include/linux/memcontrol.h

   1 /* memcontrol.h - 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  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  */
  19
  20 #ifndef _LINUX_MEMCONTROL_H
  21 #define _LINUX_MEMCONTROL_H
  22 #include <linux/cgroup.h>
  23 #include <linux/vm_event_item.h>
  24 #include <linux/hardirq.h>
  25 #include <linux/jump_label.h>
  26 #include <linux/page_counter.h>
  27 #include <linux/vmpressure.h>
  28 #include <linux/eventfd.h>
  29 #include <linux/mmzone.h>
  30 #include <linux/writeback.h>
  31
  32 struct mem_cgroup;
  33 struct page;
  34 struct mm_struct;
  35 struct kmem_cache;
  36
  37 /*
  38  * The corresponding mem_cgroup_stat_names is defined in mm/memcontrol.c,
  39  * These two lists should keep in accord with each other.
  40  */
  41 enum mem_cgroup_stat_index {
  42         /*
  43          * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
  44          */
  45         MEM_CGROUP_STAT_CACHE,          /* # of pages charged as cache */
  46         MEM_CGROUP_STAT_RSS,            /* # of pages charged as anon rss */
  47         MEM_CGROUP_STAT_RSS_HUGE,       /* # of pages charged as anon huge */
  48         MEM_CGROUP_STAT_FILE_MAPPED,    /* # of pages charged as file rss */
  49         MEM_CGROUP_STAT_DIRTY,          /* # of dirty pages in page cache */
  50         MEM_CGROUP_STAT_WRITEBACK,      /* # of pages under writeback */
  51         MEM_CGROUP_STAT_SWAP,           /* # of pages, swapped out */
  52         MEM_CGROUP_STAT_NSTATS,
  53 };
  54
  55 struct mem_cgroup_reclaim_cookie {
  56         struct zone *zone;
  57         int priority;
  58         unsigned int generation;
  59 };
  60
  61 enum mem_cgroup_events_index {
  62         MEM_CGROUP_EVENTS_PGPGIN,       /* # of pages paged in */
  63         MEM_CGROUP_EVENTS_PGPGOUT,      /* # of pages paged out */
  64         MEM_CGROUP_EVENTS_PGFAULT,      /* # of page-faults */
  65         MEM_CGROUP_EVENTS_PGMAJFAULT,   /* # of major page-faults */
  66         MEM_CGROUP_EVENTS_NSTATS,
  67         /* default hierarchy events */
  68         MEMCG_LOW = MEM_CGROUP_EVENTS_NSTATS,
  69         MEMCG_HIGH,
  70         MEMCG_MAX,
  71         MEMCG_OOM,
  72         MEMCG_NR_EVENTS,
  73 };
  74
  75 /*
  76  * Per memcg event counter is incremented at every pagein/pageout. With THP,
  77  * it will be incremated by the number of pages. This counter is used for
  78  * for trigger some periodic events. This is straightforward and better
  79  * than using jiffies etc. to handle periodic memcg event.
  80  */
  81 enum mem_cgroup_events_target {
  82         MEM_CGROUP_TARGET_THRESH,
  83         MEM_CGROUP_TARGET_SOFTLIMIT,
  84         MEM_CGROUP_TARGET_NUMAINFO,
  85         MEM_CGROUP_NTARGETS,
  86 };
  87
  88 /*
  89  * Bits in struct cg_proto.flags
  90  */
  91 enum cg_proto_flags {
  92         /* Currently active and new sockets should be assigned to cgroups */
  93         MEMCG_SOCK_ACTIVE,
  94         /* It was ever activated; we must disarm static keys on destruction */
  95         MEMCG_SOCK_ACTIVATED,
  96 };
  97
  98 struct cg_proto {
  99         struct page_counter     memory_allocated;       /* Current allocated memory. */
 100         struct percpu_counter   sockets_allocated;      /* Current number of sockets. */
 101         int                     memory_pressure;
 102         long                    sysctl_mem[3];
 103         unsigned long           flags;
 104         /*
 105          * memcg field is used to find which memcg we belong directly
 106          * Each memcg struct can hold more than one cg_proto, so container_of
 107          * won't really cut.
 108          *
 109          * The elegant solution would be having an inverse function to
 110          * proto_cgroup in struct proto, but that means polluting the structure
 111          * for everybody, instead of just for memcg users.
 112          */
 113         struct mem_cgroup       *memcg;
 114 };
 115
 116 #ifdef CONFIG_MEMCG
 117 struct mem_cgroup_stat_cpu {
 118         long count[MEM_CGROUP_STAT_NSTATS];
 119         unsigned long events[MEMCG_NR_EVENTS];
 120         unsigned long nr_page_events;
 121         unsigned long targets[MEM_CGROUP_NTARGETS];
 122 };
 123
 124 struct mem_cgroup_reclaim_iter {
 125         struct mem_cgroup *position;
 126         /* scan generation, increased every round-trip */
 127         unsigned int generation;
 128 };
 129
 130 /*
 131  * per-zone information in memory controller.
 132  */
 133 struct mem_cgroup_per_zone {
 134         struct lruvec           lruvec;
 135         unsigned long           lru_size[NR_LRU_LISTS];
 136
 137         struct mem_cgroup_reclaim_iter  iter[DEF_PRIORITY + 1];
 138
 139         struct rb_node          tree_node;      /* RB tree node */
 140         unsigned long           usage_in_excess;/* Set to the value by which */
 141                                                 /* the soft limit is exceeded*/
 142         bool                    on_tree;
 143         struct mem_cgroup       *memcg;         /* Back pointer, we cannot */
 144                                                 /* use container_of        */
 145 };
 146
 147 struct mem_cgroup_per_node {
 148         struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
 149 };
 150
 151 struct mem_cgroup_threshold {
 152         struct eventfd_ctx *eventfd;
 153         unsigned long threshold;
 154 };
 155
 156 /* For threshold */
 157 struct mem_cgroup_threshold_ary {
 158         /* An array index points to threshold just below or equal to usage. */
 159         int current_threshold;
 160         /* Size of entries[] */
 161         unsigned int size;
 162         /* Array of thresholds */
 163         struct mem_cgroup_threshold entries[0];
 164 };
 165
 166 struct mem_cgroup_thresholds {
 167         /* Primary thresholds array */
 168         struct mem_cgroup_threshold_ary *primary;
 169         /*
 170          * Spare threshold array.
 171          * This is needed to make mem_cgroup_unregister_event() "never fail".
 172          * It must be able to store at least primary->size - 1 entries.
 173          */
 174         struct mem_cgroup_threshold_ary *spare;
 175 };
 176
 177 /*
 178  * The memory controller data structure. The memory controller controls both
 179  * page cache and RSS per cgroup. We would eventually like to provide
 180  * statistics based on the statistics developed by Rik Van Riel for clock-pro,
 181  * to help the administrator determine what knobs to tune.
 182  */
 183 struct mem_cgroup {
 184         struct cgroup_subsys_state css;
 185
 186         /* Accounted resources */
 187         struct page_counter memory;
 188         struct page_counter memsw;
 189         struct page_counter kmem;
 190
 191         /* Normal memory consumption range */
 192         unsigned long low;
 193         unsigned long high;
 194
 195         unsigned long soft_limit;
 196
 197         /* vmpressure notifications */
 198         struct vmpressure vmpressure;
 199
 200         /* css_online() has been completed */
 201         int initialized;
 202
 203         /*
 204          * Should the accounting and control be hierarchical, per subtree?
 205          */
 206         bool use_hierarchy;
 207
 208         /* protected by memcg_oom_lock */
 209         bool            oom_lock;
 210         int             under_oom;
 211
 212         int     swappiness;
 213         /* OOM-Killer disable */
 214         int             oom_kill_disable;
 215
 216         /* protect arrays of thresholds */
 217         struct mutex thresholds_lock;
 218
 219         /* thresholds for memory usage. RCU-protected */
 220         struct mem_cgroup_thresholds thresholds;
 221
 222         /* thresholds for mem+swap usage. RCU-protected */
 223         struct mem_cgroup_thresholds memsw_thresholds;
 224
 225         /* For oom notifier event fd */
 226         struct list_head oom_notify;
 227
 228         /*
 229          * Should we move charges of a task when a task is moved into this
 230          * mem_cgroup ? And what type of charges should we move ?
 231          */
 232         unsigned long move_charge_at_immigrate;
 233         /*
 234          * set > 0 if pages under this cgroup are moving to other cgroup.
 235          */
 236         atomic_t                moving_account;
 237         /* taken only while moving_account > 0 */
 238         spinlock_t              move_lock;
 239         struct task_struct      *move_lock_task;
 240         unsigned long           move_lock_flags;
 241         /*
 242          * percpu counter.
 243          */
 244         struct mem_cgroup_stat_cpu __percpu *stat;
 245         spinlock_t pcp_counter_lock;
 246
 247 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
 248         struct cg_proto tcp_mem;
 249 #endif
 250 #if defined(CONFIG_MEMCG_KMEM)
 251         /* Index in the kmem_cache->memcg_params.memcg_caches array */
 252         int kmemcg_id;
 253         bool kmem_acct_activated;
 254         bool kmem_acct_active;
 255 #endif
 256
 257         int last_scanned_node;
 258 #if MAX_NUMNODES > 1
 259         nodemask_t      scan_nodes;
 260         atomic_t        numainfo_events;
 261         atomic_t        numainfo_updating;
 262 #endif
 263
 264 #ifdef CONFIG_CGROUP_WRITEBACK
 265         struct list_head cgwb_list;
 266         struct wb_domain cgwb_domain;
 267 #endif
 268
 269         /* List of events which userspace want to receive */
 270         struct list_head event_list;
 271         spinlock_t event_list_lock;
 272
 273         struct mem_cgroup_per_node *nodeinfo[0];
 274         /* WARNING: nodeinfo must be the last member here */
 275 };
 276 extern struct cgroup_subsys_state *mem_cgroup_root_css;
 277
 278 /**
 279  * mem_cgroup_events - count memory events against a cgroup
 280  * @memcg: the memory cgroup
 281  * @idx: the event index
 282  * @nr: the number of events to account for
 283  */
 284 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
 285                        enum mem_cgroup_events_index idx,
 286                        unsigned int nr)
 287 {
 288         this_cpu_add(memcg->stat->events[idx], nr);
 289 }
 290
 291 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
 292
 293 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
 294                           gfp_t gfp_mask, struct mem_cgroup **memcgp);
 295 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
 296                               bool lrucare);
 297 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg);
 298 void mem_cgroup_uncharge(struct page *page);
 299 void mem_cgroup_uncharge_list(struct list_head *page_list);
 300
 301 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage,
 302                         bool lrucare);
 303
 304 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
 305 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
 306
 307 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
 308 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
 309 struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
 310
 311 static inline
 312 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
 313         return css ? container_of(css, struct mem_cgroup, css) : NULL;
 314 }
 315
 316 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
 317                                    struct mem_cgroup *,
 318                                    struct mem_cgroup_reclaim_cookie *);
 319 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
 320
 321 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
 322                               struct mem_cgroup *root)
 323 {
 324         if (root == memcg)
 325                 return true;
 326         if (!root->use_hierarchy)
 327                 return false;
 328         return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
 329 }
 330
 331 static inline bool mm_match_cgroup(struct mm_struct *mm,
 332                                    struct mem_cgroup *memcg)
 333 {
 334         struct mem_cgroup *task_memcg;
 335         bool match = false;
 336
 337         rcu_read_lock();
 338         task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 339         if (task_memcg)
 340                 match = mem_cgroup_is_descendant(task_memcg, memcg);
 341         rcu_read_unlock();
 342         return match;
 343 }
 344
 345 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
 346 ino_t page_cgroup_ino(struct page *page);
 347
 348 static inline bool mem_cgroup_disabled(void)
 349 {
 350         if (memory_cgrp_subsys.disabled)
 351                 return true;
 352         return false;
 353 }
 354
 355 /*
 356  * For memory reclaim.
 357  */
 358 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
 359
 360 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
 361                 int nr_pages);
 362
 363 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
 364 {
 365         struct mem_cgroup_per_zone *mz;
 366         struct mem_cgroup *memcg;
 367
 368         if (mem_cgroup_disabled())
 369                 return true;
 370
 371         mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
 372         memcg = mz->memcg;
 373
 374         return !!(memcg->css.flags & CSS_ONLINE);
 375 }
 376
 377 static inline
 378 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
 379 {
 380         struct mem_cgroup_per_zone *mz;
 381
 382         mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
 383         return mz->lru_size[lru];
 384 }
 385
 386 static inline int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
 387 {
 388         unsigned long inactive_ratio;
 389         unsigned long inactive;
 390         unsigned long active;
 391         unsigned long gb;
 392
 393         inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
 394         active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
 395
 396         gb = (inactive + active) >> (30 - PAGE_SHIFT);
 397         if (gb)
 398                 inactive_ratio = int_sqrt(10 * gb);
 399         else
 400                 inactive_ratio = 1;
 401
 402         return inactive * inactive_ratio < active;
 403 }
 404
 405 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
 406                                 struct task_struct *p);
 407
 408 static inline void mem_cgroup_oom_enable(void)
 409 {
 410         WARN_ON(current->memcg_oom.may_oom);
 411         current->memcg_oom.may_oom = 1;
 412 }
 413
 414 static inline void mem_cgroup_oom_disable(void)
 415 {
 416         WARN_ON(!current->memcg_oom.may_oom);
 417         current->memcg_oom.may_oom = 0;
 418 }
 419
 420 static inline bool task_in_memcg_oom(struct task_struct *p)
 421 {
 422         return p->memcg_oom.memcg;
 423 }
 424
 425 bool mem_cgroup_oom_synchronize(bool wait);
 426
 427 #ifdef CONFIG_MEMCG_SWAP
 428 extern int do_swap_account;
 429 #endif
 430
 431 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
 432 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
 433
 434 /**
 435  * mem_cgroup_update_page_stat - update page state statistics
 436  * @memcg: memcg to account against
 437  * @idx: page state item to account
 438  * @val: number of pages (positive or negative)
 439  *
 440  * See mem_cgroup_begin_page_stat() for locking requirements.
 441  */
 442 static inline void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
 443                                  enum mem_cgroup_stat_index idx, int val)
 444 {
 445         VM_BUG_ON(!rcu_read_lock_held());
 446
 447         if (memcg)
 448                 this_cpu_add(memcg->stat->count[idx], val);
 449 }
 450
 451 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
 452                                             enum mem_cgroup_stat_index idx)
 453 {
 454         mem_cgroup_update_page_stat(memcg, idx, 1);
 455 }
 456
 457 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
 458                                             enum mem_cgroup_stat_index idx)
 459 {
 460         mem_cgroup_update_page_stat(memcg, idx, -1);
 461 }
 462
 463 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
 464                                                 gfp_t gfp_mask,
 465                                                 unsigned long *total_scanned);
 466
 467 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
 468                                              enum vm_event_item idx)
 469 {
 470         struct mem_cgroup *memcg;
 471
 472         if (mem_cgroup_disabled())
 473                 return;
 474
 475         rcu_read_lock();
 476         memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 477         if (unlikely(!memcg))
 478                 goto out;
 479
 480         switch (idx) {
 481         case PGFAULT:
 482                 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
 483                 break;
 484         case PGMAJFAULT:
 485                 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
 486                 break;
 487         default:
 488                 BUG();
 489         }
 490 out:
 491         rcu_read_unlock();
 492 }
 493 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 494 void mem_cgroup_split_huge_fixup(struct page *head);
 495 #endif
 496
 497 #else /* CONFIG_MEMCG */
 498 struct mem_cgroup;
 499
 500 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
 501                                      enum mem_cgroup_events_index idx,
 502                                      unsigned int nr)
 503 {
 504 }
 505
 506 static inline bool mem_cgroup_low(struct mem_cgroup *root,
 507                                   struct mem_cgroup *memcg)
 508 {
 509         return false;
 510 }
 511
 512 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
 513                                         gfp_t gfp_mask,
 514                                         struct mem_cgroup **memcgp)
 515 {
 516         *memcgp = NULL;
 517         return 0;
 518 }
 519
 520 static inline void mem_cgroup_commit_charge(struct page *page,
 521                                             struct mem_cgroup *memcg,
 522                                             bool lrucare)
 523 {
 524 }
 525
 526 static inline void mem_cgroup_cancel_charge(struct page *page,
 527                                             struct mem_cgroup *memcg)
 528 {
 529 }
 530
 531 static inline void mem_cgroup_uncharge(struct page *page)
 532 {
 533 }
 534
 535 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
 536 {
 537 }
 538
 539 static inline void mem_cgroup_migrate(struct page *oldpage,
 540                                       struct page *newpage,
 541                                       bool lrucare)
 542 {
 543 }
 544
 545 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
 546                                                     struct mem_cgroup *memcg)
 547 {
 548         return &zone->lruvec;
 549 }
 550
 551 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
 552                                                     struct zone *zone)
 553 {
 554         return &zone->lruvec;
 555 }
 556
 557 static inline bool mm_match_cgroup(struct mm_struct *mm,
 558                 struct mem_cgroup *memcg)
 559 {
 560         return true;
 561 }
 562
 563 static inline bool task_in_mem_cgroup(struct task_struct *task,
 564                                       const struct mem_cgroup *memcg)
 565 {
 566         return true;
 567 }
 568
 569 static inline struct mem_cgroup *
 570 mem_cgroup_iter(struct mem_cgroup *root,
 571                 struct mem_cgroup *prev,
 572                 struct mem_cgroup_reclaim_cookie *reclaim)
 573 {
 574         return NULL;
 575 }
 576
 577 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
 578                                          struct mem_cgroup *prev)
 579 {
 580 }
 581
 582 static inline bool mem_cgroup_disabled(void)
 583 {
 584         return true;
 585 }
 586
 587 static inline int
 588 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
 589 {
 590         return 1;
 591 }
 592
 593 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
 594 {
 595         return true;
 596 }
 597
 598 static inline unsigned long
 599 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
 600 {
 601         return 0;
 602 }
 603
 604 static inline void
 605 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
 606                               int increment)
 607 {
 608 }
 609
 610 static inline void
 611 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
 612 {
 613 }
 614
 615 static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
 616 {
 617         return NULL;
 618 }
 619
 620 static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
 621 {
 622 }
 623
 624 static inline void mem_cgroup_oom_enable(void)
 625 {
 626 }
 627
 628 static inline void mem_cgroup_oom_disable(void)
 629 {
 630 }
 631
 632 static inline bool task_in_memcg_oom(struct task_struct *p)
 633 {
 634         return false;
 635 }
 636
 637 static inline bool mem_cgroup_oom_synchronize(bool wait)
 638 {
 639         return false;
 640 }
 641
 642 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
 643                                             enum mem_cgroup_stat_index idx)
 644 {
 645 }
 646
 647 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
 648                                             enum mem_cgroup_stat_index idx)
 649 {
 650 }
 651
 652 static inline
 653 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
 654                                             gfp_t gfp_mask,
 655                                             unsigned long *total_scanned)
 656 {
 657         return 0;
 658 }
 659
 660 static inline void mem_cgroup_split_huge_fixup(struct page *head)
 661 {
 662 }
 663
 664 static inline
 665 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
 666 {
 667 }
 668 #endif /* CONFIG_MEMCG */
 669
 670 enum {
 671         UNDER_LIMIT,
 672         SOFT_LIMIT,
 673         OVER_LIMIT,
 674 };
 675
 676 #ifdef CONFIG_CGROUP_WRITEBACK
 677
 678 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
 679 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
 680 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pavail,
 681                          unsigned long *pdirty, unsigned long *pwriteback);
 682
 683 #else   /* CONFIG_CGROUP_WRITEBACK */
 684
 685 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
 686 {
 687         return NULL;
 688 }
 689
 690 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
 691                                        unsigned long *pavail,
 692                                        unsigned long *pdirty,
 693                                        unsigned long *pwriteback)
 694 {
 695 }
 696
 697 #endif  /* CONFIG_CGROUP_WRITEBACK */
 698
 699 struct sock;
 700 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
 701 void sock_update_memcg(struct sock *sk);
 702 void sock_release_memcg(struct sock *sk);
 703 #else
 704 static inline void sock_update_memcg(struct sock *sk)
 705 {
 706 }
 707 static inline void sock_release_memcg(struct sock *sk)
 708 {
 709 }
 710 #endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
 711
 712 #ifdef CONFIG_MEMCG_KMEM
 713 extern struct static_key memcg_kmem_enabled_key;
 714
 715 extern int memcg_nr_cache_ids;
 716 void memcg_get_cache_ids(void);
 717 void memcg_put_cache_ids(void);
 718
 719 /*
 720  * Helper macro to loop through all memcg-specific caches. Callers must still
 721  * check if the cache is valid (it is either valid or NULL).
 722  * the slab_mutex must be held when looping through those caches
 723  */
 724 #define for_each_memcg_cache_index(_idx)        \
 725         for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
 726
 727 static inline bool memcg_kmem_enabled(void)
 728 {
 729         return static_key_false(&memcg_kmem_enabled_key);
 730 }
 731
 732 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
 733 {
 734         return memcg->kmem_acct_active;
 735 }
 736
 737 /*
 738  * In general, we'll do everything in our power to not incur in any overhead
 739  * for non-memcg users for the kmem functions. Not even a function call, if we
 740  * can avoid it.
 741  *
 742  * Therefore, we'll inline all those functions so that in the best case, we'll
 743  * see that kmemcg is off for everybody and proceed quickly.  If it is on,
 744  * we'll still do most of the flag checking inline. We check a lot of
 745  * conditions, but because they are pretty simple, they are expected to be
 746  * fast.
 747  */
 748 bool __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg,
 749                                         int order);
 750 void __memcg_kmem_commit_charge(struct page *page,
 751                                        struct mem_cgroup *memcg, int order);
 752 void __memcg_kmem_uncharge_pages(struct page *page, int order);
 753
 754 /*
 755  * helper for acessing a memcg's index. It will be used as an index in the
 756  * child cache array in kmem_cache, and also to derive its name. This function
 757  * will return -1 when this is not a kmem-limited memcg.
 758  */
 759 static inline int memcg_cache_id(struct mem_cgroup *memcg)
 760 {
 761         return memcg ? memcg->kmemcg_id : -1;
 762 }
 763
 764 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep);
 765 void __memcg_kmem_put_cache(struct kmem_cache *cachep);
 766
 767 struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr);
 768
 769 int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp,
 770                       unsigned long nr_pages);
 771 void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages);
 772
 773 /**
 774  * memcg_kmem_newpage_charge: verify if a new kmem allocation is allowed.
 775  * @gfp: the gfp allocation flags.
 776  * @memcg: a pointer to the memcg this was charged against.
 777  * @order: allocation order.
 778  *
 779  * returns true if the memcg where the current task belongs can hold this
 780  * allocation.
 781  *
 782  * We return true automatically if this allocation is not to be accounted to
 783  * any memcg.
 784  */
 785 static inline bool
 786 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
 787 {
 788         if (!memcg_kmem_enabled())
 789                 return true;
 790
 791         if (gfp & __GFP_NOACCOUNT)
 792                 return true;
 793         /*
 794          * __GFP_NOFAIL allocations will move on even if charging is not
 795          * possible. Therefore we don't even try, and have this allocation
 796          * unaccounted. We could in theory charge it forcibly, but we hope
 797          * those allocations are rare, and won't be worth the trouble.
 798          */
 799         if (gfp & __GFP_NOFAIL)
 800                 return true;
 801         if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
 802                 return true;
 803
 804         /* If the test is dying, just let it go. */
 805         if (unlikely(fatal_signal_pending(current)))
 806                 return true;
 807
 808         return __memcg_kmem_newpage_charge(gfp, memcg, order);
 809 }
 810
 811 /**
 812  * memcg_kmem_uncharge_pages: uncharge pages from memcg
 813  * @page: pointer to struct page being freed
 814  * @order: allocation order.
 815  */
 816 static inline void
 817 memcg_kmem_uncharge_pages(struct page *page, int order)
 818 {
 819         if (memcg_kmem_enabled())
 820                 __memcg_kmem_uncharge_pages(page, order);
 821 }
 822
 823 /**
 824  * memcg_kmem_commit_charge: embeds correct memcg in a page
 825  * @page: pointer to struct page recently allocated
 826  * @memcg: the memcg structure we charged against
 827  * @order: allocation order.
 828  *
 829  * Needs to be called after memcg_kmem_newpage_charge, regardless of success or
 830  * failure of the allocation. if @page is NULL, this function will revert the
 831  * charges. Otherwise, it will commit @page to @memcg.
 832  */
 833 static inline void
 834 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
 835 {
 836         if (memcg_kmem_enabled() && memcg)
 837                 __memcg_kmem_commit_charge(page, memcg, order);
 838 }
 839
 840 /**
 841  * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
 842  * @cachep: the original global kmem cache
 843  * @gfp: allocation flags.
 844  *
 845  * All memory allocated from a per-memcg cache is charged to the owner memcg.
 846  */
 847 static __always_inline struct kmem_cache *
 848 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
 849 {
 850         if (!memcg_kmem_enabled())
 851                 return cachep;
 852         if (gfp & __GFP_NOACCOUNT)
 853                 return cachep;
 854         if (gfp & __GFP_NOFAIL)
 855                 return cachep;
 856         if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
 857                 return cachep;
 858         if (unlikely(fatal_signal_pending(current)))
 859                 return cachep;
 860
 861         return __memcg_kmem_get_cache(cachep);
 862 }
 863
 864 static __always_inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
 865 {
 866         if (memcg_kmem_enabled())
 867                 __memcg_kmem_put_cache(cachep);
 868 }
 869
 870 static __always_inline struct mem_cgroup *mem_cgroup_from_kmem(void *ptr)
 871 {
 872         if (!memcg_kmem_enabled())
 873                 return NULL;
 874         return __mem_cgroup_from_kmem(ptr);
 875 }
 876 #else
 877 #define for_each_memcg_cache_index(_idx)        \
 878         for (; NULL; )
 879
 880 static inline bool memcg_kmem_enabled(void)
 881 {
 882         return false;
 883 }
 884
 885 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
 886 {
 887         return false;
 888 }
 889
 890 static inline bool
 891 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
 892 {
 893         return true;
 894 }
 895
 896 static inline void memcg_kmem_uncharge_pages(struct page *page, int order)
 897 {
 898 }
 899
 900 static inline void
 901 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
 902 {
 903 }
 904
 905 static inline int memcg_cache_id(struct mem_cgroup *memcg)
 906 {
 907         return -1;
 908 }
 909
 910 static inline void memcg_get_cache_ids(void)
 911 {
 912 }
 913
 914 static inline void memcg_put_cache_ids(void)
 915 {
 916 }
 917
 918 static inline struct kmem_cache *
 919 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
 920 {
 921         return cachep;
 922 }
 923
 924 static inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
 925 {
 926 }
 927
 928 static inline struct mem_cgroup *mem_cgroup_from_kmem(void *ptr)
 929 {
 930         return NULL;
 931 }
 932 #endif /* CONFIG_MEMCG_KMEM */
 933 #endif /* _LINUX_MEMCONTROL_H */
 934