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
  27 struct mem_cgroup;
  28 struct page;
  29 struct mm_struct;
  30 struct kmem_cache;
  31
  32 /*
  33  * The corresponding mem_cgroup_stat_names is defined in mm/memcontrol.c,
  34  * These two lists should keep in accord with each other.
  35  */
  36 enum mem_cgroup_stat_index {
  37         /*
  38          * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
  39          */
  40         MEM_CGROUP_STAT_CACHE,          /* # of pages charged as cache */
  41         MEM_CGROUP_STAT_RSS,            /* # of pages charged as anon rss */
  42         MEM_CGROUP_STAT_RSS_HUGE,       /* # of pages charged as anon huge */
  43         MEM_CGROUP_STAT_FILE_MAPPED,    /* # of pages charged as file rss */
  44         MEM_CGROUP_STAT_DIRTY,          /* # of dirty pages in page cache */
  45         MEM_CGROUP_STAT_WRITEBACK,      /* # of pages under writeback */
  46         MEM_CGROUP_STAT_SWAP,           /* # of pages, swapped out */
  47         MEM_CGROUP_STAT_NSTATS,
  48 };
  49
  50 struct mem_cgroup_reclaim_cookie {
  51         struct zone *zone;
  52         int priority;
  53         unsigned int generation;
  54 };
  55
  56 enum mem_cgroup_events_index {
  57         MEM_CGROUP_EVENTS_PGPGIN,       /* # of pages paged in */
  58         MEM_CGROUP_EVENTS_PGPGOUT,      /* # of pages paged out */
  59         MEM_CGROUP_EVENTS_PGFAULT,      /* # of page-faults */
  60         MEM_CGROUP_EVENTS_PGMAJFAULT,   /* # of major page-faults */
  61         MEM_CGROUP_EVENTS_NSTATS,
  62         /* default hierarchy events */
  63         MEMCG_LOW = MEM_CGROUP_EVENTS_NSTATS,
  64         MEMCG_HIGH,
  65         MEMCG_MAX,
  66         MEMCG_OOM,
  67         MEMCG_NR_EVENTS,
  68 };
  69
  70 #ifdef CONFIG_MEMCG
  71 extern struct cgroup_subsys_state *mem_cgroup_root_css;
  72
  73 void mem_cgroup_events(struct mem_cgroup *memcg,
  74                        enum mem_cgroup_events_index idx,
  75                        unsigned int nr);
  76
  77 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
  78
  79 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
  80                           gfp_t gfp_mask, struct mem_cgroup **memcgp);
  81 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
  82                               bool lrucare);
  83 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg);
  84 void mem_cgroup_uncharge(struct page *page);
  85 void mem_cgroup_uncharge_list(struct list_head *page_list);
  86
  87 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage,
  88                         bool lrucare);
  89
  90 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
  91 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
  92
  93 bool mem_cgroup_is_descendant(struct mem_cgroup *memcg,
  94                               struct mem_cgroup *root);
  95 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg);
  96
  97 extern struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page);
  98 extern struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
  99
 100 extern struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
 101 extern struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css);
 102
 103 static inline bool mm_match_cgroup(struct mm_struct *mm,
 104                                    struct mem_cgroup *memcg)
 105 {
 106         struct mem_cgroup *task_memcg;
 107         bool match = false;
 108
 109         rcu_read_lock();
 110         task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 111         if (task_memcg)
 112                 match = mem_cgroup_is_descendant(task_memcg, memcg);
 113         rcu_read_unlock();
 114         return match;
 115 }
 116
 117 extern struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg);
 118 extern struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page);
 119
 120 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
 121                                    struct mem_cgroup *,
 122                                    struct mem_cgroup_reclaim_cookie *);
 123 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
 124
 125 /*
 126  * For memory reclaim.
 127  */
 128 int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec);
 129 bool mem_cgroup_lruvec_online(struct lruvec *lruvec);
 130 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
 131 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list);
 132 void mem_cgroup_update_lru_size(struct lruvec *, enum lru_list, int);
 133 extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
 134                                         struct task_struct *p);
 135
 136 static inline void mem_cgroup_oom_enable(void)
 137 {
 138         WARN_ON(current->memcg_oom.may_oom);
 139         current->memcg_oom.may_oom = 1;
 140 }
 141
 142 static inline void mem_cgroup_oom_disable(void)
 143 {
 144         WARN_ON(!current->memcg_oom.may_oom);
 145         current->memcg_oom.may_oom = 0;
 146 }
 147
 148 static inline bool task_in_memcg_oom(struct task_struct *p)
 149 {
 150         return p->memcg_oom.memcg;
 151 }
 152
 153 bool mem_cgroup_oom_synchronize(bool wait);
 154
 155 #ifdef CONFIG_MEMCG_SWAP
 156 extern int do_swap_account;
 157 #endif
 158
 159 static inline bool mem_cgroup_disabled(void)
 160 {
 161         if (memory_cgrp_subsys.disabled)
 162                 return true;
 163         return false;
 164 }
 165
 166 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page);
 167 void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
 168                                  enum mem_cgroup_stat_index idx, int val);
 169 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg);
 170
 171 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
 172                                             enum mem_cgroup_stat_index idx)
 173 {
 174         mem_cgroup_update_page_stat(memcg, idx, 1);
 175 }
 176
 177 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
 178                                             enum mem_cgroup_stat_index idx)
 179 {
 180         mem_cgroup_update_page_stat(memcg, idx, -1);
 181 }
 182
 183 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
 184                                                 gfp_t gfp_mask,
 185                                                 unsigned long *total_scanned);
 186
 187 void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx);
 188 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
 189                                              enum vm_event_item idx)
 190 {
 191         if (mem_cgroup_disabled())
 192                 return;
 193         __mem_cgroup_count_vm_event(mm, idx);
 194 }
 195 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 196 void mem_cgroup_split_huge_fixup(struct page *head);
 197 #endif
 198
 199 #else /* CONFIG_MEMCG */
 200 struct mem_cgroup;
 201
 202 #define mem_cgroup_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
 203
 204 static inline void mem_cgroup_events(struct mem_cgroup *memcg,
 205                                      enum mem_cgroup_events_index idx,
 206                                      unsigned int nr)
 207 {
 208 }
 209
 210 static inline bool mem_cgroup_low(struct mem_cgroup *root,
 211                                   struct mem_cgroup *memcg)
 212 {
 213         return false;
 214 }
 215
 216 static inline int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
 217                                         gfp_t gfp_mask,
 218                                         struct mem_cgroup **memcgp)
 219 {
 220         *memcgp = NULL;
 221         return 0;
 222 }
 223
 224 static inline void mem_cgroup_commit_charge(struct page *page,
 225                                             struct mem_cgroup *memcg,
 226                                             bool lrucare)
 227 {
 228 }
 229
 230 static inline void mem_cgroup_cancel_charge(struct page *page,
 231                                             struct mem_cgroup *memcg)
 232 {
 233 }
 234
 235 static inline void mem_cgroup_uncharge(struct page *page)
 236 {
 237 }
 238
 239 static inline void mem_cgroup_uncharge_list(struct list_head *page_list)
 240 {
 241 }
 242
 243 static inline void mem_cgroup_migrate(struct page *oldpage,
 244                                       struct page *newpage,
 245                                       bool lrucare)
 246 {
 247 }
 248
 249 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
 250                                                     struct mem_cgroup *memcg)
 251 {
 252         return &zone->lruvec;
 253 }
 254
 255 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
 256                                                     struct zone *zone)
 257 {
 258         return &zone->lruvec;
 259 }
 260
 261 static inline struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
 262 {
 263         return NULL;
 264 }
 265
 266 static inline bool mm_match_cgroup(struct mm_struct *mm,
 267                 struct mem_cgroup *memcg)
 268 {
 269         return true;
 270 }
 271
 272 static inline bool task_in_mem_cgroup(struct task_struct *task,
 273                                       const struct mem_cgroup *memcg)
 274 {
 275         return true;
 276 }
 277
 278 static inline struct cgroup_subsys_state
 279                 *mem_cgroup_css(struct mem_cgroup *memcg)
 280 {
 281         return NULL;
 282 }
 283
 284 static inline struct mem_cgroup *
 285 mem_cgroup_iter(struct mem_cgroup *root,
 286                 struct mem_cgroup *prev,
 287                 struct mem_cgroup_reclaim_cookie *reclaim)
 288 {
 289         return NULL;
 290 }
 291
 292 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
 293                                          struct mem_cgroup *prev)
 294 {
 295 }
 296
 297 static inline bool mem_cgroup_disabled(void)
 298 {
 299         return true;
 300 }
 301
 302 static inline int
 303 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
 304 {
 305         return 1;
 306 }
 307
 308 static inline bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
 309 {
 310         return true;
 311 }
 312
 313 static inline unsigned long
 314 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
 315 {
 316         return 0;
 317 }
 318
 319 static inline void
 320 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
 321                               int increment)
 322 {
 323 }
 324
 325 static inline void
 326 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
 327 {
 328 }
 329
 330 static inline struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
 331 {
 332         return NULL;
 333 }
 334
 335 static inline void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
 336 {
 337 }
 338
 339 static inline void mem_cgroup_oom_enable(void)
 340 {
 341 }
 342
 343 static inline void mem_cgroup_oom_disable(void)
 344 {
 345 }
 346
 347 static inline bool task_in_memcg_oom(struct task_struct *p)
 348 {
 349         return false;
 350 }
 351
 352 static inline bool mem_cgroup_oom_synchronize(bool wait)
 353 {
 354         return false;
 355 }
 356
 357 static inline void mem_cgroup_inc_page_stat(struct mem_cgroup *memcg,
 358                                             enum mem_cgroup_stat_index idx)
 359 {
 360 }
 361
 362 static inline void mem_cgroup_dec_page_stat(struct mem_cgroup *memcg,
 363                                             enum mem_cgroup_stat_index idx)
 364 {
 365 }
 366
 367 static inline
 368 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
 369                                             gfp_t gfp_mask,
 370                                             unsigned long *total_scanned)
 371 {
 372         return 0;
 373 }
 374
 375 static inline void mem_cgroup_split_huge_fixup(struct page *head)
 376 {
 377 }
 378
 379 static inline
 380 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
 381 {
 382 }
 383 #endif /* CONFIG_MEMCG */
 384
 385 enum {
 386         UNDER_LIMIT,
 387         SOFT_LIMIT,
 388         OVER_LIMIT,
 389 };
 390
 391 #ifdef CONFIG_CGROUP_WRITEBACK
 392
 393 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg);
 394 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb);
 395 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pavail,
 396                          unsigned long *pdirty, unsigned long *pwriteback);
 397
 398 #else   /* CONFIG_CGROUP_WRITEBACK */
 399
 400 static inline struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
 401 {
 402         return NULL;
 403 }
 404
 405 static inline void mem_cgroup_wb_stats(struct bdi_writeback *wb,
 406                                        unsigned long *pavail,
 407                                        unsigned long *pdirty,
 408                                        unsigned long *pwriteback)
 409 {
 410 }
 411
 412 #endif  /* CONFIG_CGROUP_WRITEBACK */
 413
 414 struct sock;
 415 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
 416 void sock_update_memcg(struct sock *sk);
 417 void sock_release_memcg(struct sock *sk);
 418 #else
 419 static inline void sock_update_memcg(struct sock *sk)
 420 {
 421 }
 422 static inline void sock_release_memcg(struct sock *sk)
 423 {
 424 }
 425 #endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
 426
 427 #ifdef CONFIG_MEMCG_KMEM
 428 extern struct static_key memcg_kmem_enabled_key;
 429
 430 extern int memcg_nr_cache_ids;
 431 extern void memcg_get_cache_ids(void);
 432 extern void memcg_put_cache_ids(void);
 433
 434 /*
 435  * Helper macro to loop through all memcg-specific caches. Callers must still
 436  * check if the cache is valid (it is either valid or NULL).
 437  * the slab_mutex must be held when looping through those caches
 438  */
 439 #define for_each_memcg_cache_index(_idx)        \
 440         for ((_idx) = 0; (_idx) < memcg_nr_cache_ids; (_idx)++)
 441
 442 static inline bool memcg_kmem_enabled(void)
 443 {
 444         return static_key_false(&memcg_kmem_enabled_key);
 445 }
 446
 447 bool memcg_kmem_is_active(struct mem_cgroup *memcg);
 448
 449 /*
 450  * In general, we'll do everything in our power to not incur in any overhead
 451  * for non-memcg users for the kmem functions. Not even a function call, if we
 452  * can avoid it.
 453  *
 454  * Therefore, we'll inline all those functions so that in the best case, we'll
 455  * see that kmemcg is off for everybody and proceed quickly.  If it is on,
 456  * we'll still do most of the flag checking inline. We check a lot of
 457  * conditions, but because they are pretty simple, they are expected to be
 458  * fast.
 459  */
 460 bool __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg,
 461                                         int order);
 462 void __memcg_kmem_commit_charge(struct page *page,
 463                                        struct mem_cgroup *memcg, int order);
 464 void __memcg_kmem_uncharge_pages(struct page *page, int order);
 465
 466 int memcg_cache_id(struct mem_cgroup *memcg);
 467
 468 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep);
 469 void __memcg_kmem_put_cache(struct kmem_cache *cachep);
 470
 471 struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr);
 472
 473 int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp,
 474                       unsigned long nr_pages);
 475 void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages);
 476
 477 /**
 478  * memcg_kmem_newpage_charge: verify if a new kmem allocation is allowed.
 479  * @gfp: the gfp allocation flags.
 480  * @memcg: a pointer to the memcg this was charged against.
 481  * @order: allocation order.
 482  *
 483  * returns true if the memcg where the current task belongs can hold this
 484  * allocation.
 485  *
 486  * We return true automatically if this allocation is not to be accounted to
 487  * any memcg.
 488  */
 489 static inline bool
 490 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
 491 {
 492         if (!memcg_kmem_enabled())
 493                 return true;
 494
 495         if (gfp & __GFP_NOACCOUNT)
 496                 return true;
 497         /*
 498          * __GFP_NOFAIL allocations will move on even if charging is not
 499          * possible. Therefore we don't even try, and have this allocation
 500          * unaccounted. We could in theory charge it forcibly, but we hope
 501          * those allocations are rare, and won't be worth the trouble.
 502          */
 503         if (gfp & __GFP_NOFAIL)
 504                 return true;
 505         if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
 506                 return true;
 507
 508         /* If the test is dying, just let it go. */
 509         if (unlikely(fatal_signal_pending(current)))
 510                 return true;
 511
 512         return __memcg_kmem_newpage_charge(gfp, memcg, order);
 513 }
 514
 515 /**
 516  * memcg_kmem_uncharge_pages: uncharge pages from memcg
 517  * @page: pointer to struct page being freed
 518  * @order: allocation order.
 519  */
 520 static inline void
 521 memcg_kmem_uncharge_pages(struct page *page, int order)
 522 {
 523         if (memcg_kmem_enabled())
 524                 __memcg_kmem_uncharge_pages(page, order);
 525 }
 526
 527 /**
 528  * memcg_kmem_commit_charge: embeds correct memcg in a page
 529  * @page: pointer to struct page recently allocated
 530  * @memcg: the memcg structure we charged against
 531  * @order: allocation order.
 532  *
 533  * Needs to be called after memcg_kmem_newpage_charge, regardless of success or
 534  * failure of the allocation. if @page is NULL, this function will revert the
 535  * charges. Otherwise, it will commit @page to @memcg.
 536  */
 537 static inline void
 538 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
 539 {
 540         if (memcg_kmem_enabled() && memcg)
 541                 __memcg_kmem_commit_charge(page, memcg, order);
 542 }
 543
 544 /**
 545  * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
 546  * @cachep: the original global kmem cache
 547  * @gfp: allocation flags.
 548  *
 549  * All memory allocated from a per-memcg cache is charged to the owner memcg.
 550  */
 551 static __always_inline struct kmem_cache *
 552 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
 553 {
 554         if (!memcg_kmem_enabled())
 555                 return cachep;
 556         if (gfp & __GFP_NOACCOUNT)
 557                 return cachep;
 558         if (gfp & __GFP_NOFAIL)
 559                 return cachep;
 560         if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
 561                 return cachep;
 562         if (unlikely(fatal_signal_pending(current)))
 563                 return cachep;
 564
 565         return __memcg_kmem_get_cache(cachep);
 566 }
 567
 568 static __always_inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
 569 {
 570         if (memcg_kmem_enabled())
 571                 __memcg_kmem_put_cache(cachep);
 572 }
 573
 574 static __always_inline struct mem_cgroup *mem_cgroup_from_kmem(void *ptr)
 575 {
 576         if (!memcg_kmem_enabled())
 577                 return NULL;
 578         return __mem_cgroup_from_kmem(ptr);
 579 }
 580 #else
 581 #define for_each_memcg_cache_index(_idx)        \
 582         for (; NULL; )
 583
 584 static inline bool memcg_kmem_enabled(void)
 585 {
 586         return false;
 587 }
 588
 589 static inline bool memcg_kmem_is_active(struct mem_cgroup *memcg)
 590 {
 591         return false;
 592 }
 593
 594 static inline bool
 595 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
 596 {
 597         return true;
 598 }
 599
 600 static inline void memcg_kmem_uncharge_pages(struct page *page, int order)
 601 {
 602 }
 603
 604 static inline void
 605 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
 606 {
 607 }
 608
 609 static inline int memcg_cache_id(struct mem_cgroup *memcg)
 610 {
 611         return -1;
 612 }
 613
 614 static inline void memcg_get_cache_ids(void)
 615 {
 616 }
 617
 618 static inline void memcg_put_cache_ids(void)
 619 {
 620 }
 621
 622 static inline struct kmem_cache *
 623 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
 624 {
 625         return cachep;
 626 }
 627
 628 static inline void memcg_kmem_put_cache(struct kmem_cache *cachep)
 629 {
 630 }
 631
 632 static inline struct mem_cgroup *mem_cgroup_from_kmem(void *ptr)
 633 {
 634         return NULL;
 635 }
 636 #endif /* CONFIG_MEMCG_KMEM */
 637 #endif /* _LINUX_MEMCONTROL_H */
 638