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 { /* ZONE_DEVICE pages */
145 /** @pgmap: Points to the hosting device page map. */
146 struct dev_pagemap *pgmap;
147 void *zone_device_data;
148 /*
149 * ZONE_DEVICE private pages are counted as being
150 * mapped so the next 3 words hold the mapping, index,
151 * and private fields from the source anonymous or
152 * page cache page while the page is migrated to device
153 * private memory.
154 * ZONE_DEVICE MEMORY_DEVICE_FS_DAX pages also
155 * use the mapping, index, and private fields when
156 * pmem backed DAX files are mapped.
157 */
158 };
159
160 /** @rcu_head: You can use this to free a page by RCU. */
161 struct rcu_head rcu_head;
162 };
163
164 union { /* This union is 4 bytes in size. */
165 /*
166 * If the page can be mapped to userspace, encodes the number
167 * of times this page is referenced by a page table.
168 */
169 atomic_t _mapcount;
170
171 /*
172 * If the page is neither PageSlab nor mappable to userspace,
173 * the value stored here may help determine what this page
174 * is used for. See page-flags.h for a list of page types
175 * which are currently stored here.
176 */
177 unsigned int page_type;
178 };
179
180 /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
181 atomic_t _refcount;
182
183 #ifdef CONFIG_MEMCG
184 unsigned long memcg_data;
185 #endif
186
187 /*
188 * On machines where all RAM is mapped into kernel address space,
189 * we can simply calculate the virtual address. On machines with
190 * highmem some memory is mapped into kernel virtual memory
191 * dynamically, so we need a place to store that address.
192 * Note that this field could be 16 bits on x86 ... ;)
193 *
194 * Architectures with slow multiplication can define
195 * WANT_PAGE_VIRTUAL in asm/page.h
196 */
197 #if defined(WANT_PAGE_VIRTUAL)
198 void *virtual; /* Kernel virtual address (NULL if
199 not kmapped, ie. highmem) */
200 #endif /* WANT_PAGE_VIRTUAL */
201
202 #ifdef CONFIG_KMSAN
203 /*
204 * KMSAN metadata for this page:
205 * - shadow page: every bit indicates whether the corresponding
206 * bit of the original page is initialized (0) or not (1);
207 * - origin page: every 4 bytes contain an id of the stack trace
208 * where the uninitialized value was created.
209 */
210 struct page *kmsan_shadow;
211 struct page *kmsan_origin;
212 #endif
213
214 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
215 int _last_cpupid;
216 #endif
217 } _struct_page_alignment;
218
219 /*
220 * struct encoded_page - a nonexistent type marking this pointer
221 *
222 * An 'encoded_page' pointer is a pointer to a regular 'struct page', but
223 * with the low bits of the pointer indicating extra context-dependent
224 * information. Not super-common, but happens in mmu_gather and mlock
225 * handling, and this acts as a type system check on that use.
226 *
227 * We only really have two guaranteed bits in general, although you could
228 * play with 'struct page' alignment (see CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
229 * for more.
230 *
231 * Use the supplied helper functions to endcode/decode the pointer and bits.
232 */
233 struct encoded_page;
234 #define ENCODE_PAGE_BITS 3ul
encode_page(struct page * page,unsigned long flags)235 static __always_inline struct encoded_page *encode_page(struct page *page, unsigned long flags)
236 {
237 BUILD_BUG_ON(flags > ENCODE_PAGE_BITS);
238 return (struct encoded_page *)(flags | (unsigned long)page);
239 }
240
encoded_page_flags(struct encoded_page * page)241 static inline unsigned long encoded_page_flags(struct encoded_page *page)
242 {
243 return ENCODE_PAGE_BITS & (unsigned long)page;
244 }
245
encoded_page_ptr(struct encoded_page * page)246 static inline struct page *encoded_page_ptr(struct encoded_page *page)
247 {
248 return (struct page *)(~ENCODE_PAGE_BITS & (unsigned long)page);
249 }
250
251 /*
252 * A swap entry has to fit into a "unsigned long", as the entry is hidden
253 * in the "index" field of the swapper address space.
254 */
255 typedef struct {
256 unsigned long val;
257 } swp_entry_t;
258
259 /**
260 * struct folio - Represents a contiguous set of bytes.
261 * @flags: Identical to the page flags.
262 * @lru: Least Recently Used list; tracks how recently this folio was used.
263 * @mlock_count: Number of times this folio has been pinned by mlock().
264 * @mapping: The file this page belongs to, or refers to the anon_vma for
265 * anonymous memory.
266 * @index: Offset within the file, in units of pages. For anonymous memory,
267 * this is the index from the beginning of the mmap.
268 * @private: Filesystem per-folio data (see folio_attach_private()).
269 * @swap: Used for swp_entry_t if folio_test_swapcache().
270 * @_mapcount: Do not access this member directly. Use folio_mapcount() to
271 * find out how many times this folio is mapped by userspace.
272 * @_refcount: Do not access this member directly. Use folio_ref_count()
273 * to find how many references there are to this folio.
274 * @memcg_data: Memory Control Group data.
275 * @_entire_mapcount: Do not use directly, call folio_entire_mapcount().
276 * @_nr_pages_mapped: Do not use directly, call folio_mapcount().
277 * @_pincount: Do not use directly, call folio_maybe_dma_pinned().
278 * @_folio_nr_pages: Do not use directly, call folio_nr_pages().
279 * @_hugetlb_subpool: Do not use directly, use accessor in hugetlb.h.
280 * @_hugetlb_cgroup: Do not use directly, use accessor in hugetlb_cgroup.h.
281 * @_hugetlb_cgroup_rsvd: Do not use directly, use accessor in hugetlb_cgroup.h.
282 * @_hugetlb_hwpoison: Do not use directly, call raw_hwp_list_head().
283 * @_deferred_list: Folios to be split under memory pressure.
284 *
285 * A folio is a physically, virtually and logically contiguous set
286 * of bytes. It is a power-of-two in size, and it is aligned to that
287 * same power-of-two. It is at least as large as %PAGE_SIZE. If it is
288 * in the page cache, it is at a file offset which is a multiple of that
289 * power-of-two. It may be mapped into userspace at an address which is
290 * at an arbitrary page offset, but its kernel virtual address is aligned
291 * to its size.
292 */
293 struct folio {
294 /* private: don't document the anon union */
295 union {
296 struct {
297 /* public: */
298 unsigned long flags;
299 union {
300 struct list_head lru;
301 /* private: avoid cluttering the output */
302 struct {
303 void *__filler;
304 /* public: */
305 unsigned int mlock_count;
306 /* private: */
307 };
308 /* public: */
309 };
310 struct address_space *mapping;
311 pgoff_t index;
312 union {
313 void *private;
314 swp_entry_t swap;
315 };
316 atomic_t _mapcount;
317 atomic_t _refcount;
318 #ifdef CONFIG_MEMCG
319 unsigned long memcg_data;
320 #endif
321 /* private: the union with struct page is transitional */
322 };
323 struct page page;
324 };
325 union {
326 struct {
327 unsigned long _flags_1;
328 unsigned long _head_1;
329 unsigned long _folio_avail;
330 /* public: */
331 atomic_t _entire_mapcount;
332 atomic_t _nr_pages_mapped;
333 atomic_t _pincount;
334 #ifdef CONFIG_64BIT
335 unsigned int _folio_nr_pages;
336 #endif
337 /* private: the union with struct page is transitional */
338 };
339 struct page __page_1;
340 };
341 union {
342 struct {
343 unsigned long _flags_2;
344 unsigned long _head_2;
345 /* public: */
346 void *_hugetlb_subpool;
347 void *_hugetlb_cgroup;
348 void *_hugetlb_cgroup_rsvd;
349 void *_hugetlb_hwpoison;
350 /* private: the union with struct page is transitional */
351 };
352 struct {
353 unsigned long _flags_2a;
354 unsigned long _head_2a;
355 /* public: */
356 struct list_head _deferred_list;
357 /* private: the union with struct page is transitional */
358 };
359 struct page __page_2;
360 };
361 };
362
363 #define FOLIO_MATCH(pg, fl) \
364 static_assert(offsetof(struct page, pg) == offsetof(struct folio, fl))
365 FOLIO_MATCH(flags, flags);
366 FOLIO_MATCH(lru, lru);
367 FOLIO_MATCH(mapping, mapping);
368 FOLIO_MATCH(compound_head, lru);
369 FOLIO_MATCH(index, index);
370 FOLIO_MATCH(private, private);
371 FOLIO_MATCH(_mapcount, _mapcount);
372 FOLIO_MATCH(_refcount, _refcount);
373 #ifdef CONFIG_MEMCG
374 FOLIO_MATCH(memcg_data, memcg_data);
375 #endif
376 #undef FOLIO_MATCH
377 #define FOLIO_MATCH(pg, fl) \
378 static_assert(offsetof(struct folio, fl) == \
379 offsetof(struct page, pg) + sizeof(struct page))
380 FOLIO_MATCH(flags, _flags_1);
381 FOLIO_MATCH(compound_head, _head_1);
382 #undef FOLIO_MATCH
383 #define FOLIO_MATCH(pg, fl) \
384 static_assert(offsetof(struct folio, fl) == \
385 offsetof(struct page, pg) + 2 * sizeof(struct page))
386 FOLIO_MATCH(flags, _flags_2);
387 FOLIO_MATCH(compound_head, _head_2);
388 FOLIO_MATCH(flags, _flags_2a);
389 FOLIO_MATCH(compound_head, _head_2a);
390 #undef FOLIO_MATCH
391
392 /**
393 * struct ptdesc - Memory descriptor for page tables.
394 * @__page_flags: Same as page flags. Unused for page tables.
395 * @pt_rcu_head: For freeing page table pages.
396 * @pt_list: List of used page tables. Used for s390 and x86.
397 * @_pt_pad_1: Padding that aliases with page's compound head.
398 * @pmd_huge_pte: Protected by ptdesc->ptl, used for THPs.
399 * @__page_mapping: Aliases with page->mapping. Unused for page tables.
400 * @pt_mm: Used for x86 pgds.
401 * @pt_frag_refcount: For fragmented page table tracking. Powerpc and s390 only.
402 * @_pt_pad_2: Padding to ensure proper alignment.
403 * @ptl: Lock for the page table.
404 * @__page_type: Same as page->page_type. Unused for page tables.
405 * @_refcount: Same as page refcount. Used for s390 page tables.
406 * @pt_memcg_data: Memcg data. Tracked for page tables here.
407 *
408 * This struct overlays struct page for now. Do not modify without a good
409 * understanding of the issues.
410 */
411 struct ptdesc {
412 unsigned long __page_flags;
413
414 union {
415 struct rcu_head pt_rcu_head;
416 struct list_head pt_list;
417 struct {
418 unsigned long _pt_pad_1;
419 pgtable_t pmd_huge_pte;
420 };
421 };
422 unsigned long __page_mapping;
423
424 union {
425 struct mm_struct *pt_mm;
426 atomic_t pt_frag_refcount;
427 };
428
429 union {
430 unsigned long _pt_pad_2;
431 #if ALLOC_SPLIT_PTLOCKS
432 spinlock_t *ptl;
433 #else
434 spinlock_t ptl;
435 #endif
436 };
437 unsigned int __page_type;
438 atomic_t _refcount;
439 #ifdef CONFIG_MEMCG
440 unsigned long pt_memcg_data;
441 #endif
442 };
443
444 #define TABLE_MATCH(pg, pt) \
445 static_assert(offsetof(struct page, pg) == offsetof(struct ptdesc, pt))
446 TABLE_MATCH(flags, __page_flags);
447 TABLE_MATCH(compound_head, pt_list);
448 TABLE_MATCH(compound_head, _pt_pad_1);
449 TABLE_MATCH(mapping, __page_mapping);
450 TABLE_MATCH(rcu_head, pt_rcu_head);
451 TABLE_MATCH(page_type, __page_type);
452 TABLE_MATCH(_refcount, _refcount);
453 #ifdef CONFIG_MEMCG
454 TABLE_MATCH(memcg_data, pt_memcg_data);
455 #endif
456 #undef TABLE_MATCH
457 static_assert(sizeof(struct ptdesc) <= sizeof(struct page));
458
459 #define ptdesc_page(pt) (_Generic((pt), \
460 const struct ptdesc *: (const struct page *)(pt), \
461 struct ptdesc *: (struct page *)(pt)))
462
463 #define ptdesc_folio(pt) (_Generic((pt), \
464 const struct ptdesc *: (const struct folio *)(pt), \
465 struct ptdesc *: (struct folio *)(pt)))
466
467 #define page_ptdesc(p) (_Generic((p), \
468 const struct page *: (const struct ptdesc *)(p), \
469 struct page *: (struct ptdesc *)(p)))
470
471 /*
472 * Used for sizing the vmemmap region on some architectures
473 */
474 #define STRUCT_PAGE_MAX_SHIFT (order_base_2(sizeof(struct page)))
475
476 #define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
477 #define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
478
479 /*
480 * page_private can be used on tail pages. However, PagePrivate is only
481 * checked by the VM on the head page. So page_private on the tail pages
482 * should be used for data that's ancillary to the head page (eg attaching
483 * buffer heads to tail pages after attaching buffer heads to the head page)
484 */
485 #define page_private(page) ((page)->private)
486
set_page_private(struct page * page,unsigned long private)487 static inline void set_page_private(struct page *page, unsigned long private)
488 {
489 page->private = private;
490 }
491
folio_get_private(struct folio * folio)492 static inline void *folio_get_private(struct folio *folio)
493 {
494 return folio->private;
495 }
496
497 struct page_frag_cache {
498 void * va;
499 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
500 __u16 offset;
501 __u16 size;
502 #else
503 __u32 offset;
504 #endif
505 /* we maintain a pagecount bias, so that we dont dirty cache line
506 * containing page->_refcount every time we allocate a fragment.
507 */
508 unsigned int pagecnt_bias;
509 bool pfmemalloc;
510 };
511
512 typedef unsigned long vm_flags_t;
513
514 /*
515 * A region containing a mapping of a non-memory backed file under NOMMU
516 * conditions. These are held in a global tree and are pinned by the VMAs that
517 * map parts of them.
518 */
519 struct vm_region {
520 struct rb_node vm_rb; /* link in global region tree */
521 vm_flags_t vm_flags; /* VMA vm_flags */
522 unsigned long vm_start; /* start address of region */
523 unsigned long vm_end; /* region initialised to here */
524 unsigned long vm_top; /* region allocated to here */
525 unsigned long vm_pgoff; /* the offset in vm_file corresponding to vm_start */
526 struct file *vm_file; /* the backing file or NULL */
527
528 int vm_usage; /* region usage count (access under nommu_region_sem) */
529 bool vm_icache_flushed : 1; /* true if the icache has been flushed for
530 * this region */
531 };
532
533 #ifdef CONFIG_USERFAULTFD
534 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) { NULL, })
535 struct vm_userfaultfd_ctx {
536 struct userfaultfd_ctx *ctx;
537 };
538 #else /* CONFIG_USERFAULTFD */
539 #define NULL_VM_UFFD_CTX ((struct vm_userfaultfd_ctx) {})
540 struct vm_userfaultfd_ctx {};
541 #endif /* CONFIG_USERFAULTFD */
542
543 struct anon_vma_name {
544 struct kref kref;
545 /* The name needs to be at the end because it is dynamically sized. */
546 char name[];
547 };
548
549 struct vma_lock {
550 struct rw_semaphore lock;
551 };
552
553 struct vma_numab_state {
554 unsigned long next_scan;
555 unsigned long next_pid_reset;
556 unsigned long access_pids[2];
557 };
558
559 /*
560 * This struct describes a virtual memory area. There is one of these
561 * per VM-area/task. A VM area is any part of the process virtual memory
562 * space that has a special rule for the page-fault handlers (ie a shared
563 * library, the executable area etc).
564 */
565 struct vm_area_struct {
566 /* The first cache line has the info for VMA tree walking. */
567
568 union {
569 struct {
570 /* VMA covers [vm_start; vm_end) addresses within mm */
571 unsigned long vm_start;
572 unsigned long vm_end;
573 };
574 #ifdef CONFIG_PER_VMA_LOCK
575 struct rcu_head vm_rcu; /* Used for deferred freeing. */
576 #endif
577 };
578
579 struct mm_struct *vm_mm; /* The address space we belong to. */
580 pgprot_t vm_page_prot; /* Access permissions of this VMA. */
581
582 /*
583 * Flags, see mm.h.
584 * To modify use vm_flags_{init|reset|set|clear|mod} functions.
585 */
586 union {
587 const vm_flags_t vm_flags;
588 vm_flags_t __private __vm_flags;
589 };
590
591 #ifdef CONFIG_PER_VMA_LOCK
592 /*
593 * Can only be written (using WRITE_ONCE()) while holding both:
594 * - mmap_lock (in write mode)
595 * - vm_lock->lock (in write mode)
596 * Can be read reliably while holding one of:
597 * - mmap_lock (in read or write mode)
598 * - vm_lock->lock (in read or write mode)
599 * Can be read unreliably (using READ_ONCE()) for pessimistic bailout
600 * while holding nothing (except RCU to keep the VMA struct allocated).
601 *
602 * This sequence counter is explicitly allowed to overflow; sequence
603 * counter reuse can only lead to occasional unnecessary use of the
604 * slowpath.
605 */
606 int vm_lock_seq;
607 struct vma_lock *vm_lock;
608
609 /* Flag to indicate areas detached from the mm->mm_mt tree */
610 bool detached;
611 #endif
612
613 /*
614 * For areas with an address space and backing store,
615 * linkage into the address_space->i_mmap interval tree.
616 *
617 */
618 struct {
619 struct rb_node rb;
620 unsigned long rb_subtree_last;
621 } shared;
622
623 /*
624 * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma
625 * list, after a COW of one of the file pages. A MAP_SHARED vma
626 * can only be in the i_mmap tree. An anonymous MAP_PRIVATE, stack
627 * or brk vma (with NULL file) can only be in an anon_vma list.
628 */
629 struct list_head anon_vma_chain; /* Serialized by mmap_lock &
630 * page_table_lock */
631 struct anon_vma *anon_vma; /* Serialized by page_table_lock */
632
633 /* Function pointers to deal with this struct. */
634 const struct vm_operations_struct *vm_ops;
635
636 /* Information about our backing store: */
637 unsigned long vm_pgoff; /* Offset (within vm_file) in PAGE_SIZE
638 units */
639 struct file * vm_file; /* File we map to (can be NULL). */
640 void * vm_private_data; /* was vm_pte (shared mem) */
641
642 #ifdef CONFIG_ANON_VMA_NAME
643 /*
644 * For private and shared anonymous mappings, a pointer to a null
645 * terminated string containing the name given to the vma, or NULL if
646 * unnamed. Serialized by mmap_lock. Use anon_vma_name to access.
647 */
648 struct anon_vma_name *anon_name;
649 #endif
650 #ifdef CONFIG_SWAP
651 atomic_long_t swap_readahead_info;
652 #endif
653 #ifndef CONFIG_MMU
654 struct vm_region *vm_region; /* NOMMU mapping region */
655 #endif
656 #ifdef CONFIG_NUMA
657 struct mempolicy *vm_policy; /* NUMA policy for the VMA */
658 #endif
659 #ifdef CONFIG_NUMA_BALANCING
660 struct vma_numab_state *numab_state; /* NUMA Balancing state */
661 #endif
662 struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
663 } __randomize_layout;
664
665 #ifdef CONFIG_SCHED_MM_CID
666 struct mm_cid {
667 u64 time;
668 int cid;
669 };
670 #endif
671
672 struct kioctx_table;
673 struct mm_struct {
674 struct {
675 /*
676 * Fields which are often written to are placed in a separate
677 * cache line.
678 */
679 struct {
680 /**
681 * @mm_count: The number of references to &struct
682 * mm_struct (@mm_users count as 1).
683 *
684 * Use mmgrab()/mmdrop() to modify. When this drops to
685 * 0, the &struct mm_struct is freed.
686 */
687 atomic_t mm_count;
688 } ____cacheline_aligned_in_smp;
689
690 struct maple_tree mm_mt;
691 #ifdef CONFIG_MMU
692 unsigned long (*get_unmapped_area) (struct file *filp,
693 unsigned long addr, unsigned long len,
694 unsigned long pgoff, unsigned long flags);
695 #endif
696 unsigned long mmap_base; /* base of mmap area */
697 unsigned long mmap_legacy_base; /* base of mmap area in bottom-up allocations */
698 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
699 /* Base addresses for compatible mmap() */
700 unsigned long mmap_compat_base;
701 unsigned long mmap_compat_legacy_base;
702 #endif
703 unsigned long task_size; /* size of task vm space */
704 pgd_t * pgd;
705
706 #ifdef CONFIG_MEMBARRIER
707 /**
708 * @membarrier_state: Flags controlling membarrier behavior.
709 *
710 * This field is close to @pgd to hopefully fit in the same
711 * cache-line, which needs to be touched by switch_mm().
712 */
713 atomic_t membarrier_state;
714 #endif
715
716 /**
717 * @mm_users: The number of users including userspace.
718 *
719 * Use mmget()/mmget_not_zero()/mmput() to modify. When this
720 * drops to 0 (i.e. when the task exits and there are no other
721 * temporary reference holders), we also release a reference on
722 * @mm_count (which may then free the &struct mm_struct if
723 * @mm_count also drops to 0).
724 */
725 atomic_t mm_users;
726
727 #ifdef CONFIG_SCHED_MM_CID
728 /**
729 * @pcpu_cid: Per-cpu current cid.
730 *
731 * Keep track of the currently allocated mm_cid for each cpu.
732 * The per-cpu mm_cid values are serialized by their respective
733 * runqueue locks.
734 */
735 struct mm_cid __percpu *pcpu_cid;
736 /*
737 * @mm_cid_next_scan: Next mm_cid scan (in jiffies).
738 *
739 * When the next mm_cid scan is due (in jiffies).
740 */
741 unsigned long mm_cid_next_scan;
742 #endif
743 #ifdef CONFIG_MMU
744 atomic_long_t pgtables_bytes; /* size of all page tables */
745 #endif
746 int map_count; /* number of VMAs */
747
748 spinlock_t page_table_lock; /* Protects page tables and some
749 * counters
750 */
751 /*
752 * With some kernel config, the current mmap_lock's offset
753 * inside 'mm_struct' is at 0x120, which is very optimal, as
754 * its two hot fields 'count' and 'owner' sit in 2 different
755 * cachelines, and when mmap_lock is highly contended, both
756 * of the 2 fields will be accessed frequently, current layout
757 * will help to reduce cache bouncing.
758 *
759 * So please be careful with adding new fields before
760 * mmap_lock, which can easily push the 2 fields into one
761 * cacheline.
762 */
763 struct rw_semaphore mmap_lock;
764
765 struct list_head mmlist; /* List of maybe swapped mm's. These
766 * are globally strung together off
767 * init_mm.mmlist, and are protected
768 * by mmlist_lock
769 */
770 #ifdef CONFIG_PER_VMA_LOCK
771 /*
772 * This field has lock-like semantics, meaning it is sometimes
773 * accessed with ACQUIRE/RELEASE semantics.
774 * Roughly speaking, incrementing the sequence number is
775 * equivalent to releasing locks on VMAs; reading the sequence
776 * number can be part of taking a read lock on a VMA.
777 *
778 * Can be modified under write mmap_lock using RELEASE
779 * semantics.
780 * Can be read with no other protection when holding write
781 * mmap_lock.
782 * Can be read with ACQUIRE semantics if not holding write
783 * mmap_lock.
784 */
785 int mm_lock_seq;
786 #endif
787
788
789 unsigned long hiwater_rss; /* High-watermark of RSS usage */
790 unsigned long hiwater_vm; /* High-water virtual memory usage */
791
792 unsigned long total_vm; /* Total pages mapped */
793 unsigned long locked_vm; /* Pages that have PG_mlocked set */
794 atomic64_t pinned_vm; /* Refcount permanently increased */
795 unsigned long data_vm; /* VM_WRITE & ~VM_SHARED & ~VM_STACK */
796 unsigned long exec_vm; /* VM_EXEC & ~VM_WRITE & ~VM_STACK */
797 unsigned long stack_vm; /* VM_STACK */
798 unsigned long def_flags;
799
800 /**
801 * @write_protect_seq: Locked when any thread is write
802 * protecting pages mapped by this mm to enforce a later COW,
803 * for instance during page table copying for fork().
804 */
805 seqcount_t write_protect_seq;
806
807 spinlock_t arg_lock; /* protect the below fields */
808
809 unsigned long start_code, end_code, start_data, end_data;
810 unsigned long start_brk, brk, start_stack;
811 unsigned long arg_start, arg_end, env_start, env_end;
812
813 unsigned long saved_auxv[AT_VECTOR_SIZE]; /* for /proc/PID/auxv */
814
815 struct percpu_counter rss_stat[NR_MM_COUNTERS];
816
817 struct linux_binfmt *binfmt;
818
819 /* Architecture-specific MM context */
820 mm_context_t context;
821
822 unsigned long flags; /* Must use atomic bitops to access */
823
824 #ifdef CONFIG_AIO
825 spinlock_t ioctx_lock;
826 struct kioctx_table __rcu *ioctx_table;
827 #endif
828 #ifdef CONFIG_MEMCG
829 /*
830 * "owner" points to a task that is regarded as the canonical
831 * user/owner of this mm. All of the following must be true in
832 * order for it to be changed:
833 *
834 * current == mm->owner
835 * current->mm != mm
836 * new_owner->mm == mm
837 * new_owner->alloc_lock is held
838 */
839 struct task_struct __rcu *owner;
840 #endif
841 struct user_namespace *user_ns;
842
843 /* store ref to file /proc/<pid>/exe symlink points to */
844 struct file __rcu *exe_file;
845 #ifdef CONFIG_MMU_NOTIFIER
846 struct mmu_notifier_subscriptions *notifier_subscriptions;
847 #endif
848 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
849 pgtable_t pmd_huge_pte; /* protected by page_table_lock */
850 #endif
851 #ifdef CONFIG_NUMA_BALANCING
852 /*
853 * numa_next_scan is the next time that PTEs will be remapped
854 * PROT_NONE to trigger NUMA hinting faults; such faults gather
855 * statistics and migrate pages to new nodes if necessary.
856 */
857 unsigned long numa_next_scan;
858
859 /* Restart point for scanning and remapping PTEs. */
860 unsigned long numa_scan_offset;
861
862 /* numa_scan_seq prevents two threads remapping PTEs. */
863 int numa_scan_seq;
864 #endif
865 /*
866 * An operation with batched TLB flushing is going on. Anything
867 * that can move process memory needs to flush the TLB when
868 * moving a PROT_NONE mapped page.
869 */
870 atomic_t tlb_flush_pending;
871 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
872 /* See flush_tlb_batched_pending() */
873 atomic_t tlb_flush_batched;
874 #endif
875 struct uprobes_state uprobes_state;
876 #ifdef CONFIG_PREEMPT_RT
877 struct rcu_head delayed_drop;
878 #endif
879 #ifdef CONFIG_HUGETLB_PAGE
880 atomic_long_t hugetlb_usage;
881 #endif
882 struct work_struct async_put_work;
883
884 #ifdef CONFIG_IOMMU_SVA
885 u32 pasid;
886 #endif
887 #ifdef CONFIG_KSM
888 /*
889 * Represent how many pages of this process are involved in KSM
890 * merging (not including ksm_zero_pages).
891 */
892 unsigned long ksm_merging_pages;
893 /*
894 * Represent how many pages are checked for ksm merging
895 * including merged and not merged.
896 */
897 unsigned long ksm_rmap_items;
898 /*
899 * Represent how many empty pages are merged with kernel zero
900 * pages when enabling KSM use_zero_pages.
901 */
902 unsigned long ksm_zero_pages;
903 #endif /* CONFIG_KSM */
904 #ifdef CONFIG_LRU_GEN
905 struct {
906 /* this mm_struct is on lru_gen_mm_list */
907 struct list_head list;
908 /*
909 * Set when switching to this mm_struct, as a hint of
910 * whether it has been used since the last time per-node
911 * page table walkers cleared the corresponding bits.
912 */
913 unsigned long bitmap;
914 #ifdef CONFIG_MEMCG
915 /* points to the memcg of "owner" above */
916 struct mem_cgroup *memcg;
917 #endif
918 } lru_gen;
919 #endif /* CONFIG_LRU_GEN */
920 } __randomize_layout;
921
922 /*
923 * The mm_cpumask needs to be at the end of mm_struct, because it
924 * is dynamically sized based on nr_cpu_ids.
925 */
926 unsigned long cpu_bitmap[];
927 };
928
929 #define MM_MT_FLAGS (MT_FLAGS_ALLOC_RANGE | MT_FLAGS_LOCK_EXTERN | \
930 MT_FLAGS_USE_RCU)
931 extern struct mm_struct init_mm;
932
933 /* Pointer magic because the dynamic array size confuses some compilers. */
mm_init_cpumask(struct mm_struct * mm)934 static inline void mm_init_cpumask(struct mm_struct *mm)
935 {
936 unsigned long cpu_bitmap = (unsigned long)mm;
937
938 cpu_bitmap += offsetof(struct mm_struct, cpu_bitmap);
939 cpumask_clear((struct cpumask *)cpu_bitmap);
940 }
941
942 /* Future-safe accessor for struct mm_struct's cpu_vm_mask. */
mm_cpumask(struct mm_struct * mm)943 static inline cpumask_t *mm_cpumask(struct mm_struct *mm)
944 {
945 return (struct cpumask *)&mm->cpu_bitmap;
946 }
947
948 #ifdef CONFIG_LRU_GEN
949
950 struct lru_gen_mm_list {
951 /* mm_struct list for page table walkers */
952 struct list_head fifo;
953 /* protects the list above */
954 spinlock_t lock;
955 };
956
957 void lru_gen_add_mm(struct mm_struct *mm);
958 void lru_gen_del_mm(struct mm_struct *mm);
959 #ifdef CONFIG_MEMCG
960 void lru_gen_migrate_mm(struct mm_struct *mm);
961 #endif
962
lru_gen_init_mm(struct mm_struct * mm)963 static inline void lru_gen_init_mm(struct mm_struct *mm)
964 {
965 INIT_LIST_HEAD(&mm->lru_gen.list);
966 mm->lru_gen.bitmap = 0;
967 #ifdef CONFIG_MEMCG
968 mm->lru_gen.memcg = NULL;
969 #endif
970 }
971
lru_gen_use_mm(struct mm_struct * mm)972 static inline void lru_gen_use_mm(struct mm_struct *mm)
973 {
974 /*
975 * When the bitmap is set, page reclaim knows this mm_struct has been
976 * used since the last time it cleared the bitmap. So it might be worth
977 * walking the page tables of this mm_struct to clear the accessed bit.
978 */
979 WRITE_ONCE(mm->lru_gen.bitmap, -1);
980 }
981
982 #else /* !CONFIG_LRU_GEN */
983
lru_gen_add_mm(struct mm_struct * mm)984 static inline void lru_gen_add_mm(struct mm_struct *mm)
985 {
986 }
987
lru_gen_del_mm(struct mm_struct * mm)988 static inline void lru_gen_del_mm(struct mm_struct *mm)
989 {
990 }
991
992 #ifdef CONFIG_MEMCG
lru_gen_migrate_mm(struct mm_struct * mm)993 static inline void lru_gen_migrate_mm(struct mm_struct *mm)
994 {
995 }
996 #endif
997
lru_gen_init_mm(struct mm_struct * mm)998 static inline void lru_gen_init_mm(struct mm_struct *mm)
999 {
1000 }
1001
lru_gen_use_mm(struct mm_struct * mm)1002 static inline void lru_gen_use_mm(struct mm_struct *mm)
1003 {
1004 }
1005
1006 #endif /* CONFIG_LRU_GEN */
1007
1008 struct vma_iterator {
1009 struct ma_state mas;
1010 };
1011
1012 #define VMA_ITERATOR(name, __mm, __addr) \
1013 struct vma_iterator name = { \
1014 .mas = { \
1015 .tree = &(__mm)->mm_mt, \
1016 .index = __addr, \
1017 .node = MAS_START, \
1018 }, \
1019 }
1020
vma_iter_init(struct vma_iterator * vmi,struct mm_struct * mm,unsigned long addr)1021 static inline void vma_iter_init(struct vma_iterator *vmi,
1022 struct mm_struct *mm, unsigned long addr)
1023 {
1024 mas_init(&vmi->mas, &mm->mm_mt, addr);
1025 }
1026
1027 #ifdef CONFIG_SCHED_MM_CID
1028
1029 enum mm_cid_state {
1030 MM_CID_UNSET = -1U, /* Unset state has lazy_put flag set. */
1031 MM_CID_LAZY_PUT = (1U << 31),
1032 };
1033
mm_cid_is_unset(int cid)1034 static inline bool mm_cid_is_unset(int cid)
1035 {
1036 return cid == MM_CID_UNSET;
1037 }
1038
mm_cid_is_lazy_put(int cid)1039 static inline bool mm_cid_is_lazy_put(int cid)
1040 {
1041 return !mm_cid_is_unset(cid) && (cid & MM_CID_LAZY_PUT);
1042 }
1043
mm_cid_is_valid(int cid)1044 static inline bool mm_cid_is_valid(int cid)
1045 {
1046 return !(cid & MM_CID_LAZY_PUT);
1047 }
1048
mm_cid_set_lazy_put(int cid)1049 static inline int mm_cid_set_lazy_put(int cid)
1050 {
1051 return cid | MM_CID_LAZY_PUT;
1052 }
1053
mm_cid_clear_lazy_put(int cid)1054 static inline int mm_cid_clear_lazy_put(int cid)
1055 {
1056 return cid & ~MM_CID_LAZY_PUT;
1057 }
1058
1059 /* Accessor for struct mm_struct's cidmask. */
mm_cidmask(struct mm_struct * mm)1060 static inline cpumask_t *mm_cidmask(struct mm_struct *mm)
1061 {
1062 unsigned long cid_bitmap = (unsigned long)mm;
1063
1064 cid_bitmap += offsetof(struct mm_struct, cpu_bitmap);
1065 /* Skip cpu_bitmap */
1066 cid_bitmap += cpumask_size();
1067 return (struct cpumask *)cid_bitmap;
1068 }
1069
mm_init_cid(struct mm_struct * mm)1070 static inline void mm_init_cid(struct mm_struct *mm)
1071 {
1072 int i;
1073
1074 for_each_possible_cpu(i) {
1075 struct mm_cid *pcpu_cid = per_cpu_ptr(mm->pcpu_cid, i);
1076
1077 pcpu_cid->cid = MM_CID_UNSET;
1078 pcpu_cid->time = 0;
1079 }
1080 cpumask_clear(mm_cidmask(mm));
1081 }
1082
mm_alloc_cid(struct mm_struct * mm)1083 static inline int mm_alloc_cid(struct mm_struct *mm)
1084 {
1085 mm->pcpu_cid = alloc_percpu(struct mm_cid);
1086 if (!mm->pcpu_cid)
1087 return -ENOMEM;
1088 mm_init_cid(mm);
1089 return 0;
1090 }
1091
mm_destroy_cid(struct mm_struct * mm)1092 static inline void mm_destroy_cid(struct mm_struct *mm)
1093 {
1094 free_percpu(mm->pcpu_cid);
1095 mm->pcpu_cid = NULL;
1096 }
1097
mm_cid_size(void)1098 static inline unsigned int mm_cid_size(void)
1099 {
1100 return cpumask_size();
1101 }
1102 #else /* CONFIG_SCHED_MM_CID */
mm_init_cid(struct mm_struct * mm)1103 static inline void mm_init_cid(struct mm_struct *mm) { }
mm_alloc_cid(struct mm_struct * mm)1104 static inline int mm_alloc_cid(struct mm_struct *mm) { return 0; }
mm_destroy_cid(struct mm_struct * mm)1105 static inline void mm_destroy_cid(struct mm_struct *mm) { }
mm_cid_size(void)1106 static inline unsigned int mm_cid_size(void)
1107 {
1108 return 0;
1109 }
1110 #endif /* CONFIG_SCHED_MM_CID */
1111
1112 struct mmu_gather;
1113 extern void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm);
1114 extern void tlb_gather_mmu_fullmm(struct mmu_gather *tlb, struct mm_struct *mm);
1115 extern void tlb_finish_mmu(struct mmu_gather *tlb);
1116
1117 struct vm_fault;
1118
1119 /**
1120 * typedef vm_fault_t - Return type for page fault handlers.
1121 *
1122 * Page fault handlers return a bitmask of %VM_FAULT values.
1123 */
1124 typedef __bitwise unsigned int vm_fault_t;
1125
1126 /**
1127 * enum vm_fault_reason - Page fault handlers return a bitmask of
1128 * these values to tell the core VM what happened when handling the
1129 * fault. Used to decide whether a process gets delivered SIGBUS or
1130 * just gets major/minor fault counters bumped up.
1131 *
1132 * @VM_FAULT_OOM: Out Of Memory
1133 * @VM_FAULT_SIGBUS: Bad access
1134 * @VM_FAULT_MAJOR: Page read from storage
1135 * @VM_FAULT_HWPOISON: Hit poisoned small page
1136 * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
1137 * in upper bits
1138 * @VM_FAULT_SIGSEGV: segmentation fault
1139 * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
1140 * @VM_FAULT_LOCKED: ->fault locked the returned page
1141 * @VM_FAULT_RETRY: ->fault blocked, must retry
1142 * @VM_FAULT_FALLBACK: huge page fault failed, fall back to small
1143 * @VM_FAULT_DONE_COW: ->fault has fully handled COW
1144 * @VM_FAULT_NEEDDSYNC: ->fault did not modify page tables and needs
1145 * fsync() to complete (for synchronous page faults
1146 * in DAX)
1147 * @VM_FAULT_COMPLETED: ->fault completed, meanwhile mmap lock released
1148 * @VM_FAULT_HINDEX_MASK: mask HINDEX value
1149 *
1150 */
1151 enum vm_fault_reason {
1152 VM_FAULT_OOM = (__force vm_fault_t)0x000001,
1153 VM_FAULT_SIGBUS = (__force vm_fault_t)0x000002,
1154 VM_FAULT_MAJOR = (__force vm_fault_t)0x000004,
1155 VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
1156 VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
1157 VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
1158 VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
1159 VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
1160 VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
1161 VM_FAULT_FALLBACK = (__force vm_fault_t)0x000800,
1162 VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000,
1163 VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000,
1164 VM_FAULT_COMPLETED = (__force vm_fault_t)0x004000,
1165 VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000,
1166 };
1167
1168 /* Encode hstate index for a hwpoisoned large page */
1169 #define VM_FAULT_SET_HINDEX(x) ((__force vm_fault_t)((x) << 16))
1170 #define VM_FAULT_GET_HINDEX(x) (((__force unsigned int)(x) >> 16) & 0xf)
1171
1172 #define VM_FAULT_ERROR (VM_FAULT_OOM | VM_FAULT_SIGBUS | \
1173 VM_FAULT_SIGSEGV | VM_FAULT_HWPOISON | \
1174 VM_FAULT_HWPOISON_LARGE | VM_FAULT_FALLBACK)
1175
1176 #define VM_FAULT_RESULT_TRACE \
1177 { VM_FAULT_OOM, "OOM" }, \
1178 { VM_FAULT_SIGBUS, "SIGBUS" }, \
1179 { VM_FAULT_MAJOR, "MAJOR" }, \
1180 { VM_FAULT_HWPOISON, "HWPOISON" }, \
1181 { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
1182 { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
1183 { VM_FAULT_NOPAGE, "NOPAGE" }, \
1184 { VM_FAULT_LOCKED, "LOCKED" }, \
1185 { VM_FAULT_RETRY, "RETRY" }, \
1186 { VM_FAULT_FALLBACK, "FALLBACK" }, \
1187 { VM_FAULT_DONE_COW, "DONE_COW" }, \
1188 { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }, \
1189 { VM_FAULT_COMPLETED, "COMPLETED" }
1190
1191 struct vm_special_mapping {
1192 const char *name; /* The name, e.g. "[vdso]". */
1193
1194 /*
1195 * If .fault is not provided, this points to a
1196 * NULL-terminated array of pages that back the special mapping.
1197 *
1198 * This must not be NULL unless .fault is provided.
1199 */
1200 struct page **pages;
1201
1202 /*
1203 * If non-NULL, then this is called to resolve page faults
1204 * on the special mapping. If used, .pages is not checked.
1205 */
1206 vm_fault_t (*fault)(const struct vm_special_mapping *sm,
1207 struct vm_area_struct *vma,
1208 struct vm_fault *vmf);
1209
1210 int (*mremap)(const struct vm_special_mapping *sm,
1211 struct vm_area_struct *new_vma);
1212 };
1213
1214 enum tlb_flush_reason {
1215 TLB_FLUSH_ON_TASK_SWITCH,
1216 TLB_REMOTE_SHOOTDOWN,
1217 TLB_LOCAL_SHOOTDOWN,
1218 TLB_LOCAL_MM_SHOOTDOWN,
1219 TLB_REMOTE_SEND_IPI,
1220 NR_TLB_FLUSH_REASONS,
1221 };
1222
1223 /**
1224 * enum fault_flag - Fault flag definitions.
1225 * @FAULT_FLAG_WRITE: Fault was a write fault.
1226 * @FAULT_FLAG_MKWRITE: Fault was mkwrite of existing PTE.
1227 * @FAULT_FLAG_ALLOW_RETRY: Allow to retry the fault if blocked.
1228 * @FAULT_FLAG_RETRY_NOWAIT: Don't drop mmap_lock and wait when retrying.
1229 * @FAULT_FLAG_KILLABLE: The fault task is in SIGKILL killable region.
1230 * @FAULT_FLAG_TRIED: The fault has been tried once.
1231 * @FAULT_FLAG_USER: The fault originated in userspace.
1232 * @FAULT_FLAG_REMOTE: The fault is not for current task/mm.
1233 * @FAULT_FLAG_INSTRUCTION: The fault was during an instruction fetch.
1234 * @FAULT_FLAG_INTERRUPTIBLE: The fault can be interrupted by non-fatal signals.
1235 * @FAULT_FLAG_UNSHARE: The fault is an unsharing request to break COW in a
1236 * COW mapping, making sure that an exclusive anon page is
1237 * mapped after the fault.
1238 * @FAULT_FLAG_ORIG_PTE_VALID: whether the fault has vmf->orig_pte cached.
1239 * We should only access orig_pte if this flag set.
1240 * @FAULT_FLAG_VMA_LOCK: The fault is handled under VMA lock.
1241 *
1242 * About @FAULT_FLAG_ALLOW_RETRY and @FAULT_FLAG_TRIED: we can specify
1243 * whether we would allow page faults to retry by specifying these two
1244 * fault flags correctly. Currently there can be three legal combinations:
1245 *
1246 * (a) ALLOW_RETRY and !TRIED: this means the page fault allows retry, and
1247 * this is the first try
1248 *
1249 * (b) ALLOW_RETRY and TRIED: this means the page fault allows retry, and
1250 * we've already tried at least once
1251 *
1252 * (c) !ALLOW_RETRY and !TRIED: this means the page fault does not allow retry
1253 *
1254 * The unlisted combination (!ALLOW_RETRY && TRIED) is illegal and should never
1255 * be used. Note that page faults can be allowed to retry for multiple times,
1256 * in which case we'll have an initial fault with flags (a) then later on
1257 * continuous faults with flags (b). We should always try to detect pending
1258 * signals before a retry to make sure the continuous page faults can still be
1259 * interrupted if necessary.
1260 *
1261 * The combination FAULT_FLAG_WRITE|FAULT_FLAG_UNSHARE is illegal.
1262 * FAULT_FLAG_UNSHARE is ignored and treated like an ordinary read fault when
1263 * applied to mappings that are not COW mappings.
1264 */
1265 enum fault_flag {
1266 FAULT_FLAG_WRITE = 1 << 0,
1267 FAULT_FLAG_MKWRITE = 1 << 1,
1268 FAULT_FLAG_ALLOW_RETRY = 1 << 2,
1269 FAULT_FLAG_RETRY_NOWAIT = 1 << 3,
1270 FAULT_FLAG_KILLABLE = 1 << 4,
1271 FAULT_FLAG_TRIED = 1 << 5,
1272 FAULT_FLAG_USER = 1 << 6,
1273 FAULT_FLAG_REMOTE = 1 << 7,
1274 FAULT_FLAG_INSTRUCTION = 1 << 8,
1275 FAULT_FLAG_INTERRUPTIBLE = 1 << 9,
1276 FAULT_FLAG_UNSHARE = 1 << 10,
1277 FAULT_FLAG_ORIG_PTE_VALID = 1 << 11,
1278 FAULT_FLAG_VMA_LOCK = 1 << 12,
1279 };
1280
1281 typedef unsigned int __bitwise zap_flags_t;
1282
1283 /*
1284 * FOLL_PIN and FOLL_LONGTERM may be used in various combinations with each
1285 * other. Here is what they mean, and how to use them:
1286 *
1287 *
1288 * FIXME: For pages which are part of a filesystem, mappings are subject to the
1289 * lifetime enforced by the filesystem and we need guarantees that longterm
1290 * users like RDMA and V4L2 only establish mappings which coordinate usage with
1291 * the filesystem. Ideas for this coordination include revoking the longterm
1292 * pin, delaying writeback, bounce buffer page writeback, etc. As FS DAX was
1293 * added after the problem with filesystems was found FS DAX VMAs are
1294 * specifically failed. Filesystem pages are still subject to bugs and use of
1295 * FOLL_LONGTERM should be avoided on those pages.
1296 *
1297 * In the CMA case: long term pins in a CMA region would unnecessarily fragment
1298 * that region. And so, CMA attempts to migrate the page before pinning, when
1299 * FOLL_LONGTERM is specified.
1300 *
1301 * FOLL_PIN indicates that a special kind of tracking (not just page->_refcount,
1302 * but an additional pin counting system) will be invoked. This is intended for
1303 * anything that gets a page reference and then touches page data (for example,
1304 * Direct IO). This lets the filesystem know that some non-file-system entity is
1305 * potentially changing the pages' data. In contrast to FOLL_GET (whose pages
1306 * are released via put_page()), FOLL_PIN pages must be released, ultimately, by
1307 * a call to unpin_user_page().
1308 *
1309 * FOLL_PIN is similar to FOLL_GET: both of these pin pages. They use different
1310 * and separate refcounting mechanisms, however, and that means that each has
1311 * its own acquire and release mechanisms:
1312 *
1313 * FOLL_GET: get_user_pages*() to acquire, and put_page() to release.
1314 *
1315 * FOLL_PIN: pin_user_pages*() to acquire, and unpin_user_pages to release.
1316 *
1317 * FOLL_PIN and FOLL_GET are mutually exclusive for a given function call.
1318 * (The underlying pages may experience both FOLL_GET-based and FOLL_PIN-based
1319 * calls applied to them, and that's perfectly OK. This is a constraint on the
1320 * callers, not on the pages.)
1321 *
1322 * FOLL_PIN should be set internally by the pin_user_pages*() APIs, never
1323 * directly by the caller. That's in order to help avoid mismatches when
1324 * releasing pages: get_user_pages*() pages must be released via put_page(),
1325 * while pin_user_pages*() pages must be released via unpin_user_page().
1326 *
1327 * Please see Documentation/core-api/pin_user_pages.rst for more information.
1328 */
1329
1330 enum {
1331 /* check pte is writable */
1332 FOLL_WRITE = 1 << 0,
1333 /* do get_page on page */
1334 FOLL_GET = 1 << 1,
1335 /* give error on hole if it would be zero */
1336 FOLL_DUMP = 1 << 2,
1337 /* get_user_pages read/write w/o permission */
1338 FOLL_FORCE = 1 << 3,
1339 /*
1340 * if a disk transfer is needed, start the IO and return without waiting
1341 * upon it
1342 */
1343 FOLL_NOWAIT = 1 << 4,
1344 /* do not fault in pages */
1345 FOLL_NOFAULT = 1 << 5,
1346 /* check page is hwpoisoned */
1347 FOLL_HWPOISON = 1 << 6,
1348 /* don't do file mappings */
1349 FOLL_ANON = 1 << 7,
1350 /*
1351 * FOLL_LONGTERM indicates that the page will be held for an indefinite
1352 * time period _often_ under userspace control. This is in contrast to
1353 * iov_iter_get_pages(), whose usages are transient.
1354 */
1355 FOLL_LONGTERM = 1 << 8,
1356 /* split huge pmd before returning */
1357 FOLL_SPLIT_PMD = 1 << 9,
1358 /* allow returning PCI P2PDMA pages */
1359 FOLL_PCI_P2PDMA = 1 << 10,
1360 /* allow interrupts from generic signals */
1361 FOLL_INTERRUPTIBLE = 1 << 11,
1362 /*
1363 * Always honor (trigger) NUMA hinting faults.
1364 *
1365 * FOLL_WRITE implicitly honors NUMA hinting faults because a
1366 * PROT_NONE-mapped page is not writable (exceptions with FOLL_FORCE
1367 * apply). get_user_pages_fast_only() always implicitly honors NUMA
1368 * hinting faults.
1369 */
1370 FOLL_HONOR_NUMA_FAULT = 1 << 12,
1371
1372 /* See also internal only FOLL flags in mm/internal.h */
1373 };
1374
1375 #endif /* _LINUX_MM_TYPES_H */
1376