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_inactive_file_is_low(struct lruvec *lruvec);
 120 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg);
 121 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list);
 122 void mem_cgroup_update_lru_size(struct lruvec *, enum lru_list, int);
 123 extern void mem_cgroup_print_oom_info(struct mem_cgroup *memcg,
 124                                         struct task_struct *p);
 125 extern void mem_cgroup_replace_page_cache(struct page *oldpage,
 126                                         struct page *newpage);
 127
 128 #ifdef CONFIG_MEMCG_SWAP
 129 extern int do_swap_account;
 130 #endif
 131
 132 static inline bool mem_cgroup_disabled(void)
 133 {
 134         if (mem_cgroup_subsys.disabled)
 135                 return true;
 136         return false;
 137 }
 138
 139 void __mem_cgroup_begin_update_page_stat(struct page *page, bool *locked,
 140                                          unsigned long *flags);
 141
 142 extern atomic_t memcg_moving;
 143
 144 static inline void mem_cgroup_begin_update_page_stat(struct page *page,
 145                                         bool *locked, unsigned long *flags)
 146 {
 147         if (mem_cgroup_disabled())
 148                 return;
 149         rcu_read_lock();
 150         *locked = false;
 151         if (atomic_read(&memcg_moving))
 152                 __mem_cgroup_begin_update_page_stat(page, locked, flags);
 153 }
 154
 155 void __mem_cgroup_end_update_page_stat(struct page *page,
 156                                 unsigned long *flags);
 157 static inline void mem_cgroup_end_update_page_stat(struct page *page,
 158                                         bool *locked, unsigned long *flags)
 159 {
 160         if (mem_cgroup_disabled())
 161                 return;
 162         if (*locked)
 163                 __mem_cgroup_end_update_page_stat(page, flags);
 164         rcu_read_unlock();
 165 }
 166
 167 void mem_cgroup_update_page_stat(struct page *page,
 168                                  enum mem_cgroup_page_stat_item idx,
 169                                  int val);
 170
 171 static inline void mem_cgroup_inc_page_stat(struct page *page,
 172                                             enum mem_cgroup_page_stat_item idx)
 173 {
 174         mem_cgroup_update_page_stat(page, idx, 1);
 175 }
 176
 177 static inline void mem_cgroup_dec_page_stat(struct page *page,
 178                                             enum mem_cgroup_page_stat_item idx)
 179 {
 180         mem_cgroup_update_page_stat(page, 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 #ifdef CONFIG_DEBUG_VM
 200 bool mem_cgroup_bad_page_check(struct page *page);
 201 void mem_cgroup_print_bad_page(struct page *page);
 202 #endif
 203 #else /* CONFIG_MEMCG */
 204 struct mem_cgroup;
 205
 206 static inline int mem_cgroup_newpage_charge(struct page *page,
 207                                         struct mm_struct *mm, gfp_t gfp_mask)
 208 {
 209         return 0;
 210 }
 211
 212 static inline int mem_cgroup_cache_charge(struct page *page,
 213                                         struct mm_struct *mm, gfp_t gfp_mask)
 214 {
 215         return 0;
 216 }
 217
 218 static inline int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
 219                 struct page *page, gfp_t gfp_mask, struct mem_cgroup **memcgp)
 220 {
 221         return 0;
 222 }
 223
 224 static inline void mem_cgroup_commit_charge_swapin(struct page *page,
 225                                           struct mem_cgroup *memcg)
 226 {
 227 }
 228
 229 static inline void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
 230 {
 231 }
 232
 233 static inline void mem_cgroup_uncharge_start(void)
 234 {
 235 }
 236
 237 static inline void mem_cgroup_uncharge_end(void)
 238 {
 239 }
 240
 241 static inline void mem_cgroup_uncharge_page(struct page *page)
 242 {
 243 }
 244
 245 static inline void mem_cgroup_uncharge_cache_page(struct page *page)
 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 struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
 267 {
 268         return NULL;
 269 }
 270
 271 static inline bool mm_match_cgroup(struct mm_struct *mm,
 272                 struct mem_cgroup *memcg)
 273 {
 274         return true;
 275 }
 276
 277 static inline int task_in_mem_cgroup(struct task_struct *task,
 278                                      const struct mem_cgroup *memcg)
 279 {
 280         return 1;
 281 }
 282
 283 static inline struct cgroup_subsys_state
 284                 *mem_cgroup_css(struct mem_cgroup *memcg)
 285 {
 286         return NULL;
 287 }
 288
 289 static inline void
 290 mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
 291                              struct mem_cgroup **memcgp)
 292 {
 293 }
 294
 295 static inline void mem_cgroup_end_migration(struct mem_cgroup *memcg,
 296                 struct page *oldpage, struct page *newpage, bool migration_ok)
 297 {
 298 }
 299
 300 static inline struct mem_cgroup *
 301 mem_cgroup_iter(struct mem_cgroup *root,
 302                 struct mem_cgroup *prev,
 303                 struct mem_cgroup_reclaim_cookie *reclaim)
 304 {
 305         return NULL;
 306 }
 307
 308 static inline void mem_cgroup_iter_break(struct mem_cgroup *root,
 309                                          struct mem_cgroup *prev)
 310 {
 311 }
 312
 313 static inline bool mem_cgroup_disabled(void)
 314 {
 315         return true;
 316 }
 317
 318 static inline int
 319 mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
 320 {
 321         return 1;
 322 }
 323
 324 static inline int
 325 mem_cgroup_inactive_file_is_low(struct lruvec *lruvec)
 326 {
 327         return 1;
 328 }
 329
 330 static inline unsigned long
 331 mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
 332 {
 333         return 0;
 334 }
 335
 336 static inline void
 337 mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
 338                               int increment)
 339 {
 340 }
 341
 342 static inline void
 343 mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
 344 {
 345 }
 346
 347 static inline void mem_cgroup_begin_update_page_stat(struct page *page,
 348                                         bool *locked, unsigned long *flags)
 349 {
 350 }
 351
 352 static inline void mem_cgroup_end_update_page_stat(struct page *page,
 353                                         bool *locked, unsigned long *flags)
 354 {
 355 }
 356
 357 static inline void mem_cgroup_inc_page_stat(struct page *page,
 358                                             enum mem_cgroup_page_stat_item idx)
 359 {
 360 }
 361
 362 static inline void mem_cgroup_dec_page_stat(struct page *page,
 363                                             enum mem_cgroup_page_stat_item 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 static inline void mem_cgroup_replace_page_cache(struct page *oldpage,
 384                                 struct page *newpage)
 385 {
 386 }
 387 #endif /* CONFIG_MEMCG */
 388
 389 #if !defined(CONFIG_MEMCG) || !defined(CONFIG_DEBUG_VM)
 390 static inline bool
 391 mem_cgroup_bad_page_check(struct page *page)
 392 {
 393         return false;
 394 }
 395
 396 static inline void
 397 mem_cgroup_print_bad_page(struct page *page)
 398 {
 399 }
 400 #endif
 401
 402 enum {
 403         UNDER_LIMIT,
 404         SOFT_LIMIT,
 405         OVER_LIMIT,
 406 };
 407
 408 struct sock;
 409 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
 410 void sock_update_memcg(struct sock *sk);
 411 void sock_release_memcg(struct sock *sk);
 412 #else
 413 static inline void sock_update_memcg(struct sock *sk)
 414 {
 415 }
 416 static inline void sock_release_memcg(struct sock *sk)
 417 {
 418 }
 419 #endif /* CONFIG_INET && CONFIG_MEMCG_KMEM */
 420
 421 #ifdef CONFIG_MEMCG_KMEM
 422 extern struct static_key memcg_kmem_enabled_key;
 423
 424 extern int memcg_limited_groups_array_size;
 425
 426 /*
 427  * Helper macro to loop through all memcg-specific caches. Callers must still
 428  * check if the cache is valid (it is either valid or NULL).
 429  * the slab_mutex must be held when looping through those caches
 430  */
 431 #define for_each_memcg_cache_index(_idx)        \
 432         for ((_idx) = 0; i < memcg_limited_groups_array_size; (_idx)++)
 433
 434 static inline bool memcg_kmem_enabled(void)
 435 {
 436         return static_key_false(&memcg_kmem_enabled_key);
 437 }
 438
 439 /*
 440  * In general, we'll do everything in our power to not incur in any overhead
 441  * for non-memcg users for the kmem functions. Not even a function call, if we
 442  * can avoid it.
 443  *
 444  * Therefore, we'll inline all those functions so that in the best case, we'll
 445  * see that kmemcg is off for everybody and proceed quickly.  If it is on,
 446  * we'll still do most of the flag checking inline. We check a lot of
 447  * conditions, but because they are pretty simple, they are expected to be
 448  * fast.
 449  */
 450 bool __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg,
 451                                         int order);
 452 void __memcg_kmem_commit_charge(struct page *page,
 453                                        struct mem_cgroup *memcg, int order);
 454 void __memcg_kmem_uncharge_pages(struct page *page, int order);
 455
 456 int memcg_cache_id(struct mem_cgroup *memcg);
 457 int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
 458                          struct kmem_cache *root_cache);
 459 void memcg_release_cache(struct kmem_cache *cachep);
 460 void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep);
 461
 462 int memcg_update_cache_size(struct kmem_cache *s, int num_groups);
 463 void memcg_update_array_size(int num_groups);
 464
 465 struct kmem_cache *
 466 __memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp);
 467
 468 void mem_cgroup_destroy_cache(struct kmem_cache *cachep);
 469 void kmem_cache_destroy_memcg_children(struct kmem_cache *s);
 470
 471 /**
 472  * memcg_kmem_newpage_charge: verify if a new kmem allocation is allowed.
 473  * @gfp: the gfp allocation flags.
 474  * @memcg: a pointer to the memcg this was charged against.
 475  * @order: allocation order.
 476  *
 477  * returns true if the memcg where the current task belongs can hold this
 478  * allocation.
 479  *
 480  * We return true automatically if this allocation is not to be accounted to
 481  * any memcg.
 482  */
 483 static inline bool
 484 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
 485 {
 486         if (!memcg_kmem_enabled())
 487                 return true;
 488
 489         /*
 490          * __GFP_NOFAIL allocations will move on even if charging is not
 491          * possible. Therefore we don't even try, and have this allocation
 492          * unaccounted. We could in theory charge it with
 493          * res_counter_charge_nofail, but we hope those allocations are rare,
 494          * and won't be worth the trouble.
 495          */
 496         if (!(gfp & __GFP_KMEMCG) || (gfp & __GFP_NOFAIL))
 497                 return true;
 498         if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
 499                 return true;
 500
 501         /* If the test is dying, just let it go. */
 502         if (unlikely(fatal_signal_pending(current)))
 503                 return true;
 504
 505         return __memcg_kmem_newpage_charge(gfp, memcg, order);
 506 }
 507
 508 /**
 509  * memcg_kmem_uncharge_pages: uncharge pages from memcg
 510  * @page: pointer to struct page being freed
 511  * @order: allocation order.
 512  *
 513  * there is no need to specify memcg here, since it is embedded in page_cgroup
 514  */
 515 static inline void
 516 memcg_kmem_uncharge_pages(struct page *page, int order)
 517 {
 518         if (memcg_kmem_enabled())
 519                 __memcg_kmem_uncharge_pages(page, order);
 520 }
 521
 522 /**
 523  * memcg_kmem_commit_charge: embeds correct memcg in a page
 524  * @page: pointer to struct page recently allocated
 525  * @memcg: the memcg structure we charged against
 526  * @order: allocation order.
 527  *
 528  * Needs to be called after memcg_kmem_newpage_charge, regardless of success or
 529  * failure of the allocation. if @page is NULL, this function will revert the
 530  * charges. Otherwise, it will commit the memcg given by @memcg to the
 531  * corresponding page_cgroup.
 532  */
 533 static inline void
 534 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
 535 {
 536         if (memcg_kmem_enabled() && memcg)
 537                 __memcg_kmem_commit_charge(page, memcg, order);
 538 }
 539
 540 /**
 541  * memcg_kmem_get_cache: selects the correct per-memcg cache for allocation
 542  * @cachep: the original global kmem cache
 543  * @gfp: allocation flags.
 544  *
 545  * This function assumes that the task allocating, which determines the memcg
 546  * in the page allocator, belongs to the same cgroup throughout the whole
 547  * process.  Misacounting can happen if the task calls memcg_kmem_get_cache()
 548  * while belonging to a cgroup, and later on changes. This is considered
 549  * acceptable, and should only happen upon task migration.
 550  *
 551  * Before the cache is created by the memcg core, there is also a possible
 552  * imbalance: the task belongs to a memcg, but the cache being allocated from
 553  * is the global cache, since the child cache is not yet guaranteed to be
 554  * ready. This case is also fine, since in this case the GFP_KMEMCG will not be
 555  * passed and the page allocator will not attempt any cgroup accounting.
 556  */
 557 static __always_inline struct kmem_cache *
 558 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
 559 {
 560         if (!memcg_kmem_enabled())
 561                 return cachep;
 562         if (gfp & __GFP_NOFAIL)
 563                 return cachep;
 564         if (in_interrupt() || (!current->mm) || (current->flags & PF_KTHREAD))
 565                 return cachep;
 566         if (unlikely(fatal_signal_pending(current)))
 567                 return cachep;
 568
 569         return __memcg_kmem_get_cache(cachep, gfp);
 570 }
 571 #else
 572 #define for_each_memcg_cache_index(_idx)        \
 573         for (; NULL; )
 574
 575 static inline bool memcg_kmem_enabled(void)
 576 {
 577         return false;
 578 }
 579
 580 static inline bool
 581 memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **memcg, int order)
 582 {
 583         return true;
 584 }
 585
 586 static inline void memcg_kmem_uncharge_pages(struct page *page, int order)
 587 {
 588 }
 589
 590 static inline void
 591 memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, int order)
 592 {
 593 }
 594
 595 static inline int memcg_cache_id(struct mem_cgroup *memcg)
 596 {
 597         return -1;
 598 }
 599
 600 static inline int
 601 memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s,
 602                      struct kmem_cache *root_cache)
 603 {
 604         return 0;
 605 }
 606
 607 static inline void memcg_release_cache(struct kmem_cache *cachep)
 608 {
 609 }
 610
 611 static inline void memcg_cache_list_add(struct mem_cgroup *memcg,
 612                                         struct kmem_cache *s)
 613 {
 614 }
 615
 616 static inline struct kmem_cache *
 617 memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp)
 618 {
 619         return cachep;
 620 }
 621
 622 static inline void kmem_cache_destroy_memcg_children(struct kmem_cache *s)
 623 {
 624 }
 625 #endif /* CONFIG_MEMCG_KMEM */
 626 #endif /* _LINUX_MEMCONTROL_H */
 627