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/ip.h>
  55#include <net/tcp_memcontrol.h>
  56
  57#include <asm/uaccess.h>
  58
  59#include <trace/events/vmscan.h>
  60
  61struct cgroup_subsys mem_cgroup_subsys __read_mostly;
  62#define MEM_CGROUP_RECLAIM_RETRIES      5
  63static struct mem_cgroup *root_mem_cgroup __read_mostly;
  64
  65#ifdef CONFIG_MEMCG_SWAP
  66/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
  67int do_swap_account __read_mostly;
  68
  69/* for remember boot option*/
  70#ifdef CONFIG_MEMCG_SWAP_ENABLED
  71static int really_do_swap_account __initdata = 1;
  72#else
  73static int really_do_swap_account __initdata = 0;
  74#endif
  75
  76#else
  77#define do_swap_account         0
  78#endif
  79
  80
  81/*
  82 * Statistics for memory cgroup.
  83 */
  84enum mem_cgroup_stat_index {
  85        /*
  86         * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
  87         */
  88        MEM_CGROUP_STAT_CACHE,     /* # of pages charged as cache */
  89        MEM_CGROUP_STAT_RSS,       /* # of pages charged as anon rss */
  90        MEM_CGROUP_STAT_FILE_MAPPED,  /* # of pages charged as file rss */
  91        MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */
  92        MEM_CGROUP_STAT_NSTATS,
  93};
  94
  95static const char * const mem_cgroup_stat_names[] = {
  96        "cache",
  97        "rss",
  98        "mapped_file",
  99        "swap",
 100};
 101
 102enum mem_cgroup_events_index {
 103        MEM_CGROUP_EVENTS_PGPGIN,       /* # of pages paged in */
 104        MEM_CGROUP_EVENTS_PGPGOUT,      /* # of pages paged out */
 105        MEM_CGROUP_EVENTS_PGFAULT,      /* # of page-faults */
 106        MEM_CGROUP_EVENTS_PGMAJFAULT,   /* # of major page-faults */
 107        MEM_CGROUP_EVENTS_NSTATS,
 108};
 109
 110static const char * const mem_cgroup_events_names[] = {
 111        "pgpgin",
 112        "pgpgout",
 113        "pgfault",
 114        "pgmajfault",
 115};
 116
 117/*
 118 * Per memcg event counter is incremented at every pagein/pageout. With THP,
 119 * it will be incremated by the number of pages. This counter is used for
 120 * for trigger some periodic events. This is straightforward and better
 121 * than using jiffies etc. to handle periodic memcg event.
 122 */
 123enum mem_cgroup_events_target {
 124        MEM_CGROUP_TARGET_THRESH,
 125        MEM_CGROUP_TARGET_SOFTLIMIT,
 126        MEM_CGROUP_TARGET_NUMAINFO,
 127        MEM_CGROUP_NTARGETS,
 128};
 129#define THRESHOLDS_EVENTS_TARGET 128
 130#define SOFTLIMIT_EVENTS_TARGET 1024
 131#define NUMAINFO_EVENTS_TARGET  1024
 132
 133struct mem_cgroup_stat_cpu {
 134        long count[MEM_CGROUP_STAT_NSTATS];
 135        unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
 136        unsigned long nr_page_events;
 137        unsigned long targets[MEM_CGROUP_NTARGETS];
 138};
 139
 140struct mem_cgroup_reclaim_iter {
 141        /* css_id of the last scanned hierarchy member */
 142        int position;
 143        /* scan generation, increased every round-trip */
 144        unsigned int generation;
 145};
 146
 147/*
 148 * per-zone information in memory controller.
 149 */
 150struct mem_cgroup_per_zone {
 151        struct lruvec           lruvec;
 152        unsigned long           lru_size[NR_LRU_LISTS];
 153
 154        struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1];
 155
 156        struct rb_node          tree_node;      /* RB tree node */
 157        unsigned long long      usage_in_excess;/* Set to the value by which */
 158                                                /* the soft limit is exceeded*/
 159        bool                    on_tree;
 160        struct mem_cgroup       *memcg;         /* Back pointer, we cannot */
 161                                                /* use container_of        */
 162};
 163
 164struct mem_cgroup_per_node {
 165        struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
 166};
 167
 168struct mem_cgroup_lru_info {
 169        struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
 170};
 171
 172/*
 173 * Cgroups above their limits are maintained in a RB-Tree, independent of
 174 * their hierarchy representation
 175 */
 176
 177struct mem_cgroup_tree_per_zone {
 178        struct rb_root rb_root;
 179        spinlock_t lock;
 180};
 181
 182struct mem_cgroup_tree_per_node {
 183        struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
 184};
 185
 186struct mem_cgroup_tree {
 187        struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
 188};
 189
 190static struct mem_cgroup_tree soft_limit_tree __read_mostly;
 191
 192struct mem_cgroup_threshold {
 193        struct eventfd_ctx *eventfd;
 194        u64 threshold;
 195};
 196
 197/* For threshold */
 198struct mem_cgroup_threshold_ary {
 199        /* An array index points to threshold just below or equal to usage. */
 200        int current_threshold;
 201        /* Size of entries[] */
 202        unsigned int size;
 203        /* Array of thresholds */
 204        struct mem_cgroup_threshold entries[0];
 205};
 206
 207struct mem_cgroup_thresholds {
 208        /* Primary thresholds array */
 209        struct mem_cgroup_threshold_ary *primary;
 210        /*
 211         * Spare threshold array.
 212         * This is needed to make mem_cgroup_unregister_event() "never fail".
 213         * It must be able to store at least primary->size - 1 entries.
 214         */
 215        struct mem_cgroup_threshold_ary *spare;
 216};
 217
 218/* for OOM */
 219struct mem_cgroup_eventfd_list {
 220        struct list_head list;
 221        struct eventfd_ctx *eventfd;
 222};
 223
 224static void mem_cgroup_threshold(struct mem_cgroup *memcg);
 225static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
 226
 227/*
 228 * The memory controller data structure. The memory controller controls both
 229 * page cache and RSS per cgroup. We would eventually like to provide
 230 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
 231 * to help the administrator determine what knobs to tune.
 232 *
 233 * TODO: Add a water mark for the memory controller. Reclaim will begin when
 234 * we hit the water mark. May be even add a low water mark, such that
 235 * no reclaim occurs from a cgroup at it's low water mark, this is
 236 * a feature that will be implemented much later in the future.
 237 */
 238struct mem_cgroup {
 239        struct cgroup_subsys_state css;
 240        /*
 241         * the counter to account for memory usage
 242         */
 243        struct res_counter res;
 244
 245        union {
 246                /*
 247                 * the counter to account for mem+swap usage.
 248                 */
 249                struct res_counter memsw;
 250
 251                /*
 252                 * rcu_freeing is used only when freeing struct mem_cgroup,
 253                 * so put it into a union to avoid wasting more memory.
 254                 * It must be disjoint from the css field.  It could be
 255                 * in a union with the res field, but res plays a much
 256                 * larger part in mem_cgroup life than memsw, and might
 257                 * be of interest, even at time of free, when debugging.
 258                 * So share rcu_head with the less interesting memsw.
 259                 */
 260                struct rcu_head rcu_freeing;
 261                /*
 262                 * We also need some space for a worker in deferred freeing.
 263                 * By the time we call it, rcu_freeing is no longer in use.
 264                 */
 265                struct work_struct work_freeing;
 266        };
 267
 268        /*
 269         * Per cgroup active and inactive list, similar to the
 270         * per zone LRU lists.
 271         */
 272        struct mem_cgroup_lru_info info;
 273        int last_scanned_node;
 274#if MAX_NUMNODES > 1
 275        nodemask_t      scan_nodes;
 276        atomic_t        numainfo_events;
 277        atomic_t        numainfo_updating;
 278#endif
 279        /*
 280         * Should the accounting and control be hierarchical, per subtree?
 281         */
 282        bool use_hierarchy;
 283
 284        bool            oom_lock;
 285        atomic_t        under_oom;
 286
 287        atomic_t        refcnt;
 288
 289        int     swappiness;
 290        /* OOM-Killer disable */
 291        int             oom_kill_disable;
 292
 293        /* set when res.limit == memsw.limit */
 294        bool            memsw_is_minimum;
 295
 296        /* protect arrays of thresholds */
 297        struct mutex thresholds_lock;
 298
 299        /* thresholds for memory usage. RCU-protected */
 300        struct mem_cgroup_thresholds thresholds;
 301
 302        /* thresholds for mem+swap usage. RCU-protected */
 303        struct mem_cgroup_thresholds memsw_thresholds;
 304
 305        /* For oom notifier event fd */
 306        struct list_head oom_notify;
 307
 308        /*
 309         * Should we move charges of a task when a task is moved into this
 310         * mem_cgroup ? And what type of charges should we move ?
 311         */
 312        unsigned long   move_charge_at_immigrate;
 313        /*
 314         * set > 0 if pages under this cgroup are moving to other cgroup.
 315         */
 316        atomic_t        moving_account;
 317        /* taken only while moving_account > 0 */
 318        spinlock_t      move_lock;
 319        /*
 320         * percpu counter.
 321         */
 322        struct mem_cgroup_stat_cpu __percpu *stat;
 323        /*
 324         * used when a cpu is offlined or other synchronizations
 325         * See mem_cgroup_read_stat().
 326         */
 327        struct mem_cgroup_stat_cpu nocpu_base;
 328        spinlock_t pcp_counter_lock;
 329
 330#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
 331        struct tcp_memcontrol tcp_mem;
 332#endif
 333};
 334
 335/* Stuffs for move charges at task migration. */
 336/*
 337 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
 338 * left-shifted bitmap of these types.
 339 */
 340enum move_type {
 341        MOVE_CHARGE_TYPE_ANON,  /* private anonymous page and swap of it */
 342        MOVE_CHARGE_TYPE_FILE,  /* file page(including tmpfs) and swap of it */
 343        NR_MOVE_TYPE,
 344};
 345
 346/* "mc" and its members are protected by cgroup_mutex */
 347static struct move_charge_struct {
 348        spinlock_t        lock; /* for from, to */
 349        struct mem_cgroup *from;
 350        struct mem_cgroup *to;
 351        unsigned long precharge;
 352        unsigned long moved_charge;
 353        unsigned long moved_swap;
 354        struct task_struct *moving_task;        /* a task moving charges */
 355        wait_queue_head_t waitq;                /* a waitq for other context */
 356} mc = {
 357        .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
 358        .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
 359};
 360
 361static bool move_anon(void)
 362{
 363        return test_bit(MOVE_CHARGE_TYPE_ANON,
 364                                        &mc.to->move_charge_at_immigrate);
 365}
 366
 367static bool move_file(void)
 368{
 369        return test_bit(MOVE_CHARGE_TYPE_FILE,
 370                                        &mc.to->move_charge_at_immigrate);
 371}
 372
 373/*
 374 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
 375 * limit reclaim to prevent infinite loops, if they ever occur.
 376 */
 377#define MEM_CGROUP_MAX_RECLAIM_LOOPS            100
 378#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
 379
 380enum charge_type {
 381        MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
 382        MEM_CGROUP_CHARGE_TYPE_ANON,
 383        MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
 384        MEM_CGROUP_CHARGE_TYPE_DROP,    /* a page was unused swap cache */
 385        NR_CHARGE_TYPE,
 386};
 387
 388/* for encoding cft->private value on file */
 389#define _MEM                    (0)
 390#define _MEMSWAP                (1)
 391#define _OOM_TYPE               (2)
 392#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
 393#define MEMFILE_TYPE(val)       ((val) >> 16 & 0xffff)
 394#define MEMFILE_ATTR(val)       ((val) & 0xffff)
 395/* Used for OOM nofiier */
 396#define OOM_CONTROL             (0)
 397
 398/*
 399 * Reclaim flags for mem_cgroup_hierarchical_reclaim
 400 */
 401#define MEM_CGROUP_RECLAIM_NOSWAP_BIT   0x0
 402#define MEM_CGROUP_RECLAIM_NOSWAP       (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
 403#define MEM_CGROUP_RECLAIM_SHRINK_BIT   0x1
 404#define MEM_CGROUP_RECLAIM_SHRINK       (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
 405
 406static void mem_cgroup_get(struct mem_cgroup *memcg);
 407static void mem_cgroup_put(struct mem_cgroup *memcg);
 408
 409static inline
 410struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
 411{
 412        return container_of(s, struct mem_cgroup, css);
 413}
 414
 415static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
 416{
 417        return (memcg == root_mem_cgroup);
 418}
 419
 420/* Writing them here to avoid exposing memcg's inner layout */
 421#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
 422
 423void sock_update_memcg(struct sock *sk)
 424{
 425        if (mem_cgroup_sockets_enabled) {
 426                struct mem_cgroup *memcg;
 427                struct cg_proto *cg_proto;
 428
 429                BUG_ON(!sk->sk_prot->proto_cgroup);
 430
 431                /* Socket cloning can throw us here with sk_cgrp already
 432                 * filled. It won't however, necessarily happen from
 433                 * process context. So the test for root memcg given
 434                 * the current task's memcg won't help us in this case.
 435                 *
 436                 * Respecting the original socket's memcg is a better
 437                 * decision in this case.
 438                 */
 439                if (sk->sk_cgrp) {
 440                        BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
 441                        mem_cgroup_get(sk->sk_cgrp->memcg);
 442                        return;
 443                }
 444
 445                rcu_read_lock();
 446                memcg = mem_cgroup_from_task(current);
 447                cg_proto = sk->sk_prot->proto_cgroup(memcg);
 448                if (!mem_cgroup_is_root(memcg) && memcg_proto_active(cg_proto)) {
 449                        mem_cgroup_get(memcg);
 450                        sk->sk_cgrp = cg_proto;
 451                }
 452                rcu_read_unlock();
 453        }
 454}
 455EXPORT_SYMBOL(sock_update_memcg);
 456
 457void sock_release_memcg(struct sock *sk)
 458{
 459        if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
 460                struct mem_cgroup *memcg;
 461                WARN_ON(!sk->sk_cgrp->memcg);
 462                memcg = sk->sk_cgrp->memcg;
 463                mem_cgroup_put(memcg);
 464        }
 465}
 466
 467struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
 468{
 469        if (!memcg || mem_cgroup_is_root(memcg))
 470                return NULL;
 471
 472        return &memcg->tcp_mem.cg_proto;
 473}
 474EXPORT_SYMBOL(tcp_proto_cgroup);
 475
 476static void disarm_sock_keys(struct mem_cgroup *memcg)
 477{
 478        if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto))
 479                return;
 480        static_key_slow_dec(&memcg_socket_limit_enabled);
 481}
 482#else
 483static void disarm_sock_keys(struct mem_cgroup *memcg)
 484{
 485}
 486#endif
 487
 488static void drain_all_stock_async(struct mem_cgroup *memcg);
 489
 490static struct mem_cgroup_per_zone *
 491mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid)
 492{
 493        return &memcg->info.nodeinfo[nid]->zoneinfo[zid];
 494}
 495
 496struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
 497{
 498        return &memcg->css;
 499}
 500
 501static struct mem_cgroup_per_zone *
 502page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page)
 503{
 504        int nid = page_to_nid(page);
 505        int zid = page_zonenum(page);
 506
 507        return mem_cgroup_zoneinfo(memcg, nid, zid);
 508}
 509
 510static struct mem_cgroup_tree_per_zone *
 511soft_limit_tree_node_zone(int nid, int zid)
 512{
 513        return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
 514}
 515
 516static struct mem_cgroup_tree_per_zone *
 517soft_limit_tree_from_page(struct page *page)
 518{
 519        int nid = page_to_nid(page);
 520        int zid = page_zonenum(page);
 521
 522        return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
 523}
 524
 525static void
 526__mem_cgroup_insert_exceeded(struct mem_cgroup *memcg,
 527                                struct mem_cgroup_per_zone *mz,
 528                                struct mem_cgroup_tree_per_zone *mctz,
 529                                unsigned long long new_usage_in_excess)
 530{
 531        struct rb_node **p = &mctz->rb_root.rb_node;
 532        struct rb_node *parent = NULL;
 533        struct mem_cgroup_per_zone *mz_node;
 534
 535        if (mz->on_tree)
 536                return;
 537
 538        mz->usage_in_excess = new_usage_in_excess;
 539        if (!mz->usage_in_excess)
 540                return;
 541        while (*p) {
 542                parent = *p;
 543                mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
 544                                        tree_node);
 545                if (mz->usage_in_excess < mz_node->usage_in_excess)
 546                        p = &(*p)->rb_left;
 547                /*
 548                 * We can't avoid mem cgroups that are over their soft
 549                 * limit by the same amount
 550                 */
 551                else if (mz->usage_in_excess >= mz_node->usage_in_excess)
 552                        p = &(*p)->rb_right;
 553        }
 554        rb_link_node(&mz->tree_node, parent, p);
 555        rb_insert_color(&mz->tree_node, &mctz->rb_root);
 556        mz->on_tree = true;
 557}
 558
 559static void
 560__mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
 561                                struct mem_cgroup_per_zone *mz,
 562                                struct mem_cgroup_tree_per_zone *mctz)
 563{
 564        if (!mz->on_tree)
 565                return;
 566        rb_erase(&mz->tree_node, &mctz->rb_root);
 567        mz->on_tree = false;
 568}
 569
 570static void
 571mem_cgroup_remove_exceeded(struct mem_cgroup *memcg,
 572                                struct mem_cgroup_per_zone *mz,
 573                                struct mem_cgroup_tree_per_zone *mctz)
 574{
 575        spin_lock(&mctz->lock);
 576        __mem_cgroup_remove_exceeded(memcg, mz, mctz);
 577        spin_unlock(&mctz->lock);
 578}
 579
 580
 581static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
 582{
 583        unsigned long long excess;
 584        struct mem_cgroup_per_zone *mz;
 585        struct mem_cgroup_tree_per_zone *mctz;
 586        int nid = page_to_nid(page);
 587        int zid = page_zonenum(page);
 588        mctz = soft_limit_tree_from_page(page);
 589
 590        /*
 591         * Necessary to update all ancestors when hierarchy is used.
 592         * because their event counter is not touched.
 593         */
 594        for (; memcg; memcg = parent_mem_cgroup(memcg)) {
 595                mz = mem_cgroup_zoneinfo(memcg, nid, zid);
 596                excess = res_counter_soft_limit_excess(&memcg->res);
 597                /*
 598                 * We have to update the tree if mz is on RB-tree or
 599                 * mem is over its softlimit.
 600                 */
 601                if (excess || mz->on_tree) {
 602                        spin_lock(&mctz->lock);
 603                        /* if on-tree, remove it */
 604                        if (mz->on_tree)
 605                                __mem_cgroup_remove_exceeded(memcg, mz, mctz);
 606                        /*
 607                         * Insert again. mz->usage_in_excess will be updated.
 608                         * If excess is 0, no tree ops.
 609                         */
 610                        __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess);
 611                        spin_unlock(&mctz->lock);
 612                }
 613        }
 614}
 615
 616static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
 617{
 618        int node, zone;
 619        struct mem_cgroup_per_zone *mz;
 620        struct mem_cgroup_tree_per_zone *mctz;
 621
 622        for_each_node(node) {
 623                for (zone = 0; zone < MAX_NR_ZONES; zone++) {
 624                        mz = mem_cgroup_zoneinfo(memcg, node, zone);
 625                        mctz = soft_limit_tree_node_zone(node, zone);
 626                        mem_cgroup_remove_exceeded(memcg, mz, mctz);
 627                }
 628        }
 629}
 630
 631static struct mem_cgroup_per_zone *
 632__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
 633{
 634        struct rb_node *rightmost = NULL;
 635        struct mem_cgroup_per_zone *mz;
 636
 637retry:
 638        mz = NULL;
 639        rightmost = rb_last(&mctz->rb_root);
 640        if (!rightmost)
 641                goto done;              /* Nothing to reclaim from */
 642
 643        mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
 644        /*
 645         * Remove the node now but someone else can add it back,
 646         * we will to add it back at the end of reclaim to its correct
 647         * position in the tree.
 648         */
 649        __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
 650        if (!res_counter_soft_limit_excess(&mz->memcg->res) ||
 651                !css_tryget(&mz->memcg->css))
 652                goto retry;
 653done:
 654        return mz;
 655}
 656
 657static struct mem_cgroup_per_zone *
 658mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
 659{
 660        struct mem_cgroup_per_zone *mz;
 661
 662        spin_lock(&mctz->lock);
 663        mz = __mem_cgroup_largest_soft_limit_node(mctz);
 664        spin_unlock(&mctz->lock);
 665        return mz;
 666}
 667
 668/*
 669 * Implementation Note: reading percpu statistics for memcg.
 670 *
 671 * Both of vmstat[] and percpu_counter has threshold and do periodic
 672 * synchronization to implement "quick" read. There are trade-off between
 673 * reading cost and precision of value. Then, we may have a chance to implement
 674 * a periodic synchronizion of counter in memcg's counter.
 675 *
 676 * But this _read() function is used for user interface now. The user accounts
 677 * memory usage by memory cgroup and he _always_ requires exact value because
 678 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
 679 * have to visit all online cpus and make sum. So, for now, unnecessary
 680 * synchronization is not implemented. (just implemented for cpu hotplug)
 681 *
 682 * If there are kernel internal actions which can make use of some not-exact
 683 * value, and reading all cpu value can be performance bottleneck in some
 684 * common workload, threashold and synchonization as vmstat[] should be
 685 * implemented.
 686 */
 687static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
 688                                 enum mem_cgroup_stat_index idx)
 689{
 690        long val = 0;
 691        int cpu;
 692
 693        get_online_cpus();
 694        for_each_online_cpu(cpu)
 695                val += per_cpu(memcg->stat->count[idx], cpu);
 696#ifdef CONFIG_HOTPLUG_CPU
 697        spin_lock(&memcg->pcp_counter_lock);
 698        val += memcg->nocpu_base.count[idx];
 699        spin_unlock(&memcg->pcp_counter_lock);
 700#endif
 701        put_online_cpus();
 702        return val;
 703}
 704
 705static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
 706                                         bool charge)
 707{
 708        int val = (charge) ? 1 : -1;
 709        this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
 710}
 711
 712static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
 713                                            enum mem_cgroup_events_index idx)
 714{
 715        unsigned long val = 0;
 716        int cpu;
 717
 718        for_each_online_cpu(cpu)
 719                val += per_cpu(memcg->stat->events[idx], cpu);
 720#ifdef CONFIG_HOTPLUG_CPU
 721        spin_lock(&memcg->pcp_counter_lock);
 722        val += memcg->nocpu_base.events[idx];
 723        spin_unlock(&memcg->pcp_counter_lock);
 724#endif
 725        return val;
 726}
 727
 728static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
 729                                         bool anon, int nr_pages)
 730{
 731        preempt_disable();
 732
 733        /*
 734         * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
 735         * counted as CACHE even if it's on ANON LRU.
 736         */
 737        if (anon)
 738                __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
 739                                nr_pages);
 740        else
 741                __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
 742                                nr_pages);
 743
 744        /* pagein of a big page is an event. So, ignore page size */
 745        if (nr_pages > 0)
 746                __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
 747        else {
 748                __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
 749                nr_pages = -nr_pages; /* for event */
 750        }
 751
 752        __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
 753
 754        preempt_enable();
 755}
 756
 757unsigned long
 758mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
 759{
 760        struct mem_cgroup_per_zone *mz;
 761
 762        mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
 763        return mz->lru_size[lru];
 764}
 765
 766static unsigned long
 767mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid,
 768                        unsigned int lru_mask)
 769{
 770        struct mem_cgroup_per_zone *mz;
 771        enum lru_list lru;
 772        unsigned long ret = 0;
 773
 774        mz = mem_cgroup_zoneinfo(memcg, nid, zid);
 775
 776        for_each_lru(lru) {
 777                if (BIT(lru) & lru_mask)
 778                        ret += mz->lru_size[lru];
 779        }
 780        return ret;
 781}
 782
 783static unsigned long
 784mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
 785                        int nid, unsigned int lru_mask)
 786{
 787        u64 total = 0;
 788        int zid;
 789
 790        for (zid = 0; zid < MAX_NR_ZONES; zid++)
 791                total += mem_cgroup_zone_nr_lru_pages(memcg,
 792                                                nid, zid, lru_mask);
 793
 794        return total;
 795}
 796
 797static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
 798                        unsigned int lru_mask)
 799{
 800        int nid;
 801        u64 total = 0;
 802
 803        for_each_node_state(nid, N_HIGH_MEMORY)
 804                total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
 805        return total;
 806}
 807
 808static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
 809                                       enum mem_cgroup_events_target target)
 810{
 811        unsigned long val, next;
 812
 813        val = __this_cpu_read(memcg->stat->nr_page_events);
 814        next = __this_cpu_read(memcg->stat->targets[target]);
 815        /* from time_after() in jiffies.h */
 816        if ((long)next - (long)val < 0) {
 817                switch (target) {
 818                case MEM_CGROUP_TARGET_THRESH:
 819                        next = val + THRESHOLDS_EVENTS_TARGET;
 820                        break;
 821                case MEM_CGROUP_TARGET_SOFTLIMIT:
 822                        next = val + SOFTLIMIT_EVENTS_TARGET;
 823                        break;
 824                case MEM_CGROUP_TARGET_NUMAINFO:
 825                        next = val + NUMAINFO_EVENTS_TARGET;
 826                        break;
 827                default:
 828                        break;
 829                }
 830                __this_cpu_write(memcg->stat->targets[target], next);
 831                return true;
 832        }
 833        return false;
 834}
 835
 836/*
 837 * Check events in order.
 838 *
 839 */
 840static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
 841{
 842        preempt_disable();
 843        /* threshold event is triggered in finer grain than soft limit */
 844        if (unlikely(mem_cgroup_event_ratelimit(memcg,
 845                                                MEM_CGROUP_TARGET_THRESH))) {
 846                bool do_softlimit;
 847                bool do_numainfo __maybe_unused;
 848
 849                do_softlimit = mem_cgroup_event_ratelimit(memcg,
 850                                                MEM_CGROUP_TARGET_SOFTLIMIT);
 851#if MAX_NUMNODES > 1
 852                do_numainfo = mem_cgroup_event_ratelimit(memcg,
 853                                                MEM_CGROUP_TARGET_NUMAINFO);
 854#endif
 855                preempt_enable();
 856
 857                mem_cgroup_threshold(memcg);
 858                if (unlikely(do_softlimit))
 859                        mem_cgroup_update_tree(memcg, page);
 860#if MAX_NUMNODES > 1
 861                if (unlikely(do_numainfo))
 862                        atomic_inc(&memcg->numainfo_events);
 863#endif
 864        } else
 865                preempt_enable();
 866}
 867
 868struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
 869{
 870        return mem_cgroup_from_css(
 871                cgroup_subsys_state(cont, mem_cgroup_subsys_id));
 872}
 873
 874struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
 875{
 876        /*
 877         * mm_update_next_owner() may clear mm->owner to NULL
 878         * if it races with swapoff, page migration, etc.
 879         * So this can be called with p == NULL.
 880         */
 881        if (unlikely(!p))
 882                return NULL;
 883
 884        return mem_cgroup_from_css(task_subsys_state(p, mem_cgroup_subsys_id));
 885}
 886
 887struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
 888{
 889        struct mem_cgroup *memcg = NULL;
 890
 891        if (!mm)
 892                return NULL;
 893        /*
 894         * Because we have no locks, mm->owner's may be being moved to other
 895         * cgroup. We use css_tryget() here even if this looks
 896         * pessimistic (rather than adding locks here).
 897         */
 898        rcu_read_lock();
 899        do {
 900                memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 901                if (unlikely(!memcg))
 902                        break;
 903        } while (!css_tryget(&memcg->css));
 904        rcu_read_unlock();
 905        return memcg;
 906}
 907
 908/**
 909 * mem_cgroup_iter - iterate over memory cgroup hierarchy
 910 * @root: hierarchy root
 911 * @prev: previously returned memcg, NULL on first invocation
 912 * @reclaim: cookie for shared reclaim walks, NULL for full walks
 913 *
 914 * Returns references to children of the hierarchy below @root, or
 915 * @root itself, or %NULL after a full round-trip.
 916 *
 917 * Caller must pass the return value in @prev on subsequent
 918 * invocations for reference counting, or use mem_cgroup_iter_break()
 919 * to cancel a hierarchy walk before the round-trip is complete.
 920 *
 921 * Reclaimers can specify a zone and a priority level in @reclaim to
 922 * divide up the memcgs in the hierarchy among all concurrent
 923 * reclaimers operating on the same zone and priority.
 924 */
 925struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
 926                                   struct mem_cgroup *prev,
 927                                   struct mem_cgroup_reclaim_cookie *reclaim)
 928{
 929        struct mem_cgroup *memcg = NULL;
 930        int id = 0;
 931
 932        if (mem_cgroup_disabled())
 933                return NULL;
 934
 935        if (!root)
 936                root = root_mem_cgroup;
 937
 938        if (prev && !reclaim)
 939                id = css_id(&prev->css);
 940
 941        if (prev && prev != root)
 942                css_put(&prev->css);
 943
 944        if (!root->use_hierarchy && root != root_mem_cgroup) {
 945                if (prev)
 946                        return NULL;
 947                return root;
 948        }
 949
 950        while (!memcg) {
 951                struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
 952                struct cgroup_subsys_state *css;
 953
 954                if (reclaim) {
 955                        int nid = zone_to_nid(reclaim->zone);
 956                        int zid = zone_idx(reclaim->zone);
 957                        struct mem_cgroup_per_zone *mz;
 958
 959                        mz = mem_cgroup_zoneinfo(root, nid, zid);
 960                        iter = &mz->reclaim_iter[reclaim->priority];
 961                        if (prev && reclaim->generation != iter->generation)
 962                                return NULL;
 963                        id = iter->position;
 964                }
 965
 966                rcu_read_lock();
 967                css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id);
 968                if (css) {
 969                        if (css == &root->css || css_tryget(css))
 970                                memcg = mem_cgroup_from_css(css);
 971                } else
 972                        id = 0;
 973                rcu_read_unlock();
 974
 975                if (reclaim) {
 976                        iter->position = id;
 977                        if (!css)
 978                                iter->generation++;
 979                        else if (!prev && memcg)
 980                                reclaim->generation = iter->generation;
 981                }
 982
 983                if (prev && !css)
 984                        return NULL;
 985        }
 986        return memcg;
 987}
 988
 989/**
 990 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
 991 * @root: hierarchy root
 992 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
 993 */
 994void mem_cgroup_iter_break(struct mem_cgroup *root,
 995                           struct mem_cgroup *prev)
 996{
 997        if (!root)
 998                root = root_mem_cgroup;
 999        if (prev && prev != root)
1000                css_put(&prev->css);
1001}
1002
1003/*
1004 * Iteration constructs for visiting all cgroups (under a tree).  If
1005 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1006 * be used for reference counting.
1007 */
1008#define for_each_mem_cgroup_tree(iter, root)            \
1009        for (iter = mem_cgroup_iter(root, NULL, NULL);  \
1010             iter != NULL;                              \
1011             iter = mem_cgroup_iter(root, iter, NULL))
1012
1013#define for_each_mem_cgroup(iter)                       \
1014        for (iter = mem_cgroup_iter(NULL, NULL, NULL);  \
1015             iter != NULL;                              \
1016             iter = mem_cgroup_iter(NULL, iter, NULL))
1017
1018void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
1019{
1020        struct mem_cgroup *memcg;
1021
1022        if (!mm)
1023                return;
1024
1025        rcu_read_lock();
1026        memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1027        if (unlikely(!memcg))
1028                goto out;
1029
1030        switch (idx) {
1031        case PGFAULT:
1032                this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1033                break;
1034        case PGMAJFAULT:
1035                this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
1036                break;
1037        default:
1038                BUG();
1039        }
1040out:
1041        rcu_read_unlock();
1042}
1043EXPORT_SYMBOL(mem_cgroup_count_vm_event);
1044
1045/**
1046 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1047 * @zone: zone of the wanted lruvec
1048 * @memcg: memcg of the wanted lruvec
1049 *
1050 * Returns the lru list vector holding pages for the given @zone and
1051 * @mem.  This can be the global zone lruvec, if the memory controller
1052 * is disabled.
1053 */
1054struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1055                                      struct mem_cgroup *memcg)
1056{
1057        struct mem_cgroup_per_zone *mz;
1058        struct lruvec *lruvec;
1059
1060        if (mem_cgroup_disabled()) {
1061                lruvec = &zone->lruvec;
1062                goto out;
1063        }
1064
1065        mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone));
1066        lruvec = &mz->lruvec;
1067out:
1068        /*
1069         * Since a node can be onlined after the mem_cgroup was created,
1070         * we have to be prepared to initialize lruvec->zone here;
1071         * and if offlined then reonlined, we need to reinitialize it.
1072         */
1073        if (unlikely(lruvec->zone != zone))
1074                lruvec->zone = zone;
1075        return lruvec;
1076}
1077
1078/*
1079 * Following LRU functions are allowed to be used without PCG_LOCK.
1080 * Operations are called by routine of global LRU independently from memcg.
1081 * What we have to take care of here is validness of pc->mem_cgroup.
1082 *
1083 * Changes to pc->mem_cgroup happens when
1084 * 1. charge
1085 * 2. moving account
1086 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1087 * It is added to LRU before charge.
1088 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1089 * When moving account, the page is not on LRU. It's isolated.
1090 */
1091
1092/**
1093 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
1094 * @page: the page
1095 * @zone: zone of the page
1096 */
1097struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
1098{
1099        struct mem_cgroup_per_zone *mz;
1100        struct mem_cgroup *memcg;
1101        struct page_cgroup *pc;
1102        struct lruvec *lruvec;
1103
1104        if (mem_cgroup_disabled()) {
1105                lruvec = &zone->lruvec;
1106                goto out;
1107        }
1108
1109        pc = lookup_page_cgroup(page);
1110        memcg = pc->mem_cgroup;
1111
1112        /*
1113         * Surreptitiously switch any uncharged offlist page to root:
1114         * an uncharged page off lru does nothing to secure
1115         * its former mem_cgroup from sudden removal.
1116         *
1117         * Our caller holds lru_lock, and PageCgroupUsed is updated
1118         * under page_cgroup lock: between them, they make all uses
1119         * of pc->mem_cgroup safe.
1120         */
1121        if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup)
1122                pc->mem_cgroup = memcg = root_mem_cgroup;
1123
1124        mz = page_cgroup_zoneinfo(memcg, page);
1125        lruvec = &mz->lruvec;
1126out:
1127        /*
1128         * Since a node can be onlined after the mem_cgroup was created,
1129         * we have to be prepared to initialize lruvec->zone here;
1130         * and if offlined then reonlined, we need to reinitialize it.
1131         */
1132        if (unlikely(lruvec->zone != zone))
1133                lruvec->zone = zone;
1134        return lruvec;
1135}
1136
1137/**
1138 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1139 * @lruvec: mem_cgroup per zone lru vector
1140 * @lru: index of lru list the page is sitting on
1141 * @nr_pages: positive when adding or negative when removing
1142 *
1143 * This function must be called when a page is added to or removed from an
1144 * lru list.
1145 */
1146void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1147                                int nr_pages)
1148{
1149        struct mem_cgroup_per_zone *mz;
1150        unsigned long *lru_size;
1151
1152        if (mem_cgroup_disabled())
1153                return;
1154
1155        mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1156        lru_size = mz->lru_size + lru;
1157        *lru_size += nr_pages;
1158        VM_BUG_ON((long)(*lru_size) < 0);
1159}
1160
1161/*
1162 * Checks whether given mem is same or in the root_mem_cgroup's
1163 * hierarchy subtree
1164 */
1165bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1166                                  struct mem_cgroup *memcg)
1167{
1168        if (root_memcg == memcg)
1169                return true;
1170        if (!root_memcg->use_hierarchy || !memcg)
1171                return false;
1172        return css_is_ancestor(&memcg->css, &root_memcg->css);
1173}
1174
1175static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
1176                                       struct mem_cgroup *memcg)
1177{
1178        bool ret;
1179
1180        rcu_read_lock();
1181        ret = __mem_cgroup_same_or_subtree(root_memcg, memcg);
1182        rcu_read_unlock();
1183        return ret;
1184}
1185
1186int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg)
1187{
1188        int ret;
1189        struct mem_cgroup *curr = NULL;
1190        struct task_struct *p;
1191
1192        p = find_lock_task_mm(task);
1193        if (p) {
1194                curr = try_get_mem_cgroup_from_mm(p->mm);
1195                task_unlock(p);
1196        } else {
1197                /*
1198                 * All threads may have already detached their mm's, but the oom
1199                 * killer still needs to detect if they have already been oom
1200                 * killed to prevent needlessly killing additional tasks.
1201                 */
1202                task_lock(task);
1203                curr = mem_cgroup_from_task(task);
1204                if (curr)
1205                        css_get(&curr->css);
1206                task_unlock(task);
1207        }
1208        if (!curr)
1209                return 0;
1210        /*
1211         * We should check use_hierarchy of "memcg" not "curr". Because checking
1212         * use_hierarchy of "curr" here make this function true if hierarchy is
1213         * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1214         * hierarchy(even if use_hierarchy is disabled in "memcg").
1215         */
1216        ret = mem_cgroup_same_or_subtree(memcg, curr);
1217        css_put(&curr->css);
1218        return ret;
1219}
1220
1221int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
1222{
1223        unsigned long inactive_ratio;
1224        unsigned long inactive;
1225        unsigned long active;
1226        unsigned long gb;
1227
1228        inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1229        active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
1230
1231        gb = (inactive + active) >> (30 - PAGE_SHIFT);
1232        if (gb)
1233                inactive_ratio = int_sqrt(10 * gb);
1234        else
1235                inactive_ratio = 1;
1236
1237        return inactive * inactive_ratio < active;
1238}
1239
1240int mem_cgroup_inactive_file_is_low(struct lruvec *lruvec)
1241{
1242        unsigned long active;
1243        unsigned long inactive;
1244
1245        inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_FILE);
1246        active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_FILE);
1247
1248        return (active > inactive);
1249}
1250
1251#define mem_cgroup_from_res_counter(counter, member)    \
1252        container_of(counter, struct mem_cgroup, member)
1253
1254/**
1255 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1256 * @memcg: the memory cgroup
1257 *
1258 * Returns the maximum amount of memory @mem can be charged with, in
1259 * pages.
1260 */
1261static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
1262{
1263        unsigned long long margin;
1264
1265        margin = res_counter_margin(&memcg->res);
1266        if (do_swap_account)
1267                margin = min(margin, res_counter_margin(&memcg->memsw));
1268        return margin >> PAGE_SHIFT;
1269}
1270
1271int mem_cgroup_swappiness(struct mem_cgroup *memcg)
1272{
1273        struct cgroup *cgrp = memcg->css.cgroup;
1274
1275        /* root ? */
1276        if (cgrp->parent == NULL)
1277                return vm_swappiness;
1278
1279        return memcg->swappiness;
1280}
1281
1282/*
1283 * memcg->moving_account is used for checking possibility that some thread is
1284 * calling move_account(). When a thread on CPU-A starts moving pages under
1285 * a memcg, other threads should check memcg->moving_account under
1286 * rcu_read_lock(), like this:
1287 *
1288 *         CPU-A                                    CPU-B
1289 *                                              rcu_read_lock()
1290 *         memcg->moving_account+1              if (memcg->mocing_account)
1291 *                                                   take heavy locks.
1292 *         synchronize_rcu()                    update something.
1293 *                                              rcu_read_unlock()
1294 *         start move here.
1295 */
1296
1297/* for quick checking without looking up memcg */
1298atomic_t memcg_moving __read_mostly;
1299
1300static void mem_cgroup_start_move(struct mem_cgroup *memcg)
1301{
1302        atomic_inc(&memcg_moving);
1303        atomic_inc(&memcg->moving_account);
1304        synchronize_rcu();
1305}
1306
1307static void mem_cgroup_end_move(struct mem_cgroup *memcg)
1308{
1309        /*
1310         * Now, mem_cgroup_clear_mc() may call this function with NULL.
1311         * We check NULL in callee rather than caller.
1312         */
1313        if (memcg) {
1314                atomic_dec(&memcg_moving);
1315                atomic_dec(&memcg->moving_account);
1316        }
1317}
1318
1319/*
1320 * 2 routines for checking "mem" is under move_account() or not.
1321 *
1322 * mem_cgroup_stolen() -  checking whether a cgroup is mc.from or not. This
1323 *                        is used for avoiding races in accounting.  If true,
1324 *                        pc->mem_cgroup may be overwritten.
1325 *
1326 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1327 *                        under hierarchy of moving cgroups. This is for
1328 *                        waiting at hith-memory prressure caused by "move".
1329 */
1330
1331static bool mem_cgroup_stolen(struct mem_cgroup *memcg)
1332{
1333        VM_BUG_ON(!rcu_read_lock_held());
1334        return atomic_read(&memcg->moving_account) > 0;
1335}
1336
1337static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
1338{
1339        struct mem_cgroup *from;
1340        struct mem_cgroup *to;
1341        bool ret = false;
1342        /*
1343         * Unlike task_move routines, we access mc.to, mc.from not under
1344         * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1345         */
1346        spin_lock(&mc.lock);
1347        from = mc.from;
1348        to = mc.to;
1349        if (!from)
1350                goto unlock;
1351
1352        ret = mem_cgroup_same_or_subtree(memcg, from)
1353                || mem_cgroup_same_or_subtree(memcg, to);
1354unlock:
1355        spin_unlock(&mc.lock);
1356        return ret;
1357}
1358
1359static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
1360{
1361        if (mc.moving_task && current != mc.moving_task) {
1362                if (mem_cgroup_under_move(memcg)) {
1363                        DEFINE_WAIT(wait);
1364                        prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1365                        /* moving charge context might have finished. */
1366                        if (mc.moving_task)
1367                                schedule();
1368                        finish_wait(&mc.waitq, &wait);
1369                        return true;
1370                }
1371        }
1372        return false;
1373}
1374
1375/*
1376 * Take this lock when
1377 * - a code tries to modify page's memcg while it's USED.
1378 * - a code tries to modify page state accounting in a memcg.
1379 * see mem_cgroup_stolen(), too.
1380 */
1381static void move_lock_mem_cgroup(struct mem_cgroup *memcg,
1382                                  unsigned long *flags)
1383{
1384        spin_lock_irqsave(&memcg->move_lock, *flags);
1385}
1386
1387static void move_unlock_mem_cgroup(struct mem_cgroup *memcg,
1388                                unsigned long *flags)
1389{
1390        spin_unlock_irqrestore(&memcg->move_lock, *flags);
1391}
1392
1393/**
1394 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1395 * @memcg: The memory cgroup that went over limit
1396 * @p: Task that is going to be killed
1397 *
1398 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1399 * enabled
1400 */
1401void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1402{
1403        struct cgroup *task_cgrp;
1404        struct cgroup *mem_cgrp;
1405        /*
1406         * Need a buffer in BSS, can't rely on allocations. The code relies
1407         * on the assumption that OOM is serialized for memory controller.
1408         * If this assumption is broken, revisit this code.
1409         */
1410        static char memcg_name[PATH_MAX];
1411        int ret;
1412
1413        if (!memcg || !p)
1414                return;
1415
1416        rcu_read_lock();
1417
1418        mem_cgrp = memcg->css.cgroup;
1419        task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
1420
1421        ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
1422        if (ret < 0) {
1423                /*
1424                 * Unfortunately, we are unable to convert to a useful name
1425                 * But we'll still print out the usage information
1426                 */
1427                rcu_read_unlock();
1428                goto done;
1429        }
1430        rcu_read_unlock();
1431
1432        printk(KERN_INFO "Task in %s killed", memcg_name);
1433
1434        rcu_read_lock();
1435        ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
1436        if (ret < 0) {
1437                rcu_read_unlock();
1438                goto done;
1439        }
1440        rcu_read_unlock();
1441
1442        /*
1443         * Continues from above, so we don't need an KERN_ level
1444         */
1445        printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
1446done:
1447
1448        printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
1449                res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
1450                res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
1451                res_counter_read_u64(&memcg->res, RES_FAILCNT));
1452        printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
1453                "failcnt %llu\n",
1454                res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
1455                res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
1456                res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
1457}
1458
1459/*
1460 * This function returns the number of memcg under hierarchy tree. Returns
1461 * 1(self count) if no children.
1462 */
1463static int mem_cgroup_count_children(struct mem_cgroup *memcg)
1464{
1465        int num = 0;
1466        struct mem_cgroup *iter;
1467
1468        for_each_mem_cgroup_tree(iter, memcg)
1469                num++;
1470        return num;
1471}
1472
1473/*
1474 * Return the memory (and swap, if configured) limit for a memcg.
1475 */
1476static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
1477{
1478        u64 limit;
1479
1480        limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
1481
1482        /*
1483         * Do not consider swap space if we cannot swap due to swappiness
1484         */
1485        if (mem_cgroup_swappiness(memcg)) {
1486                u64 memsw;
1487
1488                limit += total_swap_pages << PAGE_SHIFT;
1489                memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
1490
1491                /*
1492                 * If memsw is finite and limits the amount of swap space
1493                 * available to this memcg, return that limit.
1494                 */
1495                limit = min(limit, memsw);
1496        }
1497
1498        return limit;
1499}
1500
1501void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1502                              int order)
1503{
1504        struct mem_cgroup *iter;
1505        unsigned long chosen_points = 0;
1506        unsigned long totalpages;
1507        unsigned int points = 0;
1508        struct task_struct *chosen = NULL;
1509
1510        /*
1511         * If current has a pending SIGKILL, then automatically select it.  The
1512         * goal is to allow it to allocate so that it may quickly exit and free
1513         * its memory.
1514         */
1515        if (fatal_signal_pending(current)) {
1516                set_thread_flag(TIF_MEMDIE);
1517                return;
1518        }
1519
1520        check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
1521        totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
1522        for_each_mem_cgroup_tree(iter, memcg) {
1523                struct cgroup *cgroup = iter->css.cgroup;
1524                struct cgroup_iter it;
1525                struct task_struct *task;
1526
1527                cgroup_iter_start(cgroup, &it);
1528                while ((task = cgroup_iter_next(cgroup, &it))) {
1529                        switch (oom_scan_process_thread(task, totalpages, NULL,
1530                                                        false)) {
1531                        case OOM_SCAN_SELECT:
1532                                if (chosen)
1533                                        put_task_struct(chosen);
1534                                chosen = task;
1535                                chosen_points = ULONG_MAX;
1536                                get_task_struct(chosen);
1537                                /* fall through */
1538                        case OOM_SCAN_CONTINUE:
1539                                continue;
1540                        case OOM_SCAN_ABORT:
1541                                cgroup_iter_end(cgroup, &it);
1542                                mem_cgroup_iter_break(memcg, iter);
1543                                if (chosen)
1544                                        put_task_struct(chosen);
1545                                return;
1546                        case OOM_SCAN_OK:
1547                                break;
1548                        };
1549                        points = oom_badness(task, memcg, NULL, totalpages);
1550                        if (points > chosen_points) {
1551                                if (chosen)
1552                                        put_task_struct(chosen);
1553                                chosen = task;
1554                                chosen_points = points;
1555                                get_task_struct(chosen);
1556                        }
1557                }
1558                cgroup_iter_end(cgroup, &it);
1559        }
1560
1561        if (!chosen)
1562                return;
1563        points = chosen_points * 1000 / totalpages;
1564        oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1565                         NULL, "Memory cgroup out of memory");
1566}
1567
1568static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg,
1569                                        gfp_t gfp_mask,
1570                                        unsigned long flags)
1571{
1572        unsigned long total = 0;
1573        bool noswap = false;
1574        int loop;
1575
1576        if (flags & MEM_CGROUP_RECLAIM_NOSWAP)
1577                noswap = true;
1578        if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum)
1579                noswap = true;
1580
1581        for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) {
1582                if (loop)
1583                        drain_all_stock_async(memcg);
1584                total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap);
1585                /*
1586                 * Allow limit shrinkers, which are triggered directly
1587                 * by userspace, to catch signals and stop reclaim
1588                 * after minimal progress, regardless of the margin.
1589                 */
1590                if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK))
1591                        break;
1592                if (mem_cgroup_margin(memcg))
1593                        break;
1594                /*
1595                 * If nothing was reclaimed after two attempts, there
1596                 * may be no reclaimable pages in this hierarchy.
1597                 */
1598                if (loop && !total)
1599                        break;
1600        }
1601        return total;
1602}
1603
1604/**
1605 * test_mem_cgroup_node_reclaimable
1606 * @memcg: the target memcg
1607 * @nid: the node ID to be checked.
1608 * @noswap : specify true here if the user wants flle only information.
1609 *
1610 * This function returns whether the specified memcg contains any
1611 * reclaimable pages on a node. Returns true if there are any reclaimable
1612 * pages in the node.
1613 */
1614static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
1615                int nid, bool noswap)
1616{
1617        if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
1618                return true;
1619        if (noswap || !total_swap_pages)
1620                return false;
1621        if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
1622                return true;
1623        return false;
1624
1625}
1626#if MAX_NUMNODES > 1
1627
1628/*
1629 * Always updating the nodemask is not very good - even if we have an empty
1630 * list or the wrong list here, we can start from some node and traverse all
1631 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1632 *
1633 */
1634static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
1635{
1636        int nid;
1637        /*
1638         * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1639         * pagein/pageout changes since the last update.
1640         */
1641        if (!atomic_read(&memcg->numainfo_events))
1642                return;
1643        if (atomic_inc_return(&memcg->numainfo_updating) > 1)
1644                return;
1645
1646        /* make a nodemask where this memcg uses memory from */
1647        memcg->scan_nodes = node_states[N_HIGH_MEMORY];
1648
1649        for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) {
1650
1651                if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1652                        node_clear(nid, memcg->scan_nodes);
1653        }
1654
1655        atomic_set(&memcg->numainfo_events, 0);
1656        atomic_set(&memcg->numainfo_updating, 0);
1657}
1658
1659/*
1660 * Selecting a node where we start reclaim from. Because what we need is just
1661 * reducing usage counter, start from anywhere is O,K. Considering
1662 * memory reclaim from current node, there are pros. and cons.
1663 *
1664 * Freeing memory from current node means freeing memory from a node which
1665 * we'll use or we've used. So, it may make LRU bad. And if several threads
1666 * hit limits, it will see a contention on a node. But freeing from remote
1667 * node means more costs for memory reclaim because of memory latency.
1668 *
1669 * Now, we use round-robin. Better algorithm is welcomed.
1670 */
1671int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1672{
1673        int node;
1674
1675        mem_cgroup_may_update_nodemask(memcg);
1676        node = memcg->last_scanned_node;
1677
1678        node = next_node(node, memcg->scan_nodes);
1679        if (node == MAX_NUMNODES)
1680                node = first_node(memcg->scan_nodes);
1681        /*
1682         * We call this when we hit limit, not when pages are added to LRU.
1683         * No LRU may hold pages because all pages are UNEVICTABLE or
1684         * memcg is too small and all pages are not on LRU. In that case,
1685         * we use curret node.
1686         */
1687        if (unlikely(node == MAX_NUMNODES))
1688                node = numa_node_id();
1689
1690        memcg->last_scanned_node = node;
1691        return node;
1692}
1693
1694/*
1695 * Check all nodes whether it contains reclaimable pages or not.
1696 * For quick scan, we make use of scan_nodes. This will allow us to skip
1697 * unused nodes. But scan_nodes is lazily updated and may not cotain
1698 * enough new information. We need to do double check.
1699 */
1700static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
1701{
1702        int nid;
1703
1704        /*
1705         * quick check...making use of scan_node.
1706         * We can skip unused nodes.
1707         */
1708        if (!nodes_empty(memcg->scan_nodes)) {
1709                for (nid = first_node(memcg->scan_nodes);
1710                     nid < MAX_NUMNODES;
1711                     nid = next_node(nid, memcg->scan_nodes)) {
1712
1713                        if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1714                                return true;
1715                }
1716        }
1717        /*
1718         * Check rest of nodes.
1719         */
1720        for_each_node_state(nid, N_HIGH_MEMORY) {
1721                if (node_isset(nid, memcg->scan_nodes))
1722                        continue;
1723                if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap))
1724                        return true;
1725        }
1726        return false;
1727}
1728
1729#else
1730int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1731{
1732        return 0;
1733}
1734
1735static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap)
1736{
1737        return test_mem_cgroup_node_reclaimable(memcg, 0, noswap);
1738}
1739#endif
1740
1741static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1742                                   struct zone *zone,
1743                                   gfp_t gfp_mask,
1744                                   unsigned long *total_scanned)
1745{
1746        struct mem_cgroup *victim = NULL;
1747        int total = 0;
1748        int loop = 0;
1749        unsigned long excess;
1750        unsigned long nr_scanned;
1751        struct mem_cgroup_reclaim_cookie reclaim = {
1752                .zone = zone,
1753                .priority = 0,
1754        };
1755
1756        excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT;
1757
1758        while (1) {
1759                victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1760                if (!victim) {
1761                        loop++;
1762                        if (loop >= 2) {
1763                                /*
1764                                 * If we have not been able to reclaim
1765                                 * anything, it might because there are
1766                                 * no reclaimable pages under this hierarchy
1767                                 */
1768                                if (!total)
1769                                        break;
1770                                /*
1771                                 * We want to do more targeted reclaim.
1772                                 * excess >> 2 is not to excessive so as to
1773                                 * reclaim too much, nor too less that we keep
1774                                 * coming back to reclaim from this cgroup
1775                                 */
1776                                if (total >= (excess >> 2) ||
1777                                        (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1778                                        break;
1779                        }
1780                        continue;
1781                }
1782                if (!mem_cgroup_reclaimable(victim, false))
1783                        continue;
1784                total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1785                                                     zone, &nr_scanned);
1786                *total_scanned += nr_scanned;
1787                if (!res_counter_soft_limit_excess(&root_memcg->res))
1788                        break;
1789        }
1790        mem_cgroup_iter_break(root_memcg, victim);
1791        return total;
1792}
1793
1794/*
1795 * Check OOM-Killer is already running under our hierarchy.
1796 * If someone is running, return false.
1797 * Has to be called with memcg_oom_lock
1798 */
1799static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg)
1800{
1801        struct mem_cgroup *iter, *failed = NULL;
1802
1803        for_each_mem_cgroup_tree(iter, memcg) {
1804                if (iter->oom_lock) {
1805                        /*
1806                         * this subtree of our hierarchy is already locked
1807                         * so we cannot give a lock.
1808                         */
1809                        failed = iter;
1810                        mem_cgroup_iter_break(memcg, iter);
1811                        break;
1812                } else
1813                        iter->oom_lock = true;
1814        }
1815
1816        if (!failed)
1817                return true;
1818
1819        /*
1820         * OK, we failed to lock the whole subtree so we have to clean up
1821         * what we set up to the failing subtree
1822         */
1823        for_each_mem_cgroup_tree(iter, memcg) {
1824                if (iter == failed) {
1825                        mem_cgroup_iter_break(memcg, iter);
1826                        break;
1827                }
1828                iter->oom_lock = false;
1829        }
1830        return false;
1831}
1832
1833/*
1834 * Has to be called with memcg_oom_lock
1835 */
1836static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
1837{
1838        struct mem_cgroup *iter;
1839
1840        for_each_mem_cgroup_tree(iter, memcg)
1841                iter->oom_lock = false;
1842        return 0;
1843}
1844
1845static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
1846{
1847        struct mem_cgroup *iter;
1848
1849        for_each_mem_cgroup_tree(iter, memcg)
1850                atomic_inc(&iter->under_oom);
1851}
1852
1853static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
1854{
1855        struct mem_cgroup *iter;
1856
1857        /*
1858         * When a new child is created while the hierarchy is under oom,
1859         * mem_cgroup_oom_lock() may not be called. We have to use
1860         * atomic_add_unless() here.
1861         */
1862        for_each_mem_cgroup_tree(iter, memcg)
1863                atomic_add_unless(&iter->under_oom, -1, 0);
1864}
1865
1866static DEFINE_SPINLOCK(memcg_oom_lock);
1867static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1868
1869struct oom_wait_info {
1870        struct mem_cgroup *memcg;
1871        wait_queue_t    wait;
1872};
1873
1874static int memcg_oom_wake_function(wait_queue_t *wait,
1875        unsigned mode, int sync, void *arg)
1876{
1877        struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1878        struct mem_cgroup *oom_wait_memcg;
1879        struct oom_wait_info *oom_wait_info;
1880
1881        oom_wait_info = container_of(wait, struct oom_wait_info, wait);
1882        oom_wait_memcg = oom_wait_info->memcg;
1883
1884        /*
1885         * Both of oom_wait_info->memcg and wake_memcg are stable under us.
1886         * Then we can use css_is_ancestor without taking care of RCU.
1887         */
1888        if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg)
1889                && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg))
1890                return 0;
1891        return autoremove_wake_function(wait, mode, sync, arg);
1892}
1893
1894static void memcg_wakeup_oom(struct mem_cgroup *memcg)
1895{
1896        /* for filtering, pass "memcg" as argument. */
1897        __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
1898}
1899
1900static void memcg_oom_recover(struct mem_cgroup *memcg)
1901{
1902        if (memcg && atomic_read(&memcg->under_oom))
1903                memcg_wakeup_oom(memcg);
1904}
1905
1906/*
1907 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1908 */
1909static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask,
1910                                  int order)
1911{
1912        struct oom_wait_info owait;
1913        bool locked, need_to_kill;
1914
1915        owait.memcg = memcg;
1916        owait.wait.flags = 0;
1917        owait.wait.func = memcg_oom_wake_function;
1918        owait.wait.private = current;
1919        INIT_LIST_HEAD(&owait.wait.task_list);
1920        need_to_kill = true;
1921        mem_cgroup_mark_under_oom(memcg);
1922
1923        /* At first, try to OOM lock hierarchy under memcg.*/
1924        spin_lock(&memcg_oom_lock);
1925        locked = mem_cgroup_oom_lock(memcg);
1926        /*
1927         * Even if signal_pending(), we can't quit charge() loop without
1928         * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1929         * under OOM is always welcomed, use TASK_KILLABLE here.
1930         */
1931        prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
1932        if (!locked || memcg->oom_kill_disable)
1933                need_to_kill = false;
1934        if (locked)
1935                mem_cgroup_oom_notify(memcg);
1936        spin_unlock(&memcg_oom_lock);
1937
1938        if (need_to_kill) {
1939                finish_wait(&memcg_oom_waitq, &owait.wait);
1940                mem_cgroup_out_of_memory(memcg, mask, order);
1941        } else {
1942                schedule();
1943                finish_wait(&memcg_oom_waitq, &owait.wait);
1944        }
1945        spin_lock(&memcg_oom_lock);
1946        if (locked)
1947                mem_cgroup_oom_unlock(memcg);
1948        memcg_wakeup_oom(memcg);
1949        spin_unlock(&memcg_oom_lock);
1950
1951        mem_cgroup_unmark_under_oom(memcg);
1952
1953        if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
1954                return false;
1955        /* Give chance to dying process */
1956        schedule_timeout_uninterruptible(1);
1957        return true;
1958}
1959
1960/*
1961 * Currently used to update mapped file statistics, but the routine can be
1962 * generalized to update other statistics as well.
1963 *
1964 * Notes: Race condition
1965 *
1966 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1967 * it tends to be costly. But considering some conditions, we doesn't need
1968 * to do so _always_.
1969 *
1970 * Considering "charge", lock_page_cgroup() is not required because all
1971 * file-stat operations happen after a page is attached to radix-tree. There
1972 * are no race with "charge".
1973 *
1974 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1975 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1976 * if there are race with "uncharge". Statistics itself is properly handled
1977 * by flags.
1978 *
1979 * Considering "move", this is an only case we see a race. To make the race
1980 * small, we check mm->moving_account and detect there are possibility of race
1981 * If there is, we take a lock.
1982 */
1983
1984void __mem_cgroup_begin_update_page_stat(struct page *page,
1985                                bool *locked, unsigned long *flags)
1986{
1987        struct mem_cgroup *memcg;
1988        struct page_cgroup *pc;
1989
1990        pc = lookup_page_cgroup(page);
1991again:
1992        memcg = pc->mem_cgroup;
1993        if (unlikely(!memcg || !PageCgroupUsed(pc)))
1994                return;
1995        /*
1996         * If this memory cgroup is not under account moving, we don't
1997         * need to take move_lock_mem_cgroup(). Because we already hold
1998         * rcu_read_lock(), any calls to move_account will be delayed until
1999         * rcu_read_unlock() if mem_cgroup_stolen() == true.
2000         */
2001        if (!mem_cgroup_stolen(memcg))
2002                return;
2003
2004        move_lock_mem_cgroup(memcg, flags);
2005        if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) {
2006                move_unlock_mem_cgroup(memcg, flags);
2007                goto again;
2008        }
2009        *locked = true;
2010}
2011
2012void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags)
2013{
2014        struct page_cgroup *pc = lookup_page_cgroup(page);
2015
2016        /*
2017         * It's guaranteed that pc->mem_cgroup never changes while
2018         * lock is held because a routine modifies pc->mem_cgroup
2019         * should take move_lock_mem_cgroup().
2020         */
2021        move_unlock_mem_cgroup(pc->mem_cgroup, flags);
2022}
2023
2024void mem_cgroup_update_page_stat(struct page *page,
2025                                 enum mem_cgroup_page_stat_item idx, int val)
2026{
2027        struct mem_cgroup *memcg;
2028        struct page_cgroup *pc = lookup_page_cgroup(page);
2029        unsigned long uninitialized_var(flags);
2030
2031        if (mem_cgroup_disabled())
2032                return;
2033
2034        memcg = pc->mem_cgroup;
2035        if (unlikely(!memcg || !PageCgroupUsed(pc)))
2036                return;
2037
2038        switch (idx) {
2039        case MEMCG_NR_FILE_MAPPED:
2040                idx = MEM_CGROUP_STAT_FILE_MAPPED;
2041                break;
2042        default:
2043                BUG();
2044        }
2045
2046        this_cpu_add(memcg->stat->count[idx], val);
2047}
2048
2049/*
2050 * size of first charge trial. "32" comes from vmscan.c's magic value.
2051 * TODO: maybe necessary to use big numbers in big irons.
2052 */
2053#define CHARGE_BATCH    32U
2054struct memcg_stock_pcp {
2055        struct mem_cgroup *cached; /* this never be root cgroup */
2056        unsigned int nr_pages;
2057        struct work_struct work;
2058        unsigned long flags;
2059#define FLUSHING_CACHED_CHARGE  0
2060};
2061static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
2062static DEFINE_MUTEX(percpu_charge_mutex);
2063
2064/*
2065 * Try to consume stocked charge on this cpu. If success, one page is consumed
2066 * from local stock and true is returned. If the stock is 0 or charges from a
2067 * cgroup which is not current target, returns false. This stock will be
2068 * refilled.
2069 */
2070static bool consume_stock(struct mem_cgroup *memcg)
2071{
2072        struct memcg_stock_pcp *stock;
2073        bool ret = true;
2074
2075        stock = &get_cpu_var(memcg_stock);
2076        if (memcg == stock->cached && stock->nr_pages)
2077                stock->nr_pages--;
2078        else /* need to call res_counter_charge */
2079                ret = false;
2080        put_cpu_var(memcg_stock);
2081        return ret;
2082}
2083
2084/*
2085 * Returns stocks cached in percpu to res_counter and reset cached information.
2086 */
2087static void drain_stock(struct memcg_stock_pcp *stock)
2088{
2089        struct mem_cgroup *old = stock->cached;
2090
2091        if (stock->nr_pages) {
2092                unsigned long bytes = stock->nr_pages * PAGE_SIZE;
2093
2094                res_counter_uncharge(&old->res, bytes);
2095                if (do_swap_account)
2096                        res_counter_uncharge(&old->memsw, bytes);
2097                stock->nr_pages = 0;
2098        }
2099        stock->cached = NULL;
2100}
2101
2102/*
2103 * This must be called under preempt disabled or must be called by
2104 * a thread which is pinned to local cpu.
2105 */
2106static void drain_local_stock(struct work_struct *dummy)
2107{
2108        struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
2109        drain_stock(stock);
2110        clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
2111}
2112
2113/*
2114 * Cache charges(val) which is from res_counter, to local per_cpu area.
2115 * This will be consumed by consume_stock() function, later.
2116 */
2117static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
2118{
2119        struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2120
2121        if (stock->cached != memcg) { /* reset if necessary */
2122                drain_stock(stock);
2123                stock->cached = memcg;
2124        }
2125        stock->nr_pages += nr_pages;
2126        put_cpu_var(memcg_stock);
2127}
2128
2129/*
2130 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2131 * of the hierarchy under it. sync flag says whether we should block
2132 * until the work is done.
2133 */
2134static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
2135{
2136        int cpu, curcpu;
2137
2138        /* Notify other cpus that system-wide "drain" is running */
2139        get_online_cpus();
2140        curcpu = get_cpu();
2141        for_each_online_cpu(cpu) {
2142                struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
2143                struct mem_cgroup *memcg;
2144
2145                memcg = stock->cached;
2146                if (!memcg || !stock->nr_pages)
2147                        continue;
2148                if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
2149                        continue;
2150                if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2151                        if (cpu == curcpu)
2152                                drain_local_stock(&stock->work);
2153                        else
2154                                schedule_work_on(cpu, &stock->work);
2155                }
2156        }
2157        put_cpu();
2158
2159        if (!sync)
2160                goto out;
2161
2162        for_each_online_cpu(cpu) {
2163                struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
2164                if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
2165                        flush_work(&stock->work);
2166        }
2167out:
2168        put_online_cpus();
2169}
2170
2171/*
2172 * Tries to drain stocked charges in other cpus. This function is asynchronous
2173 * and just put a work per cpu for draining localy on each cpu. Caller can
2174 * expects some charges will be back to res_counter later but cannot wait for
2175 * it.
2176 */
2177static void drain_all_stock_async(struct mem_cgroup *root_memcg)
2178{
2179        /*
2180         * If someone calls draining, avoid adding more kworker runs.
2181         */
2182        if (!mutex_trylock(&percpu_charge_mutex))
2183                return;
2184        drain_all_stock(root_memcg, false);
2185        mutex_unlock(&percpu_charge_mutex);
2186}
2187
2188/* This is a synchronous drain interface. */
2189static void drain_all_stock_sync(struct mem_cgroup *root_memcg)
2190{
2191        /* called when force_empty is called */
2192        mutex_lock(&percpu_charge_mutex);
2193        drain_all_stock(root_memcg, true);
2194        mutex_unlock(&percpu_charge_mutex);
2195}
2196
2197/*
2198 * This function drains percpu counter value from DEAD cpu and
2199 * move it to local cpu. Note that this function can be preempted.
2200 */
2201static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
2202{
2203        int i;
2204
2205        spin_lock(&memcg->pcp_counter_lock);
2206        for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
2207                long x = per_cpu(memcg->stat->count[i], cpu);
2208
2209                per_cpu(memcg->stat->count[i], cpu) = 0;
2210                memcg->nocpu_base.count[i] += x;
2211        }
2212        for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
2213                unsigned long x = per_cpu(memcg->stat->events[i], cpu);
2214
2215                per_cpu(memcg->stat->events[i], cpu) = 0;
2216                memcg->nocpu_base.events[i] += x;
2217        }
2218        spin_unlock(&memcg->pcp_counter_lock);
2219}
2220
2221static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
2222                                        unsigned long action,
2223                                        void *hcpu)
2224{
2225        int cpu = (unsigned long)hcpu;
2226        struct memcg_stock_pcp *stock;
2227        struct mem_cgroup *iter;
2228
2229        if (action == CPU_ONLINE)
2230                return NOTIFY_OK;
2231
2232        if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
2233                return NOTIFY_OK;
2234
2235        for_each_mem_cgroup(iter)
2236                mem_cgroup_drain_pcp_counter(iter, cpu);
2237
2238        stock = &per_cpu(memcg_stock, cpu);
2239        drain_stock(stock);
2240        return NOTIFY_OK;
2241}
2242
2243
2244/* See __mem_cgroup_try_charge() for details */
2245enum {
2246        CHARGE_OK,              /* success */
2247        CHARGE_RETRY,           /* need to retry but retry is not bad */
2248        CHARGE_NOMEM,           /* we can't do more. return -ENOMEM */
2249        CHARGE_WOULDBLOCK,      /* GFP_WAIT wasn't set and no enough res. */
2250        CHARGE_OOM_DIE,         /* the current is killed because of OOM */
2251};
2252
2253static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
2254                                unsigned int nr_pages, bool oom_check)
2255{
2256        unsigned long csize = nr_pages * PAGE_SIZE;
2257        struct mem_cgroup *mem_over_limit;
2258        struct res_counter *fail_res;
2259        unsigned long flags = 0;
2260        int ret;
2261
2262        ret = res_counter_charge(&memcg->res, csize, &fail_res);
2263
2264        if (likely(!ret)) {
2265                if (!do_swap_account)
2266                        return CHARGE_OK;
2267                ret = res_counter_charge(&memcg->memsw, csize, &fail_res);
2268                if (likely(!ret))
2269                        return CHARGE_OK;
2270
2271                res_counter_uncharge(&memcg->res, csize);
2272                mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
2273                flags |= MEM_CGROUP_RECLAIM_NOSWAP;
2274        } else
2275                mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
2276        /*
2277         * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2278         * of regular pages (CHARGE_BATCH), or a single regular page (1).
2279         *
2280         * Never reclaim on behalf of optional batching, retry with a
2281         * single page instead.
2282         */
2283        if (nr_pages == CHARGE_BATCH)
2284                return CHARGE_RETRY;
2285
2286        if (!(gfp_mask & __GFP_WAIT))
2287                return CHARGE_WOULDBLOCK;
2288
2289        ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
2290        if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
2291                return CHARGE_RETRY;
2292        /*
2293         * Even though the limit is exceeded at this point, reclaim
2294         * may have been able to free some pages.  Retry the charge
2295         * before killing the task.
2296         *
2297         * Only for regular pages, though: huge pages are rather
2298         * unlikely to succeed so close to the limit, and we fall back
2299         * to regular pages anyway in case of failure.
2300         */
2301        if (nr_pages == 1 && ret)
2302                return CHARGE_RETRY;
2303
2304        /*
2305         * At task move, charge accounts can be doubly counted. So, it's
2306         * better to wait until the end of task_move if something is going on.
2307         */
2308        if (mem_cgroup_wait_acct_move(mem_over_limit))
2309                return CHARGE_RETRY;
2310
2311        /* If we don't need to call oom-killer at el, return immediately */
2312        if (!oom_check)
2313                return CHARGE_NOMEM;
2314        /* check OOM */
2315        if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize)))
2316                return CHARGE_OOM_DIE;
2317
2318        return CHARGE_RETRY;
2319}
2320
2321/*
2322 * __mem_cgroup_try_charge() does
2323 * 1. detect memcg to be charged against from passed *mm and *ptr,
2324 * 2. update res_counter
2325 * 3. call memory reclaim if necessary.
2326 *
2327 * In some special case, if the task is fatal, fatal_signal_pending() or
2328 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2329 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2330 * as possible without any hazards. 2: all pages should have a valid
2331 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2332 * pointer, that is treated as a charge to root_mem_cgroup.
2333 *
2334 * So __mem_cgroup_try_charge() will return
2335 *  0       ...  on success, filling *ptr with a valid memcg pointer.
2336 *  -ENOMEM ...  charge failure because of resource limits.
2337 *  -EINTR  ...  if thread is fatal. *ptr is filled with root_mem_cgroup.
2338 *
2339 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2340 * the oom-killer can be invoked.
2341 */
2342static int __mem_cgroup_try_charge(struct mm_struct *mm,
2343                                   gfp_t gfp_mask,
2344                                   unsigned int nr_pages,
2345                                   struct mem_cgroup **ptr,
2346                                   bool oom)
2347{
2348        unsigned int batch = max(CHARGE_BATCH, nr_pages);
2349        int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2350        struct mem_cgroup *memcg = NULL;
2351        int ret;
2352
2353        /*
2354         * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2355         * in system level. So, allow to go ahead dying process in addition to
2356         * MEMDIE process.
2357         */
2358        if (unlikely(test_thread_flag(TIF_MEMDIE)
2359                     || fatal_signal_pending(current)))
2360                goto bypass;
2361
2362        /*
2363         * We always charge the cgroup the mm_struct belongs to.
2364         * The mm_struct's mem_cgroup changes on task migration if the
2365         * thread group leader migrates. It's possible that mm is not
2366         * set, if so charge the root memcg (happens for pagecache usage).
2367         */
2368        if (!*ptr && !mm)
2369                *ptr = root_mem_cgroup;
2370again:
2371        if (*ptr) { /* css should be a valid one */
2372                memcg = *ptr;
2373                VM_BUG_ON(css_is_removed(&memcg->css));
2374                if (mem_cgroup_is_root(memcg))
2375                        goto done;
2376                if (nr_pages == 1 && consume_stock(memcg))
2377                        goto done;
2378                css_get(&memcg->css);
2379        } else {
2380                struct task_struct *p;
2381
2382                rcu_read_lock();
2383                p = rcu_dereference(mm->owner);
2384                /*
2385                 * Because we don't have task_lock(), "p" can exit.
2386                 * In that case, "memcg" can point to root or p can be NULL with
2387                 * race with swapoff. Then, we have small risk of mis-accouning.
2388                 * But such kind of mis-account by race always happens because
2389                 * we don't have cgroup_mutex(). It's overkill and we allo that
2390                 * small race, here.
2391                 * (*) swapoff at el will charge against mm-struct not against
2392                 * task-struct. So, mm->owner can be NULL.
2393                 */
2394                memcg = mem_cgroup_from_task(p);
2395                if (!memcg)
2396                        memcg = root_mem_cgroup;
2397                if (mem_cgroup_is_root(memcg)) {
2398                        rcu_read_unlock();
2399                        goto done;
2400                }
2401                if (nr_pages == 1 && consume_stock(memcg)) {
2402                        /*
2403                         * It seems dagerous to access memcg without css_get().
2404                         * But considering how consume_stok works, it's not
2405                         * necessary. If consume_stock success, some charges
2406                         * from this memcg are cached on this cpu. So, we
2407                         * don't need to call css_get()/css_tryget() before
2408                         * calling consume_stock().
2409                         */
2410                        rcu_read_unlock();
2411                        goto done;
2412                }
2413                /* after here, we may be blocked. we need to get refcnt */
2414                if (!css_tryget(&memcg->css)) {
2415                        rcu_read_unlock();
2416                        goto again;
2417                }
2418                rcu_read_unlock();
2419        }
2420
2421        do {
2422                bool oom_check;
2423
2424                /* If killed, bypass charge */
2425                if (fatal_signal_pending(current)) {
2426                        css_put(&memcg->css);
2427                        goto bypass;
2428                }
2429
2430                oom_check = false;
2431                if (oom && !nr_oom_retries) {
2432                        oom_check = true;
2433                        nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
2434                }
2435
2436                ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check);
2437                switch (ret) {
2438                case CHARGE_OK:
2439                        break;
2440                case CHARGE_RETRY: /* not in OOM situation but retry */
2441                        batch = nr_pages;
2442                        css_put(&memcg->css);
2443                        memcg = NULL;
2444                        goto again;
2445                case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
2446                        css_put(&memcg->css);
2447                        goto nomem;
2448                case CHARGE_NOMEM: /* OOM routine works */
2449                        if (!oom) {
2450                                css_put(&memcg->css);
2451                                goto nomem;
2452                        }
2453                        /* If oom, we never return -ENOMEM */
2454                        nr_oom_retries--;
2455                        break;
2456                case CHARGE_OOM_DIE: /* Killed by OOM Killer */
2457                        css_put(&memcg->css);
2458                        goto bypass;
2459                }
2460        } while (ret != CHARGE_OK);
2461
2462        if (batch > nr_pages)
2463                refill_stock(memcg, batch - nr_pages);
2464        css_put(&memcg->css);
2465done:
2466        *ptr = memcg;
2467        return 0;
2468nomem:
2469        *ptr = NULL;
2470        return -ENOMEM;
2471bypass:
2472        *ptr = root_mem_cgroup;
2473        return -EINTR;
2474}
2475
2476/*
2477 * Somemtimes we have to undo a charge we got by try_charge().
2478 * This function is for that and do uncharge, put css's refcnt.
2479 * gotten by try_charge().
2480 */
2481static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg,
2482                                       unsigned int nr_pages)
2483{
2484        if (!mem_cgroup_is_root(memcg)) {
2485                unsigned long bytes = nr_pages * PAGE_SIZE;
2486
2487                res_counter_uncharge(&memcg->res, bytes);
2488                if (do_swap_account)
2489                        res_counter_uncharge(&memcg->memsw, bytes);
2490        }
2491}
2492
2493/*
2494 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2495 * This is useful when moving usage to parent cgroup.
2496 */
2497static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg,
2498                                        unsigned int nr_pages)
2499{
2500        unsigned long bytes = nr_pages * PAGE_SIZE;
2501
2502        if (mem_cgroup_is_root(memcg))
2503                return;
2504
2505        res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes);
2506        if (do_swap_account)
2507                res_counter_uncharge_until(&memcg->memsw,
2508                                                memcg->memsw.parent, bytes);
2509}
2510
2511/*
2512 * A helper function to get mem_cgroup from ID. must be called under
2513 * rcu_read_lock(). The caller must check css_is_removed() or some if
2514 * it's concern. (dropping refcnt from swap can be called against removed
2515 * memcg.)
2516 */
2517static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2518{
2519        struct cgroup_subsys_state *css;
2520
2521        /* ID 0 is unused ID */
2522        if (!id)
2523                return NULL;
2524        css = css_lookup(&mem_cgroup_subsys, id);
2525        if (!css)
2526                return NULL;
2527        return mem_cgroup_from_css(css);
2528}
2529
2530struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
2531{
2532        struct mem_cgroup *memcg = NULL;
2533        struct page_cgroup *pc;
2534        unsigned short id;
2535        swp_entry_t ent;
2536
2537        VM_BUG_ON(!PageLocked(page));
2538
2539        pc = lookup_page_cgroup(page);
2540        lock_page_cgroup(pc);
2541        if (PageCgroupUsed(pc)) {
2542                memcg = pc->mem_cgroup;
2543                if (memcg && !css_tryget(&memcg->css))
2544                        memcg = NULL;
2545        } else if (PageSwapCache(page)) {
2546                ent.val = page_private(page);
2547                id = lookup_swap_cgroup_id(ent);
2548                rcu_read_lock();
2549                memcg = mem_cgroup_lookup(id);
2550                if (memcg && !css_tryget(&memcg->css))
2551                        memcg = NULL;
2552                rcu_read_unlock();
2553        }
2554        unlock_page_cgroup(pc);
2555        return memcg;
2556}
2557
2558static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
2559                                       struct page *page,
2560                                       unsigned int nr_pages,
2561                                       enum charge_type ctype,
2562                                       bool lrucare)
2563{
2564        struct page_cgroup *pc = lookup_page_cgroup(page);
2565        struct zone *uninitialized_var(zone);
2566        struct lruvec *lruvec;
2567        bool was_on_lru = false;
2568        bool anon;
2569
2570        lock_page_cgroup(pc);
2571        VM_BUG_ON(PageCgroupUsed(pc));
2572        /*
2573         * we don't need page_cgroup_lock about tail pages, becase they are not
2574         * accessed by any other context at this point.
2575         */
2576
2577        /*
2578         * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2579         * may already be on some other mem_cgroup's LRU.  Take care of it.
2580         */
2581        if (lrucare) {
2582                zone = page_zone(page);
2583                spin_lock_irq(&zone->lru_lock);
2584                if (PageLRU(page)) {
2585                        lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
2586                        ClearPageLRU(page);
2587                        del_page_from_lru_list(page, lruvec, page_lru(page));
2588                        was_on_lru = true;
2589                }
2590        }
2591
2592        pc->mem_cgroup = memcg;
2593        /*
2594         * We access a page_cgroup asynchronously without lock_page_cgroup().
2595         * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2596         * is accessed after testing USED bit. To make pc->mem_cgroup visible
2597         * before USED bit, we need memory barrier here.
2598         * See mem_cgroup_add_lru_list(), etc.
2599         */
2600        smp_wmb();
2601        SetPageCgroupUsed(pc);
2602
2603        if (lrucare) {
2604                if (was_on_lru) {
2605                        lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup);
2606                        VM_BUG_ON(PageLRU(page));
2607                        SetPageLRU(page);
2608                        add_page_to_lru_list(page, lruvec, page_lru(page));
2609                }
2610                spin_unlock_irq(&zone->lru_lock);
2611        }
2612
2613        if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON)
2614                anon = true;
2615        else
2616                anon = false;
2617
2618        mem_cgroup_charge_statistics(memcg, anon, nr_pages);
2619        unlock_page_cgroup(pc);
2620
2621        /*
2622         * "charge_statistics" updated event counter. Then, check it.
2623         * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2624         * if they exceeds softlimit.
2625         */
2626        memcg_check_events(memcg, page);
2627}
2628
2629#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2630
2631#define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
2632/*
2633 * Because tail pages are not marked as "used", set it. We're under
2634 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2635 * charge/uncharge will be never happen and move_account() is done under
2636 * compound_lock(), so we don't have to take care of races.
2637 */
2638void mem_cgroup_split_huge_fixup(struct page *head)
2639{
2640        struct page_cgroup *head_pc = lookup_page_cgroup(head);
2641        struct page_cgroup *pc;
2642        int i;
2643
2644        if (mem_cgroup_disabled())
2645                return;
2646        for (i = 1; i < HPAGE_PMD_NR; i++) {
2647                pc = head_pc + i;
2648                pc->mem_cgroup = head_pc->mem_cgroup;
2649                smp_wmb();/* see __commit_charge() */
2650                pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT;
2651        }
2652}
2653#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2654
2655/**
2656 * mem_cgroup_move_account - move account of the page
2657 * @page: the page
2658 * @nr_pages: number of regular pages (>1 for huge pages)
2659 * @pc: page_cgroup of the page.
2660 * @from: mem_cgroup which the page is moved from.
2661 * @to: mem_cgroup which the page is moved to. @from != @to.
2662 *
2663 * The caller must confirm following.
2664 * - page is not on LRU (isolate_page() is useful.)
2665 * - compound_lock is held when nr_pages > 1
2666 *
2667 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
2668 * from old cgroup.
2669 */
2670static int mem_cgroup_move_account(struct page *page,
2671                                   unsigned int nr_pages,
2672                                   struct page_cgroup *pc,
2673                                   struct mem_cgroup *from,
2674                                   struct mem_cgroup *to)
2675{
2676        unsigned long flags;
2677        int ret;
2678        bool anon = PageAnon(page);
2679
2680        VM_BUG_ON(from == to);
2681        VM_BUG_ON(PageLRU(page));
2682        /*
2683         * The page is isolated from LRU. So, collapse function
2684         * will not handle this page. But page splitting can happen.
2685         * Do this check under compound_page_lock(). The caller should
2686         * hold it.
2687         */
2688        ret = -EBUSY;
2689        if (nr_pages > 1 && !PageTransHuge(page))
2690                goto out;
2691
2692        lock_page_cgroup(pc);
2693
2694        ret = -EINVAL;
2695        if (!PageCgroupUsed(pc) || pc->mem_cgroup != from)
2696                goto unlock;
2697
2698        move_lock_mem_cgroup(from, &flags);
2699
2700        if (!anon && page_mapped(page)) {
2701                /* Update mapped_file data for mem_cgroup */
2702                preempt_disable();
2703                __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2704                __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
2705                preempt_enable();
2706        }
2707        mem_cgroup_charge_statistics(from, anon, -nr_pages);
2708
2709        /* caller should have done css_get */
2710        pc->mem_cgroup = to;
2711        mem_cgroup_charge_statistics(to, anon, nr_pages);
2712        /*
2713         * We charges against "to" which may not have any tasks. Then, "to"
2714         * can be under rmdir(). But in current implementation, caller of
2715         * this function is just force_empty() and move charge, so it's
2716         * guaranteed that "to" is never removed. So, we don't check rmdir
2717         * status here.
2718         */
2719        move_unlock_mem_cgroup(from, &flags);
2720        ret = 0;
2721unlock:
2722        unlock_page_cgroup(pc);
2723        /*
2724         * check events
2725         */
2726        memcg_check_events(to, page);
2727        memcg_check_events(from, page);
2728out:
2729        return ret;
2730}
2731
2732/*
2733 * move charges to its parent.
2734 */
2735
2736static int mem_cgroup_move_parent(struct page *page,
2737                                  struct page_cgroup *pc,
2738                                  struct mem_cgroup *child)
2739{
2740        struct mem_cgroup *parent;
2741        unsigned int nr_pages;
2742        unsigned long uninitialized_var(flags);
2743        int ret;
2744
2745        /* Is ROOT ? */
2746        if (mem_cgroup_is_root(child))
2747                return -EINVAL;
2748
2749        ret = -EBUSY;
2750        if (!get_page_unless_zero(page))
2751                goto out;
2752        if (isolate_lru_page(page))
2753                goto put;
2754
2755        nr_pages = hpage_nr_pages(page);
2756
2757        parent = parent_mem_cgroup(child);
2758        /*
2759         * If no parent, move charges to root cgroup.
2760         */
2761        if (!parent)
2762                parent = root_mem_cgroup;
2763
2764        if (nr_pages > 1)
2765                flags = compound_lock_irqsave(page);
2766
2767        ret = mem_cgroup_move_account(page, nr_pages,
2768                                pc, child, parent);
2769        if (!ret)
2770                __mem_cgroup_cancel_local_charge(child, nr_pages);
2771
2772        if (nr_pages > 1)
2773                compound_unlock_irqrestore(page, flags);
2774        putback_lru_page(page);
2775put:
2776        put_page(page);
2777out:
2778        return ret;
2779}
2780
2781/*
2782 * Charge the memory controller for page usage.
2783 * Return
2784 * 0 if the charge was successful
2785 * < 0 if the cgroup is over its limit
2786 */
2787static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
2788                                gfp_t gfp_mask, enum charge_type ctype)
2789{
2790        struct mem_cgroup *memcg = NULL;
2791        unsigned int nr_pages = 1;
2792        bool oom = true;
2793        int ret;
2794
2795        if (PageTransHuge(page)) {
2796                nr_pages <<= compound_order(page);
2797                VM_BUG_ON(!PageTransHuge(page));
2798                /*
2799                 * Never OOM-kill a process for a huge page.  The
2800                 * fault handler will fall back to regular pages.
2801                 */
2802                oom = false;
2803        }
2804
2805        ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom);
2806        if (ret == -ENOMEM)
2807                return ret;
2808        __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false);
2809        return 0;
2810}
2811
2812int mem_cgroup_newpage_charge(struct page *page,
2813                              struct mm_struct *mm, gfp_t gfp_mask)
2814{
2815        if (mem_cgroup_disabled())
2816                return 0;
2817        VM_BUG_ON(page_mapped(page));
2818        VM_BUG_ON(page->mapping && !PageAnon(page));
2819        VM_BUG_ON(!mm);
2820        return mem_cgroup_charge_common(page, mm, gfp_mask,
2821                                        MEM_CGROUP_CHARGE_TYPE_ANON);
2822}
2823
2824/*
2825 * While swap-in, try_charge -> commit or cancel, the page is locked.
2826 * And when try_charge() successfully returns, one refcnt to memcg without
2827 * struct page_cgroup is acquired. This refcnt will be consumed by
2828 * "commit()" or removed by "cancel()"
2829 */
2830static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
2831                                          struct page *page,
2832                                          gfp_t mask,
2833                                          struct mem_cgroup **memcgp)
2834{
2835        struct mem_cgroup *memcg;
2836        struct page_cgroup *pc;
2837        int ret;
2838
2839        pc = lookup_page_cgroup(page);
2840        /*
2841         * Every swap fault against a single page tries to charge the
2842         * page, bail as early as possible.  shmem_unuse() encounters
2843         * already charged pages, too.  The USED bit is protected by
2844         * the page lock, which serializes swap cache removal, which
2845         * in turn serializes uncharging.
2846         */
2847        if (PageCgroupUsed(pc))
2848                return 0;
2849        if (!do_swap_account)
2850                goto charge_cur_mm;
2851        memcg = try_get_mem_cgroup_from_page(page);
2852        if (!memcg)
2853                goto charge_cur_mm;
2854        *memcgp = memcg;
2855        ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true);
2856        css_put(&memcg->css);
2857        if (ret == -EINTR)
2858                ret = 0;
2859        return ret;
2860charge_cur_mm:
2861        ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
2862        if (ret == -EINTR)
2863                ret = 0;
2864        return ret;
2865}
2866
2867int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page,
2868                                 gfp_t gfp_mask, struct mem_cgroup **memcgp)
2869{
2870        *memcgp = NULL;
2871        if (mem_cgroup_disabled())
2872                return 0;
2873        /*
2874         * A racing thread's fault, or swapoff, may have already
2875         * updated the pte, and even removed page from swap cache: in
2876         * those cases unuse_pte()'s pte_same() test will fail; but
2877         * there's also a KSM case which does need to charge the page.
2878         */
2879        if (!PageSwapCache(page)) {
2880                int ret;
2881
2882                ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true);
2883                if (ret == -EINTR)
2884                        ret = 0;
2885                return ret;
2886        }
2887        return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
2888}
2889
2890void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
2891{
2892        if (mem_cgroup_disabled())
2893                return;
2894        if (!memcg)
2895                return;
2896        __mem_cgroup_cancel_charge(memcg, 1);
2897}
2898
2899static void
2900__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg,
2901                                        enum charge_type ctype)
2902{
2903        if (mem_cgroup_disabled())
2904                return;
2905        if (!memcg)
2906                return;
2907        cgroup_exclude_rmdir(&memcg->css);
2908
2909        __mem_cgroup_commit_charge(memcg, page, 1, ctype, true);
2910        /*
2911         * Now swap is on-memory. This means this page may be
2912         * counted both as mem and swap....double count.
2913         * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2914         * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2915         * may call delete_from_swap_cache() before reach here.
2916         */
2917        if (do_swap_account && PageSwapCache(page)) {
2918                swp_entry_t ent = {.val = page_private(page)};
2919                mem_cgroup_uncharge_swap(ent);
2920        }
2921        /*
2922         * At swapin, we may charge account against cgroup which has no tasks.
2923         * So, rmdir()->pre_destroy() can be called while we do this charge.
2924         * In that case, we need to call pre_destroy() again. check it here.
2925         */
2926        cgroup_release_and_wakeup_rmdir(&memcg->css);
2927}
2928
2929void mem_cgroup_commit_charge_swapin(struct page *page,
2930                                     struct mem_cgroup *memcg)
2931{
2932        __mem_cgroup_commit_charge_swapin(page, memcg,
2933                                          MEM_CGROUP_CHARGE_TYPE_ANON);
2934}
2935
2936int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
2937                                gfp_t gfp_mask)
2938{
2939        struct mem_cgroup *memcg = NULL;
2940        enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
2941        int ret;
2942
2943        if (mem_cgroup_disabled())
2944                return 0;
2945        if (PageCompound(page))
2946                return 0;
2947
2948        if (!PageSwapCache(page))
2949                ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
2950        else { /* page is swapcache/shmem */
2951                ret = __mem_cgroup_try_charge_swapin(mm, page,
2952                                                     gfp_mask, &memcg);
2953                if (!ret)
2954                        __mem_cgroup_commit_charge_swapin(page, memcg, type);
2955        }
2956        return ret;
2957}
2958
2959static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg,
2960                                   unsigned int nr_pages,
2961                                   const enum charge_type ctype)
2962{
2963        struct memcg_batch_info *batch = NULL;
2964        bool uncharge_memsw = true;
2965
2966        /* If swapout, usage of swap doesn't decrease */
2967        if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
2968                uncharge_memsw = false;
2969
2970        batch = &current->memcg_batch;
2971        /*
2972         * In usual, we do css_get() when we remember memcg pointer.
2973         * But in this case, we keep res->usage until end of a series of
2974         * uncharges. Then, it's ok to ignore memcg's refcnt.
2975         */
2976        if (!batch->memcg)
2977                batch->memcg = memcg;
2978        /*
2979         * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2980         * In those cases, all pages freed continuously can be expected to be in
2981         * the same cgroup and we have chance to coalesce uncharges.
2982         * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2983         * because we want to do uncharge as soon as possible.
2984         */
2985
2986        if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
2987                goto direct_uncharge;
2988
2989        if (nr_pages > 1)
2990                goto direct_uncharge;
2991
2992        /*
2993         * In typical case, batch->memcg == mem. This means we can
2994         * merge a series of uncharges to an uncharge of res_counter.
2995         * If not, we uncharge res_counter ony by one.
2996         */
2997        if (batch->memcg != memcg)
2998                goto direct_uncharge;
2999        /* remember freed charge and uncharge it later */
3000        batch->nr_pages++;
3001        if (uncharge_memsw)
3002                batch->memsw_nr_pages++;
3003        return;
3004direct_uncharge:
3005        res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE);
3006        if (uncharge_memsw)
3007                res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE);
3008        if (unlikely(batch->memcg != memcg))
3009                memcg_oom_recover(memcg);
3010}
3011
3012/*
3013 * uncharge if !page_mapped(page)
3014 */
3015static struct mem_cgroup *
3016__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype,
3017                             bool end_migration)
3018{
3019        struct mem_cgroup *memcg = NULL;
3020        unsigned int nr_pages = 1;
3021        struct page_cgroup *pc;
3022        bool anon;
3023
3024        if (mem_cgroup_disabled())
3025                return NULL;
3026
3027        VM_BUG_ON(PageSwapCache(page));
3028
3029        if (PageTransHuge(page)) {
3030                nr_pages <<= compound_order(page);
3031                VM_BUG_ON(!PageTransHuge(page));
3032        }
3033        /*
3034         * Check if our page_cgroup is valid
3035         */
3036        pc = lookup_page_cgroup(page);
3037        if (unlikely(!PageCgroupUsed(pc)))
3038                return NULL;
3039
3040        lock_page_cgroup(pc);
3041
3042        memcg = pc->mem_cgroup;
3043
3044        if (!PageCgroupUsed(pc))
3045                goto unlock_out;
3046
3047        anon = PageAnon(page);
3048
3049        switch (ctype) {
3050        case MEM_CGROUP_CHARGE_TYPE_ANON:
3051                /*
3052                 * Generally PageAnon tells if it's the anon statistics to be
3053                 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
3054                 * used before page reached the stage of being marked PageAnon.
3055                 */
3056                anon = true;
3057                /* fallthrough */
3058        case MEM_CGROUP_CHARGE_TYPE_DROP:
3059                /* See mem_cgroup_prepare_migration() */
3060                if (page_mapped(page))
3061                        goto unlock_out;
3062                /*
3063                 * Pages under migration may not be uncharged.  But
3064                 * end_migration() /must/ be the one uncharging the
3065                 * unused post-migration page and so it has to call
3066                 * here with the migration bit still set.  See the
3067                 * res_counter handling below.
3068                 */
3069                if (!end_migration && PageCgroupMigration(pc))
3070                        goto unlock_out;
3071                break;
3072        case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
3073                if (!PageAnon(page)) {  /* Shared memory */
3074                        if (page->mapping && !page_is_file_cache(page))
3075                                goto unlock_out;
3076                } else if (page_mapped(page)) /* Anon */
3077                                goto unlock_out;
3078                break;
3079        default:
3080                break;
3081        }
3082
3083        mem_cgroup_charge_statistics(memcg, anon, -nr_pages);
3084
3085        ClearPageCgroupUsed(pc);
3086        /*
3087         * pc->mem_cgroup is not cleared here. It will be accessed when it's
3088         * freed from LRU. This is safe because uncharged page is expected not
3089         * to be reused (freed soon). Exception is SwapCache, it's handled by
3090         * special functions.
3091         */
3092
3093        unlock_page_cgroup(pc);
3094        /*
3095         * even after unlock, we have memcg->res.usage here and this memcg
3096         * will never be freed.
3097         */
3098        memcg_check_events(memcg, page);
3099        if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
3100                mem_cgroup_swap_statistics(memcg, true);
3101                mem_cgroup_get(memcg);
3102        }
3103        /*
3104         * Migration does not charge the res_counter for the
3105         * replacement page, so leave it alone when phasing out the
3106         * page that is unused after the migration.
3107         */
3108        if (!end_migration && !mem_cgroup_is_root(memcg))
3109                mem_cgroup_do_uncharge(memcg, nr_pages, ctype);
3110
3111        return memcg;
3112
3113unlock_out:
3114        unlock_page_cgroup(pc);
3115        return NULL;
3116}
3117
3118void mem_cgroup_uncharge_page(struct page *page)
3119{
3120        /* early check. */
3121        if (page_mapped(page))
3122                return;
3123        VM_BUG_ON(page->mapping && !PageAnon(page));
3124        if (PageSwapCache(page))
3125                return;
3126        __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false);
3127}
3128
3129void mem_cgroup_uncharge_cache_page(struct page *page)
3130{
3131        VM_BUG_ON(page_mapped(page));
3132        VM_BUG_ON(page->mapping);
3133        __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false);
3134}
3135
3136/*
3137 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3138 * In that cases, pages are freed continuously and we can expect pages
3139 * are in the same memcg. All these calls itself limits the number of
3140 * pages freed at once, then uncharge_start/end() is called properly.
3141 * This may be called prural(2) times in a context,
3142 */
3143
3144void mem_cgroup_uncharge_start(void)
3145{
3146        current->memcg_batch.do_batch++;
3147        /* We can do nest. */
3148        if (current->memcg_batch.do_batch == 1) {
3149                current->memcg_batch.memcg = NULL;
3150                current->memcg_batch.nr_pages = 0;
3151                current->memcg_batch.memsw_nr_pages = 0;
3152        }
3153}
3154
3155void mem_cgroup_uncharge_end(void)
3156{
3157        struct memcg_batch_info *batch = &current->memcg_batch;
3158
3159        if (!batch->do_batch)
3160                return;
3161
3162        batch->do_batch--;
3163        if (batch->do_batch) /* If stacked, do nothing. */
3164                return;
3165
3166        if (!batch->memcg)
3167                return;
3168        /*
3169         * This "batch->memcg" is valid without any css_get/put etc...
3170         * bacause we hide charges behind us.
3171         */
3172        if (batch->nr_pages)
3173                res_counter_uncharge(&batch->memcg->res,
3174                                     batch->nr_pages * PAGE_SIZE);
3175        if (batch->memsw_nr_pages)
3176                res_counter_uncharge(&batch->memcg->memsw,
3177                                     batch->memsw_nr_pages * PAGE_SIZE);
3178        memcg_oom_recover(batch->memcg);
3179        /* forget this pointer (for sanity check) */
3180        batch->memcg = NULL;
3181}
3182
3183#ifdef CONFIG_SWAP
3184/*
3185 * called after __delete_from_swap_cache() and drop "page" account.
3186 * memcg information is recorded to swap_cgroup of "ent"
3187 */
3188void
3189mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
3190{
3191        struct mem_cgroup *memcg;
3192        int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
3193
3194        if (!swapout) /* this was a swap cache but the swap is unused ! */
3195                ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
3196
3197        memcg = __mem_cgroup_uncharge_common(page, ctype, false);
3198
3199        /*
3200         * record memcg information,  if swapout && memcg != NULL,
3201         * mem_cgroup_get() was called in uncharge().
3202         */
3203        if (do_swap_account && swapout && memcg)
3204                swap_cgroup_record(ent, css_id(&memcg->css));
3205}
3206#endif
3207
3208#ifdef CONFIG_MEMCG_SWAP
3209/*
3210 * called from swap_entry_free(). remove record in swap_cgroup and
3211 * uncharge "memsw" account.
3212 */
3213void mem_cgroup_uncharge_swap(swp_entry_t ent)
3214{
3215        struct mem_cgroup *memcg;
3216        unsigned short id;
3217
3218        if (!do_swap_account)
3219                return;
3220
3221        id = swap_cgroup_record(ent, 0);
3222        rcu_read_lock();
3223        memcg = mem_cgroup_lookup(id);
3224        if (memcg) {
3225                /*
3226                 * We uncharge this because swap is freed.
3227                 * This memcg can be obsolete one. We avoid calling css_tryget
3228                 */
3229                if (!mem_cgroup_is_root(memcg))
3230                        res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
3231                mem_cgroup_swap_statistics(memcg, false);
3232                mem_cgroup_put(memcg);
3233        }
3234        rcu_read_unlock();
3235}
3236
3237/**
3238 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3239 * @entry: swap entry to be moved
3240 * @from:  mem_cgroup which the entry is moved from
3241 * @to:  mem_cgroup which the entry is moved to
3242 *
3243 * It succeeds only when the swap_cgroup's record for this entry is the same
3244 * as the mem_cgroup's id of @from.
3245 *
3246 * Returns 0 on success, -EINVAL on failure.
3247 *
3248 * The caller must have charged to @to, IOW, called res_counter_charge() about
3249 * both res and memsw, and called css_get().
3250 */
3251static int mem_cgroup_move_swap_account(swp_entry_t entry,
3252                                struct mem_cgroup *from, struct mem_cgroup *to)
3253{
3254        unsigned short old_id, new_id;
3255
3256        old_id = css_id(&from->css);
3257        new_id = css_id(&to->css);
3258
3259        if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
3260                mem_cgroup_swap_statistics(from, false);
3261                mem_cgroup_swap_statistics(to, true);
3262                /*
3263                 * This function is only called from task migration context now.
3264                 * It postpones res_counter and refcount handling till the end
3265                 * of task migration(mem_cgroup_clear_mc()) for performance
3266                 * improvement. But we cannot postpone mem_cgroup_get(to)
3267                 * because if the process that has been moved to @to does
3268                 * swap-in, the refcount of @to might be decreased to 0.
3269                 */
3270                mem_cgroup_get(to);
3271                return 0;
3272        }
3273        return -EINVAL;
3274}
3275#else
3276static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
3277                                struct mem_cgroup *from, struct mem_cgroup *to)
3278{
3279        return -EINVAL;
3280}
3281#endif
3282
3283/*
3284 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3285 * page belongs to.
3286 */
3287void mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
3288                                  struct mem_cgroup **memcgp)
3289{
3290        struct mem_cgroup *memcg = NULL;
3291        struct page_cgroup *pc;
3292        enum charge_type ctype;
3293
3294        *memcgp = NULL;
3295
3296        VM_BUG_ON(PageTransHuge(page));
3297        if (mem_cgroup_disabled())
3298                return;
3299
3300        pc = lookup_page_cgroup(page);
3301        lock_page_cgroup(pc);
3302        if (PageCgroupUsed(pc)) {
3303                memcg = pc->mem_cgroup;
3304                css_get(&memcg->css);
3305                /*
3306                 * At migrating an anonymous page, its mapcount goes down
3307                 * to 0 and uncharge() will be called. But, even if it's fully
3308                 * unmapped, migration may fail and this page has to be
3309                 * charged again. We set MIGRATION flag here and delay uncharge
3310                 * until end_migration() is called
3311                 *
3312                 * Corner Case Thinking
3313                 * A)
3314                 * When the old page was mapped as Anon and it's unmap-and-freed
3315                 * while migration was ongoing.
3316                 * If unmap finds the old page, uncharge() of it will be delayed
3317                 * until end_migration(). If unmap finds a new page, it's
3318                 * uncharged when it make mapcount to be 1->0. If unmap code
3319                 * finds swap_migration_entry, the new page will not be mapped
3320                 * and end_migration() will find it(mapcount==0).
3321                 *
3322                 * B)
3323                 * When the old page was mapped but migraion fails, the kernel
3324                 * remaps it. A charge for it is kept by MIGRATION flag even
3325                 * if mapcount goes down to 0. We can do remap successfully
3326                 * without charging it again.
3327                 *
3328                 * C)
3329                 * The "old" page is under lock_page() until the end of
3330                 * migration, so, the old page itself will not be swapped-out.
3331                 * If the new page is swapped out before end_migraton, our
3332                 * hook to usual swap-out path will catch the event.
3333                 */
3334                if (PageAnon(page))
3335                        SetPageCgroupMigration(pc);
3336        }
3337        unlock_page_cgroup(pc);
3338        /*
3339         * If the page is not charged at this point,
3340         * we return here.
3341         */
3342        if (!memcg)
3343                return;
3344
3345        *memcgp = memcg;
3346        /*
3347         * We charge new page before it's used/mapped. So, even if unlock_page()
3348         * is called before end_migration, we can catch all events on this new
3349         * page. In the case new page is migrated but not remapped, new page's
3350         * mapcount will be finally 0 and we call uncharge in end_migration().
3351         */
3352        if (PageAnon(page))
3353                ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
3354        else
3355                ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
3356        /*
3357         * The page is committed to the memcg, but it's not actually
3358         * charged to the res_counter since we plan on replacing the
3359         * old one and only one page is going to be left afterwards.
3360         */
3361        __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false);
3362}
3363
3364/* remove redundant charge if migration failed*/
3365void mem_cgroup_end_migration(struct mem_cgroup *memcg,
3366        struct page *oldpage, struct page *newpage, bool migration_ok)
3367{
3368        struct page *used, *unused;
3369        struct page_cgroup *pc;
3370        bool anon;
3371
3372        if (!memcg)
3373                return;
3374        /* blocks rmdir() */
3375        cgroup_exclude_rmdir(&memcg->css);
3376        if (!migration_ok) {
3377                used = oldpage;
3378                unused = newpage;
3379        } else {
3380                used = newpage;
3381                unused = oldpage;
3382        }
3383        anon = PageAnon(used);
3384        __mem_cgroup_uncharge_common(unused,
3385                                     anon ? MEM_CGROUP_CHARGE_TYPE_ANON
3386                                     : MEM_CGROUP_CHARGE_TYPE_CACHE,
3387                                     true);
3388        css_put(&memcg->css);
3389        /*
3390         * We disallowed uncharge of pages under migration because mapcount
3391         * of the page goes down to zero, temporarly.
3392         * Clear the flag and check the page should be charged.
3393         */
3394        pc = lookup_page_cgroup(oldpage);
3395        lock_page_cgroup(pc);
3396        ClearPageCgroupMigration(pc);
3397        unlock_page_cgroup(pc);
3398
3399        /*
3400         * If a page is a file cache, radix-tree replacement is very atomic
3401         * and we can skip this check. When it was an Anon page, its mapcount
3402         * goes down to 0. But because we added MIGRATION flage, it's not
3403         * uncharged yet. There are several case but page->mapcount check
3404         * and USED bit check in mem_cgroup_uncharge_page() will do enough
3405         * check. (see prepare_charge() also)
3406         */
3407        if (anon)
3408                mem_cgroup_uncharge_page(used);
3409        /*
3410         * At migration, we may charge account against cgroup which has no
3411         * tasks.
3412         * So, rmdir()->pre_destroy() can be called while we do this charge.
3413         * In that case, we need to call pre_destroy() again. check it here.
3414         */
3415        cgroup_release_and_wakeup_rmdir(&memcg->css);
3416}
3417
3418/*
3419 * At replace page cache, newpage is not under any memcg but it's on
3420 * LRU. So, this function doesn't touch res_counter but handles LRU
3421 * in correct way. Both pages are locked so we cannot race with uncharge.
3422 */
3423void mem_cgroup_replace_page_cache(struct page *oldpage,
3424                                  struct page *newpage)
3425{
3426        struct mem_cgroup *memcg = NULL;
3427        struct page_cgroup *pc;
3428        enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE;
3429
3430        if (mem_cgroup_disabled())
3431                return;
3432
3433        pc = lookup_page_cgroup(oldpage);
3434        /* fix accounting on old pages */
3435        lock_page_cgroup(pc);
3436        if (PageCgroupUsed(pc)) {
3437                memcg = pc->mem_cgroup;
3438                mem_cgroup_charge_statistics(memcg, false, -1);
3439                ClearPageCgroupUsed(pc);
3440        }
3441        unlock_page_cgroup(pc);
3442
3443        /*
3444         * When called from shmem_replace_page(), in some cases the
3445         * oldpage has already been charged, and in some cases not.
3446         */
3447        if (!memcg)
3448                return;
3449        /*
3450         * Even if newpage->mapping was NULL before starting replacement,
3451         * the newpage may be on LRU(or pagevec for LRU) already. We lock
3452         * LRU while we overwrite pc->mem_cgroup.
3453         */
3454        __mem_cgroup_commit_charge(memcg, newpage, 1, type, true);
3455}
3456
3457#ifdef CONFIG_DEBUG_VM
3458static struct page_cgroup *lookup_page_cgroup_used(struct page *page)
3459{
3460        struct page_cgroup *pc;
3461
3462        pc = lookup_page_cgroup(page);
3463        /*
3464         * Can be NULL while feeding pages into the page allocator for
3465         * the first time, i.e. during boot or memory hotplug;
3466         * or when mem_cgroup_disabled().
3467         */
3468        if (likely(pc) && PageCgroupUsed(pc))
3469                return pc;
3470        return NULL;
3471}
3472
3473bool mem_cgroup_bad_page_check(struct page *page)
3474{
3475        if (mem_cgroup_disabled())
3476                return false;
3477
3478        return lookup_page_cgroup_used(page) != NULL;
3479}
3480
3481void mem_cgroup_print_bad_page(struct page *page)
3482{
3483        struct page_cgroup *pc;
3484
3485        pc = lookup_page_cgroup_used(page);
3486        if (pc) {
3487                printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3488                       pc, pc->flags, pc->mem_cgroup);
3489        }
3490}
3491#endif
3492
3493static DEFINE_MUTEX(set_limit_mutex);
3494
3495static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
3496                                unsigned long long val)
3497{
3498        int retry_count;
3499        u64 memswlimit, memlimit;
3500        int ret = 0;
3501        int children = mem_cgroup_count_children(memcg);
3502        u64 curusage, oldusage;
3503        int enlarge;
3504
3505        /*
3506         * For keeping hierarchical_reclaim simple, how long we should retry
3507         * is depends on callers. We set our retry-count to be function
3508         * of # of children which we should visit in this loop.
3509         */
3510        retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
3511
3512        oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3513
3514        enlarge = 0;
3515        while (retry_count) {
3516                if (signal_pending(current)) {
3517                        ret = -EINTR;
3518                        break;
3519                }
3520                /*
3521                 * Rather than hide all in some function, I do this in
3522                 * open coded manner. You see what this really does.
3523                 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3524                 */
3525                mutex_lock(&set_limit_mutex);
3526                memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3527                if (memswlimit < val) {
3528                        ret = -EINVAL;
3529                        mutex_unlock(&set_limit_mutex);
3530                        break;
3531                }
3532
3533                memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3534                if (memlimit < val)
3535                        enlarge = 1;
3536
3537                ret = res_counter_set_limit(&memcg->res, val);
3538                if (!ret) {
3539                        if (memswlimit == val)
3540                                memcg->memsw_is_minimum = true;
3541                        else
3542                                memcg->memsw_is_minimum = false;
3543                }
3544                mutex_unlock(&set_limit_mutex);
3545
3546                if (!ret)
3547                        break;
3548
3549                mem_cgroup_reclaim(memcg, GFP_KERNEL,
3550                                   MEM_CGROUP_RECLAIM_SHRINK);
3551                curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
3552                /* Usage is reduced ? */
3553                if (curusage >= oldusage)
3554                        retry_count--;
3555                else
3556                        oldusage = curusage;
3557        }
3558        if (!ret && enlarge)
3559                memcg_oom_recover(memcg);
3560
3561        return ret;
3562}
3563
3564static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
3565                                        unsigned long long val)
3566{
3567        int retry_count;
3568        u64 memlimit, memswlimit, oldusage, curusage;
3569        int children = mem_cgroup_count_children(memcg);
3570        int ret = -EBUSY;
3571        int enlarge = 0;
3572
3573        /* see mem_cgroup_resize_res_limit */
3574        retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
3575        oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
3576        while (retry_count) {
3577                if (signal_pending(current)) {
3578                        ret = -EINTR;
3579                        break;
3580                }
3581                /*
3582                 * Rather than hide all in some function, I do this in
3583                 * open coded manner. You see what this really does.
3584                 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3585                 */
3586                mutex_lock(&set_limit_mutex);
3587                memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
3588                if (memlimit > val) {
3589                        ret = -EINVAL;
3590                        mutex_unlock(&set_limit_mutex);
3591                        break;
3592                }
3593                memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
3594                if (memswlimit < val)
3595                        enlarge = 1;
3596                ret = res_counter_set_limit(&memcg->memsw, val);
3597                if (!ret) {
3598                        if (memlimit == val)
3599                                memcg->memsw_is_minimum = true;
3600                        else
3601                                memcg->memsw_is_minimum = false;
3602                }
3603                mutex_unlock(&set_limit_mutex);
3604
3605                if (!ret)
3606                        break;
3607
3608                mem_cgroup_reclaim(memcg, GFP_KERNEL,
3609                                   MEM_CGROUP_RECLAIM_NOSWAP |
3610                                   MEM_CGROUP_RECLAIM_SHRINK);
3611                curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
3612                /* Usage is reduced ? */
3613                if (curusage >= oldusage)
3614                        retry_count--;
3615                else
3616                        oldusage = curusage;
3617        }
3618        if (!ret && enlarge)
3619                memcg_oom_recover(memcg);
3620        return ret;
3621}
3622
3623unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
3624                                            gfp_t gfp_mask,
3625                                            unsigned long *total_scanned)
3626{
3627        unsigned long nr_reclaimed = 0;
3628        struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3629        unsigned long reclaimed;
3630        int loop = 0;
3631        struct mem_cgroup_tree_per_zone *mctz;
3632        unsigned long long excess;
3633        unsigned long nr_scanned;
3634
3635        if (order > 0)
3636                return 0;
3637
3638        mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
3639        /*
3640         * This loop can run a while, specially if mem_cgroup's continuously
3641         * keep exceeding their soft limit and putting the system under
3642         * pressure
3643         */
3644        do {
3645                if (next_mz)
3646                        mz = next_mz;
3647                else
3648                        mz = mem_cgroup_largest_soft_limit_node(mctz);
3649                if (!mz)
3650                        break;
3651
3652                nr_scanned = 0;
3653                reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
3654                                                    gfp_mask, &nr_scanned);
3655                nr_reclaimed += reclaimed;
3656                *total_scanned += nr_scanned;
3657                spin_lock(&mctz->lock);
3658
3659                /*
3660                 * If we failed to reclaim anything from this memory cgroup
3661                 * it is time to move on to the next cgroup
3662                 */
3663                next_mz = NULL;
3664                if (!reclaimed) {
3665                        do {
3666                                /*
3667                                 * Loop until we find yet another one.
3668                                 *
3669                                 * By the time we get the soft_limit lock
3670                                 * again, someone might have aded the
3671                                 * group back on the RB tree. Iterate to
3672                                 * make sure we get a different mem.
3673                                 * mem_cgroup_largest_soft_limit_node returns
3674                                 * NULL if no other cgroup is present on
3675                                 * the tree
3676                                 */
3677                                next_mz =
3678                                __mem_cgroup_largest_soft_limit_node(mctz);
3679                                if (next_mz == mz)
3680                                        css_put(&next_mz->memcg->css);
3681                                else /* next_mz == NULL or other memcg */
3682                                        break;
3683                        } while (1);
3684                }
3685                __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz);
3686                excess = res_counter_soft_limit_excess(&mz->memcg->res);
3687                /*
3688                 * One school of thought says that we should not add
3689                 * back the node to the tree if reclaim returns 0.
3690                 * But our reclaim could return 0, simply because due
3691                 * to priority we are exposing a smaller subset of
3692                 * memory to reclaim from. Consider this as a longer
3693                 * term TODO.
3694                 */
3695                /* If excess == 0, no tree ops */
3696                __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess);
3697                spin_unlock(&mctz->lock);
3698                css_put(&mz->memcg->css);
3699                loop++;
3700                /*
3701                 * Could not reclaim anything and there are no more
3702                 * mem cgroups to try or we seem to be looping without
3703                 * reclaiming anything.
3704                 */
3705                if (!nr_reclaimed &&
3706                        (next_mz == NULL ||
3707                        loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3708                        break;
3709        } while (!nr_reclaimed);
3710        if (next_mz)
3711                css_put(&next_mz->memcg->css);
3712        return nr_reclaimed;
3713}
3714
3715/*
3716 * Traverse a specified page_cgroup list and try to drop them all.  This doesn't
3717 * reclaim the pages page themselves - it just removes the page_cgroups.
3718 * Returns true if some page_cgroups were not freed, indicating that the caller
3719 * must retry this operation.
3720 */
3721static bool mem_cgroup_force_empty_list(struct mem_cgroup *memcg,
3722                                int node, int zid, enum lru_list lru)
3723{
3724        struct lruvec *lruvec;
3725        unsigned long flags, loop;
3726        struct list_head *list;
3727        struct page *busy;
3728        struct zone *zone;
3729
3730        zone = &NODE_DATA(node)->node_zones[zid];
3731        lruvec = mem_cgroup_zone_lruvec(zone, memcg);
3732        list = &lruvec->lists[lru];
3733
3734        loop = mem_cgroup_get_lru_size(lruvec, lru);
3735        /* give some margin against EBUSY etc...*/
3736        loop += 256;
3737        busy = NULL;
3738        while (loop--) {
3739                struct page_cgroup *pc;
3740                struct page *page;
3741
3742                spin_lock_irqsave(&zone->lru_lock, flags);
3743                if (list_empty(list)) {
3744                        spin_unlock_irqrestore(&zone->lru_lock, flags);
3745                        break;
3746                }
3747                page = list_entry(list->prev, struct page, lru);
3748                if (busy == page) {
3749                        list_move(&page->lru, list);
3750                        busy = NULL;
3751                        spin_unlock_irqrestore(&zone->lru_lock, flags);
3752                        continue;
3753                }
3754                spin_unlock_irqrestore(&zone->lru_lock, flags);
3755
3756                pc = lookup_page_cgroup(page);
3757
3758                if (mem_cgroup_move_parent(page, pc, memcg)) {
3759                        /* found lock contention or "pc" is obsolete. */
3760                        busy = page;
3761                        cond_resched();
3762                } else
3763                        busy = NULL;
3764        }
3765        return !list_empty(list);
3766}
3767
3768/*
3769 * make mem_cgroup's charge to be 0 if there is no task.
3770 * This enables deleting this mem_cgroup.
3771 */
3772static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all)
3773{
3774        int ret;
3775        int node, zid, shrink;
3776        int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
3777        struct cgroup *cgrp = memcg->css.cgroup;
3778
3779        css_get(&memcg->css);
3780
3781        shrink = 0;
3782        /* should free all ? */
3783        if (free_all)
3784                goto try_to_free;
3785move_account:
3786        do {
3787                ret = -EBUSY;
3788                if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
3789                        goto out;
3790                /* This is for making all *used* pages to be on LRU. */
3791                lru_add_drain_all();
3792                drain_all_stock_sync(memcg);
3793                ret = 0;
3794                mem_cgroup_start_move(memcg);
3795                for_each_node_state(node, N_HIGH_MEMORY) {
3796                        for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
3797                                enum lru_list lru;
3798                                for_each_lru(lru) {
3799                                        ret = mem_cgroup_force_empty_list(memcg,
3800                                                        node, zid, lru);
3801                                        if (ret)
3802                                                break;
3803                                }
3804                        }
3805                        if (ret)
3806                                break;
3807                }
3808                mem_cgroup_end_move(memcg);
3809                memcg_oom_recover(memcg);
3810                cond_resched();
3811        /* "ret" should also be checked to ensure all lists are empty. */
3812        } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0 || ret);
3813out:
3814        css_put(&memcg->css);
3815        return ret;
3816
3817try_to_free:
3818        /* returns EBUSY if there is a task or if we come here twice. */
3819        if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
3820                ret = -EBUSY;
3821                goto out;
3822        }
3823        /* we call try-to-free pages for make this cgroup empty */
3824        lru_add_drain_all();
3825        /* try to free all pages in this cgroup */
3826        shrink = 1;
3827        while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) {
3828                int progress;
3829
3830                if (signal_pending(current)) {
3831                        ret = -EINTR;
3832                        goto out;
3833                }
3834                progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL,
3835                                                false);
3836                if (!progress) {
3837                        nr_retries--;
3838                        /* maybe some writeback is necessary */
3839                        congestion_wait(BLK_RW_ASYNC, HZ/10);
3840                }
3841
3842        }
3843        lru_add_drain();
3844        /* try move_account...there may be some *locked* pages. */
3845        goto move_account;
3846}
3847
3848static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
3849{
3850        return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
3851}
3852
3853
3854static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
3855{
3856        return mem_cgroup_from_cont(cont)->use_hierarchy;
3857}
3858
3859static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
3860                                        u64 val)
3861{
3862        int retval = 0;
3863        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3864        struct cgroup *parent = cont->parent;
3865        struct mem_cgroup *parent_memcg = NULL;
3866
3867        if (parent)
3868                parent_memcg = mem_cgroup_from_cont(parent);
3869
3870        cgroup_lock();
3871
3872        if (memcg->use_hierarchy == val)
3873                goto out;
3874
3875        /*
3876         * If parent's use_hierarchy is set, we can't make any modifications
3877         * in the child subtrees. If it is unset, then the change can
3878         * occur, provided the current cgroup has no children.
3879         *
3880         * For the root cgroup, parent_mem is NULL, we allow value to be
3881         * set if there are no children.
3882         */
3883        if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
3884                                (val == 1 || val == 0)) {
3885                if (list_empty(&cont->children))
3886                        memcg->use_hierarchy = val;
3887                else
3888                        retval = -EBUSY;
3889        } else
3890                retval = -EINVAL;
3891
3892out:
3893        cgroup_unlock();
3894
3895        return retval;
3896}
3897
3898
3899static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg,
3900                                               enum mem_cgroup_stat_index idx)
3901{
3902        struct mem_cgroup *iter;
3903        long val = 0;
3904
3905        /* Per-cpu values can be negative, use a signed accumulator */
3906        for_each_mem_cgroup_tree(iter, memcg)
3907                val += mem_cgroup_read_stat(iter, idx);
3908
3909        if (val < 0) /* race ? */
3910                val = 0;
3911        return val;
3912}
3913
3914static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
3915{
3916        u64 val;
3917
3918        if (!mem_cgroup_is_root(memcg)) {
3919                if (!swap)
3920                        return res_counter_read_u64(&memcg->res, RES_USAGE);
3921                else
3922                        return res_counter_read_u64(&memcg->memsw, RES_USAGE);
3923        }
3924
3925        val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE);
3926        val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS);
3927
3928        if (swap)
3929                val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP);
3930
3931        return val << PAGE_SHIFT;
3932}
3933
3934static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft,
3935                               struct file *file, char __user *buf,
3936                               size_t nbytes, loff_t *ppos)
3937{
3938        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3939        char str[64];
3940        u64 val;
3941        int type, name, len;
3942
3943        type = MEMFILE_TYPE(cft->private);
3944        name = MEMFILE_ATTR(cft->private);
3945
3946        if (!do_swap_account && type == _MEMSWAP)
3947                return -EOPNOTSUPP;
3948
3949        switch (type) {
3950        case _MEM:
3951                if (name == RES_USAGE)
3952                        val = mem_cgroup_usage(memcg, false);
3953                else
3954                        val = res_counter_read_u64(&memcg->res, name);
3955                break;
3956        case _MEMSWAP:
3957                if (name == RES_USAGE)
3958                        val = mem_cgroup_usage(memcg, true);
3959                else
3960                        val = res_counter_read_u64(&memcg->memsw, name);
3961                break;
3962        default:
3963                BUG();
3964        }
3965
3966        len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val);
3967        return simple_read_from_buffer(buf, nbytes, ppos, str, len);
3968}
3969/*
3970 * The user of this function is...
3971 * RES_LIMIT.
3972 */
3973static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
3974                            const char *buffer)
3975{
3976        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
3977        int type, name;
3978        unsigned long long val;
3979        int ret;
3980
3981        type = MEMFILE_TYPE(cft->private);
3982        name = MEMFILE_ATTR(cft->private);
3983
3984        if (!do_swap_account && type == _MEMSWAP)
3985                return -EOPNOTSUPP;
3986
3987        switch (name) {
3988        case RES_LIMIT:
3989                if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3990                        ret = -EINVAL;
3991                        break;
3992                }
3993                /* This function does all necessary parse...reuse it */
3994                ret = res_counter_memparse_write_strategy(buffer, &val);
3995                if (ret)
3996                        break;
3997                if (type == _MEM)
3998                        ret = mem_cgroup_resize_limit(memcg, val);
3999                else
4000                        ret = mem_cgroup_resize_memsw_limit(memcg, val);
4001                break;
4002        case RES_SOFT_LIMIT:
4003                ret = res_counter_memparse_write_strategy(buffer, &val);
4004                if (ret)
4005                        break;
4006                /*
4007                 * For memsw, soft limits are hard to implement in terms
4008                 * of semantics, for now, we support soft limits for
4009                 * control without swap
4010                 */
4011                if (type == _MEM)
4012                        ret = res_counter_set_soft_limit(&memcg->res, val);
4013                else
4014                        ret = -EINVAL;
4015                break;
4016        default:
4017                ret = -EINVAL; /* should be BUG() ? */
4018                break;
4019        }
4020        return ret;
4021}
4022
4023static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
4024                unsigned long long *mem_limit, unsigned long long *memsw_limit)
4025{
4026        struct cgroup *cgroup;
4027        unsigned long long min_limit, min_memsw_limit, tmp;
4028
4029        min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
4030        min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4031        cgroup = memcg->css.cgroup;
4032        if (!memcg->use_hierarchy)
4033                goto out;
4034
4035        while (cgroup->parent) {
4036                cgroup = cgroup->parent;
4037                memcg = mem_cgroup_from_cont(cgroup);
4038                if (!memcg->use_hierarchy)
4039                        break;
4040                tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
4041                min_limit = min(min_limit, tmp);
4042                tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
4043                min_memsw_limit = min(min_memsw_limit, tmp);
4044        }
4045out:
4046        *mem_limit = min_limit;
4047        *memsw_limit = min_memsw_limit;
4048}
4049
4050static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
4051{
4052        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
4053        int type, name;
4054
4055        type = MEMFILE_TYPE(event);
4056        name = MEMFILE_ATTR(event);
4057
4058        if (!do_swap_account && type == _MEMSWAP)
4059                return -EOPNOTSUPP;
4060
4061        switch (name) {
4062        case RES_MAX_USAGE:
4063                if (type == _MEM)
4064                        res_counter_reset_max(&memcg->res);
4065                else
4066                        res_counter_reset_max(&memcg->memsw);
4067                break;
4068        case RES_FAILCNT:
4069                if (type == _MEM)
4070                        res_counter_reset_failcnt(&memcg->res);
4071                else
4072                        res_counter_reset_failcnt(&memcg->memsw);
4073                break;
4074        }
4075
4076        return 0;
4077}
4078
4079static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
4080                                        struct cftype *cft)
4081{
4082        return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
4083}
4084
4085#ifdef CONFIG_MMU
4086static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4087                                        struct cftype *cft, u64 val)
4088{
4089        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4090
4091        if (val >= (1 << NR_MOVE_TYPE))
4092                return -EINVAL;
4093        /*
4094         * We check this value several times in both in can_attach() and
4095         * attach(), so we need cgroup lock to prevent this value from being
4096         * inconsistent.
4097         */
4098        cgroup_lock();
4099        memcg->move_charge_at_immigrate = val;
4100        cgroup_unlock();
4101
4102        return 0;
4103}
4104#else
4105static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
4106                                        struct cftype *cft, u64 val)
4107{
4108        return -ENOSYS;
4109}
4110#endif
4111
4112#ifdef CONFIG_NUMA
4113static int memcg_numa_stat_show(struct cgroup *cont, struct cftype *cft,
4114                                      struct seq_file *m)
4115{
4116        int nid;
4117        unsigned long total_nr, file_nr, anon_nr, unevictable_nr;
4118        unsigned long node_nr;
4119        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
4120
4121        total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL);
4122        seq_printf(m, "total=%lu", total_nr);
4123        for_each_node_state(nid, N_HIGH_MEMORY) {
4124                node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL);
4125                seq_printf(m, " N%d=%lu", nid, node_nr);
4126        }
4127        seq_putc(m, '\n');
4128
4129        file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE);
4130        seq_printf(m, "file=%lu", file_nr);
4131        for_each_node_state(nid, N_HIGH_MEMORY) {
4132                node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
4133                                LRU_ALL_FILE);
4134                seq_printf(m, " N%d=%lu", nid, node_nr);
4135        }
4136        seq_putc(m, '\n');
4137
4138        anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON);
4139        seq_printf(m, "anon=%lu", anon_nr);
4140        for_each_node_state(nid, N_HIGH_MEMORY) {
4141                node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
4142                                LRU_ALL_ANON);
4143                seq_printf(m, " N%d=%lu", nid, node_nr);
4144        }
4145        seq_putc(m, '\n');
4146
4147        unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE));
4148        seq_printf(m, "unevictable=%lu", unevictable_nr);
4149        for_each_node_state(nid, N_HIGH_MEMORY) {
4150                node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
4151                                BIT(LRU_UNEVICTABLE));
4152                seq_printf(m, " N%d=%lu", nid, node_nr);
4153        }
4154        seq_putc(m, '\n');
4155        return 0;
4156}
4157#endif /* CONFIG_NUMA */
4158
4159static const char * const mem_cgroup_lru_names[] = {
4160        "inactive_anon",
4161        "active_anon",
4162        "inactive_file",
4163        "active_file",
4164        "unevictable",
4165};
4166
4167static inline void mem_cgroup_lru_names_not_uptodate(void)
4168{
4169        BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
4170}
4171
4172static int memcg_stat_show(struct cgroup *cont, struct cftype *cft,
4173                                 struct seq_file *m)
4174{
4175        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
4176        struct mem_cgroup *mi;
4177        unsigned int i;
4178
4179        for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
4180                if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
4181                        continue;
4182                seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
4183                           mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
4184        }
4185
4186        for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
4187                seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
4188                           mem_cgroup_read_events(memcg, i));
4189
4190        for (i = 0; i < NR_LRU_LISTS; i++)
4191                seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
4192                           mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
4193
4194        /* Hierarchical information */
4195        {
4196                unsigned long long limit, memsw_limit;
4197                memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit);
4198                seq_printf(m, "hierarchical_memory_limit %llu\n", limit);
4199                if (do_swap_account)
4200                        seq_printf(m, "hierarchical_memsw_limit %llu\n",
4201                                   memsw_limit);
4202        }
4203
4204        for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
4205                long long val = 0;
4206
4207                if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
4208                        continue;
4209                for_each_mem_cgroup_tree(mi, memcg)
4210                        val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
4211                seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
4212        }
4213
4214        for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
4215                unsigned long long val = 0;
4216
4217                for_each_mem_cgroup_tree(mi, memcg)
4218                        val += mem_cgroup_read_events(mi, i);
4219                seq_printf(m, "total_%s %llu\n",
4220                           mem_cgroup_events_names[i], val);
4221        }
4222
4223        for (i = 0; i < NR_LRU_LISTS; i++) {
4224                unsigned long long val = 0;
4225
4226                for_each_mem_cgroup_tree(mi, memcg)
4227                        val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
4228                seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
4229        }
4230
4231#ifdef CONFIG_DEBUG_VM
4232        {
4233                int nid, zid;
4234                struct mem_cgroup_per_zone *mz;
4235                struct zone_reclaim_stat *rstat;
4236                unsigned long recent_rotated[2] = {0, 0};
4237                unsigned long recent_scanned[2] = {0, 0};
4238
4239                for_each_online_node(nid)
4240                        for (zid = 0; zid < MAX_NR_ZONES; zid++) {
4241                                mz = mem_cgroup_zoneinfo(memcg, nid, zid);
4242                                rstat = &mz->lruvec.reclaim_stat;
4243
4244                                recent_rotated[0] += rstat->recent_rotated[0];
4245                                recent_rotated[1] += rstat->recent_rotated[1];
4246                                recent_scanned[0] += rstat->recent_scanned[0];
4247                                recent_scanned[1] += rstat->recent_scanned[1];
4248                        }
4249                seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
4250                seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
4251                seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
4252                seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
4253        }
4254#endif
4255
4256        return 0;
4257}
4258
4259static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
4260{
4261        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4262
4263        return mem_cgroup_swappiness(memcg);
4264}
4265
4266static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
4267                                       u64 val)
4268{
4269        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4270        struct mem_cgroup *parent;
4271
4272        if (val > 100)
4273                return -EINVAL;
4274
4275        if (cgrp->parent == NULL)
4276                return -EINVAL;
4277
4278        parent = mem_cgroup_from_cont(cgrp->parent);
4279
4280        cgroup_lock();
4281
4282        /* If under hierarchy, only empty-root can set this value */
4283        if ((parent->use_hierarchy) ||
4284            (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4285                cgroup_unlock();
4286                return -EINVAL;
4287        }
4288
4289        memcg->swappiness = val;
4290
4291        cgroup_unlock();
4292
4293        return 0;
4294}
4295
4296static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
4297{
4298        struct mem_cgroup_threshold_ary *t;
4299        u64 usage;
4300        int i;
4301
4302        rcu_read_lock();
4303        if (!swap)
4304                t = rcu_dereference(memcg->thresholds.primary);
4305        else
4306                t = rcu_dereference(memcg->memsw_thresholds.primary);
4307
4308        if (!t)
4309                goto unlock;
4310
4311        usage = mem_cgroup_usage(memcg, swap);
4312
4313        /*
4314         * current_threshold points to threshold just below or equal to usage.
4315         * If it's not true, a threshold was crossed after last
4316         * call of __mem_cgroup_threshold().
4317         */
4318        i = t->current_threshold;
4319
4320        /*
4321         * Iterate backward over array of thresholds starting from
4322         * current_threshold and check if a threshold is crossed.
4323         * If none of thresholds below usage is crossed, we read
4324         * only one element of the array here.
4325         */
4326        for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
4327                eventfd_signal(t->entries[i].eventfd, 1);
4328
4329        /* i = current_threshold + 1 */
4330        i++;
4331
4332        /*
4333         * Iterate forward over array of thresholds starting from
4334         * current_threshold+1 and check if a threshold is crossed.
4335         * If none of thresholds above usage is crossed, we read
4336         * only one element of the array here.
4337         */
4338        for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
4339                eventfd_signal(t->entries[i].eventfd, 1);
4340
4341        /* Update current_threshold */
4342        t->current_threshold = i - 1;
4343unlock:
4344        rcu_read_unlock();
4345}
4346
4347static void mem_cgroup_threshold(struct mem_cgroup *memcg)
4348{
4349        while (memcg) {
4350                __mem_cgroup_threshold(memcg, false);
4351                if (do_swap_account)
4352                        __mem_cgroup_threshold(memcg, true);
4353
4354                memcg = parent_mem_cgroup(memcg);
4355        }
4356}
4357
4358static int compare_thresholds(const void *a, const void *b)
4359{
4360        const struct mem_cgroup_threshold *_a = a;
4361        const struct mem_cgroup_threshold *_b = b;
4362
4363        return _a->threshold - _b->threshold;
4364}
4365
4366static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
4367{
4368        struct mem_cgroup_eventfd_list *ev;
4369
4370        list_for_each_entry(ev, &memcg->oom_notify, list)
4371                eventfd_signal(ev->eventfd, 1);
4372        return 0;
4373}
4374
4375static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
4376{
4377        struct mem_cgroup *iter;
4378
4379        for_each_mem_cgroup_tree(iter, memcg)
4380                mem_cgroup_oom_notify_cb(iter);
4381}
4382
4383static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
4384        struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4385{
4386        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4387        struct mem_cgroup_thresholds *thresholds;
4388        struct mem_cgroup_threshold_ary *new;
4389        int type = MEMFILE_TYPE(cft->private);
4390        u64 threshold, usage;
4391        int i, size, ret;
4392
4393        ret = res_counter_memparse_write_strategy(args, &threshold);
4394        if (ret)
4395                return ret;
4396
4397        mutex_lock(&memcg->thresholds_lock);
4398
4399        if (type == _MEM)
4400                thresholds = &memcg->thresholds;
4401        else if (type == _MEMSWAP)
4402                thresholds = &memcg->memsw_thresholds;
4403        else
4404                BUG();
4405
4406        usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4407
4408        /* Check if a threshold crossed before adding a new one */
4409        if (thresholds->primary)
4410                __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4411
4412        size = thresholds->primary ? thresholds->primary->size + 1 : 1;
4413
4414        /* Allocate memory for new array of thresholds */
4415        new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
4416                        GFP_KERNEL);
4417        if (!new) {
4418                ret = -ENOMEM;
4419                goto unlock;
4420        }
4421        new->size = size;
4422
4423        /* Copy thresholds (if any) to new array */
4424        if (thresholds->primary) {
4425                memcpy(new->entries, thresholds->primary->entries, (size - 1) *
4426                                sizeof(struct mem_cgroup_threshold));
4427        }
4428
4429        /* Add new threshold */
4430        new->entries[size - 1].eventfd = eventfd;
4431        new->entries[size - 1].threshold = threshold;
4432
4433        /* Sort thresholds. Registering of new threshold isn't time-critical */
4434        sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
4435                        compare_thresholds, NULL);
4436
4437        /* Find current threshold */
4438        new->current_threshold = -1;
4439        for (i = 0; i < size; i++) {
4440                if (new->entries[i].threshold <= usage) {
4441                        /*
4442                         * new->current_threshold will not be used until
4443                         * rcu_assign_pointer(), so it's safe to increment
4444                         * it here.
4445                         */
4446                        ++new->current_threshold;
4447                } else
4448                        break;
4449        }
4450
4451        /* Free old spare buffer and save old primary buffer as spare */
4452        kfree(thresholds->spare);
4453        thresholds->spare = thresholds->primary;
4454
4455        rcu_assign_pointer(thresholds->primary, new);
4456
4457        /* To be sure that nobody uses thresholds */
4458        synchronize_rcu();
4459
4460unlock:
4461        mutex_unlock(&memcg->thresholds_lock);
4462
4463        return ret;
4464}
4465
4466static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
4467        struct cftype *cft, struct eventfd_ctx *eventfd)
4468{
4469        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4470        struct mem_cgroup_thresholds *thresholds;
4471        struct mem_cgroup_threshold_ary *new;
4472        int type = MEMFILE_TYPE(cft->private);
4473        u64 usage;
4474        int i, j, size;
4475
4476        mutex_lock(&memcg->thresholds_lock);
4477        if (type == _MEM)
4478                thresholds = &memcg->thresholds;
4479        else if (type == _MEMSWAP)
4480                thresholds = &memcg->memsw_thresholds;
4481        else
4482                BUG();
4483
4484        if (!thresholds->primary)
4485                goto unlock;
4486
4487        usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
4488
4489        /* Check if a threshold crossed before removing */
4490        __mem_cgroup_threshold(memcg, type == _MEMSWAP);
4491
4492        /* Calculate new number of threshold */
4493        size = 0;
4494        for (i = 0; i < thresholds->primary->size; i++) {
4495                if (thresholds->primary->entries[i].eventfd != eventfd)
4496                        size++;
4497        }
4498
4499        new = thresholds->spare;
4500
4501        /* Set thresholds array to NULL if we don't have thresholds */
4502        if (!size) {
4503                kfree(new);
4504                new = NULL;
4505                goto swap_buffers;
4506        }
4507
4508        new->size = size;
4509
4510        /* Copy thresholds and find current threshold */
4511        new->current_threshold = -1;
4512        for (i = 0, j = 0; i < thresholds->primary->size; i++) {
4513                if (thresholds->primary->entries[i].eventfd == eventfd)
4514                        continue;
4515
4516                new->entries[j] = thresholds->primary->entries[i];
4517                if (new->entries[j].threshold <= usage) {
4518                        /*
4519                         * new->current_threshold will not be used
4520                         * until rcu_assign_pointer(), so it's safe to increment
4521                         * it here.
4522                         */
4523                        ++new->current_threshold;
4524                }
4525                j++;
4526        }
4527
4528swap_buffers:
4529        /* Swap primary and spare array */
4530        thresholds->spare = thresholds->primary;
4531        /* If all events are unregistered, free the spare array */
4532        if (!new) {
4533                kfree(thresholds->spare);
4534                thresholds->spare = NULL;
4535        }
4536
4537        rcu_assign_pointer(thresholds->primary, new);
4538
4539        /* To be sure that nobody uses thresholds */
4540        synchronize_rcu();
4541unlock:
4542        mutex_unlock(&memcg->thresholds_lock);
4543}
4544
4545static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
4546        struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
4547{
4548        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4549        struct mem_cgroup_eventfd_list *event;
4550        int type = MEMFILE_TYPE(cft->private);
4551
4552        BUG_ON(type != _OOM_TYPE);
4553        event = kmalloc(sizeof(*event), GFP_KERNEL);
4554        if (!event)
4555                return -ENOMEM;
4556
4557        spin_lock(&memcg_oom_lock);
4558
4559        event->eventfd = eventfd;
4560        list_add(&event->list, &memcg->oom_notify);
4561
4562        /* already in OOM ? */
4563        if (atomic_read(&memcg->under_oom))
4564                eventfd_signal(eventfd, 1);
4565        spin_unlock(&memcg_oom_lock);
4566
4567        return 0;
4568}
4569
4570static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
4571        struct cftype *cft, struct eventfd_ctx *eventfd)
4572{
4573        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4574        struct mem_cgroup_eventfd_list *ev, *tmp;
4575        int type = MEMFILE_TYPE(cft->private);
4576
4577        BUG_ON(type != _OOM_TYPE);
4578
4579        spin_lock(&memcg_oom_lock);
4580
4581        list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
4582                if (ev->eventfd == eventfd) {
4583                        list_del(&ev->list);
4584                        kfree(ev);
4585                }
4586        }
4587
4588        spin_unlock(&memcg_oom_lock);
4589}
4590
4591static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
4592        struct cftype *cft,  struct cgroup_map_cb *cb)
4593{
4594        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4595
4596        cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable);
4597
4598        if (atomic_read(&memcg->under_oom))
4599                cb->fill(cb, "under_oom", 1);
4600        else
4601                cb->fill(cb, "under_oom", 0);
4602        return 0;
4603}
4604
4605static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
4606        struct cftype *cft, u64 val)
4607{
4608        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
4609        struct mem_cgroup *parent;
4610
4611        /* cannot set to root cgroup and only 0 and 1 are allowed */
4612        if (!cgrp->parent || !((val == 0) || (val == 1)))
4613                return -EINVAL;
4614
4615        parent = mem_cgroup_from_cont(cgrp->parent);
4616
4617        cgroup_lock();
4618        /* oom-kill-disable is a flag for subhierarchy. */
4619        if ((parent->use_hierarchy) ||
4620            (memcg->use_hierarchy && !list_empty(&cgrp->children))) {
4621                cgroup_unlock();
4622                return -EINVAL;
4623        }
4624        memcg->oom_kill_disable = val;
4625        if (!val)
4626                memcg_oom_recover(memcg);
4627        cgroup_unlock();
4628        return 0;
4629}
4630
4631#ifdef CONFIG_MEMCG_KMEM
4632static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
4633{
4634        return mem_cgroup_sockets_init(memcg, ss);
4635};
4636
4637static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
4638{
4639        mem_cgroup_sockets_destroy(memcg);
4640}
4641#else
4642static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
4643{
4644        return 0;
4645}
4646
4647static void kmem_cgroup_destroy(struct mem_cgroup *memcg)
4648{
4649}
4650#endif
4651
4652static struct cftype mem_cgroup_files[] = {
4653        {
4654                .name = "usage_in_bytes",
4655                .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
4656                .read = mem_cgroup_read,
4657                .register_event = mem_cgroup_usage_register_event,
4658                .unregister_event = mem_cgroup_usage_unregister_event,
4659        },
4660        {
4661                .name = "max_usage_in_bytes",
4662                .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
4663                .trigger = mem_cgroup_reset,
4664                .read = mem_cgroup_read,
4665        },
4666        {
4667                .name = "limit_in_bytes",
4668                .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
4669                .write_string = mem_cgroup_write,
4670                .read = mem_cgroup_read,
4671        },
4672        {
4673                .name = "soft_limit_in_bytes",
4674                .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4675                .write_string = mem_cgroup_write,
4676                .read = mem_cgroup_read,
4677        },
4678        {
4679                .name = "failcnt",
4680                .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
4681                .trigger = mem_cgroup_reset,
4682                .read = mem_cgroup_read,
4683        },
4684        {
4685                .name = "stat",
4686                .read_seq_string = memcg_stat_show,
4687        },
4688        {
4689                .name = "force_empty",
4690                .trigger = mem_cgroup_force_empty_write,
4691        },
4692        {
4693                .name = "use_hierarchy",
4694                .write_u64 = mem_cgroup_hierarchy_write,
4695                .read_u64 = mem_cgroup_hierarchy_read,
4696        },
4697        {
4698                .name = "swappiness",
4699                .read_u64 = mem_cgroup_swappiness_read,
4700                .write_u64 = mem_cgroup_swappiness_write,
4701        },
4702        {
4703                .name = "move_charge_at_immigrate",
4704                .read_u64 = mem_cgroup_move_charge_read,
4705                .write_u64 = mem_cgroup_move_charge_write,
4706        },
4707        {
4708                .name = "oom_control",
4709                .read_map = mem_cgroup_oom_control_read,
4710                .write_u64 = mem_cgroup_oom_control_write,
4711                .register_event = mem_cgroup_oom_register_event,
4712                .unregister_event = mem_cgroup_oom_unregister_event,
4713                .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4714        },
4715#ifdef CONFIG_NUMA
4716        {
4717                .name = "numa_stat",
4718                .read_seq_string = memcg_numa_stat_show,
4719        },
4720#endif
4721#ifdef CONFIG_MEMCG_SWAP
4722        {
4723                .name = "memsw.usage_in_bytes",
4724                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
4725                .read = mem_cgroup_read,
4726                .register_event = mem_cgroup_usage_register_event,
4727                .unregister_event = mem_cgroup_usage_unregister_event,
4728        },
4729        {
4730                .name = "memsw.max_usage_in_bytes",
4731                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
4732                .trigger = mem_cgroup_reset,
4733                .read = mem_cgroup_read,
4734        },
4735        {
4736                .name = "memsw.limit_in_bytes",
4737                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
4738                .write_string = mem_cgroup_write,
4739                .read = mem_cgroup_read,
4740        },
4741        {
4742                .name = "memsw.failcnt",
4743                .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
4744                .trigger = mem_cgroup_reset,
4745                .read = mem_cgroup_read,
4746        },
4747#endif
4748        { },    /* terminate */
4749};
4750
4751static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
4752{
4753        struct mem_cgroup_per_node *pn;
4754        struct mem_cgroup_per_zone *mz;
4755        int zone, tmp = node;
4756        /*
4757         * This routine is called against possible nodes.
4758         * But it's BUG to call kmalloc() against offline node.
4759         *
4760         * TODO: this routine can waste much memory for nodes which will
4761         *       never be onlined. It's better to use memory hotplug callback
4762         *       function.
4763         */
4764        if (!node_state(node, N_NORMAL_MEMORY))
4765                tmp = -1;
4766        pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
4767        if (!pn)
4768                return 1;
4769
4770        for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4771                mz = &pn->zoneinfo[zone];
4772                lruvec_init(&mz->lruvec);
4773                mz->usage_in_excess = 0;
4774                mz->on_tree = false;
4775                mz->memcg = memcg;
4776        }
4777        memcg->info.nodeinfo[node] = pn;
4778        return 0;
4779}
4780
4781static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
4782{
4783        kfree(memcg->info.nodeinfo[node]);
4784}
4785
4786static struct mem_cgroup *mem_cgroup_alloc(void)
4787{
4788        struct mem_cgroup *memcg;
4789        int size = sizeof(struct mem_cgroup);
4790
4791        /* Can be very big if MAX_NUMNODES is very big */
4792        if (size < PAGE_SIZE)
4793                memcg = kzalloc(size, GFP_KERNEL);
4794        else
4795                memcg = vzalloc(size);
4796
4797        if (!memcg)
4798                return NULL;
4799
4800        memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4801        if (!memcg->stat)
4802                goto out_free;
4803        spin_lock_init(&memcg->pcp_counter_lock);
4804        return memcg;
4805
4806out_free:
4807        if (size < PAGE_SIZE)
4808                kfree(memcg);
4809        else
4810                vfree(memcg);
4811        return NULL;
4812}
4813
4814/*
4815 * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU,
4816 * but in process context.  The work_freeing structure is overlaid
4817 * on the rcu_freeing structure, which itself is overlaid on memsw.
4818 */
4819static void free_work(struct work_struct *work)
4820{
4821        struct mem_cgroup *memcg;
4822        int size = sizeof(struct mem_cgroup);
4823
4824        memcg = container_of(work, struct mem_cgroup, work_freeing);
4825        /*
4826         * We need to make sure that (at least for now), the jump label
4827         * destruction code runs outside of the cgroup lock. This is because
4828         * get_online_cpus(), which is called from the static_branch update,
4829         * can't be called inside the cgroup_lock. cpusets are the ones
4830         * enforcing this dependency, so if they ever change, we might as well.
4831         *
4832         * schedule_work() will guarantee this happens. Be careful if you need
4833         * to move this code around, and make sure it is outside
4834         * the cgroup_lock.
4835         */
4836        disarm_sock_keys(memcg);
4837        if (size < PAGE_SIZE)
4838                kfree(memcg);
4839        else
4840                vfree(memcg);
4841}
4842
4843static void free_rcu(struct rcu_head *rcu_head)
4844{
4845        struct mem_cgroup *memcg;
4846
4847        memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing);
4848        INIT_WORK(&memcg->work_freeing, free_work);
4849        schedule_work(&memcg->work_freeing);
4850}
4851
4852/*
4853 * At destroying mem_cgroup, references from swap_cgroup can remain.
4854 * (scanning all at force_empty is too costly...)
4855 *
4856 * Instead of clearing all references at force_empty, we remember
4857 * the number of reference from swap_cgroup and free mem_cgroup when
4858 * it goes down to 0.
4859 *
4860 * Removal of cgroup itself succeeds regardless of refs from swap.
4861 */
4862
4863static void __mem_cgroup_free(struct mem_cgroup *memcg)
4864{
4865        int node;
4866
4867        mem_cgroup_remove_from_trees(memcg);
4868        free_css_id(&mem_cgroup_subsys, &memcg->css);
4869
4870        for_each_node(node)
4871                free_mem_cgroup_per_zone_info(memcg, node);
4872
4873        free_percpu(memcg->stat);
4874        call_rcu(&memcg->rcu_freeing, free_rcu);
4875}
4876
4877static void mem_cgroup_get(struct mem_cgroup *memcg)
4878{
4879        atomic_inc(&memcg->refcnt);
4880}
4881
4882static void __mem_cgroup_put(struct mem_cgroup *memcg, int count)
4883{
4884        if (atomic_sub_and_test(count, &memcg->refcnt)) {
4885                struct mem_cgroup *parent = parent_mem_cgroup(memcg);
4886                __mem_cgroup_free(memcg);
4887                if (parent)
4888                        mem_cgroup_put(parent);
4889        }
4890}
4891
4892static void mem_cgroup_put(struct mem_cgroup *memcg)
4893{
4894        __mem_cgroup_put(memcg, 1);
4895}
4896
4897/*
4898 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4899 */
4900struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
4901{
4902        if (!memcg->res.parent)
4903                return NULL;
4904        return mem_cgroup_from_res_counter(memcg->res.parent, res);
4905}
4906EXPORT_SYMBOL(parent_mem_cgroup);
4907
4908#ifdef CONFIG_MEMCG_SWAP
4909static void __init enable_swap_cgroup(void)
4910{
4911        if (!mem_cgroup_disabled() && really_do_swap_account)
4912                do_swap_account = 1;
4913}
4914#else
4915static void __init enable_swap_cgroup(void)
4916{
4917}
4918#endif
4919
4920static int mem_cgroup_soft_limit_tree_init(void)
4921{
4922        struct mem_cgroup_tree_per_node *rtpn;
4923        struct mem_cgroup_tree_per_zone *rtpz;
4924        int tmp, node, zone;
4925
4926        for_each_node(node) {
4927                tmp = node;
4928                if (!node_state(node, N_NORMAL_MEMORY))
4929                        tmp = -1;
4930                rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
4931                if (!rtpn)
4932                        goto err_cleanup;
4933
4934                soft_limit_tree.rb_tree_per_node[node] = rtpn;
4935
4936                for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4937                        rtpz = &rtpn->rb_tree_per_zone[zone];
4938                        rtpz->rb_root = RB_ROOT;
4939                        spin_lock_init(&rtpz->lock);
4940                }
4941        }
4942        return 0;
4943
4944err_cleanup:
4945        for_each_node(node) {
4946                if (!soft_limit_tree.rb_tree_per_node[node])
4947                        break;
4948                kfree(soft_limit_tree.rb_tree_per_node[node]);
4949                soft_limit_tree.rb_tree_per_node[node] = NULL;
4950        }
4951        return 1;
4952
4953}
4954
4955static struct cgroup_subsys_state * __ref
4956mem_cgroup_create(struct cgroup *cont)
4957{
4958        struct mem_cgroup *memcg, *parent;
4959        long error = -ENOMEM;
4960        int node;
4961
4962        memcg = mem_cgroup_alloc();
4963        if (!memcg)
4964                return ERR_PTR(error);
4965
4966        for_each_node(node)
4967                if (alloc_mem_cgroup_per_zone_info(memcg, node))
4968                        goto free_out;
4969
4970        /* root ? */
4971        if (cont->parent == NULL) {
4972                int cpu;
4973                enable_swap_cgroup();
4974                parent = NULL;
4975                if (mem_cgroup_soft_limit_tree_init())
4976                        goto free_out;
4977                root_mem_cgroup = memcg;
4978                for_each_possible_cpu(cpu) {
4979                        struct memcg_stock_pcp *stock =
4980                                                &per_cpu(memcg_stock, cpu);
4981                        INIT_WORK(&stock->work, drain_local_stock);
4982                }
4983                hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
4984        } else {
4985                parent = mem_cgroup_from_cont(cont->parent);
4986                memcg->use_hierarchy = parent->use_hierarchy;
4987                memcg->oom_kill_disable = parent->oom_kill_disable;
4988        }
4989
4990        if (parent && parent->use_hierarchy) {
4991                res_counter_init(&memcg->res, &parent->res);
4992                res_counter_init(&memcg->memsw, &parent->memsw);
4993                /*
4994                 * We increment refcnt of the parent to ensure that we can
4995                 * safely access it on res_counter_charge/uncharge.
4996                 * This refcnt will be decremented when freeing this
4997                 * mem_cgroup(see mem_cgroup_put).
4998                 */
4999                mem_cgroup_get(parent);
5000        } else {
5001                res_counter_init(&memcg->res, NULL);
5002                res_counter_init(&memcg->memsw, NULL);
5003                /*
5004                 * Deeper hierachy with use_hierarchy == false doesn't make
5005                 * much sense so let cgroup subsystem know about this
5006                 * unfortunate state in our controller.
5007                 */
5008                if (parent && parent != root_mem_cgroup)
5009                        mem_cgroup_subsys.broken_hierarchy = true;
5010        }
5011        memcg->last_scanned_node = MAX_NUMNODES;
5012        INIT_LIST_HEAD(&memcg->oom_notify);
5013
5014        if (parent)
5015                memcg->swappiness = mem_cgroup_swappiness(parent);
5016        atomic_set(&memcg->refcnt, 1);
5017        memcg->move_charge_at_immigrate = 0;
5018        mutex_init(&memcg->thresholds_lock);
5019        spin_lock_init(&memcg->move_lock);
5020
5021        error = memcg_init_kmem(memcg, &mem_cgroup_subsys);
5022        if (error) {
5023                /*
5024                 * We call put now because our (and parent's) refcnts
5025                 * are already in place. mem_cgroup_put() will internally
5026                 * call __mem_cgroup_free, so return directly
5027                 */
5028                mem_cgroup_put(memcg);
5029                return ERR_PTR(error);
5030        }
5031        return &memcg->css;
5032free_out:
5033        __mem_cgroup_free(memcg);
5034        return ERR_PTR(error);
5035}
5036
5037static int mem_cgroup_pre_destroy(struct cgroup *cont)
5038{
5039        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
5040
5041        return mem_cgroup_force_empty(memcg, false);
5042}
5043
5044static void mem_cgroup_destroy(struct cgroup *cont)
5045{
5046        struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
5047
5048        kmem_cgroup_destroy(memcg);
5049
5050        mem_cgroup_put(memcg);
5051}
5052
5053#ifdef CONFIG_MMU
5054/* Handlers for move charge at task migration. */
5055#define PRECHARGE_COUNT_AT_ONCE 256
5056static int mem_cgroup_do_precharge(unsigned long count)
5057{
5058        int ret = 0;
5059        int batch_count = PRECHARGE_COUNT_AT_ONCE;
5060        struct mem_cgroup *memcg = mc.to;
5061
5062        if (mem_cgroup_is_root(memcg)) {
5063                mc.precharge += count;
5064                /* we don't need css_get for root */
5065                return ret;
5066        }
5067        /* try to charge at once */
5068        if (count > 1) {
5069                struct res_counter *dummy;
5070                /*
5071                 * "memcg" cannot be under rmdir() because we've already checked
5072                 * by cgroup_lock_live_cgroup() that it is not removed and we
5073                 * are still under the same cgroup_mutex. So we can postpone
5074                 * css_get().
5075                 */
5076                if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy))
5077                        goto one_by_one;
5078                if (do_swap_account && res_counter_charge(&memcg->memsw,
5079                                                PAGE_SIZE * count, &dummy)) {
5080                        res_counter_uncharge(&memcg->res, PAGE_SIZE * count);
5081                        goto one_by_one;
5082                }
5083                mc.precharge += count;
5084                return ret;
5085        }
5086one_by_one:
5087        /* fall back to one by one charge */
5088        while (count--) {
5089                if (signal_pending(current)) {
5090                        ret = -EINTR;
5091                        break;
5092                }
5093                if (!batch_count--) {
5094                        batch_count = PRECHARGE_COUNT_AT_ONCE;
5095                        cond_resched();
5096                }
5097                ret = __mem_cgroup_try_charge(NULL,
5098                                        GFP_KERNEL, 1, &memcg, false);
5099                if (ret)
5100                        /* mem_cgroup_clear_mc() will do uncharge later */
5101                        return ret;
5102                mc.precharge++;
5103        }
5104        return ret;
5105}
5106
5107/**
5108 * get_mctgt_type - get target type of moving charge
5109 * @vma: the vma the pte to be checked belongs
5110 * @addr: the address corresponding to the pte to be checked
5111 * @ptent: the pte to be checked
5112 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5113 *
5114 * Returns
5115 *   0(MC_TARGET_NONE): if the pte is not a target for move charge.
5116 *   1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5117 *     move charge. if @target is not NULL, the page is stored in target->page
5118 *     with extra refcnt got(Callers should handle it).
5119 *   2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5120 *     target for charge migration. if @target is not NULL, the entry is stored
5121 *     in target->ent.
5122 *
5123 * Called with pte lock held.
5124 */
5125union mc_target {
5126        struct page     *page;
5127        swp_entry_t     ent;
5128};
5129
5130enum mc_target_type {
5131        MC_TARGET_NONE = 0,
5132        MC_TARGET_PAGE,
5133        MC_TARGET_SWAP,
5134};
5135
5136static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
5137                                                unsigned long addr, pte_t ptent)
5138{
5139        struct page *page = vm_normal_page(vma, addr, ptent);
5140
5141        if (!page || !page_mapped(page))
5142                return NULL;
5143        if (PageAnon(page)) {
5144                /* we don't move shared anon */
5145                if (!move_anon())
5146                        return NULL;
5147        } else if (!move_file())
5148                /* we ignore mapcount for file pages */
5149                return NULL;
5150        if (!get_page_unless_zero(page))
5151                return NULL;
5152
5153        return page;
5154}
5155
5156#ifdef CONFIG_SWAP
5157static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5158                        unsigned long addr, pte_t ptent, swp_entry_t *entry)
5159{
5160        struct page *page = NULL;
5161        swp_entry_t ent = pte_to_swp_entry(ptent);
5162
5163        if (!move_anon() || non_swap_entry(ent))
5164                return NULL;
5165        /*
5166         * Because lookup_swap_cache() updates some statistics counter,
5167         * we call find_get_page() with swapper_space directly.
5168         */
5169        page = find_get_page(&swapper_space, ent.val);
5170        if (do_swap_account)
5171                entry->val = ent.val;
5172
5173        return page;
5174}
5175#else
5176static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
5177                        unsigned long addr, pte_t ptent, swp_entry_t *entry)
5178{
5179        return NULL;
5180}
5181#endif
5182
5183static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
5184                        unsigned long addr, pte_t ptent, swp_entry_t *entry)
5185{
5186        struct page *page = NULL;
5187        struct address_space *mapping;
5188        pgoff_t pgoff;
5189
5190        if (!vma->vm_file) /* anonymous vma */
5191                return NULL;
5192        if (!move_file())
5193                return NULL;
5194
5195        mapping = vma->vm_file->f_mapping;
5196        if (pte_none(ptent))
5197                pgoff = linear_page_index(vma, addr);
5198        else /* pte_file(ptent) is true */
5199                pgoff = pte_to_pgoff(ptent);
5200
5201        /* page is moved even if it's not RSS of this task(page-faulted). */
5202        page = find_get_page(mapping, pgoff);
5203
5204#ifdef CONFIG_SWAP
5205        /* shmem/tmpfs may report page out on swap: account for that too. */
5206        if (radix_tree_exceptional_entry(page)) {
5207                swp_entry_t swap = radix_to_swp_entry(page);
5208                if (do_swap_account)
5209                        *entry = swap;
5210                page = find_get_page(&swapper_space, swap.val);
5211        }
5212#endif
5213        return page;
5214}
5215
5216static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
5217                unsigned long addr, pte_t ptent, union mc_target *target)
5218{
5219        struct page *page = NULL;
5220        struct page_cgroup *pc;
5221        enum mc_target_type ret = MC_TARGET_NONE;
5222        swp_entry_t ent = { .val = 0 };
5223
5224        if (pte_present(ptent))
5225                page = mc_handle_present_pte(vma, addr, ptent);
5226        else if (is_swap_pte(ptent))
5227                page = mc_handle_swap_pte(vma, addr, ptent, &ent);
5228        else if (pte_none(ptent) || pte_file(ptent))
5229                page = mc_handle_file_pte(vma, addr, ptent, &ent);
5230
5231        if (!page && !ent.val)
5232                return ret;
5233        if (page) {
5234                pc = lookup_page_cgroup(page);
5235                /*
5236                 * Do only loose check w/o page_cgroup lock.
5237                 * mem_cgroup_move_account() checks the pc is valid or not under
5238                 * the lock.
5239                 */
5240                if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5241                        ret = MC_TARGET_PAGE;
5242                        if (target)
5243                                target->page = page;
5244                }
5245                if (!ret || !target)
5246                        put_page(page);
5247        }
5248        /* There is a swap entry and a page doesn't exist or isn't charged */
5249        if (ent.val && !ret &&
5250                        css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) {
5251                ret = MC_TARGET_SWAP;
5252                if (target)
5253                        target->ent = ent;
5254        }
5255        return ret;
5256}
5257
5258#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5259/*
5260 * We don't consider swapping or file mapped pages because THP does not
5261 * support them for now.
5262 * Caller should make sure that pmd_trans_huge(pmd) is true.
5263 */
5264static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5265                unsigned long addr, pmd_t pmd, union mc_target *target)
5266{
5267        struct page *page = NULL;
5268        struct page_cgroup *pc;
5269        enum mc_target_type ret = MC_TARGET_NONE;
5270
5271        page = pmd_page(pmd);
5272        VM_BUG_ON(!page || !PageHead(page));
5273        if (!move_anon())
5274                return ret;
5275        pc = lookup_page_cgroup(page);
5276        if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
5277                ret = MC_TARGET_PAGE;
5278                if (target) {
5279                        get_page(page);
5280                        target->page = page;
5281                }
5282        }
5283        return ret;
5284}
5285#else
5286static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
5287                unsigned long addr, pmd_t pmd, union mc_target *target)
5288{
5289        return MC_TARGET_NONE;
5290}
5291#endif
5292
5293static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
5294                                        unsigned long addr, unsigned long end,
5295                                        struct mm_walk *walk)
5296{
5297        struct vm_area_struct *vma = walk->private;
5298        pte_t *pte;
5299        spinlock_t *ptl;
5300
5301        if (pmd_trans_huge_lock(pmd, vma) == 1) {
5302                if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
5303                        mc.precharge += HPAGE_PMD_NR;
5304                spin_unlock(&vma->vm_mm->page_table_lock);
5305                return 0;
5306        }
5307
5308        if (pmd_trans_unstable(pmd))
5309                return 0;
5310        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5311        for (; addr != end; pte++, addr += PAGE_SIZE)
5312                if (get_mctgt_type(vma, addr, *pte, NULL))
5313                        mc.precharge++; /* increment precharge temporarily */
5314        pte_unmap_unlock(pte - 1, ptl);
5315        cond_resched();
5316
5317        return 0;
5318}
5319
5320static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
5321{
5322        unsigned long precharge;
5323        struct vm_area_struct *vma;
5324
5325        down_read(&mm->mmap_sem);
5326        for (vma = mm->mmap; vma; vma = vma->vm_next) {
5327                struct mm_walk mem_cgroup_count_precharge_walk = {
5328                        .pmd_entry = mem_cgroup_count_precharge_pte_range,
5329                        .mm = mm,
5330                        .private = vma,
5331                };
5332                if (is_vm_hugetlb_page(vma))
5333                        continue;
5334                walk_page_range(vma->vm_start, vma->vm_end,
5335                                        &mem_cgroup_count_precharge_walk);
5336        }
5337        up_read(&mm->mmap_sem);
5338
5339        precharge = mc.precharge;
5340        mc.precharge = 0;
5341
5342        return precharge;
5343}
5344
5345static int mem_cgroup_precharge_mc(struct mm_struct *mm)
5346{
5347        unsigned long precharge = mem_cgroup_count_precharge(mm);
5348
5349        VM_BUG_ON(mc.moving_task);
5350        mc.moving_task = current;
5351        return mem_cgroup_do_precharge(precharge);
5352}
5353
5354/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5355static void __mem_cgroup_clear_mc(void)
5356{
5357        struct mem_cgroup *from = mc.from;
5358        struct mem_cgroup *to = mc.to;
5359
5360        /* we must uncharge all the leftover precharges from mc.to */
5361        if (mc.precharge) {
5362                __mem_cgroup_cancel_charge(mc.to, mc.precharge);
5363                mc.precharge = 0;
5364        }
5365        /*
5366         * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5367         * we must uncharge here.
5368         */
5369        if (mc.moved_charge) {
5370                __mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
5371                mc.moved_charge = 0;
5372        }
5373        /* we must fixup refcnts and charges */
5374        if (mc.moved_swap) {
5375                /* uncharge swap account from the old cgroup */
5376                if (!mem_cgroup_is_root(mc.from))
5377                        res_counter_uncharge(&mc.from->memsw,
5378                                                PAGE_SIZE * mc.moved_swap);
5379                __mem_cgroup_put(mc.from, mc.moved_swap);
5380
5381                if (!mem_cgroup_is_root(mc.to)) {
5382                        /*
5383                         * we charged both to->res and to->memsw, so we should
5384                         * uncharge to->res.
5385                         */
5386                        res_counter_uncharge(&mc.to->res,
5387                                                PAGE_SIZE * mc.moved_swap);
5388                }
5389                /* we've already done mem_cgroup_get(mc.to) */
5390                mc.moved_swap = 0;
5391        }
5392        memcg_oom_recover(from);
5393        memcg_oom_recover(to);
5394        wake_up_all(&mc.waitq);
5395}
5396
5397static void mem_cgroup_clear_mc(void)
5398{
5399        struct mem_cgroup *from = mc.from;
5400
5401        /*
5402         * we must clear moving_task before waking up waiters at the end of
5403         * task migration.
5404         */
5405        mc.moving_task = NULL;
5406        __mem_cgroup_clear_mc();
5407        spin_lock(&mc.lock);
5408        mc.from = NULL;
5409        mc.to = NULL;
5410        spin_unlock(&mc.lock);
5411        mem_cgroup_end_move(from);
5412}
5413
5414static int mem_cgroup_can_attach(struct cgroup *cgroup,
5415                                 struct cgroup_taskset *tset)
5416{
5417        struct task_struct *p = cgroup_taskset_first(tset);
5418        int ret = 0;
5419        struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup);
5420
5421        if (memcg->move_charge_at_immigrate) {
5422                struct mm_struct *mm;
5423                struct mem_cgroup *from = mem_cgroup_from_task(p);
5424
5425                VM_BUG_ON(from == memcg);
5426
5427                mm = get_task_mm(p);
5428                if (!mm)
5429                        return 0;
5430                /* We move charges only when we move a owner of the mm */
5431                if (mm->owner == p) {
5432                        VM_BUG_ON(mc.from);
5433                        VM_BUG_ON(mc.to);
5434                        VM_BUG_ON(mc.precharge);
5435                        VM_BUG_ON(mc.moved_charge);
5436                        VM_BUG_ON(mc.moved_swap);
5437                        mem_cgroup_start_move(from);
5438                        spin_lock(&mc.lock);
5439                        mc.from = from;
5440                        mc.to = memcg;
5441                        spin_unlock(&mc.lock);
5442                        /* We set mc.moving_task later */
5443
5444                        ret = mem_cgroup_precharge_mc(mm);
5445                        if (ret)
5446                                mem_cgroup_clear_mc();
5447                }
5448                mmput(mm);
5449        }
5450        return ret;
5451}
5452
5453static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5454                                     struct cgroup_taskset *tset)
5455{
5456        mem_cgroup_clear_mc();
5457}
5458
5459static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5460                                unsigned long addr, unsigned long end,
5461                                struct mm_walk *walk)
5462{
5463        int ret = 0;
5464        struct vm_area_struct *vma = walk->private;
5465        pte_t *pte;
5466        spinlock_t *ptl;
5467        enum mc_target_type target_type;
5468        union mc_target target;
5469        struct page *page;
5470        struct page_cgroup *pc;
5471
5472        /*
5473         * We don't take compound_lock() here but no race with splitting thp
5474         * happens because:
5475         *  - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5476         *    under splitting, which means there's no concurrent thp split,
5477         *  - if another thread runs into split_huge_page() just after we
5478         *    entered this if-block, the thread must wait for page table lock
5479         *    to be unlocked in __split_huge_page_splitting(), where the main
5480         *    part of thp split is not executed yet.
5481         */
5482        if (pmd_trans_huge_lock(pmd, vma) == 1) {
5483                if (mc.precharge < HPAGE_PMD_NR) {
5484                        spin_unlock(&vma->vm_mm->page_table_lock);
5485                        return 0;
5486                }
5487                target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
5488                if (target_type == MC_TARGET_PAGE) {
5489                        page = target.page;
5490                        if (!isolate_lru_page(page)) {
5491                                pc = lookup_page_cgroup(page);
5492                                if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
5493                                                        pc, mc.from, mc.to)) {
5494                                        mc.precharge -= HPAGE_PMD_NR;
5495                                        mc.moved_charge += HPAGE_PMD_NR;
5496                                }
5497                                putback_lru_page(page);
5498                        }
5499                        put_page(page);
5500                }
5501                spin_unlock(&vma->vm_mm->page_table_lock);
5502                return 0;
5503        }
5504
5505        if (pmd_trans_unstable(pmd))
5506                return 0;
5507retry:
5508        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5509        for (; addr != end; addr += PAGE_SIZE) {
5510                pte_t ptent = *(pte++);
5511                swp_entry_t ent;
5512
5513                if (!mc.precharge)
5514                        break;
5515
5516                switch (get_mctgt_type(vma, addr, ptent, &target)) {
5517                case MC_TARGET_PAGE:
5518                        page = target.page;
5519                        if (isolate_lru_page(page))
5520                                goto put;
5521                        pc = lookup_page_cgroup(page);
5522                        if (!mem_cgroup_move_account(page, 1, pc,
5523                                                     mc.from, mc.to)) {
5524                                mc.precharge--;
5525                                /* we uncharge from mc.from later. */
5526                                mc.moved_charge++;
5527                        }
5528                        putback_lru_page(page);
5529put:                    /* get_mctgt_type() gets the page */
5530                        put_page(page);
5531                        break;
5532                case MC_TARGET_SWAP:
5533                        ent = target.ent;
5534                        if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
5535                                mc.precharge--;
5536                                /* we fixup refcnts and charges later. */
5537                                mc.moved_swap++;
5538                        }
5539                        break;
5540                default:
5541                        break;
5542                }
5543        }
5544        pte_unmap_unlock(pte - 1, ptl);
5545        cond_resched();
5546
5547        if (addr != end) {
5548                /*
5549                 * We have consumed all precharges we got in can_attach().
5550                 * We try charge one by one, but don't do any additional
5551                 * charges to mc.to if we have failed in charge once in attach()
5552                 * phase.
5553                 */
5554                ret = mem_cgroup_do_precharge(1);
5555                if (!ret)
5556                        goto retry;
5557        }
5558
5559        return ret;
5560}
5561
5562static void mem_cgroup_move_charge(struct mm_struct *mm)
5563{
5564        struct vm_area_struct *vma;
5565
5566        lru_add_drain_all();
5567retry:
5568        if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5569                /*
5570                 * Someone who are holding the mmap_sem might be waiting in
5571                 * waitq. So we cancel all extra charges, wake up all waiters,
5572                 * and retry. Because we cancel precharges, we might not be able
5573                 * to move enough charges, but moving charge is a best-effort
5574                 * feature anyway, so it wouldn't be a big problem.
5575                 */
5576                __mem_cgroup_clear_mc();
5577                cond_resched();
5578                goto retry;
5579        }
5580        for (vma = mm->mmap; vma; vma = vma->vm_next) {
5581                int ret;
5582                struct mm_walk mem_cgroup_move_charge_walk = {
5583                        .pmd_entry = mem_cgroup_move_charge_pte_range,
5584                        .mm = mm,
5585                        .private = vma,
5586                };
5587                if (is_vm_hugetlb_page(vma))
5588                        continue;
5589                ret = walk_page_range(vma->vm_start, vma->vm_end,
5590                                                &mem_cgroup_move_charge_walk);
5591                if (ret)
5592                        /*
5593                         * means we have consumed all precharges and failed in
5594                         * doing additional charge. Just abandon here.
5595                         */
5596                        break;
5597        }
5598        up_read(&mm->mmap_sem);
5599}
5600
5601static void mem_cgroup_move_task(struct cgroup *cont,
5602                                 struct cgroup_taskset *tset)
5603{
5604        struct task_struct *p = cgroup_taskset_first(tset);
5605        struct mm_struct *mm = get_task_mm(p);
5606
5607        if (mm) {
5608                if (mc.to)
5609                        mem_cgroup_move_charge(mm);
5610                mmput(mm);
5611        }
5612        if (mc.to)
5613                mem_cgroup_clear_mc();
5614}
5615#else   /* !CONFIG_MMU */
5616static int mem_cgroup_can_attach(struct cgroup *cgroup,
5617                                 struct cgroup_taskset *tset)
5618{
5619        return 0;
5620}
5621static void mem_cgroup_cancel_attach(struct cgroup *cgroup,
5622                                     struct cgroup_taskset *tset)
5623{
5624}
5625static void mem_cgroup_move_task(struct cgroup *cont,
5626                                 struct cgroup_taskset *tset)
5627{
5628}
5629#endif
5630
5631struct cgroup_subsys mem_cgroup_subsys = {
5632        .name = "memory",
5633        .subsys_id = mem_cgroup_subsys_id,
5634        .create = mem_cgroup_create,
5635        .pre_destroy = mem_cgroup_pre_destroy,
5636        .destroy = mem_cgroup_destroy,
5637        .can_attach = mem_cgroup_can_attach,
5638        .cancel_attach = mem_cgroup_cancel_attach,
5639        .attach = mem_cgroup_move_task,
5640        .base_cftypes = mem_cgroup_files,
5641        .early_init = 0,
5642        .use_id = 1,
5643        .__DEPRECATED_clear_css_refs = true,
5644};
5645
5646#ifdef CONFIG_MEMCG_SWAP
5647static int __init enable_swap_account(char *s)
5648{
5649        /* consider enabled if no parameter or 1 is given */
5650        if (!strcmp(s, "1"))
5651                really_do_swap_account = 1;
5652        else if (!strcmp(s, "0"))
5653                really_do_swap_account = 0;
5654        return 1;
5655}
5656__setup("swapaccount=", enable_swap_account);
5657
5658#endif
5659
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.