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_cgroup;
  29 struct page;
  30 struct mm_struct;
  31 struct kmem_cache;
  32
  33 /* Stats that can be updated by kernel. */
  34 enum mem_cgroup_page_stat_item {
  35         MEMCG_NR_FILE_MAPPED, /* # of pages charged as file rss */
  36 };
  37
  38 struct mem_cgroup_reclaim_cookie {
  39         struct zone *zone;
  40         int priority;
  41         unsigned int generation;
  42 };
  43
  44 #ifdef CONFIG_MEMCG
  45 /*
  46  * All "charge" functions with gfp_mask should use GFP_KERNEL or
  47  * (gfp_mask & GFP_RECLAIM_MASK). In current implementatin, memcg doesn't
  48  * alloc memory but reclaims memory from all available zones. So, "where I want
  49  * memory from" bits of gfp_mask has no meaning. So any bits of that field is
  50  * available but adding a rule is better. charge functions' gfp_mask should
  51  * be set to GFP_KERNEL or gfp_mask & GFP_RECLAIM_MASK for avoiding ambiguous
  52  * codes.
  53  * (Of course, if memcg does memory allocation in future, GFP_KERNEL is sane.)
  54  */
  55
  56 extern int mem_cgroup_newpage_charge(struct page *page, struct mm_struct *mm,
  57                                 gfp_t gfp_mask);
  58 /* for swap handling */
  59 extern int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
  60                 struct page *page, gfp_t mask, struct mem_cgroup **memcgp);
  61 extern void mem_cgroup_commit_charge_swapin(struct page *page,
  62                                         struct mem_cgroup *memcg);
  63 extern void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg);
  64
  65 extern int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
  66                                         gfp_t gfp_mask);
  67
  68 struct lruvec *mem_cgroup_zone_lruvec(struct zone *, struct mem_cgroup *);
  69 struct lruvec *mem_cgroup_page_lruvec(struct page *, struct zone *);
  70
  71 /* For coalescing uncharge for reducing memcg' overhead*/
  72 extern void mem_cgroup_uncharge_start(void);
  73 extern void mem_cgroup_uncharge_end(void);
  74
  75 extern void mem_cgroup_uncharge_page(struct page *page);
  76 extern void mem_cgroup_uncharge_cache_page(struct page *page);
  77
  78 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg,
  79                                   struct mem_cgroup *memcg);
  80 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg);
  81
  82 extern struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page);
  83 extern struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p);
  84 extern struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm);
  85
  86 extern struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg);
  87 extern struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont);
  88
  89 static inline
  90 bool mm_match_cgroup(const struct mm_struct *mm, const struct mem_cgroup *memcg)
  91 {
  92         struct mem_cgroup *task_memcg;
  93         bool match;
  94
  95         rcu_read_lock();
  96         task_memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
  97         match = __mem_cgroup_same_or_subtree(memcg, task_memcg);
  98         rcu_read_unlock();
  99         return match;
 100 }
 101
 102 extern struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg);
 103
 104 extern void
 105 mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
 106                              struct mem_cgroup **memcgp);
 107 extern void mem_cgroup_end_migration(struct mem_cgroup *memcg,
 108         struct page *oldpage, struct page *newpage, bool migration_ok);
 109
 110 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *,
 111                                    struct mem_cgroup *,
 112                                    struct mem_cgroup_reclaim_cookie *);
 113 void mem_cgroup_iter_break(struct mem_cgroup *, struct mem_cgroup *);
 114
 115 /*
 116  * For memory reclaim.
 117  */
 118 int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec);
 119 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
 120 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list);
 121 void mem_cgroup_update_lru_size(struct lruvec *, enum lru_list, int);
 122 extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
 123                                         struct task_struct *p);
 124 extern void mem_cgroup_replace_page_cache(struct page *oldpage,
 125                                         struct page *newpage);
 126
 127 #ifdef CONFIG_MEMCG_SWAP
 128 extern int do_swap_account;
 129 #endif
 130
 131 static inline bool mem_cgroup_disabled(void)
 132 {
 133         if (mem_cgroup_subsys.disabled)
 134                 return true;
 135         return false;
 136 }
 137
 138 void __mem_cgroup_begin_update_page_stat(struct page *page, bool *locked,
 139                                          unsigned long *flags);
 140
 141 extern atomic_t memcg_moving;
 142
 143 static inline void mem_cgroup_begin_update_page_stat(struct page *page,
 144                                         bool *locked, unsigned long *flags)
 145 {
 146         if (mem_cgroup_disabled())
 147                 return;
 148         rcu_read_lock();
 149         *locked = false;
 150         if (atomic_read(&memcg_moving))
 151                 __mem_cgroup_begin_update_page_stat(page, locked, flags);
 152 }
 153
 154 void __mem_cgroup_end_update_page_stat(struct page *page,
 155                                 unsigned long *flags);
 156 static inline void mem_cgroup_end_update_page_stat(struct page *page,
 157                                         bool *locked, unsigned long *flags)
 158 {
 159         if (mem_cgroup_disabled())
 160                 return;
 161         if (*locked)
 162                 __mem_cgroup_end_update_page_stat(page, flags);
 163         rcu_read_unlock();
 164 }
 165
 166 void mem_cgroup_update_page_stat(struct page *page,
 167                                  enum mem_cgroup_page_stat_item idx,
 168                                  int val);
 169
 170 static inline void mem_cgroup_inc_page_stat(struct page *page,
 171                                             enum mem_cgroup_page_stat_item idx)
 172 {
 173         mem_cgroup_update_page_stat(page, idx, 1);
 174 }
 175
 176 static inline void mem_cgroup_dec_page_stat(struct page *page,
 177                                             enum mem_cgroup_page_stat_item idx)
 178 {
 179         mem_cgroup_update_page_stat(page, idx, -1);
 180 }
 181
 182 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
 183                                                 gfp_t gfp_mask,
 184                                                 unsigned long *total_scanned);
 185
 186 void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx);
 187 static inline void mem_cgroup_count_vm_event(struct mm_struct *mm,
 188                                              enum vm_event_item idx)
 189 {
 190         if (mem_cgroup_disabled())
 191                 return;
 192         __mem_cgroup_count_vm_event(mm, idx);
 193 }
 194 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 195 void mem_cgroup_split_huge_fixup(struct page *head);
 196 #endif
 197
 198 #ifdef CONFIG_DEBUG_VM
 199 bool mem_cgroup_bad_page_check(struct page *page);
 200 void mem_cgroup_print_bad_page(struct page *page);
 201 #endif
 202 #else /* CONFIG_MEMCG */
 203 struct mem_cgroup;
 204
 205 static inline int mem_cgroup_newpage_charge(struct page *page,
 206                                         struct mm_struct *mm, gfp_t gfp_mask)
 207 {
 208         return 0;
 209 }
 210
 211 static inline int mem_cgroup_cache_charge(struct page *page,
 212                                         struct mm_struct *mm, gfp_t gfp_mask)
 213 {
 214         return 0;
 215 }
 216
 217 static inline int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
 218                 struct page *page, gfp_t gfp_mask, struct mem_cgroup **memcgp)
 219 {
 220         return 0;
 221 }
 222
 223 static inline void mem_cgroup_commit_charge_swapin(struct page *page,
 224                                           struct mem_cgroup *memcg)
 225 {
 226 }
 227
 228 static inline void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
 229 {
 230 }
 231
 232 static inline void mem_cgroup_uncharge_start(void)
 233 {
 234 }
 235
 236 static inline void mem_cgroup_uncharge_end(void)
 237 {
 238 }
 239
 240 static inline void mem_cgroup_uncharge_page(struct page *page)
 241 {
 242 }
 243
 244 static inline void mem_cgroup_uncharge_cache_page(struct page *page)
 245 {
 246 }
 247
 248 static inline struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
 249                                                     struct mem_cgroup *memcg)
 250 {
 251         return &zone->lruvec;
 252 }
 253
 254 static inline struct lruvec *mem_cgroup_page_lruvec(struct page *page,
 255                                                     struct zone *zone)
 256 {
 257         return &zone->lruvec;
 258 }
 259
 260 static inline struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
 261 {
 262         return NULL;
 263 }
 264
 265 static inline struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
 266 {
 267         return NULL;
 268 }
 269
 270 static inline bool mm_match_cgroup(struct mm_struct *mm,
 271                 struct mem_cgroup *memcg)
 272 {
 273         return true;
 274 }
 275
 276 static inline int task_in_mem_cgroup(struct task_struct *task,
 277                                      const struct mem_cgroup *memcg)
 278 {
 279         return 1;
 280 }
 281
 282 static inline struct cgroup_subsys_state
 283                 *mem_cgroup_css(struct mem_cgroup *memcg)
 284 {
 285         return NULL;
 286 }
 287
 288 static inline void
 289 mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
 290                              struct mem_cgroup **memcgp)
 291 {
 292 }
 293
 294 static inline void mem_cgroup_end_migration(struct mem_cgroup *memcg,
 295                 struct page *oldpage, struct page *newpage, bool migration_ok)
 296 {
 297 }
 298
 299 static inline struct mem_cgroup *
 300 mem_cgroup_iter(struct mem_cgroup *root,
 301                 struct mem_cgroup *prev,
 302                 struct mem_cgroup_reclaim_cookie *reclaim)
 303 {
 304         return NULL;
 305 }
 306
 307 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
 308                                          struct mem_cgroup *prev)
 309 {
 310 }
 311
 312 static inline bool mem_cgroup_disabled(void)
 313 {
 314         return true;
 315 }
 316
 317 static inline int
 318 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
 319 {
 320         return 1;
 321 }
 322
 323 static inline unsigned long
 324 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
 325 {
 326         return 0;
 327 }
 328
 329 static inline void
 330 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
 331                               int increment)
 332 {
 333 }
 334
 335 static inline void
 336 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
 337 {
 338 }
 339
 340 static inline void mem_cgroup_begin_update_page_stat(struct page *page,
 341                                         bool *locked, unsigned long *flags)
 342 {
 343 }
 344
 345 static inline void mem_cgroup_end_update_page_stat(struct page *page,
 346                                         bool *locked, unsigned long *flags)
 347 {
 348 }
 349
 350 static inline void mem_cgroup_inc_page_stat(struct page *page,
 351                                             enum mem_cgroup_page_stat_item idx)
 352 {
 353 }
 354
 355 static inline void mem_cgroup_dec_page_stat(struct page *page,
 356                                             enum mem_cgroup_page_stat_item idx)
 357 {
 358 }
 359
 360 static inline
 361 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
 362                                             gfp_t gfp_mask,
 363                                             unsigned long *total_scanned)
 364 {
 365         return 0;
 366 }
 367
 368 static inline void mem_cgroup_split_huge_fixup(struct page *head)
 369 {
 370 }
 371
 372 static inline
 373 void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
 374 {
 375 }
 376 static inline void mem_cgroup_replace_page_cache(struct page *oldpage,
 377                                 struct page *newpage)
 378 {
 379 }
 380 #endif /* CONFIG_MEMCG */
 381
 382 #if !defined(CONFIG_MEMCG) || !defined(CONFIG_DEBUG_VM)
 383 static inline bool
 384 mem_cgroup_bad_page_check(struct page *page)
 385 {
 386         return false;
 387 }
 388
 389 static inline void
 390 mem_cgroup_print_bad_page(struct page *page)
 391 {
 392 }
 393 #endif
 394
 395 enum {
 396         UNDER_LIMIT,
 397         SOFT_LIMIT,
 398         OVER_LIMIT,
 399 };
 400
 401 struct sock;
 402 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
 403 void sock_update_memcg(struct sock *sk);
 404 void sock_release_memcg(struct sock *sk);
 405 #else
 406 static inline void sock_update_memcg(struct sock *sk)
 407 {
 408 }
 409 static inline void sock_release_memcg(struct sock *sk)
 410 {
 411 }
 412 #endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
 413
 414 #ifdef CONFIG_MEMCG_KMEM
 415 extern struct static_key memcg_kmem_enabled_key;
 416
 417 extern int memcg_limited_groups_array_size;
 418
 419 /*
 420  * Helper macro to loop through all memcg-specific caches. Callers must still
 421  * check if the cache is valid (it is either valid or NULL).
 422  * the slab_mutex must be held when looping through those caches
 423  */
 424 #define for_each_memcg_cache_index(_idx)        \
 425         for ((_idx) = 0; (_idx) < memcg_limited_groups_array_size; (_idx)++)
 426
 427 static inline bool memcg_kmem_enabled(void)
 428 {
 429         return static_key_false(&memcg_kmem_enabled_key);
 430 }
 431
 432 /*
 433  * In general, we'll do everything in our power to not incur in any overhead
 434  * for non-memcg users for the kmem functions. Not even a function call, if we
 435  * can avoid it.
 436  *
 437  * Therefore, we'll inline all those functions so that in the best case, we'll
 438  * see that kmemcg is off for everybody and proceed quickly.  If it is on,
 439  * we'll still do most of the flag checking inline. We check a lot of
 440  * conditions, but because they are pretty simple, they are expected to be
 441  * fast.
 442  */
 443 bool __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg,
 444                                         int order);
 445 void __memcg_kmem_commit_charge(struct page *page,
 446                                        struct mem_cgroup *memcg, int order);
 447 void __memcg_kmem_uncharge_pages(struct page *page, int order);
 448
 449 int memcg_cache_id(struct mem_cgroup *memcg);
 450 int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
 451                          struct kmem_cache *root_cache);
 452 void memcg_release_cache(struct kmem_cache *cachep);
 453 void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep);
 454
 455 int memcg_update_cache_size(struct kmem_cache *s, int num_groups);
 456 void memcg_update_array_size(int num_groups);
 457
 458 struct kmem_cache *
 459 __memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp);
 460
 461 void mem_cgroup_destroy_cache(struct kmem_cache *cachep);
 462 void kmem_cache_destroy_memcg_children(struct kmem_cache *s);
 463
 464 /**
 465  * memcg_kmem_newpage_charge: verify if a new kmem allocation is allowed.
 466  * @gfp: the gfp allocation flags.
 467  * @memcg: a pointer to the memcg this was charged against.
 468  * @order: allocation order.
 469  *
 470  * returns true if the memcg where the current task belongs can hold this
 471  * allocation.
 472  *
 473  * We return true automatically if this allocation is not to be accounted to
 474  * any memcg.
 475  */
 476 static inline bool
 477 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
 478 {
 479         if (!memcg_kmem_enabled())
 480                 return true;
 481
 482         /*
 483          * __GFP_NOFAIL allocations will move on even if charging is not
 484          * possible. Therefore we don't even try, and have this allocation
 485          * unaccounted. We could in theory charge it with
 486          * res_counter_charge_nofail, but we hope those allocations are rare,
 487          * and won't be worth the trouble.
 488          */
 489         if (!(gfp & __GFP_KMEMCG) || (gfp & __GFP_NOFAIL))
 490                 return true;
 491         if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
 492                 return true;
 493
 494         /* If the test is dying, just let it go. */
 495         if (unlikely(fatal_signal_pending(current)))
 496                 return true;
 497
 498         return __memcg_kmem_newpage_charge(gfp, memcg, order);
 499 }
 500
 501 /**
 502  * memcg_kmem_uncharge_pages: uncharge pages from memcg
 503  * @page: pointer to struct page being freed
 504  * @order: allocation order.
 505  *
 506  * there is no need to specify memcg here, since it is embedded in page_cgroup
 507  */
 508 static inline void
 509 memcg_kmem_uncharge_pages(struct page *page, int order)
 510 {
 511         if (memcg_kmem_enabled())
 512                 __memcg_kmem_uncharge_pages(page, order);
 513 }
 514
 515 /**
 516  * memcg_kmem_commit_charge: embeds correct memcg in a page
 517  * @page: pointer to struct page recently allocated
 518  * @memcg: the memcg structure we charged against
 519  * @order: allocation order.
 520  *
 521  * Needs to be called after memcg_kmem_newpage_charge, regardless of success or
 522  * failure of the allocation. if @page is NULL, this function will revert the
 523  * charges. Otherwise, it will commit the memcg given by @memcg to the
 524  * corresponding page_cgroup.
 525  */
 526 static inline void
 527 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
 528 {
 529         if (memcg_kmem_enabled() && memcg)
 530                 __memcg_kmem_commit_charge(page, memcg, order);
 531 }
 532
 533 /**
 534  * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
 535  * @cachep: the original global kmem cache
 536  * @gfp: allocation flags.
 537  *
 538  * This function assumes that the task allocating, which determines the memcg
 539  * in the page allocator, belongs to the same cgroup throughout the whole
 540  * process.  Misacounting can happen if the task calls memcg_kmem_get_cache()
 541  * while belonging to a cgroup, and later on changes. This is considered
 542  * acceptable, and should only happen upon task migration.
 543  *
 544  * Before the cache is created by the memcg core, there is also a possible
 545  * imbalance: the task belongs to a memcg, but the cache being allocated from
 546  * is the global cache, since the child cache is not yet guaranteed to be
 547  * ready. This case is also fine, since in this case the GFP_KMEMCG will not be
 548  * passed and the page allocator will not attempt any cgroup accounting.
 549  */
 550 static __always_inline struct kmem_cache *
 551 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
 552 {
 553         if (!memcg_kmem_enabled())
 554                 return cachep;
 555         if (gfp & __GFP_NOFAIL)
 556                 return cachep;
 557         if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
 558                 return cachep;
 559         if (unlikely(fatal_signal_pending(current)))
 560                 return cachep;
 561
 562         return __memcg_kmem_get_cache(cachep, gfp);
 563 }
 564 #else
 565 #define for_each_memcg_cache_index(_idx)        \
 566         for (; NULL; )
 567
 568 static inline bool memcg_kmem_enabled(void)
 569 {
 570         return false;
 571 }
 572
 573 static inline bool
 574 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
 575 {
 576         return true;
 577 }
 578
 579 static inline void memcg_kmem_uncharge_pages(struct page *page, int order)
 580 {
 581 }
 582
 583 static inline void
 584 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
 585 {
 586 }
 587
 588 static inline int memcg_cache_id(struct mem_cgroup *memcg)
 589 {
 590         return -1;
 591 }
 592
 593 static inline int
 594 memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
 595                      struct kmem_cache *root_cache)
 596 {
 597         return 0;
 598 }
 599
 600 static inline void memcg_release_cache(struct kmem_cache *cachep)
 601 {
 602 }
 603
 604 static inline void memcg_cache_list_add(struct mem_cgroup *memcg,
 605                                         struct kmem_cache *s)
 606 {
 607 }
 608
 609 static inline struct kmem_cache *
 610 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
 611 {
 612         return cachep;
 613 }
 614
 615 static inline void kmem_cache_destroy_memcg_children(struct kmem_cache *s)
 616 {
 617 }
 618 #endif /* CONFIG_MEMCG_KMEM */
 619 #endif /* _LINUX_MEMCONTROL_H */
 620