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 * Kernel Memory Controller
  14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
  15 * Authors: Glauber Costa and Suleiman Souhlal
  16 *
  17 * Native page reclaim
  18 * Charge lifetime sanitation
  19 * Lockless page tracking & accounting
  20 * Unified hierarchy configuration model
  21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
  22 *
  23 * This program is free software; you can redistribute it and/or modify
  24 * it under the terms of the GNU General Public License as published by
  25 * the Free Software Foundation; either version 2 of the License, or
  26 * (at your option) any later version.
  27 *
  28 * This program is distributed in the hope that it will be useful,
  29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  31 * GNU General Public License for more details.
  32 */
  33
  34#include <linux/page_counter.h>
  35#include <linux/memcontrol.h>
  36#include <linux/cgroup.h>
  37#include <linux/mm.h>
  38#include <linux/hugetlb.h>
  39#include <linux/pagemap.h>
  40#include <linux/smp.h>
  41#include <linux/page-flags.h>
  42#include <linux/backing-dev.h>
  43#include <linux/bit_spinlock.h>
  44#include <linux/rcupdate.h>
  45#include <linux/limits.h>
  46#include <linux/export.h>
  47#include <linux/mutex.h>
  48#include <linux/rbtree.h>
  49#include <linux/slab.h>
  50#include <linux/swap.h>
  51#include <linux/swapops.h>
  52#include <linux/spinlock.h>
  53#include <linux/eventfd.h>
  54#include <linux/poll.h>
  55#include <linux/sort.h>
  56#include <linux/fs.h>
  57#include <linux/seq_file.h>
  58#include <linux/vmpressure.h>
  59#include <linux/mm_inline.h>
  60#include <linux/swap_cgroup.h>
  61#include <linux/cpu.h>
  62#include <linux/oom.h>
  63#include <linux/lockdep.h>
  64#include <linux/file.h>
  65#include <linux/tracehook.h>
  66#include "internal.h"
  67#include <net/sock.h>
  68#include <net/ip.h>
  69#include "slab.h"
  70
  71#include <asm/uaccess.h>
  72
  73#include <trace/events/vmscan.h>
  74
  75struct cgroup_subsys memory_cgrp_subsys __read_mostly;
  76EXPORT_SYMBOL(memory_cgrp_subsys);
  77
  78struct mem_cgroup *root_mem_cgroup __read_mostly;
  79
  80#define MEM_CGROUP_RECLAIM_RETRIES      5
  81
  82/* Socket memory accounting disabled? */
  83static bool cgroup_memory_nosocket;
  84
  85/* Kernel memory accounting disabled? */
  86static bool cgroup_memory_nokmem;
  87
  88/* Whether the swap controller is active */
  89#ifdef CONFIG_MEMCG_SWAP
  90int do_swap_account __read_mostly;
  91#else
  92#define do_swap_account         0
  93#endif
  94
  95/* Whether legacy memory+swap accounting is active */
  96static bool do_memsw_account(void)
  97{
  98        return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account;
  99}
 100
 101static const char * const mem_cgroup_stat_names[] = {
 102        "cache",
 103        "rss",
 104        "rss_huge",
 105        "mapped_file",
 106        "dirty",
 107        "writeback",
 108        "swap",
 109};
 110
 111static const char * const mem_cgroup_events_names[] = {
 112        "pgpgin",
 113        "pgpgout",
 114        "pgfault",
 115        "pgmajfault",
 116};
 117
 118static const char * const mem_cgroup_lru_names[] = {
 119        "inactive_anon",
 120        "active_anon",
 121        "inactive_file",
 122        "active_file",
 123        "unevictable",
 124};
 125
 126#define THRESHOLDS_EVENTS_TARGET 128
 127#define SOFTLIMIT_EVENTS_TARGET 1024
 128#define NUMAINFO_EVENTS_TARGET  1024
 129
 130/*
 131 * Cgroups above their limits are maintained in a RB-Tree, independent of
 132 * their hierarchy representation
 133 */
 134
 135struct mem_cgroup_tree_per_node {
 136        struct rb_root rb_root;
 137        spinlock_t lock;
 138};
 139
 140struct mem_cgroup_tree {
 141        struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
 142};
 143
 144static struct mem_cgroup_tree soft_limit_tree __read_mostly;
 145
 146/* for OOM */
 147struct mem_cgroup_eventfd_list {
 148        struct list_head list;
 149        struct eventfd_ctx *eventfd;
 150};
 151
 152/*
 153 * cgroup_event represents events which userspace want to receive.
 154 */
 155struct mem_cgroup_event {
 156        /*
 157         * memcg which the event belongs to.
 158         */
 159        struct mem_cgroup *memcg;
 160        /*
 161         * eventfd to signal userspace about the event.
 162         */
 163        struct eventfd_ctx *eventfd;
 164        /*
 165         * Each of these stored in a list by the cgroup.
 166         */
 167        struct list_head list;
 168        /*
 169         * register_event() callback will be used to add new userspace
 170         * waiter for changes related to this event.  Use eventfd_signal()
 171         * on eventfd to send notification to userspace.
 172         */
 173        int (*register_event)(struct mem_cgroup *memcg,
 174                              struct eventfd_ctx *eventfd, const char *args);
 175        /*
 176         * unregister_event() callback will be called when userspace closes
 177         * the eventfd or on cgroup removing.  This callback must be set,
 178         * if you want provide notification functionality.
 179         */
 180        void (*unregister_event)(struct mem_cgroup *memcg,
 181                                 struct eventfd_ctx *eventfd);
 182        /*
 183         * All fields below needed to unregister event when
 184         * userspace closes eventfd.
 185         */
 186        poll_table pt;
 187        wait_queue_head_t *wqh;
 188        wait_queue_t wait;
 189        struct work_struct remove;
 190};
 191
 192static void mem_cgroup_threshold(struct mem_cgroup *memcg);
 193static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
 194
 195/* Stuffs for move charges at task migration. */
 196/*
 197 * Types of charges to be moved.
 198 */
 199#define MOVE_ANON       0x1U
 200#define MOVE_FILE       0x2U
 201#define MOVE_MASK       (MOVE_ANON | MOVE_FILE)
 202
 203/* "mc" and its members are protected by cgroup_mutex */
 204static struct move_charge_struct {
 205        spinlock_t        lock; /* for from, to */
 206        struct mm_struct  *mm;
 207        struct mem_cgroup *from;
 208        struct mem_cgroup *to;
 209        unsigned long flags;
 210        unsigned long precharge;
 211        unsigned long moved_charge;
 212        unsigned long moved_swap;
 213        struct task_struct *moving_task;        /* a task moving charges */
 214        wait_queue_head_t waitq;                /* a waitq for other context */
 215} mc = {
 216        .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
 217        .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
 218};
 219
 220/*
 221 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
 222 * limit reclaim to prevent infinite loops, if they ever occur.
 223 */
 224#define MEM_CGROUP_MAX_RECLAIM_LOOPS            100
 225#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
 226
 227enum charge_type {
 228        MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
 229        MEM_CGROUP_CHARGE_TYPE_ANON,
 230        MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
 231        MEM_CGROUP_CHARGE_TYPE_DROP,    /* a page was unused swap cache */
 232        NR_CHARGE_TYPE,
 233};
 234
 235/* for encoding cft->private value on file */
 236enum res_type {
 237        _MEM,
 238        _MEMSWAP,
 239        _OOM_TYPE,
 240        _KMEM,
 241        _TCP,
 242};
 243
 244#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
 245#define MEMFILE_TYPE(val)       ((val) >> 16 & 0xffff)
 246#define MEMFILE_ATTR(val)       ((val) & 0xffff)
 247/* Used for OOM nofiier */
 248#define OOM_CONTROL             (0)
 249
 250/* Some nice accessors for the vmpressure. */
 251struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
 252{
 253        if (!memcg)
 254                memcg = root_mem_cgroup;
 255        return &memcg->vmpressure;
 256}
 257
 258struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
 259{
 260        return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
 261}
 262
 263static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
 264{
 265        return (memcg == root_mem_cgroup);
 266}
 267
 268#ifndef CONFIG_SLOB
 269/*
 270 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
 271 * The main reason for not using cgroup id for this:
 272 *  this works better in sparse environments, where we have a lot of memcgs,
 273 *  but only a few kmem-limited. Or also, if we have, for instance, 200
 274 *  memcgs, and none but the 200th is kmem-limited, we'd have to have a
 275 *  200 entry array for that.
 276 *
 277 * The current size of the caches array is stored in memcg_nr_cache_ids. It
 278 * will double each time we have to increase it.
 279 */
 280static DEFINE_IDA(memcg_cache_ida);
 281int memcg_nr_cache_ids;
 282
 283/* Protects memcg_nr_cache_ids */
 284static DECLARE_RWSEM(memcg_cache_ids_sem);
 285
 286void memcg_get_cache_ids(void)
 287{
 288        down_read(&memcg_cache_ids_sem);
 289}
 290
 291void memcg_put_cache_ids(void)
 292{
 293        up_read(&memcg_cache_ids_sem);
 294}
 295
 296/*
 297 * MIN_SIZE is different than 1, because we would like to avoid going through
 298 * the alloc/free process all the time. In a small machine, 4 kmem-limited
 299 * cgroups is a reasonable guess. In the future, it could be a parameter or
 300 * tunable, but that is strictly not necessary.
 301 *
 302 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
 303 * this constant directly from cgroup, but it is understandable that this is
 304 * better kept as an internal representation in cgroup.c. In any case, the
 305 * cgrp_id space is not getting any smaller, and we don't have to necessarily
 306 * increase ours as well if it increases.
 307 */
 308#define MEMCG_CACHES_MIN_SIZE 4
 309#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
 310
 311/*
 312 * A lot of the calls to the cache allocation functions are expected to be
 313 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
 314 * conditional to this static branch, we'll have to allow modules that does
 315 * kmem_cache_alloc and the such to see this symbol as well
 316 */
 317DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
 318EXPORT_SYMBOL(memcg_kmem_enabled_key);
 319
 320#endif /* !CONFIG_SLOB */
 321
 322/**
 323 * mem_cgroup_css_from_page - css of the memcg associated with a page
 324 * @page: page of interest
 325 *
 326 * If memcg is bound to the default hierarchy, css of the memcg associated
 327 * with @page is returned.  The returned css remains associated with @page
 328 * until it is released.
 329 *
 330 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
 331 * is returned.
 332 */
 333struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
 334{
 335        struct mem_cgroup *memcg;
 336
 337        memcg = page->mem_cgroup;
 338
 339        if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
 340                memcg = root_mem_cgroup;
 341
 342        return &memcg->css;
 343}
 344
 345/**
 346 * page_cgroup_ino - return inode number of the memcg a page is charged to
 347 * @page: the page
 348 *
 349 * Look up the closest online ancestor of the memory cgroup @page is charged to
 350 * and return its inode number or 0 if @page is not charged to any cgroup. It
 351 * is safe to call this function without holding a reference to @page.
 352 *
 353 * Note, this function is inherently racy, because there is nothing to prevent
 354 * the cgroup inode from getting torn down and potentially reallocated a moment
 355 * after page_cgroup_ino() returns, so it only should be used by callers that
 356 * do not care (such as procfs interfaces).
 357 */
 358ino_t page_cgroup_ino(struct page *page)
 359{
 360        struct mem_cgroup *memcg;
 361        unsigned long ino = 0;
 362
 363        rcu_read_lock();
 364        memcg = READ_ONCE(page->mem_cgroup);
 365        while (memcg && !(memcg->css.flags & CSS_ONLINE))
 366                memcg = parent_mem_cgroup(memcg);
 367        if (memcg)
 368                ino = cgroup_ino(memcg->css.cgroup);
 369        rcu_read_unlock();
 370        return ino;
 371}
 372
 373static struct mem_cgroup_per_node *
 374mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page)
 375{
 376        int nid = page_to_nid(page);
 377
 378        return memcg->nodeinfo[nid];
 379}
 380
 381static struct mem_cgroup_tree_per_node *
 382soft_limit_tree_node(int nid)
 383{
 384        return soft_limit_tree.rb_tree_per_node[nid];
 385}
 386
 387static struct mem_cgroup_tree_per_node *
 388soft_limit_tree_from_page(struct page *page)
 389{
 390        int nid = page_to_nid(page);
 391
 392        return soft_limit_tree.rb_tree_per_node[nid];
 393}
 394
 395static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz,
 396                                         struct mem_cgroup_tree_per_node *mctz,
 397                                         unsigned long new_usage_in_excess)
 398{
 399        struct rb_node **p = &mctz->rb_root.rb_node;
 400        struct rb_node *parent = NULL;
 401        struct mem_cgroup_per_node *mz_node;
 402
 403        if (mz->on_tree)
 404                return;
 405
 406        mz->usage_in_excess = new_usage_in_excess;
 407        if (!mz->usage_in_excess)
 408                return;
 409        while (*p) {
 410                parent = *p;
 411                mz_node = rb_entry(parent, struct mem_cgroup_per_node,
 412                                        tree_node);
 413                if (mz->usage_in_excess < mz_node->usage_in_excess)
 414                        p = &(*p)->rb_left;
 415                /*
 416                 * We can't avoid mem cgroups that are over their soft
 417                 * limit by the same amount
 418                 */
 419                else if (mz->usage_in_excess >= mz_node->usage_in_excess)
 420                        p = &(*p)->rb_right;
 421        }
 422        rb_link_node(&mz->tree_node, parent, p);
 423        rb_insert_color(&mz->tree_node, &mctz->rb_root);
 424        mz->on_tree = true;
 425}
 426
 427static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
 428                                         struct mem_cgroup_tree_per_node *mctz)
 429{
 430        if (!mz->on_tree)
 431                return;
 432        rb_erase(&mz->tree_node, &mctz->rb_root);
 433        mz->on_tree = false;
 434}
 435
 436static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
 437                                       struct mem_cgroup_tree_per_node *mctz)
 438{
 439        unsigned long flags;
 440
 441        spin_lock_irqsave(&mctz->lock, flags);
 442        __mem_cgroup_remove_exceeded(mz, mctz);
 443        spin_unlock_irqrestore(&mctz->lock, flags);
 444}
 445
 446static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
 447{
 448        unsigned long nr_pages = page_counter_read(&memcg->memory);
 449        unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
 450        unsigned long excess = 0;
 451
 452        if (nr_pages > soft_limit)
 453                excess = nr_pages - soft_limit;
 454
 455        return excess;
 456}
 457
 458static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
 459{
 460        unsigned long excess;
 461        struct mem_cgroup_per_node *mz;
 462        struct mem_cgroup_tree_per_node *mctz;
 463
 464        mctz = soft_limit_tree_from_page(page);
 465        /*
 466         * Necessary to update all ancestors when hierarchy is used.
 467         * because their event counter is not touched.
 468         */
 469        for (; memcg; memcg = parent_mem_cgroup(memcg)) {
 470                mz = mem_cgroup_page_nodeinfo(memcg, page);
 471                excess = soft_limit_excess(memcg);
 472                /*
 473                 * We have to update the tree if mz is on RB-tree or
 474                 * mem is over its softlimit.
 475                 */
 476                if (excess || mz->on_tree) {
 477                        unsigned long flags;
 478
 479                        spin_lock_irqsave(&mctz->lock, flags);
 480                        /* if on-tree, remove it */
 481                        if (mz->on_tree)
 482                                __mem_cgroup_remove_exceeded(mz, mctz);
 483                        /*
 484                         * Insert again. mz->usage_in_excess will be updated.
 485                         * If excess is 0, no tree ops.
 486                         */
 487                        __mem_cgroup_insert_exceeded(mz, mctz, excess);
 488                        spin_unlock_irqrestore(&mctz->lock, flags);
 489                }
 490        }
 491}
 492
 493static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
 494{
 495        struct mem_cgroup_tree_per_node *mctz;
 496        struct mem_cgroup_per_node *mz;
 497        int nid;
 498
 499        for_each_node(nid) {
 500                mz = mem_cgroup_nodeinfo(memcg, nid);
 501                mctz = soft_limit_tree_node(nid);
 502                mem_cgroup_remove_exceeded(mz, mctz);
 503        }
 504}
 505
 506static struct mem_cgroup_per_node *
 507__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
 508{
 509        struct rb_node *rightmost = NULL;
 510        struct mem_cgroup_per_node *mz;
 511
 512retry:
 513        mz = NULL;
 514        rightmost = rb_last(&mctz->rb_root);
 515        if (!rightmost)
 516                goto done;              /* Nothing to reclaim from */
 517
 518        mz = rb_entry(rightmost, struct mem_cgroup_per_node, tree_node);
 519        /*
 520         * Remove the node now but someone else can add it back,
 521         * we will to add it back at the end of reclaim to its correct
 522         * position in the tree.
 523         */
 524        __mem_cgroup_remove_exceeded(mz, mctz);
 525        if (!soft_limit_excess(mz->memcg) ||
 526            !css_tryget_online(&mz->memcg->css))
 527                goto retry;
 528done:
 529        return mz;
 530}
 531
 532static struct mem_cgroup_per_node *
 533mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
 534{
 535        struct mem_cgroup_per_node *mz;
 536
 537        spin_lock_irq(&mctz->lock);
 538        mz = __mem_cgroup_largest_soft_limit_node(mctz);
 539        spin_unlock_irq(&mctz->lock);
 540        return mz;
 541}
 542
 543/*
 544 * Return page count for single (non recursive) @memcg.
 545 *
 546 * Implementation Note: reading percpu statistics for memcg.
 547 *
 548 * Both of vmstat[] and percpu_counter has threshold and do periodic
 549 * synchronization to implement "quick" read. There are trade-off between
 550 * reading cost and precision of value. Then, we may have a chance to implement
 551 * a periodic synchronization of counter in memcg's counter.
 552 *
 553 * But this _read() function is used for user interface now. The user accounts
 554 * memory usage by memory cgroup and he _always_ requires exact value because
 555 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
 556 * have to visit all online cpus and make sum. So, for now, unnecessary
 557 * synchronization is not implemented. (just implemented for cpu hotplug)
 558 *
 559 * If there are kernel internal actions which can make use of some not-exact
 560 * value, and reading all cpu value can be performance bottleneck in some
 561 * common workload, threshold and synchronization as vmstat[] should be
 562 * implemented.
 563 */
 564static unsigned long
 565mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx)
 566{
 567        long val = 0;
 568        int cpu;
 569
 570        /* Per-cpu values can be negative, use a signed accumulator */
 571        for_each_possible_cpu(cpu)
 572                val += per_cpu(memcg->stat->count[idx], cpu);
 573        /*
 574         * Summing races with updates, so val may be negative.  Avoid exposing
 575         * transient negative values.
 576         */
 577        if (val < 0)
 578                val = 0;
 579        return val;
 580}
 581
 582static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
 583                                            enum mem_cgroup_events_index idx)
 584{
 585        unsigned long val = 0;
 586        int cpu;
 587
 588        for_each_possible_cpu(cpu)
 589                val += per_cpu(memcg->stat->events[idx], cpu);
 590        return val;
 591}
 592
 593static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
 594                                         struct page *page,
 595                                         bool compound, int nr_pages)
 596{
 597        /*
 598         * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
 599         * counted as CACHE even if it's on ANON LRU.
 600         */
 601        if (PageAnon(page))
 602                __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
 603                                nr_pages);
 604        else
 605                __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
 606                                nr_pages);
 607
 608        if (compound) {
 609                VM_BUG_ON_PAGE(!PageTransHuge(page), page);
 610                __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
 611                                nr_pages);
 612        }
 613
 614        /* pagein of a big page is an event. So, ignore page size */
 615        if (nr_pages > 0)
 616                __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
 617        else {
 618                __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
 619                nr_pages = -nr_pages; /* for event */
 620        }
 621
 622        __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
 623}
 624
 625unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
 626                                           int nid, unsigned int lru_mask)
 627{
 628        unsigned long nr = 0;
 629        struct mem_cgroup_per_node *mz;
 630        enum lru_list lru;
 631
 632        VM_BUG_ON((unsigned)nid >= nr_node_ids);
 633
 634        for_each_lru(lru) {
 635                if (!(BIT(lru) & lru_mask))
 636                        continue;
 637                mz = mem_cgroup_nodeinfo(memcg, nid);
 638                nr += mz->lru_size[lru];
 639        }
 640        return nr;
 641}
 642
 643static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
 644                        unsigned int lru_mask)
 645{
 646        unsigned long nr = 0;
 647        int nid;
 648
 649        for_each_node_state(nid, N_MEMORY)
 650                nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
 651        return nr;
 652}
 653
 654static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
 655                                       enum mem_cgroup_events_target target)
 656{
 657        unsigned long val, next;
 658
 659        val = __this_cpu_read(memcg->stat->nr_page_events);
 660        next = __this_cpu_read(memcg->stat->targets[target]);
 661        /* from time_after() in jiffies.h */
 662        if ((long)next - (long)val < 0) {
 663                switch (target) {
 664                case MEM_CGROUP_TARGET_THRESH:
 665                        next = val + THRESHOLDS_EVENTS_TARGET;
 666                        break;
 667                case MEM_CGROUP_TARGET_SOFTLIMIT:
 668                        next = val + SOFTLIMIT_EVENTS_TARGET;
 669                        break;
 670                case MEM_CGROUP_TARGET_NUMAINFO:
 671                        next = val + NUMAINFO_EVENTS_TARGET;
 672                        break;
 673                default:
 674                        break;
 675                }
 676                __this_cpu_write(memcg->stat->targets[target], next);
 677                return true;
 678        }
 679        return false;
 680}
 681
 682/*
 683 * Check events in order.
 684 *
 685 */
 686static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
 687{
 688        /* threshold event is triggered in finer grain than soft limit */
 689        if (unlikely(mem_cgroup_event_ratelimit(memcg,
 690                                                MEM_CGROUP_TARGET_THRESH))) {
 691                bool do_softlimit;
 692                bool do_numainfo __maybe_unused;
 693
 694                do_softlimit = mem_cgroup_event_ratelimit(memcg,
 695                                                MEM_CGROUP_TARGET_SOFTLIMIT);
 696#if MAX_NUMNODES > 1
 697                do_numainfo = mem_cgroup_event_ratelimit(memcg,
 698                                                MEM_CGROUP_TARGET_NUMAINFO);
 699#endif
 700                mem_cgroup_threshold(memcg);
 701                if (unlikely(do_softlimit))
 702                        mem_cgroup_update_tree(memcg, page);
 703#if MAX_NUMNODES > 1
 704                if (unlikely(do_numainfo))
 705                        atomic_inc(&memcg->numainfo_events);
 706#endif
 707        }
 708}
 709
 710struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
 711{
 712        /*
 713         * mm_update_next_owner() may clear mm->owner to NULL
 714         * if it races with swapoff, page migration, etc.
 715         * So this can be called with p == NULL.
 716         */
 717        if (unlikely(!p))
 718                return NULL;
 719
 720        return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
 721}
 722EXPORT_SYMBOL(mem_cgroup_from_task);
 723
 724static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
 725{
 726        struct mem_cgroup *memcg = NULL;
 727
 728        rcu_read_lock();
 729        do {
 730                /*
 731                 * Page cache insertions can happen withou an
 732                 * actual mm context, e.g. during disk probing
 733                 * on boot, loopback IO, acct() writes etc.
 734                 */
 735                if (unlikely(!mm))
 736                        memcg = root_mem_cgroup;
 737                else {
 738                        memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 739                        if (unlikely(!memcg))
 740                                memcg = root_mem_cgroup;
 741                }
 742        } while (!css_tryget_online(&memcg->css));
 743        rcu_read_unlock();
 744        return memcg;
 745}
 746
 747/**
 748 * mem_cgroup_iter - iterate over memory cgroup hierarchy
 749 * @root: hierarchy root
 750 * @prev: previously returned memcg, NULL on first invocation
 751 * @reclaim: cookie for shared reclaim walks, NULL for full walks
 752 *
 753 * Returns references to children of the hierarchy below @root, or
 754 * @root itself, or %NULL after a full round-trip.
 755 *
 756 * Caller must pass the return value in @prev on subsequent
 757 * invocations for reference counting, or use mem_cgroup_iter_break()
 758 * to cancel a hierarchy walk before the round-trip is complete.
 759 *
 760 * Reclaimers can specify a zone and a priority level in @reclaim to
 761 * divide up the memcgs in the hierarchy among all concurrent
 762 * reclaimers operating on the same zone and priority.
 763 */
 764struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
 765                                   struct mem_cgroup *prev,
 766                                   struct mem_cgroup_reclaim_cookie *reclaim)
 767{
 768        struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
 769        struct cgroup_subsys_state *css = NULL;
 770        struct mem_cgroup *memcg = NULL;
 771        struct mem_cgroup *pos = NULL;
 772
 773        if (mem_cgroup_disabled())
 774                return NULL;
 775
 776        if (!root)
 777                root = root_mem_cgroup;
 778
 779        if (prev && !reclaim)
 780                pos = prev;
 781
 782        if (!root->use_hierarchy && root != root_mem_cgroup) {
 783                if (prev)
 784                        goto out;
 785                return root;
 786        }
 787
 788        rcu_read_lock();
 789
 790        if (reclaim) {
 791                struct mem_cgroup_per_node *mz;
 792
 793                mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id);
 794                iter = &mz->iter[reclaim->priority];
 795
 796                if (prev && reclaim->generation != iter->generation)
 797                        goto out_unlock;
 798
 799                while (1) {
 800                        pos = READ_ONCE(iter->position);
 801                        if (!pos || css_tryget(&pos->css))
 802                                break;
 803                        /*
 804                         * css reference reached zero, so iter->position will
 805                         * be cleared by ->css_released. However, we should not
 806                         * rely on this happening soon, because ->css_released
 807                         * is called from a work queue, and by busy-waiting we
 808                         * might block it. So we clear iter->position right
 809                         * away.
 810                         */
 811                        (void)cmpxchg(&iter->position, pos, NULL);
 812                }
 813        }
 814
 815        if (pos)
 816                css = &pos->css;
 817
 818        for (;;) {
 819                css = css_next_descendant_pre(css, &root->css);
 820                if (!css) {
 821                        /*
 822                         * Reclaimers share the hierarchy walk, and a
 823                         * new one might jump in right at the end of
 824                         * the hierarchy - make sure they see at least
 825                         * one group and restart from the beginning.
 826                         */
 827                        if (!prev)
 828                                continue;
 829                        break;
 830                }
 831
 832                /*
 833                 * Verify the css and acquire a reference.  The root
 834                 * is provided by the caller, so we know it's alive
 835                 * and kicking, and don't take an extra reference.
 836                 */
 837                memcg = mem_cgroup_from_css(css);
 838
 839                if (css == &root->css)
 840                        break;
 841
 842                if (css_tryget(css))
 843                        break;
 844
 845                memcg = NULL;
 846        }
 847
 848        if (reclaim) {
 849                /*
 850                 * The position could have already been updated by a competing
 851                 * thread, so check that the value hasn't changed since we read
 852                 * it to avoid reclaiming from the same cgroup twice.
 853                 */
 854                (void)cmpxchg(&iter->position, pos, memcg);
 855
 856                if (pos)
 857                        css_put(&pos->css);
 858
 859                if (!memcg)
 860                        iter->generation++;
 861                else if (!prev)
 862                        reclaim->generation = iter->generation;
 863        }
 864
 865out_unlock:
 866        rcu_read_unlock();
 867out:
 868        if (prev && prev != root)
 869                css_put(&prev->css);
 870
 871        return memcg;
 872}
 873
 874/**
 875 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
 876 * @root: hierarchy root
 877 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
 878 */
 879void mem_cgroup_iter_break(struct mem_cgroup *root,
 880                           struct mem_cgroup *prev)
 881{
 882        if (!root)
 883                root = root_mem_cgroup;
 884        if (prev && prev != root)
 885                css_put(&prev->css);
 886}
 887
 888static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
 889{
 890        struct mem_cgroup *memcg = dead_memcg;
 891        struct mem_cgroup_reclaim_iter *iter;
 892        struct mem_cgroup_per_node *mz;
 893        int nid;
 894        int i;
 895
 896        while ((memcg = parent_mem_cgroup(memcg))) {
 897                for_each_node(nid) {
 898                        mz = mem_cgroup_nodeinfo(memcg, nid);
 899                        for (i = 0; i <= DEF_PRIORITY; i++) {
 900                                iter = &mz->iter[i];
 901                                cmpxchg(&iter->position,
 902                                        dead_memcg, NULL);
 903                        }
 904                }
 905        }
 906}
 907
 908/*
 909 * Iteration constructs for visiting all cgroups (under a tree).  If
 910 * loops are exited prematurely (break), mem_cgroup_iter_break() must
 911 * be used for reference counting.
 912 */
 913#define for_each_mem_cgroup_tree(iter, root)            \
 914        for (iter = mem_cgroup_iter(root, NULL, NULL);  \
 915             iter != NULL;                              \
 916             iter = mem_cgroup_iter(root, iter, NULL))
 917
 918#define for_each_mem_cgroup(iter)                       \
 919        for (iter = mem_cgroup_iter(NULL, NULL, NULL);  \
 920             iter != NULL;                              \
 921             iter = mem_cgroup_iter(NULL, iter, NULL))
 922
 923/**
 924 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
 925 * @memcg: hierarchy root
 926 * @fn: function to call for each task
 927 * @arg: argument passed to @fn
 928 *
 929 * This function iterates over tasks attached to @memcg or to any of its
 930 * descendants and calls @fn for each task. If @fn returns a non-zero
 931 * value, the function breaks the iteration loop and returns the value.
 932 * Otherwise, it will iterate over all tasks and return 0.
 933 *
 934 * This function must not be called for the root memory cgroup.
 935 */
 936int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
 937                          int (*fn)(struct task_struct *, void *), void *arg)
 938{
 939        struct mem_cgroup *iter;
 940        int ret = 0;
 941
 942        BUG_ON(memcg == root_mem_cgroup);
 943
 944        for_each_mem_cgroup_tree(iter, memcg) {
 945                struct css_task_iter it;
 946                struct task_struct *task;
 947
 948                css_task_iter_start(&iter->css, &it);
 949                while (!ret && (task = css_task_iter_next(&it)))
 950                        ret = fn(task, arg);
 951                css_task_iter_end(&it);
 952                if (ret) {
 953                        mem_cgroup_iter_break(memcg, iter);
 954                        break;
 955                }
 956        }
 957        return ret;
 958}
 959
 960/**
 961 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
 962 * @page: the page
 963 * @zone: zone of the page
 964 *
 965 * This function is only safe when following the LRU page isolation
 966 * and putback protocol: the LRU lock must be held, and the page must
 967 * either be PageLRU() or the caller must have isolated/allocated it.
 968 */
 969struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat)
 970{
 971        struct mem_cgroup_per_node *mz;
 972        struct mem_cgroup *memcg;
 973        struct lruvec *lruvec;
 974
 975        if (mem_cgroup_disabled()) {
 976                lruvec = &pgdat->lruvec;
 977                goto out;
 978        }
 979
 980        memcg = page->mem_cgroup;
 981        /*
 982         * Swapcache readahead pages are added to the LRU - and
 983         * possibly migrated - before they are charged.
 984         */
 985        if (!memcg)
 986                memcg = root_mem_cgroup;
 987
 988        mz = mem_cgroup_page_nodeinfo(memcg, page);
 989        lruvec = &mz->lruvec;
 990out:
 991        /*
 992         * Since a node can be onlined after the mem_cgroup was created,
 993         * we have to be prepared to initialize lruvec->zone here;
 994         * and if offlined then reonlined, we need to reinitialize it.
 995         */
 996        if (unlikely(lruvec->pgdat != pgdat))
 997                lruvec->pgdat = pgdat;
 998        return lruvec;
 999}
1000
1001/**
1002 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1003 * @lruvec: mem_cgroup per zone lru vector
1004 * @lru: index of lru list the page is sitting on
1005 * @nr_pages: positive when adding or negative when removing
1006 *
1007 * This function must be called under lru_lock, just before a page is added
1008 * to or just after a page is removed from an lru list (that ordering being
1009 * so as to allow it to check that lru_size 0 is consistent with list_empty).
1010 */
1011void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1012                                int nr_pages)
1013{
1014        struct mem_cgroup_per_node *mz;
1015        unsigned long *lru_size;
1016        long size;
1017        bool empty;
1018
1019        if (mem_cgroup_disabled())
1020                return;
1021
1022        mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
1023        lru_size = mz->lru_size + lru;
1024        empty = list_empty(lruvec->lists + lru);
1025
1026        if (nr_pages < 0)
1027                *lru_size += nr_pages;
1028
1029        size = *lru_size;
1030        if (WARN_ONCE(size < 0 || empty != !size,
1031                "%s(%p, %d, %d): lru_size %ld but %sempty\n",
1032                __func__, lruvec, lru, nr_pages, size, empty ? "" : "not ")) {
1033                VM_BUG_ON(1);
1034                *lru_size = 0;
1035        }
1036
1037        if (nr_pages > 0)
1038                *lru_size += nr_pages;
1039}
1040
1041bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
1042{
1043        struct mem_cgroup *task_memcg;
1044        struct task_struct *p;
1045        bool ret;
1046
1047        p = find_lock_task_mm(task);
1048        if (p) {
1049                task_memcg = get_mem_cgroup_from_mm(p->mm);
1050                task_unlock(p);
1051        } else {
1052                /*
1053                 * All threads may have already detached their mm's, but the oom
1054                 * killer still needs to detect if they have already been oom
1055                 * killed to prevent needlessly killing additional tasks.
1056                 */
1057                rcu_read_lock();
1058                task_memcg = mem_cgroup_from_task(task);
1059                css_get(&task_memcg->css);
1060                rcu_read_unlock();
1061        }
1062        ret = mem_cgroup_is_descendant(task_memcg, memcg);
1063        css_put(&task_memcg->css);
1064        return ret;
1065}
1066
1067/**
1068 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1069 * @memcg: the memory cgroup
1070 *
1071 * Returns the maximum amount of memory @mem can be charged with, in
1072 * pages.
1073 */
1074static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
1075{
1076        unsigned long margin = 0;
1077        unsigned long count;
1078        unsigned long limit;
1079
1080        count = page_counter_read(&memcg->memory);
1081        limit = READ_ONCE(memcg->memory.limit);
1082        if (count < limit)
1083                margin = limit - count;
1084
1085        if (do_memsw_account()) {
1086                count = page_counter_read(&memcg->memsw);
1087                limit = READ_ONCE(memcg->memsw.limit);
1088                if (count <= limit)
1089                        margin = min(margin, limit - count);
1090                else
1091                        margin = 0;
1092        }
1093
1094        return margin;
1095}
1096
1097/*
1098 * A routine for checking "mem" is under move_account() or not.
1099 *
1100 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1101 * moving cgroups. This is for waiting at high-memory pressure
1102 * caused by "move".
1103 */
1104static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
1105{
1106        struct mem_cgroup *from;
1107        struct mem_cgroup *to;
1108        bool ret = false;
1109        /*
1110         * Unlike task_move routines, we access mc.to, mc.from not under
1111         * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1112         */
1113        spin_lock(&mc.lock);
1114        from = mc.from;
1115        to = mc.to;
1116        if (!from)
1117                goto unlock;
1118
1119        ret = mem_cgroup_is_descendant(from, memcg) ||
1120                mem_cgroup_is_descendant(to, memcg);
1121unlock:
1122        spin_unlock(&mc.lock);
1123        return ret;
1124}
1125
1126static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
1127{
1128        if (mc.moving_task && current != mc.moving_task) {
1129                if (mem_cgroup_under_move(memcg)) {
1130                        DEFINE_WAIT(wait);
1131                        prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1132                        /* moving charge context might have finished. */
1133                        if (mc.moving_task)
1134                                schedule();
1135                        finish_wait(&mc.waitq, &wait);
1136                        return true;
1137                }
1138        }
1139        return false;
1140}
1141
1142#define K(x) ((x) << (PAGE_SHIFT-10))
1143/**
1144 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
1145 * @memcg: The memory cgroup that went over limit
1146 * @p: Task that is going to be killed
1147 *
1148 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1149 * enabled
1150 */
1151void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1152{
1153        struct mem_cgroup *iter;
1154        unsigned int i;
1155
1156        rcu_read_lock();
1157
1158        if (p) {
1159                pr_info("Task in ");
1160                pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1161                pr_cont(" killed as a result of limit of ");
1162        } else {
1163                pr_info("Memory limit reached of cgroup ");
1164        }
1165
1166        pr_cont_cgroup_path(memcg->css.cgroup);
1167        pr_cont("\n");
1168
1169        rcu_read_unlock();
1170
1171        pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1172                K((u64)page_counter_read(&memcg->memory)),
1173                K((u64)memcg->memory.limit), memcg->memory.failcnt);
1174        pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1175                K((u64)page_counter_read(&memcg->memsw)),
1176                K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1177        pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1178                K((u64)page_counter_read(&memcg->kmem)),
1179                K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
1180
1181        for_each_mem_cgroup_tree(iter, memcg) {
1182                pr_info("Memory cgroup stats for ");
1183                pr_cont_cgroup_path(iter->css.cgroup);
1184                pr_cont(":");
1185
1186                for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1187                        if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1188                                continue;
1189                        pr_cont(" %s:%luKB", mem_cgroup_stat_names[i],
1190                                K(mem_cgroup_read_stat(iter, i)));
1191                }
1192
1193                for (i = 0; i < NR_LRU_LISTS; i++)
1194                        pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1195                                K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1196
1197                pr_cont("\n");
1198        }
1199}
1200
1201/*
1202 * This function returns the number of memcg under hierarchy tree. Returns
1203 * 1(self count) if no children.
1204 */
1205static int mem_cgroup_count_children(struct mem_cgroup *memcg)
1206{
1207        int num = 0;
1208        struct mem_cgroup *iter;
1209
1210        for_each_mem_cgroup_tree(iter, memcg)
1211                num++;
1212        return num;
1213}
1214
1215/*
1216 * Return the memory (and swap, if configured) limit for a memcg.
1217 */
1218unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
1219{
1220        unsigned long limit;
1221
1222        limit = memcg->memory.limit;
1223        if (mem_cgroup_swappiness(memcg)) {
1224                unsigned long memsw_limit;
1225                unsigned long swap_limit;
1226
1227                memsw_limit = memcg->memsw.limit;
1228                swap_limit = memcg->swap.limit;
1229                swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1230                limit = min(limit + swap_limit, memsw_limit);
1231        }
1232        return limit;
1233}
1234
1235static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1236                                     int order)
1237{
1238        struct oom_control oc = {
1239                .zonelist = NULL,
1240                .nodemask = NULL,
1241                .memcg = memcg,
1242                .gfp_mask = gfp_mask,
1243                .order = order,
1244        };
1245        bool ret;
1246
1247        mutex_lock(&oom_lock);
1248        ret = out_of_memory(&oc);
1249        mutex_unlock(&oom_lock);
1250        return ret;
1251}
1252
1253#if MAX_NUMNODES > 1
1254
1255/**
1256 * test_mem_cgroup_node_reclaimable
1257 * @memcg: the target memcg
1258 * @nid: the node ID to be checked.
1259 * @noswap : specify true here if the user wants flle only information.
1260 *
1261 * This function returns whether the specified memcg contains any
1262 * reclaimable pages on a node. Returns true if there are any reclaimable
1263 * pages in the node.
1264 */
1265static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
1266                int nid, bool noswap)
1267{
1268        if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
1269                return true;
1270        if (noswap || !total_swap_pages)
1271                return false;
1272        if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
1273                return true;
1274        return false;
1275
1276}
1277
1278/*
1279 * Always updating the nodemask is not very good - even if we have an empty
1280 * list or the wrong list here, we can start from some node and traverse all
1281 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1282 *
1283 */
1284static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
1285{
1286        int nid;
1287        /*
1288         * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1289         * pagein/pageout changes since the last update.
1290         */
1291        if (!atomic_read(&memcg->numainfo_events))
1292                return;
1293        if (atomic_inc_return(&memcg->numainfo_updating) > 1)
1294                return;
1295
1296        /* make a nodemask where this memcg uses memory from */
1297        memcg->scan_nodes = node_states[N_MEMORY];
1298
1299        for_each_node_mask(nid, node_states[N_MEMORY]) {
1300
1301                if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1302                        node_clear(nid, memcg->scan_nodes);
1303        }
1304
1305        atomic_set(&memcg->numainfo_events, 0);
1306        atomic_set(&memcg->numainfo_updating, 0);
1307}
1308
1309/*
1310 * Selecting a node where we start reclaim from. Because what we need is just
1311 * reducing usage counter, start from anywhere is O,K. Considering
1312 * memory reclaim from current node, there are pros. and cons.
1313 *
1314 * Freeing memory from current node means freeing memory from a node which
1315 * we'll use or we've used. So, it may make LRU bad. And if several threads
1316 * hit limits, it will see a contention on a node. But freeing from remote
1317 * node means more costs for memory reclaim because of memory latency.
1318 *
1319 * Now, we use round-robin. Better algorithm is welcomed.
1320 */
1321int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1322{
1323        int node;
1324
1325        mem_cgroup_may_update_nodemask(memcg);
1326        node = memcg->last_scanned_node;
1327
1328        node = next_node_in(node, memcg->scan_nodes);
1329        /*
1330         * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1331         * last time it really checked all the LRUs due to rate limiting.
1332         * Fallback to the current node in that case for simplicity.
1333         */
1334        if (unlikely(node == MAX_NUMNODES))
1335                node = numa_node_id();
1336
1337        memcg->last_scanned_node = node;
1338        return node;
1339}
1340#else
1341int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1342{
1343        return 0;
1344}
1345#endif
1346
1347static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1348                                   pg_data_t *pgdat,
1349                                   gfp_t gfp_mask,
1350                                   unsigned long *total_scanned)
1351{
1352        struct mem_cgroup *victim = NULL;
1353        int total = 0;
1354        int loop = 0;
1355        unsigned long excess;
1356        unsigned long nr_scanned;
1357        struct mem_cgroup_reclaim_cookie reclaim = {
1358                .pgdat = pgdat,
1359                .priority = 0,
1360        };
1361
1362        excess = soft_limit_excess(root_memcg);
1363
1364        while (1) {
1365                victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1366                if (!victim) {
1367                        loop++;
1368                        if (loop >= 2) {
1369                                /*
1370                                 * If we have not been able to reclaim
1371                                 * anything, it might because there are
1372                                 * no reclaimable pages under this hierarchy
1373                                 */
1374                                if (!total)
1375                                        break;
1376                                /*
1377                                 * We want to do more targeted reclaim.
1378                                 * excess >> 2 is not to excessive so as to
1379                                 * reclaim too much, nor too less that we keep
1380                                 * coming back to reclaim from this cgroup
1381                                 */
1382                                if (total >= (excess >> 2) ||
1383                                        (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1384                                        break;
1385                        }
1386                        continue;
1387                }
1388                total += mem_cgroup_shrink_node(victim, gfp_mask, false,
1389                                        pgdat, &nr_scanned);
1390                *total_scanned += nr_scanned;
1391                if (!soft_limit_excess(root_memcg))
1392                        break;
1393        }
1394        mem_cgroup_iter_break(root_memcg, victim);
1395        return total;
1396}
1397
1398#ifdef CONFIG_LOCKDEP
1399static struct lockdep_map memcg_oom_lock_dep_map = {
1400        .name = "memcg_oom_lock",
1401};
1402#endif
1403
1404static DEFINE_SPINLOCK(memcg_oom_lock);
1405
1406/*
1407 * Check OOM-Killer is already running under our hierarchy.
1408 * If someone is running, return false.
1409 */
1410static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
1411{
1412        struct mem_cgroup *iter, *failed = NULL;
1413
1414        spin_lock(&memcg_oom_lock);
1415
1416        for_each_mem_cgroup_tree(iter, memcg) {
1417                if (iter->oom_lock) {
1418                        /*
1419                         * this subtree of our hierarchy is already locked
1420                         * so we cannot give a lock.
1421                         */
1422                        failed = iter;
1423                        mem_cgroup_iter_break(memcg, iter);
1424                        break;
1425                } else
1426                        iter->oom_lock = true;
1427        }
1428
1429        if (failed) {
1430                /*
1431                 * OK, we failed to lock the whole subtree so we have
1432                 * to clean up what we set up to the failing subtree
1433                 */
1434                for_each_mem_cgroup_tree(iter, memcg) {
1435                        if (iter == failed) {
1436                                mem_cgroup_iter_break(memcg, iter);
1437                                break;
1438                        }
1439                        iter->oom_lock = false;
1440                }
1441        } else
1442                mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
1443
1444        spin_unlock(&memcg_oom_lock);
1445
1446        return !failed;
1447}
1448
1449static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
1450{
1451        struct mem_cgroup *iter;
1452
1453        spin_lock(&memcg_oom_lock);
1454        mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
1455        for_each_mem_cgroup_tree(iter, memcg)
1456                iter->oom_lock = false;
1457        spin_unlock(&memcg_oom_lock);
1458}
1459
1460static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
1461{
1462        struct mem_cgroup *iter;
1463
1464        spin_lock(&memcg_oom_lock);
1465        for_each_mem_cgroup_tree(iter, memcg)
1466                iter->under_oom++;
1467        spin_unlock(&memcg_oom_lock);
1468}
1469
1470static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
1471{
1472        struct mem_cgroup *iter;
1473
1474        /*
1475         * When a new child is created while the hierarchy is under oom,
1476         * mem_cgroup_oom_lock() may not be called. Watch for underflow.
1477         */
1478        spin_lock(&memcg_oom_lock);
1479        for_each_mem_cgroup_tree(iter, memcg)
1480                if (iter->under_oom > 0)
1481                        iter->under_oom--;
1482        spin_unlock(&memcg_oom_lock);
1483}
1484
1485static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1486
1487struct oom_wait_info {
1488        struct mem_cgroup *memcg;
1489        wait_queue_t    wait;
1490};
1491
1492static int memcg_oom_wake_function(wait_queue_t *wait,
1493        unsigned mode, int sync, void *arg)
1494{
1495        struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1496        struct mem_cgroup *oom_wait_memcg;
1497        struct oom_wait_info *oom_wait_info;
1498
1499        oom_wait_info = container_of(wait, struct oom_wait_info, wait);
1500        oom_wait_memcg = oom_wait_info->memcg;
1501
1502        if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1503            !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
1504                return 0;
1505        return autoremove_wake_function(wait, mode, sync, arg);
1506}
1507
1508static void memcg_oom_recover(struct mem_cgroup *memcg)
1509{
1510        /*
1511         * For the following lockless ->under_oom test, the only required
1512         * guarantee is that it must see the state asserted by an OOM when
1513         * this function is called as a result of userland actions
1514         * triggered by the notification of the OOM.  This is trivially
1515         * achieved by invoking mem_cgroup_mark_under_oom() before
1516         * triggering notification.
1517         */
1518        if (memcg && memcg->under_oom)
1519                __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
1520}
1521
1522static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
1523{
1524        if (!current->memcg_may_oom)
1525                return;
1526        /*
1527         * We are in the middle of the charge context here, so we
1528         * don't want to block when potentially sitting on a callstack
1529         * that holds all kinds of filesystem and mm locks.
1530         *
1531         * Also, the caller may handle a failed allocation gracefully
1532         * (like optional page cache readahead) and so an OOM killer
1533         * invocation might not even be necessary.
1534         *
1535         * That's why we don't do anything here except remember the
1536         * OOM context and then deal with it at the end of the page
1537         * fault when the stack is unwound, the locks are released,
1538         * and when we know whether the fault was overall successful.
1539         */
1540        css_get(&memcg->css);
1541        current->memcg_in_oom = memcg;
1542        current->memcg_oom_gfp_mask = mask;
1543        current->memcg_oom_order = order;
1544}
1545
1546/**
1547 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1548 * @handle: actually kill/wait or just clean up the OOM state
1549 *
1550 * This has to be called at the end of a page fault if the memcg OOM
1551 * handler was enabled.
1552 *
1553 * Memcg supports userspace OOM handling where failed allocations must
1554 * sleep on a waitqueue until the userspace task resolves the
1555 * situation.  Sleeping directly in the charge context with all kinds
1556 * of locks held is not a good idea, instead we remember an OOM state
1557 * in the task and mem_cgroup_oom_synchronize() has to be called at
1558 * the end of the page fault to complete the OOM handling.
1559 *
1560 * Returns %true if an ongoing memcg OOM situation was detected and
1561 * completed, %false otherwise.
1562 */
1563bool mem_cgroup_oom_synchronize(bool handle)
1564{
1565        struct mem_cgroup *memcg = current->memcg_in_oom;
1566        struct oom_wait_info owait;
1567        bool locked;
1568
1569        /* OOM is global, do not handle */
1570        if (!memcg)
1571                return false;
1572
1573        if (!handle)
1574                goto cleanup;
1575
1576        owait.memcg = memcg;
1577        owait.wait.flags = 0;
1578        owait.wait.func = memcg_oom_wake_function;
1579        owait.wait.private = current;
1580        INIT_LIST_HEAD(&owait.wait.task_list);
1581
1582        prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
1583        mem_cgroup_mark_under_oom(memcg);
1584
1585        locked = mem_cgroup_oom_trylock(memcg);
1586
1587        if (locked)
1588                mem_cgroup_oom_notify(memcg);
1589
1590        if (locked && !memcg->oom_kill_disable) {
1591                mem_cgroup_unmark_under_oom(memcg);
1592                finish_wait(&memcg_oom_waitq, &owait.wait);
1593                mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1594                                         current->memcg_oom_order);
1595        } else {
1596                schedule();
1597                mem_cgroup_unmark_under_oom(memcg);
1598                finish_wait(&memcg_oom_waitq, &owait.wait);
1599        }
1600
1601        if (locked) {
1602                mem_cgroup_oom_unlock(memcg);
1603                /*
1604                 * There is no guarantee that an OOM-lock contender
1605                 * sees the wakeups triggered by the OOM kill
1606                 * uncharges.  Wake any sleepers explicitely.
1607                 */
1608                memcg_oom_recover(memcg);
1609        }
1610cleanup:
1611        current->memcg_in_oom = NULL;
1612        css_put(&memcg->css);
1613        return true;
1614}
1615
1616/**
1617 * lock_page_memcg - lock a page->mem_cgroup binding
1618 * @page: the page
1619 *
1620 * This function protects unlocked LRU pages from being moved to
1621 * another cgroup and stabilizes their page->mem_cgroup binding.
1622 */
1623void lock_page_memcg(struct page *page)
1624{
1625        struct mem_cgroup *memcg;
1626        unsigned long flags;
1627
1628        /*
1629         * The RCU lock is held throughout the transaction.  The fast
1630         * path can get away without acquiring the memcg->move_lock
1631         * because page moving starts with an RCU grace period.
1632         */
1633        rcu_read_lock();
1634
1635        if (mem_cgroup_disabled())
1636                return;
1637again:
1638        memcg = page->mem_cgroup;
1639        if (unlikely(!memcg))
1640                return;
1641
1642        if (atomic_read(&memcg->moving_account) <= 0)
1643                return;
1644
1645        spin_lock_irqsave(&memcg->move_lock, flags);
1646        if (memcg != page->mem_cgroup) {
1647                spin_unlock_irqrestore(&memcg->move_lock, flags);
1648                goto again;
1649        }
1650
1651        /*
1652         * When charge migration first begins, we can have locked and
1653         * unlocked page stat updates happening concurrently.  Track
1654         * the task who has the lock for unlock_page_memcg().
1655         */
1656        memcg->move_lock_task = current;
1657        memcg->move_lock_flags = flags;
1658
1659        return;
1660}
1661EXPORT_SYMBOL(lock_page_memcg);
1662
1663/**
1664 * unlock_page_memcg - unlock a page->mem_cgroup binding
1665 * @page: the page
1666 */
1667void unlock_page_memcg(struct page *page)
1668{
1669        struct mem_cgroup *memcg = page->mem_cgroup;
1670
1671        if (memcg && memcg->move_lock_task == current) {
1672                unsigned long flags = memcg->move_lock_flags;
1673
1674                memcg->move_lock_task = NULL;
1675                memcg->move_lock_flags = 0;
1676
1677                spin_unlock_irqrestore(&memcg->move_lock, flags);
1678        }
1679
1680        rcu_read_unlock();
1681}
1682EXPORT_SYMBOL(unlock_page_memcg);
1683
1684/*
1685 * size of first charge trial. "32" comes from vmscan.c's magic value.
1686 * TODO: maybe necessary to use big numbers in big irons.
1687 */
1688#define CHARGE_BATCH    32U
1689struct memcg_stock_pcp {
1690        struct mem_cgroup *cached; /* this never be root cgroup */
1691        unsigned int nr_pages;
1692        struct work_struct work;
1693        unsigned long flags;
1694#define FLUSHING_CACHED_CHARGE  0
1695};
1696static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
1697static DEFINE_MUTEX(percpu_charge_mutex);
1698
1699/**
1700 * consume_stock: Try to consume stocked charge on this cpu.
1701 * @memcg: memcg to consume from.
1702 * @nr_pages: how many pages to charge.
1703 *
1704 * The charges will only happen if @memcg matches the current cpu's memcg
1705 * stock, and at least @nr_pages are available in that stock.  Failure to
1706 * service an allocation will refill the stock.
1707 *
1708 * returns true if successful, false otherwise.
1709 */
1710static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
1711{
1712        struct memcg_stock_pcp *stock;
1713        unsigned long flags;
1714        bool ret = false;
1715
1716        if (nr_pages > CHARGE_BATCH)
1717                return ret;
1718
1719        local_irq_save(flags);
1720
1721        stock = this_cpu_ptr(&memcg_stock);
1722        if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
1723                stock->nr_pages -= nr_pages;
1724                ret = true;
1725        }
1726
1727        local_irq_restore(flags);
1728
1729        return ret;
1730}
1731
1732/*
1733 * Returns stocks cached in percpu and reset cached information.
1734 */
1735static void drain_stock(struct memcg_stock_pcp *stock)
1736{
1737        struct mem_cgroup *old = stock->cached;
1738
1739        if (stock->nr_pages) {
1740                page_counter_uncharge(&old->memory, stock->nr_pages);
1741                if (do_memsw_account())
1742                        page_counter_uncharge(&old->memsw, stock->nr_pages);
1743                css_put_many(&old->css, stock->nr_pages);
1744                stock->nr_pages = 0;
1745        }
1746        stock->cached = NULL;
1747}
1748
1749static void drain_local_stock(struct work_struct *dummy)
1750{
1751        struct memcg_stock_pcp *stock;
1752        unsigned long flags;
1753
1754        local_irq_save(flags);
1755
1756        stock = this_cpu_ptr(&memcg_stock);
1757        drain_stock(stock);
1758        clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
1759
1760        local_irq_restore(flags);
1761}
1762
1763/*
1764 * Cache charges(val) to local per_cpu area.
1765 * This will be consumed by consume_stock() function, later.
1766 */
1767static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
1768{
1769        struct memcg_stock_pcp *stock;
1770        unsigned long flags;
1771
1772        local_irq_save(flags);
1773
1774        stock = this_cpu_ptr(&memcg_stock);
1775        if (stock->cached != memcg) { /* reset if necessary */
1776                drain_stock(stock);
1777                stock->cached = memcg;
1778        }
1779        stock->nr_pages += nr_pages;
1780
1781        local_irq_restore(flags);
1782}
1783
1784/*
1785 * Drains all per-CPU charge caches for given root_memcg resp. subtree
1786 * of the hierarchy under it.
1787 */
1788static void drain_all_stock(struct mem_cgroup *root_memcg)
1789{
1790        int cpu, curcpu;
1791
1792        /* If someone's already draining, avoid adding running more workers. */
1793        if (!mutex_trylock(&percpu_charge_mutex))
1794                return;
1795        /* Notify other cpus that system-wide "drain" is running */
1796        get_online_cpus();
1797        curcpu = get_cpu();
1798        for_each_online_cpu(cpu) {
1799                struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
1800                struct mem_cgroup *memcg;
1801
1802                memcg = stock->cached;
1803                if (!memcg || !stock->nr_pages)
1804                        continue;
1805                if (!mem_cgroup_is_descendant(memcg, root_memcg))
1806                        continue;
1807                if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1808                        if (cpu == curcpu)
1809                                drain_local_stock(&stock->work);
1810                        else
1811                                schedule_work_on(cpu, &stock->work);
1812                }
1813        }
1814        put_cpu();
1815        put_online_cpus();
1816        mutex_unlock(&percpu_charge_mutex);
1817}
1818
1819static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
1820                                        unsigned long action,
1821                                        void *hcpu)
1822{
1823        int cpu = (unsigned long)hcpu;
1824        struct memcg_stock_pcp *stock;
1825
1826        if (action == CPU_ONLINE)
1827                return NOTIFY_OK;
1828
1829        if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
1830                return NOTIFY_OK;
1831
1832        stock = &per_cpu(memcg_stock, cpu);
1833        drain_stock(stock);
1834        return NOTIFY_OK;
1835}
1836
1837static void reclaim_high(struct mem_cgroup *memcg,
1838                         unsigned int nr_pages,
1839                         gfp_t gfp_mask)
1840{
1841        do {
1842                if (page_counter_read(&memcg->memory) <= memcg->high)
1843                        continue;
1844                mem_cgroup_events(memcg, MEMCG_HIGH, 1);
1845                try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1846        } while ((memcg = parent_mem_cgroup(memcg)));
1847}
1848
1849static void high_work_func(struct work_struct *work)
1850{
1851        struct mem_cgroup *memcg;
1852
1853        memcg = container_of(work, struct mem_cgroup, high_work);
1854        reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1855}
1856
1857/*
1858 * Scheduled by try_charge() to be executed from the userland return path
1859 * and reclaims memory over the high limit.
1860 */
1861void mem_cgroup_handle_over_high(void)
1862{
1863        unsigned int nr_pages = current->memcg_nr_pages_over_high;
1864        struct mem_cgroup *memcg;
1865
1866        if (likely(!nr_pages))
1867                return;
1868
1869        memcg = get_mem_cgroup_from_mm(current->mm);
1870        reclaim_high(memcg, nr_pages, GFP_KERNEL);
1871        css_put(&memcg->css);
1872        current->memcg_nr_pages_over_high = 0;
1873}
1874
1875static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1876                      unsigned int nr_pages)
1877{
1878        unsigned int batch = max(CHARGE_BATCH, nr_pages);
1879        int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1880        struct mem_cgroup *mem_over_limit;
1881        struct page_counter *counter;
1882        unsigned long nr_reclaimed;
1883        bool may_swap = true;
1884        bool drained = false;
1885
1886        if (mem_cgroup_is_root(memcg))
1887                return 0;
1888retry:
1889        if (consume_stock(memcg, nr_pages))
1890                return 0;
1891
1892        if (!do_memsw_account() ||
1893            page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1894                if (page_counter_try_charge(&memcg->memory, batch, &counter))
1895                        goto done_restock;
1896                if (do_memsw_account())
1897                        page_counter_uncharge(&memcg->memsw, batch);
1898                mem_over_limit = mem_cgroup_from_counter(counter, memory);
1899        } else {
1900                mem_over_limit = mem_cgroup_from_counter(counter, memsw);
1901                may_swap = false;
1902        }
1903
1904        if (batch > nr_pages) {
1905                batch = nr_pages;
1906                goto retry;
1907        }
1908
1909        /*
1910         * Unlike in global OOM situations, memcg is not in a physical
1911         * memory shortage.  Allow dying and OOM-killed tasks to
1912         * bypass the last charges so that they can exit quickly and
1913         * free their memory.
1914         */
1915        if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1916                     fatal_signal_pending(current) ||
1917                     current->flags & PF_EXITING))
1918                goto force;
1919
1920        /*
1921         * Prevent unbounded recursion when reclaim operations need to
1922         * allocate memory. This might exceed the limits temporarily,
1923         * but we prefer facilitating memory reclaim and getting back
1924         * under the limit over triggering OOM kills in these cases.
1925         */
1926        if (unlikely(current->flags & PF_MEMALLOC))
1927                goto force;
1928
1929        if (unlikely(task_in_memcg_oom(current)))
1930                goto nomem;
1931
1932        if (!gfpflags_allow_blocking(gfp_mask))
1933                goto nomem;
1934
1935        mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
1936
1937        nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1938                                                    gfp_mask, may_swap);
1939
1940        if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
1941                goto retry;
1942
1943        if (!drained) {
1944                drain_all_stock(mem_over_limit);
1945                drained = true;
1946                goto retry;
1947        }
1948
1949        if (gfp_mask & __GFP_NORETRY)
1950                goto nomem;
1951        /*
1952         * Even though the limit is exceeded at this point, reclaim
1953         * may have been able to free some pages.  Retry the charge
1954         * before killing the task.
1955         *
1956         * Only for regular pages, though: huge pages are rather
1957         * unlikely to succeed so close to the limit, and we fall back
1958         * to regular pages anyway in case of failure.
1959         */
1960        if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
1961                goto retry;
1962        /*
1963         * At task move, charge accounts can be doubly counted. So, it's
1964         * better to wait until the end of task_move if something is going on.
1965         */
1966        if (mem_cgroup_wait_acct_move(mem_over_limit))
1967                goto retry;
1968
1969        if (nr_retries--)
1970                goto retry;
1971
1972        if (gfp_mask & __GFP_NOFAIL)
1973                goto force;
1974
1975        if (fatal_signal_pending(current))
1976                goto force;
1977
1978        mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
1979
1980        mem_cgroup_oom(mem_over_limit, gfp_mask,
1981                       get_order(nr_pages * PAGE_SIZE));
1982nomem:
1983        if (!(gfp_mask & __GFP_NOFAIL))
1984                return -ENOMEM;
1985force:
1986        /*
1987         * The allocation either can't fail or will lead to more memory
1988         * being freed very soon.  Allow memory usage go over the limit
1989         * temporarily by force charging it.
1990         */
1991        page_counter_charge(&memcg->memory, nr_pages);
1992        if (do_memsw_account())
1993                page_counter_charge(&memcg->memsw, nr_pages);
1994        css_get_many(&memcg->css, nr_pages);
1995
1996        return 0;
1997
1998done_restock:
1999        css_get_many(&memcg->css, batch);
2000        if (batch > nr_pages)
2001                refill_stock(memcg, batch - nr_pages);
2002
2003        /*
2004         * If the hierarchy is above the normal consumption range, schedule
2005         * reclaim on returning to userland.  We can perform reclaim here
2006         * if __GFP_RECLAIM but let's always punt for simplicity and so that
2007         * GFP_KERNEL can consistently be used during reclaim.  @memcg is
2008         * not recorded as it most likely matches current's and won't
2009         * change in the meantime.  As high limit is checked again before
2010         * reclaim, the cost of mismatch is negligible.
2011         */
2012        do {
2013                if (page_counter_read(&memcg->memory) > memcg->high) {
2014                        /* Don't bother a random interrupted task */
2015                        if (in_interrupt()) {
2016                                schedule_work(&memcg->high_work);
2017                                break;
2018                        }
2019                        current->memcg_nr_pages_over_high += batch;
2020                        set_notify_resume(current);
2021                        break;
2022                }
2023        } while ((memcg = parent_mem_cgroup(memcg)));
2024
2025        return 0;
2026}
2027
2028static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
2029{
2030        if (mem_cgroup_is_root(memcg))
2031                return;
2032
2033        page_counter_uncharge(&memcg->memory, nr_pages);
2034        if (do_memsw_account())
2035                page_counter_uncharge(&memcg->memsw, nr_pages);
2036
2037        css_put_many(&memcg->css, nr_pages);
2038}
2039
2040static void lock_page_lru(struct page *page, int *isolated)
2041{
2042        struct zone *zone = page_zone(page);
2043
2044        spin_lock_irq(zone_lru_lock(zone));
2045        if (PageLRU(page)) {
2046                struct lruvec *lruvec;
2047
2048                lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
2049                ClearPageLRU(page);
2050                del_page_from_lru_list(page, lruvec, page_lru(page));
2051                *isolated = 1;
2052        } else
2053                *isolated = 0;
2054}
2055
2056static void unlock_page_lru(struct page *page, int isolated)
2057{
2058        struct zone *zone = page_zone(page);
2059
2060        if (isolated) {
2061                struct lruvec *lruvec;
2062
2063                lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
2064                VM_BUG_ON_PAGE(PageLRU(page), page);
2065                SetPageLRU(page);
2066                add_page_to_lru_list(page, lruvec, page_lru(page));
2067        }
2068        spin_unlock_irq(zone_lru_lock(zone));
2069}
2070
2071static void commit_charge(struct page *page, struct mem_cgroup *memcg,
2072                          bool lrucare)
2073{
2074        int isolated;
2075
2076        VM_BUG_ON_PAGE(page->mem_cgroup, page);
2077
2078        /*
2079         * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2080         * may already be on some other mem_cgroup's LRU.  Take care of it.
2081         */
2082        if (lrucare)
2083                lock_page_lru(page, &isolated);
2084
2085        /*
2086         * Nobody should be changing or seriously looking at
2087         * page->mem_cgroup at this point:
2088         *
2089         * - the page is uncharged
2090         *
2091         * - the page is off-LRU
2092         *
2093         * - an anonymous fault has exclusive page access, except for
2094         *   a locked page table
2095         *
2096         * - a page cache insertion, a swapin fault, or a migration
2097         *   have the page locked
2098         */
2099        page->mem_cgroup = memcg;
2100
2101        if (lrucare)
2102                unlock_page_lru(page, isolated);
2103}
2104
2105#ifndef CONFIG_SLOB
2106static int memcg_alloc_cache_id(void)
2107{
2108        int id, size;
2109        int err;
2110
2111        id = ida_simple_get(&memcg_cache_ida,
2112                            0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2113        if (id < 0)
2114                return id;
2115
2116        if (id < memcg_nr_cache_ids)
2117                return id;
2118
2119        /*
2120         * There's no space for the new id in memcg_caches arrays,
2121         * so we have to grow them.
2122         */
2123        down_write(&memcg_cache_ids_sem);
2124
2125        size = 2 * (id + 1);
2126        if (size < MEMCG_CACHES_MIN_SIZE)
2127                size = MEMCG_CACHES_MIN_SIZE;
2128        else if (size > MEMCG_CACHES_MAX_SIZE)
2129                size = MEMCG_CACHES_MAX_SIZE;
2130
2131        err = memcg_update_all_caches(size);
2132        if (!err)
2133                err = memcg_update_all_list_lrus(size);
2134        if (!err)
2135                memcg_nr_cache_ids = size;
2136
2137        up_write(&memcg_cache_ids_sem);
2138
2139        if (err) {
2140                ida_simple_remove(&memcg_cache_ida, id);
2141                return err;
2142        }
2143        return id;
2144}
2145
2146static void memcg_free_cache_id(int id)
2147{
2148        ida_simple_remove(&memcg_cache_ida, id);
2149}
2150
2151struct memcg_kmem_cache_create_work {
2152        struct mem_cgroup *memcg;
2153        struct kmem_cache *cachep;
2154        struct work_struct work;
2155};
2156
2157static void memcg_kmem_cache_create_func(struct work_struct *w)
2158{
2159        struct memcg_kmem_cache_create_work *cw =
2160                container_of(w, struct memcg_kmem_cache_create_work, work);
2161        struct mem_cgroup *memcg = cw->memcg;
2162        struct kmem_cache *cachep = cw->cachep;
2163
2164        memcg_create_kmem_cache(memcg, cachep);
2165
2166        css_put(&memcg->css);
2167        kfree(cw);
2168}
2169
2170/*
2171 * Enqueue the creation of a per-memcg kmem_cache.
2172 */
2173static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2174                                               struct kmem_cache *cachep)
2175{
2176        struct memcg_kmem_cache_create_work *cw;
2177
2178        cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
2179        if (!cw)
2180                return;
2181
2182        css_get(&memcg->css);
2183
2184        cw->memcg = memcg;
2185        cw->cachep = cachep;
2186        INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
2187
2188        schedule_work(&cw->work);
2189}
2190
2191static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2192                                             struct kmem_cache *cachep)
2193{
2194        /*
2195         * We need to stop accounting when we kmalloc, because if the
2196         * corresponding kmalloc cache is not yet created, the first allocation
2197         * in __memcg_schedule_kmem_cache_create will recurse.
2198         *
2199         * However, it is better to enclose the whole function. Depending on
2200         * the debugging options enabled, INIT_WORK(), for instance, can
2201         * trigger an allocation. This too, will make us recurse. Because at
2202         * this point we can't allow ourselves back into memcg_kmem_get_cache,
2203         * the safest choice is to do it like this, wrapping the whole function.
2204         */
2205        current->memcg_kmem_skip_account = 1;
2206        __memcg_schedule_kmem_cache_create(memcg, cachep);
2207        current->memcg_kmem_skip_account = 0;
2208}
2209
2210static inline bool memcg_kmem_bypass(void)
2211{
2212        if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
2213                return true;
2214        return false;
2215}
2216
2217/**
2218 * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2219 * @cachep: the original global kmem cache
2220 *
2221 * Return the kmem_cache we're supposed to use for a slab allocation.
2222 * We try to use the current memcg's version of the cache.
2223 *
2224 * If the cache does not exist yet, if we are the first user of it, we
2225 * create it asynchronously in a workqueue and let the current allocation
2226 * go through with the original cache.
2227 *
2228 * This function takes a reference to the cache it returns to assure it
2229 * won't get destroyed while we are working with it. Once the caller is
2230 * done with it, memcg_kmem_put_cache() must be called to release the
2231 * reference.
2232 */
2233struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep)
2234{
2235        struct mem_cgroup *memcg;
2236        struct kmem_cache *memcg_cachep;
2237        int kmemcg_id;
2238
2239        VM_BUG_ON(!is_root_cache(cachep));
2240
2241        if (memcg_kmem_bypass())
2242                return cachep;
2243
2244        if (current->memcg_kmem_skip_account)
2245                return cachep;
2246
2247        memcg = get_mem_cgroup_from_mm(current->mm);
2248        kmemcg_id = READ_ONCE(memcg->kmemcg_id);
2249        if (kmemcg_id < 0)
2250                goto out;
2251
2252        memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
2253        if (likely(memcg_cachep))
2254                return memcg_cachep;
2255
2256        /*
2257         * If we are in a safe context (can wait, and not in interrupt
2258         * context), we could be be predictable and return right away.
2259         * This would guarantee that the allocation being performed
2260         * already belongs in the new cache.
2261         *
2262         * However, there are some clashes that can arrive from locking.
2263         * For instance, because we acquire the slab_mutex while doing
2264         * memcg_create_kmem_cache, this means no further allocation
2265         * could happen with the slab_mutex held. So it's better to
2266         * defer everything.
2267         */
2268        memcg_schedule_kmem_cache_create(memcg, cachep);
2269out:
2270        css_put(&memcg->css);
2271        return cachep;
2272}
2273
2274/**
2275 * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2276 * @cachep: the cache returned by memcg_kmem_get_cache
2277 */
2278void memcg_kmem_put_cache(struct kmem_cache *cachep)
2279{
2280        if (!is_root_cache(cachep))
2281                css_put(&cachep->memcg_params.memcg->css);
2282}
2283
2284/**
2285 * memcg_kmem_charge: charge a kmem page
2286 * @page: page to charge
2287 * @gfp: reclaim mode
2288 * @order: allocation order
2289 * @memcg: memory cgroup to charge
2290 *
2291 * Returns 0 on success, an error code on failure.
2292 */
2293int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2294                            struct mem_cgroup *memcg)
2295{
2296        unsigned int nr_pages = 1 << order;
2297        struct page_counter *counter;
2298        int ret;
2299
2300        ret = try_charge(memcg, gfp, nr_pages);
2301        if (ret)
2302                return ret;
2303
2304        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2305            !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2306                cancel_charge(memcg, nr_pages);
2307                return -ENOMEM;
2308        }
2309
2310        page->mem_cgroup = memcg;
2311
2312        return 0;
2313}
2314
2315/**
2316 * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2317 * @page: page to charge
2318 * @gfp: reclaim mode
2319 * @order: allocation order
2320 *
2321 * Returns 0 on success, an error code on failure.
2322 */
2323int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
2324{
2325        struct mem_cgroup *memcg;
2326        int ret = 0;
2327
2328        if (memcg_kmem_bypass())
2329                return 0;
2330
2331        memcg = get_mem_cgroup_from_mm(current->mm);
2332        if (!mem_cgroup_is_root(memcg)) {
2333                ret = memcg_kmem_charge_memcg(page, gfp, order, memcg);
2334                if (!ret)
2335                        __SetPageKmemcg(page);
2336        }
2337        css_put(&memcg->css);
2338        return ret;
2339}
2340/**
2341 * memcg_kmem_uncharge: uncharge a kmem page
2342 * @page: page to uncharge
2343 * @order: allocation order
2344 */
2345void memcg_kmem_uncharge(struct page *page, int order)
2346{
2347        struct mem_cgroup *memcg = page->mem_cgroup;
2348        unsigned int nr_pages = 1 << order;
2349
2350        if (!memcg)
2351                return;
2352
2353        VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
2354
2355        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2356                page_counter_uncharge(&memcg->kmem, nr_pages);
2357
2358        page_counter_uncharge(&memcg->memory, nr_pages);
2359        if (do_memsw_account())
2360                page_counter_uncharge(&memcg->memsw, nr_pages);
2361
2362        page->mem_cgroup = NULL;
2363
2364        /* slab pages do not have PageKmemcg flag set */
2365        if (PageKmemcg(page))
2366                __ClearPageKmemcg(page);
2367
2368        css_put_many(&memcg->css, nr_pages);
2369}
2370#endif /* !CONFIG_SLOB */
2371
2372#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2373
2374/*
2375 * Because tail pages are not marked as "used", set it. We're under
2376 * zone_lru_lock and migration entries setup in all page mappings.
2377 */
2378void mem_cgroup_split_huge_fixup(struct page *head)
2379{
2380        int i;
2381
2382        if (mem_cgroup_disabled())
2383                return;
2384
2385        for (i = 1; i < HPAGE_PMD_NR; i++)
2386                head[i].mem_cgroup = head->mem_cgroup;
2387
2388        __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
2389                       HPAGE_PMD_NR);
2390}
2391#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2392
2393#ifdef CONFIG_MEMCG_SWAP
2394static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2395                                         bool charge)
2396{
2397        int val = (charge) ? 1 : -1;
2398        this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
2399}
2400
2401/**
2402 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2403 * @entry: swap entry to be moved
2404 * @from:  mem_cgroup which the entry is moved from
2405 * @to:  mem_cgroup which the entry is moved to
2406 *
2407 * It succeeds only when the swap_cgroup's record for this entry is the same
2408 * as the mem_cgroup's id of @from.
2409 *
2410 * Returns 0 on success, -EINVAL on failure.
2411 *
2412 * The caller must have charged to @to, IOW, called page_counter_charge() about
2413 * both res and memsw, and called css_get().
2414 */
2415static int mem_cgroup_move_swap_account(swp_entry_t entry,
2416                                struct mem_cgroup *from, struct mem_cgroup *to)
2417{
2418        unsigned short old_id, new_id;
2419
2420        old_id = mem_cgroup_id(from);
2421        new_id = mem_cgroup_id(to);
2422
2423        if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
2424                mem_cgroup_swap_statistics(from, false);
2425                mem_cgroup_swap_statistics(to, true);
2426                return 0;
2427        }
2428        return -EINVAL;
2429}
2430#else
2431static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
2432                                struct mem_cgroup *from, struct mem_cgroup *to)
2433{
2434        return -EINVAL;
2435}
2436#endif
2437
2438static DEFINE_MUTEX(memcg_limit_mutex);
2439
2440static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
2441                                   unsigned long limit)
2442{
2443        unsigned long curusage;
2444        unsigned long oldusage;
2445        bool enlarge = false;
2446        int retry_count;
2447        int ret;
2448
2449        /*
2450         * For keeping hierarchical_reclaim simple, how long we should retry
2451         * is depends on callers. We set our retry-count to be function
2452         * of # of children which we should visit in this loop.
2453         */
2454        retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2455                      mem_cgroup_count_children(memcg);
2456
2457        oldusage = page_counter_read(&memcg->memory);
2458
2459        do {
2460                if (signal_pending(current)) {
2461                        ret = -EINTR;
2462                        break;
2463                }
2464
2465                mutex_lock(&memcg_limit_mutex);
2466                if (limit > memcg->memsw.limit) {
2467                        mutex_unlock(&memcg_limit_mutex);
2468                        ret = -EINVAL;
2469                        break;
2470                }
2471                if (limit > memcg->memory.limit)
2472                        enlarge = true;
2473                ret = page_counter_limit(&memcg->memory, limit);
2474                mutex_unlock(&memcg_limit_mutex);
2475
2476                if (!ret)
2477                        break;
2478
2479                try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2480
2481                curusage = page_counter_read(&memcg->memory);
2482                /* Usage is reduced ? */
2483                if (curusage >= oldusage)
2484                        retry_count--;
2485                else
2486                        oldusage = curusage;
2487        } while (retry_count);
2488
2489        if (!ret && enlarge)
2490                memcg_oom_recover(memcg);
2491
2492        return ret;
2493}
2494
2495static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2496                                         unsigned long limit)
2497{
2498        unsigned long curusage;
2499        unsigned long oldusage;
2500        bool enlarge = false;
2501        int retry_count;
2502        int ret;
2503
2504        /* see mem_cgroup_resize_res_limit */
2505        retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2506                      mem_cgroup_count_children(memcg);
2507
2508        oldusage = page_counter_read(&memcg->memsw);
2509
2510        do {
2511                if (signal_pending(current)) {
2512                        ret = -EINTR;
2513                        break;
2514                }
2515
2516                mutex_lock(&memcg_limit_mutex);
2517                if (limit < memcg->memory.limit) {
2518                        mutex_unlock(&memcg_limit_mutex);
2519                        ret = -EINVAL;
2520                        break;
2521                }
2522                if (limit > memcg->memsw.limit)
2523                        enlarge = true;
2524                ret = page_counter_limit(&memcg->memsw, limit);
2525                mutex_unlock(&memcg_limit_mutex);
2526
2527                if (!ret)
2528                        break;
2529
2530                try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2531
2532                curusage = page_counter_read(&memcg->memsw);
2533                /* Usage is reduced ? */
2534                if (curusage >= oldusage)
2535                        retry_count--;
2536                else
2537                        oldusage = curusage;
2538        } while (retry_count);
2539
2540        if (!ret && enlarge)
2541                memcg_oom_recover(memcg);
2542
2543        return ret;
2544}
2545
2546unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
2547                                            gfp_t gfp_mask,
2548                                            unsigned long *total_scanned)
2549{
2550        unsigned long nr_reclaimed = 0;
2551        struct mem_cgroup_per_node *mz, *next_mz = NULL;
2552        unsigned long reclaimed;
2553        int loop = 0;
2554        struct mem_cgroup_tree_per_node *mctz;
2555        unsigned long excess;
2556        unsigned long nr_scanned;
2557
2558        if (order > 0)
2559                return 0;
2560
2561        mctz = soft_limit_tree_node(pgdat->node_id);
2562
2563        /*
2564         * Do not even bother to check the largest node if the root
2565         * is empty. Do it lockless to prevent lock bouncing. Races
2566         * are acceptable as soft limit is best effort anyway.
2567         */
2568        if (RB_EMPTY_ROOT(&mctz->rb_root))
2569                return 0;
2570
2571        /*
2572         * This loop can run a while, specially if mem_cgroup's continuously
2573         * keep exceeding their soft limit and putting the system under
2574         * pressure
2575         */
2576        do {
2577                if (next_mz)
2578                        mz = next_mz;
2579                else
2580                        mz = mem_cgroup_largest_soft_limit_node(mctz);
2581                if (!mz)
2582                        break;
2583
2584                nr_scanned = 0;
2585                reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
2586                                                    gfp_mask, &nr_scanned);
2587                nr_reclaimed += reclaimed;
2588                *total_scanned += nr_scanned;
2589                spin_lock_irq(&mctz->lock);
2590                __mem_cgroup_remove_exceeded(mz, mctz);
2591
2592                /*
2593                 * If we failed to reclaim anything from this memory cgroup
2594                 * it is time to move on to the next cgroup
2595                 */
2596                next_mz = NULL;
2597                if (!reclaimed)
2598                        next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2599
2600                excess = soft_limit_excess(mz->memcg);
2601                /*
2602                 * One school of thought says that we should not add
2603                 * back the node to the tree if reclaim returns 0.
2604                 * But our reclaim could return 0, simply because due
2605                 * to priority we are exposing a smaller subset of
2606                 * memory to reclaim from. Consider this as a longer
2607                 * term TODO.
2608                 */
2609                /* If excess == 0, no tree ops */
2610                __mem_cgroup_insert_exceeded(mz, mctz, excess);
2611                spin_unlock_irq(&mctz->lock);
2612                css_put(&mz->memcg->css);
2613                loop++;
2614                /*
2615                 * Could not reclaim anything and there are no more
2616                 * mem cgroups to try or we seem to be looping without
2617                 * reclaiming anything.
2618                 */
2619                if (!nr_reclaimed &&
2620                        (next_mz == NULL ||
2621                        loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2622                        break;
2623        } while (!nr_reclaimed);
2624        if (next_mz)
2625                css_put(&next_mz->memcg->css);
2626        return nr_reclaimed;
2627}
2628
2629/*
2630 * Test whether @memcg has children, dead or alive.  Note that this
2631 * function doesn't care whether @memcg has use_hierarchy enabled and
2632 * returns %true if there are child csses according to the cgroup
2633 * hierarchy.  Testing use_hierarchy is the caller's responsiblity.
2634 */
2635static inline bool memcg_has_children(struct mem_cgroup *memcg)
2636{
2637        bool ret;
2638
2639        rcu_read_lock();
2640        ret = css_next_child(NULL, &memcg->css);
2641        rcu_read_unlock();
2642        return ret;
2643}
2644
2645/*
2646 * Reclaims as many pages from the given memcg as possible.
2647 *
2648 * Caller is responsible for holding css reference for memcg.
2649 */
2650static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2651{
2652        int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
2653
2654        /* we call try-to-free pages for make this cgroup empty */
2655        lru_add_drain_all();
2656        /* try to free all pages in this cgroup */
2657        while (nr_retries && page_counter_read(&memcg->memory)) {
2658                int progress;
2659
2660                if (signal_pending(current))
2661                        return -EINTR;
2662
2663                progress = try_to_free_mem_cgroup_pages(memcg, 1,
2664                                                        GFP_KERNEL, true);
2665                if (!progress) {
2666                        nr_retries--;
2667                        /* maybe some writeback is necessary */
2668                        congestion_wait(BLK_RW_ASYNC, HZ/10);
2669                }
2670
2671        }
2672
2673        return 0;
2674}
2675
2676static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2677                                            char *buf, size_t nbytes,
2678                                            loff_t off)
2679{
2680        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
2681
2682        if (mem_cgroup_is_root(memcg))
2683                return -EINVAL;
2684        return mem_cgroup_force_empty(memcg) ?: nbytes;
2685}
2686
2687static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2688                                     struct cftype *cft)
2689{
2690        return mem_cgroup_from_css(css)->use_hierarchy;
2691}
2692
2693static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2694                                      struct cftype *cft, u64 val)
2695{
2696        int retval = 0;
2697        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
2698        struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
2699
2700        if (memcg->use_hierarchy == val)
2701                return 0;
2702
2703        /*
2704         * If parent's use_hierarchy is set, we can't make any modifications
2705         * in the child subtrees. If it is unset, then the change can
2706         * occur, provided the current cgroup has no children.
2707         *
2708         * For the root cgroup, parent_mem is NULL, we allow value to be
2709         * set if there are no children.
2710         */
2711        if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
2712                                (val == 1 || val == 0)) {
2713                if (!memcg_has_children(memcg))
2714                        memcg->use_hierarchy = val;
2715                else
2716                        retval = -EBUSY;
2717        } else
2718                retval = -EINVAL;
2719
2720        return retval;
2721}
2722
2723static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
2724{
2725        struct mem_cgroup *iter;
2726        int i;
2727
2728        memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
2729
2730        for_each_mem_cgroup_tree(iter, memcg) {
2731                for (i = 0; i < MEMCG_NR_STAT; i++)
2732                        stat[i] += mem_cgroup_read_stat(iter, i);
2733        }
2734}
2735
2736static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
2737{
2738        struct mem_cgroup *iter;
2739        int i;
2740
2741        memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
2742
2743        for_each_mem_cgroup_tree(iter, memcg) {
2744                for (i = 0; i < MEMCG_NR_EVENTS; i++)
2745                        events[i] += mem_cgroup_read_events(iter, i);
2746        }
2747}
2748
2749static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
2750{
2751        unsigned long val = 0;
2752
2753        if (mem_cgroup_is_root(memcg)) {
2754                struct mem_cgroup *iter;
2755
2756                for_each_mem_cgroup_tree(iter, memcg) {
2757                        val += mem_cgroup_read_stat(iter,
2758                                        MEM_CGROUP_STAT_CACHE);
2759                        val += mem_cgroup_read_stat(iter,
2760                                        MEM_CGROUP_STAT_RSS);
2761                        if (swap)
2762                                val += mem_cgroup_read_stat(iter,
2763                                                MEM_CGROUP_STAT_SWAP);
2764                }
2765        } else {
2766                if (!swap)
2767                        val = page_counter_read(&memcg->memory);
2768                else
2769                        val = page_counter_read(&memcg->memsw);
2770        }
2771        return val;
2772}
2773
2774enum {
2775        RES_USAGE,
2776        RES_LIMIT,
2777        RES_MAX_USAGE,
2778        RES_FAILCNT,
2779        RES_SOFT_LIMIT,
2780};
2781
2782static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
2783                               struct cftype *cft)
2784{
2785        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
2786        struct page_counter *counter;
2787
2788        switch (MEMFILE_TYPE(cft->private)) {
2789        case _MEM:
2790                counter = &memcg->memory;
2791                break;
2792        case _MEMSWAP:
2793                counter = &memcg->memsw;
2794                break;
2795        case _KMEM:
2796                counter = &memcg->kmem;
2797                break;
2798        case _TCP:
2799                counter = &memcg->tcpmem;
2800                break;
2801        default:
2802                BUG();
2803        }
2804
2805        switch (MEMFILE_ATTR(cft->private)) {
2806        case RES_USAGE:
2807                if (counter == &memcg->memory)
2808                        return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
2809                if (counter == &memcg->memsw)
2810                        return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
2811                return (u64)page_counter_read(counter) * PAGE_SIZE;
2812        case RES_LIMIT:
2813                return (u64)counter->limit * PAGE_SIZE;
2814        case RES_MAX_USAGE:
2815                return (u64)counter->watermark * PAGE_SIZE;
2816        case RES_FAILCNT:
2817                return counter->failcnt;
2818        case RES_SOFT_LIMIT:
2819                return (u64)memcg->soft_limit * PAGE_SIZE;
2820        default:
2821                BUG();
2822        }
2823}
2824
2825#ifndef CONFIG_SLOB
2826static int memcg_online_kmem(struct mem_cgroup *memcg)
2827{
2828        int memcg_id;
2829
2830        if (cgroup_memory_nokmem)
2831                return 0;
2832
2833        BUG_ON(memcg->kmemcg_id >= 0);
2834        BUG_ON(memcg->kmem_state);
2835
2836        memcg_id = memcg_alloc_cache_id();
2837        if (memcg_id < 0)
2838                return memcg_id;
2839
2840        static_branch_inc(&memcg_kmem_enabled_key);
2841        /*
2842         * A memory cgroup is considered kmem-online as soon as it gets
2843         * kmemcg_id. Setting the id after enabling static branching will
2844         * guarantee no one starts accounting before all call sites are
2845         * patched.
2846         */
2847        memcg->kmemcg_id = memcg_id;
2848        memcg->kmem_state = KMEM_ONLINE;
2849
2850        return 0;
2851}
2852
2853static void memcg_offline_kmem(struct mem_cgroup *memcg)
2854{
2855        struct cgroup_subsys_state *css;
2856        struct mem_cgroup *parent, *child;
2857        int kmemcg_id;
2858
2859        if (memcg->kmem_state != KMEM_ONLINE)
2860                return;
2861        /*
2862         * Clear the online state before clearing memcg_caches array
2863         * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2864         * guarantees that no cache will be created for this cgroup
2865         * after we are done (see memcg_create_kmem_cache()).
2866         */
2867        memcg->kmem_state = KMEM_ALLOCATED;
2868
2869        memcg_deactivate_kmem_caches(memcg);
2870
2871        kmemcg_id = memcg->kmemcg_id;
2872        BUG_ON(kmemcg_id < 0);
2873
2874        parent = parent_mem_cgroup(memcg);
2875        if (!parent)
2876                parent = root_mem_cgroup;
2877
2878        /*
2879         * Change kmemcg_id of this cgroup and all its descendants to the
2880         * parent's id, and then move all entries from this cgroup's list_lrus
2881         * to ones of the parent. After we have finished, all list_lrus
2882         * corresponding to this cgroup are guaranteed to remain empty. The
2883         * ordering is imposed by list_lru_node->lock taken by
2884         * memcg_drain_all_list_lrus().
2885         */
2886        rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
2887        css_for_each_descendant_pre(css, &memcg->css) {
2888                child = mem_cgroup_from_css(css);
2889                BUG_ON(child->kmemcg_id != kmemcg_id);
2890                child->kmemcg_id = parent->kmemcg_id;
2891                if (!memcg->use_hierarchy)
2892                        break;
2893        }
2894        rcu_read_unlock();
2895
2896        memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2897
2898        memcg_free_cache_id(kmemcg_id);
2899}
2900
2901static void memcg_free_kmem(struct mem_cgroup *memcg)
2902{
2903        /* css_alloc() failed, offlining didn't happen */
2904        if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2905                memcg_offline_kmem(memcg);
2906
2907        if (memcg->kmem_state == KMEM_ALLOCATED) {
2908                memcg_destroy_kmem_caches(memcg);
2909                static_branch_dec(&memcg_kmem_enabled_key);
2910                WARN_ON(page_counter_read(&memcg->kmem));
2911        }
2912}
2913#else
2914static int memcg_online_kmem(struct mem_cgroup *memcg)
2915{
2916        return 0;
2917}
2918static void memcg_offline_kmem(struct mem_cgroup *memcg)
2919{
2920}
2921static void memcg_free_kmem(struct mem_cgroup *memcg)
2922{
2923}
2924#endif /* !CONFIG_SLOB */
2925
2926static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2927                                   unsigned long limit)
2928{
2929        int ret;
2930
2931        mutex_lock(&memcg_limit_mutex);
2932        ret = page_counter_limit(&memcg->kmem, limit);
2933        mutex_unlock(&memcg_limit_mutex);
2934        return ret;
2935}
2936
2937static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2938{
2939        int ret;
2940
2941        mutex_lock(&memcg_limit_mutex);
2942
2943        ret = page_counter_limit(&memcg->tcpmem, limit);
2944        if (ret)
2945                goto out;
2946
2947        if (!memcg->tcpmem_active) {
2948                /*
2949                 * The active flag needs to be written after the static_key
2950                 * update. This is what guarantees that the socket activation
2951                 * function is the last one to run. See mem_cgroup_sk_alloc()
2952                 * for details, and note that we don't mark any socket as
2953                 * belonging to this memcg until that flag is up.
2954                 *
2955                 * We need to do this, because static_keys will span multiple
2956                 * sites, but we can't control their order. If we mark a socket
2957                 * as accounted, but the accounting functions are not patched in
2958                 * yet, we'll lose accounting.
2959                 *
2960                 * We never race with the readers in mem_cgroup_sk_alloc(),
2961                 * because when this value change, the code to process it is not
2962                 * patched in yet.
2963                 */
2964                static_branch_inc(&memcg_sockets_enabled_key);
2965                memcg->tcpmem_active = true;
2966        }
2967out:
2968        mutex_unlock(&memcg_limit_mutex);
2969        return ret;
2970}
2971
2972/*
2973 * The user of this function is...
2974 * RES_LIMIT.
2975 */
2976static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2977                                char *buf, size_t nbytes, loff_t off)
2978{
2979        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
2980        unsigned long nr_pages;
2981        int ret;
2982
2983        buf = strstrip(buf);
2984        ret = page_counter_memparse(buf, "-1", &nr_pages);
2985        if (ret)
2986                return ret;
2987
2988        switch (MEMFILE_ATTR(of_cft(of)->private)) {
2989        case RES_LIMIT:
2990                if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2991                        ret = -EINVAL;
2992                        break;
2993                }
2994                switch (MEMFILE_TYPE(of_cft(of)->private)) {
2995                case _MEM:
2996                        ret = mem_cgroup_resize_limit(memcg, nr_pages);
2997                        break;
2998                case _MEMSWAP:
2999                        ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3000                        break;
3001                case _KMEM:
3002                        ret = memcg_update_kmem_limit(memcg, nr_pages);
3003                        break;
3004                case _TCP:
3005                        ret = memcg_update_tcp_limit(memcg, nr_pages);
3006                        break;
3007                }
3008                break;
3009        case RES_SOFT_LIMIT:
3010                memcg->soft_limit = nr_pages;
3011                ret = 0;
3012                break;
3013        }
3014        return ret ?: nbytes;
3015}
3016
3017static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3018                                size_t nbytes, loff_t off)
3019{
3020        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
3021        struct page_counter *counter;
3022
3023        switch (MEMFILE_TYPE(of_cft(of)->private)) {
3024        case _MEM:
3025                counter = &memcg->memory;
3026                break;
3027        case _MEMSWAP:
3028                counter = &memcg->memsw;
3029                break;
3030        case _KMEM:
3031                counter = &memcg->kmem;
3032                break;
3033        case _TCP:
3034                counter = &memcg->tcpmem;
3035                break;
3036        default:
3037                BUG();
3038        }
3039
3040        switch (MEMFILE_ATTR(of_cft(of)->private)) {
3041        case RES_MAX_USAGE:
3042                page_counter_reset_watermark(counter);
3043                break;
3044        case RES_FAILCNT:
3045                counter->failcnt = 0;
3046                break;
3047        default:
3048                BUG();
3049        }
3050
3051        return nbytes;
3052}
3053
3054static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
3055                                        struct cftype *cft)
3056{
3057        return mem_cgroup_from_css(css)->move_charge_at_immigrate;
3058}
3059
3060#ifdef CONFIG_MMU
3061static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
3062                                        struct cftype *cft, u64 val)
3063{
3064        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3065
3066        if (val & ~MOVE_MASK)
3067                return -EINVAL;
3068
3069        /*
3070         * No kind of locking is needed in here, because ->can_attach() will
3071         * check this value once in the beginning of the process, and then carry
3072         * on with stale data. This means that changes to this value will only
3073         * affect task migrations starting after the change.
3074         */
3075        memcg->move_charge_at_immigrate = val;
3076        return 0;
3077}
3078#else
3079static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
3080                                        struct cftype *cft, u64 val)
3081{
3082        return -ENOSYS;
3083}
3084#endif
3085
3086#ifdef CONFIG_NUMA
3087static int memcg_numa_stat_show(struct seq_file *m, void *v)
3088{
3089        struct numa_stat {
3090                const char *name;
3091                unsigned int lru_mask;
3092        };
3093
3094        static const struct numa_stat stats[] = {
3095                { "total", LRU_ALL },
3096                { "file", LRU_ALL_FILE },
3097                { "anon", LRU_ALL_ANON },
3098                { "unevictable", BIT(LRU_UNEVICTABLE) },
3099        };
3100        const struct numa_stat *stat;
3101        int nid;
3102        unsigned long nr;
3103        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
3104
3105        for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3106                nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3107                seq_printf(m, "%s=%lu", stat->name, nr);
3108                for_each_node_state(nid, N_MEMORY) {
3109                        nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3110                                                          stat->lru_mask);
3111                        seq_printf(m, " N%d=%lu", nid, nr);
3112                }
3113                seq_putc(m, '\n');
3114        }
3115
3116        for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3117                struct mem_cgroup *iter;
3118
3119                nr = 0;
3120                for_each_mem_cgroup_tree(iter, memcg)
3121                        nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3122                seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3123                for_each_node_state(nid, N_MEMORY) {
3124                        nr = 0;
3125                        for_each_mem_cgroup_tree(iter, memcg)
3126                                nr += mem_cgroup_node_nr_lru_pages(
3127                                        iter, nid, stat->lru_mask);
3128                        seq_printf(m, " N%d=%lu", nid, nr);
3129                }
3130                seq_putc(m, '\n');
3131        }
3132
3133        return 0;
3134}
3135#endif /* CONFIG_NUMA */
3136
3137static int memcg_stat_show(struct seq_file *m, void *v)
3138{
3139        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
3140        unsigned long memory, memsw;
3141        struct mem_cgroup *mi;
3142        unsigned int i;
3143
3144        BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3145                     MEM_CGROUP_STAT_NSTATS);
3146        BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3147                     MEM_CGROUP_EVENTS_NSTATS);
3148        BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3149
3150        for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
3151                if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
3152                        continue;
3153                seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i],
3154                           mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
3155        }
3156
3157        for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3158                seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3159                           mem_cgroup_read_events(memcg, i));
3160
3161        for (i = 0; i < NR_LRU_LISTS; i++)
3162                seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3163                           mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3164
3165        /* Hierarchical information */
3166        memory = memsw = PAGE_COUNTER_MAX;
3167        for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3168                memory = min(memory, mi->memory.limit);
3169                memsw = min(memsw, mi->memsw.limit);
3170        }
3171        seq_printf(m, "hierarchical_memory_limit %llu\n",
3172                   (u64)memory * PAGE_SIZE);
3173        if (do_memsw_account())
3174                seq_printf(m, "hierarchical_memsw_limit %llu\n",
3175                           (u64)memsw * PAGE_SIZE);
3176
3177        for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
3178                unsigned long long val = 0;
3179
3180                if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account())
3181                        continue;
3182                for_each_mem_cgroup_tree(mi, memcg)
3183                        val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
3184                seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val);
3185        }
3186
3187        for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3188                unsigned long long val = 0;
3189
3190                for_each_mem_cgroup_tree(mi, memcg)
3191                        val += mem_cgroup_read_events(mi, i);
3192                seq_printf(m, "total_%s %llu\n",
3193                           mem_cgroup_events_names[i], val);
3194        }
3195
3196        for (i = 0; i < NR_LRU_LISTS; i++) {
3197                unsigned long long val = 0;
3198
3199                for_each_mem_cgroup_tree(mi, memcg)
3200                        val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3201                seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
3202        }
3203
3204#ifdef CONFIG_DEBUG_VM
3205        {
3206                pg_data_t *pgdat;
3207                struct mem_cgroup_per_node *mz;
3208                struct zone_reclaim_stat *rstat;
3209                unsigned long recent_rotated[2] = {0, 0};
3210                unsigned long recent_scanned[2] = {0, 0};
3211
3212                for_each_online_pgdat(pgdat) {
3213                        mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
3214                        rstat = &mz->lruvec.reclaim_stat;
3215
3216                        recent_rotated[0] += rstat->recent_rotated[0];
3217                        recent_rotated[1] += rstat->recent_rotated[1];
3218                        recent_scanned[0] += rstat->recent_scanned[0];
3219                        recent_scanned[1] += rstat->recent_scanned[1];
3220                }
3221                seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3222                seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3223                seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3224                seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
3225        }
3226#endif
3227
3228        return 0;
3229}
3230
3231static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3232                                      struct cftype *cft)
3233{
3234        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3235
3236        return mem_cgroup_swappiness(memcg);
3237}
3238
3239static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3240                                       struct cftype *cft, u64 val)
3241{
3242        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3243
3244        if (val > 100)
3245                return -EINVAL;
3246
3247        if (css->parent)
3248                memcg->swappiness = val;
3249        else
3250                vm_swappiness = val;
3251
3252        return 0;
3253}
3254
3255static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3256{
3257        struct mem_cgroup_threshold_ary *t;
3258        unsigned long usage;
3259        int i;
3260
3261        rcu_read_lock();
3262        if (!swap)
3263                t = rcu_dereference(memcg->thresholds.primary);
3264        else
3265                t = rcu_dereference(memcg->memsw_thresholds.primary);
3266
3267        if (!t)
3268                goto unlock;
3269
3270        usage = mem_cgroup_usage(memcg, swap);
3271
3272        /*
3273         * current_threshold points to threshold just below or equal to usage.
3274         * If it's not true, a threshold was crossed after last
3275         * call of __mem_cgroup_threshold().
3276         */
3277        i = t->current_threshold;
3278
3279        /*
3280         * Iterate backward over array of thresholds starting from
3281         * current_threshold and check if a threshold is crossed.
3282         * If none of thresholds below usage is crossed, we read
3283         * only one element of the array here.
3284         */
3285        for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3286                eventfd_signal(t->entries[i].eventfd, 1);
3287
3288        /* i = current_threshold + 1 */
3289        i++;
3290
3291        /*
3292         * Iterate forward over array of thresholds starting from
3293         * current_threshold+1 and check if a threshold is crossed.
3294         * If none of thresholds above usage is crossed, we read
3295         * only one element of the array here.
3296         */
3297        for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3298                eventfd_signal(t->entries[i].eventfd, 1);
3299
3300        /* Update current_threshold */
3301        t->current_threshold = i - 1;
3302unlock:
3303        rcu_read_unlock();
3304}
3305
3306static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3307{
3308        while (memcg) {
3309                __mem_cgroup_threshold(memcg, false);
3310                if (do_memsw_account())
3311                        __mem_cgroup_threshold(memcg, true);
3312
3313                memcg = parent_mem_cgroup(memcg);
3314        }
3315}
3316
3317static int compare_thresholds(const void *a, const void *b)
3318{
3319        const struct mem_cgroup_threshold *_a = a;
3320        const struct mem_cgroup_threshold *_b = b;
3321
3322        if (_a->threshold > _b->threshold)
3323                return 1;
3324
3325        if (_a->threshold < _b->threshold)
3326                return -1;
3327
3328        return 0;
3329}
3330
3331static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
3332{
3333        struct mem_cgroup_eventfd_list *ev;
3334
3335        spin_lock(&memcg_oom_lock);
3336
3337        list_for_each_entry(ev, &memcg->oom_notify, list)
3338                eventfd_signal(ev->eventfd, 1);
3339
3340        spin_unlock(&memcg_oom_lock);
3341        return 0;
3342}
3343
3344static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
3345{
3346        struct mem_cgroup *iter;
3347
3348        for_each_mem_cgroup_tree(iter, memcg)
3349                mem_cgroup_oom_notify_cb(iter);
3350}
3351
3352static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
3353        struct eventfd_ctx *eventfd, const char *args, enum res_type type)
3354{
3355        struct mem_cgroup_thresholds *thresholds;
3356        struct mem_cgroup_threshold_ary *new;
3357        unsigned long threshold;
3358        unsigned long usage;
3359        int i, size, ret;
3360
3361        ret = page_counter_memparse(args, "-1", &threshold);
3362        if (ret)
3363                return ret;
3364
3365        mutex_lock(&memcg->thresholds_lock);
3366
3367        if (type == _MEM) {
3368                thresholds = &memcg->thresholds;
3369                usage = mem_cgroup_usage(memcg, false);
3370        } else if (type == _MEMSWAP) {
3371                thresholds = &memcg->memsw_thresholds;
3372                usage = mem_cgroup_usage(memcg, true);
3373        } else
3374                BUG();
3375
3376        /* Check if a threshold crossed before adding a new one */
3377        if (thresholds->primary)
3378                __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3379
3380        size = thresholds->primary ? thresholds->primary->size + 1 : 1;
3381
3382        /* Allocate memory for new array of thresholds */
3383        new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
3384                        GFP_KERNEL);
3385        if (!new) {
3386                ret = -ENOMEM;
3387                goto unlock;
3388        }
3389        new->size = size;
3390
3391        /* Copy thresholds (if any) to new array */
3392        if (thresholds->primary) {
3393                memcpy(new->entries, thresholds->primary->entries, (size - 1) *
3394                                sizeof(struct mem_cgroup_threshold));
3395        }
3396
3397        /* Add new threshold */
3398        new->entries[size - 1].eventfd = eventfd;
3399        new->entries[size - 1].threshold = threshold;
3400
3401        /* Sort thresholds. Registering of new threshold isn't time-critical */
3402        sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
3403                        compare_thresholds, NULL);
3404
3405        /* Find current threshold */
3406        new->current_threshold = -1;
3407        for (i = 0; i < size; i++) {
3408                if (new->entries[i].threshold <= usage) {
3409                        /*
3410                         * new->current_threshold will not be used until
3411                         * rcu_assign_pointer(), so it's safe to increment
3412                         * it here.
3413                         */
3414                        ++new->current_threshold;
3415                } else
3416                        break;
3417        }
3418
3419        /* Free old spare buffer and save old primary buffer as spare */
3420        kfree(thresholds->spare);
3421        thresholds->spare = thresholds->primary;
3422
3423        rcu_assign_pointer(thresholds->primary, new);
3424
3425        /* To be sure that nobody uses thresholds */
3426        synchronize_rcu();
3427
3428unlock:
3429        mutex_unlock(&memcg->thresholds_lock);
3430
3431        return ret;
3432}
3433
3434static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
3435        struct eventfd_ctx *eventfd, const char *args)
3436{
3437        return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
3438}
3439
3440static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
3441        struct eventfd_ctx *eventfd, const char *args)
3442{
3443        return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
3444}
3445
3446static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3447        struct eventfd_ctx *eventfd, enum res_type type)
3448{
3449        struct mem_cgroup_thresholds *thresholds;
3450        struct mem_cgroup_threshold_ary *new;
3451        unsigned long usage;
3452        int i, j, size;
3453
3454        mutex_lock(&memcg->thresholds_lock);
3455
3456        if (type == _MEM) {
3457                thresholds = &memcg->thresholds;
3458                usage = mem_cgroup_usage(memcg, false);
3459        } else if (type == _MEMSWAP) {
3460                thresholds = &memcg->memsw_thresholds;
3461                usage = mem_cgroup_usage(memcg, true);
3462        } else
3463                BUG();
3464
3465        if (!thresholds->primary)
3466                goto unlock;
3467
3468        /* Check if a threshold crossed before removing */
3469        __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3470
3471        /* Calculate new number of threshold */
3472        size = 0;
3473        for (i = 0; i < thresholds->primary->size; i++) {
3474                if (thresholds->primary->entries[i].eventfd != eventfd)
3475                        size++;
3476        }
3477
3478        new = thresholds->spare;
3479
3480        /* Set thresholds array to NULL if we don't have thresholds */
3481        if (!size) {
3482                kfree(new);
3483                new = NULL;
3484                goto swap_buffers;
3485        }
3486
3487        new->size = size;
3488
3489        /* Copy thresholds and find current threshold */
3490        new->current_threshold = -1;
3491        for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3492                if (thresholds->primary->entries[i].eventfd == eventfd)
3493                        continue;
3494
3495                new->entries[j] = thresholds->primary->entries[i];
3496                if (new->entries[j].threshold <= usage) {
3497                        /*
3498                         * new->current_threshold will not be used
3499                         * until rcu_assign_pointer(), so it's safe to increment
3500                         * it here.
3501                         */
3502                        ++new->current_threshold;
3503                }
3504                j++;
3505        }
3506
3507swap_buffers:
3508        /* Swap primary and spare array */
3509        thresholds->spare = thresholds->primary;
3510
3511        rcu_assign_pointer(thresholds->primary, new);
3512
3513        /* To be sure that nobody uses thresholds */
3514        synchronize_rcu();
3515
3516        /* If all events are unregistered, free the spare array */
3517        if (!new) {
3518                kfree(thresholds->spare);
3519                thresholds->spare = NULL;
3520        }
3521unlock:
3522        mutex_unlock(&memcg->thresholds_lock);
3523}
3524
3525static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3526        struct eventfd_ctx *eventfd)
3527{
3528        return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
3529}
3530
3531static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3532        struct eventfd_ctx *eventfd)
3533{
3534        return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
3535}
3536
3537static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
3538        struct eventfd_ctx *eventfd, const char *args)
3539{
3540        struct mem_cgroup_eventfd_list *event;
3541
3542        event = kmalloc(sizeof(*event), GFP_KERNEL);
3543        if (!event)
3544                return -ENOMEM;
3545
3546        spin_lock(&memcg_oom_lock);
3547
3548        event->eventfd = eventfd;
3549        list_add(&event->list, &memcg->oom_notify);
3550
3551        /* already in OOM ? */
3552        if (memcg->under_oom)
3553                eventfd_signal(eventfd, 1);
3554        spin_unlock(&memcg_oom_lock);
3555
3556        return 0;
3557}
3558
3559static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
3560        struct eventfd_ctx *eventfd)
3561{
3562        struct mem_cgroup_eventfd_list *ev, *tmp;
3563
3564        spin_lock(&memcg_oom_lock);
3565
3566        list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
3567                if (ev->eventfd == eventfd) {
3568                        list_del(&ev->list);
3569                        kfree(ev);
3570                }
3571        }
3572
3573        spin_unlock(&memcg_oom_lock);
3574}
3575
3576static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
3577{
3578        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
3579
3580        seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
3581        seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
3582        return 0;
3583}
3584
3585static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
3586        struct cftype *cft, u64 val)
3587{
3588        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3589
3590        /* cannot set to root cgroup and only 0 and 1 are allowed */
3591        if (!css->parent || !((val == 0) || (val == 1)))
3592                return -EINVAL;
3593
3594        memcg->oom_kill_disable = val;
3595        if (!val)
3596                memcg_oom_recover(memcg);
3597
3598        return 0;
3599}
3600
3601#ifdef CONFIG_CGROUP_WRITEBACK
3602
3603struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3604{
3605        return &memcg->cgwb_list;
3606}
3607
3608static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3609{
3610        return wb_domain_init(&memcg->cgwb_domain, gfp);
3611}
3612
3613static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3614{
3615        wb_domain_exit(&memcg->cgwb_domain);
3616}
3617
3618static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3619{
3620        wb_domain_size_changed(&memcg->cgwb_domain);
3621}
3622
3623struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3624{
3625        struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3626
3627        if (!memcg->css.parent)
3628                return NULL;
3629
3630        return &memcg->cgwb_domain;
3631}
3632
3633/**
3634 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3635 * @wb: bdi_writeback in question
3636 * @pfilepages: out parameter for number of file pages
3637 * @pheadroom: out parameter for number of allocatable pages according to memcg
3638 * @pdirty: out parameter for number of dirty pages
3639 * @pwriteback: out parameter for number of pages under writeback
3640 *
3641 * Determine the numbers of file, headroom, dirty, and writeback pages in
3642 * @wb's memcg.  File, dirty and writeback are self-explanatory.  Headroom
3643 * is a bit more involved.
3644 *
3645 * A memcg's headroom is "min(max, high) - used".  In the hierarchy, the
3646 * headroom is calculated as the lowest headroom of itself and the
3647 * ancestors.  Note that this doesn't consider the actual amount of
3648 * available memory in the system.  The caller should further cap
3649 * *@pheadroom accordingly.
3650 */
3651void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3652                         unsigned long *pheadroom, unsigned long *pdirty,
3653                         unsigned long *pwriteback)
3654{
3655        struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3656        struct mem_cgroup *parent;
3657
3658        *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY);
3659
3660        /* this should eventually include NR_UNSTABLE_NFS */
3661        *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK);
3662        *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3663                                                     (1 << LRU_ACTIVE_FILE));
3664        *pheadroom = PAGE_COUNTER_MAX;
3665
3666        while ((parent = parent_mem_cgroup(memcg))) {
3667                unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3668                unsigned long used = page_counter_read(&memcg->memory);
3669
3670                *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
3671                memcg = parent;
3672        }
3673}
3674
3675#else   /* CONFIG_CGROUP_WRITEBACK */
3676
3677static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3678{
3679        return 0;
3680}
3681
3682static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3683{
3684}
3685
3686static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3687{
3688}
3689
3690#endif  /* CONFIG_CGROUP_WRITEBACK */
3691
3692/*
3693 * DO NOT USE IN NEW FILES.
3694 *
3695 * "cgroup.event_control" implementation.
3696 *
3697 * This is way over-engineered.  It tries to support fully configurable
3698 * events for each user.  Such level of flexibility is completely
3699 * unnecessary especially in the light of the planned unified hierarchy.
3700 *
3701 * Please deprecate this and replace with something simpler if at all
3702 * possible.
3703 */
3704
3705/*
3706 * Unregister event and free resources.
3707 *
3708 * Gets called from workqueue.
3709 */
3710static void memcg_event_remove(struct work_struct *work)
3711{
3712        struct mem_cgroup_event *event =
3713                container_of(work, struct mem_cgroup_event, remove);
3714        struct mem_cgroup *memcg = event->memcg;
3715
3716        remove_wait_queue(event->wqh, &event->wait);
3717
3718        event->unregister_event(memcg, event->eventfd);
3719
3720        /* Notify userspace the event is going away. */
3721        eventfd_signal(event->eventfd, 1);
3722
3723        eventfd_ctx_put(event->eventfd);
3724        kfree(event);
3725        css_put(&memcg->css);
3726}
3727
3728/*
3729 * Gets called on POLLHUP on eventfd when user closes it.
3730 *
3731 * Called with wqh->lock held and interrupts disabled.
3732 */
3733static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
3734                            int sync, void *key)
3735{
3736        struct mem_cgroup_event *event =
3737                container_of(wait, struct mem_cgroup_event, wait);
3738        struct mem_cgroup *memcg = event->memcg;
3739        unsigned long flags = (unsigned long)key;
3740
3741        if (flags & POLLHUP) {
3742                /*
3743                 * If the event has been detached at cgroup removal, we
3744                 * can simply return knowing the other side will cleanup
3745                 * for us.
3746                 *
3747                 * We can't race against event freeing since the other
3748                 * side will require wqh->lock via remove_wait_queue(),
3749                 * which we hold.
3750                 */
3751                spin_lock(&memcg->event_list_lock);
3752                if (!list_empty(&event->list)) {
3753                        list_del_init(&event->list);
3754                        /*
3755                         * We are in atomic context, but cgroup_event_remove()
3756                         * may sleep, so we have to call it in workqueue.
3757                         */
3758                        schedule_work(&event->remove);
3759                }
3760                spin_unlock(&memcg->event_list_lock);
3761        }
3762
3763        return 0;
3764}
3765
3766static void memcg_event_ptable_queue_proc(struct file *file,
3767                wait_queue_head_t *wqh, poll_table *pt)
3768{
3769        struct mem_cgroup_event *event =
3770                container_of(pt, struct mem_cgroup_event, pt);
3771
3772        event->wqh = wqh;
3773        add_wait_queue(wqh, &event->wait);
3774}
3775
3776/*
3777 * DO NOT USE IN NEW FILES.
3778 *
3779 * Parse input and register new cgroup event handler.
3780 *
3781 * Input must be in format '<event_fd> <control_fd> <args>'.
3782 * Interpretation of args is defined by control file implementation.
3783 */
3784static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3785                                         char *buf, size_t nbytes, loff_t off)
3786{
3787        struct cgroup_subsys_state *css = of_css(of);
3788        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3789        struct mem_cgroup_event *event;
3790        struct cgroup_subsys_state *cfile_css;
3791        unsigned int efd, cfd;
3792        struct fd efile;
3793        struct fd cfile;
3794        const char *name;
3795        char *endp;
3796        int ret;
3797
3798        buf = strstrip(buf);
3799
3800        efd = simple_strtoul(buf, &endp, 10);
3801        if (*endp != ' ')
3802                return -EINVAL;
3803        buf = endp + 1;
3804
3805        cfd = simple_strtoul(buf, &endp, 10);
3806        if ((*endp != ' ') && (*endp != '\0'))
3807                return -EINVAL;
3808        buf = endp + 1;
3809
3810        event = kzalloc(sizeof(*event), GFP_KERNEL);
3811        if (!event)
3812                return -ENOMEM;
3813
3814        event->memcg = memcg;
3815        INIT_LIST_HEAD(&event->list);
3816        init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3817        init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3818        INIT_WORK(&event->remove, memcg_event_remove);
3819
3820        efile = fdget(efd);
3821        if (!efile.file) {
3822                ret = -EBADF;
3823                goto out_kfree;
3824        }
3825
3826        event->eventfd = eventfd_ctx_fileget(efile.file);
3827        if (IS_ERR(event->eventfd)) {
3828                ret = PTR_ERR(event->eventfd);
3829                goto out_put_efile;
3830        }
3831
3832        cfile = fdget(cfd);
3833        if (!cfile.file) {
3834                ret = -EBADF;
3835                goto out_put_eventfd;
3836        }
3837
3838        /* the process need read permission on control file */
3839        /* AV: shouldn't we check that it's been opened for read instead? */
3840        ret = inode_permission(file_inode(cfile.file), MAY_READ);
3841        if (ret < 0)
3842                goto out_put_cfile;
3843
3844        /*
3845         * Determine the event callbacks and set them in @event.  This used
3846         * to be done via struct cftype but cgroup core no longer knows
3847         * about these events.  The following is crude but the whole thing
3848         * is for compatibility anyway.
3849         *
3850         * DO NOT ADD NEW FILES.
3851         */
3852        name = cfile.file->f_path.dentry->d_name.name;
3853
3854        if (!strcmp(name, "memory.usage_in_bytes")) {
3855                event->register_event = mem_cgroup_usage_register_event;
3856                event->unregister_event = mem_cgroup_usage_unregister_event;
3857        } else if (!strcmp(name, "memory.oom_control")) {
3858                event->register_event = mem_cgroup_oom_register_event;
3859                event->unregister_event = mem_cgroup_oom_unregister_event;
3860        } else if (!strcmp(name, "memory.pressure_level")) {
3861                event->register_event = vmpressure_register_event;
3862                event->unregister_event = vmpressure_unregister_event;
3863        } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
3864                event->register_event = memsw_cgroup_usage_register_event;
3865                event->unregister_event = memsw_cgroup_usage_unregister_event;
3866        } else {
3867                ret = -EINVAL;
3868                goto out_put_cfile;
3869        }
3870
3871        /*
3872         * Verify @cfile should belong to @css.  Also, remaining events are
3873         * automatically removed on cgroup destruction but the removal is
3874         * asynchronous, so take an extra ref on @css.
3875         */
3876        cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
3877                                               &memory_cgrp_subsys);
3878        ret = -EINVAL;
3879        if (IS_ERR(cfile_css))
3880                goto out_put_cfile;
3881        if (cfile_css != css) {
3882                css_put(cfile_css);
3883                goto out_put_cfile;
3884        }
3885
3886        ret = event->register_event(memcg, event->eventfd, buf);
3887        if (ret)
3888                goto out_put_css;
3889
3890        efile.file->f_op->poll(efile.file, &event->pt);
3891
3892        spin_lock(&memcg->event_list_lock);
3893        list_add(&event->list, &memcg->event_list);
3894        spin_unlock(&memcg->event_list_lock);
3895
3896        fdput(cfile);
3897        fdput(efile);
3898
3899        return nbytes;
3900
3901out_put_css:
3902        css_put(css);
3903out_put_cfile:
3904        fdput(cfile);
3905out_put_eventfd:
3906        eventfd_ctx_put(event->eventfd);
3907out_put_efile:
3908        fdput(efile);
3909out_kfree:
3910        kfree(event);
3911
3912        return ret;
3913}
3914
3915static struct cftype mem_cgroup_legacy_files[] = {
3916        {
3917                .name = "usage_in_bytes",
3918                .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
3919                .read_u64 = mem_cgroup_read_u64,
3920        },
3921        {
3922                .name = "max_usage_in_bytes",
3923                .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
3924                .write = mem_cgroup_reset,
3925                .read_u64 = mem_cgroup_read_u64,
3926        },
3927        {
3928                .name = "limit_in_bytes",
3929                .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
3930                .write = mem_cgroup_write,
3931                .read_u64 = mem_cgroup_read_u64,
3932        },
3933        {
3934                .name = "soft_limit_in_bytes",
3935                .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
3936                .write = mem_cgroup_write,
3937                .read_u64 = mem_cgroup_read_u64,
3938        },
3939        {
3940                .name = "failcnt",
3941                .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
3942                .write = mem_cgroup_reset,
3943                .read_u64 = mem_cgroup_read_u64,
3944        },
3945        {
3946                .name = "stat",
3947                .seq_show = memcg_stat_show,
3948        },
3949        {
3950                .name = "force_empty",
3951                .write = mem_cgroup_force_empty_write,
3952        },
3953        {
3954                .name = "use_hierarchy",
3955                .write_u64 = mem_cgroup_hierarchy_write,
3956                .read_u64 = mem_cgroup_hierarchy_read,
3957        },
3958        {
3959                .name = "cgroup.event_control",         /* XXX: for compat */
3960                .write = memcg_write_event_control,
3961                .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
3962        },
3963        {
3964                .name = "swappiness",
3965                .read_u64 = mem_cgroup_swappiness_read,
3966                .write_u64 = mem_cgroup_swappiness_write,
3967        },
3968        {
3969                .name = "move_charge_at_immigrate",
3970                .read_u64 = mem_cgroup_move_charge_read,
3971                .write_u64 = mem_cgroup_move_charge_write,
3972        },
3973        {
3974                .name = "oom_control",
3975                .seq_show = mem_cgroup_oom_control_read,
3976                .write_u64 = mem_cgroup_oom_control_write,
3977                .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3978        },
3979        {
3980                .name = "pressure_level",
3981        },
3982#ifdef CONFIG_NUMA
3983        {
3984                .name = "numa_stat",
3985                .seq_show = memcg_numa_stat_show,
3986        },
3987#endif
3988        {
3989                .name = "kmem.limit_in_bytes",
3990                .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
3991                .write = mem_cgroup_write,
3992                .read_u64 = mem_cgroup_read_u64,
3993        },
3994        {
3995                .name = "kmem.usage_in_bytes",
3996                .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
3997                .read_u64 = mem_cgroup_read_u64,
3998        },
3999        {
4000                .name = "kmem.failcnt",
4001                .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
4002                .write = mem_cgroup_reset,
4003                .read_u64 = mem_cgroup_read_u64,
4004        },
4005        {
4006                .name = "kmem.max_usage_in_bytes",
4007                .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
4008                .write = mem_cgroup_reset,
4009                .read_u64 = mem_cgroup_read_u64,
4010        },
4011#ifdef CONFIG_SLABINFO
4012        {
4013                .name = "kmem.slabinfo",
4014                .seq_start = slab_start,
4015                .seq_next = slab_next,
4016                .seq_stop = slab_stop,
4017                .seq_show = memcg_slab_show,
4018        },
4019#endif
4020        {
4021                .name = "kmem.tcp.limit_in_bytes",
4022                .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4023                .write = mem_cgroup_write,
4024                .read_u64 = mem_cgroup_read_u64,
4025        },
4026        {
4027                .name = "kmem.tcp.usage_in_bytes",
4028                .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4029                .read_u64 = mem_cgroup_read_u64,
4030        },
4031        {
4032                .name = "kmem.tcp.failcnt",
4033                .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4034                .write = mem_cgroup_reset,
4035                .read_u64 = mem_cgroup_read_u64,
4036        },
4037        {
4038                .name = "kmem.tcp.max_usage_in_bytes",
4039                .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4040                .write = mem_cgroup_reset,
4041                .read_u64 = mem_cgroup_read_u64,
4042        },
4043        { },    /* terminate */
4044};
4045
4046/*
4047 * Private memory cgroup IDR
4048 *
4049 * Swap-out records and page cache shadow entries need to store memcg
4050 * references in constrained space, so we maintain an ID space that is
4051 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4052 * memory-controlled cgroups to 64k.
4053 *
4054 * However, there usually are many references to the oflline CSS after
4055 * the cgroup has been destroyed, such as page cache or reclaimable
4056 * slab objects, that don't need to hang on to the ID. We want to keep
4057 * those dead CSS from occupying IDs, or we might quickly exhaust the
4058 * relatively small ID space and prevent the creation of new cgroups
4059 * even when there are much fewer than 64k cgroups - possibly none.
4060 *
4061 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4062 * be freed and recycled when it's no longer needed, which is usually
4063 * when the CSS is offlined.
4064 *
4065 * The only exception to that are records of swapped out tmpfs/shmem
4066 * pages that need to be attributed to live ancestors on swapin. But
4067 * those references are manageable from userspace.
4068 */
4069
4070static DEFINE_IDR(mem_cgroup_idr);
4071
4072static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n)
4073{
4074        VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0);
4075        atomic_add(n, &memcg->id.ref);
4076}
4077
4078static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
4079{
4080        VM_BUG_ON(atomic_read(&memcg->id.ref) < n);
4081        if (atomic_sub_and_test(n, &memcg->id.ref)) {
4082                idr_remove(&mem_cgroup_idr, memcg->id.id);
4083                memcg->id.id = 0;
4084
4085                /* Memcg ID pins CSS */
4086                css_put(&memcg->css);
4087        }
4088}
4089
4090static inline void mem_cgroup_id_get(struct mem_cgroup *memcg)
4091{
4092        mem_cgroup_id_get_many(memcg, 1);
4093}
4094
4095static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
4096{
4097        mem_cgroup_id_put_many(memcg, 1);
4098}
4099
4100/**
4101 * mem_cgroup_from_id - look up a memcg from a memcg id
4102 * @id: the memcg id to look up
4103 *
4104 * Caller must hold rcu_read_lock().
4105 */
4106struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
4107{
4108        WARN_ON_ONCE(!rcu_read_lock_held());
4109        return idr_find(&mem_cgroup_idr, id);
4110}
4111
4112static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
4113{
4114        struct mem_cgroup_per_node *pn;
4115        int tmp = node;
4116        /*
4117         * This routine is called against possible nodes.
4118         * But it's BUG to call kmalloc() against offline node.
4119         *
4120         * TODO: this routine can waste much memory for nodes which will
4121         *       never be onlined. It's better to use memory hotplug callback
4122         *       function.
4123         */
4124        if (!node_state(node, N_NORMAL_MEMORY))
4125                tmp = -1;
4126        pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
4127        if (!pn)
4128                return 1;
4129
4130        lruvec_init(&pn->lruvec);
4131        pn->usage_in_excess = 0;
4132        pn->on_tree = false;
4133        pn->memcg = memcg;
4134
4135        memcg->nodeinfo[node] = pn;
4136        return 0;
4137}
4138
4139static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
4140{
4141        kfree(memcg->nodeinfo[node]);
4142}
4143
4144static void mem_cgroup_free(struct mem_cgroup *memcg)
4145{
4146        int node;
4147
4148        memcg_wb_domain_exit(memcg);
4149        for_each_node(node)
4150                free_mem_cgroup_per_node_info(memcg, node);
4151        free_percpu(memcg->stat);
4152        kfree(memcg);
4153}
4154
4155static struct mem_cgroup *mem_cgroup_alloc(void)
4156{
4157        struct mem_cgroup *memcg;
4158        size_t size;
4159        int node;
4160
4161        size = sizeof(struct mem_cgroup);
4162        size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
4163
4164        memcg = kzalloc(size, GFP_KERNEL);
4165        if (!memcg)
4166                return NULL;
4167
4168        memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
4169                                 1, MEM_CGROUP_ID_MAX,
4170                                 GFP_KERNEL);
4171        if (memcg->id.id < 0)
4172                goto fail;
4173
4174        memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4175        if (!memcg->stat)
4176                goto fail;
4177
4178        for_each_node(node)
4179                if (alloc_mem_cgroup_per_node_info(memcg, node))
4180                        goto fail;
4181
4182        if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4183                goto fail;
4184
4185        INIT_WORK(&memcg->high_work, high_work_func);
4186        memcg->last_scanned_node = MAX_NUMNODES;
4187        INIT_LIST_HEAD(&memcg->oom_notify);
4188        mutex_init(&memcg->thresholds_lock);
4189        spin_lock_init(&memcg->move_lock);
4190        vmpressure_init(&memcg->vmpressure);
4191        INIT_LIST_HEAD(&memcg->event_list);
4192        spin_lock_init(&memcg->event_list_lock);
4193        memcg->socket_pressure = jiffies;
4194#ifndef CONFIG_SLOB
4195        memcg->kmemcg_id = -1;
4196#endif
4197#ifdef CONFIG_CGROUP_WRITEBACK
4198        INIT_LIST_HEAD(&memcg->cgwb_list);
4199#endif
4200        idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
4201        return memcg;
4202fail:
4203        if (memcg->id.id > 0)
4204                idr_remove(&mem_cgroup_idr, memcg->id.id);
4205        mem_cgroup_free(memcg);
4206        return NULL;
4207}
4208
4209static struct cgroup_subsys_state * __ref
4210mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
4211{
4212        struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4213        struct mem_cgroup *memcg;
4214        long error = -ENOMEM;
4215
4216        memcg = mem_cgroup_alloc();
4217        if (!memcg)
4218                return ERR_PTR(error);
4219
4220        memcg->high = PAGE_COUNTER_MAX;
4221        memcg->soft_limit = PAGE_COUNTER_MAX;
4222        if (parent) {
4223                memcg->swappiness = mem_cgroup_swappiness(parent);
4224                memcg->oom_kill_disable = parent->oom_kill_disable;
4225        }
4226        if (parent && parent->use_hierarchy) {
4227                memcg->use_hierarchy = true;
4228                page_counter_init(&memcg->memory, &parent->memory);
4229                page_counter_init(&memcg->swap, &parent->swap);
4230                page_counter_init(&memcg->memsw, &parent->memsw);
4231                page_counter_init(&memcg->kmem, &parent->kmem);
4232                page_counter_init(&memcg->tcpmem, &parent->tcpmem);
4233        } else {
4234                page_counter_init(&memcg->memory, NULL);
4235                page_counter_init(&memcg->swap, NULL);
4236                page_counter_init(&memcg->memsw, NULL);
4237                page_counter_init(&memcg->kmem, NULL);
4238                page_counter_init(&memcg->tcpmem, NULL);
4239                /*
4240                 * Deeper hierachy with use_hierarchy == false doesn't make
4241                 * much sense so let cgroup subsystem know about this
4242                 * unfortunate state in our controller.
4243                 */
4244                if (parent != root_mem_cgroup)
4245                        memory_cgrp_subsys.broken_hierarchy = true;
4246        }
4247
4248        /* The following stuff does not apply to the root */
4249        if (!parent) {
4250                root_mem_cgroup = memcg;
4251                return &memcg->css;
4252        }
4253
4254        error = memcg_online_kmem(memcg);
4255        if (error)
4256                goto fail;
4257
4258        if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
4259                static_branch_inc(&memcg_sockets_enabled_key);
4260
4261        return &memcg->css;
4262fail:
4263        mem_cgroup_free(memcg);
4264        return ERR_PTR(-ENOMEM);
4265}
4266
4267static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
4268{
4269        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4270
4271        /* Online state pins memcg ID, memcg ID pins CSS */
4272        atomic_set(&memcg->id.ref, 1);
4273        css_get(css);
4274        return 0;
4275}
4276
4277static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
4278{
4279        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4280        struct mem_cgroup_event *event, *tmp;
4281
4282        /*
4283         * Unregister events and notify userspace.
4284         * Notify userspace about cgroup removing only after rmdir of cgroup
4285         * directory to avoid race between userspace and kernelspace.
4286         */
4287        spin_lock(&memcg->event_list_lock);
4288        list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
4289                list_del_init(&event->list);
4290                schedule_work(&event->remove);
4291        }
4292        spin_unlock(&memcg->event_list_lock);
4293
4294        memcg_offline_kmem(memcg);
4295        wb_memcg_offline(memcg);
4296
4297        mem_cgroup_id_put(memcg);
4298}
4299
4300static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4301{
4302        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4303
4304        invalidate_reclaim_iterators(memcg);
4305}
4306
4307static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
4308{
4309        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4310
4311        if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
4312                static_branch_dec(&memcg_sockets_enabled_key);
4313
4314        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
4315                static_branch_dec(&memcg_sockets_enabled_key);
4316
4317        vmpressure_cleanup(&memcg->vmpressure);
4318        cancel_work_sync(&memcg->high_work);
4319        mem_cgroup_remove_from_trees(memcg);
4320        memcg_free_kmem(memcg);
4321        mem_cgroup_free(memcg);
4322}
4323
4324/**
4325 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4326 * @css: the target css
4327 *
4328 * Reset the states of the mem_cgroup associated with @css.  This is
4329 * invoked when the userland requests disabling on the default hierarchy
4330 * but the memcg is pinned through dependency.  The memcg should stop
4331 * applying policies and should revert to the vanilla state as it may be
4332 * made visible again.
4333 *
4334 * The current implementation only resets the essential configurations.
4335 * This needs to be expanded to cover all the visible parts.
4336 */
4337static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4338{
4339        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4340
4341        page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4342        page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4343        page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4344        page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4345        page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
4346        memcg->low = 0;
4347        memcg->high = PAGE_COUNTER_MAX;
4348        memcg->soft_limit = PAGE_COUNTER_MAX;
4349        memcg_wb_domain_size_changed(memcg);
4350}
4351
4352#ifdef CONFIG_MMU
4353/* Handlers for move charge at task migration. */
4354static int mem_cgroup_do_precharge(unsigned long count)
4355{
4356        int ret;
4357
4358        /* Try a single bulk charge without reclaim first, kswapd may wake */
4359        ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
4360        if (!ret) {
4361                mc.precharge += count;
4362                return ret;
4363        }
4364
4365        /* Try charges one by one with reclaim */
4366        while (count--) {
4367                ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
4368                if (ret)
4369                        return ret;
4370                mc.precharge++;
4371                cond_resched();
4372        }
4373        return 0;
4374}
4375
4376union mc_target {
4377        struct page     *page;
4378        swp_entry_t     ent;
4379};
4380
4381enum mc_target_type {
4382        MC_TARGET_NONE = 0,
4383        MC_TARGET_PAGE,
4384        MC_TARGET_SWAP,
4385};
4386
4387static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4388                                                unsigned long addr, pte_t ptent)
4389{
4390        struct page *page = vm_normal_page(vma, addr, ptent);
4391
4392        if (!page || !page_mapped(page))
4393                return NULL;
4394        if (PageAnon(page)) {
4395                if (!(mc.flags & MOVE_ANON))
4396                        return NULL;
4397        } else {
4398                if (!(mc.flags & MOVE_FILE))
4399                        return NULL;
4400        }
4401        if (!get_page_unless_zero(page))
4402                return NULL;
4403
4404        return page;
4405}
4406
4407#ifdef CONFIG_SWAP
4408static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4409                        pte_t ptent, swp_entry_t *entry)
4410{
4411        struct page *page = NULL;
4412        swp_entry_t ent = pte_to_swp_entry(ptent);
4413
4414        if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
4415                return NULL;
4416        /*
4417         * Because lookup_swap_cache() updates some statistics counter,
4418         * we call find_get_page() with swapper_space directly.
4419         */
4420        page = find_get_page(swap_address_space(ent), swp_offset(ent));
4421        if (do_memsw_account())
4422                entry->val = ent.val;
4423
4424        return page;
4425}
4426#else
4427static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4428                        pte_t ptent, swp_entry_t *entry)
4429{
4430        return NULL;
4431}
4432#endif
4433
4434static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4435                        unsigned long addr, pte_t ptent, swp_entry_t *entry)
4436{
4437        struct page *page = NULL;
4438        struct address_space *mapping;
4439        pgoff_t pgoff;
4440
4441        if (!vma->vm_file) /* anonymous vma */
4442                return NULL;
4443        if (!(mc.flags & MOVE_FILE))
4444                return NULL;
4445
4446        mapping = vma->vm_file->f_mapping;
4447        pgoff = linear_page_index(vma, addr);
4448
4449        /* page is moved even if it's not RSS of this task(page-faulted). */
4450#ifdef CONFIG_SWAP
4451        /* shmem/tmpfs may report page out on swap: account for that too. */
4452        if (shmem_mapping(mapping)) {
4453                page = find_get_entry(mapping, pgoff);
4454                if (radix_tree_exceptional_entry(page)) {
4455                        swp_entry_t swp = radix_to_swp_entry(page);
4456                        if (do_memsw_account())
4457                                *entry = swp;
4458                        page = find_get_page(swap_address_space(swp),
4459                                             swp_offset(swp));
4460                }
4461        } else
4462                page = find_get_page(mapping, pgoff);
4463#else
4464        page = find_get_page(mapping, pgoff);
4465#endif
4466        return page;
4467}
4468
4469/**
4470 * mem_cgroup_move_account - move account of the page
4471 * @page: the page
4472 * @compound: charge the page as compound or small page
4473 * @from: mem_cgroup which the page is moved from.
4474 * @to: mem_cgroup which the page is moved to. @from != @to.
4475 *
4476 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
4477 *
4478 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4479 * from old cgroup.
4480 */
4481static int mem_cgroup_move_account(struct page *page,
4482                                   bool compound,
4483                                   struct mem_cgroup *from,
4484                                   struct mem_cgroup *to)
4485{
4486        unsigned long flags;
4487        unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
4488        int ret;
4489        bool anon;
4490
4491        VM_BUG_ON(from == to);
4492        VM_BUG_ON_PAGE(PageLRU(page), page);
4493        VM_BUG_ON(compound && !PageTransHuge(page));
4494
4495        /*
4496         * Prevent mem_cgroup_migrate() from looking at
4497         * page->mem_cgroup of its source page while we change it.
4498         */
4499        ret = -EBUSY;
4500        if (!trylock_page(page))
4501                goto out;
4502
4503        ret = -EINVAL;
4504        if (page->mem_cgroup != from)
4505                goto out_unlock;
4506
4507        anon = PageAnon(page);
4508
4509        spin_lock_irqsave(&from->move_lock, flags);
4510
4511        if (!anon && page_mapped(page)) {
4512                __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4513                               nr_pages);
4514                __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4515                               nr_pages);
4516        }
4517
4518        /*
4519         * move_lock grabbed above and caller set from->moving_account, so
4520         * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4521         * So mapping should be stable for dirty pages.
4522         */
4523        if (!anon && PageDirty(page)) {
4524                struct address_space *mapping = page_mapping(page);
4525
4526                if (mapping_cap_account_dirty(mapping)) {
4527                        __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY],
4528                                       nr_pages);
4529                        __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY],
4530                                       nr_pages);
4531                }
4532        }
4533
4534        if (PageWriteback(page)) {
4535                __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4536                               nr_pages);
4537                __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4538                               nr_pages);
4539        }
4540
4541        /*
4542         * It is safe to change page->mem_cgroup here because the page
4543         * is referenced, charged, and isolated - we can't race with
4544         * uncharging, charging, migration, or LRU putback.
4545         */
4546
4547        /* caller should have done css_get */
4548        page->mem_cgroup = to;
4549        spin_unlock_irqrestore(&from->move_lock, flags);
4550
4551        ret = 0;
4552
4553        local_irq_disable();
4554        mem_cgroup_charge_statistics(to, page, compound, nr_pages);
4555        memcg_check_events(to, page);
4556        mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
4557        memcg_check_events(from, page);
4558        local_irq_enable();
4559out_unlock:
4560        unlock_page(page);
4561out:
4562        return ret;
4563}
4564
4565/**
4566 * get_mctgt_type - get target type of moving charge
4567 * @vma: the vma the pte to be checked belongs
4568 * @addr: the address corresponding to the pte to be checked
4569 * @ptent: the pte to be checked
4570 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4571 *
4572 * Returns
4573 *   0(MC_TARGET_NONE): if the pte is not a target for move charge.
4574 *   1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4575 *     move charge. if @target is not NULL, the page is stored in target->page
4576 *     with extra refcnt got(Callers should handle it).
4577 *   2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4578 *     target for charge migration. if @target is not NULL, the entry is stored
4579 *     in target->ent.
4580 *
4581 * Called with pte lock held.
4582 */
4583
4584static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
4585                unsigned long addr, pte_t ptent, union mc_target *target)
4586{
4587        struct page *page = NULL;
4588        enum mc_target_type ret = MC_TARGET_NONE;
4589        swp_entry_t ent = { .val = 0 };
4590
4591        if (pte_present(ptent))
4592                page = mc_handle_present_pte(vma, addr, ptent);
4593        else if (is_swap_pte(ptent))
4594                page = mc_handle_swap_pte(vma, ptent, &ent);
4595        else if (pte_none(ptent))
4596                page = mc_handle_file_pte(vma, addr, ptent, &ent);
4597
4598        if (!page && !ent.val)
4599                return ret;
4600        if (page) {
4601                /*
4602                 * Do only loose check w/o serialization.
4603                 * mem_cgroup_move_account() checks the page is valid or
4604                 * not under LRU exclusion.
4605                 */
4606                if (page->mem_cgroup == mc.from) {
4607                        ret = MC_TARGET_PAGE;
4608                        if (target)
4609                                target->page = page;
4610                }
4611                if (!ret || !target)
4612                        put_page(page);
4613        }
4614        /* There is a swap entry and a page doesn't exist or isn't charged */
4615        if (ent.val && !ret &&
4616            mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
4617                ret = MC_TARGET_SWAP;
4618                if (target)
4619                        target->ent = ent;
4620        }
4621        return ret;
4622}
4623
4624#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4625/*
4626 * We don't consider swapping or file mapped pages because THP does not
4627 * support them for now.
4628 * Caller should make sure that pmd_trans_huge(pmd) is true.
4629 */
4630static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4631                unsigned long addr, pmd_t pmd, union mc_target *target)
4632{
4633        struct page *page = NULL;
4634        enum mc_target_type ret = MC_TARGET_NONE;
4635
4636        page = pmd_page(pmd);
4637        VM_BUG_ON_PAGE(!page || !PageHead(page), page);
4638        if (!(mc.flags & MOVE_ANON))
4639                return ret;
4640        if (page->mem_cgroup == mc.from) {
4641                ret = MC_TARGET_PAGE;
4642                if (target) {
4643                        get_page(page);
4644                        target->page = page;
4645                }
4646        }
4647        return ret;
4648}
4649#else
4650static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4651                unsigned long addr, pmd_t pmd, union mc_target *target)
4652{
4653        return MC_TARGET_NONE;
4654}
4655#endif
4656
4657static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4658                                        unsigned long addr, unsigned long end,
4659                                        struct mm_walk *walk)
4660{
4661        struct vm_area_struct *vma = walk->vma;
4662        pte_t *pte;
4663        spinlock_t *ptl;
4664
4665        ptl = pmd_trans_huge_lock(pmd, vma);
4666        if (ptl) {
4667                if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4668                        mc.precharge += HPAGE_PMD_NR;
4669                spin_unlock(ptl);
4670                return 0;
4671        }
4672
4673        if (pmd_trans_unstable(pmd))
4674                return 0;
4675        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4676        for (; addr != end; pte++, addr += PAGE_SIZE)
4677                if (get_mctgt_type(vma, addr, *pte, NULL))
4678                        mc.precharge++; /* increment precharge temporarily */
4679        pte_unmap_unlock(pte - 1, ptl);
4680        cond_resched();
4681
4682        return 0;
4683}
4684
4685static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4686{
4687        unsigned long precharge;
4688
4689        struct mm_walk mem_cgroup_count_precharge_walk = {
4690                .pmd_entry = mem_cgroup_count_precharge_pte_range,
4691                .mm = mm,
4692        };
4693        down_read(&mm->mmap_sem);
4694        walk_page_range(0, mm->highest_vm_end,
4695                        &mem_cgroup_count_precharge_walk);
4696        up_read(&mm->mmap_sem);
4697
4698        precharge = mc.precharge;
4699        mc.precharge = 0;
4700
4701        return precharge;
4702}
4703
4704static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4705{
4706        unsigned long precharge = mem_cgroup_count_precharge(mm);
4707
4708        VM_BUG_ON(mc.moving_task);
4709        mc.moving_task = current;
4710        return mem_cgroup_do_precharge(precharge);
4711}
4712
4713/* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4714static void __mem_cgroup_clear_mc(void)
4715{
4716        struct mem_cgroup *from = mc.from;
4717        struct mem_cgroup *to = mc.to;
4718
4719        /* we must uncharge all the leftover precharges from mc.to */
4720        if (mc.precharge) {
4721                cancel_charge(mc.to, mc.precharge);
4722                mc.precharge = 0;
4723        }
4724        /*
4725         * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4726         * we must uncharge here.
4727         */
4728        if (mc.moved_charge) {
4729                cancel_charge(mc.from, mc.moved_charge);
4730                mc.moved_charge = 0;
4731        }
4732        /* we must fixup refcnts and charges */
4733        if (mc.moved_swap) {
4734                /* uncharge swap account from the old cgroup */
4735                if (!mem_cgroup_is_root(mc.from))
4736                        page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
4737
4738                mem_cgroup_id_put_many(mc.from, mc.moved_swap);
4739
4740                /*
4741                 * we charged both to->memory and to->memsw, so we
4742                 * should uncharge to->memory.
4743                 */
4744                if (!mem_cgroup_is_root(mc.to))
4745                        page_counter_uncharge(&mc.to->memory, mc.moved_swap);
4746
4747                mem_cgroup_id_get_many(mc.to, mc.moved_swap);
4748                css_put_many(&mc.to->css, mc.moved_swap);
4749
4750                mc.moved_swap = 0;
4751        }
4752        memcg_oom_recover(from);
4753        memcg_oom_recover(to);
4754        wake_up_all(&mc.waitq);
4755}
4756
4757static void mem_cgroup_clear_mc(void)
4758{
4759        struct mm_struct *mm = mc.mm;
4760
4761        /*
4762         * we must clear moving_task before waking up waiters at the end of
4763         * task migration.
4764         */
4765        mc.moving_task = NULL;
4766        __mem_cgroup_clear_mc();
4767        spin_lock(&mc.lock);
4768        mc.from = NULL;
4769        mc.to = NULL;
4770        mc.mm = NULL;
4771        spin_unlock(&mc.lock);
4772
4773        mmput(mm);
4774}
4775
4776static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
4777{
4778        struct cgroup_subsys_state *css;
4779        struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
4780        struct mem_cgroup *from;
4781        struct task_struct *leader, *p;
4782        struct mm_struct *mm;
4783        unsigned long move_flags;
4784        int ret = 0;
4785
4786        /* charge immigration isn't supported on the default hierarchy */
4787        if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
4788                return 0;
4789
4790        /*
4791         * Multi-process migrations only happen on the default hierarchy
4792         * where charge immigration is not used.  Perform charge
4793         * immigration if @tset contains a leader and whine if there are
4794         * multiple.
4795         */
4796        p = NULL;
4797        cgroup_taskset_for_each_leader(leader, css, tset) {
4798                WARN_ON_ONCE(p);
4799                p = leader;
4800                memcg = mem_cgroup_from_css(css);
4801        }
4802        if (!p)
4803                return 0;
4804
4805        /*
4806         * We are now commited to this value whatever it is. Changes in this
4807         * tunable will only affect upcoming migrations, not the current one.
4808         * So we need to save it, and keep it going.
4809         */
4810        move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4811        if (!move_flags)
4812                return 0;
4813
4814        from = mem_cgroup_from_task(p);
4815
4816        VM_BUG_ON(from == memcg);
4817
4818        mm = get_task_mm(p);
4819        if (!mm)
4820                return 0;
4821        /* We move charges only when we move a owner of the mm */
4822        if (mm->owner == p) {
4823                VM_BUG_ON(mc.from);
4824                VM_BUG_ON(mc.to);
4825                VM_BUG_ON(mc.precharge);
4826                VM_BUG_ON(mc.moved_charge);
4827                VM_BUG_ON(mc.moved_swap);
4828
4829                spin_lock(&mc.lock);
4830                mc.mm = mm;
4831                mc.from = from;
4832                mc.to = memcg;
4833                mc.flags = move_flags;
4834                spin_unlock(&mc.lock);
4835                /* We set mc.moving_task later */
4836
4837                ret = mem_cgroup_precharge_mc(mm);
4838                if (ret)
4839                        mem_cgroup_clear_mc();
4840        } else {
4841                mmput(mm);
4842        }
4843        return ret;
4844}
4845
4846static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
4847{
4848        if (mc.to)
4849                mem_cgroup_clear_mc();
4850}
4851
4852static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4853                                unsigned long addr, unsigned long end,
4854                                struct mm_walk *walk)
4855{
4856        int ret = 0;
4857        struct vm_area_struct *vma = walk->vma;
4858        pte_t *pte;
4859        spinlock_t *ptl;
4860        enum mc_target_type target_type;
4861        union mc_target target;
4862        struct page *page;
4863
4864        ptl = pmd_trans_huge_lock(pmd, vma);
4865        if (ptl) {
4866                if (mc.precharge < HPAGE_PMD_NR) {
4867                        spin_unlock(ptl);
4868                        return 0;
4869                }
4870                target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4871                if (target_type == MC_TARGET_PAGE) {
4872                        page = target.page;
4873                        if (!isolate_lru_page(page)) {
4874                                if (!mem_cgroup_move_account(page, true,
4875                                                             mc.from, mc.to)) {
4876                                        mc.precharge -= HPAGE_PMD_NR;
4877                                        mc.moved_charge += HPAGE_PMD_NR;
4878                                }
4879                                putback_lru_page(page);
4880                        }
4881                        put_page(page);
4882                }
4883                spin_unlock(ptl);
4884                return 0;
4885        }
4886
4887        if (pmd_trans_unstable(pmd))
4888                return 0;
4889retry:
4890        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4891        for (; addr != end; addr += PAGE_SIZE) {
4892                pte_t ptent = *(pte++);
4893                swp_entry_t ent;
4894
4895                if (!mc.precharge)
4896                        break;
4897
4898                switch (get_mctgt_type(vma, addr, ptent, &target)) {
4899                case MC_TARGET_PAGE:
4900                        page = target.page;
4901                        /*
4902                         * We can have a part of the split pmd here. Moving it
4903                         * can be done but it would be too convoluted so simply
4904                         * ignore such a partial THP and keep it in original
4905                         * memcg. There should be somebody mapping the head.
4906                         */
4907                        if (PageTransCompound(page))
4908                                goto put;
4909                        if (isolate_lru_page(page))
4910                                goto put;
4911                        if (!mem_cgroup_move_account(page, false,
4912                                                mc.from, mc.to)) {
4913                                mc.precharge--;
4914                                /* we uncharge from mc.from later. */
4915                                mc.moved_charge++;
4916                        }
4917                        putback_lru_page(page);
4918put:                    /* get_mctgt_type() gets the page */
4919                        put_page(page);
4920                        break;
4921                case MC_TARGET_SWAP:
4922                        ent = target.ent;
4923                        if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
4924                                mc.precharge--;
4925                                /* we fixup refcnts and charges later. */
4926                                mc.moved_swap++;
4927                        }
4928                        break;
4929                default:
4930                        break;
4931                }
4932        }
4933        pte_unmap_unlock(pte - 1, ptl);
4934        cond_resched();
4935
4936        if (addr != end) {
4937                /*
4938                 * We have consumed all precharges we got in can_attach().
4939                 * We try charge one by one, but don't do any additional
4940                 * charges to mc.to if we have failed in charge once in attach()
4941                 * phase.
4942                 */
4943                ret = mem_cgroup_do_precharge(1);
4944                if (!ret)
4945                        goto retry;
4946        }
4947
4948        return ret;
4949}
4950
4951static void mem_cgroup_move_charge(void)
4952{
4953        struct mm_walk mem_cgroup_move_charge_walk = {
4954                .pmd_entry = mem_cgroup_move_charge_pte_range,
4955                .mm = mc.mm,
4956        };
4957
4958        lru_add_drain_all();
4959        /*
4960         * Signal lock_page_memcg() to take the memcg's move_lock
4961         * while we're moving its pages to another memcg. Then wait
4962         * for already started RCU-only updates to finish.
4963         */
4964        atomic_inc(&mc.from->moving_account);
4965        synchronize_rcu();
4966retry:
4967        if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
4968                /*
4969                 * Someone who are holding the mmap_sem might be waiting in
4970                 * waitq. So we cancel all extra charges, wake up all waiters,
4971                 * and retry. Because we cancel precharges, we might not be able
4972                 * to move enough charges, but moving charge is a best-effort
4973                 * feature anyway, so it wouldn't be a big problem.
4974                 */
4975                __mem_cgroup_clear_mc();
4976                cond_resched();
4977                goto retry;
4978        }
4979        /*
4980         * When we have consumed all precharges and failed in doing
4981         * additional charge, the page walk just aborts.
4982         */
4983        walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk);
4984
4985        up_read(&mc.mm->mmap_sem);
4986        atomic_dec(&mc.from->moving_account);
4987}
4988
4989static void mem_cgroup_move_task(void)
4990{
4991        if (mc.to) {
4992                mem_cgroup_move_charge();
4993                mem_cgroup_clear_mc();
4994        }
4995}
4996#else   /* !CONFIG_MMU */
4997static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
4998{
4999        return 0;
5000}
5001static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
5002{
5003}
5004static void mem_cgroup_move_task(void)
5005{
5006}
5007#endif
5008
5009/*
5010 * Cgroup retains root cgroups across [un]mount cycles making it necessary
5011 * to verify whether we're attached to the default hierarchy on each mount
5012 * attempt.
5013 */
5014static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
5015{
5016        /*
5017         * use_hierarchy is forced on the default hierarchy.  cgroup core
5018         * guarantees that @root doesn't have any children, so turning it
5019         * on for the root memcg is enough.
5020         */
5021        if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5022                root_mem_cgroup->use_hierarchy = true;
5023        else
5024                root_mem_cgroup->use_hierarchy = false;
5025}
5026
5027static u64 memory_current_read(struct cgroup_subsys_state *css,
5028                               struct cftype *cft)
5029{
5030        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5031
5032        return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
5033}
5034
5035static int memory_low_show(struct seq_file *m, void *v)
5036{
5037        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5038        unsigned long low = READ_ONCE(memcg->low);
5039
5040        if (low == PAGE_COUNTER_MAX)
5041                seq_puts(m, "max\n");
5042        else
5043                seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5044
5045        return 0;
5046}
5047
5048static ssize_t memory_low_write(struct kernfs_open_file *of,
5049                                char *buf, size_t nbytes, loff_t off)
5050{
5051        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5052        unsigned long low;
5053        int err;
5054
5055        buf = strstrip(buf);
5056        err = page_counter_memparse(buf, "max", &low);
5057        if (err)
5058                return err;
5059
5060        memcg->low = low;
5061
5062        return nbytes;
5063}
5064
5065static int memory_high_show(struct seq_file *m, void *v)
5066{
5067        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5068        unsigned long high = READ_ONCE(memcg->high);
5069
5070        if (high == PAGE_COUNTER_MAX)
5071                seq_puts(m, "max\n");
5072        else
5073                seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5074
5075        return 0;
5076}
5077
5078static ssize_t memory_high_write(struct kernfs_open_file *of,
5079                                 char *buf, size_t nbytes, loff_t off)
5080{
5081        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5082        unsigned long nr_pages;
5083        unsigned long high;
5084        int err;
5085
5086        buf = strstrip(buf);
5087        err = page_counter_memparse(buf, "max", &high);
5088        if (err)
5089                return err;
5090
5091        memcg->high = high;
5092
5093        nr_pages = page_counter_read(&memcg->memory);
5094        if (nr_pages > high)
5095                try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5096                                             GFP_KERNEL, true);
5097
5098        memcg_wb_domain_size_changed(memcg);
5099        return nbytes;
5100}
5101
5102static int memory_max_show(struct seq_file *m, void *v)
5103{
5104        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5105        unsigned long max = READ_ONCE(memcg->memory.limit);
5106
5107        if (max == PAGE_COUNTER_MAX)
5108                seq_puts(m, "max\n");
5109        else
5110                seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5111
5112        return 0;
5113}
5114
5115static ssize_t memory_max_write(struct kernfs_open_file *of,
5116                                char *buf, size_t nbytes, loff_t off)
5117{
5118        struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5119        unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5120        bool drained = false;
5121        unsigned long max;
5122        int err;
5123
5124        buf = strstrip(buf);
5125        err = page_counter_memparse(buf, "max", &max);
5126        if (err)
5127                return err;
5128
5129        xchg(&memcg->memory.limit, max);
5130
5131        for (;;) {
5132                unsigned long nr_pages = page_counter_read(&memcg->memory);
5133
5134                if (nr_pages <= max)
5135                        break;
5136
5137                if (signal_pending(current)) {
5138                        err = -EINTR;
5139                        break;
5140                }
5141
5142                if (!drained) {
5143                        drain_all_stock(memcg);
5144                        drained = true;
5145                        continue;
5146                }
5147
5148                if (nr_reclaims) {
5149                        if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5150                                                          GFP_KERNEL, true))
5151                                nr_reclaims--;
5152                        continue;
5153                }
5154
5155                mem_cgroup_events(memcg, MEMCG_OOM, 1);
5156                if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5157                        break;
5158        }
5159
5160        memcg_wb_domain_size_changed(memcg);
5161        return nbytes;
5162}
5163
5164static int memory_events_show(struct seq_file *m, void *v)
5165{
5166        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5167
5168        seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5169        seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5170        seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5171        seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5172
5173        return 0;
5174}
5175
5176static int memory_stat_show(struct seq_file *m, void *v)
5177{
5178        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5179        unsigned long stat[MEMCG_NR_STAT];
5180        unsigned long events[MEMCG_NR_EVENTS];
5181        int i;
5182
5183        /*
5184         * Provide statistics on the state of the memory subsystem as
5185         * well as cumulative event counters that show past behavior.
5186         *
5187         * This list is ordered following a combination of these gradients:
5188         * 1) generic big picture -> specifics and details
5189         * 2) reflecting userspace activity -> reflecting kernel heuristics
5190         *
5191         * Current memory state:
5192         */
5193
5194        tree_stat(memcg, stat);
5195        tree_events(memcg, events);
5196
5197        seq_printf(m, "anon %llu\n",
5198                   (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE);
5199        seq_printf(m, "file %llu\n",
5200                   (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE);
5201        seq_printf(m, "kernel_stack %llu\n",
5202                   (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024);
5203        seq_printf(m, "slab %llu\n",
5204                   (u64)(stat[MEMCG_SLAB_RECLAIMABLE] +
5205                         stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
5206        seq_printf(m, "sock %llu\n",
5207                   (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
5208
5209        seq_printf(m, "file_mapped %llu\n",
5210                   (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE);
5211        seq_printf(m, "file_dirty %llu\n",
5212                   (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE);
5213        seq_printf(m, "file_writeback %llu\n",
5214                   (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE);
5215
5216        for (i = 0; i < NR_LRU_LISTS; i++) {
5217                struct mem_cgroup *mi;
5218                unsigned long val = 0;
5219
5220                for_each_mem_cgroup_tree(mi, memcg)
5221                        val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5222                seq_printf(m, "%s %llu\n",
5223                           mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5224        }
5225
5226        seq_printf(m, "slab_reclaimable %llu\n",
5227                   (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE);
5228        seq_printf(m, "slab_unreclaimable %llu\n",
5229                   (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5230
5231        /* Accumulated memory events */
5232
5233        seq_printf(m, "pgfault %lu\n",
5234                   events[MEM_CGROUP_EVENTS_PGFAULT]);
5235        seq_printf(m, "pgmajfault %lu\n",
5236                   events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
5237
5238        return 0;
5239}
5240
5241static struct cftype memory_files[] = {
5242        {
5243                .name = "current",
5244                .flags = CFTYPE_NOT_ON_ROOT,
5245                .read_u64 = memory_current_read,
5246        },
5247        {
5248                .name = "low",
5249                .flags = CFTYPE_NOT_ON_ROOT,
5250                .seq_show = memory_low_show,
5251                .write = memory_low_write,
5252        },
5253        {
5254                .name = "high",
5255                .flags = CFTYPE_NOT_ON_ROOT,
5256                .seq_show = memory_high_show,
5257                .write = memory_high_write,
5258        },
5259        {
5260                .name = "max",
5261                .flags = CFTYPE_NOT_ON_ROOT,
5262                .seq_show = memory_max_show,
5263                .write = memory_max_write,
5264        },
5265        {
5266                .name = "events",
5267                .flags = CFTYPE_NOT_ON_ROOT,
5268                .file_offset = offsetof(struct mem_cgroup, events_file),
5269                .seq_show = memory_events_show,
5270        },
5271        {
5272                .name = "stat",
5273                .flags = CFTYPE_NOT_ON_ROOT,
5274                .seq_show = memory_stat_show,
5275        },
5276        { }     /* terminate */
5277};
5278
5279struct cgroup_subsys memory_cgrp_subsys = {
5280        .css_alloc = mem_cgroup_css_alloc,
5281        .css_online = mem_cgroup_css_online,
5282        .css_offline = mem_cgroup_css_offline,
5283        .css_released = mem_cgroup_css_released,
5284        .css_free = mem_cgroup_css_free,
5285        .css_reset = mem_cgroup_css_reset,
5286        .can_attach = mem_cgroup_can_attach,
5287        .cancel_attach = mem_cgroup_cancel_attach,
5288        .post_attach = mem_cgroup_move_task,
5289        .bind = mem_cgroup_bind,
5290        .dfl_cftypes = memory_files,
5291        .legacy_cftypes = mem_cgroup_legacy_files,
5292        .early_init = 0,
5293};
5294
5295/**
5296 * mem_cgroup_low - check if memory consumption is below the normal range
5297 * @root: the highest ancestor to consider
5298 * @memcg: the memory cgroup to check
5299 *
5300 * Returns %true if memory consumption of @memcg, and that of all
5301 * configurable ancestors up to @root, is below the normal range.
5302 */
5303bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5304{
5305        if (mem_cgroup_disabled())
5306                return false;
5307
5308        /*
5309         * The toplevel group doesn't have a configurable range, so
5310         * it's never low when looked at directly, and it is not
5311         * considered an ancestor when assessing the hierarchy.
5312         */
5313
5314        if (memcg == root_mem_cgroup)
5315                return false;
5316
5317        if (page_counter_read(&memcg->memory) >= memcg->low)
5318                return false;
5319
5320        while (memcg != root) {
5321                memcg = parent_mem_cgroup(memcg);
5322
5323                if (memcg == root_mem_cgroup)
5324                        break;
5325
5326                if (page_counter_read(&memcg->memory) >= memcg->low)
5327                        return false;
5328        }
5329        return true;
5330}
5331
5332/**
5333 * mem_cgroup_try_charge - try charging a page
5334 * @page: page to charge
5335 * @mm: mm context of the victim
5336 * @gfp_mask: reclaim mode
5337 * @memcgp: charged memcg return
5338 * @compound: charge the page as compound or small page
5339 *
5340 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5341 * pages according to @gfp_mask if necessary.
5342 *
5343 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5344 * Otherwise, an error code is returned.
5345 *
5346 * After page->mapping has been set up, the caller must finalize the
5347 * charge with mem_cgroup_commit_charge().  Or abort the transaction
5348 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5349 */
5350int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
5351                          gfp_t gfp_mask, struct mem_cgroup **memcgp,
5352                          bool compound)
5353{
5354        struct mem_cgroup *memcg = NULL;
5355        unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
5356        int ret = 0;
5357
5358        if (mem_cgroup_disabled())
5359                goto out;
5360
5361        if (PageSwapCache(page)) {
5362                /*
5363                 * Every swap fault against a single page tries to charge the
5364                 * page, bail as early as possible.  shmem_unuse() encounters
5365                 * already charged pages, too.  The USED bit is protected by
5366                 * the page lock, which serializes swap cache removal, which
5367                 * in turn serializes uncharging.
5368                 */
5369                VM_BUG_ON_PAGE(!PageLocked(page), page);
5370                if (page->mem_cgroup)
5371                        goto out;
5372
5373                if (do_swap_account) {
5374                        swp_entry_t ent = { .val = page_private(page), };
5375                        unsigned short id = lookup_swap_cgroup_id(ent);
5376
5377                        rcu_read_lock();
5378                        memcg = mem_cgroup_from_id(id);
5379                        if (memcg && !css_tryget_online(&memcg->css))
5380                                memcg = NULL;
5381                        rcu_read_unlock();
5382                }
5383        }
5384
5385        if (!memcg)
5386                memcg = get_mem_cgroup_from_mm(mm);
5387
5388        ret = try_charge(memcg, gfp_mask, nr_pages);
5389
5390        css_put(&memcg->css);
5391out:
5392        *memcgp = memcg;
5393        return ret;
5394}
5395
5396/**
5397 * mem_cgroup_commit_charge - commit a page charge
5398 * @page: page to charge
5399 * @memcg: memcg to charge the page to
5400 * @lrucare: page might be on LRU already
5401 * @compound: charge the page as compound or small page
5402 *
5403 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5404 * after page->mapping has been set up.  This must happen atomically
5405 * as part of the page instantiation, i.e. under the page table lock
5406 * for anonymous pages, under the page lock for page and swap cache.
5407 *
5408 * In addition, the page must not be on the LRU during the commit, to
5409 * prevent racing with task migration.  If it might be, use @lrucare.
5410 *
5411 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5412 */
5413void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
5414                              bool lrucare, bool compound)
5415{
5416        unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
5417
5418        VM_BUG_ON_PAGE(!page->mapping, page);
5419        VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5420
5421        if (mem_cgroup_disabled())
5422                return;
5423        /*
5424         * Swap faults will attempt to charge the same page multiple
5425         * times.  But reuse_swap_page() might have removed the page
5426         * from swapcache already, so we can't check PageSwapCache().
5427         */
5428        if (!memcg)
5429                return;
5430
5431        commit_charge(page, memcg, lrucare);
5432
5433        local_irq_disable();
5434        mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
5435        memcg_check_events(memcg, page);
5436        local_irq_enable();
5437
5438        if (do_memsw_account() && PageSwapCache(page)) {
5439                swp_entry_t entry = { .val = page_private(page) };
5440                /*
5441                 * The swap entry might not get freed for a long time,
5442                 * let's not wait for it.  The page already received a
5443                 * memory+swap charge, drop the swap entry duplicate.
5444                 */
5445                mem_cgroup_uncharge_swap(entry);
5446        }
5447}
5448
5449/**
5450 * mem_cgroup_cancel_charge - cancel a page charge
5451 * @page: page to charge
5452 * @memcg: memcg to charge the page to
5453 * @compound: charge the page as compound or small page
5454 *
5455 * Cancel a charge transaction started by mem_cgroup_try_charge().
5456 */
5457void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5458                bool compound)
5459{
5460        unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
5461
5462        if (mem_cgroup_disabled())
5463                return;
5464        /*
5465         * Swap faults will attempt to charge the same page multiple
5466         * times.  But reuse_swap_page() might have removed the page
5467         * from swapcache already, so we can't check PageSwapCache().
5468         */
5469        if (!memcg)
5470                return;
5471
5472        cancel_charge(memcg, nr_pages);
5473}
5474
5475static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
5476                           unsigned long nr_anon, unsigned long nr_file,
5477                           unsigned long nr_huge, unsigned long nr_kmem,
5478                           struct page *dummy_page)
5479{
5480        unsigned long nr_pages = nr_anon + nr_file + nr_kmem;
5481        unsigned long flags;
5482
5483        if (!mem_cgroup_is_root(memcg)) {
5484                page_counter_uncharge(&memcg->memory, nr_pages);
5485                if (do_memsw_account())
5486                        page_counter_uncharge(&memcg->memsw, nr_pages);
5487                if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem)
5488                        page_counter_uncharge(&memcg->kmem, nr_kmem);
5489                memcg_oom_recover(memcg);
5490        }
5491
5492        local_irq_save(flags);
5493        __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5494        __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5495        __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5496        __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
5497        __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
5498        memcg_check_events(memcg, dummy_page);
5499        local_irq_restore(flags);
5500
5501        if (!mem_cgroup_is_root(memcg))
5502                css_put_many(&memcg->css, nr_pages);
5503}
5504
5505static void uncharge_list(struct list_head *page_list)
5506{
5507        struct mem_cgroup *memcg = NULL;
5508        unsigned long nr_anon = 0;
5509        unsigned long nr_file = 0;
5510        unsigned long nr_huge = 0;
5511        unsigned long nr_kmem = 0;
5512        unsigned long pgpgout = 0;
5513        struct list_head *next;
5514        struct page *page;
5515
5516        /*
5517         * Note that the list can be a single page->lru; hence the
5518         * do-while loop instead of a simple list_for_each_entry().
5519         */
5520        next = page_list->next;
5521        do {
5522                page = list_entry(next, struct page, lru);
5523                next = page->lru.next;
5524
5525                VM_BUG_ON_PAGE(PageLRU(page), page);
5526                VM_BUG_ON_PAGE(page_count(page), page);
5527
5528                if (!page->mem_cgroup)
5529                        continue;
5530
5531                /*
5532                 * Nobody should be changing or seriously looking at
5533                 * page->mem_cgroup at this point, we have fully
5534                 * exclusive access to the page.
5535                 */
5536
5537                if (memcg != page->mem_cgroup) {
5538                        if (memcg) {
5539                                uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5540                                               nr_huge, nr_kmem, page);
5541                                pgpgout = nr_anon = nr_file =
5542                                        nr_huge = nr_kmem = 0;
5543                        }
5544                        memcg = page->mem_cgroup;
5545                }
5546
5547                if (!PageKmemcg(page)) {
5548                        unsigned int nr_pages = 1;
5549
5550                        if (PageTransHuge(page)) {
5551                                nr_pages <<= compound_order(page);
5552                                nr_huge += nr_pages;
5553                        }
5554                        if (PageAnon(page))
5555                                nr_anon += nr_pages;
5556                        else
5557                                nr_file += nr_pages;
5558                        pgpgout++;
5559                } else {
5560                        nr_kmem += 1 << compound_order(page);
5561                        __ClearPageKmemcg(page);
5562                }
5563
5564                page->mem_cgroup = NULL;
5565        } while (next != page_list);
5566
5567        if (memcg)
5568                uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5569                               nr_huge, nr_kmem, page);
5570}
5571
5572/**
5573 * mem_cgroup_uncharge - uncharge a page
5574 * @page: page to uncharge
5575 *
5576 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5577 * mem_cgroup_commit_charge().
5578 */
5579void mem_cgroup_uncharge(struct page *page)
5580{
5581        if (mem_cgroup_disabled())
5582                return;
5583
5584        /* Don't touch page->lru of any random page, pre-check: */
5585        if (!page->mem_cgroup)
5586                return;
5587
5588        INIT_LIST_HEAD(&page->lru);
5589        uncharge_list(&page->lru);
5590}
5591
5592/**
5593 * mem_cgroup_uncharge_list - uncharge a list of page
5594 * @page_list: list of pages to uncharge
5595 *
5596 * Uncharge a list of pages previously charged with
5597 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5598 */
5599void mem_cgroup_uncharge_list(struct list_head *page_list)
5600{
5601        if (mem_cgroup_disabled())
5602                return;
5603
5604        if (!list_empty(page_list))
5605                uncharge_list(page_list);
5606}
5607
5608/**
5609 * mem_cgroup_migrate - charge a page's replacement
5610 * @oldpage: currently circulating page
5611 * @newpage: replacement page
5612 *
5613 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5614 * be uncharged upon free.
5615 *
5616 * Both pages must be locked, @newpage->mapping must be set up.
5617 */
5618void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
5619{
5620        struct mem_cgroup *memcg;
5621        unsigned int nr_pages;
5622        bool compound;
5623        unsigned long flags;
5624
5625        VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5626        VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
5627        VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
5628        VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5629                       newpage);
5630
5631        if (mem_cgroup_disabled())
5632                return;
5633
5634        /* Page cache replacement: new page already charged? */
5635        if (newpage->mem_cgroup)
5636                return;
5637
5638        /* Swapcache readahead pages can get replaced before being charged */
5639        memcg = oldpage->mem_cgroup;
5640        if (!memcg)
5641                return;
5642
5643        /* Force-charge the new page. The old one will be freed soon */
5644        compound = PageTransHuge(newpage);
5645        nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5646
5647        page_counter_charge(&memcg->memory, nr_pages);
5648        if (do_memsw_account())
5649                page_counter_charge(&memcg->memsw, nr_pages);
5650        css_get_many(&memcg->css, nr_pages);
5651
5652        commit_charge(newpage, memcg, false);
5653
5654        local_irq_save(flags);
5655        mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5656        memcg_check_events(memcg, newpage);
5657        local_irq_restore(flags);
5658}
5659
5660DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
5661EXPORT_SYMBOL(memcg_sockets_enabled_key);
5662
5663void mem_cgroup_sk_alloc(struct sock *sk)
5664{
5665        struct mem_cgroup *memcg;
5666
5667        if (!mem_cgroup_sockets_enabled)
5668                return;
5669
5670        /*
5671         * Socket cloning can throw us here with sk_memcg already
5672         * filled. It won't however, necessarily happen from
5673         * process context. So the test for root memcg given
5674         * the current task's memcg won't help us in this case.
5675         *
5676         * Respecting the original socket's memcg is a better
5677         * decision in this case.
5678         */
5679        if (sk->sk_memcg) {
5680                BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5681                css_get(&sk->sk_memcg->css);
5682                return;
5683        }
5684
5685        rcu_read_lock();
5686        memcg = mem_cgroup_from_task(current);
5687        if (memcg == root_mem_cgroup)
5688                goto out;
5689        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
5690                goto out;
5691        if (css_tryget_online(&memcg->css))
5692                sk->sk_memcg = memcg;
5693out:
5694        rcu_read_unlock();
5695}
5696
5697void mem_cgroup_sk_free(struct sock *sk)
5698{
5699        if (sk->sk_memcg)
5700                css_put(&sk->sk_memcg->css);
5701}
5702
5703/**
5704 * mem_cgroup_charge_skmem - charge socket memory
5705 * @memcg: memcg to charge
5706 * @nr_pages: number of pages to charge
5707 *
5708 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5709 * @memcg's configured limit, %false if the charge had to be forced.
5710 */
5711bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5712{
5713        gfp_t gfp_mask = GFP_KERNEL;
5714
5715        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
5716                struct page_counter *fail;
5717
5718                if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5719                        memcg->tcpmem_pressure = 0;
5720                        return true;
5721                }
5722                page_counter_charge(&memcg->tcpmem, nr_pages);
5723                memcg->tcpmem_pressure = 1;
5724                return false;
5725        }
5726
5727        /* Don't block in the packet receive path */
5728        if (in_softirq())
5729                gfp_mask = GFP_NOWAIT;
5730
5731        this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5732
5733        if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5734                return true;
5735
5736        try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
5737        return false;
5738}
5739
5740/**
5741 * mem_cgroup_uncharge_skmem - uncharge socket memory
5742 * @memcg - memcg to uncharge
5743 * @nr_pages - number of pages to uncharge
5744 */
5745void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5746{
5747        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
5748                page_counter_uncharge(&memcg->tcpmem, nr_pages);
5749                return;
5750        }
5751
5752        this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5753
5754        page_counter_uncharge(&memcg->memory, nr_pages);
5755        css_put_many(&memcg->css, nr_pages);
5756}
5757
5758static int __init cgroup_memory(char *s)
5759{
5760        char *token;
5761
5762        while ((token = strsep(&s, ",")) != NULL) {
5763                if (!*token)
5764                        continue;
5765                if (!strcmp(token, "nosocket"))
5766                        cgroup_memory_nosocket = true;
5767                if (!strcmp(token, "nokmem"))
5768                        cgroup_memory_nokmem = true;
5769        }
5770        return 0;
5771}
5772__setup("cgroup.memory=", cgroup_memory);
5773
5774/*
5775 * subsys_initcall() for memory controller.
5776 *
5777 * Some parts like hotcpu_notifier() have to be initialized from this context
5778 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5779 * everything that doesn't depend on a specific mem_cgroup structure should
5780 * be initialized from here.
5781 */
5782static int __init mem_cgroup_init(void)
5783{
5784        int cpu, node;
5785
5786        hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
5787
5788        for_each_possible_cpu(cpu)
5789                INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5790                          drain_local_stock);
5791
5792        for_each_node(node) {
5793                struct mem_cgroup_tree_per_node *rtpn;
5794
5795                rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5796                                    node_online(node) ? node : NUMA_NO_NODE);
5797
5798                rtpn->rb_root = RB_ROOT;
5799                spin_lock_init(&rtpn->lock);
5800                soft_limit_tree.rb_tree_per_node[node] = rtpn;
5801        }
5802
5803        return 0;
5804}
5805subsys_initcall(mem_cgroup_init);
5806
5807#ifdef CONFIG_MEMCG_SWAP
5808static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
5809{
5810        while (!atomic_inc_not_zero(&memcg->id.ref)) {
5811                /*
5812                 * The root cgroup cannot be destroyed, so it's refcount must
5813                 * always be >= 1.
5814                 */
5815                if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
5816                        VM_BUG_ON(1);
5817                        break;
5818                }
5819                memcg = parent_mem_cgroup(memcg);
5820                if (!memcg)
5821                        memcg = root_mem_cgroup;
5822        }
5823        return memcg;
5824}
5825
5826/**
5827 * mem_cgroup_swapout - transfer a memsw charge to swap
5828 * @page: page whose memsw charge to transfer
5829 * @entry: swap entry to move the charge to
5830 *
5831 * Transfer the memsw charge of @page to @entry.
5832 */
5833void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5834{
5835        struct mem_cgroup *memcg, *swap_memcg;
5836        unsigned short oldid;
5837
5838        VM_BUG_ON_PAGE(PageLRU(page), page);
5839        VM_BUG_ON_PAGE(page_count(page), page);
5840
5841        if (!do_memsw_account())
5842                return;
5843
5844        memcg = page->mem_cgroup;
5845
5846        /* Readahead page, never charged */
5847        if (!memcg)
5848                return;
5849
5850        /*
5851         * In case the memcg owning these pages has been offlined and doesn't
5852         * have an ID allocated to it anymore, charge the closest online
5853         * ancestor for the swap instead and transfer the memory+swap charge.
5854         */
5855        swap_memcg = mem_cgroup_id_get_online(memcg);
5856        oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg));
5857        VM_BUG_ON_PAGE(oldid, page);
5858        mem_cgroup_swap_statistics(swap_memcg, true);
5859
5860        page->mem_cgroup = NULL;
5861
5862        if (!mem_cgroup_is_root(memcg))
5863                page_counter_uncharge(&memcg->memory, 1);
5864
5865        if (memcg != swap_memcg) {
5866                if (!mem_cgroup_is_root(swap_memcg))
5867                        page_counter_charge(&swap_memcg->memsw, 1);
5868                page_counter_uncharge(&memcg->memsw, 1);
5869        }
5870
5871        /*
5872         * Interrupts should be disabled here because the caller holds the
5873         * mapping->tree_lock lock which is taken with interrupts-off. It is
5874         * important here to have the interrupts disabled because it is the
5875         * only synchronisation we have for udpating the per-CPU variables.
5876         */
5877        VM_BUG_ON(!irqs_disabled());
5878        mem_cgroup_charge_statistics(memcg, page, false, -1);
5879        memcg_check_events(memcg, page);
5880
5881        if (!mem_cgroup_is_root(memcg))
5882                css_put(&memcg->css);
5883}
5884
5885/*
5886 * mem_cgroup_try_charge_swap - try charging a swap entry
5887 * @page: page being added to swap
5888 * @entry: swap entry to charge
5889 *
5890 * Try to charge @entry to the memcg that @page belongs to.
5891 *
5892 * Returns 0 on success, -ENOMEM on failure.
5893 */
5894int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5895{
5896        struct mem_cgroup *memcg;
5897        struct page_counter *counter;
5898        unsigned short oldid;
5899
5900        if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5901                return 0;
5902
5903        memcg = page->mem_cgroup;
5904
5905        /* Readahead page, never charged */
5906        if (!memcg)
5907                return 0;
5908
5909        memcg = mem_cgroup_id_get_online(memcg);
5910
5911        if (!mem_cgroup_is_root(memcg) &&
5912            !page_counter_try_charge(&memcg->swap, 1, &counter)) {
5913                mem_cgroup_id_put(memcg);
5914                return -ENOMEM;
5915        }
5916
5917        oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5918        VM_BUG_ON_PAGE(oldid, page);
5919        mem_cgroup_swap_statistics(memcg, true);
5920
5921        return 0;
5922}
5923
5924/**
5925 * mem_cgroup_uncharge_swap - uncharge a swap entry
5926 * @entry: swap entry to uncharge
5927 *
5928 * Drop the swap charge associated with @entry.
5929 */
5930void mem_cgroup_uncharge_swap(swp_entry_t entry)
5931{
5932        struct mem_cgroup *memcg;
5933        unsigned short id;
5934
5935        if (!do_swap_account)
5936                return;
5937
5938        id = swap_cgroup_record(entry, 0);
5939        rcu_read_lock();
5940        memcg = mem_cgroup_from_id(id);
5941        if (memcg) {
5942                if (!mem_cgroup_is_root(memcg)) {
5943                        if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5944                                page_counter_uncharge(&memcg->swap, 1);
5945                        else
5946                                page_counter_uncharge(&memcg->memsw, 1);
5947                }
5948                mem_cgroup_swap_statistics(memcg, false);
5949                mem_cgroup_id_put(memcg);
5950        }
5951        rcu_read_unlock();
5952}
5953
5954long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
5955{
5956        long nr_swap_pages = get_nr_swap_pages();
5957
5958        if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5959                return nr_swap_pages;
5960        for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5961                nr_swap_pages = min_t(long, nr_swap_pages,
5962                                      READ_ONCE(memcg->swap.limit) -
5963                                      page_counter_read(&memcg->swap));
5964        return nr_swap_pages;
5965}
5966
5967bool mem_cgroup_swap_full(struct page *page)
5968{
5969        struct mem_cgroup *memcg;
5970
5971        VM_BUG_ON_PAGE(!PageLocked(page), page);
5972
5973        if (vm_swap_full())
5974                return true;
5975        if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
5976                return false;
5977
5978        memcg = page->mem_cgroup;
5979        if (!memcg)
5980                return false;
5981
5982        for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
5983                if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
5984                        return true;
5985
5986        return false;
5987}
5988
5989/* for remember boot option*/
5990#ifdef CONFIG_MEMCG_SWAP_ENABLED
5991static int really_do_swap_account __initdata = 1;
5992#else
5993static int really_do_swap_account __initdata;
5994#endif
5995
5996static int __init enable_swap_account(char *s)
5997{
5998        if (!strcmp(s, "1"))
5999                really_do_swap_account = 1;
6000        else if (!strcmp(s, "0"))
6001                really_do_swap_account = 0;
6002        return 1;
6003}
6004__setup("swapaccount=", enable_swap_account);
6005
6006static u64 swap_current_read(struct cgroup_subsys_state *css,
6007                             struct cftype *cft)
6008{
6009        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6010
6011        return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
6012}
6013
6014static int swap_max_show(struct seq_file *m, void *v)
6015{
6016        struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
6017        unsigned long max = READ_ONC