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
 246 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
 247         struct cg_proto tcp_mem;
 248 #endif
 249 #if defined(CONFIG_MEMCG_KMEM)
 250         /* Index in the kmem_cache->memcg_params.memcg_caches array */
 251         int kmemcg_id;
 252         bool kmem_acct_activated;
 253         bool kmem_acct_active;
 254 #endif
 255
 256         int last_scanned_node;
 257 #if MAX_NUMNODES > 1
 258         nodemask_t      scan_nodes;
 259         atomic_t        numainfo_events;
 260         atomic_t        numainfo_updating;
 261 #endif
 262
 263 #ifdef CONFIG_CGROUP_WRITEBACK
 264         struct list_head cgwb_list;
 265         struct wb_domain cgwb_domain;
 266 #endif
 267
 268         /* List of events which userspace want to receive */
 269         struct list_head event_list;
 270         spinlock_t event_list_lock;
 271
 272         struct mem_cgroup_per_node *nodeinfo[0];
 273         /* WARNING: nodeinfo must be the last member here */
 274 };
 275 extern struct cgroup_subsys_state *mem_cgroup_root_css;
 276
 277 /**
 278  * mem_cgroup_events - count memory events against a cgroup
 279  * @memcg: the memory cgroup
 280  * @idx: the event index
 281  * @nr: the number of events to account for
 282  */
 283 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
 284                        enum mem_cgroup_events_index idx,
 285                        unsigned int nr)
 286 {
 287         this_cpu_add(memcg->stat->events[idx], nr);
 288 }
 289
 290 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
 291
 292 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
 293                           gfp_t gfp_mask, struct mem_cgroup **memcgp);
 294 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
 295                               bool lrucare);
 296 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg);
 297 void mem_cgroup_uncharge(struct page *page);
 298 void mem_cgroup_uncharge_list(struct list_head *page_list);
 299
 300 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage,
 301                         bool lrucare);
 302
 303 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
 304 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
 305
 306 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
 307 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
 308 struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
 309
 310 static inline
 311 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
 312         return css ? container_of(css, struct mem_cgroup, css) : NULL;
 313 }
 314
 315 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
 316                                    struct mem_cgroup *,
 317                                    struct mem_cgroup_reclaim_cookie *);
 318 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
 319
 320 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
 321                               struct mem_cgroup *root)
 322 {
 323         if (root == memcg)
 324                 return true;
 325         if (!root->use_hierarchy)
 326                 return false;
 327         return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
 328 }
 329
 330 static inline bool mm_match_cgroup(struct mm_struct *mm,
 331                                    struct mem_cgroup *memcg)
 332 {
 333         struct mem_cgroup *task_memcg;
 334         bool match = false;
 335
 336         rcu_read_lock();
 337         task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 338         if (task_memcg)
 339                 match = mem_cgroup_is_descendant(task_memcg, memcg);
 340         rcu_read_unlock();
 341         return match;
 342 }
 343
 344 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
 345 ino_t page_cgroup_ino(struct page *page);
 346
 347 static inline bool mem_cgroup_disabled(void)
 348 {
 349         if (memory_cgrp_subsys.disabled)
 350                 return true;
 351         return false;
 352 }
 353
 354 /*
 355  * For memory reclaim.
 356  */
 357 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
 358
 359 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
 360                 int nr_pages);
 361
 362 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
 363 {
 364         struct mem_cgroup_per_zone *mz;
 365         struct mem_cgroup *memcg;
 366
 367         if (mem_cgroup_disabled())
 368                 return true;
 369
 370         mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
 371         memcg = mz->memcg;
 372
 373         return !!(memcg->css.flags & CSS_ONLINE);
 374 }
 375
 376 static inline
 377 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
 378 {
 379         struct mem_cgroup_per_zone *mz;
 380
 381         mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
 382         return mz->lru_size[lru];
 383 }
 384
 385 static inline int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
 386 {
 387         unsigned long inactive_ratio;
 388         unsigned long inactive;
 389         unsigned long active;
 390         unsigned long gb;
 391
 392         inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
 393         active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
 394
 395         gb = (inactive + active) >> (30 - PAGE_SHIFT);
 396         if (gb)
 397                 inactive_ratio = int_sqrt(10 * gb);
 398         else
 399                 inactive_ratio = 1;
 400
 401         return inactive * inactive_ratio < active;
 402 }
 403
 404 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
 405                                 struct task_struct *p);
 406
 407 static inline void mem_cgroup_oom_enable(void)
 408 {
 409         WARN_ON(current->memcg_oom.may_oom);
 410         current->memcg_oom.may_oom = 1;
 411 }
 412
 413 static inline void mem_cgroup_oom_disable(void)
 414 {
 415         WARN_ON(!current->memcg_oom.may_oom);
 416         current->memcg_oom.may_oom = 0;
 417 }
 418
 419 static inline bool task_in_memcg_oom(struct task_struct *p)
 420 {
 421         return p->memcg_oom.memcg;
 422 }
 423
 424 bool mem_cgroup_oom_synchronize(bool wait);
 425
 426 #ifdef CONFIG_MEMCG_SWAP
 427 extern int do_swap_account;
 428 #endif
 429
 430 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
 431 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
 432
 433 /**
 434  * mem_cgroup_update_page_stat - update page state statistics
 435  * @memcg: memcg to account against
 436  * @idx: page state item to account
 437  * @val: number of pages (positive or negative)
 438  *
 439  * See mem_cgroup_begin_page_stat() for locking requirements.
 440  */
 441 static inline void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
 442                                  enum mem_cgroup_stat_index idx, int val)
 443 {
 444         VM_BUG_ON(!rcu_read_lock_held());
 445
 446         if (memcg)
 447                 this_cpu_add(memcg->stat->count[idx], val);
 448 }
 449
 450 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
 451                                             enum mem_cgroup_stat_index idx)
 452 {
 453         mem_cgroup_update_page_stat(memcg, idx, 1);
 454 }
 455
 456 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
 457                                             enum mem_cgroup_stat_index idx)
 458 {
 459         mem_cgroup_update_page_stat(memcg, idx, -1);
 460 }
 461
 462 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
 463                                                 gfp_t gfp_mask,
 464                                                 unsigned long *total_scanned);
 465
 466 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
 467                                              enum vm_event_item idx)
 468 {
 469         struct mem_cgroup *memcg;
 470
 471         if (mem_cgroup_disabled())
 472                 return;
 473
 474         rcu_read_lock();
 475         memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 476         if (unlikely(!memcg))
 477                 goto out;
 478
 479         switch (idx) {
 480         case PGFAULT:
 481                 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
 482                 break;
 483         case PGMAJFAULT:
 484                 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
 485                 break;
 486         default:
 487                 BUG();
 488         }
 489 out:
 490         rcu_read_unlock();
 491 }
 492 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 493 void mem_cgroup_split_huge_fixup(struct page *head);
 494 #endif
 495
 496 #else /* CONFIG_MEMCG */
 497 struct mem_cgroup;
 498
 499 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
 500                                      enum mem_cgroup_events_index idx,
 501                                      unsigned int nr)
 502 {
 503 }
 504
 505 static inline bool mem_cgroup_low(struct mem_cgroup *root,
 506                                   struct mem_cgroup *memcg)
 507 {
 508         return false;
 509 }
 510
 511 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
 512                                         gfp_t gfp_mask,
 513                                         struct mem_cgroup **memcgp)
 514 {
 515         *memcgp = NULL;
 516         return 0;
 517 }
 518
 519 static inline void mem_cgroup_commit_charge(struct page *page,
 520                                             struct mem_cgroup *memcg,
 521                                             bool lrucare)
 522 {
 523 }
 524
 525 static inline void mem_cgroup_cancel_charge(struct page *page,
 526                                             struct mem_cgroup *memcg)
 527 {
 528 }
 529
 530 static inline void mem_cgroup_uncharge(struct page *page)
 531 {
 532 }
 533
 534 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
 535 {
 536 }
 537
 538 static inline void mem_cgroup_migrate(struct page *oldpage,
 539                                       struct page *newpage,
 540                                       bool lrucare)
 541 {
 542 }
 543
 544 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
 545                                                     struct mem_cgroup *memcg)
 546 {
 547         return &zone->lruvec;
 548 }
 549
 550 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
 551                                                     struct zone *zone)
 552 {
 553         return &zone->lruvec;
 554 }
 555
 556 static inline bool mm_match_cgroup(struct mm_struct *mm,
 557                 struct mem_cgroup *memcg)
 558 {
 559         return true;
 560 }
 561
 562 static inline bool task_in_mem_cgroup(struct task_struct *task,
 563                                       const struct mem_cgroup *memcg)
 564 {
 565         return true;
 566 }
 567
 568 static inline struct mem_cgroup *
 569 mem_cgroup_iter(struct mem_cgroup *root,
 570                 struct mem_cgroup *prev,
 571                 struct mem_cgroup_reclaim_cookie *reclaim)
 572 {
 573         return NULL;
 574 }
 575
 576 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
 577                                          struct mem_cgroup *prev)
 578 {
 579 }
 580
 581 static inline bool mem_cgroup_disabled(void)
 582 {
 583         return true;
 584 }
 585
 586 static inline int
 587 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
 588 {
 589         return 1;
 590 }
 591
 592 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
 593 {
 594         return true;
 595 }
 596
 597 static inline unsigned long
 598 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
 599 {
 600         return 0;
 601 }
 602
 603 static inline void
 604 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
 605                               int increment)
 606 {
 607 }
 608
 609 static inline void
 610 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
 611 {
 612 }
 613
 614 static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
 615 {
 616         return NULL;
 617 }
 618
 619 static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
 620 {
 621 }
 622
 623 static inline void mem_cgroup_oom_enable(void)
 624 {
 625 }
 626
 627 static inline void mem_cgroup_oom_disable(void)
 628 {
 629 }
 630
 631 static inline bool task_in_memcg_oom(struct task_struct *p)
 632 {
 633         return false;
 634 }
 635
 636 static inline bool mem_cgroup_oom_synchronize(bool wait)
 637 {
 638         return false;
 639 }
 640
 641 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
 642                                             enum mem_cgroup_stat_index idx)
 643 {
 644 }
 645
 646 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
 647                                             enum mem_cgroup_stat_index idx)
 648 {
 649 }
 650
 651 static inline
 652 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
 653                                             gfp_t gfp_mask,
 654                                             unsigned long *total_scanned)
 655 {
 656         return 0;
 657 }
 658
 659 static inline void mem_cgroup_split_huge_fixup(struct page *head)
 660 {
 661 }
 662
 663 static inline
 664 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
 665 {
 666 }
 667 #endif /* CONFIG_MEMCG */
 668
 669 enum {
 670         UNDER_LIMIT,
 671         SOFT_LIMIT,
 672         OVER_LIMIT,
 673 };
 674
 675 #ifdef CONFIG_CGROUP_WRITEBACK
 676
 677 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
 678 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
 679 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pavail,
 680                          unsigned long *pdirty, unsigned long *pwriteback);
 681
 682 #else   /* CONFIG_CGROUP_WRITEBACK */
 683
 684 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
 685 {
 686         return NULL;
 687 }
 688
 689 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
 690                                        unsigned long *pavail,
 691                                        unsigned long *pdirty,
 692                                        unsigned long *pwriteback)
 693 {
 694 }
 695
 696 #endif  /* CONFIG_CGROUP_WRITEBACK */
 697
 698 struct sock;
 699 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
 700 void sock_update_memcg(struct sock *sk);
 701 void sock_release_memcg(struct sock *sk);
 702 #else
 703 static inline void sock_update_memcg(struct sock *sk)
 704 {
 705 }
 706 static inline void sock_release_memcg(struct sock *sk)
 707 {
 708 }
 709 #endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
 710
 711 #ifdef CONFIG_MEMCG_KMEM
 712 extern struct static_key memcg_kmem_enabled_key;
 713
 714 extern int memcg_nr_cache_ids;
 715 void memcg_get_cache_ids(void);
 716 void memcg_put_cache_ids(void);
 717
 718 /*
 719  * Helper macro to loop through all memcg-specific caches. Callers must still
 720  * check if the cache is valid (it is either valid or NULL).
 721  * the slab_mutex must be held when looping through those caches
 722  */
 723 #define for_each_memcg_cache_index(_idx)        \
 724         for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
 725
 726 static inline bool memcg_kmem_enabled(void)
 727 {
 728         return static_key_false(&memcg_kmem_enabled_key);
 729 }
 730
 731 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
 732 {
 733         return memcg->kmem_acct_active;
 734 }
 735
 736 /*
 737  * In general, we'll do everything in our power to not incur in any overhead
 738  * for non-memcg users for the kmem functions. Not even a function call, if we
 739  * can avoid it.
 740  *
 741  * Therefore, we'll inline all those functions so that in the best case, we'll
 742  * see that kmemcg is off for everybody and proceed quickly.  If it is on,
 743  * we'll still do most of the flag checking inline. We check a lot of
 744  * conditions, but because they are pretty simple, they are expected to be
 745  * fast.
 746  */
 747 bool __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg,
 748                                         int order);
 749 void __memcg_kmem_commit_charge(struct page *page,
 750                                        struct mem_cgroup *memcg, int order);
 751 void __memcg_kmem_uncharge_pages(struct page *page, int order);
 752
 753 /*
 754  * helper for acessing a memcg's index. It will be used as an index in the
 755  * child cache array in kmem_cache, and also to derive its name. This function
 756  * will return -1 when this is not a kmem-limited memcg.
 757  */
 758 static inline int memcg_cache_id(struct mem_cgroup *memcg)
 759 {
 760         return memcg ? memcg->kmemcg_id : -1;
 761 }
 762
 763 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep);
 764 void __memcg_kmem_put_cache(struct kmem_cache *cachep);
 765
 766 struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr);
 767
 768 int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp,
 769                       unsigned long nr_pages);
 770 void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages);
 771
 772 /**
 773  * memcg_kmem_newpage_charge: verify if a new kmem allocation is allowed.
 774  * @gfp: the gfp allocation flags.
 775  * @memcg: a pointer to the memcg this was charged against.
 776  * @order: allocation order.
 777  *
 778  * returns true if the memcg where the current task belongs can hold this
 779  * allocation.
 780  *
 781  * We return true automatically if this allocation is not to be accounted to
 782  * any memcg.
 783  */
 784 static inline bool
 785 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
 786 {
 787         if (!memcg_kmem_enabled())
 788                 return true;
 789
 790         if (gfp & __GFP_NOACCOUNT)
 791                 return true;
 792         /*
 793          * __GFP_NOFAIL allocations will move on even if charging is not
 794          * possible. Therefore we don't even try, and have this allocation
 795          * unaccounted. We could in theory charge it forcibly, but we hope
 796          * those allocations are rare, and won't be worth the trouble.
 797          */
 798         if (gfp & __GFP_NOFAIL)
 799                 return true;
 800         if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
 801                 return true;
 802
 803         /* If the test is dying, just let it go. */
 804         if (unlikely(fatal_signal_pending(current)))
 805                 return true;
 806
 807         return __memcg_kmem_newpage_charge(gfp, memcg, order);
 808 }
 809
 810 /**
 811  * memcg_kmem_uncharge_pages: uncharge pages from memcg
 812  * @page: pointer to struct page being freed
 813  * @order: allocation order.
 814  */
 815 static inline void
 816 memcg_kmem_uncharge_pages(struct page *page, int order)
 817 {
 818         if (memcg_kmem_enabled())
 819                 __memcg_kmem_uncharge_pages(page, order);
 820 }
 821
 822 /**
 823  * memcg_kmem_commit_charge: embeds correct memcg in a page
 824  * @page: pointer to struct page recently allocated
 825  * @memcg: the memcg structure we charged against
 826  * @order: allocation order.
 827  *
 828  * Needs to be called after memcg_kmem_newpage_charge, regardless of success or
 829  * failure of the allocation. if @page is NULL, this function will revert the
 830  * charges. Otherwise, it will commit @page to @memcg.
 831  */
 832 static inline void
 833 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
 834 {
 835         if (memcg_kmem_enabled() && memcg)
 836                 __memcg_kmem_commit_charge(page, memcg, order);
 837 }
 838
 839 /**
 840  * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
 841  * @cachep: the original global kmem cache
 842  * @gfp: allocation flags.
 843  *
 844  * All memory allocated from a per-memcg cache is charged to the owner memcg.
 845  */
 846 static __always_inline struct kmem_cache *
 847 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
 848 {
 849         if (!memcg_kmem_enabled())
 850                 return cachep;
 851         if (gfp & __GFP_NOACCOUNT)
 852                 return cachep;
 853         if (gfp & __GFP_NOFAIL)
 854                 return cachep;
 855         if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
 856                 return cachep;
 857         if (unlikely(fatal_signal_pending(current)))
 858                 return cachep;
 859
 860         return __memcg_kmem_get_cache(cachep);
 861 }
 862
 863 static __always_inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
 864 {
 865         if (memcg_kmem_enabled())
 866                 __memcg_kmem_put_cache(cachep);
 867 }
 868
 869 static __always_inline struct mem_cgroup *mem_cgroup_from_kmem(void *ptr)
 870 {
 871         if (!memcg_kmem_enabled())
 872                 return NULL;
 873         return __mem_cgroup_from_kmem(ptr);
 874 }
 875 #else
 876 #define for_each_memcg_cache_index(_idx)        \
 877         for (; NULL; )
 878
 879 static inline bool memcg_kmem_enabled(void)
 880 {
 881         return false;
 882 }
 883
 884 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
 885 {
 886         return false;
 887 }
 888
 889 static inline bool
 890 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
 891 {
 892         return true;
 893 }
 894
 895 static inline void memcg_kmem_uncharge_pages(struct page *page, int order)
 896 {
 897 }
 898
 899 static inline void
 900 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
 901 {
 902 }
 903
 904 static inline int memcg_cache_id(struct mem_cgroup *memcg)
 905 {
 906         return -1;
 907 }
 908
 909 static inline void memcg_get_cache_ids(void)
 910 {
 911 }
 912
 913 static inline void memcg_put_cache_ids(void)
 914 {
 915 }
 916
 917 static inline struct kmem_cache *
 918 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
 919 {
 920         return cachep;
 921 }
 922
 923 static inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
 924 {
 925 }
 926
 927 static inline struct mem_cgroup *mem_cgroup_from_kmem(void *ptr)
 928 {
 929         return NULL;
 930 }
 931 #endif /* CONFIG_MEMCG_KMEM */
 932 #endif /* _LINUX_MEMCONTROL_H */
 933