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 <linux/workqueue.h>
  16 #include <asm/page.h>
  17 #include <asm/mmu.h>
  18
  19 #ifndef AT_VECTOR_SIZE_ARCH
  20 #define AT_VECTOR_SIZE_ARCH 0
  21 #endif
  22 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
  23
  24 struct address_space;
  25 struct mem_cgroup;
  26
  27 #define USE_SPLIT_PTE_PTLOCKS   (NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS)
  28 #define USE_SPLIT_PMD_PTLOCKS   (USE_SPLIT_PTE_PTLOCKS && \
  29                 IS_ENABLED(CONFIG_ARCH_ENABLE_SPLIT_PMD_PTLOCK))
  30 #define ALLOC_SPLIT_PTLOCKS     (SPINLOCK_SIZE > BITS_PER_LONG/8)
  31
  32 /*
  33  * Each physical page in the system has a struct page associated with
  34  * it to keep track of whatever it is we are using the page for at the
  35  * moment. Note that we have no way to track which tasks are using
  36  * a page, though if it is a pagecache page, rmap structures can tell us
  37  * who is mapping it.
  38  *
  39  * The objects in struct page are organized in double word blocks in
  40  * order to allows us to use atomic double word operations on portions
  41  * of struct page. That is currently only used by slub but the arrangement
  42  * allows the use of atomic double word operations on the flags/mapping
  43  * and lru list pointers also.
  44  */
  45 struct page {
  46         /* First double word block */
  47         unsigned long flags;            /* Atomic flags, some possibly
  48                                          * updated asynchronously */
  49         union {
  50                 struct address_space *mapping;  /* If low bit clear, points to
  51                                                  * inode address_space, or NULL.
  52                                                  * If page mapped as anonymous
  53                                                  * memory, low bit is set, and
  54                                                  * it points to anon_vma object:
  55                                                  * see PAGE_MAPPING_ANON below.
  56                                                  */
  57                 void *s_mem;                    /* slab first object */
  58                 atomic_t compound_mapcount;     /* first tail page */
  59                 /* page_deferred_list().next     -- second tail page */
  60         };
  61
  62         /* Second double word */
  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 _refcount separate from slub cmpxchg_double data.
  77                  * As the rest of the double word is protected by slab_lock
  78                  * but _refcount is not.
  79                  */
  80                 unsigned counters;
  81 #endif
  82                 struct {
  83
  84                         union {
  85                                 /*
  86                                  * Count of ptes mapped in mms, to show when
  87                                  * page is mapped & limit reverse map searches.
  88                                  */
  89                                 atomic_t _mapcount;
  90
  91                                 unsigned int active;            /* SLAB */
  92                                 struct {                        /* SLUB */
  93                                         unsigned inuse:16;
  94                                         unsigned objects:15;
  95                                         unsigned frozen:1;
  96                                 };
  97                                 int units;                      /* SLOB */
  98                         };
  99                         /*
 100                          * Usage count, *USE WRAPPER FUNCTION* when manual
 101                          * accounting. See page_ref.h
 102                          */
 103                         atomic_t _refcount;
 104                 };
 105         };
 106
 107         /*
 108          * Third double word block
 109          *
 110          * WARNING: bit 0 of the first word encode PageTail(). That means
 111          * the rest users of the storage space MUST NOT use the bit to
 112          * avoid collision and false-positive PageTail().
 113          */
 114         union {
 115                 struct list_head lru;   /* Pageout list, eg. active_list
 116                                          * protected by zone->lru_lock !
 117                                          * Can be used as a generic list
 118                                          * by the page owner.
 119                                          */
 120                 struct dev_pagemap *pgmap; /* ZONE_DEVICE pages are never on an
 121                                             * lru or handled by a slab
 122                                             * allocator, this points to the
 123                                             * hosting device page map.
 124                                             */
 125                 struct {                /* slub per cpu partial pages */
 126                         struct page *next;      /* Next partial slab */
 127 #ifdef CONFIG_64BIT
 128                         int pages;      /* Nr of partial slabs left */
 129                         int pobjects;   /* Approximate # of objects */
 130 #else
 131                         short int pages;
 132                         short int pobjects;
 133 #endif
 134                 };
 135
 136                 struct rcu_head rcu_head;       /* Used by SLAB
 137                                                  * when destroying via RCU
 138                                                  */
 139                 /* Tail pages of compound page */
 140                 struct {
 141                         unsigned long compound_head; /* If bit zero is set */
 142
 143                         /* First tail page only */
 144 #ifdef CONFIG_64BIT
 145                         /*
 146                          * On 64 bit system we have enough space in struct page
 147                          * to encode compound_dtor and compound_order with
 148                          * unsigned int. It can help compiler generate better or
 149                          * smaller code on some archtectures.
 150                          */
 151                         unsigned int compound_dtor;
 152                         unsigned int compound_order;
 153 #else
 154                         unsigned short int compound_dtor;
 155                         unsigned short int compound_order;
 156 #endif
 157                 };
 158
 159 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
 160                 struct {
 161                         unsigned long __pad;    /* do not overlay pmd_huge_pte
 162                                                  * with compound_head to avoid
 163                                                  * possible bit 0 collision.
 164                                                  */
 165                         pgtable_t pmd_huge_pte; /* protected by page->ptl */
 166                 };
 167 #endif
 168         };
 169
 170         /* Remainder is not double word aligned */
 171         union {
 172                 unsigned long private;          /* Mapping-private opaque data:
 173                                                  * usually used for buffer_heads
 174                                                  * if PagePrivate set; used for
 175                                                  * swp_entry_t if PageSwapCache;
 176                                                  * indicates order in the buddy
 177                                                  * system if PG_buddy is set.
 178                                                  */
 179 #if USE_SPLIT_PTE_PTLOCKS
 180 #if ALLOC_SPLIT_PTLOCKS
 181                 spinlock_t *ptl;
 182 #else
 183                 spinlock_t ptl;
 184 #endif
 185 #endif
 186                 struct kmem_cache *slab_cache;  /* SL[AU]B: Pointer to slab */
 187         };
 188
 189 #ifdef CONFIG_MEMCG
 190         struct mem_cgroup *mem_cgroup;
 191 #endif
 192
 193         /*
 194          * On machines where all RAM is mapped into kernel address space,
 195          * we can simply calculate the virtual address. On machines with
 196          * highmem some memory is mapped into kernel virtual memory
 197          * dynamically, so we need a place to store that address.
 198          * Note that this field could be 16 bits on x86 ... ;)
 199          *
 200          * Architectures with slow multiplication can define
 201          * WANT_PAGE_VIRTUAL in asm/page.h
 202          */
 203 #if defined(WANT_PAGE_VIRTUAL)
 204         void *virtual;                  /* Kernel virtual address (NULL if
 205                                            not kmapped, ie. highmem) */
 206 #endif /* WANT_PAGE_VIRTUAL */
 207
 208 #ifdef CONFIG_KMEMCHECK
 209         /*
 210          * kmemcheck wants to track the status of each byte in a page; this
 211          * is a pointer to such a status block. NULL if not tracked.
 212          */
 213         void *shadow;
 214 #endif
 215
 216 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 217         int _last_cpupid;
 218 #endif
 219 }
 220 /*
 221  * The struct page can be forced to be double word aligned so that atomic ops
 222  * on double words work. The SLUB allocator can make use of such a feature.
 223  */
 224 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
 225         __aligned(2 * sizeof(unsigned long))
 226 #endif
 227 ;
 228
 229 struct page_frag {
 230         struct page *page;
 231 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 232         __u32 offset;
 233         __u32 size;
 234 #else
 235         __u16 offset;
 236         __u16 size;
 237 #endif
 238 };
 239
 240 #define PAGE_FRAG_CACHE_MAX_SIZE        __ALIGN_MASK(32768, ~PAGE_MASK)
 241 #define PAGE_FRAG_CACHE_MAX_ORDER       get_order(PAGE_FRAG_CACHE_MAX_SIZE)
 242
 243 struct page_frag_cache {
 244         void * va;
 245 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
 246         __u16 offset;
 247         __u16 size;
 248 #else
 249         __u32 offset;
 250 #endif
 251         /* we maintain a pagecount bias, so that we dont dirty cache line
 252          * containing page->_refcount every time we allocate a fragment.
 253          */
 254         unsigned int            pagecnt_bias;
 255         bool pfmemalloc;
 256 };
 257
 258 typedef unsigned long vm_flags_t;
 259
 260 /*
 261  * A region containing a mapping of a non-memory backed file under NOMMU
 262  * conditions.  These are held in a global tree and are pinned by the VMAs that
 263  * map parts of them.
 264  */
 265 struct vm_region {
 266         struct rb_node  vm_rb;          /* link in global region tree */
 267         vm_flags_t      vm_flags;       /* VMA vm_flags */
 268         unsigned long   vm_start;       /* start address of region */
 269         unsigned long   vm_end;         /* region initialised to here */
 270         unsigned long   vm_top;         /* region allocated to here */
 271         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 272         struct file     *vm_file;       /* the backing file or NULL */
 273
 274         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 275         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 276                                                 * this region */
 277 };
 278
 279 #ifdef CONFIG_USERFAULTFD
 280 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
 281 struct vm_userfaultfd_ctx {
 282         struct userfaultfd_ctx *ctx;
 283 };
 284 #else /* CONFIG_USERFAULTFD */
 285 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
 286 struct vm_userfaultfd_ctx {};
 287 #endif /* CONFIG_USERFAULTFD */
 288
 289 /*
 290  * This struct defines a memory VMM memory area. There is one of these
 291  * per VM-area/task.  A VM area is any part of the process virtual memory
 292  * space that has a special rule for the page-fault handlers (ie a shared
 293  * library, the executable area etc).
 294  */
 295 struct vm_area_struct {
 296         /* The first cache line has the info for VMA tree walking. */
 297
 298         unsigned long vm_start;         /* Our start address within vm_mm. */
 299         unsigned long vm_end;           /* The first byte after our end address
 300                                            within vm_mm. */
 301
 302         /* linked list of VM areas per task, sorted by address */
 303         struct vm_area_struct *vm_next, *vm_prev;
 304
 305         struct rb_node vm_rb;
 306
 307         /*
 308          * Largest free memory gap in bytes to the left of this VMA.
 309          * Either between this VMA and vma->vm_prev, or between one of the
 310          * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
 311          * get_unmapped_area find a free area of the right size.
 312          */
 313         unsigned long rb_subtree_gap;
 314
 315         /* Second cache line starts here. */
 316
 317         struct mm_struct *vm_mm;        /* The address space we belong to. */
 318         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 319         unsigned long vm_flags;         /* Flags, see mm.h. */
 320
 321         /*
 322          * For areas with an address space and backing store,
 323          * linkage into the address_space->i_mmap interval tree.
 324          */
 325         struct {
 326                 struct rb_node rb;
 327                 unsigned long rb_subtree_last;
 328         } shared;
 329
 330         /*
 331          * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 332          * list, after a COW of one of the file pages.  A MAP_SHARED vma
 333          * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 334          * or brk vma (with NULL file) can only be in an anon_vma list.
 335          */
 336         struct list_head anon_vma_chain; /* Serialized by mmap_sem &
 337                                           * page_table_lock */
 338         struct anon_vma *anon_vma;      /* Serialized by page_table_lock */
 339
 340         /* Function pointers to deal with this struct. */
 341         const struct vm_operations_struct *vm_ops;
 342
 343         /* Information about our backing store: */
 344         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 345                                            units */
 346         struct file * vm_file;          /* File we map to (can be NULL). */
 347         void * vm_private_data;         /* was vm_pte (shared mem) */
 348
 349 #ifndef CONFIG_MMU
 350         struct vm_region *vm_region;    /* NOMMU mapping region */
 351 #endif
 352 #ifdef CONFIG_NUMA
 353         struct mempolicy *vm_policy;    /* NUMA policy for the VMA */
 354 #endif
 355         struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
 356 };
 357
 358 struct core_thread {
 359         struct task_struct *task;
 360         struct core_thread *next;
 361 };
 362
 363 struct core_state {
 364         atomic_t nr_threads;
 365         struct core_thread dumper;
 366         struct completion startup;
 367 };
 368
 369 enum {
 370         MM_FILEPAGES,   /* Resident file mapping pages */
 371         MM_ANONPAGES,   /* Resident anonymous pages */
 372         MM_SWAPENTS,    /* Anonymous swap entries */
 373         MM_SHMEMPAGES,  /* Resident shared memory pages */
 374         NR_MM_COUNTERS
 375 };
 376
 377 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
 378 #define SPLIT_RSS_COUNTING
 379 /* per-thread cached information, */
 380 struct task_rss_stat {
 381         int events;     /* for synchronization threshold */
 382         int count[NR_MM_COUNTERS];
 383 };
 384 #endif /* USE_SPLIT_PTE_PTLOCKS */
 385
 386 struct mm_rss_stat {
 387         atomic_long_t count[NR_MM_COUNTERS];
 388 };
 389
 390 struct kioctx_table;
 391 struct mm_struct {
 392         struct vm_area_struct *mmap;            /* list of VMAs */
 393         struct rb_root mm_rb;
 394         u32 vmacache_seqnum;                   /* per-thread vmacache */
 395 #ifdef CONFIG_MMU
 396         unsigned long (*get_unmapped_area) (struct file *filp,
 397                                 unsigned long addr, unsigned long len,
 398                                 unsigned long pgoff, unsigned long flags);
 399 #endif
 400         unsigned long mmap_base;                /* base of mmap area */
 401         unsigned long mmap_legacy_base;         /* base of mmap area in bottom-up allocations */
 402         unsigned long task_size;                /* size of task vm space */
 403         unsigned long highest_vm_end;           /* highest vma end address */
 404         pgd_t * pgd;
 405         atomic_t mm_users;                      /* How many users with user space? */
 406         atomic_t mm_count;                      /* How many references to "struct mm_struct" (users count as 1) */
 407         atomic_long_t nr_ptes;                  /* PTE page table pages */
 408 #if CONFIG_PGTABLE_LEVELS > 2
 409         atomic_long_t nr_pmds;                  /* PMD page table pages */
 410 #endif
 411         int map_count;                          /* number of VMAs */
 412
 413         spinlock_t page_table_lock;             /* Protects page tables and some counters */
 414         struct rw_semaphore mmap_sem;
 415
 416         struct list_head mmlist;                /* List of maybe swapped mm's.  These are globally strung
 417                                                  * together off init_mm.mmlist, and are protected
 418                                                  * by mmlist_lock
 419                                                  */
 420
 421
 422         unsigned long hiwater_rss;      /* High-watermark of RSS usage */
 423         unsigned long hiwater_vm;       /* High-water virtual memory usage */
 424
 425         unsigned long total_vm;         /* Total pages mapped */
 426         unsigned long locked_vm;        /* Pages that have PG_mlocked set */
 427         unsigned long pinned_vm;        /* Refcount permanently increased */
 428         unsigned long data_vm;          /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
 429         unsigned long exec_vm;          /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
 430         unsigned long stack_vm;         /* VM_STACK */
 431         unsigned long def_flags;
 432         unsigned long start_code, end_code, start_data, end_data;
 433         unsigned long start_brk, brk, start_stack;
 434         unsigned long arg_start, arg_end, env_start, env_end;
 435
 436         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 437
 438         /*
 439          * Special counters, in some configurations protected by the
 440          * page_table_lock, in other configurations by being atomic.
 441          */
 442         struct mm_rss_stat rss_stat;
 443
 444         struct linux_binfmt *binfmt;
 445
 446         cpumask_var_t cpu_vm_mask_var;
 447
 448         /* Architecture-specific MM context */
 449         mm_context_t context;
 450
 451         unsigned long flags; /* Must use atomic bitops to access the bits */
 452
 453         struct core_state *core_state; /* coredumping support */
 454 #ifdef CONFIG_AIO
 455         spinlock_t                      ioctx_lock;
 456         struct kioctx_table __rcu       *ioctx_table;
 457 #endif
 458 #ifdef CONFIG_MEMCG
 459         /*
 460          * "owner" points to a task that is regarded as the canonical
 461          * user/owner of this mm. All of the following must be true in
 462          * order for it to be changed:
 463          *
 464          * current == mm->owner
 465          * current->mm != mm
 466          * new_owner->mm == mm
 467          * new_owner->alloc_lock is held
 468          */
 469         struct task_struct __rcu *owner;
 470 #endif
 471
 472         /* store ref to file /proc/<pid>/exe symlink points to */
 473         struct file __rcu *exe_file;
 474 #ifdef CONFIG_MMU_NOTIFIER
 475         struct mmu_notifier_mm *mmu_notifier_mm;
 476 #endif
 477 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 478         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 479 #endif
 480 #ifdef CONFIG_CPUMASK_OFFSTACK
 481         struct cpumask cpumask_allocation;
 482 #endif
 483 #ifdef CONFIG_NUMA_BALANCING
 484         /*
 485          * numa_next_scan is the next time that the PTEs will be marked
 486          * pte_numa. NUMA hinting faults will gather statistics and migrate
 487          * pages to new nodes if necessary.
 488          */
 489         unsigned long numa_next_scan;
 490
 491         /* Restart point for scanning and setting pte_numa */
 492         unsigned long numa_scan_offset;
 493
 494         /* numa_scan_seq prevents two threads setting pte_numa */
 495         int numa_scan_seq;
 496 #endif
 497 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 498         /*
 499          * An operation with batched TLB flushing is going on. Anything that
 500          * can move process memory needs to flush the TLB when moving a
 501          * PROT_NONE or PROT_NUMA mapped page.
 502          */
 503         bool tlb_flush_pending;
 504 #endif
 505         struct uprobes_state uprobes_state;
 506 #ifdef CONFIG_X86_INTEL_MPX
 507         /* address of the bounds directory */
 508         void __user *bd_addr;
 509 #endif
 510 #ifdef CONFIG_HUGETLB_PAGE
 511         atomic_long_t hugetlb_usage;
 512 #endif
 513 #ifdef CONFIG_MMU
 514         struct work_struct async_put_work;
 515 #endif
 516 };
 517
 518 static inline void mm_init_cpumask(struct mm_struct *mm)
 519 {
 520 #ifdef CONFIG_CPUMASK_OFFSTACK
 521         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 522 #endif
 523         cpumask_clear(mm->cpu_vm_mask_var);
 524 }
 525
 526 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 527 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 528 {
 529         return mm->cpu_vm_mask_var;
 530 }
 531
 532 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 533 /*
 534  * Memory barriers to keep this state in sync are graciously provided by
 535  * the page table locks, outside of which no page table modifications happen.
 536  * The barriers below prevent the compiler from re-ordering the instructions
 537  * around the memory barriers that are already present in the code.
 538  */
 539 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 540 {
 541         barrier();
 542         return mm->tlb_flush_pending;
 543 }
 544 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 545 {
 546         mm->tlb_flush_pending = true;
 547
 548         /*
 549          * Guarantee that the tlb_flush_pending store does not leak into the
 550          * critical section updating the page tables
 551          */
 552         smp_mb__before_spinlock();
 553 }
 554 /* Clearing is done after a TLB flush, which also provides a barrier. */
 555 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 556 {
 557         barrier();
 558         mm->tlb_flush_pending = false;
 559 }
 560 #else
 561 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 562 {
 563         return false;
 564 }
 565 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 566 {
 567 }
 568 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 569 {
 570 }
 571 #endif
 572
 573 struct vm_fault;
 574
 575 struct vm_special_mapping {
 576         const char *name;       /* The name, e.g. "[vdso]". */
 577
 578         /*
 579          * If .fault is not provided, this points to a
 580          * NULL-terminated array of pages that back the special mapping.
 581          *
 582          * This must not be NULL unless .fault is provided.
 583          */
 584         struct page **pages;
 585
 586         /*
 587          * If non-NULL, then this is called to resolve page faults
 588          * on the special mapping.  If used, .pages is not checked.
 589          */
 590         int (*fault)(const struct vm_special_mapping *sm,
 591                      struct vm_area_struct *vma,
 592                      struct vm_fault *vmf);
 593
 594         int (*mremap)(const struct vm_special_mapping *sm,
 595                      struct vm_area_struct *new_vma);
 596 };
 597
 598 enum tlb_flush_reason {
 599         TLB_FLUSH_ON_TASK_SWITCH,
 600         TLB_REMOTE_SHOOTDOWN,
 601         TLB_LOCAL_SHOOTDOWN,
 602         TLB_LOCAL_MM_SHOOTDOWN,
 603         TLB_REMOTE_SEND_IPI,
 604         NR_TLB_FLUSH_REASONS,
 605 };
 606
 607  /*
 608   * A swap entry has to fit into a "unsigned long", as the entry is hidden
 609   * in the "index" field of the swapper address space.
 610   */
 611 typedef struct {
 612         unsigned long val;
 613 } swp_entry_t;
 614
 615 #endif /* _LINUX_MM_TYPES_H */