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 #ifdef CONFIG_64BIT
 147                         /*
 148                          * On 64 bit system we have enough space in struct page
 149                          * to encode compound_dtor and compound_order with
 150                          * unsigned int. It can help compiler generate better or
 151                          * smaller code on some archtectures.
 152                          */
 153                         unsigned int compound_dtor;
 154                         unsigned int compound_order;
 155 #else
 156                         unsigned short int compound_dtor;
 157                         unsigned short int compound_order;
 158 #endif
 159                 };
 160
 161 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
 162                 struct {
 163                         unsigned long __pad;    /* do not overlay pmd_huge_pte
 164                                                  * with compound_head to avoid
 165                                                  * possible bit 0 collision.
 166                                                  */
 167                         pgtable_t pmd_huge_pte; /* protected by page->ptl */
 168                 };
 169 #endif
 170         };
 171
 172         /* Remainder is not double word aligned */
 173         union {
 174                 unsigned long private;          /* Mapping-private opaque data:
 175                                                  * usually used for buffer_heads
 176                                                  * if PagePrivate set; used for
 177                                                  * swp_entry_t if PageSwapCache;
 178                                                  * indicates order in the buddy
 179                                                  * system if PG_buddy is set.
 180                                                  */
 181 #if USE_SPLIT_PTE_PTLOCKS
 182 #if ALLOC_SPLIT_PTLOCKS
 183                 spinlock_t *ptl;
 184 #else
 185                 spinlock_t ptl;
 186 #endif
 187 #endif
 188                 struct kmem_cache *slab_cache;  /* SL[AU]B: Pointer to slab */
 189         };
 190
 191 #ifdef CONFIG_MEMCG
 192         struct mem_cgroup *mem_cgroup;
 193 #endif
 194
 195         /*
 196          * On machines where all RAM is mapped into kernel address space,
 197          * we can simply calculate the virtual address. On machines with
 198          * highmem some memory is mapped into kernel virtual memory
 199          * dynamically, so we need a place to store that address.
 200          * Note that this field could be 16 bits on x86 ... ;)
 201          *
 202          * Architectures with slow multiplication can define
 203          * WANT_PAGE_VIRTUAL in asm/page.h
 204          */
 205 #if defined(WANT_PAGE_VIRTUAL)
 206         void *virtual;                  /* Kernel virtual address (NULL if
 207                                            not kmapped, ie. highmem) */
 208 #endif /* WANT_PAGE_VIRTUAL */
 209
 210 #ifdef CONFIG_KMEMCHECK
 211         /*
 212          * kmemcheck wants to track the status of each byte in a page; this
 213          * is a pointer to such a status block. NULL if not tracked.
 214          */
 215         void *shadow;
 216 #endif
 217
 218 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 219         int _last_cpupid;
 220 #endif
 221 }
 222 /*
 223  * The struct page can be forced to be double word aligned so that atomic ops
 224  * on double words work. The SLUB allocator can make use of such a feature.
 225  */
 226 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
 227         __aligned(2 * sizeof(unsigned long))
 228 #endif
 229 ;
 230
 231 struct page_frag {
 232         struct page *page;
 233 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 234         __u32 offset;
 235         __u32 size;
 236 #else
 237         __u16 offset;
 238         __u16 size;
 239 #endif
 240 };
 241
 242 #define PAGE_FRAG_CACHE_MAX_SIZE        __ALIGN_MASK(32768, ~PAGE_MASK)
 243 #define PAGE_FRAG_CACHE_MAX_ORDER       get_order(PAGE_FRAG_CACHE_MAX_SIZE)
 244
 245 struct page_frag_cache {
 246         void * va;
 247 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
 248         __u16 offset;
 249         __u16 size;
 250 #else
 251         __u32 offset;
 252 #endif
 253         /* we maintain a pagecount bias, so that we dont dirty cache line
 254          * containing page->_count every time we allocate a fragment.
 255          */
 256         unsigned int            pagecnt_bias;
 257         bool pfmemalloc;
 258 };
 259
 260 typedef unsigned long vm_flags_t;
 261
 262 /*
 263  * A region containing a mapping of a non-memory backed file under NOMMU
 264  * conditions.  These are held in a global tree and are pinned by the VMAs that
 265  * map parts of them.
 266  */
 267 struct vm_region {
 268         struct rb_node  vm_rb;          /* link in global region tree */
 269         vm_flags_t      vm_flags;       /* VMA vm_flags */
 270         unsigned long   vm_start;       /* start address of region */
 271         unsigned long   vm_end;         /* region initialised to here */
 272         unsigned long   vm_top;         /* region allocated to here */
 273         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 274         struct file     *vm_file;       /* the backing file or NULL */
 275
 276         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 277         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 278                                                 * this region */
 279 };
 280
 281 #ifdef CONFIG_USERFAULTFD
 282 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
 283 struct vm_userfaultfd_ctx {
 284         struct userfaultfd_ctx *ctx;
 285 };
 286 #else /* CONFIG_USERFAULTFD */
 287 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
 288 struct vm_userfaultfd_ctx {};
 289 #endif /* CONFIG_USERFAULTFD */
 290
 291 /*
 292  * This struct defines a memory VMM memory area. There is one of these
 293  * per VM-area/task.  A VM area is any part of the process virtual memory
 294  * space that has a special rule for the page-fault handlers (ie a shared
 295  * library, the executable area etc).
 296  */
 297 struct vm_area_struct {
 298         /* The first cache line has the info for VMA tree walking. */
 299
 300         unsigned long vm_start;         /* Our start address within vm_mm. */
 301         unsigned long vm_end;           /* The first byte after our end address
 302                                            within vm_mm. */
 303
 304         /* linked list of VM areas per task, sorted by address */
 305         struct vm_area_struct *vm_next, *vm_prev;
 306
 307         struct rb_node vm_rb;
 308
 309         /*
 310          * Largest free memory gap in bytes to the left of this VMA.
 311          * Either between this VMA and vma->vm_prev, or between one of the
 312          * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
 313          * get_unmapped_area find a free area of the right size.
 314          */
 315         unsigned long rb_subtree_gap;
 316
 317         /* Second cache line starts here. */
 318
 319         struct mm_struct *vm_mm;        /* The address space we belong to. */
 320         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 321         unsigned long vm_flags;         /* Flags, see mm.h. */
 322
 323         /*
 324          * For areas with an address space and backing store,
 325          * linkage into the address_space->i_mmap interval tree.
 326          *
 327          * For private anonymous mappings, a pointer to a null terminated string
 328          * in the user process containing the name given to the vma, or NULL
 329          * if unnamed.
 330          */
 331         union {
 332                 struct {
 333                         struct rb_node rb;
 334                         unsigned long rb_subtree_last;
 335                 } shared;
 336                 const char __user *anon_name;
 337         };
 338
 339         /*
 340          * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 341          * list, after a COW of one of the file pages.  A MAP_SHARED vma
 342          * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 343          * or brk vma (with NULL file) can only be in an anon_vma list.
 344          */
 345         struct list_head anon_vma_chain; /* Serialized by mmap_sem &
 346                                           * page_table_lock */
 347         struct anon_vma *anon_vma;      /* Serialized by page_table_lock */
 348
 349         /* Function pointers to deal with this struct. */
 350         const struct vm_operations_struct *vm_ops;
 351
 352         /* Information about our backing store: */
 353         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 354                                            units, *not* PAGE_CACHE_SIZE */
 355         struct file * vm_file;          /* File we map to (can be NULL). */
 356         void * vm_private_data;         /* was vm_pte (shared mem) */
 357
 358 #ifndef CONFIG_MMU
 359         struct vm_region *vm_region;    /* NOMMU mapping region */
 360 #endif
 361 #ifdef CONFIG_NUMA
 362         struct mempolicy *vm_policy;    /* NUMA policy for the VMA */
 363 #endif
 364         struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
 365 };
 366
 367 struct core_thread {
 368         struct task_struct *task;
 369         struct core_thread *next;
 370 };
 371
 372 struct core_state {
 373         atomic_t nr_threads;
 374         struct core_thread dumper;
 375         struct completion startup;
 376 };
 377
 378 enum {
 379         MM_FILEPAGES,
 380         MM_ANONPAGES,
 381         MM_SWAPENTS,
 382         NR_MM_COUNTERS
 383 };
 384
 385 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
 386 #define SPLIT_RSS_COUNTING
 387 /* per-thread cached information, */
 388 struct task_rss_stat {
 389         int events;     /* for synchronization threshold */
 390         int count[NR_MM_COUNTERS];
 391 };
 392 #endif /* USE_SPLIT_PTE_PTLOCKS */
 393
 394 struct mm_rss_stat {
 395         atomic_long_t count[NR_MM_COUNTERS];
 396 };
 397
 398 struct kioctx_table;
 399 struct mm_struct {
 400         struct vm_area_struct *mmap;            /* list of VMAs */
 401         struct rb_root mm_rb;
 402         u32 vmacache_seqnum;                   /* per-thread vmacache */
 403 #ifdef CONFIG_MMU
 404         unsigned long (*get_unmapped_area) (struct file *filp,
 405                                 unsigned long addr, unsigned long len,
 406                                 unsigned long pgoff, unsigned long flags);
 407 #endif
 408         unsigned long mmap_base;                /* base of mmap area */
 409         unsigned long mmap_legacy_base;         /* base of mmap area in bottom-up allocations */
 410         unsigned long task_size;                /* size of task vm space */
 411         unsigned long highest_vm_end;           /* highest vma end address */
 412         pgd_t * pgd;
 413         atomic_t mm_users;                      /* How many users with user space? */
 414         atomic_t mm_count;                      /* How many references to "struct mm_struct" (users count as 1) */
 415         atomic_long_t nr_ptes;                  /* PTE page table pages */
 416 #if CONFIG_PGTABLE_LEVELS > 2
 417         atomic_long_t nr_pmds;                  /* PMD page table pages */
 418 #endif
 419         int map_count;                          /* number of VMAs */
 420
 421         spinlock_t page_table_lock;             /* Protects page tables and some counters */
 422         struct rw_semaphore mmap_sem;
 423
 424         struct list_head mmlist;                /* List of maybe swapped mm's.  These are globally strung
 425                                                  * together off init_mm.mmlist, and are protected
 426                                                  * by mmlist_lock
 427                                                  */
 428
 429
 430         unsigned long hiwater_rss;      /* High-watermark of RSS usage */
 431         unsigned long hiwater_vm;       /* High-water virtual memory usage */
 432
 433         unsigned long total_vm;         /* Total pages mapped */
 434         unsigned long locked_vm;        /* Pages that have PG_mlocked set */
 435         unsigned long pinned_vm;        /* Refcount permanently increased */
 436         unsigned long shared_vm;        /* Shared pages (files) */
 437         unsigned long exec_vm;          /* VM_EXEC & ~VM_WRITE */
 438         unsigned long stack_vm;         /* VM_GROWSUP/DOWN */
 439         unsigned long def_flags;
 440         unsigned long start_code, end_code, start_data, end_data;
 441         unsigned long start_brk, brk, start_stack;
 442         unsigned long arg_start, arg_end, env_start, env_end;
 443
 444         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 445
 446         /*
 447          * Special counters, in some configurations protected by the
 448          * page_table_lock, in other configurations by being atomic.
 449          */
 450         struct mm_rss_stat rss_stat;
 451
 452         struct linux_binfmt *binfmt;
 453
 454         cpumask_var_t cpu_vm_mask_var;
 455
 456         /* Architecture-specific MM context */
 457         mm_context_t context;
 458
 459         unsigned long flags; /* Must use atomic bitops to access the bits */
 460
 461         struct core_state *core_state; /* coredumping support */
 462 #ifdef CONFIG_AIO
 463         spinlock_t                      ioctx_lock;
 464         struct kioctx_table __rcu       *ioctx_table;
 465 #endif
 466 #ifdef CONFIG_MEMCG
 467         /*
 468          * "owner" points to a task that is regarded as the canonical
 469          * user/owner of this mm. All of the following must be true in
 470          * order for it to be changed:
 471          *
 472          * current == mm->owner
 473          * current->mm != mm
 474          * new_owner->mm == mm
 475          * new_owner->alloc_lock is held
 476          */
 477         struct task_struct __rcu *owner;
 478 #endif
 479         struct user_namespace *user_ns;
 480
 481         /* store ref to file /proc/<pid>/exe symlink points to */
 482         struct file __rcu *exe_file;
 483 #ifdef CONFIG_MMU_NOTIFIER
 484         struct mmu_notifier_mm *mmu_notifier_mm;
 485 #endif
 486 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 487         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 488 #endif
 489 #ifdef CONFIG_CPUMASK_OFFSTACK
 490         struct cpumask cpumask_allocation;
 491 #endif
 492 #ifdef CONFIG_NUMA_BALANCING
 493         /*
 494          * numa_next_scan is the next time that the PTEs will be marked
 495          * pte_numa. NUMA hinting faults will gather statistics and migrate
 496          * pages to new nodes if necessary.
 497          */
 498         unsigned long numa_next_scan;
 499
 500         /* Restart point for scanning and setting pte_numa */
 501         unsigned long numa_scan_offset;
 502
 503         /* numa_scan_seq prevents two threads setting pte_numa */
 504         int numa_scan_seq;
 505 #endif
 506 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 507         /*
 508          * An operation with batched TLB flushing is going on. Anything that
 509          * can move process memory needs to flush the TLB when moving a
 510          * PROT_NONE or PROT_NUMA mapped page.
 511          */
 512         bool tlb_flush_pending;
 513 #endif
 514 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
 515         /* See flush_tlb_batched_pending() */
 516         bool tlb_flush_batched;
 517 #endif
 518         struct uprobes_state uprobes_state;
 519 #ifdef CONFIG_X86_INTEL_MPX
 520         /* address of the bounds directory */
 521         void __user *bd_addr;
 522 #endif
 523 #ifdef CONFIG_HUGETLB_PAGE
 524         atomic_long_t hugetlb_usage;
 525 #endif
 526 };
 527
 528 static inline void mm_init_cpumask(struct mm_struct *mm)
 529 {
 530 #ifdef CONFIG_CPUMASK_OFFSTACK
 531         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 532 #endif
 533         cpumask_clear(mm->cpu_vm_mask_var);
 534 }
 535
 536 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 537 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 538 {
 539         return mm->cpu_vm_mask_var;
 540 }
 541
 542 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 543 /*
 544  * Memory barriers to keep this state in sync are graciously provided by
 545  * the page table locks, outside of which no page table modifications happen.
 546  * The barriers below prevent the compiler from re-ordering the instructions
 547  * around the memory barriers that are already present in the code.
 548  */
 549 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 550 {
 551         barrier();
 552         return mm->tlb_flush_pending;
 553 }
 554 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 555 {
 556         mm->tlb_flush_pending = true;
 557
 558         /*
 559          * Guarantee that the tlb_flush_pending store does not leak into the
 560          * critical section updating the page tables
 561          */
 562         smp_mb__before_spinlock();
 563 }
 564 /* Clearing is done after a TLB flush, which also provides a barrier. */
 565 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 566 {
 567         barrier();
 568         mm->tlb_flush_pending = false;
 569 }
 570 #else
 571 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 572 {
 573         return false;
 574 }
 575 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 576 {
 577 }
 578 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 579 {
 580 }
 581 #endif
 582
 583 struct vm_special_mapping
 584 {
 585         const char *name;
 586         struct page **pages;
 587 };
 588
 589 enum tlb_flush_reason {
 590         TLB_FLUSH_ON_TASK_SWITCH,
 591         TLB_REMOTE_SHOOTDOWN,
 592         TLB_LOCAL_SHOOTDOWN,
 593         TLB_LOCAL_MM_SHOOTDOWN,
 594         TLB_REMOTE_SEND_IPI,
 595         NR_TLB_FLUSH_REASONS,
 596 };
 597
 598  /*
 599   * A swap entry has to fit into a "unsigned long", as the entry is hidden
 600   * in the "index" field of the swapper address space.
 601   */
 602 typedef struct {
 603         unsigned long val;
 604 } swp_entry_t;
 605
 606 /* Return the name for an anonymous mapping or NULL for a file-backed mapping */
 607 static inline const char __user *vma_get_anon_name(struct vm_area_struct *vma)
 608 {
 609         if (vma->vm_file)
 610                 return NULL;
 611
 612         return vma->anon_name;
 613 }
 614
 615 #endif /* _LINUX_MM_TYPES_H */