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
 309 struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page);
 310 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
 311
 312 struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
 313 static inline
 314 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css){
 315         return css ? container_of(css, struct mem_cgroup, css) : NULL;
 316 }
 317
 318 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
 319                                    struct mem_cgroup *,
 320                                    struct mem_cgroup_reclaim_cookie *);
 321 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
 322
 323 static inline bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
 324                               struct mem_cgroup *root)
 325 {
 326         if (root == memcg)
 327                 return true;
 328         if (!root->use_hierarchy)
 329                 return false;
 330         return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
 331 }
 332
 333 static inline bool mm_match_cgroup(struct mm_struct *mm,
 334                                    struct mem_cgroup *memcg)
 335 {
 336         struct mem_cgroup *task_memcg;
 337         bool match = false;
 338
 339         rcu_read_lock();
 340         task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 341         if (task_memcg)
 342                 match = mem_cgroup_is_descendant(task_memcg, memcg);
 343         rcu_read_unlock();
 344         return match;
 345 }
 346
 347 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
 348
 349 static inline bool mem_cgroup_disabled(void)
 350 {
 351         if (memory_cgrp_subsys.disabled)
 352                 return true;
 353         return false;
 354 }
 355
 356 /*
 357  * For memory reclaim.
 358  */
 359 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
 360
 361 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
 362                 int nr_pages);
 363
 364 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
 365 {
 366         struct mem_cgroup_per_zone *mz;
 367         struct mem_cgroup *memcg;
 368
 369         if (mem_cgroup_disabled())
 370                 return true;
 371
 372         mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
 373         memcg = mz->memcg;
 374
 375         return !!(memcg->css.flags & CSS_ONLINE);
 376 }
 377
 378 static inline
 379 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
 380 {
 381         struct mem_cgroup_per_zone *mz;
 382
 383         mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
 384         return mz->lru_size[lru];
 385 }
 386
 387 static inline int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
 388 {
 389         unsigned long inactive_ratio;
 390         unsigned long inactive;
 391         unsigned long active;
 392         unsigned long gb;
 393
 394         inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
 395         active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
 396
 397         gb = (inactive + active) >> (30 - PAGE_SHIFT);
 398         if (gb)
 399                 inactive_ratio = int_sqrt(10 * gb);
 400         else
 401                 inactive_ratio = 1;
 402
 403         return inactive * inactive_ratio < active;
 404 }
 405
 406 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
 407                                 struct task_struct *p);
 408
 409 static inline void mem_cgroup_oom_enable(void)
 410 {
 411         WARN_ON(current->memcg_oom.may_oom);
 412         current->memcg_oom.may_oom = 1;
 413 }
 414
 415 static inline void mem_cgroup_oom_disable(void)
 416 {
 417         WARN_ON(!current->memcg_oom.may_oom);
 418         current->memcg_oom.may_oom = 0;
 419 }
 420
 421 static inline bool task_in_memcg_oom(struct task_struct *p)
 422 {
 423         return p->memcg_oom.memcg;
 424 }
 425
 426 bool mem_cgroup_oom_synchronize(bool wait);
 427
 428 #ifdef CONFIG_MEMCG_SWAP
 429 extern int do_swap_account;
 430 #endif
 431
 432 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
 433 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
 434
 435 /**
 436  * mem_cgroup_update_page_stat - update page state statistics
 437  * @memcg: memcg to account against
 438  * @idx: page state item to account
 439  * @val: number of pages (positive or negative)
 440  *
 441  * See mem_cgroup_begin_page_stat() for locking requirements.
 442  */
 443 static inline void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
 444                                  enum mem_cgroup_stat_index idx, int val)
 445 {
 446         VM_BUG_ON(!rcu_read_lock_held());
 447
 448         if (memcg)
 449                 this_cpu_add(memcg->stat->count[idx], val);
 450 }
 451
 452 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
 453                                             enum mem_cgroup_stat_index idx)
 454 {
 455         mem_cgroup_update_page_stat(memcg, idx, 1);
 456 }
 457
 458 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
 459                                             enum mem_cgroup_stat_index idx)
 460 {
 461         mem_cgroup_update_page_stat(memcg, idx, -1);
 462 }
 463
 464 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
 465                                                 gfp_t gfp_mask,
 466                                                 unsigned long *total_scanned);
 467
 468 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
 469                                              enum vm_event_item idx)
 470 {
 471         struct mem_cgroup *memcg;
 472
 473         if (mem_cgroup_disabled())
 474                 return;
 475
 476         rcu_read_lock();
 477         memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 478         if (unlikely(!memcg))
 479                 goto out;
 480
 481         switch (idx) {
 482         case PGFAULT:
 483                 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
 484                 break;
 485         case PGMAJFAULT:
 486                 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
 487                 break;
 488         default:
 489                 BUG();
 490         }
 491 out:
 492         rcu_read_unlock();
 493 }
 494 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 495 void mem_cgroup_split_huge_fixup(struct page *head);
 496 #endif
 497
 498 #else /* CONFIG_MEMCG */
 499 struct mem_cgroup;
 500
 501 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
 502                                      enum mem_cgroup_events_index idx,
 503                                      unsigned int nr)
 504 {
 505 }
 506
 507 static inline bool mem_cgroup_low(struct mem_cgroup *root,
 508                                   struct mem_cgroup *memcg)
 509 {
 510         return false;
 511 }
 512
 513 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
 514                                         gfp_t gfp_mask,
 515                                         struct mem_cgroup **memcgp)
 516 {
 517         *memcgp = NULL;
 518         return 0;
 519 }
 520
 521 static inline void mem_cgroup_commit_charge(struct page *page,
 522                                             struct mem_cgroup *memcg,
 523                                             bool lrucare)
 524 {
 525 }
 526
 527 static inline void mem_cgroup_cancel_charge(struct page *page,
 528                                             struct mem_cgroup *memcg)
 529 {
 530 }
 531
 532 static inline void mem_cgroup_uncharge(struct page *page)
 533 {
 534 }
 535
 536 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
 537 {
 538 }
 539
 540 static inline void mem_cgroup_migrate(struct page *oldpage,
 541                                       struct page *newpage,
 542                                       bool lrucare)
 543 {
 544 }
 545
 546 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
 547                                                     struct mem_cgroup *memcg)
 548 {
 549         return &zone->lruvec;
 550 }
 551
 552 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
 553                                                     struct zone *zone)
 554 {
 555         return &zone->lruvec;
 556 }
 557
 558 static inline struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
 559 {
 560         return NULL;
 561 }
 562
 563 static inline bool mm_match_cgroup(struct mm_struct *mm,
 564                 struct mem_cgroup *memcg)
 565 {
 566         return true;
 567 }
 568
 569 static inline bool task_in_mem_cgroup(struct task_struct *task,
 570                                       const struct mem_cgroup *memcg)
 571 {
 572         return true;
 573 }
 574
 575 static inline struct mem_cgroup *
 576 mem_cgroup_iter(struct mem_cgroup *root,
 577                 struct mem_cgroup *prev,
 578                 struct mem_cgroup_reclaim_cookie *reclaim)
 579 {
 580         return NULL;
 581 }
 582
 583 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
 584                                          struct mem_cgroup *prev)
 585 {
 586 }
 587
 588 static inline bool mem_cgroup_disabled(void)
 589 {
 590         return true;
 591 }
 592
 593 static inline int
 594 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
 595 {
 596         return 1;
 597 }
 598
 599 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
 600 {
 601         return true;
 602 }
 603
 604 static inline unsigned long
 605 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
 606 {
 607         return 0;
 608 }
 609
 610 static inline void
 611 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
 612                               int increment)
 613 {
 614 }
 615
 616 static inline void
 617 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
 618 {
 619 }
 620
 621 static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
 622 {
 623         return NULL;
 624 }
 625
 626 static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
 627 {
 628 }
 629
 630 static inline void mem_cgroup_oom_enable(void)
 631 {
 632 }
 633
 634 static inline void mem_cgroup_oom_disable(void)
 635 {
 636 }
 637
 638 static inline bool task_in_memcg_oom(struct task_struct *p)
 639 {
 640         return false;
 641 }
 642
 643 static inline bool mem_cgroup_oom_synchronize(bool wait)
 644 {
 645         return false;
 646 }
 647
 648 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
 649                                             enum mem_cgroup_stat_index idx)
 650 {
 651 }
 652
 653 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
 654                                             enum mem_cgroup_stat_index idx)
 655 {
 656 }
 657
 658 static inline
 659 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
 660                                             gfp_t gfp_mask,
 661                                             unsigned long *total_scanned)
 662 {
 663         return 0;
 664 }
 665
 666 static inline void mem_cgroup_split_huge_fixup(struct page *head)
 667 {
 668 }
 669
 670 static inline
 671 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
 672 {
 673 }
 674 #endif /* CONFIG_MEMCG */
 675
 676 enum {
 677         UNDER_LIMIT,
 678         SOFT_LIMIT,
 679         OVER_LIMIT,
 680 };
 681
 682 #ifdef CONFIG_CGROUP_WRITEBACK
 683
 684 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
 685 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
 686 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pavail,
 687                          unsigned long *pdirty, unsigned long *pwriteback);
 688
 689 #else   /* CONFIG_CGROUP_WRITEBACK */
 690
 691 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
 692 {
 693         return NULL;
 694 }
 695
 696 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
 697                                        unsigned long *pavail,
 698                                        unsigned long *pdirty,
 699                                        unsigned long *pwriteback)
 700 {
 701 }
 702
 703 #endif  /* CONFIG_CGROUP_WRITEBACK */
 704
 705 struct sock;
 706 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
 707 void sock_update_memcg(struct sock *sk);
 708 void sock_release_memcg(struct sock *sk);
 709 #else
 710 static inline void sock_update_memcg(struct sock *sk)
 711 {
 712 }
 713 static inline void sock_release_memcg(struct sock *sk)
 714 {
 715 }
 716 #endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
 717
 718 #ifdef CONFIG_MEMCG_KMEM
 719 extern struct static_key memcg_kmem_enabled_key;
 720
 721 extern int memcg_nr_cache_ids;
 722 void memcg_get_cache_ids(void);
 723 void memcg_put_cache_ids(void);
 724
 725 /*
 726  * Helper macro to loop through all memcg-specific caches. Callers must still
 727  * check if the cache is valid (it is either valid or NULL).
 728  * the slab_mutex must be held when looping through those caches
 729  */
 730 #define for_each_memcg_cache_index(_idx)        \
 731         for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
 732
 733 static inline bool memcg_kmem_enabled(void)
 734 {
 735         return static_key_false(&memcg_kmem_enabled_key);
 736 }
 737
 738 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
 739 {
 740         return memcg->kmem_acct_active;
 741 }
 742
 743 /*
 744  * In general, we'll do everything in our power to not incur in any overhead
 745  * for non-memcg users for the kmem functions. Not even a function call, if we
 746  * can avoid it.
 747  *
 748  * Therefore, we'll inline all those functions so that in the best case, we'll
 749  * see that kmemcg is off for everybody and proceed quickly.  If it is on,
 750  * we'll still do most of the flag checking inline. We check a lot of
 751  * conditions, but because they are pretty simple, they are expected to be
 752  * fast.
 753  */
 754 bool __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg,
 755                                         int order);
 756 void __memcg_kmem_commit_charge(struct page *page,
 757                                        struct mem_cgroup *memcg, int order);
 758 void __memcg_kmem_uncharge_pages(struct page *page, int order);
 759
 760 /*
 761  * helper for acessing a memcg's index. It will be used as an index in the
 762  * child cache array in kmem_cache, and also to derive its name. This function
 763  * will return -1 when this is not a kmem-limited memcg.
 764  */
 765 static inline int memcg_cache_id(struct mem_cgroup *memcg)
 766 {
 767         return memcg ? memcg->kmemcg_id : -1;
 768 }
 769
 770 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep);
 771 void __memcg_kmem_put_cache(struct kmem_cache *cachep);
 772
 773 struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr);
 774
 775 int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp,
 776                       unsigned long nr_pages);
 777 void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages);
 778
 779 /**
 780  * memcg_kmem_newpage_charge: verify if a new kmem allocation is allowed.
 781  * @gfp: the gfp allocation flags.
 782  * @memcg: a pointer to the memcg this was charged against.
 783  * @order: allocation order.
 784  *
 785  * returns true if the memcg where the current task belongs can hold this
 786  * allocation.
 787  *
 788  * We return true automatically if this allocation is not to be accounted to
 789  * any memcg.
 790  */
 791 static inline bool
 792 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
 793 {
 794         if (!memcg_kmem_enabled())
 795                 return true;
 796
 797         if (gfp & __GFP_NOACCOUNT)
 798                 return true;
 799         /*
 800          * __GFP_NOFAIL allocations will move on even if charging is not
 801          * possible. Therefore we don't even try, and have this allocation
 802          * unaccounted. We could in theory charge it forcibly, but we hope
 803          * those allocations are rare, and won't be worth the trouble.
 804          */
 805         if (gfp & __GFP_NOFAIL)
 806                 return true;
 807         if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
 808                 return true;
 809
 810         /* If the test is dying, just let it go. */
 811         if (unlikely(fatal_signal_pending(current)))
 812                 return true;
 813
 814         return __memcg_kmem_newpage_charge(gfp, memcg, order);
 815 }
 816
 817 /**
 818  * memcg_kmem_uncharge_pages: uncharge pages from memcg
 819  * @page: pointer to struct page being freed
 820  * @order: allocation order.
 821  */
 822 static inline void
 823 memcg_kmem_uncharge_pages(struct page *page, int order)
 824 {
 825         if (memcg_kmem_enabled())
 826                 __memcg_kmem_uncharge_pages(page, order);
 827 }
 828
 829 /**
 830  * memcg_kmem_commit_charge: embeds correct memcg in a page
 831  * @page: pointer to struct page recently allocated
 832  * @memcg: the memcg structure we charged against
 833  * @order: allocation order.
 834  *
 835  * Needs to be called after memcg_kmem_newpage_charge, regardless of success or
 836  * failure of the allocation. if @page is NULL, this function will revert the
 837  * charges. Otherwise, it will commit @page to @memcg.
 838  */
 839 static inline void
 840 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
 841 {
 842         if (memcg_kmem_enabled() && memcg)
 843                 __memcg_kmem_commit_charge(page, memcg, order);
 844 }
 845
 846 /**
 847  * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
 848  * @cachep: the original global kmem cache
 849  * @gfp: allocation flags.
 850  *
 851  * All memory allocated from a per-memcg cache is charged to the owner memcg.
 852  */
 853 static __always_inline struct kmem_cache *
 854 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
 855 {
 856         if (!memcg_kmem_enabled())
 857                 return cachep;
 858         if (gfp & __GFP_NOACCOUNT)
 859                 return cachep;
 860         if (gfp & __GFP_NOFAIL)
 861                 return cachep;
 862         if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
 863                 return cachep;
 864         if (unlikely(fatal_signal_pending(current)))
 865                 return cachep;
 866
 867         return __memcg_kmem_get_cache(cachep);
 868 }
 869
 870 static __always_inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
 871 {
 872         if (memcg_kmem_enabled())
 873                 __memcg_kmem_put_cache(cachep);
 874 }
 875
 876 static __always_inline struct mem_cgroup *mem_cgroup_from_kmem(void *ptr)
 877 {
 878         if (!memcg_kmem_enabled())
 879                 return NULL;
 880         return __mem_cgroup_from_kmem(ptr);
 881 }
 882 #else
 883 #define for_each_memcg_cache_index(_idx)        \
 884         for (; NULL; )
 885
 886 static inline bool memcg_kmem_enabled(void)
 887 {
 888         return false;
 889 }
 890
 891 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
 892 {
 893         return false;
 894 }
 895
 896 static inline bool
 897 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
 898 {
 899         return true;
 900 }
 901
 902 static inline void memcg_kmem_uncharge_pages(struct page *page, int order)
 903 {
 904 }
 905
 906 static inline void
 907 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
 908 {
 909 }
 910
 911 static inline int memcg_cache_id(struct mem_cgroup *memcg)
 912 {
 913         return -1;
 914 }
 915
 916 static inline void memcg_get_cache_ids(void)
 917 {
 918 }
 919
 920 static inline void memcg_put_cache_ids(void)
 921 {
 922 }
 923
 924 static inline struct kmem_cache *
 925 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
 926 {
 927         return cachep;
 928 }
 929
 930 static inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
 931 {
 932 }
 933
 934 static inline struct mem_cgroup *mem_cgroup_from_kmem(void *ptr)
 935 {
 936         return NULL;
 937 }
 938 #endif /* CONFIG_MEMCG_KMEM */
 939 #endif /* _LINUX_MEMCONTROL_H */
 940