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         struct {
  64                 union {
  65                         pgoff_t index;          /* Our offset within mapping. */
  66                         void *freelist;         /* sl[aou]b first free object */
  67                         /* page_deferred_list().prev    -- second tail page */
  68                 };
  69
  70                 union {
  71 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
  72         defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
  73                         /* Used for cmpxchg_double in slub */
  74                         unsigned long counters;
  75 #else
  76                         /*
  77                          * Keep _refcount separate from slub cmpxchg_double
  78                          * data.  As the rest of the double word is protected by
  79                          * slab_lock but _refcount is not.
  80                          */
  81                         unsigned counters;
  82 #endif
  83
  84                         struct {
  85
  86                                 union {
  87                                         /*
  88                                          * Count of ptes mapped in mms, to show
  89                                          * when page is mapped & limit reverse
  90                                          * map searches.
  91                                          */
  92                                         atomic_t _mapcount;
  93
  94                                         struct { /* SLUB */
  95                                                 unsigned inuse:16;
  96                                                 unsigned objects:15;
  97                                                 unsigned frozen:1;
  98                                         };
  99                                         int units;      /* SLOB */
 100                                 };
 101                                 /*
 102                                  * Usage count, *USE WRAPPER FUNCTION*
 103                                  * when manual accounting. See page_ref.h
 104                                  */
 105                                 atomic_t _refcount;
 106                         };
 107                         unsigned int active;    /* SLAB */
 108                 };
 109         };
 110
 111         /*
 112          * Third double word block
 113          *
 114          * WARNING: bit 0 of the first word encode PageTail(). That means
 115          * the rest users of the storage space MUST NOT use the bit to
 116          * avoid collision and false-positive PageTail().
 117          */
 118         union {
 119                 struct list_head lru;   /* Pageout list, eg. active_list
 120                                          * protected by zone->lru_lock !
 121                                          * Can be used as a generic list
 122                                          * by the page owner.
 123                                          */
 124                 struct dev_pagemap *pgmap; /* ZONE_DEVICE pages are never on an
 125                                             * lru or handled by a slab
 126                                             * allocator, this points to the
 127                                             * hosting device page map.
 128                                             */
 129                 struct {                /* slub per cpu partial pages */
 130                         struct page *next;      /* Next partial slab */
 131 #ifdef CONFIG_64BIT
 132                         int pages;      /* Nr of partial slabs left */
 133                         int pobjects;   /* Approximate # of objects */
 134 #else
 135                         short int pages;
 136                         short int pobjects;
 137 #endif
 138                 };
 139
 140                 struct rcu_head rcu_head;       /* Used by SLAB
 141                                                  * when destroying via RCU
 142                                                  */
 143                 /* Tail pages of compound page */
 144                 struct {
 145                         unsigned long compound_head; /* If bit zero is set */
 146
 147                         /* First tail page only */
 148 #ifdef CONFIG_64BIT
 149                         /*
 150                          * On 64 bit system we have enough space in struct page
 151                          * to encode compound_dtor and compound_order with
 152                          * unsigned int. It can help compiler generate better or
 153                          * smaller code on some archtectures.
 154                          */
 155                         unsigned int compound_dtor;
 156                         unsigned int compound_order;
 157 #else
 158                         unsigned short int compound_dtor;
 159                         unsigned short int compound_order;
 160 #endif
 161                 };
 162
 163 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && USE_SPLIT_PMD_PTLOCKS
 164                 struct {
 165                         unsigned long __pad;    /* do not overlay pmd_huge_pte
 166                                                  * with compound_head to avoid
 167                                                  * possible bit 0 collision.
 168                                                  */
 169                         pgtable_t pmd_huge_pte; /* protected by page->ptl */
 170                 };
 171 #endif
 172         };
 173
 174         /* Remainder is not double word aligned */
 175         union {
 176                 unsigned long private;          /* Mapping-private opaque data:
 177                                                  * usually used for buffer_heads
 178                                                  * if PagePrivate set; used for
 179                                                  * swp_entry_t if PageSwapCache;
 180                                                  * indicates order in the buddy
 181                                                  * system if PG_buddy is set.
 182                                                  */
 183 #if USE_SPLIT_PTE_PTLOCKS
 184 #if ALLOC_SPLIT_PTLOCKS
 185                 spinlock_t *ptl;
 186 #else
 187                 spinlock_t ptl;
 188 #endif
 189 #endif
 190                 struct kmem_cache *slab_cache;  /* SL[AU]B: Pointer to slab */
 191         };
 192
 193 #ifdef CONFIG_MEMCG
 194         struct mem_cgroup *mem_cgroup;
 195 #endif
 196
 197         /*
 198          * On machines where all RAM is mapped into kernel address space,
 199          * we can simply calculate the virtual address. On machines with
 200          * highmem some memory is mapped into kernel virtual memory
 201          * dynamically, so we need a place to store that address.
 202          * Note that this field could be 16 bits on x86 ... ;)
 203          *
 204          * Architectures with slow multiplication can define
 205          * WANT_PAGE_VIRTUAL in asm/page.h
 206          */
 207 #if defined(WANT_PAGE_VIRTUAL)
 208         void *virtual;                  /* Kernel virtual address (NULL if
 209                                            not kmapped, ie. highmem) */
 210 #endif /* WANT_PAGE_VIRTUAL */
 211
 212 #ifdef CONFIG_KMEMCHECK
 213         /*
 214          * kmemcheck wants to track the status of each byte in a page; this
 215          * is a pointer to such a status block. NULL if not tracked.
 216          */
 217         void *shadow;
 218 #endif
 219
 220 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 221         int _last_cpupid;
 222 #endif
 223 }
 224 /*
 225  * The struct page can be forced to be double word aligned so that atomic ops
 226  * on double words work. The SLUB allocator can make use of such a feature.
 227  */
 228 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
 229         __aligned(2 * sizeof(unsigned long))
 230 #endif
 231 ;
 232
 233 struct page_frag {
 234         struct page *page;
 235 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 236         __u32 offset;
 237         __u32 size;
 238 #else
 239         __u16 offset;
 240         __u16 size;
 241 #endif
 242 };
 243
 244 #define PAGE_FRAG_CACHE_MAX_SIZE        __ALIGN_MASK(32768, ~PAGE_MASK)
 245 #define PAGE_FRAG_CACHE_MAX_ORDER       get_order(PAGE_FRAG_CACHE_MAX_SIZE)
 246
 247 struct page_frag_cache {
 248         void * va;
 249 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
 250         __u16 offset;
 251         __u16 size;
 252 #else
 253         __u32 offset;
 254 #endif
 255         /* we maintain a pagecount bias, so that we dont dirty cache line
 256          * containing page->_refcount every time we allocate a fragment.
 257          */
 258         unsigned int            pagecnt_bias;
 259         bool pfmemalloc;
 260 };
 261
 262 typedef unsigned long vm_flags_t;
 263
 264 /*
 265  * A region containing a mapping of a non-memory backed file under NOMMU
 266  * conditions.  These are held in a global tree and are pinned by the VMAs that
 267  * map parts of them.
 268  */
 269 struct vm_region {
 270         struct rb_node  vm_rb;          /* link in global region tree */
 271         vm_flags_t      vm_flags;       /* VMA vm_flags */
 272         unsigned long   vm_start;       /* start address of region */
 273         unsigned long   vm_end;         /* region initialised to here */
 274         unsigned long   vm_top;         /* region allocated to here */
 275         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 276         struct file     *vm_file;       /* the backing file or NULL */
 277
 278         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 279         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 280                                                 * this region */
 281 };
 282
 283 #ifdef CONFIG_USERFAULTFD
 284 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
 285 struct vm_userfaultfd_ctx {
 286         struct userfaultfd_ctx *ctx;
 287 };
 288 #else /* CONFIG_USERFAULTFD */
 289 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
 290 struct vm_userfaultfd_ctx {};
 291 #endif /* CONFIG_USERFAULTFD */
 292
 293 /*
 294  * This struct defines a memory VMM memory area. There is one of these
 295  * per VM-area/task.  A VM area is any part of the process virtual memory
 296  * space that has a special rule for the page-fault handlers (ie a shared
 297  * library, the executable area etc).
 298  */
 299 struct vm_area_struct {
 300         /* The first cache line has the info for VMA tree walking. */
 301
 302         unsigned long vm_start;         /* Our start address within vm_mm. */
 303         unsigned long vm_end;           /* The first byte after our end address
 304                                            within vm_mm. */
 305
 306         /* linked list of VM areas per task, sorted by address */
 307         struct vm_area_struct *vm_next, *vm_prev;
 308
 309         struct rb_node vm_rb;
 310
 311         /*
 312          * Largest free memory gap in bytes to the left of this VMA.
 313          * Either between this VMA and vma->vm_prev, or between one of the
 314          * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
 315          * get_unmapped_area find a free area of the right size.
 316          */
 317         unsigned long rb_subtree_gap;
 318
 319         /* Second cache line starts here. */
 320
 321         struct mm_struct *vm_mm;        /* The address space we belong to. */
 322         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 323         unsigned long vm_flags;         /* Flags, see mm.h. */
 324
 325         /*
 326          * For areas with an address space and backing store,
 327          * linkage into the address_space->i_mmap interval tree.
 328          */
 329         struct {
 330                 struct rb_node rb;
 331                 unsigned long rb_subtree_last;
 332         } shared;
 333
 334         /*
 335          * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 336          * list, after a COW of one of the file pages.  A MAP_SHARED vma
 337          * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 338          * or brk vma (with NULL file) can only be in an anon_vma list.
 339          */
 340         struct list_head anon_vma_chain; /* Serialized by mmap_sem &
 341                                           * page_table_lock */
 342         struct anon_vma *anon_vma;      /* Serialized by page_table_lock */
 343
 344         /* Function pointers to deal with this struct. */
 345         const struct vm_operations_struct *vm_ops;
 346
 347         /* Information about our backing store: */
 348         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 349                                            units */
 350         struct file * vm_file;          /* File we map to (can be NULL). */
 351         void * vm_private_data;         /* was vm_pte (shared mem) */
 352
 353 #ifndef CONFIG_MMU
 354         struct vm_region *vm_region;    /* NOMMU mapping region */
 355 #endif
 356 #ifdef CONFIG_NUMA
 357         struct mempolicy *vm_policy;    /* NUMA policy for the VMA */
 358 #endif
 359         struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
 360 };
 361
 362 struct core_thread {
 363         struct task_struct *task;
 364         struct core_thread *next;
 365 };
 366
 367 struct core_state {
 368         atomic_t nr_threads;
 369         struct core_thread dumper;
 370         struct completion startup;
 371 };
 372
 373 enum {
 374         MM_FILEPAGES,   /* Resident file mapping pages */
 375         MM_ANONPAGES,   /* Resident anonymous pages */
 376         MM_SWAPENTS,    /* Anonymous swap entries */
 377         MM_SHMEMPAGES,  /* Resident shared memory pages */
 378         NR_MM_COUNTERS
 379 };
 380
 381 #if USE_SPLIT_PTE_PTLOCKS && defined(CONFIG_MMU)
 382 #define SPLIT_RSS_COUNTING
 383 /* per-thread cached information, */
 384 struct task_rss_stat {
 385         int events;     /* for synchronization threshold */
 386         int count[NR_MM_COUNTERS];
 387 };
 388 #endif /* USE_SPLIT_PTE_PTLOCKS */
 389
 390 struct mm_rss_stat {
 391         atomic_long_t count[NR_MM_COUNTERS];
 392 };
 393
 394 struct kioctx_table;
 395 struct mm_struct {
 396         struct vm_area_struct *mmap;            /* list of VMAs */
 397         struct rb_root mm_rb;
 398         u32 vmacache_seqnum;                   /* per-thread vmacache */
 399 #ifdef CONFIG_MMU
 400         unsigned long (*get_unmapped_area) (struct file *filp,
 401                                 unsigned long addr, unsigned long len,
 402                                 unsigned long pgoff, unsigned long flags);
 403 #endif
 404         unsigned long mmap_base;                /* base of mmap area */
 405         unsigned long mmap_legacy_base;         /* base of mmap area in bottom-up allocations */
 406         unsigned long task_size;                /* size of task vm space */
 407         unsigned long highest_vm_end;           /* highest vma end address */
 408         pgd_t * pgd;
 409         atomic_t mm_users;                      /* How many users with user space? */
 410         atomic_t mm_count;                      /* How many references to "struct mm_struct" (users count as 1) */
 411         atomic_long_t nr_ptes;                  /* PTE page table pages */
 412 #if CONFIG_PGTABLE_LEVELS > 2
 413         atomic_long_t nr_pmds;                  /* PMD page table pages */
 414 #endif
 415         int map_count;                          /* number of VMAs */
 416
 417         spinlock_t page_table_lock;             /* Protects page tables and some counters */
 418         struct rw_semaphore mmap_sem;
 419
 420         struct list_head mmlist;                /* List of maybe swapped mm's.  These are globally strung
 421                                                  * together off init_mm.mmlist, and are protected
 422                                                  * by mmlist_lock
 423                                                  */
 424
 425
 426         unsigned long hiwater_rss;      /* High-watermark of RSS usage */
 427         unsigned long hiwater_vm;       /* High-water virtual memory usage */
 428
 429         unsigned long total_vm;         /* Total pages mapped */
 430         unsigned long locked_vm;        /* Pages that have PG_mlocked set */
 431         unsigned long pinned_vm;        /* Refcount permanently increased */
 432         unsigned long data_vm;          /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
 433         unsigned long exec_vm;          /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
 434         unsigned long stack_vm;         /* VM_STACK */
 435         unsigned long def_flags;
 436         unsigned long start_code, end_code, start_data, end_data;
 437         unsigned long start_brk, brk, start_stack;
 438         unsigned long arg_start, arg_end, env_start, env_end;
 439
 440         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 441
 442         /*
 443          * Special counters, in some configurations protected by the
 444          * page_table_lock, in other configurations by being atomic.
 445          */
 446         struct mm_rss_stat rss_stat;
 447
 448         struct linux_binfmt *binfmt;
 449
 450         cpumask_var_t cpu_vm_mask_var;
 451
 452         /* Architecture-specific MM context */
 453         mm_context_t context;
 454
 455         unsigned long flags; /* Must use atomic bitops to access the bits */
 456
 457         struct core_state *core_state; /* coredumping support */
 458 #ifdef CONFIG_AIO
 459         spinlock_t                      ioctx_lock;
 460         struct kioctx_table __rcu       *ioctx_table;
 461 #endif
 462 #ifdef CONFIG_MEMCG
 463         /*
 464          * "owner" points to a task that is regarded as the canonical
 465          * user/owner of this mm. All of the following must be true in
 466          * order for it to be changed:
 467          *
 468          * current == mm->owner
 469          * current->mm != mm
 470          * new_owner->mm == mm
 471          * new_owner->alloc_lock is held
 472          */
 473         struct task_struct __rcu *owner;
 474 #endif
 475
 476         /* store ref to file /proc/<pid>/exe symlink points to */
 477         struct file __rcu *exe_file;
 478 #ifdef CONFIG_MMU_NOTIFIER
 479         struct mmu_notifier_mm *mmu_notifier_mm;
 480 #endif
 481 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 482         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 483 #endif
 484 #ifdef CONFIG_CPUMASK_OFFSTACK
 485         struct cpumask cpumask_allocation;
 486 #endif
 487 #ifdef CONFIG_NUMA_BALANCING
 488         /*
 489          * numa_next_scan is the next time that the PTEs will be marked
 490          * pte_numa. NUMA hinting faults will gather statistics and migrate
 491          * pages to new nodes if necessary.
 492          */
 493         unsigned long numa_next_scan;
 494
 495         /* Restart point for scanning and setting pte_numa */
 496         unsigned long numa_scan_offset;
 497
 498         /* numa_scan_seq prevents two threads setting pte_numa */
 499         int numa_scan_seq;
 500 #endif
 501 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 502         /*
 503          * An operation with batched TLB flushing is going on. Anything that
 504          * can move process memory needs to flush the TLB when moving a
 505          * PROT_NONE or PROT_NUMA mapped page.
 506          */
 507         bool tlb_flush_pending;
 508 #endif
 509         struct uprobes_state uprobes_state;
 510 #ifdef CONFIG_X86_INTEL_MPX
 511         /* address of the bounds directory */
 512         void __user *bd_addr;
 513 #endif
 514 #ifdef CONFIG_HUGETLB_PAGE
 515         atomic_long_t hugetlb_usage;
 516 #endif
 517 #ifdef CONFIG_MMU
 518         struct work_struct async_put_work;
 519 #endif
 520 };
 521
 522 static inline void mm_init_cpumask(struct mm_struct *mm)
 523 {
 524 #ifdef CONFIG_CPUMASK_OFFSTACK
 525         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 526 #endif
 527         cpumask_clear(mm->cpu_vm_mask_var);
 528 }
 529
 530 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 531 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 532 {
 533         return mm->cpu_vm_mask_var;
 534 }
 535
 536 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 537 /*
 538  * Memory barriers to keep this state in sync are graciously provided by
 539  * the page table locks, outside of which no page table modifications happen.
 540  * The barriers below prevent the compiler from re-ordering the instructions
 541  * around the memory barriers that are already present in the code.
 542  */
 543 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 544 {
 545         barrier();
 546         return mm->tlb_flush_pending;
 547 }
 548 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 549 {
 550         mm->tlb_flush_pending = true;
 551
 552         /*
 553          * Guarantee that the tlb_flush_pending store does not leak into the
 554          * critical section updating the page tables
 555          */
 556         smp_mb__before_spinlock();
 557 }
 558 /* Clearing is done after a TLB flush, which also provides a barrier. */
 559 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 560 {
 561         barrier();
 562         mm->tlb_flush_pending = false;
 563 }
 564 #else
 565 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 566 {
 567         return false;
 568 }
 569 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 570 {
 571 }
 572 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 573 {
 574 }
 575 #endif
 576
 577 struct vm_fault;
 578
 579 struct vm_special_mapping {
 580         const char *name;       /* The name, e.g. "[vdso]". */
 581
 582         /*
 583          * If .fault is not provided, this points to a
 584          * NULL-terminated array of pages that back the special mapping.
 585          *
 586          * This must not be NULL unless .fault is provided.
 587          */
 588         struct page **pages;
 589
 590         /*
 591          * If non-NULL, then this is called to resolve page faults
 592          * on the special mapping.  If used, .pages is not checked.
 593          */
 594         int (*fault)(const struct vm_special_mapping *sm,
 595                      struct vm_area_struct *vma,
 596                      struct vm_fault *vmf);
 597 };
 598
 599 enum tlb_flush_reason {
 600         TLB_FLUSH_ON_TASK_SWITCH,
 601         TLB_REMOTE_SHOOTDOWN,
 602         TLB_LOCAL_SHOOTDOWN,
 603         TLB_LOCAL_MM_SHOOTDOWN,
 604         TLB_REMOTE_SEND_IPI,
 605         NR_TLB_FLUSH_REASONS,
 606 };
 607
 608  /*
 609   * A swap entry has to fit into a "unsigned long", as the entry is hidden
 610   * in the "index" field of the swapper address space.
 611   */
 612 typedef struct {
 613         unsigned long val;
 614 } swp_entry_t;
 615
 616 #endif /* _LINUX_MM_TYPES_H */