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