include/linux/mm_types.h

   1 #ifndef _LINUX_MM_TYPES_H
   2 #define _LINUX_MM_TYPES_H
   3
   4 #include <linux/auxvec.h>
   5 #include <linux/types.h>
   6 #include <linux/threads.h>
   7 #include <linux/list.h>
   8 #include <linux/spinlock.h>
   9 #include <linux/rbtree.h>
  10 #include <linux/rwsem.h>
  11 #include <linux/completion.h>
  12 #include <linux/cpumask.h>
  13 #include <linux/uprobes.h>
  14 #include <linux/page-flags-layout.h>
  15 #include <asm/page.h>
  16 #include <asm/mmu.h>
  17
  18 #ifndef AT_VECTOR_SIZE_ARCH
  19 #define AT_VECTOR_SIZE_ARCH 0
  20 #endif
  21 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
  22
  23 struct address_space;
  24 struct mem_cgroup;
  25
  26 #define USE_SPLIT_PTE_PTLOCKS   (NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS)
  27 #define USE_SPLIT_PMD_PTLOCKS   (USE_SPLIT_PTE_PTLOCKS && \
  28                 IS_ENABLED(CONFIG_ARCH_ENABLE_SPLIT_PMD_PTLOCK))
  29 #define ALLOC_SPLIT_PTLOCKS     (SPINLOCK_SIZE > BITS_PER_LONG/8)
  30
  31 /*
  32  * Each physical page in the system has a struct page associated with
  33  * it to keep track of whatever it is we are using the page for at the
  34  * moment. Note that we have no way to track which tasks are using
  35  * a page, though if it is a pagecache page, rmap structures can tell us
  36  * who is mapping it.
  37  *
  38  * The objects in struct page are organized in double word blocks in
  39  * order to allows us to use atomic double word operations on portions
  40  * of struct page. That is currently only used by slub but the arrangement
  41  * allows the use of atomic double word operations on the flags/mapping
  42  * and lru list pointers also.
  43  */
  44 struct page {
  45         /* First double word block */
  46         unsigned long flags;            /* Atomic flags, some possibly
  47                                          * updated asynchronously */
  48         union {
  49                 struct address_space *mapping;  /* If low bit clear, points to
  50                                                  * inode address_space, or NULL.
  51                                                  * If page mapped as anonymous
  52                                                  * memory, low bit is set, and
  53                                                  * it points to anon_vma object:
  54                                                  * see PAGE_MAPPING_ANON below.
  55                                                  */
  56                 void *s_mem;                    /* slab first object */
  57                 atomic_t compound_mapcount;     /* first tail page */
  58                 /* page_deferred_list().next     -- second tail page */
  59         };
  60
  61         /* Second double word */
  62         struct {
  63                 union {
  64                         pgoff_t index;          /* Our offset within mapping. */
  65                         void *freelist;         /* sl[aou]b first free object */
  66                         /* page_deferred_list().prev    -- second tail page */
  67                 };
  68
  69                 union {
  70 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
  71         defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
  72                         /* Used for cmpxchg_double in slub */
  73                         unsigned long counters;
  74 #else
  75                         /*
  76                          * Keep _count separate from slub cmpxchg_double data.
  77                          * As the rest of the double word is protected by
  78                          * slab_lock but _count is not.
  79                          */
  80                         unsigned counters;
  81 #endif
  82
  83                         struct {
  84
  85                                 union {
  86                                         /*
  87                                          * Count of ptes mapped in mms, to show
  88                                          * when page is mapped & limit reverse
  89                                          * map searches.
  90                                          */
  91                                         atomic_t _mapcount;
  92
  93                                         struct { /* SLUB */
  94                                                 unsigned inuse:16;
  95                                                 unsigned objects:15;
  96                                                 unsigned frozen:1;
  97                                         };
  98                                         int units;      /* SLOB */
  99                                 };
 100                                 atomic_t _count;                /* Usage count, see below. */
 101                         };
 102                         unsigned int active;    /* SLAB */
 103                 };
 104         };
 105
 106         /*
 107          * Third double word block
 108          *
 109          * WARNING: bit 0 of the first word encode PageTail(). That means
 110          * the rest users of the storage space MUST NOT use the bit to
 111          * avoid collision and false-positive PageTail().
 112          */
 113         union {
 114                 struct list_head lru;   /* Pageout list, eg. active_list
 115                                          * protected by zone->lru_lock !
 116                                          * Can be used as a generic list
 117                                          * by the page owner.
 118                                          */
 119                 struct dev_pagemap *pgmap; /* ZONE_DEVICE pages are never on an
 120                                             * lru or handled by a slab
 121                                             * allocator, this points to the
 122                                             * hosting device page map.
 123                                             */
 124                 struct {                /* slub per cpu partial pages */
 125                         struct page *next;      /* Next partial slab */
 126 #ifdef CONFIG_64BIT
 127                         int pages;      /* Nr of partial slabs left */
 128                         int pobjects;   /* Approximate # of objects */
 129 #else
 130                         short int pages;
 131                         short int pobjects;
 132 #endif
 133                 };
 134
 135                 struct rcu_head rcu_head;       /* Used by SLAB
 136                                                  * when destroying via RCU
 137                                                  */
 138                 /* Tail pages of compound page */
 139                 struct {
 140                         unsigned long compound_head; /* If bit zero is set */
 141
 142                         /* First tail page only */
 143 #ifdef CONFIG_64BIT
 144                         /*
 145                          * On 64 bit system we have enough space in struct page
 146                          * to encode compound_dtor and compound_order with
 147                          * unsigned int. It can help compiler generate better or
 148                          * smaller code on some archtectures.
 149                          */
 150                         unsigned int compound_dtor;
 151                         unsigned int compound_order;
 152 #else
 153                         unsigned short int compound_dtor;
 154                         unsigned short int compound_order;
 155 #endif
 156                 };
 157
 158 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
 159                 struct {
 160                         unsigned long __pad;    /* do not overlay pmd_huge_pte
 161                                                  * with compound_head to avoid
 162                                                  * possible bit 0 collision.
 163                                                  */
 164                         pgtable_t pmd_huge_pte; /* protected by page->ptl */
 165                 };
 166 #endif
 167         };
 168
 169         /* Remainder is not double word aligned */
 170         union {
 171                 unsigned long private;          /* Mapping-private opaque data:
 172                                                  * usually used for buffer_heads
 173                                                  * if PagePrivate set; used for
 174                                                  * swp_entry_t if PageSwapCache;
 175                                                  * indicates order in the buddy
 176                                                  * system if PG_buddy is set.
 177                                                  */
 178 #if USE_SPLIT_PTE_PTLOCKS
 179 #if ALLOC_SPLIT_PTLOCKS
 180                 spinlock_t *ptl;
 181 #else
 182                 spinlock_t ptl;
 183 #endif
 184 #endif
 185                 struct kmem_cache *slab_cache;  /* SL[AU]B: Pointer to slab */
 186         };
 187
 188 #ifdef CONFIG_MEMCG
 189         struct mem_cgroup *mem_cgroup;
 190 #endif
 191
 192         /*
 193          * On machines where all RAM is mapped into kernel address space,
 194          * we can simply calculate the virtual address. On machines with
 195          * highmem some memory is mapped into kernel virtual memory
 196          * dynamically, so we need a place to store that address.
 197          * Note that this field could be 16 bits on x86 ... ;)
 198          *
 199          * Architectures with slow multiplication can define
 200          * WANT_PAGE_VIRTUAL in asm/page.h
 201          */
 202 #if defined(WANT_PAGE_VIRTUAL)
 203         void *virtual;                  /* Kernel virtual address (NULL if
 204                                            not kmapped, ie. highmem) */
 205 #endif /* WANT_PAGE_VIRTUAL */
 206
 207 #ifdef CONFIG_KMEMCHECK
 208         /*
 209          * kmemcheck wants to track the status of each byte in a page; this
 210          * is a pointer to such a status block. NULL if not tracked.
 211          */
 212         void *shadow;
 213 #endif
 214
 215 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 216         int _last_cpupid;
 217 #endif
 218 }
 219 /*
 220  * The struct page can be forced to be double word aligned so that atomic ops
 221  * on double words work. The SLUB allocator can make use of such a feature.
 222  */
 223 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
 224         __aligned(2 * sizeof(unsigned long))
 225 #endif
 226 ;
 227
 228 struct page_frag {
 229         struct page *page;
 230 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 231         __u32 offset;
 232         __u32 size;
 233 #else
 234         __u16 offset;
 235         __u16 size;
 236 #endif
 237 };
 238
 239 #define PAGE_FRAG_CACHE_MAX_SIZE        __ALIGN_MASK(32768, ~PAGE_MASK)
 240 #define PAGE_FRAG_CACHE_MAX_ORDER       get_order(PAGE_FRAG_CACHE_MAX_SIZE)
 241
 242 struct page_frag_cache {
 243         void * va;
 244 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
 245         __u16 offset;
 246         __u16 size;
 247 #else
 248         __u32 offset;
 249 #endif
 250         /* we maintain a pagecount bias, so that we dont dirty cache line
 251          * containing page->_count every time we allocate a fragment.
 252          */
 253         unsigned int            pagecnt_bias;
 254         bool pfmemalloc;
 255 };
 256
 257 typedef unsigned long vm_flags_t;
 258
 259 /*
 260  * A region containing a mapping of a non-memory backed file under NOMMU
 261  * conditions.  These are held in a global tree and are pinned by the VMAs that
 262  * map parts of them.
 263  */
 264 struct vm_region {
 265         struct rb_node  vm_rb;          /* link in global region tree */
 266         vm_flags_t      vm_flags;       /* VMA vm_flags */
 267         unsigned long   vm_start;       /* start address of region */
 268         unsigned long   vm_end;         /* region initialised to here */
 269         unsigned long   vm_top;         /* region allocated to here */
 270         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 271         struct file     *vm_file;       /* the backing file or NULL */
 272
 273         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 274         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 275                                                 * this region */
 276 };
 277
 278 #ifdef CONFIG_USERFAULTFD
 279 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
 280 struct vm_userfaultfd_ctx {
 281         struct userfaultfd_ctx *ctx;
 282 };
 283 #else /* CONFIG_USERFAULTFD */
 284 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
 285 struct vm_userfaultfd_ctx {};
 286 #endif /* CONFIG_USERFAULTFD */
 287
 288 /*
 289  * This struct defines a memory VMM memory area. There is one of these
 290  * per VM-area/task.  A VM area is any part of the process virtual memory
 291  * space that has a special rule for the page-fault handlers (ie a shared
 292  * library, the executable area etc).
 293  */
 294 struct vm_area_struct {
 295         /* The first cache line has the info for VMA tree walking. */
 296
 297         unsigned long vm_start;         /* Our start address within vm_mm. */
 298         unsigned long vm_end;           /* The first byte after our end address
 299                                            within vm_mm. */
 300
 301         /* linked list of VM areas per task, sorted by address */
 302         struct vm_area_struct *vm_next, *vm_prev;
 303
 304         struct rb_node vm_rb;
 305
 306         /*
 307          * Largest free memory gap in bytes to the left of this VMA.
 308          * Either between this VMA and vma->vm_prev, or between one of the
 309          * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
 310          * get_unmapped_area find a free area of the right size.
 311          */
 312         unsigned long rb_subtree_gap;
 313
 314         /* Second cache line starts here. */
 315
 316         struct mm_struct *vm_mm;        /* The address space we belong to. */
 317         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 318         unsigned long vm_flags;         /* Flags, see mm.h. */
 319
 320         /*
 321          * For areas with an address space and backing store,
 322          * linkage into the address_space->i_mmap interval tree.
 323          */
 324         struct {
 325                 struct rb_node rb;
 326                 unsigned long rb_subtree_last;
 327         } shared;
 328
 329         /*
 330          * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 331          * list, after a COW of one of the file pages.  A MAP_SHARED vma
 332          * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 333          * or brk vma (with NULL file) can only be in an anon_vma list.
 334          */
 335         struct list_head anon_vma_chain; /* Serialized by mmap_sem &
 336                                           * page_table_lock */
 337         struct anon_vma *anon_vma;      /* Serialized by page_table_lock */
 338
 339         /* Function pointers to deal with this struct. */
 340         const struct vm_operations_struct *vm_ops;
 341
 342         /* Information about our backing store: */
 343         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 344                                            units */
 345         struct file * vm_file;          /* File we map to (can be NULL). */
 346         void * vm_private_data;         /* was vm_pte (shared mem) */
 347
 348 #ifndef CONFIG_MMU
 349         struct vm_region *vm_region;    /* NOMMU mapping region */
 350 #endif
 351 #ifdef CONFIG_NUMA
 352         struct mempolicy *vm_policy;    /* NUMA policy for the VMA */
 353 #endif
 354         struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
 355 };
 356
 357 struct core_thread {
 358         struct task_struct *task;
 359         struct core_thread *next;
 360 };
 361
 362 struct core_state {
 363         atomic_t nr_threads;
 364         struct core_thread dumper;
 365         struct completion startup;
 366 };
 367
 368 enum {
 369         MM_FILEPAGES,   /* Resident file mapping pages */
 370         MM_ANONPAGES,   /* Resident anonymous pages */
 371         MM_SWAPENTS,    /* Anonymous swap entries */
 372         MM_SHMEMPAGES,  /* Resident shared memory pages */
 373         NR_MM_COUNTERS
 374 };
 375
 376 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
 377 #define SPLIT_RSS_COUNTING
 378 /* per-thread cached information, */
 379 struct task_rss_stat {
 380         int events;     /* for synchronization threshold */
 381         int count[NR_MM_COUNTERS];
 382 };
 383 #endif /* USE_SPLIT_PTE_PTLOCKS */
 384
 385 struct mm_rss_stat {
 386         atomic_long_t count[NR_MM_COUNTERS];
 387 };
 388
 389 struct kioctx_table;
 390 struct mm_struct {
 391         struct vm_area_struct *mmap;            /* list of VMAs */
 392         struct rb_root mm_rb;
 393         u32 vmacache_seqnum;                   /* per-thread vmacache */
 394 #ifdef CONFIG_MMU
 395         unsigned long (*get_unmapped_area) (struct file *filp,
 396                                 unsigned long addr, unsigned long len,
 397                                 unsigned long pgoff, unsigned long flags);
 398 #endif
 399         unsigned long mmap_base;                /* base of mmap area */
 400         unsigned long mmap_legacy_base;         /* base of mmap area in bottom-up allocations */
 401         unsigned long task_size;                /* size of task vm space */
 402         unsigned long highest_vm_end;           /* highest vma end address */
 403         pgd_t * pgd;
 404         atomic_t mm_users;                      /* How many users with user space? */
 405         atomic_t mm_count;                      /* How many references to "struct mm_struct" (users count as 1) */
 406         atomic_long_t nr_ptes;                  /* PTE page table pages */
 407 #if CONFIG_PGTABLE_LEVELS > 2
 408         atomic_long_t nr_pmds;                  /* PMD page table pages */
 409 #endif
 410         int map_count;                          /* number of VMAs */
 411
 412         spinlock_t page_table_lock;             /* Protects page tables and some counters */
 413         struct rw_semaphore mmap_sem;
 414
 415         struct list_head mmlist;                /* List of maybe swapped mm's.  These are globally strung
 416                                                  * together off init_mm.mmlist, and are protected
 417                                                  * by mmlist_lock
 418                                                  */
 419
 420
 421         unsigned long hiwater_rss;      /* High-watermark of RSS usage */
 422         unsigned long hiwater_vm;       /* High-water virtual memory usage */
 423
 424         unsigned long total_vm;         /* Total pages mapped */
 425         unsigned long locked_vm;        /* Pages that have PG_mlocked set */
 426         unsigned long pinned_vm;        /* Refcount permanently increased */
 427         unsigned long data_vm;          /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
 428         unsigned long exec_vm;          /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
 429         unsigned long stack_vm;         /* VM_STACK */
 430         unsigned long def_flags;
 431         unsigned long start_code, end_code, start_data, end_data;
 432         unsigned long start_brk, brk, start_stack;
 433         unsigned long arg_start, arg_end, env_start, env_end;
 434
 435         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 436
 437         /*
 438          * Special counters, in some configurations protected by the
 439          * page_table_lock, in other configurations by being atomic.
 440          */
 441         struct mm_rss_stat rss_stat;
 442
 443         struct linux_binfmt *binfmt;
 444
 445         cpumask_var_t cpu_vm_mask_var;
 446
 447         /* Architecture-specific MM context */
 448         mm_context_t context;
 449
 450         unsigned long flags; /* Must use atomic bitops to access the bits */
 451
 452         struct core_state *core_state; /* coredumping support */
 453 #ifdef CONFIG_AIO
 454         spinlock_t                      ioctx_lock;
 455         struct kioctx_table __rcu       *ioctx_table;
 456 #endif
 457 #ifdef CONFIG_MEMCG
 458         /*
 459          * "owner" points to a task that is regarded as the canonical
 460          * user/owner of this mm. All of the following must be true in
 461          * order for it to be changed:
 462          *
 463          * current == mm->owner
 464          * current->mm != mm
 465          * new_owner->mm == mm
 466          * new_owner->alloc_lock is held
 467          */
 468         struct task_struct __rcu *owner;
 469 #endif
 470
 471         /* store ref to file /proc/<pid>/exe symlink points to */
 472         struct file __rcu *exe_file;
 473 #ifdef CONFIG_MMU_NOTIFIER
 474         struct mmu_notifier_mm *mmu_notifier_mm;
 475 #endif
 476 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 477         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 478 #endif
 479 #ifdef CONFIG_CPUMASK_OFFSTACK
 480         struct cpumask cpumask_allocation;
 481 #endif
 482 #ifdef CONFIG_NUMA_BALANCING
 483         /*
 484          * numa_next_scan is the next time that the PTEs will be marked
 485          * pte_numa. NUMA hinting faults will gather statistics and migrate
 486          * pages to new nodes if necessary.
 487          */
 488         unsigned long numa_next_scan;
 489
 490         /* Restart point for scanning and setting pte_numa */
 491         unsigned long numa_scan_offset;
 492
 493         /* numa_scan_seq prevents two threads setting pte_numa */
 494         int numa_scan_seq;
 495 #endif
 496 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 497         /*
 498          * An operation with batched TLB flushing is going on. Anything that
 499          * can move process memory needs to flush the TLB when moving a
 500          * PROT_NONE or PROT_NUMA mapped page.
 501          */
 502         bool tlb_flush_pending;
 503 #endif
 504         struct uprobes_state uprobes_state;
 505 #ifdef CONFIG_X86_INTEL_MPX
 506         /* address of the bounds directory */
 507         void __user *bd_addr;
 508 #endif
 509 #ifdef CONFIG_HUGETLB_PAGE
 510         atomic_long_t hugetlb_usage;
 511 #endif
 512 };
 513
 514 static inline void mm_init_cpumask(struct mm_struct *mm)
 515 {
 516 #ifdef CONFIG_CPUMASK_OFFSTACK
 517         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 518 #endif
 519         cpumask_clear(mm->cpu_vm_mask_var);
 520 }
 521
 522 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 523 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 524 {
 525         return mm->cpu_vm_mask_var;
 526 }
 527
 528 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 529 /*
 530  * Memory barriers to keep this state in sync are graciously provided by
 531  * the page table locks, outside of which no page table modifications happen.
 532  * The barriers below prevent the compiler from re-ordering the instructions
 533  * around the memory barriers that are already present in the code.
 534  */
 535 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 536 {
 537         barrier();
 538         return mm->tlb_flush_pending;
 539 }
 540 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 541 {
 542         mm->tlb_flush_pending = true;
 543
 544         /*
 545          * Guarantee that the tlb_flush_pending store does not leak into the
 546          * critical section updating the page tables
 547          */
 548         smp_mb__before_spinlock();
 549 }
 550 /* Clearing is done after a TLB flush, which also provides a barrier. */
 551 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 552 {
 553         barrier();
 554         mm->tlb_flush_pending = false;
 555 }
 556 #else
 557 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 558 {
 559         return false;
 560 }
 561 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 562 {
 563 }
 564 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 565 {
 566 }
 567 #endif
 568
 569 struct vm_fault;
 570
 571 struct vm_special_mapping {
 572         const char *name;       /* The name, e.g. "[vdso]". */
 573
 574         /*
 575          * If .fault is not provided, this points to a
 576          * NULL-terminated array of pages that back the special mapping.
 577          *
 578          * This must not be NULL unless .fault is provided.
 579          */
 580         struct page **pages;
 581
 582         /*
 583          * If non-NULL, then this is called to resolve page faults
 584          * on the special mapping.  If used, .pages is not checked.
 585          */
 586         int (*fault)(const struct vm_special_mapping *sm,
 587                      struct vm_area_struct *vma,
 588                      struct vm_fault *vmf);
 589 };
 590
 591 enum tlb_flush_reason {
 592         TLB_FLUSH_ON_TASK_SWITCH,
 593         TLB_REMOTE_SHOOTDOWN,
 594         TLB_LOCAL_SHOOTDOWN,
 595         TLB_LOCAL_MM_SHOOTDOWN,
 596         TLB_REMOTE_SEND_IPI,
 597         NR_TLB_FLUSH_REASONS,
 598 };
 599
 600  /*
 601   * A swap entry has to fit into a "unsigned long", as the entry is hidden
 602   * in the "index" field of the swapper address space.
 603   */
 604 typedef struct {
 605         unsigned long val;
 606 } swp_entry_t;
 607
 608 #endif /* _LINUX_MM_TYPES_H */