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