include/linux/mm_types.h

   1 #ifndef _LINUX_MM_TYPES_H
   2 #define _LINUX_MM_TYPES_H
   3
   4 #include <linux/mm_types_task.h>
   5
   6 #include <linux/auxvec.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
  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 /*
  28  * Each physical page in the system has a struct page associated with
  29  * it to keep track of whatever it is we are using the page for at the
  30  * moment. Note that we have no way to track which tasks are using
  31  * a page, though if it is a pagecache page, rmap structures can tell us
  32  * who is mapping it.
  33  *
  34  * The objects in struct page are organized in double word blocks in
  35  * order to allows us to use atomic double word operations on portions
  36  * of struct page. That is currently only used by slub but the arrangement
  37  * allows the use of atomic double word operations on the flags/mapping
  38  * and lru list pointers also.
  39  */
  40 struct page {
  41         /* First double word block */
  42         unsigned long flags;            /* Atomic flags, some possibly
  43                                          * updated asynchronously */
  44         union {
  45                 struct address_space *mapping;  /* If low bit clear, points to
  46                                                  * inode address_space, or NULL.
  47                                                  * If page mapped as anonymous
  48                                                  * memory, low bit is set, and
  49                                                  * it points to anon_vma object:
  50                                                  * see PAGE_MAPPING_ANON below.
  51                                                  */
  52                 void *s_mem;                    /* slab first object */
  53                 atomic_t compound_mapcount;     /* first tail page */
  54                 /* page_deferred_list().next     -- second tail page */
  55         };
  56
  57         /* Second double word */
  58         union {
  59                 pgoff_t index;          /* Our offset within mapping. */
  60                 void *freelist;         /* sl[aou]b first free object */
  61                 /* page_deferred_list().prev    -- second tail page */
  62         };
  63
  64         union {
  65 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
  66         defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
  67                 /* Used for cmpxchg_double in slub */
  68                 unsigned long counters;
  69 #else
  70                 /*
  71                  * Keep _refcount separate from slub cmpxchg_double data.
  72                  * As the rest of the double word is protected by slab_lock
  73                  * but _refcount is not.
  74                  */
  75                 unsigned counters;
  76 #endif
  77                 struct {
  78
  79                         union {
  80                                 /*
  81                                  * Count of ptes mapped in mms, to show when
  82                                  * page is mapped & limit reverse map searches.
  83                                  *
  84                                  * Extra information about page type may be
  85                                  * stored here for pages that are never mapped,
  86                                  * in which case the value MUST BE <= -2.
  87                                  * See page-flags.h for more details.
  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 #define PAGE_FRAG_CACHE_MAX_SIZE        __ALIGN_MASK(32768, ~PAGE_MASK)
 230 #define PAGE_FRAG_CACHE_MAX_ORDER       get_order(PAGE_FRAG_CACHE_MAX_SIZE)
 231
 232 struct page_frag_cache {
 233         void * va;
 234 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
 235         __u16 offset;
 236         __u16 size;
 237 #else
 238         __u32 offset;
 239 #endif
 240         /* we maintain a pagecount bias, so that we dont dirty cache line
 241          * containing page->_refcount every time we allocate a fragment.
 242          */
 243         unsigned int            pagecnt_bias;
 244         bool pfmemalloc;
 245 };
 246
 247 typedef unsigned long vm_flags_t;
 248
 249 /*
 250  * A region containing a mapping of a non-memory backed file under NOMMU
 251  * conditions.  These are held in a global tree and are pinned by the VMAs that
 252  * map parts of them.
 253  */
 254 struct vm_region {
 255         struct rb_node  vm_rb;          /* link in global region tree */
 256         vm_flags_t      vm_flags;       /* VMA vm_flags */
 257         unsigned long   vm_start;       /* start address of region */
 258         unsigned long   vm_end;         /* region initialised to here */
 259         unsigned long   vm_top;         /* region allocated to here */
 260         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 261         struct file     *vm_file;       /* the backing file or NULL */
 262
 263         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 264         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 265                                                 * this region */
 266 };
 267
 268 #ifdef CONFIG_USERFAULTFD
 269 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
 270 struct vm_userfaultfd_ctx {
 271         struct userfaultfd_ctx *ctx;
 272 };
 273 #else /* CONFIG_USERFAULTFD */
 274 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
 275 struct vm_userfaultfd_ctx {};
 276 #endif /* CONFIG_USERFAULTFD */
 277
 278 /*
 279  * This struct defines a memory VMM memory area. There is one of these
 280  * per VM-area/task.  A VM area is any part of the process virtual memory
 281  * space that has a special rule for the page-fault handlers (ie a shared
 282  * library, the executable area etc).
 283  */
 284 struct vm_area_struct {
 285         /* The first cache line has the info for VMA tree walking. */
 286
 287         unsigned long vm_start;         /* Our start address within vm_mm. */
 288         unsigned long vm_end;           /* The first byte after our end address
 289                                            within vm_mm. */
 290
 291         /* linked list of VM areas per task, sorted by address */
 292         struct vm_area_struct *vm_next, *vm_prev;
 293
 294         struct rb_node vm_rb;
 295
 296         /*
 297          * Largest free memory gap in bytes to the left of this VMA.
 298          * Either between this VMA and vma->vm_prev, or between one of the
 299          * VMAs below us in the VMA rbtree and its ->vm_prev. This helps
 300          * get_unmapped_area find a free area of the right size.
 301          */
 302         unsigned long rb_subtree_gap;
 303
 304         /* Second cache line starts here. */
 305
 306         struct mm_struct *vm_mm;        /* The address space we belong to. */
 307         pgprot_t vm_page_prot;          /* Access permissions of this VMA. */
 308         unsigned long vm_flags;         /* Flags, see mm.h. */
 309
 310         /*
 311          * For areas with an address space and backing store,
 312          * linkage into the address_space->i_mmap interval tree.
 313          */
 314         struct {
 315                 struct rb_node rb;
 316                 unsigned long rb_subtree_last;
 317         } shared;
 318
 319         /*
 320          * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
 321          * list, after a COW of one of the file pages.  A MAP_SHARED vma
 322          * can only be in the i_mmap tree.  An anonymous MAP_PRIVATE, stack
 323          * or brk vma (with NULL file) can only be in an anon_vma list.
 324          */
 325         struct list_head anon_vma_chain; /* Serialized by mmap_sem &
 326                                           * page_table_lock */
 327         struct anon_vma *anon_vma;      /* Serialized by page_table_lock */
 328
 329         /* Function pointers to deal with this struct. */
 330         const struct vm_operations_struct *vm_ops;
 331
 332         /* Information about our backing store: */
 333         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 334                                            units */
 335         struct file * vm_file;          /* File we map to (can be NULL). */
 336         void * vm_private_data;         /* was vm_pte (shared mem) */
 337
 338 #ifndef CONFIG_MMU
 339         struct vm_region *vm_region;    /* NOMMU mapping region */
 340 #endif
 341 #ifdef CONFIG_NUMA
 342         struct mempolicy *vm_policy;    /* NUMA policy for the VMA */
 343 #endif
 344         struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
 345 };
 346
 347 struct core_thread {
 348         struct task_struct *task;
 349         struct core_thread *next;
 350 };
 351
 352 struct core_state {
 353         atomic_t nr_threads;
 354         struct core_thread dumper;
 355         struct completion startup;
 356 };
 357
 358 struct kioctx_table;
 359 struct mm_struct {
 360         struct vm_area_struct *mmap;            /* list of VMAs */
 361         struct rb_root mm_rb;
 362         u32 vmacache_seqnum;                   /* per-thread vmacache */
 363 #ifdef CONFIG_MMU
 364         unsigned long (*get_unmapped_area) (struct file *filp,
 365                                 unsigned long addr, unsigned long len,
 366                                 unsigned long pgoff, unsigned long flags);
 367 #endif
 368         unsigned long mmap_base;                /* base of mmap area */
 369         unsigned long mmap_legacy_base;         /* base of mmap area in bottom-up allocations */
 370 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
 371         /* Base adresses for compatible mmap() */
 372         unsigned long mmap_compat_base;
 373         unsigned long mmap_compat_legacy_base;
 374 #endif
 375         unsigned long task_size;                /* size of task vm space */
 376         unsigned long highest_vm_end;           /* highest vma end address */
 377         pgd_t * pgd;
 378
 379         /**
 380          * @mm_users: The number of users including userspace.
 381          *
 382          * Use mmget()/mmget_not_zero()/mmput() to modify. When this drops
 383          * to 0 (i.e. when the task exits and there are no other temporary
 384          * reference holders), we also release a reference on @mm_count
 385          * (which may then free the &struct mm_struct if @mm_count also
 386          * drops to 0).
 387          */
 388         atomic_t mm_users;
 389
 390         /**
 391          * @mm_count: The number of references to &struct mm_struct
 392          * (@mm_users count as 1).
 393          *
 394          * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
 395          * &struct mm_struct is freed.
 396          */
 397         atomic_t mm_count;
 398
 399         atomic_long_t nr_ptes;                  /* PTE page table pages */
 400 #if CONFIG_PGTABLE_LEVELS > 2
 401         atomic_long_t nr_pmds;                  /* PMD page table pages */
 402 #endif
 403         int map_count;                          /* number of VMAs */
 404
 405         spinlock_t page_table_lock;             /* Protects page tables and some counters */
 406         struct rw_semaphore mmap_sem;
 407
 408         struct list_head mmlist;                /* List of maybe swapped mm's.  These are globally strung
 409                                                  * together off init_mm.mmlist, and are protected
 410                                                  * by mmlist_lock
 411                                                  */
 412
 413
 414         unsigned long hiwater_rss;      /* High-watermark of RSS usage */
 415         unsigned long hiwater_vm;       /* High-water virtual memory usage */
 416
 417         unsigned long total_vm;         /* Total pages mapped */
 418         unsigned long locked_vm;        /* Pages that have PG_mlocked set */
 419         unsigned long pinned_vm;        /* Refcount permanently increased */
 420         unsigned long data_vm;          /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
 421         unsigned long exec_vm;          /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
 422         unsigned long stack_vm;         /* VM_STACK */
 423         unsigned long def_flags;
 424         unsigned long start_code, end_code, start_data, end_data;
 425         unsigned long start_brk, brk, start_stack;
 426         unsigned long arg_start, arg_end, env_start, env_end;
 427
 428         unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
 429
 430         /*
 431          * Special counters, in some configurations protected by the
 432          * page_table_lock, in other configurations by being atomic.
 433          */
 434         struct mm_rss_stat rss_stat;
 435
 436         struct linux_binfmt *binfmt;
 437
 438         cpumask_var_t cpu_vm_mask_var;
 439
 440         /* Architecture-specific MM context */
 441         mm_context_t context;
 442
 443         unsigned long flags; /* Must use atomic bitops to access the bits */
 444
 445         struct core_state *core_state; /* coredumping support */
 446 #ifdef CONFIG_AIO
 447         spinlock_t                      ioctx_lock;
 448         struct kioctx_table __rcu       *ioctx_table;
 449 #endif
 450 #ifdef CONFIG_MEMCG
 451         /*
 452          * "owner" points to a task that is regarded as the canonical
 453          * user/owner of this mm. All of the following must be true in
 454          * order for it to be changed:
 455          *
 456          * current == mm->owner
 457          * current->mm != mm
 458          * new_owner->mm == mm
 459          * new_owner->alloc_lock is held
 460          */
 461         struct task_struct __rcu *owner;
 462 #endif
 463         struct user_namespace *user_ns;
 464
 465         /* store ref to file /proc/<pid>/exe symlink points to */
 466         struct file __rcu *exe_file;
 467 #ifdef CONFIG_MMU_NOTIFIER
 468         struct mmu_notifier_mm *mmu_notifier_mm;
 469 #endif
 470 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 471         pgtable_t pmd_huge_pte; /* protected by page_table_lock */
 472 #endif
 473 #ifdef CONFIG_CPUMASK_OFFSTACK
 474         struct cpumask cpumask_allocation;
 475 #endif
 476 #ifdef CONFIG_NUMA_BALANCING
 477         /*
 478          * numa_next_scan is the next time that the PTEs will be marked
 479          * pte_numa. NUMA hinting faults will gather statistics and migrate
 480          * pages to new nodes if necessary.
 481          */
 482         unsigned long numa_next_scan;
 483
 484         /* Restart point for scanning and setting pte_numa */
 485         unsigned long numa_scan_offset;
 486
 487         /* numa_scan_seq prevents two threads setting pte_numa */
 488         int numa_scan_seq;
 489 #endif
 490 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 491         /*
 492          * An operation with batched TLB flushing is going on. Anything that
 493          * can move process memory needs to flush the TLB when moving a
 494          * PROT_NONE or PROT_NUMA mapped page.
 495          */
 496         bool tlb_flush_pending;
 497 #endif
 498         struct uprobes_state uprobes_state;
 499 #ifdef CONFIG_HUGETLB_PAGE
 500         atomic_long_t hugetlb_usage;
 501 #endif
 502         struct work_struct async_put_work;
 503 };
 504
 505 extern struct mm_struct init_mm;
 506
 507 static inline void mm_init_cpumask(struct mm_struct *mm)
 508 {
 509 #ifdef CONFIG_CPUMASK_OFFSTACK
 510         mm->cpu_vm_mask_var = &mm->cpumask_allocation;
 511 #endif
 512         cpumask_clear(mm->cpu_vm_mask_var);
 513 }
 514
 515 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
 516 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
 517 {
 518         return mm->cpu_vm_mask_var;
 519 }
 520
 521 #if defined(CONFIG_NUMA_BALANCING) || defined(CONFIG_COMPACTION)
 522 /*
 523  * Memory barriers to keep this state in sync are graciously provided by
 524  * the page table locks, outside of which no page table modifications happen.
 525  * The barriers below prevent the compiler from re-ordering the instructions
 526  * around the memory barriers that are already present in the code.
 527  */
 528 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 529 {
 530         barrier();
 531         return mm->tlb_flush_pending;
 532 }
 533 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 534 {
 535         mm->tlb_flush_pending = true;
 536
 537         /*
 538          * Guarantee that the tlb_flush_pending store does not leak into the
 539          * critical section updating the page tables
 540          */
 541         smp_mb__before_spinlock();
 542 }
 543 /* Clearing is done after a TLB flush, which also provides a barrier. */
 544 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 545 {
 546         barrier();
 547         mm->tlb_flush_pending = false;
 548 }
 549 #else
 550 static inline bool mm_tlb_flush_pending(struct mm_struct *mm)
 551 {
 552         return false;
 553 }
 554 static inline void set_tlb_flush_pending(struct mm_struct *mm)
 555 {
 556 }
 557 static inline void clear_tlb_flush_pending(struct mm_struct *mm)
 558 {
 559 }
 560 #endif
 561
 562 struct vm_fault;
 563
 564 struct vm_special_mapping {
 565         const char *name;       /* The name, e.g. "[vdso]". */
 566
 567         /*
 568          * If .fault is not provided, this points to a
 569          * NULL-terminated array of pages that back the special mapping.
 570          *
 571          * This must not be NULL unless .fault is provided.
 572          */
 573         struct page **pages;
 574
 575         /*
 576          * If non-NULL, then this is called to resolve page faults
 577          * on the special mapping.  If used, .pages is not checked.
 578          */
 579         int (*fault)(const struct vm_special_mapping *sm,
 580                      struct vm_area_struct *vma,
 581                      struct vm_fault *vmf);
 582
 583         int (*mremap)(const struct vm_special_mapping *sm,
 584                      struct vm_area_struct *new_vma);
 585 };
 586
 587 enum tlb_flush_reason {
 588         TLB_FLUSH_ON_TASK_SWITCH,
 589         TLB_REMOTE_SHOOTDOWN,
 590         TLB_LOCAL_SHOOTDOWN,
 591         TLB_LOCAL_MM_SHOOTDOWN,
 592         TLB_REMOTE_SEND_IPI,
 593         NR_TLB_FLUSH_REASONS,
 594 };
 595
 596  /*
 597   * A swap entry has to fit into a "unsigned long", as the entry is hidden
 598   * in the "index" field of the swapper address space.
 599   */
 600 typedef struct {
 601         unsigned long val;
 602 } swp_entry_t;
 603
 604 #endif /* _LINUX_MM_TYPES_H */