1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MM_TYPES_H
3 #define _LINUX_MM_TYPES_H
4
5 #include <linux/mm_types_task.h>
6
7 #include <linux/auxvec.h>
8 #include <linux/kref.h>
9 #include <linux/list.h>
10 #include <linux/spinlock.h>
11 #include <linux/rbtree.h>
12 #include <linux/maple_tree.h>
13 #include <linux/rwsem.h>
14 #include <linux/completion.h>
15 #include <linux/cpumask.h>
16 #include <linux/uprobes.h>
17 #include <linux/rcupdate.h>
18 #include <linux/page-flags-layout.h>
19 #include <linux/workqueue.h>
20 #include <linux/seqlock.h>
21 #include <linux/percpu_counter.h>
22
23 #include <asm/mmu.h>
24
25 #ifndef AT_VECTOR_SIZE_ARCH
26 #define AT_VECTOR_SIZE_ARCH 0
27 #endif
28 #define AT_VECTOR_SIZE (2*(AT_VECTOR_SIZE_ARCH + AT_VECTOR_SIZE_BASE + 1))
29
30 #define INIT_PASID 0
31
32 struct address_space;
33 struct mem_cgroup;
34
35 /*
36 * Each physical page in the system has a struct page associated with
37 * it to keep track of whatever it is we are using the page for at the
38 * moment. Note that we have no way to track which tasks are using
39 * a page, though if it is a pagecache page, rmap structures can tell us
40 * who is mapping it.
41 *
42 * If you allocate the page using alloc_pages(), you can use some of the
43 * space in struct page for your own purposes. The five words in the main
44 * union are available, except for bit 0 of the first word which must be
45 * kept clear. Many users use this word to store a pointer to an object
46 * which is guaranteed to be aligned. If you use the same storage as
47 * page->mapping, you must restore it to NULL before freeing the page.
48 *
49 * If your page will not be mapped to userspace, you can also use the four
50 * bytes in the mapcount union, but you must call page_mapcount_reset()
51 * before freeing it.
52 *
53 * If you want to use the refcount field, it must be used in such a way
54 * that other CPUs temporarily incrementing and then decrementing the
55 * refcount does not cause problems. On receiving the page from
56 * alloc_pages(), the refcount will be positive.
57 *
58 * If you allocate pages of order > 0, you can use some of the fields
59 * in each subpage, but you may need to restore some of their values
60 * afterwards.
61 *
62 * SLUB uses cmpxchg_double() to atomically update its freelist and counters.
63 * That requires that freelist & counters in struct slab be adjacent and
64 * double-word aligned. Because struct slab currently just reinterprets the
65 * bits of struct page, we align all struct pages to double-word boundaries,
66 * and ensure that 'freelist' is aligned within struct slab.
67 */
68 #ifdef CONFIG_HAVE_ALIGNED_STRUCT_PAGE
69 #define _struct_page_alignment __aligned(2 * sizeof(unsigned long))
70 #else
71 #define _struct_page_alignment __aligned(sizeof(unsigned long))
72 #endif
73
74 struct page {
75 unsigned long flags; /* Atomic flags, some possibly
76 * updated asynchronously */
77 /*
78 * Five words (20/40 bytes) are available in this union.
79 * WARNING: bit 0 of the first word is used for PageTail(). That
80 * means the other users of this union MUST NOT use the bit to
81 * avoid collision and false-positive PageTail().
82 */
83 union {
84 struct { /* Page cache and anonymous pages */
85 /**
86 * @lru: Pageout list, eg. active_list protected by
87 * lruvec->lru_lock. Sometimes used as a generic list
88 * by the page owner.
89 */
90 union {
91 struct list_head lru;
92
93 /* Or, for the Unevictable "LRU list" slot */
94 struct {
95 /* Always even, to negate PageTail */
96 void *__filler;
97 /* Count page's or folio's mlocks */
98 unsigned int mlock_count;
99 };
100
101 /* Or, free page */
102 struct list_head buddy_list;
103 struct list_head pcp_list;
104 };
105 /* See page-flags.h for PAGE_MAPPING_FLAGS */
106 struct address_space *mapping;
107 union {
108 pgoff_t index; /* Our offset within mapping. */
109 unsigned long share; /* share count for fsdax */
110 };
111 /**
112 * @private: Mapping-private opaque data.
113 * Usually used for buffer_heads if PagePrivate.
114 * Used for swp_entry_t if PageSwapCache.
115 * Indicates order in the buddy system if PageBuddy.
116 */
117 unsigned long private;
118 };
119 struct { /* page_pool used by netstack */
120 /**
121 * @pp_magic: magic value to avoid recycling non
122 * page_pool allocated pages.
123 */
124 unsigned long pp_magic;
125 struct page_pool *pp;
126 unsigned long _pp_mapping_pad;
127 unsigned long dma_addr;
128 union {
129 /**
130 * dma_addr_upper: might require a 64-bit
131 * value on 32-bit architectures.
132 */
133 unsigned long dma_addr_upper;
134 /**
135 * For frag page support, not supported in
136 * 32-bit architectures with 64-bit DMA.
137 */
138 atomic_long_t pp_frag_count;
139 };
140 };
141 struct { /* Tail pages of compound page */
142 unsigned long compound_head; /* Bit zero is set */
143 };
144 struct { /* Page table pages */
145 unsigned long _pt_pad_1; /* compound_head */
146 pgtable_t pmd_huge_pte; /* protected by page->ptl */
147 unsigned long _pt_pad_2; /* mapping */
148 union {
149 struct mm_struct *pt_mm; /* x86 pgds only */
150 atomic_t pt_frag_refcount; /* powerpc */
151 };
152 #if ALLOC_SPLIT_PTLOCKS
153 spinlock_t *ptl;
154 #else
155 spinlock_t ptl;
156 #endif
157 };
158 struct { /* ZONE_DEVICE pages */
159 /** @pgmap: Points to the hosting device page map. */
160 struct dev_pagemap *pgmap;
161 void *zone_device_data;
162 /*
163 * ZONE_DEVICE private pages are counted as being
164 * mapped so the next 3 words hold the mapping, index,
165 * and private fields from the source anonymous or
166 * page cache page while the page is migrated to device
167 * private memory.
168 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
169 * use the mapping, index, and private fields when
170 * pmem backed DAX files are mapped.
171 */
172 };
173
174 /** @rcu_head: You can use this to free a page by RCU. */
175 struct rcu_head rcu_head;
176 };
177
178 union { /* This union is 4 bytes in size. */
179 /*
180 * If the page can be mapped to userspace, encodes the number
181 * of times this page is referenced by a page table.
182 */
183 atomic_t _mapcount;
184
185 /*
186 * If the page is neither PageSlab nor mappable to userspace,
187 * the value stored here may help determine what this page
188 * is used for. See page-flags.h for a list of page types
189 * which are currently stored here.
190 */
191 unsigned int page_type;
192 };
193
194 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
195 atomic_t _refcount;
196
197 #ifdef CONFIG_MEMCG
198 unsigned long memcg_data;
199 #endif
200
201 /*
202 * On machines where all RAM is mapped into kernel address space,
203 * we can simply calculate the virtual address. On machines with
204 * highmem some memory is mapped into kernel virtual memory
205 * dynamically, so we need a place to store that address.
206 * Note that this field could be 16 bits on x86 ... ;)
207 *
208 * Architectures with slow multiplication can define
209 * WANT_PAGE_VIRTUAL in asm/page.h
210 */
211 #if defined(WANT_PAGE_VIRTUAL)
212 void *virtual; /* Kernel virtual address (NULL if
213 not kmapped, ie. highmem) */
214 #endif /* WANT_PAGE_VIRTUAL */
215
216 #ifdef CONFIG_KMSAN
217 /*
218 * KMSAN metadata for this page:
219 * - shadow page: every bit indicates whether the corresponding
220 * bit of the original page is initialized (0) or not (1);
221 * - origin page: every 4 bytes contain an id of the stack trace
222 * where the uninitialized value was created.
223 */
224 struct page *kmsan_shadow;
225 struct page *kmsan_origin;
226 #endif
227
228 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
229 int _last_cpupid;
230 #endif
231 } _struct_page_alignment;
232
233 /*
234 * struct encoded_page - a nonexistent type marking this pointer
235 *
236 * An 'encoded_page' pointer is a pointer to a regular 'struct page', but
237 * with the low bits of the pointer indicating extra context-dependent
238 * information. Not super-common, but happens in mmu_gather and mlock
239 * handling, and this acts as a type system check on that use.
240 *
241 * We only really have two guaranteed bits in general, although you could
242 * play with 'struct page' alignment (see CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
243 * for more.
244 *
245 * Use the supplied helper functions to endcode/decode the pointer and bits.
246 */
247 struct encoded_page;
248 #define ENCODE_PAGE_BITS 3ul
encode_page(struct page * page,unsigned long flags)249 static __always_inline struct encoded_page *encode_page(struct page *page, unsigned long flags)
250 {
251 BUILD_BUG_ON(flags > ENCODE_PAGE_BITS);
252 return (struct encoded_page *)(flags | (unsigned long)page);
253 }
254
encoded_page_flags(struct encoded_page * page)255 static inline unsigned long encoded_page_flags(struct encoded_page *page)
256 {
257 return ENCODE_PAGE_BITS & (unsigned long)page;
258 }
259
encoded_page_ptr(struct encoded_page * page)260 static inline struct page *encoded_page_ptr(struct encoded_page *page)
261 {
262 return (struct page *)(~ENCODE_PAGE_BITS & (unsigned long)page);
263 }
264
265 /**
266 * struct folio - Represents a contiguous set of bytes.
267 * @flags: Identical to the page flags.
268 * @lru: Least Recently Used list; tracks how recently this folio was used.
269 * @mlock_count: Number of times this folio has been pinned by mlock().
270 * @mapping: The file this page belongs to, or refers to the anon_vma for
271 * anonymous memory.
272 * @index: Offset within the file, in units of pages. For anonymous memory,
273 * this is the index from the beginning of the mmap.
274 * @private: Filesystem per-folio data (see folio_attach_private()).
275 * Used for swp_entry_t if folio_test_swapcache().
276 * @_mapcount: Do not access this member directly. Use folio_mapcount() to
277 * find out how many times this folio is mapped by userspace.
278 * @_refcount: Do not access this member directly. Use folio_ref_count()
279 * to find how many references there are to this folio.
280 * @memcg_data: Memory Control Group data.
281 * @_folio_dtor: Which destructor to use for this folio.
282 * @_folio_order: Do not use directly, call folio_order().
283 * @_entire_mapcount: Do not use directly, call folio_entire_mapcount().
284 * @_nr_pages_mapped: Do not use directly, call folio_mapcount().
285 * @_pincount: Do not use directly, call folio_maybe_dma_pinned().
286 * @_folio_nr_pages: Do not use directly, call folio_nr_pages().
287 * @_hugetlb_subpool: Do not use directly, use accessor in hugetlb.h.
288 * @_hugetlb_cgroup: Do not use directly, use accessor in hugetlb_cgroup.h.
289 * @_hugetlb_cgroup_rsvd: Do not use directly, use accessor in hugetlb_cgroup.h.
290 * @_hugetlb_hwpoison: Do not use directly, call raw_hwp_list_head().
291 * @_deferred_list: Folios to be split under memory pressure.
292 *
293 * A folio is a physically, virtually and logically contiguous set
294 * of bytes. It is a power-of-two in size, and it is aligned to that
295 * same power-of-two. It is at least as large as %PAGE_SIZE. If it is
296 * in the page cache, it is at a file offset which is a multiple of that
297 * power-of-two. It may be mapped into userspace at an address which is
298 * at an arbitrary page offset, but its kernel virtual address is aligned
299 * to its size.
300 */
301 struct folio {
302 /* private: don't document the anon union */
303 union {
304 struct {
305 /* public: */
306 unsigned long flags;
307 union {
308 struct list_head lru;
309 /* private: avoid cluttering the output */
310 struct {
311 void *__filler;
312 /* public: */
313 unsigned int mlock_count;
314 /* private: */
315 };
316 /* public: */
317 };
318 struct address_space *mapping;
319 pgoff_t index;
320 void *private;
321 atomic_t _mapcount;
322 atomic_t _refcount;
323 #ifdef CONFIG_MEMCG
324 unsigned long memcg_data;
325 #endif
326 /* private: the union with struct page is transitional */
327 };
328 struct page page;
329 };
330 union {
331 struct {
332 unsigned long _flags_1;
333 unsigned long _head_1;
334 /* public: */
335 unsigned char _folio_dtor;
336 unsigned char _folio_order;
337 atomic_t _entire_mapcount;
338 atomic_t _nr_pages_mapped;
339 atomic_t _pincount;
340 #ifdef CONFIG_64BIT
341 unsigned int _folio_nr_pages;
342 #endif
343 /* private: the union with struct page is transitional */
344 };
345 struct page __page_1;
346 };
347 union {
348 struct {
349 unsigned long _flags_2;
350 unsigned long _head_2;
351 /* public: */
352 void *_hugetlb_subpool;
353 void *_hugetlb_cgroup;
354 void *_hugetlb_cgroup_rsvd;
355 void *_hugetlb_hwpoison;
356 /* private: the union with struct page is transitional */
357 };
358 struct {
359 unsigned long _flags_2a;
360 unsigned long _head_2a;
361 /* public: */
362 struct list_head _deferred_list;
363 /* private: the union with struct page is transitional */
364 };
365 struct page __page_2;
366 };
367 };
368
369 #define FOLIO_MATCH(pg, fl) \
370 static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl))
371 FOLIO_MATCH(flags, flags);
372 FOLIO_MATCH(lru, lru);
373 FOLIO_MATCH(mapping, mapping);
374 FOLIO_MATCH(compound_head, lru);
375 FOLIO_MATCH(index, index);
376 FOLIO_MATCH(private, private);
377 FOLIO_MATCH(_mapcount, _mapcount);
378 FOLIO_MATCH(_refcount, _refcount);
379 #ifdef CONFIG_MEMCG
380 FOLIO_MATCH(memcg_data, memcg_data);
381 #endif
382 #undef FOLIO_MATCH
383 #define FOLIO_MATCH(pg, fl) \
384 static_assert(offsetof(struct folio, fl) == \
385 offsetof(struct page, pg) + sizeof(struct page))
386 FOLIO_MATCH(flags, _flags_1);
387 FOLIO_MATCH(compound_head, _head_1);
388 #undef FOLIO_MATCH
389 #define FOLIO_MATCH(pg, fl) \
390 static_assert(offsetof(struct folio, fl) == \
391 offsetof(struct page, pg) + 2 * sizeof(struct page))
392 FOLIO_MATCH(flags, _flags_2);
393 FOLIO_MATCH(compound_head, _head_2);
394 #undef FOLIO_MATCH
395
396 /*
397 * Used for sizing the vmemmap region on some architectures
398 */
399 #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page)))
400
401 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
402 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
403
404 /*
405 * page_private can be used on tail pages. However, PagePrivate is only
406 * checked by the VM on the head page. So page_private on the tail pages
407 * should be used for data that's ancillary to the head page (eg attaching
408 * buffer heads to tail pages after attaching buffer heads to the head page)
409 */
410 #define page_private(page) ((page)->private)
411
set_page_private(struct page * page,unsigned long private)412 static inline void set_page_private(struct page *page, unsigned long private)
413 {
414 page->private = private;
415 }
416
folio_get_private(struct folio * folio)417 static inline void *folio_get_private(struct folio *folio)
418 {
419 return folio->private;
420 }
421
422 struct page_frag_cache {
423 void * va;
424 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
425 __u16 offset;
426 __u16 size;
427 #else
428 __u32 offset;
429 #endif
430 /* we maintain a pagecount bias, so that we dont dirty cache line
431 * containing page->_refcount every time we allocate a fragment.
432 */
433 unsigned int pagecnt_bias;
434 bool pfmemalloc;
435 };
436
437 typedef unsigned long vm_flags_t;
438
439 /*
440 * A region containing a mapping of a non-memory backed file under NOMMU
441 * conditions. These are held in a global tree and are pinned by the VMAs that
442 * map parts of them.
443 */
444 struct vm_region {
445 struct rb_node vm_rb; /* link in global region tree */
446 vm_flags_t vm_flags; /* VMA vm_flags */
447 unsigned long vm_start; /* start address of region */
448 unsigned long vm_end; /* region initialised to here */
449 unsigned long vm_top; /* region allocated to here */
450 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
451 struct file *vm_file; /* the backing file or NULL */
452
453 int vm_usage; /* region usage count (access under nommu_region_sem) */
454 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
455 * this region */
456 };
457
458 #ifdef CONFIG_USERFAULTFD
459 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
460 struct vm_userfaultfd_ctx {
461 struct userfaultfd_ctx *ctx;
462 };
463 #else /* CONFIG_USERFAULTFD */
464 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
465 struct vm_userfaultfd_ctx {};
466 #endif /* CONFIG_USERFAULTFD */
467
468 struct anon_vma_name {
469 struct kref kref;
470 /* The name needs to be at the end because it is dynamically sized. */
471 char name[];
472 };
473
474 /*
475 * This struct describes a virtual memory area. There is one of these
476 * per VM-area/task. A VM area is any part of the process virtual memory
477 * space that has a special rule for the page-fault handlers (ie a shared
478 * library, the executable area etc).
479 */
480 struct vm_area_struct {
481 /* The first cache line has the info for VMA tree walking. */
482
483 unsigned long vm_start; /* Our start address within vm_mm. */
484 unsigned long vm_end; /* The first byte after our end address
485 within vm_mm. */
486
487 struct mm_struct *vm_mm; /* The address space we belong to. */
488
489 /*
490 * Access permissions of this VMA.
491 * See vmf_insert_mixed_prot() for discussion.
492 */
493 pgprot_t vm_page_prot;
494
495 /*
496 * Flags, see mm.h.
497 * To modify use vm_flags_{init|reset|set|clear|mod} functions.
498 */
499 union {
500 const vm_flags_t vm_flags;
501 vm_flags_t __private __vm_flags;
502 };
503
504 /*
505 * For areas with an address space and backing store,
506 * linkage into the address_space->i_mmap interval tree.
507 *
508 */
509 struct {
510 struct rb_node rb;
511 unsigned long rb_subtree_last;
512 } shared;
513
514 /*
515 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
516 * list, after a COW of one of the file pages. A MAP_SHARED vma
517 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
518 * or brk vma (with NULL file) can only be in an anon_vma list.
519 */
520 struct list_head anon_vma_chain; /* Serialized by mmap_lock &
521 * page_table_lock */
522 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
523
524 /* Function pointers to deal with this struct. */
525 const struct vm_operations_struct *vm_ops;
526
527 /* Information about our backing store: */
528 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
529 units */
530 struct file * vm_file; /* File we map to (can be NULL). */
531 void * vm_private_data; /* was vm_pte (shared mem) */
532
533 #ifdef CONFIG_ANON_VMA_NAME
534 /*
535 * For private and shared anonymous mappings, a pointer to a null
536 * terminated string containing the name given to the vma, or NULL if
537 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access.
538 */
539 struct anon_vma_name *anon_name;
540 #endif
541 #ifdef CONFIG_SWAP
542 atomic_long_t swap_readahead_info;
543 #endif
544 #ifndef CONFIG_MMU
545 struct vm_region *vm_region; /* NOMMU mapping region */
546 #endif
547 #ifdef CONFIG_NUMA
548 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
549 #endif
550 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
551 } __randomize_layout;
552
553 struct kioctx_table;
554 struct mm_struct {
555 struct {
556 struct maple_tree mm_mt;
557 #ifdef CONFIG_MMU
558 unsigned long (*get_unmapped_area) (struct file *filp,
559 unsigned long addr, unsigned long len,
560 unsigned long pgoff, unsigned long flags);
561 #endif
562 unsigned long mmap_base; /* base of mmap area */
563 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
564 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
565 /* Base addresses for compatible mmap() */
566 unsigned long mmap_compat_base;
567 unsigned long mmap_compat_legacy_base;
568 #endif
569 unsigned long task_size; /* size of task vm space */
570 pgd_t * pgd;
571
572 #ifdef CONFIG_MEMBARRIER
573 /**
574 * @membarrier_state: Flags controlling membarrier behavior.
575 *
576 * This field is close to @pgd to hopefully fit in the same
577 * cache-line, which needs to be touched by switch_mm().
578 */
579 atomic_t membarrier_state;
580 #endif
581
582 /**
583 * @mm_users: The number of users including userspace.
584 *
585 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
586 * drops to 0 (i.e. when the task exits and there are no other
587 * temporary reference holders), we also release a reference on
588 * @mm_count (which may then free the &struct mm_struct if
589 * @mm_count also drops to 0).
590 */
591 atomic_t mm_users;
592
593 /**
594 * @mm_count: The number of references to &struct mm_struct
595 * (@mm_users count as 1).
596 *
597 * Use mmgrab()/mmdrop() to modify. When this drops to 0, the
598 * &struct mm_struct is freed.
599 */
600 atomic_t mm_count;
601 #ifdef CONFIG_SCHED_MM_CID
602 /**
603 * @cid_lock: Protect cid bitmap updates vs lookups.
604 *
605 * Prevent situations where updates to the cid bitmap happen
606 * concurrently with lookups. Those can lead to situations
607 * where a lookup cannot find a free bit simply because it was
608 * unlucky enough to load, non-atomically, bitmap words as they
609 * were being concurrently updated by the updaters.
610 */
611 raw_spinlock_t cid_lock;
612 #endif
613 #ifdef CONFIG_MMU
614 atomic_long_t pgtables_bytes; /* size of all page tables */
615 #endif
616 int map_count; /* number of VMAs */
617
618 spinlock_t page_table_lock; /* Protects page tables and some
619 * counters
620 */
621 /*
622 * With some kernel config, the current mmap_lock's offset
623 * inside 'mm_struct' is at 0x120, which is very optimal, as
624 * its two hot fields 'count' and 'owner' sit in 2 different
625 * cachelines, and when mmap_lock is highly contended, both
626 * of the 2 fields will be accessed frequently, current layout
627 * will help to reduce cache bouncing.
628 *
629 * So please be careful with adding new fields before
630 * mmap_lock, which can easily push the 2 fields into one
631 * cacheline.
632 */
633 struct rw_semaphore mmap_lock;
634
635 struct list_head mmlist; /* List of maybe swapped mm's. These
636 * are globally strung together off
637 * init_mm.mmlist, and are protected
638 * by mmlist_lock
639 */
640
641
642 unsigned long hiwater_rss; /* High-watermark of RSS usage */
643 unsigned long hiwater_vm; /* High-water virtual memory usage */
644
645 unsigned long total_vm; /* Total pages mapped */
646 unsigned long locked_vm; /* Pages that have PG_mlocked set */
647 atomic64_t pinned_vm; /* Refcount permanently increased */
648 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
649 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
650 unsigned long stack_vm; /* VM_STACK */
651 unsigned long def_flags;
652
653 /**
654 * @write_protect_seq: Locked when any thread is write
655 * protecting pages mapped by this mm to enforce a later COW,
656 * for instance during page table copying for fork().
657 */
658 seqcount_t write_protect_seq;
659
660 spinlock_t arg_lock; /* protect the below fields */
661
662 unsigned long start_code, end_code, start_data, end_data;
663 unsigned long start_brk, brk, start_stack;
664 unsigned long arg_start, arg_end, env_start, env_end;
665
666 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
667
668 struct percpu_counter rss_stat[NR_MM_COUNTERS];
669
670 struct linux_binfmt *binfmt;
671
672 /* Architecture-specific MM context */
673 mm_context_t context;
674
675 unsigned long flags; /* Must use atomic bitops to access */
676
677 #ifdef CONFIG_AIO
678 spinlock_t ioctx_lock;
679 struct kioctx_table __rcu *ioctx_table;
680 #endif
681 #ifdef CONFIG_MEMCG
682 /*
683 * "owner" points to a task that is regarded as the canonical
684 * user/owner of this mm. All of the following must be true in
685 * order for it to be changed:
686 *
687 * current == mm->owner
688 * current->mm != mm
689 * new_owner->mm == mm
690 * new_owner->alloc_lock is held
691 */
692 struct task_struct __rcu *owner;
693 #endif
694 struct user_namespace *user_ns;
695
696 /* store ref to file /proc/<pid>/exe symlink points to */
697 struct file __rcu *exe_file;
698 #ifdef CONFIG_MMU_NOTIFIER
699 struct mmu_notifier_subscriptions *notifier_subscriptions;
700 #endif
701 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
702 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
703 #endif
704 #ifdef CONFIG_NUMA_BALANCING
705 /*
706 * numa_next_scan is the next time that PTEs will be remapped
707 * PROT_NONE to trigger NUMA hinting faults; such faults gather
708 * statistics and migrate pages to new nodes if necessary.
709 */
710 unsigned long numa_next_scan;
711
712 /* Restart point for scanning and remapping PTEs. */
713 unsigned long numa_scan_offset;
714
715 /* numa_scan_seq prevents two threads remapping PTEs. */
716 int numa_scan_seq;
717 #endif
718 /*
719 * An operation with batched TLB flushing is going on. Anything
720 * that can move process memory needs to flush the TLB when
721 * moving a PROT_NONE mapped page.
722 */
723 atomic_t tlb_flush_pending;
724 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
725 /* See flush_tlb_batched_pending() */
726 atomic_t tlb_flush_batched;
727 #endif
728 struct uprobes_state uprobes_state;
729 #ifdef CONFIG_PREEMPT_RT
730 struct rcu_head delayed_drop;
731 #endif
732 #ifdef CONFIG_HUGETLB_PAGE
733 atomic_long_t hugetlb_usage;
734 #endif
735 struct work_struct async_put_work;
736
737 #ifdef CONFIG_IOMMU_SVA
738 u32 pasid;
739 #endif
740 #ifdef CONFIG_KSM
741 /*
742 * Represent how many pages of this process are involved in KSM
743 * merging.
744 */
745 unsigned long ksm_merging_pages;
746 /*
747 * Represent how many pages are checked for ksm merging
748 * including merged and not merged.
749 */
750 unsigned long ksm_rmap_items;
751 #endif
752 #ifdef CONFIG_LRU_GEN
753 struct {
754 /* this mm_struct is on lru_gen_mm_list */
755 struct list_head list;
756 /*
757 * Set when switching to this mm_struct, as a hint of
758 * whether it has been used since the last time per-node
759 * page table walkers cleared the corresponding bits.
760 */
761 unsigned long bitmap;
762 #ifdef CONFIG_MEMCG
763 /* points to the memcg of "owner" above */
764 struct mem_cgroup *memcg;
765 #endif
766 } lru_gen;
767 #endif /* CONFIG_LRU_GEN */
768 } __randomize_layout;
769
770 /*
771 * The mm_cpumask needs to be at the end of mm_struct, because it
772 * is dynamically sized based on nr_cpu_ids.
773 */
774 unsigned long cpu_bitmap[];
775 };
776
777 #define MM_MT_FLAGS (MT_FLAGS_ALLOC_RANGE | MT_FLAGS_LOCK_EXTERN)
778 extern struct mm_struct init_mm;
779
780 /* Pointer magic because the dynamic array size confuses some compilers. */
mm_init_cpumask(struct mm_struct * mm)781 static inline void mm_init_cpumask(struct mm_struct *mm)
782 {
783 unsigned long cpu_bitmap = (unsigned long)mm;
784
785 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
786 cpumask_clear((struct cpumask *)cpu_bitmap);
787 }
788
789 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
mm_cpumask(struct mm_struct * mm)790 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
791 {
792 return (struct cpumask *)&mm->cpu_bitmap;
793 }
794
795 #ifdef CONFIG_LRU_GEN
796
797 struct lru_gen_mm_list {
798 /* mm_struct list for page table walkers */
799 struct list_head fifo;
800 /* protects the list above */
801 spinlock_t lock;
802 };
803
804 void lru_gen_add_mm(struct mm_struct *mm);
805 void lru_gen_del_mm(struct mm_struct *mm);
806 #ifdef CONFIG_MEMCG
807 void lru_gen_migrate_mm(struct mm_struct *mm);
808 #endif
809
lru_gen_init_mm(struct mm_struct * mm)810 static inline void lru_gen_init_mm(struct mm_struct *mm)
811 {
812 INIT_LIST_HEAD(&mm->lru_gen.list);
813 mm->lru_gen.bitmap = 0;
814 #ifdef CONFIG_MEMCG
815 mm->lru_gen.memcg = NULL;
816 #endif
817 }
818
lru_gen_use_mm(struct mm_struct * mm)819 static inline void lru_gen_use_mm(struct mm_struct *mm)
820 {
821 /*
822 * When the bitmap is set, page reclaim knows this mm_struct has been
823 * used since the last time it cleared the bitmap. So it might be worth
824 * walking the page tables of this mm_struct to clear the accessed bit.
825 */
826 WRITE_ONCE(mm->lru_gen.bitmap, -1);
827 }
828
829 #else /* !CONFIG_LRU_GEN */
830
lru_gen_add_mm(struct mm_struct * mm)831 static inline void lru_gen_add_mm(struct mm_struct *mm)
832 {
833 }
834
lru_gen_del_mm(struct mm_struct * mm)835 static inline void lru_gen_del_mm(struct mm_struct *mm)
836 {
837 }
838
839 #ifdef CONFIG_MEMCG
lru_gen_migrate_mm(struct mm_struct * mm)840 static inline void lru_gen_migrate_mm(struct mm_struct *mm)
841 {
842 }
843 #endif
844
lru_gen_init_mm(struct mm_struct * mm)845 static inline void lru_gen_init_mm(struct mm_struct *mm)
846 {
847 }
848
lru_gen_use_mm(struct mm_struct * mm)849 static inline void lru_gen_use_mm(struct mm_struct *mm)
850 {
851 }
852
853 #endif /* CONFIG_LRU_GEN */
854
855 struct vma_iterator {
856 struct ma_state mas;
857 };
858
859 #define VMA_ITERATOR(name, __mm, __addr) \
860 struct vma_iterator name = { \
861 .mas = { \
862 .tree = &(__mm)->mm_mt, \
863 .index = __addr, \
864 .node = MAS_START, \
865 }, \
866 }
867
vma_iter_init(struct vma_iterator * vmi,struct mm_struct * mm,unsigned long addr)868 static inline void vma_iter_init(struct vma_iterator *vmi,
869 struct mm_struct *mm, unsigned long addr)
870 {
871 mas_init(&vmi->mas, &mm->mm_mt, addr);
872 }
873
874 #ifdef CONFIG_SCHED_MM_CID
875 /* Accessor for struct mm_struct's cidmask. */
mm_cidmask(struct mm_struct * mm)876 static inline cpumask_t *mm_cidmask(struct mm_struct *mm)
877 {
878 unsigned long cid_bitmap = (unsigned long)mm;
879
880 cid_bitmap += offsetof(struct mm_struct, cpu_bitmap);
881 /* Skip cpu_bitmap */
882 cid_bitmap += cpumask_size();
883 return (struct cpumask *)cid_bitmap;
884 }
885
mm_init_cid(struct mm_struct * mm)886 static inline void mm_init_cid(struct mm_struct *mm)
887 {
888 raw_spin_lock_init(&mm->cid_lock);
889 cpumask_clear(mm_cidmask(mm));
890 }
891
mm_cid_size(void)892 static inline unsigned int mm_cid_size(void)
893 {
894 return cpumask_size();
895 }
896 #else /* CONFIG_SCHED_MM_CID */
mm_init_cid(struct mm_struct * mm)897 static inline void mm_init_cid(struct mm_struct *mm) { }
mm_cid_size(void)898 static inline unsigned int mm_cid_size(void)
899 {
900 return 0;
901 }
902 #endif /* CONFIG_SCHED_MM_CID */
903
904 struct mmu_gather;
905 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
906 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
907 extern void tlb_finish_mmu(struct mmu_gather *tlb);
908
909 struct vm_fault;
910
911 /**
912 * typedef vm_fault_t - Return type for page fault handlers.
913 *
914 * Page fault handlers return a bitmask of %VM_FAULT values.
915 */
916 typedef __bitwise unsigned int vm_fault_t;
917
918 /**
919 * enum vm_fault_reason - Page fault handlers return a bitmask of
920 * these values to tell the core VM what happened when handling the
921 * fault. Used to decide whether a process gets delivered SIGBUS or
922 * just gets major/minor fault counters bumped up.
923 *
924 * @VM_FAULT_OOM: Out Of Memory
925 * @VM_FAULT_SIGBUS: Bad access
926 * @VM_FAULT_MAJOR: Page read from storage
927 * @VM_FAULT_HWPOISON: Hit poisoned small page
928 * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
929 * in upper bits
930 * @VM_FAULT_SIGSEGV: segmentation fault
931 * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
932 * @VM_FAULT_LOCKED: ->fault locked the returned page
933 * @VM_FAULT_RETRY: ->fault blocked, must retry
934 * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small
935 * @VM_FAULT_DONE_COW: ->fault has fully handled COW
936 * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs
937 * fsync() to complete (for synchronous page faults
938 * in DAX)
939 * @VM_FAULT_COMPLETED: ->fault completed, meanwhile mmap lock released
940 * @VM_FAULT_HINDEX_MASK: mask HINDEX value
941 *
942 */
943 enum vm_fault_reason {
944 VM_FAULT_OOM = (__force vm_fault_t)0x000001,
945 VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002,
946 VM_FAULT_MAJOR = (__force vm_fault_t)0x000004,
947 VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
948 VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
949 VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
950 VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
951 VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
952 VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
953 VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800,
954 VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000,
955 VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000,
956 VM_FAULT_COMPLETED = (__force vm_fault_t)0x004000,
957 VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000,
958 };
959
960 /* Encode hstate index for a hwpoisoned large page */
961 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
962 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
963
964 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \
965 VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \
966 VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
967
968 #define VM_FAULT_RESULT_TRACE \
969 { VM_FAULT_OOM, "OOM" }, \
970 { VM_FAULT_SIGBUS, "SIGBUS" }, \
971 { VM_FAULT_MAJOR, "MAJOR" }, \
972 { VM_FAULT_HWPOISON, "HWPOISON" }, \
973 { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
974 { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
975 { VM_FAULT_NOPAGE, "NOPAGE" }, \
976 { VM_FAULT_LOCKED, "LOCKED" }, \
977 { VM_FAULT_RETRY, "RETRY" }, \
978 { VM_FAULT_FALLBACK, "FALLBACK" }, \
979 { VM_FAULT_DONE_COW, "DONE_COW" }, \
980 { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }
981
982 struct vm_special_mapping {
983 const char *name; /* The name, e.g. "[vdso]". */
984
985 /*
986 * If .fault is not provided, this points to a
987 * NULL-terminated array of pages that back the special mapping.
988 *
989 * This must not be NULL unless .fault is provided.
990 */
991 struct page **pages;
992
993 /*
994 * If non-NULL, then this is called to resolve page faults
995 * on the special mapping. If used, .pages is not checked.
996 */
997 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
998 struct vm_area_struct *vma,
999 struct vm_fault *vmf);
1000
1001 int (*mremap)(const struct vm_special_mapping *sm,
1002 struct vm_area_struct *new_vma);
1003 };
1004
1005 enum tlb_flush_reason {
1006 TLB_FLUSH_ON_TASK_SWITCH,
1007 TLB_REMOTE_SHOOTDOWN,
1008 TLB_LOCAL_SHOOTDOWN,
1009 TLB_LOCAL_MM_SHOOTDOWN,
1010 TLB_REMOTE_SEND_IPI,
1011 NR_TLB_FLUSH_REASONS,
1012 };
1013
1014 /*
1015 * A swap entry has to fit into a "unsigned long", as the entry is hidden
1016 * in the "index" field of the swapper address space.
1017 */
1018 typedef struct {
1019 unsigned long val;
1020 } swp_entry_t;
1021
1022 /**
1023 * enum fault_flag - Fault flag definitions.
1024 * @FAULT_FLAG_WRITE: Fault was a write fault.
1025 * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE.
1026 * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked.
1027 * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying.
1028 * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region.
1029 * @FAULT_FLAG_TRIED: The fault has been tried once.
1030 * @FAULT_FLAG_USER: The fault originated in userspace.
1031 * @FAULT_FLAG_REMOTE: The fault is not for current task/mm.
1032 * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch.
1033 * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals.
1034 * @FAULT_FLAG_UNSHARE: The fault is an unsharing request to break COW in a
1035 * COW mapping, making sure that an exclusive anon page is
1036 * mapped after the fault.
1037 * @FAULT_FLAG_ORIG_PTE_VALID: whether the fault has vmf->orig_pte cached.
1038 * We should only access orig_pte if this flag set.
1039 *
1040 * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify
1041 * whether we would allow page faults to retry by specifying these two
1042 * fault flags correctly. Currently there can be three legal combinations:
1043 *
1044 * (a) ALLOW_RETRY and !TRIED: this means the page fault allows retry, and
1045 * this is the first try
1046 *
1047 * (b) ALLOW_RETRY and TRIED: this means the page fault allows retry, and
1048 * we've already tried at least once
1049 *
1050 * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry
1051 *
1052 * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never
1053 * be used. Note that page faults can be allowed to retry for multiple times,
1054 * in which case we'll have an initial fault with flags (a) then later on
1055 * continuous faults with flags (b). We should always try to detect pending
1056 * signals before a retry to make sure the continuous page faults can still be
1057 * interrupted if necessary.
1058 *
1059 * The combination FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE is illegal.
1060 * FAULT_FLAG_UNSHARE is ignored and treated like an ordinary read fault when
1061 * applied to mappings that are not COW mappings.
1062 */
1063 enum fault_flag {
1064 FAULT_FLAG_WRITE = 1 << 0,
1065 FAULT_FLAG_MKWRITE = 1 << 1,
1066 FAULT_FLAG_ALLOW_RETRY = 1 << 2,
1067 FAULT_FLAG_RETRY_NOWAIT = 1 << 3,
1068 FAULT_FLAG_KILLABLE = 1 << 4,
1069 FAULT_FLAG_TRIED = 1 << 5,
1070 FAULT_FLAG_USER = 1 << 6,
1071 FAULT_FLAG_REMOTE = 1 << 7,
1072 FAULT_FLAG_INSTRUCTION = 1 << 8,
1073 FAULT_FLAG_INTERRUPTIBLE = 1 << 9,
1074 FAULT_FLAG_UNSHARE = 1 << 10,
1075 FAULT_FLAG_ORIG_PTE_VALID = 1 << 11,
1076 };
1077
1078 typedef unsigned int __bitwise zap_flags_t;
1079
1080 /*
1081 * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each
1082 * other. Here is what they mean, and how to use them:
1083 *
1084 *
1085 * FIXME: For pages which are part of a filesystem, mappings are subject to the
1086 * lifetime enforced by the filesystem and we need guarantees that longterm
1087 * users like RDMA and V4L2 only establish mappings which coordinate usage with
1088 * the filesystem. Ideas for this coordination include revoking the longterm
1089 * pin, delaying writeback, bounce buffer page writeback, etc. As FS DAX was
1090 * added after the problem with filesystems was found FS DAX VMAs are
1091 * specifically failed. Filesystem pages are still subject to bugs and use of
1092 * FOLL_LONGTERM should be avoided on those pages.
1093 *
1094 * In the CMA case: long term pins in a CMA region would unnecessarily fragment
1095 * that region. And so, CMA attempts to migrate the page before pinning, when
1096 * FOLL_LONGTERM is specified.
1097 *
1098 * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount,
1099 * but an additional pin counting system) will be invoked. This is intended for
1100 * anything that gets a page reference and then touches page data (for example,
1101 * Direct IO). This lets the filesystem know that some non-file-system entity is
1102 * potentially changing the pages' data. In contrast to FOLL_GET (whose pages
1103 * are released via put_page()), FOLL_PIN pages must be released, ultimately, by
1104 * a call to unpin_user_page().
1105 *
1106 * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different
1107 * and separate refcounting mechanisms, however, and that means that each has
1108 * its own acquire and release mechanisms:
1109 *
1110 * FOLL_GET: get_user_pages*() to acquire, and put_page() to release.
1111 *
1112 * FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release.
1113 *
1114 * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call.
1115 * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based
1116 * calls applied to them, and that's perfectly OK. This is a constraint on the
1117 * callers, not on the pages.)
1118 *
1119 * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never
1120 * directly by the caller. That's in order to help avoid mismatches when
1121 * releasing pages: get_user_pages*() pages must be released via put_page(),
1122 * while pin_user_pages*() pages must be released via unpin_user_page().
1123 *
1124 * Please see Documentation/core-api/pin_user_pages.rst for more information.
1125 */
1126
1127 enum {
1128 /* check pte is writable */
1129 FOLL_WRITE = 1 << 0,
1130 /* do get_page on page */
1131 FOLL_GET = 1 << 1,
1132 /* give error on hole if it would be zero */
1133 FOLL_DUMP = 1 << 2,
1134 /* get_user_pages read/write w/o permission */
1135 FOLL_FORCE = 1 << 3,
1136 /*
1137 * if a disk transfer is needed, start the IO and return without waiting
1138 * upon it
1139 */
1140 FOLL_NOWAIT = 1 << 4,
1141 /* do not fault in pages */
1142 FOLL_NOFAULT = 1 << 5,
1143 /* check page is hwpoisoned */
1144 FOLL_HWPOISON = 1 << 6,
1145 /* don't do file mappings */
1146 FOLL_ANON = 1 << 7,
1147 /*
1148 * FOLL_LONGTERM indicates that the page will be held for an indefinite
1149 * time period _often_ under userspace control. This is in contrast to
1150 * iov_iter_get_pages(), whose usages are transient.
1151 */
1152 FOLL_LONGTERM = 1 << 8,
1153 /* split huge pmd before returning */
1154 FOLL_SPLIT_PMD = 1 << 9,
1155 /* allow returning PCI P2PDMA pages */
1156 FOLL_PCI_P2PDMA = 1 << 10,
1157 /* allow interrupts from generic signals */
1158 FOLL_INTERRUPTIBLE = 1 << 11,
1159
1160 /* See also internal only FOLL flags in mm/internal.h */
1161 };
1162
1163 #endif /* _LINUX_MM_TYPES_H */
1164