgroup_oom_notify_cb(iter);
4387}
4388
4389static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
4390        struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4391{
4392        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4393        struct mem_cgroup_thresholds *thresholds;
4394        struct mem_cgroup_threshold_ary *new;
4395        int type = MEMFILE_TYPE(cft->private);
4396        u64 threshold, usage;
4397        int i, size, ret;
4398
4399        ret = res_counter_memparse_write_strategy(args, &threshold);
4400        if (ret)
4401                return ret;
4402
4403        mutex_lock(&memcg->thresholds_lock);
4404
4405        if (type == _MEM)
4406                thresholds = &memcg->thresholds;
4407        else if (type == _MEMSWAP)
4408                thresholds = &memcg->memsw_thresholds;
4409        else
4410                BUG();
4411
4412        usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4413
4414        /* Check if a threshold crossed before adding a new one */
4415        if (thresholds->primary)
4416                __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4417
4418        size = thresholds->primary ? thresholds->primary->size + 1 : 1;
4419
4420        /* Allocate memory for new array of thresholds */
4421        new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
4422                        GFP_KERNEL);
4423        if (!new) {
4424                ret = -ENOMEM;
4425                goto unlock;
4426        }
4427        new->size = size;
4428
4429        /* Copy thresholds (if any) to new array */
4430        if (thresholds->primary) {
4431                memcpy(new->entries, thresholds->primary->entries, (size - 1) *
4432                                sizeof(struct mem_cgroup_threshold));
4433        }
4434
4435        /* Add new threshold */
4436        new->entries[size - 1].eventfd = eventfd;
4437        new->entries[size - 1].threshold = threshold;
4438
4439        /* Sort thresholds. Registering of new threshold isn't time-critical */
4440        sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
4441                        compare_thresholds, NULL);
4442
4443        /* Find current threshold */
4444        new->current_threshold = -1;
4445        for (i = 0; i < size; i++) {
4446                if (new->entries[i].threshold <= usage) {
4447                        /*
4448                         * new->current_threshold will not be used until
4449                         * rcu_assign_pointer(), so it's safe to increment
4450                         * it here.
4451                         */
4452                        ++new->current_threshold;
4453                } else
4454                        break;
4455        }
4456
4457        /* Free old spare buffer and save old primary buffer as spare */
4458        kfree(thresholds->spare);
4459        thresholds->spare = thresholds->primary;
4460
4461        rcu_assign_pointer(thresholds->primary, new);
4462
4463        /* To be sure that nobody uses thresholds */
4464        synchronize_rcu();
4465
4466unlock:
4467        mutex_unlock(&memcg->thresholds_lock);
4468
4469        return ret;
4470}
4471
4472static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
4473        struct cftype *cft, struct eventfd_ctx *eventfd)
4474{
4475        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4476        struct mem_cgroup_thresholds *thresholds;
4477        struct mem_cgroup_threshold_ary *new;
4478        int type = MEMFILE_TYPE(cft->private);
4479        u64 usage;
4480        int i, j, size;
4481
4482        mutex_lock(&memcg->thresholds_lock);
4483        if (type == _MEM)
4484                thresholds = &memcg->thresholds;
4485        else if (type == _MEMSWAP)
4486                thresholds = &memcg->memsw_thresholds;
4487        else
4488                BUG();
4489
4490        if (!thresholds->primary)
4491                goto unlock;
4492
4493        usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4494
4495        /* Check if a threshold crossed before removing */
4496        __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4497
4498        /* Calculate new number of threshold */
4499        size = 0;
4500        for (i = 0; i < thresholds->primary->size; i++) {
4501                if (thresholds->primary->entries[i].eventfd != eventfd)
4502                        size++;
4503        }
4504
4505        new = thresholds->spare;
4506
4507        /* Set thresholds array to NULL if we don't have thresholds */
4508        if (!size) {
4509                kfree(new);
4510                new = NULL;
4511                goto swap_buffers;
4512        }
4513
4514        new->size = size;
4515
4516        /* Copy thresholds and find current threshold */
4517        new->current_threshold = -1;
4518        for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4519                if (thresholds->primary->entries[i].eventfd == eventfd)
4520                        continue;
4521
4522                new->entries[j] = thresholds->primary->entries[i];
4523                if (new->entries[j].threshold <= usage) {
4524                        /*
4525                         * new->current_threshold will not be used
4526                         * until rcu_assign_pointer(), so it's safe to increment
4527                         * it here.
4528                         */
4529                        ++new->current_threshold;
4530                }
4531                j++;
4532        }
4533
4534swap_buffers:
4535        /* Swap primary and spare array */
4536        thresholds->spare = thresholds->primary;
4537        /* If all events are unregistered, free the spare array */
4538        if (!new) {
4539                kfree(thresholds->spare);
4540                thresholds->spare = NULL;
4541        }
4542
4543        rcu_assign_pointer(thresholds->primary, new);
4544
4545        /* To be sure that nobody uses thresholds */
4546        synchronize_rcu();
4547unlock:
4548        mutex_unlock(&memcg->thresholds_lock);
4549}
4550
4551static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4552        struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4553{
4554        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4555        struct mem_cgroup_eventfd_list *event;
4556        int type = MEMFILE_TYPE(cft->private);
4557
4558        BUG_ON(type != _OOM_TYPE);
4559        event = kmalloc(sizeof(*event), GFP_KERNEL);
4560        if (!event)
4561                return -ENOMEM;
4562
4563        spin_lock(&memcg_oom_lock);
4564
4565        event->eventfd = eventfd;
4566        list_add(&event->list, &memcg->oom_notify);
4567
4568        /* already in OOM ? */
4569        if (atomic_read(&memcg->under_oom))
4570                eventfd_signal(eventfd, 1);
4571        spin_unlock(&memcg_oom_lock);
4572
4573        return 0;
4574}
4575
4576static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
4577        struct cftype *cft, struct eventfd_ctx *eventfd)
4578{
4579        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4580        struct mem_cgroup_eventfd_list *ev, *tmp;
4581        int type = MEMFILE_TYPE(cft->private);
4582
4583        BUG_ON(type != _OOM_TYPE);
4584
4585        spin_lock(&memcg_oom_lock);
4586
4587        list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
4588                if (ev->eventfd == eventfd) {
4589                        list_del(&ev->list);
4590                        kfree(ev);
4591                }
4592        }
4593
4594        spin_unlock(&memcg_oom_lock);
4595}
4596
4597static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4598        struct cftype *cft,  struct cgroup_map_cb *cb)
4599{
4600        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4601
4602        cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
4603
4604        if (atomic_read(&memcg->under_oom))
4605                cb->fill(cb, "under_oom", 1);
4606        else
4607                cb->fill(cb, "under_oom", 0);
4608        return 0;
4609}
4610
4611static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4612        struct cftype *cft, u64 val)
4613{
4614        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4615        struct mem_cgroup *parent;
4616
4617        /* cannot set to root cgroup and only 0 and 1 are allowed */
4618        if (!cgrp->parent || !((val == 0) || (val == 1)))
4619                return -EINVAL;
4620
4621        parent = mem_cgroup_from_cont(cgrp->parent);
4622
4623        cgroup_lock();
4624        /* oom-kill-disable is a flag for subhierarchy. */
4625        if ((parent->use_hierarchy) ||
4626            (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4627                cgroup_unlock();
4628                return -EINVAL;
4629        }
4630        memcg->oom_kill_disable = val;
4631        if (!val)
4632                memcg_oom_recover(memcg);
4633        cgroup_unlock();
4634        return 0;
4635}
4636
4637#ifdef CONFIG_MEMCG_KMEM
4638static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
4639{
4640        return mem_cgroup_sockets_init(memcg, ss);
4641};
4642
4643static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
4644{
4645        mem_cgroup_sockets_destroy(memcg);
4646}
4647#else
4648static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
4649{
4650        return 0;
4651}
4652
4653static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
4654{
4655}
4656#endif
4657
4658static struct cftype mem_cgroup_files[] = {
4659        {
4660                .name = "usage_in_bytes",
4661                .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
4662                .read = mem_cgroup_read,
4663                .register_event = mem_cgroup_usage_register_event,
4664                .unregister_event = mem_cgroup_usage_unregister_event,
4665        },
4666        {
4667                .name = "max_usage_in_bytes",
4668                .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
4669                .trigger = mem_cgroup_reset,
4670                .read = mem_cgroup_read,
4671        },
4672        {
4673                .name = "limit_in_bytes",
4674                .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
4675                .write_string = mem_cgroup_write,
4676                .read = mem_cgroup_read,
4677        },
4678        {
4679                .name = "soft_limit_in_bytes",
4680                .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4681                .write_string = mem_cgroup_write,
4682                .read = mem_cgroup_read,
4683        },
4684        {
4685                .name = "failcnt",
4686                .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
4687                .trigger = mem_cgroup_reset,
4688                .read = mem_cgroup_read,
4689        },
4690        {
4691                .name = "stat",
4692                .read_seq_string = memcg_stat_show,
4693        },
4694        {
4695                .name = "force_empty",
4696                .trigger = mem_cgroup_force_empty_write,
4697        },
4698        {
4699                .name = "use_hierarchy",
4700                .write_u64 = mem_cgroup_hierarchy_write,
4701                .read_u64 = mem_cgroup_hierarchy_read,
4702        },
4703        {
4704                .name = "swappiness",
4705                .read_u64 = mem_cgroup_swappiness_read,
4706                .write_u64 = mem_cgroup_swappiness_write,
4707        },
4708        {
4709                .name = "move_charge_at_immigrate",
4710                .read_u64 = mem_cgroup_move_charge_read,
4711                .write_u64 = mem_cgroup_move_charge_write,
4712        },
4713        {
4714                .name = "oom_control",
4715                .read_map = mem_cgroup_oom_control_read,
4716                .write_u64 = mem_cgroup_oom_control_write,
4717                .register_event = mem_cgroup_oom_register_event,
4718                .unregister_event = mem_cgroup_oom_unregister_event,
4719                .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4720        },
4721#ifdef CONFIG_NUMA
4722        {
4723                .name = "numa_stat",
4724                .read_seq_string = memcg_numa_stat_show,
4725        },
4726#endif
4727#ifdef CONFIG_MEMCG_SWAP
4728        {
4729                .name = "memsw.usage_in_bytes",
4730                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
4731                .read = mem_cgroup_read,
4732                .register_event = mem_cgroup_usage_register_event,
4733                .unregister_event = mem_cgroup_usage_unregister_event,
4734        },
4735        {
4736                .name = "memsw.max_usage_in_bytes",
4737                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4738                .trigger = mem_cgroup_reset,
4739                .read = mem_cgroup_read,
4740        },
4741        {
4742                .name = "memsw.limit_in_bytes",
4743                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4744                .write_string = mem_cgroup_write,
4745                .read = mem_cgroup_read,
4746        },
4747        {
4748                .name = "memsw.failcnt",
4749                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4750                .trigger = mem_cgroup_reset,
4751                .read = mem_cgroup_read,
4752        },
4753#endif
4754        { },    /* terminate */
4755};
4756
4757static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
4758{
4759        struct mem_cgroup_per_node *pn;
4760        struct mem_cgroup_per_zone *mz;
4761        int zone, tmp = node;
4762        /*
4763         * This routine is called against possible nodes.
4764         * But it's BUG to call kmalloc() against offline node.
4765         *
4766         * TODO: this routine can waste much memory for nodes which will
4767         *       never be onlined. It's better to use memory hotplug callback
4768         *       function.
4769         */
4770        if (!node_state(node, N_NORMAL_MEMORY))
4771                tmp = -1;
4772        pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
4773        if (!pn)
4774                return 1;
4775
4776        for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4777                mz = &pn->zoneinfo[zone];
4778                lruvec_init(&mz->lruvec);
4779                mz->usage_in_excess = 0;
4780                mz->on_tree = false;
4781                mz->memcg = memcg;
4782        }
4783        memcg->info.nodeinfo[node] = pn;
4784        return 0;
4785}
4786
4787static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
4788{
4789        kfree(memcg->info.nodeinfo[node]);
4790}
4791
4792static struct mem_cgroup *mem_cgroup_alloc(void)
4793{
4794        struct mem_cgroup *memcg;
4795        int size = sizeof(struct mem_cgroup);
4796
4797        /* Can be very big if MAX_NUMNODES is very big */
4798        if (size < PAGE_SIZE)
4799                memcg = kzalloc(size, GFP_KERNEL);
4800        else
4801                memcg = vzalloc(size);
4802
4803        if (!memcg)
4804                return NULL;
4805
4806        memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4807        if (!memcg->stat)
4808                goto out_free;
4809        spin_lock_init(&memcg->pcp_counter_lock);
4810        return memcg;
4811
4812out_free:
4813        if (size < PAGE_SIZE)
4814                kfree(memcg);
4815        else
4816                vfree(memcg);
4817        return NULL;
4818}
4819
4820/*
4821 * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU,
4822 * but in process context.  The work_freeing structure is overlaid
4823 * on the rcu_freeing structure, which itself is overlaid on memsw.
4824 */
4825static void free_work(struct work_struct *work)
4826{
4827        struct mem_cgroup *memcg;
4828        int size = sizeof(struct mem_cgroup);
4829
4830        memcg = container_of(work, struct mem_cgroup, work_freeing);
4831        /*
4832         * We need to make sure that (at least for now), the jump label
4833         * destruction code runs outside of the cgroup lock. This is because
4834         * get_online_cpus(), which is called from the static_branch update,
4835         * can't be called inside the cgroup_lock. cpusets are the ones
4836         * enforcing this dependency, so if they ever change, we might as well.
4837         *
4838         * schedule_work() will guarantee this happens. Be careful if you need
4839         * to move this code around, and make sure it is outside
4840         * the cgroup_lock.
4841         */
4842        disarm_sock_keys(memcg);
4843        if (size < PAGE_SIZE)
4844                kfree(memcg);
4845        else
4846                vfree(memcg);
4847}
4848
4849static void free_rcu(struct rcu_head *rcu_head)
4850{
4851        struct mem_cgroup *memcg;
4852
4853        memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing);
4854        INIT_WORK(&memcg->work_freeing, free_work);
4855        schedule_work(&memcg->work_freeing);
4856}
4857
4858/*
4859 * At destroying mem_cgroup, references from swap_cgroup can remain.
4860 * (scanning all at force_empty is too costly...)
4861 *
4862 * Instead of clearing all references at force_empty, we remember
4863 * the number of reference from swap_cgroup and free mem_cgroup when
4864 * it goes down to 0.
4865 *
4866 * Removal of cgroup itself succeeds regardless of refs from swap.
4867 */
4868
4869static void __mem_cgroup_free(struct mem_cgroup *memcg)
4870{
4871        int node;
4872
4873        mem_cgroup_remove_from_trees(memcg);
4874        free_css_id(&mem_cgroup_subsys, &memcg->css);
4875
4876        for_each_node(node)
4877                free_mem_cgroup_per_zone_info(memcg, node);
4878
4879        free_percpu(memcg->stat);
4880        call_rcu(&memcg->rcu_freeing, free_rcu);
4881}
4882
4883static void mem_cgroup_get(struct mem_cgroup *memcg)
4884{
4885        atomic_inc(&memcg->refcnt);
4886}
4887
4888static void __mem_cgroup_put(struct mem_cgroup *memcg, int count)
4889{
4890        if (atomic_sub_and_test(count, &memcg->refcnt)) {
4891                struct mem_cgroup *parent = parent_mem_cgroup(memcg);
4892                __mem_cgroup_free(memcg);
4893                if (parent)
4894                        mem_cgroup_put(parent);
4895        }
4896}
4897
4898static void mem_cgroup_put(struct mem_cgroup *memcg)
4899{
4900        __mem_cgroup_put(memcg, 1);
4901}
4902
4903/*
4904 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4905 */
4906struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
4907{
4908        if (!memcg->res.parent)
4909                return NULL;
4910        return mem_cgroup_from_res_counter(memcg->res.parent, res);
4911}
4912EXPORT_SYMBOL(parent_mem_cgroup);
4913
4914#ifdef CONFIG_MEMCG_SWAP
4915static void __init enable_swap_cgroup(void)
4916{
4917        if (!mem_cgroup_disabled() && really_do_swap_account)
4918                do_swap_account = 1;
4919}
4920#else
4921static void __init enable_swap_cgroup(void)
4922{
4923}
4924#endif
4925
4926static int mem_cgroup_soft_limit_tree_init(void)
4927{
4928        struct mem_cgroup_tree_per_node *rtpn;
4929        struct mem_cgroup_tree_per_zone *rtpz;
4930        int tmp, node, zone;
4931
4932        for_each_node(node) {
4933                tmp = node;
4934                if (!node_state(node, N_NORMAL_MEMORY))
4935                        tmp = -1;
4936                rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4937                if (!rtpn)
4938                        goto err_cleanup;
4939
4940                soft_limit_tree.rb_tree_per_node[node] = rtpn;
4941
4942                for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4943                        rtpz = &rtpn->rb_tree_per_zone[zone];
4944                        rtpz->rb_root = RB_ROOT;
4945                        spin_lock_init(&rtpz->lock);
4946                }
4947        }
4948        return 0;
4949
4950err_cleanup:
4951        for_each_node(node) {
4952                if (!soft_limit_tree.rb_tree_per_node[node])
4953                        break;
4954                kfree(soft_limit_tree.rb_tree_per_node[node]);
4955                soft_limit_tree.rb_tree_per_node[node] = NULL;
4956        }
4957        return 1;
4958
4959}
4960
4961static struct cgroup_subsys_state * __ref
4962mem_cgroup_create(struct cgroup *cont)
4963{
4964        struct mem_cgroup *memcg, *parent;
4965        long error = -ENOMEM;
4966        int node;
4967
4968        memcg = mem_cgroup_alloc();
4969        if (!memcg)
4970                return ERR_PTR(error);
4971
4972        for_each_node(node)
4973                if (alloc_mem_cgroup_per_zone_info(memcg, node))
4974                        goto free_out;
4975
4976        /* root ? */
4977        if (cont->parent == NULL) {
4978                int cpu;
4979                enable_swap_cgroup();
4980                parent = NULL;
4981                if (mem_cgroup_soft_limit_tree_init())
4982                        goto free_out;
4983                root_mem_cgroup = memcg;
4984                for_each_possible_cpu(cpu) {
4985                        struct memcg_stock_pcp *stock =
4986                                                &per_cpu(memcg_stock, cpu);
4987                        INIT_WORK(&stock->work, drain_local_stock);
4988                }
4989                hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
4990        } else {
4991                parent = mem_cgroup_from_cont(cont->parent);
4992                memcg->use_hierarchy = parent->use_hierarchy;
4993                memcg->oom_kill_disable = parent->oom_kill_disable;
4994        }
4995
4996        if (parent && parent->use_hierarchy) {
4997                res_counter_init(&memcg->res, &parent->res);
4998                res_counter_init(&memcg->memsw, &parent->memsw);
4999                /*
5000                 * We increment refcnt of the parent to ensure that we can
5001                 * safely access it on res_counter_charge/uncharge.
5002                 * This refcnt will be decremented when freeing this
5003                 * mem_cgroup(see mem_cgroup_put).
5004                 */
5005                mem_cgroup_get(parent);
5006        } else {
5007                res_counter_init(&memcg->res, NULL);
5008                res_counter_init(&memcg->memsw, NULL);
5009        }
5010        memcg->last_scanned_node = MAX_NUMNODES;
5011        INIT_LIST_HEAD(&memcg->oom_notify);
5012
5013        if (parent)
5014                memcg->swappiness = mem_cgroup_swappiness(parent);
5015        atomic_set(&memcg->refcnt, 1);
5016        memcg->move_charge_at_immigrate = 0;
5017        mutex_init(&memcg->thresholds_lock);
5018        spin_lock_init(&memcg->move_lock);
5019
5020        error = memcg_init_kmem(memcg, &mem_cgroup_subsys);
5021        if (error) {
5022                /*
5023                 * We call put now because our (and parent's) refcnts
5024                 * are already in place. mem_cgroup_put() will internally
5025                 * call __mem_cgroup_free, so return directly
5026                 */
5027                mem_cgroup_put(memcg);
5028                return ERR_PTR(error);
5029        }
5030        return &memcg->css;
5031free_out:
5032        __mem_cgroup_free(memcg);
5033        return ERR_PTR(error);
5034}
5035
5036static int mem_cgroup_pre_destroy(struct cgroup *cont)
5037{
5038        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
5039
5040        return mem_cgroup_force_empty(memcg, false);
5041}
5042
5043static void mem_cgroup_destroy(struct cgroup *cont)
5044{
5045        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
5046
5047        kmem_cgroup_destroy(memcg);
5048
5049        mem_cgroup_put(memcg);
5050}
5051
5052#ifdef CONFIG_MMU
5053/* Handlers for move charge at task migration. */
5054#define PRECHARGE_COUNT_AT_ONCE 256
5055static int mem_cgroup_do_precharge(unsigned long count)
5056{
5057        int ret = 0;
5058        int batch_count = PRECHARGE_COUNT_AT_ONCE;
5059        struct mem_cgroup *memcg = mc.to;
5060
5061        if (mem_cgroup_is_root(memcg)) {
5062                mc.precharge += count;
5063                /* we don't need css_get for root */
5064                return ret;
5065        }
5066        /* try to charge at once */
5067        if (count > 1) {
5068                struct res_counter *dummy;
5069                /*
5070                 * "memcg" cannot be under rmdir() because we've already checked
5071                 * by cgroup_lock_live_cgroup() that it is not removed and we
5072                 * are still under the same cgroup_mutex. So we can postpone
5073                 * css_get().
5074                 */
5075                if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
5076                        goto one_by_one;
5077                if (do_swap_account && res_counter_charge(&memcg->memsw,
5078                                                PAGE_SIZE * count, &dummy)) {
5079                        res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
5080                        goto one_by_one;
5081                }
5082                mc.precharge += count;
5083                return ret;
5084        }
5085one_by_one:
5086        /* fall back to one by one charge */
5087        while (count--) {
5088                if (signal_pending(current)) {
5089                        ret = -EINTR;
5090                        break;
5091                }
5092                if (!batch_count--) {
5093                        batch_count = PRECHARGE_COUNT_AT_ONCE;
5094                        cond_resched();
5095                }
5096                ret = __mem_cgroup_try_charge(NULL,
5097                                        GFP_KERNEL, 1, &memcg, false);
5098                if (ret)
5099                        /* mem_cgroup_clear_mc() will do uncharge later */
5100                        return ret;
5101                mc.precharge++;
5102        }
5103        return ret;
5104}
5105
5106/**
5107 * get_mctgt_type - get target type of moving charge
5108 * @vma: the vma the pte to be checked belongs
5109 * @addr: the address corresponding to the pte to be checked
5110 * @ptent: the pte to be checked
5111 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5112 *
5113 * Returns
5114 *   0(MC_TARGET_NONE): if the pte is not a target for move charge.
5115 *   1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5116 *     move charge. if @target is not NULL, the page is stored in target->page
5117 *     with extra refcnt got(Callers should handle it).
5118 *   2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5119 *     target for charge migration. if @target is not NULL, the entry is stored
5120 *     in target->ent.
5121 *
5122 * Called with pte lock held.
5123 */
5124union mc_target {
5125        struct page     *page;
5126        swp_entry_t     ent;
5127};
5128
5129enum mc_target_type {
5130        MC_TARGET_NONE = 0,
5131        MC_TARGET_PAGE,
5132        MC_TARGET_SWAP,
5133};
5134
5135static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5136                                                unsigned long addr, pte_t ptent)
5137{
5138        struct page *page = vm_normal_page(vma, addr, ptent);
5139
5140        if (!page || !page_mapped(page))
5141                return NULL;
5142        if (PageAnon(page)) {
5143                /* we don't move shared anon */
5144                if (!move_anon())
5145                        return NULL;
5146        } else if (!move_file())
5147                /* we ignore mapcount for file pages */
5148                return NULL;
5149        if (!get_page_unless_zero(page))
5150                return NULL;
5151
5152        return page;
5153}
5154
5155#ifdef CONFIG_SWAP
5156static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5157                        unsigned long addr, pte_t ptent, swp_entry_t *entry)
5158{
5159        struct page *page = NULL;
5160        swp_entry_t ent = pte_to_swp_entry(ptent);
5161
5162        if (!move_anon() || non_swap_entry(ent))
5163                return NULL;
5164        /*
5165         * Because lookup_swap_cache() updates some statistics counter,
5166         * we call find_get_page() with swapper_space directly.
5167         */
5168        page = find_get_page(&swapper_space, ent.val);
5169        if (do_swap_account)
5170                entry->val = ent.val;
5171
5172        return page;
5173}
5174#else
5175static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5176                        unsigned long addr, pte_t ptent, swp_entry_t *entry)
5177{
5178        return NULL;
5179}
5180#endif
5181
5182static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5183                        unsigned long addr, pte_t ptent, swp_entry_t *entry)
5184{
5185        struct page *page = NULL;
5186        struct address_space *mapping;
5187        pgoff_t pgoff;
5188
5189        if (!vma->vm_file) /* anonymous vma */
5190                return NULL;
5191        if (!move_file())
5192                return NULL;
5193
5194        mapping = vma->vm_file->f_mapping;
5195        if (pte_none(ptent))
5196                pgoff = linear_page_index(vma, addr);
5197        else /* pte_file(ptent) is true */
5198                pgoff = pte_to_pgoff(ptent);
5199
5200        /* page is moved even if it's not RSS of this task(page-faulted). */
5201        page = find_get_page(mapping, pgoff);
5202
5203#ifdef CONFIG_SWAP
5204        /* shmem/tmpfs may report page out on swap: account for that too. */
5205        if (radix_tree_exceptional_entry(page)) {
5206                swp_entry_t swap = radix_to_swp_entry(page);
5207                if (do_swap_account)
5208                        *entry = swap;
5209                page = find_get_page(&swapper_space, swap.val);
5210        }
5211#endif
5212        return page;
5213}
5214
5215static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
5216                unsigned long addr, pte_t ptent, union mc_target *target)
5217{
5218        struct page *page = NULL;
5219        struct page_cgroup *pc;
5220        enum mc_target_type ret = MC_TARGET_NONE;
5221        swp_entry_t ent = { .val = 0 };
5222
5223        if (pte_present(ptent))
5224                page = mc_handle_present_pte(vma, addr, ptent);
5225        else if (is_swap_pte(ptent))
5226                page = mc_handle_swap_pte(vma, addr, ptent, &ent);
5227        else if (pte_none(ptent) || pte_file(ptent))
5228                page = mc_handle_file_pte(vma, addr, ptent, &ent);
5229
5230        if (!page && !ent.val)
5231                return ret;
5232        if (page) {
5233                pc = lookup_page_cgroup(page);
5234                /*
5235                 * Do only loose check w/o page_cgroup lock.
5236                 * mem_cgroup_move_account() checks the pc is valid or not under
5237                 * the lock.
5238                 */
5239                if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5240                        ret = MC_TARGET_PAGE;
5241                        if (target)
5242                                target->page = page;
5243                }
5244                if (!ret || !target)
5245                        put_page(page);
5246        }
5247        /* There is a swap entry and a page doesn't exist or isn't charged */
5248        if (ent.val && !ret &&
5249                        css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
5250                ret = MC_TARGET_SWAP;
5251                if (target)
5252                        target->ent = ent;
5253        }
5254        return ret;
5255}
5256
5257#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5258/*
5259 * We don't consider swapping or file mapped pages because THP does not
5260 * support them for now.
5261 * Caller should make sure that pmd_trans_huge(pmd) is true.
5262 */
5263static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5264                unsigned long addr, pmd_t pmd, union mc_target *target)
5265{
5266        struct page *page = NULL;
5267        struct page_cgroup *pc;
5268        enum mc_target_type ret = MC_TARGET_NONE;
5269
5270        page = pmd_page(pmd);
5271        VM_BUG_ON(!page || !PageHead(page));
5272        if (!move_anon())
5273                return ret;
5274        pc = lookup_page_cgroup(page);
5275        if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5276                ret = MC_TARGET_PAGE;
5277                if (target) {
5278                        get_page(page);
5279                        target->page = page;
5280                }
5281        }
5282        return ret;
5283}
5284#else
5285static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5286                unsigned long addr, pmd_t pmd, union mc_target *target)
5287{
5288        return MC_TARGET_NONE;
5289}
5290#endif
5291
5292static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5293                                        unsigned long addr, unsigned long end,
5294                                        struct mm_walk *walk)
5295{
5296        struct vm_area_struct *vma = walk->private;
5297        pte_t *pte;
5298        spinlock_t *ptl;
5299
5300        if (pmd_trans_huge_lock(pmd, vma) == 1) {
5301                if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
5302                        mc.precharge += HPAGE_PMD_NR;
5303                spin_unlock(&vma->vm_mm->page_table_lock);
5304                return 0;
5305        }
5306
5307        if (pmd_trans_unstable(pmd))
5308                return 0;
5309        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5310        for (; addr != end; pte++, addr += PAGE_SIZE)
5311                if (get_mctgt_type(vma, addr, *pte, NULL))
5312                        mc.precharge++; /* increment precharge temporarily */
5313        pte_unmap_unlock(pte - 1, ptl);
5314        cond_resched();
5315
5316        return 0;
5317}
5318
5319static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5320{
5321        unsigned long precharge;
5322        struct vm_area_struct *vma;
5323
5324        down_read(&mm->mmap_sem);
5325        for (vma = mm->mmap; vma; vma = vma->vm_next) {
5326                struct mm_walk mem_cgroup_count_precharge_walk = {
5327                        .pmd_entry = mem_cgroup_count_precharge_pte_range,
5328                        .mm = mm,
5329                        .private = vma,
5330                };
5331                if (is_vm_hugetlb_page(vma))
5332                        continue;
5333                walk_page_range(vma->vm_start, vma->vm_end,
5334                                        &mem_cgroup_count_precharge_walk);
5335        }
5336        up_read(&mm->mmap_sem);
5337
5338        precharge = mc.precharge;
5339        mc.precharge = 0;
5340
5341        return precharge;
5342}
5343
5344static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5345{
5346        unsigned long precharge = mem_cgroup_count_precharge(mm);
5347
5348        VM_BUG_ON(mc.moving_task);
5349        mc.moving_task = current;
5350        return mem_cgroup_do_precharge(precharge);
5351}
5352
5353/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5354static void __mem_cgroup_clear_mc(void)
5355{
5356        struct mem_cgroup *from = mc.from;
5357        struct mem_cgroup *to = mc.to;
5358
5359        /* we must uncharge all the leftover precharges from mc.to */
5360        if (mc.precharge) {
5361                __mem_cgroup_cancel_charge(mc.to, mc.precharge);
5362                mc.precharge = 0;
5363        }
5364        /*
5365         * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5366         * we must uncharge here.
5367         */
5368        if (mc.moved_charge) {
5369                __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
5370                mc.moved_charge = 0;
5371        }
5372        /* we must fixup refcnts and charges */
5373        if (mc.moved_swap) {
5374                /* uncharge swap account from the old cgroup */
5375                if (!mem_cgroup_is_root(mc.from))
5376                        res_counter_uncharge(&mc.from->memsw,
5377                                                PAGE_SIZE * mc.moved_swap);
5378                __mem_cgroup_put(mc.from, mc.moved_swap);
5379
5380                if (!mem_cgroup_is_root(mc.to)) {
5381                        /*
5382                         * we charged both to->res and to->memsw, so we should
5383                         * uncharge to->res.
5384                         */
5385                        res_counter_uncharge(&mc.to->res,
5386                                                PAGE_SIZE * mc.moved_swap);
5387                }
5388                /* we've already done mem_cgroup_get(mc.to) */
5389                mc.moved_swap = 0;
5390        }
5391        memcg_oom_recover(from);
5392        memcg_oom_recover(to);
5393        wake_up_all(&mc.waitq);
5394}
5395
5396static void mem_cgroup_clear_mc(void)
5397{
5398        struct mem_cgroup *from = mc.from;
5399
5400        /*
5401         * we must clear moving_task before waking up waiters at the end of
5402         * task migration.
5403         */
5404        mc.moving_task = NULL;
5405        __mem_cgroup_clear_mc();
5406        spin_lock(&mc.lock);
5407        mc.from = NULL;
5408        mc.to = NULL;
5409        spin_unlock(&mc.lock);
5410        mem_cgroup_end_move(from);
5411}
5412
5413static int mem_cgroup_can_attach(struct cgroup *cgroup,
5414                                 struct cgroup_taskset *tset)
5415{
5416        struct task_struct *p = cgroup_taskset_first(tset);
5417        int ret = 0;
5418        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
5419
5420        if (memcg->move_charge_at_immigrate) {
5421                struct mm_struct *mm;
5422                struct mem_cgroup *from = mem_cgroup_from_task(p);
5423
5424                VM_BUG_ON(from == memcg);
5425
5426                mm = get_task_mm(p);
5427                if (!mm)
5428                        return 0;
5429                /* We move charges only when we move a owner of the mm */
5430                if (mm->owner == p) {
5431                        VM_BUG_ON(mc.from);
5432                        VM_BUG_ON(mc.to);
5433                        VM_BUG_ON(mc.precharge);
5434                        VM_BUG_ON(mc.moved_charge);
5435                        VM_BUG_ON(mc.moved_swap);
5436                        mem_cgroup_start_move(from);
5437                        spin_lock(&mc.lock);
5438                        mc.from = from;
5439                        mc.to = memcg;
5440                        spin_unlock(&mc.lock);
5441                        /* We set mc.moving_task later */
5442
5443                        ret = mem_cgroup_precharge_mc(mm);
5444                        if (ret)
5445                                mem_cgroup_clear_mc();
5446                }
5447                mmput(mm);
5448        }
5449        return ret;
5450}
5451
5452static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5453                                     struct cgroup_taskset *tset)
5454{
5455        mem_cgroup_clear_mc();
5456}
5457
5458static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5459                                unsigned long addr, unsigned long end,
5460                                struct mm_walk *walk)
5461{
5462        int ret = 0;
5463        struct vm_area_struct *vma = walk->private;
5464        pte_t *pte;
5465        spinlock_t *ptl;
5466        enum mc_target_type target_type;
5467        union mc_target target;
5468        struct page *page;
5469        struct page_cgroup *pc;
5470
5471        /*
5472         * We don't take compound_lock() here but no race with splitting thp
5473         * happens because:
5474         *  - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5475         *    under splitting, which means there's no concurrent thp split,
5476         *  - if another thread runs into split_huge_page() just after we
5477         *    entered this if-block, the thread must wait for page table lock
5478         *    to be unlocked in __split_huge_page_splitting(), where the main
5479         *    part of thp split is not executed yet.
5480         */
5481        if (pmd_trans_huge_lock(pmd, vma) == 1) {
5482                if (mc.precharge < HPAGE_PMD_NR) {
5483                        spin_unlock(&vma->vm_mm->page_table_lock);
5484                        return 0;
5485                }
5486                target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
5487                if (target_type == MC_TARGET_PAGE) {
5488                        page = target.page;
5489                        if (!isolate_lru_page(page)) {
5490                                pc = lookup_page_cgroup(page);
5491                                if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
5492                                                        pc, mc.from, mc.to)) {
5493                                        mc.precharge -= HPAGE_PMD_NR;
5494                                        mc.moved_charge += HPAGE_PMD_NR;
5495                                }
5496                                putback_lru_page(page);
5497                        }
5498                        put_page(page);
5499                }
5500                spin_unlock(&vma->vm_mm->page_table_lock);
5501                return 0;
5502        }
5503
5504        if (pmd_trans_unstable(pmd))
5505                return 0;
5506retry:
5507        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5508        for (; addr != end; addr += PAGE_SIZE) {
5509                pte_t ptent = *(pte++);
5510                swp_entry_t ent;
5511
5512                if (!mc.precharge)
5513                        break;
5514
5515                switch (get_mctgt_type(vma, addr, ptent, &target)) {
5516                case MC_TARGET_PAGE:
5517                        page = target.page;
5518                        if (isolate_lru_page(page))
5519                                goto put;
5520                        pc = lookup_page_cgroup(page);
5521                        if (!mem_cgroup_move_account(page, 1, pc,
5522                                                     mc.from, mc.to)) {
5523                                mc.precharge--;
5524                                /* we uncharge from mc.from later. */
5525                                mc.moved_charge++;
5526                        }
5527                        putback_lru_page(page);
5528put:                    /* get_mctgt_type() gets the page */
5529                        put_page(page);
5530                        break;
5531                case MC_TARGET_SWAP:
5532                        ent = target.ent;
5533                        if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
5534                                mc.precharge--;
5535                                /* we fixup refcnts and charges later. */
5536                                mc.moved_swap++;
5537                        }
5538                        break;
5539                default:
5540                        break;
5541                }
5542        }
5543        pte_unmap_unlock(pte - 1, ptl);
5544        cond_resched();
5545
5546        if (addr != end) {
5547                /*
5548                 * We have consumed all precharges we got in can_attach().
5549                 * We try charge one by one, but don't do any additional
5550                 * charges to mc.to if we have failed in charge once in attach()
5551                 * phase.
5552                 */
5553                ret = mem_cgroup_do_precharge(1);
5554                if (!ret)
5555                        goto retry;
5556        }
5557
5558        return ret;
5559}
5560
5561static void mem_cgroup_move_charge(struct mm_struct *mm)
5562{
5563        struct vm_area_struct *vma;
5564
5565        lru_add_drain_all();
5566retry:
5567        if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5568                /*
5569                 * Someone who are holding the mmap_sem might be waiting in
5570                 * waitq. So we cancel all extra charges, wake up all waiters,
5571                 * and retry. Because we cancel precharges, we might not be able
5572                 * to move enough charges, but moving charge is a best-effort
5573                 * feature anyway, so it wouldn't be a big problem.
5574                 */
5575                __mem_cgroup_clear_mc();
5576                cond_resched();
5577                goto retry;
5578        }
5579        for (vma = mm->mmap; vma; vma = vma->vm_next) {
5580                int ret;
5581                struct mm_walk mem_cgroup_move_charge_walk = {
5582                        .pmd_entry = mem_cgroup_move_charge_pte_range,
5583                        .mm = mm,
5584                        .private = vma,
5585                };
5586                if (is_vm_hugetlb_page(vma))
5587                        continue;
5588                ret = walk_page_range(vma->vm_start, vma->vm_end,
5589                                                &mem_cgroup_move_charge_walk);
5590                if (ret)
5591                        /*
5592                         * means we have consumed all precharges and failed in
5593                         * doing additional charge. Just abandon here.
5594                         */
5595                        break;
5596        }
5597        up_read(&mm->mmap_sem);
5598}
5599
5600static void mem_cgroup_move_task(struct cgroup *cont,
5601                                 struct cgroup_taskset *tset)
5602{
5603        struct task_struct *p = cgroup_taskset_first(tset);
5604        struct mm_struct *mm = get_task_mm(p);
5605
5606        if (mm) {
5607                if (mc.to)
5608                        mem_cgroup_move_charge(mm);
5609                mmput(mm);
5610        }
5611        if (mc.to)
5612                mem_cgroup_clear_mc();
5613}
5614#else   /* !CONFIG_MMU */
5615static int mem_cgroup_can_attach(struct cgroup *cgroup,
5616                                 struct cgroup_taskset *tset)
5617{
5618        return 0;
5619}
5620static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5621                                     struct cgroup_taskset *tset)
5622{
5623}
5624static void mem_cgroup_move_task(struct cgroup *cont,
5625                                 struct cgroup_taskset *tset)
5626{
5627}
5628#endif
5629
5630struct cgroup_subsys mem_cgroup_subsys = {
5631        .name = "memory",
5632        .subsys_id = mem_cgroup_subsys_id,
5633        .create = mem_cgroup_create,
5634        .pre_destroy = mem_cgroup_pre_destroy,
5635        .destroy = mem_cgroup_destroy,
5636        .can_attach = mem_cgroup_can_attach,
5637        .cancel_attach = mem_cgroup_cancel_attach,
5638        .attach = mem_cgroup_move_task,
5639        .base_cftypes = mem_cgroup_files,
5640        .early_init = 0,
5641        .use_id = 1,
5642        .__DEPRECATED_clear_css_refs = true,
5643};
5644
5645#ifdef CONFIG_MEMCG_SWAP
5646static int __init enable_swap_account(char *s)
5647{
5648        /* consider enabled if no parameter or 1 is given */
5649        if (!strcmp(s, "1"))
5650                really_do_swap_account = 1;
5651        else if (!strcmp(s, "0"))
5652                really_do_swap_account = 0;
5653        return 1;
5654}
5655__setup("swapaccount=", enable_swap_account);
5656
5657#endif
5658
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.