1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _LINUX_MM_H
3#define _LINUX_MM_H
4
5#include <linux/errno.h>
6#include <linux/mmdebug.h>
7#include <linux/gfp.h>
8#include <linux/bug.h>
9#include <linux/list.h>
10#include <linux/mmzone.h>
11#include <linux/rbtree.h>
12#include <linux/atomic.h>
13#include <linux/debug_locks.h>
14#include <linux/mm_types.h>
15#include <linux/mmap_lock.h>
16#include <linux/range.h>
17#include <linux/pfn.h>
18#include <linux/percpu-refcount.h>
19#include <linux/bit_spinlock.h>
20#include <linux/shrinker.h>
21#include <linux/resource.h>
22#include <linux/page_ext.h>
23#include <linux/err.h>
24#include <linux/page-flags.h>
25#include <linux/page_ref.h>
26#include <linux/overflow.h>
27#include <linux/sizes.h>
28#include <linux/sched.h>
29#include <linux/pgtable.h>
30#include <linux/kasan.h>
31#include <linux/memremap.h>
32#include <linux/slab.h>
33
34struct mempolicy;
35struct anon_vma;
36struct anon_vma_chain;
37struct user_struct;
38struct pt_regs;
39struct folio_batch;
40
41extern int sysctl_page_lock_unfairness;
42
43void mm_core_init(void);
44void init_mm_internals(void);
45
46#ifndef CONFIG_NUMA /* Don't use mapnrs, do it properly */
47extern unsigned long max_mapnr;
48
49static inline void set_max_mapnr(unsigned long limit)
50{
51 max_mapnr = limit;
52}
53#else
54static inline void set_max_mapnr(unsigned long limit) { }
55#endif
56
57extern atomic_long_t _totalram_pages;
58static inline unsigned long totalram_pages(void)
59{
60 return (unsigned long)atomic_long_read(v: &_totalram_pages);
61}
62
63static inline void totalram_pages_inc(void)
64{
65 atomic_long_inc(v: &_totalram_pages);
66}
67
68static inline void totalram_pages_dec(void)
69{
70 atomic_long_dec(v: &_totalram_pages);
71}
72
73static inline void totalram_pages_add(long count)
74{
75 atomic_long_add(i: count, v: &_totalram_pages);
76}
77
78extern void * high_memory;
79extern int page_cluster;
80extern const int page_cluster_max;
81
82#ifdef CONFIG_SYSCTL
83extern int sysctl_legacy_va_layout;
84#else
85#define sysctl_legacy_va_layout 0
86#endif
87
88#ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
89extern const int mmap_rnd_bits_min;
90extern int mmap_rnd_bits_max __ro_after_init;
91extern int mmap_rnd_bits __read_mostly;
92#endif
93#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
94extern const int mmap_rnd_compat_bits_min;
95extern const int mmap_rnd_compat_bits_max;
96extern int mmap_rnd_compat_bits __read_mostly;
97#endif
98
99#include <asm/page.h>
100#include <asm/processor.h>
101
102#ifndef __pa_symbol
103#define __pa_symbol(x) __pa(RELOC_HIDE((unsigned long)(x), 0))
104#endif
105
106#ifndef page_to_virt
107#define page_to_virt(x) __va(PFN_PHYS(page_to_pfn(x)))
108#endif
109
110#ifndef lm_alias
111#define lm_alias(x) __va(__pa_symbol(x))
112#endif
113
114/*
115 * To prevent common memory management code establishing
116 * a zero page mapping on a read fault.
117 * This macro should be defined within <asm/pgtable.h>.
118 * s390 does this to prevent multiplexing of hardware bits
119 * related to the physical page in case of virtualization.
120 */
121#ifndef mm_forbids_zeropage
122#define mm_forbids_zeropage(X) (0)
123#endif
124
125/*
126 * On some architectures it is expensive to call memset() for small sizes.
127 * If an architecture decides to implement their own version of
128 * mm_zero_struct_page they should wrap the defines below in a #ifndef and
129 * define their own version of this macro in <asm/pgtable.h>
130 */
131#if BITS_PER_LONG == 64
132/* This function must be updated when the size of struct page grows above 96
133 * or reduces below 56. The idea that compiler optimizes out switch()
134 * statement, and only leaves move/store instructions. Also the compiler can
135 * combine write statements if they are both assignments and can be reordered,
136 * this can result in several of the writes here being dropped.
137 */
138#define mm_zero_struct_page(pp) __mm_zero_struct_page(pp)
139static inline void __mm_zero_struct_page(struct page *page)
140{
141 unsigned long *_pp = (void *)page;
142
143 /* Check that struct page is either 56, 64, 72, 80, 88 or 96 bytes */
144 BUILD_BUG_ON(sizeof(struct page) & 7);
145 BUILD_BUG_ON(sizeof(struct page) < 56);
146 BUILD_BUG_ON(sizeof(struct page) > 96);
147
148 switch (sizeof(struct page)) {
149 case 96:
150 _pp[11] = 0;
151 fallthrough;
152 case 88:
153 _pp[10] = 0;
154 fallthrough;
155 case 80:
156 _pp[9] = 0;
157 fallthrough;
158 case 72:
159 _pp[8] = 0;
160 fallthrough;
161 case 64:
162 _pp[7] = 0;
163 fallthrough;
164 case 56:
165 _pp[6] = 0;
166 _pp[5] = 0;
167 _pp[4] = 0;
168 _pp[3] = 0;
169 _pp[2] = 0;
170 _pp[1] = 0;
171 _pp[0] = 0;
172 }
173}
174#else
175#define mm_zero_struct_page(pp) ((void)memset((pp), 0, sizeof(struct page)))
176#endif
177
178/*
179 * Default maximum number of active map areas, this limits the number of vmas
180 * per mm struct. Users can overwrite this number by sysctl but there is a
181 * problem.
182 *
183 * When a program's coredump is generated as ELF format, a section is created
184 * per a vma. In ELF, the number of sections is represented in unsigned short.
185 * This means the number of sections should be smaller than 65535 at coredump.
186 * Because the kernel adds some informative sections to a image of program at
187 * generating coredump, we need some margin. The number of extra sections is
188 * 1-3 now and depends on arch. We use "5" as safe margin, here.
189 *
190 * ELF extended numbering allows more than 65535 sections, so 16-bit bound is
191 * not a hard limit any more. Although some userspace tools can be surprised by
192 * that.
193 */
194#define MAPCOUNT_ELF_CORE_MARGIN (5)
195#define DEFAULT_MAX_MAP_COUNT (USHRT_MAX - MAPCOUNT_ELF_CORE_MARGIN)
196
197extern int sysctl_max_map_count;
198
199extern unsigned long sysctl_user_reserve_kbytes;
200extern unsigned long sysctl_admin_reserve_kbytes;
201
202extern int sysctl_overcommit_memory;
203extern int sysctl_overcommit_ratio;
204extern unsigned long sysctl_overcommit_kbytes;
205
206int overcommit_ratio_handler(struct ctl_table *, int, void *, size_t *,
207 loff_t *);
208int overcommit_kbytes_handler(struct ctl_table *, int, void *, size_t *,
209 loff_t *);
210int overcommit_policy_handler(struct ctl_table *, int, void *, size_t *,
211 loff_t *);
212
213#if defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP)
214#define nth_page(page,n) pfn_to_page(page_to_pfn((page)) + (n))
215#define folio_page_idx(folio, p) (page_to_pfn(p) - folio_pfn(folio))
216#else
217#define nth_page(page,n) ((page) + (n))
218#define folio_page_idx(folio, p) ((p) - &(folio)->page)
219#endif
220
221/* to align the pointer to the (next) page boundary */
222#define PAGE_ALIGN(addr) ALIGN(addr, PAGE_SIZE)
223
224/* to align the pointer to the (prev) page boundary */
225#define PAGE_ALIGN_DOWN(addr) ALIGN_DOWN(addr, PAGE_SIZE)
226
227/* test whether an address (unsigned long or pointer) is aligned to PAGE_SIZE */
228#define PAGE_ALIGNED(addr) IS_ALIGNED((unsigned long)(addr), PAGE_SIZE)
229
230static inline struct folio *lru_to_folio(struct list_head *head)
231{
232 return list_entry((head)->prev, struct folio, lru);
233}
234
235void setup_initial_init_mm(void *start_code, void *end_code,
236 void *end_data, void *brk);
237
238/*
239 * Linux kernel virtual memory manager primitives.
240 * The idea being to have a "virtual" mm in the same way
241 * we have a virtual fs - giving a cleaner interface to the
242 * mm details, and allowing different kinds of memory mappings
243 * (from shared memory to executable loading to arbitrary
244 * mmap() functions).
245 */
246
247struct vm_area_struct *vm_area_alloc(struct mm_struct *);
248struct vm_area_struct *vm_area_dup(struct vm_area_struct *);
249void vm_area_free(struct vm_area_struct *);
250/* Use only if VMA has no other users */
251void __vm_area_free(struct vm_area_struct *vma);
252
253#ifndef CONFIG_MMU
254extern struct rb_root nommu_region_tree;
255extern struct rw_semaphore nommu_region_sem;
256
257extern unsigned int kobjsize(const void *objp);
258#endif
259
260/*
261 * vm_flags in vm_area_struct, see mm_types.h.
262 * When changing, update also include/trace/events/mmflags.h
263 */
264#define VM_NONE 0x00000000
265
266#define VM_READ 0x00000001 /* currently active flags */
267#define VM_WRITE 0x00000002
268#define VM_EXEC 0x00000004
269#define VM_SHARED 0x00000008
270
271/* mprotect() hardcodes VM_MAYREAD >> 4 == VM_READ, and so for r/w/x bits. */
272#define VM_MAYREAD 0x00000010 /* limits for mprotect() etc */
273#define VM_MAYWRITE 0x00000020
274#define VM_MAYEXEC 0x00000040
275#define VM_MAYSHARE 0x00000080
276
277#define VM_GROWSDOWN 0x00000100 /* general info on the segment */
278#ifdef CONFIG_MMU
279#define VM_UFFD_MISSING 0x00000200 /* missing pages tracking */
280#else /* CONFIG_MMU */
281#define VM_MAYOVERLAY 0x00000200 /* nommu: R/O MAP_PRIVATE mapping that might overlay a file mapping */
282#define VM_UFFD_MISSING 0
283#endif /* CONFIG_MMU */
284#define VM_PFNMAP 0x00000400 /* Page-ranges managed without "struct page", just pure PFN */
285#define VM_UFFD_WP 0x00001000 /* wrprotect pages tracking */
286
287#define VM_LOCKED 0x00002000
288#define VM_IO 0x00004000 /* Memory mapped I/O or similar */
289
290 /* Used by sys_madvise() */
291#define VM_SEQ_READ 0x00008000 /* App will access data sequentially */
292#define VM_RAND_READ 0x00010000 /* App will not benefit from clustered reads */
293
294#define VM_DONTCOPY 0x00020000 /* Do not copy this vma on fork */
295#define VM_DONTEXPAND 0x00040000 /* Cannot expand with mremap() */
296#define VM_LOCKONFAULT 0x00080000 /* Lock the pages covered when they are faulted in */
297#define VM_ACCOUNT 0x00100000 /* Is a VM accounted object */
298#define VM_NORESERVE 0x00200000 /* should the VM suppress accounting */
299#define VM_HUGETLB 0x00400000 /* Huge TLB Page VM */
300#define VM_SYNC 0x00800000 /* Synchronous page faults */
301#define VM_ARCH_1 0x01000000 /* Architecture-specific flag */
302#define VM_WIPEONFORK 0x02000000 /* Wipe VMA contents in child. */
303#define VM_DONTDUMP 0x04000000 /* Do not include in the core dump */
304
305#ifdef CONFIG_MEM_SOFT_DIRTY
306# define VM_SOFTDIRTY 0x08000000 /* Not soft dirty clean area */
307#else
308# define VM_SOFTDIRTY 0
309#endif
310
311#define VM_MIXEDMAP 0x10000000 /* Can contain "struct page" and pure PFN pages */
312#define VM_HUGEPAGE 0x20000000 /* MADV_HUGEPAGE marked this vma */
313#define VM_NOHUGEPAGE 0x40000000 /* MADV_NOHUGEPAGE marked this vma */
314#define VM_MERGEABLE 0x80000000 /* KSM may merge identical pages */
315
316#ifdef CONFIG_ARCH_USES_HIGH_VMA_FLAGS
317#define VM_HIGH_ARCH_BIT_0 32 /* bit only usable on 64-bit architectures */
318#define VM_HIGH_ARCH_BIT_1 33 /* bit only usable on 64-bit architectures */
319#define VM_HIGH_ARCH_BIT_2 34 /* bit only usable on 64-bit architectures */
320#define VM_HIGH_ARCH_BIT_3 35 /* bit only usable on 64-bit architectures */
321#define VM_HIGH_ARCH_BIT_4 36 /* bit only usable on 64-bit architectures */
322#define VM_HIGH_ARCH_BIT_5 37 /* bit only usable on 64-bit architectures */
323#define VM_HIGH_ARCH_0 BIT(VM_HIGH_ARCH_BIT_0)
324#define VM_HIGH_ARCH_1 BIT(VM_HIGH_ARCH_BIT_1)
325#define VM_HIGH_ARCH_2 BIT(VM_HIGH_ARCH_BIT_2)
326#define VM_HIGH_ARCH_3 BIT(VM_HIGH_ARCH_BIT_3)
327#define VM_HIGH_ARCH_4 BIT(VM_HIGH_ARCH_BIT_4)
328#define VM_HIGH_ARCH_5 BIT(VM_HIGH_ARCH_BIT_5)
329#endif /* CONFIG_ARCH_USES_HIGH_VMA_FLAGS */
330
331#ifdef CONFIG_ARCH_HAS_PKEYS
332# define VM_PKEY_SHIFT VM_HIGH_ARCH_BIT_0
333# define VM_PKEY_BIT0 VM_HIGH_ARCH_0 /* A protection key is a 4-bit value */
334# define VM_PKEY_BIT1 VM_HIGH_ARCH_1 /* on x86 and 5-bit value on ppc64 */
335# define VM_PKEY_BIT2 VM_HIGH_ARCH_2
336# define VM_PKEY_BIT3 VM_HIGH_ARCH_3
337#ifdef CONFIG_PPC
338# define VM_PKEY_BIT4 VM_HIGH_ARCH_4
339#else
340# define VM_PKEY_BIT4 0
341#endif
342#endif /* CONFIG_ARCH_HAS_PKEYS */
343
344#ifdef CONFIG_X86_USER_SHADOW_STACK
345/*
346 * VM_SHADOW_STACK should not be set with VM_SHARED because of lack of
347 * support core mm.
348 *
349 * These VMAs will get a single end guard page. This helps userspace protect
350 * itself from attacks. A single page is enough for current shadow stack archs
351 * (x86). See the comments near alloc_shstk() in arch/x86/kernel/shstk.c
352 * for more details on the guard size.
353 */
354# define VM_SHADOW_STACK VM_HIGH_ARCH_5
355#else
356# define VM_SHADOW_STACK VM_NONE
357#endif
358
359#if defined(CONFIG_X86)
360# define VM_PAT VM_ARCH_1 /* PAT reserves whole VMA at once (x86) */
361#elif defined(CONFIG_PPC)
362# define VM_SAO VM_ARCH_1 /* Strong Access Ordering (powerpc) */
363#elif defined(CONFIG_PARISC)
364# define VM_GROWSUP VM_ARCH_1
365#elif defined(CONFIG_SPARC64)
366# define VM_SPARC_ADI VM_ARCH_1 /* Uses ADI tag for access control */
367# define VM_ARCH_CLEAR VM_SPARC_ADI
368#elif defined(CONFIG_ARM64)
369# define VM_ARM64_BTI VM_ARCH_1 /* BTI guarded page, a.k.a. GP bit */
370# define VM_ARCH_CLEAR VM_ARM64_BTI
371#elif !defined(CONFIG_MMU)
372# define VM_MAPPED_COPY VM_ARCH_1 /* T if mapped copy of data (nommu mmap) */
373#endif
374
375#if defined(CONFIG_ARM64_MTE)
376# define VM_MTE VM_HIGH_ARCH_0 /* Use Tagged memory for access control */
377# define VM_MTE_ALLOWED VM_HIGH_ARCH_1 /* Tagged memory permitted */
378#else
379# define VM_MTE VM_NONE
380# define VM_MTE_ALLOWED VM_NONE
381#endif
382
383#ifndef VM_GROWSUP
384# define VM_GROWSUP VM_NONE
385#endif
386
387#ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
388# define VM_UFFD_MINOR_BIT 38
389# define VM_UFFD_MINOR BIT(VM_UFFD_MINOR_BIT) /* UFFD minor faults */
390#else /* !CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
391# define VM_UFFD_MINOR VM_NONE
392#endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
393
394/*
395 * This flag is used to connect VFIO to arch specific KVM code. It
396 * indicates that the memory under this VMA is safe for use with any
397 * non-cachable memory type inside KVM. Some VFIO devices, on some
398 * platforms, are thought to be unsafe and can cause machine crashes
399 * if KVM does not lock down the memory type.
400 */
401#ifdef CONFIG_64BIT
402#define VM_ALLOW_ANY_UNCACHED_BIT 39
403#define VM_ALLOW_ANY_UNCACHED BIT(VM_ALLOW_ANY_UNCACHED_BIT)
404#else
405#define VM_ALLOW_ANY_UNCACHED VM_NONE
406#endif
407
408/* Bits set in the VMA until the stack is in its final location */
409#define VM_STACK_INCOMPLETE_SETUP (VM_RAND_READ | VM_SEQ_READ | VM_STACK_EARLY)
410
411#define TASK_EXEC ((current->personality & READ_IMPLIES_EXEC) ? VM_EXEC : 0)
412
413/* Common data flag combinations */
414#define VM_DATA_FLAGS_TSK_EXEC (VM_READ | VM_WRITE | TASK_EXEC | \
415 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC)
416#define VM_DATA_FLAGS_NON_EXEC (VM_READ | VM_WRITE | VM_MAYREAD | \
417 VM_MAYWRITE | VM_MAYEXEC)
418#define VM_DATA_FLAGS_EXEC (VM_READ | VM_WRITE | VM_EXEC | \
419 VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC)
420
421#ifndef VM_DATA_DEFAULT_FLAGS /* arch can override this */
422#define VM_DATA_DEFAULT_FLAGS VM_DATA_FLAGS_EXEC
423#endif
424
425#ifndef VM_STACK_DEFAULT_FLAGS /* arch can override this */
426#define VM_STACK_DEFAULT_FLAGS VM_DATA_DEFAULT_FLAGS
427#endif
428
429#define VM_STARTGAP_FLAGS (VM_GROWSDOWN | VM_SHADOW_STACK)
430
431#ifdef CONFIG_STACK_GROWSUP
432#define VM_STACK VM_GROWSUP
433#define VM_STACK_EARLY VM_GROWSDOWN
434#else
435#define VM_STACK VM_GROWSDOWN
436#define VM_STACK_EARLY 0
437#endif
438
439#define VM_STACK_FLAGS (VM_STACK | VM_STACK_DEFAULT_FLAGS | VM_ACCOUNT)
440
441/* VMA basic access permission flags */
442#define VM_ACCESS_FLAGS (VM_READ | VM_WRITE | VM_EXEC)
443
444
445/*
446 * Special vmas that are non-mergable, non-mlock()able.
447 */
448#define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_PFNMAP | VM_MIXEDMAP)
449
450/* This mask prevents VMA from being scanned with khugepaged */
451#define VM_NO_KHUGEPAGED (VM_SPECIAL | VM_HUGETLB)
452
453/* This mask defines which mm->def_flags a process can inherit its parent */
454#define VM_INIT_DEF_MASK VM_NOHUGEPAGE
455
456/* This mask represents all the VMA flag bits used by mlock */
457#define VM_LOCKED_MASK (VM_LOCKED | VM_LOCKONFAULT)
458
459/* Arch-specific flags to clear when updating VM flags on protection change */
460#ifndef VM_ARCH_CLEAR
461# define VM_ARCH_CLEAR VM_NONE
462#endif
463#define VM_FLAGS_CLEAR (ARCH_VM_PKEY_FLAGS | VM_ARCH_CLEAR)
464
465/*
466 * mapping from the currently active vm_flags protection bits (the
467 * low four bits) to a page protection mask..
468 */
469
470/*
471 * The default fault flags that should be used by most of the
472 * arch-specific page fault handlers.
473 */
474#define FAULT_FLAG_DEFAULT (FAULT_FLAG_ALLOW_RETRY | \
475 FAULT_FLAG_KILLABLE | \
476 FAULT_FLAG_INTERRUPTIBLE)
477
478/**
479 * fault_flag_allow_retry_first - check ALLOW_RETRY the first time
480 * @flags: Fault flags.
481 *
482 * This is mostly used for places where we want to try to avoid taking
483 * the mmap_lock for too long a time when waiting for another condition
484 * to change, in which case we can try to be polite to release the
485 * mmap_lock in the first round to avoid potential starvation of other
486 * processes that would also want the mmap_lock.
487 *
488 * Return: true if the page fault allows retry and this is the first
489 * attempt of the fault handling; false otherwise.
490 */
491static inline bool fault_flag_allow_retry_first(enum fault_flag flags)
492{
493 return (flags & FAULT_FLAG_ALLOW_RETRY) &&
494 (!(flags & FAULT_FLAG_TRIED));
495}
496
497#define FAULT_FLAG_TRACE \
498 { FAULT_FLAG_WRITE, "WRITE" }, \
499 { FAULT_FLAG_MKWRITE, "MKWRITE" }, \
500 { FAULT_FLAG_ALLOW_RETRY, "ALLOW_RETRY" }, \
501 { FAULT_FLAG_RETRY_NOWAIT, "RETRY_NOWAIT" }, \
502 { FAULT_FLAG_KILLABLE, "KILLABLE" }, \
503 { FAULT_FLAG_TRIED, "TRIED" }, \
504 { FAULT_FLAG_USER, "USER" }, \
505 { FAULT_FLAG_REMOTE, "REMOTE" }, \
506 { FAULT_FLAG_INSTRUCTION, "INSTRUCTION" }, \
507 { FAULT_FLAG_INTERRUPTIBLE, "INTERRUPTIBLE" }, \
508 { FAULT_FLAG_VMA_LOCK, "VMA_LOCK" }
509
510/*
511 * vm_fault is filled by the pagefault handler and passed to the vma's
512 * ->fault function. The vma's ->fault is responsible for returning a bitmask
513 * of VM_FAULT_xxx flags that give details about how the fault was handled.
514 *
515 * MM layer fills up gfp_mask for page allocations but fault handler might
516 * alter it if its implementation requires a different allocation context.
517 *
518 * pgoff should be used in favour of virtual_address, if possible.
519 */
520struct vm_fault {
521 const struct {
522 struct vm_area_struct *vma; /* Target VMA */
523 gfp_t gfp_mask; /* gfp mask to be used for allocations */
524 pgoff_t pgoff; /* Logical page offset based on vma */
525 unsigned long address; /* Faulting virtual address - masked */
526 unsigned long real_address; /* Faulting virtual address - unmasked */
527 };
528 enum fault_flag flags; /* FAULT_FLAG_xxx flags
529 * XXX: should really be 'const' */
530 pmd_t *pmd; /* Pointer to pmd entry matching
531 * the 'address' */
532 pud_t *pud; /* Pointer to pud entry matching
533 * the 'address'
534 */
535 union {
536 pte_t orig_pte; /* Value of PTE at the time of fault */
537 pmd_t orig_pmd; /* Value of PMD at the time of fault,
538 * used by PMD fault only.
539 */
540 };
541
542 struct page *cow_page; /* Page handler may use for COW fault */
543 struct page *page; /* ->fault handlers should return a
544 * page here, unless VM_FAULT_NOPAGE
545 * is set (which is also implied by
546 * VM_FAULT_ERROR).
547 */
548 /* These three entries are valid only while holding ptl lock */
549 pte_t *pte; /* Pointer to pte entry matching
550 * the 'address'. NULL if the page
551 * table hasn't been allocated.
552 */
553 spinlock_t *ptl; /* Page table lock.
554 * Protects pte page table if 'pte'
555 * is not NULL, otherwise pmd.
556 */
557 pgtable_t prealloc_pte; /* Pre-allocated pte page table.
558 * vm_ops->map_pages() sets up a page
559 * table from atomic context.
560 * do_fault_around() pre-allocates
561 * page table to avoid allocation from
562 * atomic context.
563 */
564};
565
566/*
567 * These are the virtual MM functions - opening of an area, closing and
568 * unmapping it (needed to keep files on disk up-to-date etc), pointer
569 * to the functions called when a no-page or a wp-page exception occurs.
570 */
571struct vm_operations_struct {
572 void (*open)(struct vm_area_struct * area);
573 /**
574 * @close: Called when the VMA is being removed from the MM.
575 * Context: User context. May sleep. Caller holds mmap_lock.
576 */
577 void (*close)(struct vm_area_struct * area);
578 /* Called any time before splitting to check if it's allowed */
579 int (*may_split)(struct vm_area_struct *area, unsigned long addr);
580 int (*mremap)(struct vm_area_struct *area);
581 /*
582 * Called by mprotect() to make driver-specific permission
583 * checks before mprotect() is finalised. The VMA must not
584 * be modified. Returns 0 if mprotect() can proceed.
585 */
586 int (*mprotect)(struct vm_area_struct *vma, unsigned long start,
587 unsigned long end, unsigned long newflags);
588 vm_fault_t (*fault)(struct vm_fault *vmf);
589 vm_fault_t (*huge_fault)(struct vm_fault *vmf, unsigned int order);
590 vm_fault_t (*map_pages)(struct vm_fault *vmf,
591 pgoff_t start_pgoff, pgoff_t end_pgoff);
592 unsigned long (*pagesize)(struct vm_area_struct * area);
593
594 /* notification that a previously read-only page is about to become
595 * writable, if an error is returned it will cause a SIGBUS */
596 vm_fault_t (*page_mkwrite)(struct vm_fault *vmf);
597
598 /* same as page_mkwrite when using VM_PFNMAP|VM_MIXEDMAP */
599 vm_fault_t (*pfn_mkwrite)(struct vm_fault *vmf);
600
601 /* called by access_process_vm when get_user_pages() fails, typically
602 * for use by special VMAs. See also generic_access_phys() for a generic
603 * implementation useful for any iomem mapping.
604 */
605 int (*access)(struct vm_area_struct *vma, unsigned long addr,
606 void *buf, int len, int write);
607
608 /* Called by the /proc/PID/maps code to ask the vma whether it
609 * has a special name. Returning non-NULL will also cause this
610 * vma to be dumped unconditionally. */
611 const char *(*name)(struct vm_area_struct *vma);
612
613#ifdef CONFIG_NUMA
614 /*
615 * set_policy() op must add a reference to any non-NULL @new mempolicy
616 * to hold the policy upon return. Caller should pass NULL @new to
617 * remove a policy and fall back to surrounding context--i.e. do not
618 * install a MPOL_DEFAULT policy, nor the task or system default
619 * mempolicy.
620 */
621 int (*set_policy)(struct vm_area_struct *vma, struct mempolicy *new);
622
623 /*
624 * get_policy() op must add reference [mpol_get()] to any policy at
625 * (vma,addr) marked as MPOL_SHARED. The shared policy infrastructure
626 * in mm/mempolicy.c will do this automatically.
627 * get_policy() must NOT add a ref if the policy at (vma,addr) is not
628 * marked as MPOL_SHARED. vma policies are protected by the mmap_lock.
629 * If no [shared/vma] mempolicy exists at the addr, get_policy() op
630 * must return NULL--i.e., do not "fallback" to task or system default
631 * policy.
632 */
633 struct mempolicy *(*get_policy)(struct vm_area_struct *vma,
634 unsigned long addr, pgoff_t *ilx);
635#endif
636 /*
637 * Called by vm_normal_page() for special PTEs to find the
638 * page for @addr. This is useful if the default behavior
639 * (using pte_page()) would not find the correct page.
640 */
641 struct page *(*find_special_page)(struct vm_area_struct *vma,
642 unsigned long addr);
643};
644
645#ifdef CONFIG_NUMA_BALANCING
646static inline void vma_numab_state_init(struct vm_area_struct *vma)
647{
648 vma->numab_state = NULL;
649}
650static inline void vma_numab_state_free(struct vm_area_struct *vma)
651{
652 kfree(objp: vma->numab_state);
653}
654#else
655static inline void vma_numab_state_init(struct vm_area_struct *vma) {}
656static inline void vma_numab_state_free(struct vm_area_struct *vma) {}
657#endif /* CONFIG_NUMA_BALANCING */
658
659#ifdef CONFIG_PER_VMA_LOCK
660/*
661 * Try to read-lock a vma. The function is allowed to occasionally yield false
662 * locked result to avoid performance overhead, in which case we fall back to
663 * using mmap_lock. The function should never yield false unlocked result.
664 */
665static inline bool vma_start_read(struct vm_area_struct *vma)
666{
667 /*
668 * Check before locking. A race might cause false locked result.
669 * We can use READ_ONCE() for the mm_lock_seq here, and don't need
670 * ACQUIRE semantics, because this is just a lockless check whose result
671 * we don't rely on for anything - the mm_lock_seq read against which we
672 * need ordering is below.
673 */
674 if (READ_ONCE(vma->vm_lock_seq) == READ_ONCE(vma->vm_mm->mm_lock_seq))
675 return false;
676
677 if (unlikely(down_read_trylock(&vma->vm_lock->lock) == 0))
678 return false;
679
680 /*
681 * Overflow might produce false locked result.
682 * False unlocked result is impossible because we modify and check
683 * vma->vm_lock_seq under vma->vm_lock protection and mm->mm_lock_seq
684 * modification invalidates all existing locks.
685 *
686 * We must use ACQUIRE semantics for the mm_lock_seq so that if we are
687 * racing with vma_end_write_all(), we only start reading from the VMA
688 * after it has been unlocked.
689 * This pairs with RELEASE semantics in vma_end_write_all().
690 */
691 if (unlikely(vma->vm_lock_seq == smp_load_acquire(&vma->vm_mm->mm_lock_seq))) {
692 up_read(sem: &vma->vm_lock->lock);
693 return false;
694 }
695 return true;
696}
697
698static inline void vma_end_read(struct vm_area_struct *vma)
699{
700 rcu_read_lock(); /* keeps vma alive till the end of up_read */
701 up_read(sem: &vma->vm_lock->lock);
702 rcu_read_unlock();
703}
704
705/* WARNING! Can only be used if mmap_lock is expected to be write-locked */
706static bool __is_vma_write_locked(struct vm_area_struct *vma, int *mm_lock_seq)
707{
708 mmap_assert_write_locked(mm: vma->vm_mm);
709
710 /*
711 * current task is holding mmap_write_lock, both vma->vm_lock_seq and
712 * mm->mm_lock_seq can't be concurrently modified.
713 */
714 *mm_lock_seq = vma->vm_mm->mm_lock_seq;
715 return (vma->vm_lock_seq == *mm_lock_seq);
716}
717
718/*
719 * Begin writing to a VMA.
720 * Exclude concurrent readers under the per-VMA lock until the currently
721 * write-locked mmap_lock is dropped or downgraded.
722 */
723static inline void vma_start_write(struct vm_area_struct *vma)
724{
725 int mm_lock_seq;
726
727 if (__is_vma_write_locked(vma, mm_lock_seq: &mm_lock_seq))
728 return;
729
730 down_write(sem: &vma->vm_lock->lock);
731 /*
732 * We should use WRITE_ONCE() here because we can have concurrent reads
733 * from the early lockless pessimistic check in vma_start_read().
734 * We don't really care about the correctness of that early check, but
735 * we should use WRITE_ONCE() for cleanliness and to keep KCSAN happy.
736 */
737 WRITE_ONCE(vma->vm_lock_seq, mm_lock_seq);
738 up_write(sem: &vma->vm_lock->lock);
739}
740
741static inline void vma_assert_write_locked(struct vm_area_struct *vma)
742{
743 int mm_lock_seq;
744
745 VM_BUG_ON_VMA(!__is_vma_write_locked(vma, &mm_lock_seq), vma);
746}
747
748static inline void vma_assert_locked(struct vm_area_struct *vma)
749{
750 if (!rwsem_is_locked(sem: &vma->vm_lock->lock))
751 vma_assert_write_locked(vma);
752}
753
754static inline void vma_mark_detached(struct vm_area_struct *vma, bool detached)
755{
756 /* When detaching vma should be write-locked */
757 if (detached)
758 vma_assert_write_locked(vma);
759 vma->detached = detached;
760}
761
762static inline void release_fault_lock(struct vm_fault *vmf)
763{
764 if (vmf->flags & FAULT_FLAG_VMA_LOCK)
765 vma_end_read(vma: vmf->vma);
766 else
767 mmap_read_unlock(mm: vmf->vma->vm_mm);
768}
769
770static inline void assert_fault_locked(struct vm_fault *vmf)
771{
772 if (vmf->flags & FAULT_FLAG_VMA_LOCK)
773 vma_assert_locked(vma: vmf->vma);
774 else
775 mmap_assert_locked(mm: vmf->vma->vm_mm);
776}
777
778struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
779 unsigned long address);
780
781#else /* CONFIG_PER_VMA_LOCK */
782
783static inline bool vma_start_read(struct vm_area_struct *vma)
784 { return false; }
785static inline void vma_end_read(struct vm_area_struct *vma) {}
786static inline void vma_start_write(struct vm_area_struct *vma) {}
787static inline void vma_assert_write_locked(struct vm_area_struct *vma)
788 { mmap_assert_write_locked(vma->vm_mm); }
789static inline void vma_mark_detached(struct vm_area_struct *vma,
790 bool detached) {}
791
792static inline struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
793 unsigned long address)
794{
795 return NULL;
796}
797
798static inline void vma_assert_locked(struct vm_area_struct *vma)
799{
800 mmap_assert_locked(vma->vm_mm);
801}
802
803static inline void release_fault_lock(struct vm_fault *vmf)
804{
805 mmap_read_unlock(vmf->vma->vm_mm);
806}
807
808static inline void assert_fault_locked(struct vm_fault *vmf)
809{
810 mmap_assert_locked(vmf->vma->vm_mm);
811}
812
813#endif /* CONFIG_PER_VMA_LOCK */
814
815extern const struct vm_operations_struct vma_dummy_vm_ops;
816
817/*
818 * WARNING: vma_init does not initialize vma->vm_lock.
819 * Use vm_area_alloc()/vm_area_free() if vma needs locking.
820 */
821static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm)
822{
823 memset(vma, 0, sizeof(*vma));
824 vma->vm_mm = mm;
825 vma->vm_ops = &vma_dummy_vm_ops;
826 INIT_LIST_HEAD(list: &vma->anon_vma_chain);
827 vma_mark_detached(vma, detached: false);
828 vma_numab_state_init(vma);
829}
830
831/* Use when VMA is not part of the VMA tree and needs no locking */
832static inline void vm_flags_init(struct vm_area_struct *vma,
833 vm_flags_t flags)
834{
835 ACCESS_PRIVATE(vma, __vm_flags) = flags;
836}
837
838/*
839 * Use when VMA is part of the VMA tree and modifications need coordination
840 * Note: vm_flags_reset and vm_flags_reset_once do not lock the vma and
841 * it should be locked explicitly beforehand.
842 */
843static inline void vm_flags_reset(struct vm_area_struct *vma,
844 vm_flags_t flags)
845{
846 vma_assert_write_locked(vma);
847 vm_flags_init(vma, flags);
848}
849
850static inline void vm_flags_reset_once(struct vm_area_struct *vma,
851 vm_flags_t flags)
852{
853 vma_assert_write_locked(vma);
854 WRITE_ONCE(ACCESS_PRIVATE(vma, __vm_flags), flags);
855}
856
857static inline void vm_flags_set(struct vm_area_struct *vma,
858 vm_flags_t flags)
859{
860 vma_start_write(vma);
861 ACCESS_PRIVATE(vma, __vm_flags) |= flags;
862}
863
864static inline void vm_flags_clear(struct vm_area_struct *vma,
865 vm_flags_t flags)
866{
867 vma_start_write(vma);
868 ACCESS_PRIVATE(vma, __vm_flags) &= ~flags;
869}
870
871/*
872 * Use only if VMA is not part of the VMA tree or has no other users and
873 * therefore needs no locking.
874 */
875static inline void __vm_flags_mod(struct vm_area_struct *vma,
876 vm_flags_t set, vm_flags_t clear)
877{
878 vm_flags_init(vma, flags: (vma->vm_flags | set) & ~clear);
879}
880
881/*
882 * Use only when the order of set/clear operations is unimportant, otherwise
883 * use vm_flags_{set|clear} explicitly.
884 */
885static inline void vm_flags_mod(struct vm_area_struct *vma,
886 vm_flags_t set, vm_flags_t clear)
887{
888 vma_start_write(vma);
889 __vm_flags_mod(vma, set, clear);
890}
891
892static inline void vma_set_anonymous(struct vm_area_struct *vma)
893{
894 vma->vm_ops = NULL;
895}
896
897static inline bool vma_is_anonymous(struct vm_area_struct *vma)
898{
899 return !vma->vm_ops;
900}
901
902/*
903 * Indicate if the VMA is a heap for the given task; for
904 * /proc/PID/maps that is the heap of the main task.
905 */
906static inline bool vma_is_initial_heap(const struct vm_area_struct *vma)
907{
908 return vma->vm_start < vma->vm_mm->brk &&
909 vma->vm_end > vma->vm_mm->start_brk;
910}
911
912/*
913 * Indicate if the VMA is a stack for the given task; for
914 * /proc/PID/maps that is the stack of the main task.
915 */
916static inline bool vma_is_initial_stack(const struct vm_area_struct *vma)
917{
918 /*
919 * We make no effort to guess what a given thread considers to be
920 * its "stack". It's not even well-defined for programs written
921 * languages like Go.
922 */
923 return vma->vm_start <= vma->vm_mm->start_stack &&
924 vma->vm_end >= vma->vm_mm->start_stack;
925}
926
927static inline bool vma_is_temporary_stack(struct vm_area_struct *vma)
928{
929 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
930
931 if (!maybe_stack)
932 return false;
933
934 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
935 VM_STACK_INCOMPLETE_SETUP)
936 return true;
937
938 return false;
939}
940
941static inline bool vma_is_foreign(struct vm_area_struct *vma)
942{
943 if (!current->mm)
944 return true;
945
946 if (current->mm != vma->vm_mm)
947 return true;
948
949 return false;
950}
951
952static inline bool vma_is_accessible(struct vm_area_struct *vma)
953{
954 return vma->vm_flags & VM_ACCESS_FLAGS;
955}
956
957static inline bool is_shared_maywrite(vm_flags_t vm_flags)
958{
959 return (vm_flags & (VM_SHARED | VM_MAYWRITE)) ==
960 (VM_SHARED | VM_MAYWRITE);
961}
962
963static inline bool vma_is_shared_maywrite(struct vm_area_struct *vma)
964{
965 return is_shared_maywrite(vm_flags: vma->vm_flags);
966}
967
968static inline
969struct vm_area_struct *vma_find(struct vma_iterator *vmi, unsigned long max)
970{
971 return mas_find(mas: &vmi->mas, max: max - 1);
972}
973
974static inline struct vm_area_struct *vma_next(struct vma_iterator *vmi)
975{
976 /*
977 * Uses mas_find() to get the first VMA when the iterator starts.
978 * Calling mas_next() could skip the first entry.
979 */
980 return mas_find(mas: &vmi->mas, ULONG_MAX);
981}
982
983static inline
984struct vm_area_struct *vma_iter_next_range(struct vma_iterator *vmi)
985{
986 return mas_next_range(mas: &vmi->mas, ULONG_MAX);
987}
988
989
990static inline struct vm_area_struct *vma_prev(struct vma_iterator *vmi)
991{
992 return mas_prev(mas: &vmi->mas, min: 0);
993}
994
995static inline
996struct vm_area_struct *vma_iter_prev_range(struct vma_iterator *vmi)
997{
998 return mas_prev_range(mas: &vmi->mas, max: 0);
999}
1000
1001static inline unsigned long vma_iter_addr(struct vma_iterator *vmi)
1002{
1003 return vmi->mas.index;
1004}
1005
1006static inline unsigned long vma_iter_end(struct vma_iterator *vmi)
1007{
1008 return vmi->mas.last + 1;
1009}
1010static inline int vma_iter_bulk_alloc(struct vma_iterator *vmi,
1011 unsigned long count)
1012{
1013 return mas_expected_entries(mas: &vmi->mas, nr_entries: count);
1014}
1015
1016static inline int vma_iter_clear_gfp(struct vma_iterator *vmi,
1017 unsigned long start, unsigned long end, gfp_t gfp)
1018{
1019 __mas_set_range(mas: &vmi->mas, start, last: end - 1);
1020 mas_store_gfp(mas: &vmi->mas, NULL, gfp);
1021 if (unlikely(mas_is_err(&vmi->mas)))
1022 return -ENOMEM;
1023
1024 return 0;
1025}
1026
1027/* Free any unused preallocations */
1028static inline void vma_iter_free(struct vma_iterator *vmi)
1029{
1030 mas_destroy(mas: &vmi->mas);
1031}
1032
1033static inline int vma_iter_bulk_store(struct vma_iterator *vmi,
1034 struct vm_area_struct *vma)
1035{
1036 vmi->mas.index = vma->vm_start;
1037 vmi->mas.last = vma->vm_end - 1;
1038 mas_store(mas: &vmi->mas, entry: vma);
1039 if (unlikely(mas_is_err(&vmi->mas)))
1040 return -ENOMEM;
1041
1042 return 0;
1043}
1044
1045static inline void vma_iter_invalidate(struct vma_iterator *vmi)
1046{
1047 mas_pause(mas: &vmi->mas);
1048}
1049
1050static inline void vma_iter_set(struct vma_iterator *vmi, unsigned long addr)
1051{
1052 mas_set(mas: &vmi->mas, index: addr);
1053}
1054
1055#define for_each_vma(__vmi, __vma) \
1056 while (((__vma) = vma_next(&(__vmi))) != NULL)
1057
1058/* The MM code likes to work with exclusive end addresses */
1059#define for_each_vma_range(__vmi, __vma, __end) \
1060 while (((__vma) = vma_find(&(__vmi), (__end))) != NULL)
1061
1062#ifdef CONFIG_SHMEM
1063/*
1064 * The vma_is_shmem is not inline because it is used only by slow
1065 * paths in userfault.
1066 */
1067bool vma_is_shmem(struct vm_area_struct *vma);
1068bool vma_is_anon_shmem(struct vm_area_struct *vma);
1069#else
1070static inline bool vma_is_shmem(struct vm_area_struct *vma) { return false; }
1071static inline bool vma_is_anon_shmem(struct vm_area_struct *vma) { return false; }
1072#endif
1073
1074int vma_is_stack_for_current(struct vm_area_struct *vma);
1075
1076/* flush_tlb_range() takes a vma, not a mm, and can care about flags */
1077#define TLB_FLUSH_VMA(mm,flags) { .vm_mm = (mm), .vm_flags = (flags) }
1078
1079struct mmu_gather;
1080struct inode;
1081
1082/*
1083 * compound_order() can be called without holding a reference, which means
1084 * that niceties like page_folio() don't work. These callers should be
1085 * prepared to handle wild return values. For example, PG_head may be
1086 * set before the order is initialised, or this may be a tail page.
1087 * See compaction.c for some good examples.
1088 */
1089static inline unsigned int compound_order(struct page *page)
1090{
1091 struct folio *folio = (struct folio *)page;
1092
1093 if (!test_bit(PG_head, &folio->flags))
1094 return 0;
1095 return folio->_flags_1 & 0xff;
1096}
1097
1098/**
1099 * folio_order - The allocation order of a folio.
1100 * @folio: The folio.
1101 *
1102 * A folio is composed of 2^order pages. See get_order() for the definition
1103 * of order.
1104 *
1105 * Return: The order of the folio.
1106 */
1107static inline unsigned int folio_order(struct folio *folio)
1108{
1109 if (!folio_test_large(folio))
1110 return 0;
1111 return folio->_flags_1 & 0xff;
1112}
1113
1114#include <linux/huge_mm.h>
1115
1116/*
1117 * Methods to modify the page usage count.
1118 *
1119 * What counts for a page usage:
1120 * - cache mapping (page->mapping)
1121 * - private data (page->private)
1122 * - page mapped in a task's page tables, each mapping
1123 * is counted separately
1124 *
1125 * Also, many kernel routines increase the page count before a critical
1126 * routine so they can be sure the page doesn't go away from under them.
1127 */
1128
1129/*
1130 * Drop a ref, return true if the refcount fell to zero (the page has no users)
1131 */
1132static inline int put_page_testzero(struct page *page)
1133{
1134 VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
1135 return page_ref_dec_and_test(page);
1136}
1137
1138static inline int folio_put_testzero(struct folio *folio)
1139{
1140 return put_page_testzero(page: &folio->page);
1141}
1142
1143/*
1144 * Try to grab a ref unless the page has a refcount of zero, return false if
1145 * that is the case.
1146 * This can be called when MMU is off so it must not access
1147 * any of the virtual mappings.
1148 */
1149static inline bool get_page_unless_zero(struct page *page)
1150{
1151 return page_ref_add_unless(page, nr: 1, u: 0);
1152}
1153
1154static inline struct folio *folio_get_nontail_page(struct page *page)
1155{
1156 if (unlikely(!get_page_unless_zero(page)))
1157 return NULL;
1158 return (struct folio *)page;
1159}
1160
1161extern int page_is_ram(unsigned long pfn);
1162
1163enum {
1164 REGION_INTERSECTS,
1165 REGION_DISJOINT,
1166 REGION_MIXED,
1167};
1168
1169int region_intersects(resource_size_t offset, size_t size, unsigned long flags,
1170 unsigned long desc);
1171
1172/* Support for virtually mapped pages */
1173struct page *vmalloc_to_page(const void *addr);
1174unsigned long vmalloc_to_pfn(const void *addr);
1175
1176/*
1177 * Determine if an address is within the vmalloc range
1178 *
1179 * On nommu, vmalloc/vfree wrap through kmalloc/kfree directly, so there
1180 * is no special casing required.
1181 */
1182#ifdef CONFIG_MMU
1183extern bool is_vmalloc_addr(const void *x);
1184extern int is_vmalloc_or_module_addr(const void *x);
1185#else
1186static inline bool is_vmalloc_addr(const void *x)
1187{
1188 return false;
1189}
1190static inline int is_vmalloc_or_module_addr(const void *x)
1191{
1192 return 0;
1193}
1194#endif
1195
1196/*
1197 * How many times the entire folio is mapped as a single unit (eg by a
1198 * PMD or PUD entry). This is probably not what you want, except for
1199 * debugging purposes - it does not include PTE-mapped sub-pages; look
1200 * at folio_mapcount() or page_mapcount() instead.
1201 */
1202static inline int folio_entire_mapcount(struct folio *folio)
1203{
1204 VM_BUG_ON_FOLIO(!folio_test_large(folio), folio);
1205 return atomic_read(v: &folio->_entire_mapcount) + 1;
1206}
1207
1208/*
1209 * The atomic page->_mapcount, starts from -1: so that transitions
1210 * both from it and to it can be tracked, using atomic_inc_and_test
1211 * and atomic_add_negative(-1).
1212 */
1213static inline void page_mapcount_reset(struct page *page)
1214{
1215 atomic_set(v: &(page)->_mapcount, i: -1);
1216}
1217
1218/**
1219 * page_mapcount() - Number of times this precise page is mapped.
1220 * @page: The page.
1221 *
1222 * The number of times this page is mapped. If this page is part of
1223 * a large folio, it includes the number of times this page is mapped
1224 * as part of that folio.
1225 *
1226 * The result is undefined for pages which cannot be mapped into userspace.
1227 * For example SLAB or special types of pages. See function page_has_type().
1228 * They use this field in struct page differently.
1229 */
1230static inline int page_mapcount(struct page *page)
1231{
1232 int mapcount = atomic_read(v: &page->_mapcount) + 1;
1233
1234 if (unlikely(PageCompound(page)))
1235 mapcount += folio_entire_mapcount(page_folio(page));
1236
1237 return mapcount;
1238}
1239
1240int folio_total_mapcount(struct folio *folio);
1241
1242/**
1243 * folio_mapcount() - Calculate the number of mappings of this folio.
1244 * @folio: The folio.
1245 *
1246 * A large folio tracks both how many times the entire folio is mapped,
1247 * and how many times each individual page in the folio is mapped.
1248 * This function calculates the total number of times the folio is
1249 * mapped.
1250 *
1251 * Return: The number of times this folio is mapped.
1252 */
1253static inline int folio_mapcount(struct folio *folio)
1254{
1255 if (likely(!folio_test_large(folio)))
1256 return atomic_read(v: &folio->_mapcount) + 1;
1257 return folio_total_mapcount(folio);
1258}
1259
1260static inline bool folio_large_is_mapped(struct folio *folio)
1261{
1262 /*
1263 * Reading _entire_mapcount below could be omitted if hugetlb
1264 * participated in incrementing nr_pages_mapped when compound mapped.
1265 */
1266 return atomic_read(v: &folio->_nr_pages_mapped) > 0 ||
1267 atomic_read(v: &folio->_entire_mapcount) >= 0;
1268}
1269
1270/**
1271 * folio_mapped - Is this folio mapped into userspace?
1272 * @folio: The folio.
1273 *
1274 * Return: True if any page in this folio is referenced by user page tables.
1275 */
1276static inline bool folio_mapped(struct folio *folio)
1277{
1278 if (likely(!folio_test_large(folio)))
1279 return atomic_read(v: &folio->_mapcount) >= 0;
1280 return folio_large_is_mapped(folio);
1281}
1282
1283/*
1284 * Return true if this page is mapped into pagetables.
1285 * For compound page it returns true if any sub-page of compound page is mapped,
1286 * even if this particular sub-page is not itself mapped by any PTE or PMD.
1287 */
1288static inline bool page_mapped(struct page *page)
1289{
1290 if (likely(!PageCompound(page)))
1291 return atomic_read(v: &page->_mapcount) >= 0;
1292 return folio_large_is_mapped(page_folio(page));
1293}
1294
1295static inline struct page *virt_to_head_page(const void *x)
1296{
1297 struct page *page = virt_to_page(x);
1298
1299 return compound_head(page);
1300}
1301
1302static inline struct folio *virt_to_folio(const void *x)
1303{
1304 struct page *page = virt_to_page(x);
1305
1306 return page_folio(page);
1307}
1308
1309void __folio_put(struct folio *folio);
1310
1311void put_pages_list(struct list_head *pages);
1312
1313void split_page(struct page *page, unsigned int order);
1314void folio_copy(struct folio *dst, struct folio *src);
1315
1316unsigned long nr_free_buffer_pages(void);
1317
1318void destroy_large_folio(struct folio *folio);
1319
1320/* Returns the number of bytes in this potentially compound page. */
1321static inline unsigned long page_size(struct page *page)
1322{
1323 return PAGE_SIZE << compound_order(page);
1324}
1325
1326/* Returns the number of bits needed for the number of bytes in a page */
1327static inline unsigned int page_shift(struct page *page)
1328{
1329 return PAGE_SHIFT + compound_order(page);
1330}
1331
1332/**
1333 * thp_order - Order of a transparent huge page.
1334 * @page: Head page of a transparent huge page.
1335 */
1336static inline unsigned int thp_order(struct page *page)
1337{
1338 VM_BUG_ON_PGFLAGS(PageTail(page), page);
1339 return compound_order(page);
1340}
1341
1342/**
1343 * thp_size - Size of a transparent huge page.
1344 * @page: Head page of a transparent huge page.
1345 *
1346 * Return: Number of bytes in this page.
1347 */
1348static inline unsigned long thp_size(struct page *page)
1349{
1350 return PAGE_SIZE << thp_order(page);
1351}
1352
1353#ifdef CONFIG_MMU
1354/*
1355 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1356 * servicing faults for write access. In the normal case, do always want
1357 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1358 * that do not have writing enabled, when used by access_process_vm.
1359 */
1360static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1361{
1362 if (likely(vma->vm_flags & VM_WRITE))
1363 pte = pte_mkwrite(pte, vma);
1364 return pte;
1365}
1366
1367vm_fault_t do_set_pmd(struct vm_fault *vmf, struct page *page);
1368void set_pte_range(struct vm_fault *vmf, struct folio *folio,
1369 struct page *page, unsigned int nr, unsigned long addr);
1370
1371vm_fault_t finish_fault(struct vm_fault *vmf);
1372#endif
1373
1374/*
1375 * Multiple processes may "see" the same page. E.g. for untouched
1376 * mappings of /dev/null, all processes see the same page full of
1377 * zeroes, and text pages of executables and shared libraries have
1378 * only one copy in memory, at most, normally.
1379 *
1380 * For the non-reserved pages, page_count(page) denotes a reference count.
1381 * page_count() == 0 means the page is free. page->lru is then used for
1382 * freelist management in the buddy allocator.
1383 * page_count() > 0 means the page has been allocated.
1384 *
1385 * Pages are allocated by the slab allocator in order to provide memory
1386 * to kmalloc and kmem_cache_alloc. In this case, the management of the
1387 * page, and the fields in 'struct page' are the responsibility of mm/slab.c
1388 * unless a particular usage is carefully commented. (the responsibility of
1389 * freeing the kmalloc memory is the caller's, of course).
1390 *
1391 * A page may be used by anyone else who does a __get_free_page().
1392 * In this case, page_count still tracks the references, and should only
1393 * be used through the normal accessor functions. The top bits of page->flags
1394 * and page->virtual store page management information, but all other fields
1395 * are unused and could be used privately, carefully. The management of this
1396 * page is the responsibility of the one who allocated it, and those who have
1397 * subsequently been given references to it.
1398 *
1399 * The other pages (we may call them "pagecache pages") are completely
1400 * managed by the Linux memory manager: I/O, buffers, swapping etc.
1401 * The following discussion applies only to them.
1402 *
1403 * A pagecache page contains an opaque `private' member, which belongs to the
1404 * page's address_space. Usually, this is the address of a circular list of
1405 * the page's disk buffers. PG_private must be set to tell the VM to call
1406 * into the filesystem to release these pages.
1407 *
1408 * A page may belong to an inode's memory mapping. In this case, page->mapping
1409 * is the pointer to the inode, and page->index is the file offset of the page,
1410 * in units of PAGE_SIZE.
1411 *
1412 * If pagecache pages are not associated with an inode, they are said to be
1413 * anonymous pages. These may become associated with the swapcache, and in that
1414 * case PG_swapcache is set, and page->private is an offset into the swapcache.
1415 *
1416 * In either case (swapcache or inode backed), the pagecache itself holds one
1417 * reference to the page. Setting PG_private should also increment the
1418 * refcount. The each user mapping also has a reference to the page.
1419 *
1420 * The pagecache pages are stored in a per-mapping radix tree, which is
1421 * rooted at mapping->i_pages, and indexed by offset.
1422 * Where 2.4 and early 2.6 kernels kept dirty/clean pages in per-address_space
1423 * lists, we instead now tag pages as dirty/writeback in the radix tree.
1424 *
1425 * All pagecache pages may be subject to I/O:
1426 * - inode pages may need to be read from disk,
1427 * - inode pages which have been modified and are MAP_SHARED may need
1428 * to be written back to the inode on disk,
1429 * - anonymous pages (including MAP_PRIVATE file mappings) which have been
1430 * modified may need to be swapped out to swap space and (later) to be read
1431 * back into memory.
1432 */
1433
1434#if defined(CONFIG_ZONE_DEVICE) && defined(CONFIG_FS_DAX)
1435DECLARE_STATIC_KEY_FALSE(devmap_managed_key);
1436
1437bool __put_devmap_managed_page_refs(struct page *page, int refs);
1438static inline bool put_devmap_managed_page_refs(struct page *page, int refs)
1439{
1440 if (!static_branch_unlikely(&devmap_managed_key))
1441 return false;
1442 if (!is_zone_device_page(page))
1443 return false;
1444 return __put_devmap_managed_page_refs(page, refs);
1445}
1446#else /* CONFIG_ZONE_DEVICE && CONFIG_FS_DAX */
1447static inline bool put_devmap_managed_page_refs(struct page *page, int refs)
1448{
1449 return false;
1450}
1451#endif /* CONFIG_ZONE_DEVICE && CONFIG_FS_DAX */
1452
1453static inline bool put_devmap_managed_page(struct page *page)
1454{
1455 return put_devmap_managed_page_refs(page, refs: 1);
1456}
1457
1458/* 127: arbitrary random number, small enough to assemble well */
1459#define folio_ref_zero_or_close_to_overflow(folio) \
1460 ((unsigned int) folio_ref_count(folio) + 127u <= 127u)
1461
1462/**
1463 * folio_get - Increment the reference count on a folio.
1464 * @folio: The folio.
1465 *
1466 * Context: May be called in any context, as long as you know that
1467 * you have a refcount on the folio. If you do not already have one,
1468 * folio_try_get() may be the right interface for you to use.
1469 */
1470static inline void folio_get(struct folio *folio)
1471{
1472 VM_BUG_ON_FOLIO(folio_ref_zero_or_close_to_overflow(folio), folio);
1473 folio_ref_inc(folio);
1474}
1475
1476static inline void get_page(struct page *page)
1477{
1478 folio_get(page_folio(page));
1479}
1480
1481static inline __must_check bool try_get_page(struct page *page)
1482{
1483 page = compound_head(page);
1484 if (WARN_ON_ONCE(page_ref_count(page) <= 0))
1485 return false;
1486 page_ref_inc(page);
1487 return true;
1488}
1489
1490/**
1491 * folio_put - Decrement the reference count on a folio.
1492 * @folio: The folio.
1493 *
1494 * If the folio's reference count reaches zero, the memory will be
1495 * released back to the page allocator and may be used by another
1496 * allocation immediately. Do not access the memory or the struct folio
1497 * after calling folio_put() unless you can be sure that it wasn't the
1498 * last reference.
1499 *
1500 * Context: May be called in process or interrupt context, but not in NMI
1501 * context. May be called while holding a spinlock.
1502 */
1503static inline void folio_put(struct folio *folio)
1504{
1505 if (folio_put_testzero(folio))
1506 __folio_put(folio);
1507}
1508
1509/**
1510 * folio_put_refs - Reduce the reference count on a folio.
1511 * @folio: The folio.
1512 * @refs: The amount to subtract from the folio's reference count.
1513 *
1514 * If the folio's reference count reaches zero, the memory will be
1515 * released back to the page allocator and may be used by another
1516 * allocation immediately. Do not access the memory or the struct folio
1517 * after calling folio_put_refs() unless you can be sure that these weren't
1518 * the last references.
1519 *
1520 * Context: May be called in process or interrupt context, but not in NMI
1521 * context. May be called while holding a spinlock.
1522 */
1523static inline void folio_put_refs(struct folio *folio, int refs)
1524{
1525 if (folio_ref_sub_and_test(folio, nr: refs))
1526 __folio_put(folio);
1527}
1528
1529void folios_put_refs(struct folio_batch *folios, unsigned int *refs);
1530
1531/*
1532 * union release_pages_arg - an array of pages or folios
1533 *
1534 * release_pages() releases a simple array of multiple pages, and
1535 * accepts various different forms of said page array: either
1536 * a regular old boring array of pages, an array of folios, or
1537 * an array of encoded page pointers.
1538 *
1539 * The transparent union syntax for this kind of "any of these
1540 * argument types" is all kinds of ugly, so look away.
1541 */
1542typedef union {
1543 struct page **pages;
1544 struct folio **folios;
1545 struct encoded_page **encoded_pages;
1546} release_pages_arg __attribute__ ((__transparent_union__));
1547
1548void release_pages(release_pages_arg, int nr);
1549
1550/**
1551 * folios_put - Decrement the reference count on an array of folios.
1552 * @folios: The folios.
1553 *
1554 * Like folio_put(), but for a batch of folios. This is more efficient
1555 * than writing the loop yourself as it will optimise the locks which need
1556 * to be taken if the folios are freed. The folios batch is returned
1557 * empty and ready to be reused for another batch; there is no need to
1558 * reinitialise it.
1559 *
1560 * Context: May be called in process or interrupt context, but not in NMI
1561 * context. May be called while holding a spinlock.
1562 */
1563static inline void folios_put(struct folio_batch *folios)
1564{
1565 folios_put_refs(folios, NULL);
1566}
1567
1568static inline void put_page(struct page *page)
1569{
1570 struct folio *folio = page_folio(page);
1571
1572 /*
1573 * For some devmap managed pages we need to catch refcount transition
1574 * from 2 to 1:
1575 */
1576 if (put_devmap_managed_page(page: &folio->page))
1577 return;
1578 folio_put(folio);
1579}
1580
1581/*
1582 * GUP_PIN_COUNTING_BIAS, and the associated functions that use it, overload
1583 * the page's refcount so that two separate items are tracked: the original page
1584 * reference count, and also a new count of how many pin_user_pages() calls were
1585 * made against the page. ("gup-pinned" is another term for the latter).
1586 *
1587 * With this scheme, pin_user_pages() becomes special: such pages are marked as
1588 * distinct from normal pages. As such, the unpin_user_page() call (and its
1589 * variants) must be used in order to release gup-pinned pages.
1590 *
1591 * Choice of value:
1592 *
1593 * By making GUP_PIN_COUNTING_BIAS a power of two, debugging of page reference
1594 * counts with respect to pin_user_pages() and unpin_user_page() becomes
1595 * simpler, due to the fact that adding an even power of two to the page
1596 * refcount has the effect of using only the upper N bits, for the code that
1597 * counts up using the bias value. This means that the lower bits are left for
1598 * the exclusive use of the original code that increments and decrements by one
1599 * (or at least, by much smaller values than the bias value).
1600 *
1601 * Of course, once the lower bits overflow into the upper bits (and this is
1602 * OK, because subtraction recovers the original values), then visual inspection
1603 * no longer suffices to directly view the separate counts. However, for normal
1604 * applications that don't have huge page reference counts, this won't be an
1605 * issue.
1606 *
1607 * Locking: the lockless algorithm described in folio_try_get_rcu()
1608 * provides safe operation for get_user_pages(), page_mkclean() and
1609 * other calls that race to set up page table entries.
1610 */
1611#define GUP_PIN_COUNTING_BIAS (1U << 10)
1612
1613void unpin_user_page(struct page *page);
1614void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages,
1615 bool make_dirty);
1616void unpin_user_page_range_dirty_lock(struct page *page, unsigned long npages,
1617 bool make_dirty);
1618void unpin_user_pages(struct page **pages, unsigned long npages);
1619
1620static inline bool is_cow_mapping(vm_flags_t flags)
1621{
1622 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
1623}
1624
1625#ifndef CONFIG_MMU
1626static inline bool is_nommu_shared_mapping(vm_flags_t flags)
1627{
1628 /*
1629 * NOMMU shared mappings are ordinary MAP_SHARED mappings and selected
1630 * R/O MAP_PRIVATE file mappings that are an effective R/O overlay of
1631 * a file mapping. R/O MAP_PRIVATE mappings might still modify
1632 * underlying memory if ptrace is active, so this is only possible if
1633 * ptrace does not apply. Note that there is no mprotect() to upgrade
1634 * write permissions later.
1635 */
1636 return flags & (VM_MAYSHARE | VM_MAYOVERLAY);
1637}
1638#endif
1639
1640#if defined(CONFIG_SPARSEMEM) && !defined(CONFIG_SPARSEMEM_VMEMMAP)
1641#define SECTION_IN_PAGE_FLAGS
1642#endif
1643
1644/*
1645 * The identification function is mainly used by the buddy allocator for
1646 * determining if two pages could be buddies. We are not really identifying
1647 * the zone since we could be using the section number id if we do not have
1648 * node id available in page flags.
1649 * We only guarantee that it will return the same value for two combinable
1650 * pages in a zone.
1651 */
1652static inline int page_zone_id(struct page *page)
1653{
1654 return (page->flags >> ZONEID_PGSHIFT) & ZONEID_MASK;
1655}
1656
1657#ifdef NODE_NOT_IN_PAGE_FLAGS
1658int page_to_nid(const struct page *page);
1659#else
1660static inline int page_to_nid(const struct page *page)
1661{
1662 return (PF_POISONED_CHECK(page)->flags >> NODES_PGSHIFT) & NODES_MASK;
1663}
1664#endif
1665
1666static inline int folio_nid(const struct folio *folio)
1667{
1668 return page_to_nid(page: &folio->page);
1669}
1670
1671#ifdef CONFIG_NUMA_BALANCING
1672/* page access time bits needs to hold at least 4 seconds */
1673#define PAGE_ACCESS_TIME_MIN_BITS 12
1674#if LAST_CPUPID_SHIFT < PAGE_ACCESS_TIME_MIN_BITS
1675#define PAGE_ACCESS_TIME_BUCKETS \
1676 (PAGE_ACCESS_TIME_MIN_BITS - LAST_CPUPID_SHIFT)
1677#else
1678#define PAGE_ACCESS_TIME_BUCKETS 0
1679#endif
1680
1681#define PAGE_ACCESS_TIME_MASK \
1682 (LAST_CPUPID_MASK << PAGE_ACCESS_TIME_BUCKETS)
1683
1684static inline int cpu_pid_to_cpupid(int cpu, int pid)
1685{
1686 return ((cpu & LAST__CPU_MASK) << LAST__PID_SHIFT) | (pid & LAST__PID_MASK);
1687}
1688
1689static inline int cpupid_to_pid(int cpupid)
1690{
1691 return cpupid & LAST__PID_MASK;
1692}
1693
1694static inline int cpupid_to_cpu(int cpupid)
1695{
1696 return (cpupid >> LAST__PID_SHIFT) & LAST__CPU_MASK;
1697}
1698
1699static inline int cpupid_to_nid(int cpupid)
1700{
1701 return cpu_to_node(cpu: cpupid_to_cpu(cpupid));
1702}
1703
1704static inline bool cpupid_pid_unset(int cpupid)
1705{
1706 return cpupid_to_pid(cpupid) == (-1 & LAST__PID_MASK);
1707}
1708
1709static inline bool cpupid_cpu_unset(int cpupid)
1710{
1711 return cpupid_to_cpu(cpupid) == (-1 & LAST__CPU_MASK);
1712}
1713
1714static inline bool __cpupid_match_pid(pid_t task_pid, int cpupid)
1715{
1716 return (task_pid & LAST__PID_MASK) == cpupid_to_pid(cpupid);
1717}
1718
1719#define cpupid_match_pid(task, cpupid) __cpupid_match_pid(task->pid, cpupid)
1720#ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
1721static inline int folio_xchg_last_cpupid(struct folio *folio, int cpupid)
1722{
1723 return xchg(&folio->_last_cpupid, cpupid & LAST_CPUPID_MASK);
1724}
1725
1726static inline int folio_last_cpupid(struct folio *folio)
1727{
1728 return folio->_last_cpupid;
1729}
1730static inline void page_cpupid_reset_last(struct page *page)
1731{
1732 page->_last_cpupid = -1 & LAST_CPUPID_MASK;
1733}
1734#else
1735static inline int folio_last_cpupid(struct folio *folio)
1736{
1737 return (folio->flags >> LAST_CPUPID_PGSHIFT) & LAST_CPUPID_MASK;
1738}
1739
1740int folio_xchg_last_cpupid(struct folio *folio, int cpupid);
1741
1742static inline void page_cpupid_reset_last(struct page *page)
1743{
1744 page->flags |= LAST_CPUPID_MASK << LAST_CPUPID_PGSHIFT;
1745}
1746#endif /* LAST_CPUPID_NOT_IN_PAGE_FLAGS */
1747
1748static inline int folio_xchg_access_time(struct folio *folio, int time)
1749{
1750 int last_time;
1751
1752 last_time = folio_xchg_last_cpupid(folio,
1753 time >> PAGE_ACCESS_TIME_BUCKETS);
1754 return last_time << PAGE_ACCESS_TIME_BUCKETS;
1755}
1756
1757static inline void vma_set_access_pid_bit(struct vm_area_struct *vma)
1758{
1759 unsigned int pid_bit;
1760
1761 pid_bit = hash_32(current->pid, ilog2(BITS_PER_LONG));
1762 if (vma->numab_state && !test_bit(pid_bit, &vma->numab_state->pids_active[1])) {
1763 __set_bit(pid_bit, &vma->numab_state->pids_active[1]);
1764 }
1765}
1766#else /* !CONFIG_NUMA_BALANCING */
1767static inline int folio_xchg_last_cpupid(struct folio *folio, int cpupid)
1768{
1769 return folio_nid(folio); /* XXX */
1770}
1771
1772static inline int folio_xchg_access_time(struct folio *folio, int time)
1773{
1774 return 0;
1775}
1776
1777static inline int folio_last_cpupid(struct folio *folio)
1778{
1779 return folio_nid(folio); /* XXX */
1780}
1781
1782static inline int cpupid_to_nid(int cpupid)
1783{
1784 return -1;
1785}
1786
1787static inline int cpupid_to_pid(int cpupid)
1788{
1789 return -1;
1790}
1791
1792static inline int cpupid_to_cpu(int cpupid)
1793{
1794 return -1;
1795}
1796
1797static inline int cpu_pid_to_cpupid(int nid, int pid)
1798{
1799 return -1;
1800}
1801
1802static inline bool cpupid_pid_unset(int cpupid)
1803{
1804 return true;
1805}
1806
1807static inline void page_cpupid_reset_last(struct page *page)
1808{
1809}
1810
1811static inline bool cpupid_match_pid(struct task_struct *task, int cpupid)
1812{
1813 return false;
1814}
1815
1816static inline void vma_set_access_pid_bit(struct vm_area_struct *vma)
1817{
1818}
1819#endif /* CONFIG_NUMA_BALANCING */
1820
1821#if defined(CONFIG_KASAN_SW_TAGS) || defined(CONFIG_KASAN_HW_TAGS)
1822
1823/*
1824 * KASAN per-page tags are stored xor'ed with 0xff. This allows to avoid
1825 * setting tags for all pages to native kernel tag value 0xff, as the default
1826 * value 0x00 maps to 0xff.
1827 */
1828
1829static inline u8 page_kasan_tag(const struct page *page)
1830{
1831 u8 tag = KASAN_TAG_KERNEL;
1832
1833 if (kasan_enabled()) {
1834 tag = (page->flags >> KASAN_TAG_PGSHIFT) & KASAN_TAG_MASK;
1835 tag ^= 0xff;
1836 }
1837
1838 return tag;
1839}
1840
1841static inline void page_kasan_tag_set(struct page *page, u8 tag)
1842{
1843 unsigned long old_flags, flags;
1844
1845 if (!kasan_enabled())
1846 return;
1847
1848 tag ^= 0xff;
1849 old_flags = READ_ONCE(page->flags);
1850 do {
1851 flags = old_flags;
1852 flags &= ~(KASAN_TAG_MASK << KASAN_TAG_PGSHIFT);
1853 flags |= (tag & KASAN_TAG_MASK) << KASAN_TAG_PGSHIFT;
1854 } while (unlikely(!try_cmpxchg(&page->flags, &old_flags, flags)));
1855}
1856
1857static inline void page_kasan_tag_reset(struct page *page)
1858{
1859 if (kasan_enabled())
1860 page_kasan_tag_set(page, KASAN_TAG_KERNEL);
1861}
1862
1863#else /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */
1864
1865static inline u8 page_kasan_tag(const struct page *page)
1866{
1867 return 0xff;
1868}
1869
1870static inline void page_kasan_tag_set(struct page *page, u8 tag) { }
1871static inline void page_kasan_tag_reset(struct page *page) { }
1872
1873#endif /* CONFIG_KASAN_SW_TAGS || CONFIG_KASAN_HW_TAGS */
1874
1875static inline struct zone *page_zone(const struct page *page)
1876{
1877 return &NODE_DATA(page_to_nid(page))->node_zones[page_zonenum(page)];
1878}
1879
1880static inline pg_data_t *page_pgdat(const struct page *page)
1881{
1882 return NODE_DATA(page_to_nid(page));
1883}
1884
1885static inline struct zone *folio_zone(const struct folio *folio)
1886{
1887 return page_zone(page: &folio->page);
1888}
1889
1890static inline pg_data_t *folio_pgdat(const struct folio *folio)
1891{
1892 return page_pgdat(page: &folio->page);
1893}
1894
1895#ifdef SECTION_IN_PAGE_FLAGS
1896static inline void set_page_section(struct page *page, unsigned long section)
1897{
1898 page->flags &= ~(SECTIONS_MASK << SECTIONS_PGSHIFT);
1899 page->flags |= (section & SECTIONS_MASK) << SECTIONS_PGSHIFT;
1900}
1901
1902static inline unsigned long page_to_section(const struct page *page)
1903{
1904 return (page->flags >> SECTIONS_PGSHIFT) & SECTIONS_MASK;
1905}
1906#endif
1907
1908/**
1909 * folio_pfn - Return the Page Frame Number of a folio.
1910 * @folio: The folio.
1911 *
1912 * A folio may contain multiple pages. The pages have consecutive
1913 * Page Frame Numbers.
1914 *
1915 * Return: The Page Frame Number of the first page in the folio.
1916 */
1917static inline unsigned long folio_pfn(struct folio *folio)
1918{
1919 return page_to_pfn(&folio->page);
1920}
1921
1922static inline struct folio *pfn_folio(unsigned long pfn)
1923{
1924 return page_folio(pfn_to_page(pfn));
1925}
1926
1927/**
1928 * folio_maybe_dma_pinned - Report if a folio may be pinned for DMA.
1929 * @folio: The folio.
1930 *
1931 * This function checks if a folio has been pinned via a call to
1932 * a function in the pin_user_pages() family.
1933 *
1934 * For small folios, the return value is partially fuzzy: false is not fuzzy,
1935 * because it means "definitely not pinned for DMA", but true means "probably
1936 * pinned for DMA, but possibly a false positive due to having at least
1937 * GUP_PIN_COUNTING_BIAS worth of normal folio references".
1938 *
1939 * False positives are OK, because: a) it's unlikely for a folio to
1940 * get that many refcounts, and b) all the callers of this routine are
1941 * expected to be able to deal gracefully with a false positive.
1942 *
1943 * For large folios, the result will be exactly correct. That's because
1944 * we have more tracking data available: the _pincount field is used
1945 * instead of the GUP_PIN_COUNTING_BIAS scheme.
1946 *
1947 * For more information, please see Documentation/core-api/pin_user_pages.rst.
1948 *
1949 * Return: True, if it is likely that the page has been "dma-pinned".
1950 * False, if the page is definitely not dma-pinned.
1951 */
1952static inline bool folio_maybe_dma_pinned(struct folio *folio)
1953{
1954 if (folio_test_large(folio))
1955 return atomic_read(v: &folio->_pincount) > 0;
1956
1957 /*
1958 * folio_ref_count() is signed. If that refcount overflows, then
1959 * folio_ref_count() returns a negative value, and callers will avoid
1960 * further incrementing the refcount.
1961 *
1962 * Here, for that overflow case, use the sign bit to count a little
1963 * bit higher via unsigned math, and thus still get an accurate result.
1964 */
1965 return ((unsigned int)folio_ref_count(folio)) >=
1966 GUP_PIN_COUNTING_BIAS;
1967}
1968
1969static inline bool page_maybe_dma_pinned(struct page *page)
1970{
1971 return folio_maybe_dma_pinned(page_folio(page));
1972}
1973
1974/*
1975 * This should most likely only be called during fork() to see whether we
1976 * should break the cow immediately for an anon page on the src mm.
1977 *
1978 * The caller has to hold the PT lock and the vma->vm_mm->->write_protect_seq.
1979 */
1980static inline bool folio_needs_cow_for_dma(struct vm_area_struct *vma,
1981 struct folio *folio)
1982{
1983 VM_BUG_ON(!(raw_read_seqcount(&vma->vm_mm->write_protect_seq) & 1));
1984
1985 if (!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags))
1986 return false;
1987
1988 return folio_maybe_dma_pinned(folio);
1989}
1990
1991/**
1992 * is_zero_page - Query if a page is a zero page
1993 * @page: The page to query
1994 *
1995 * This returns true if @page is one of the permanent zero pages.
1996 */
1997static inline bool is_zero_page(const struct page *page)
1998{
1999 return is_zero_pfn(page_to_pfn(page));
2000}
2001
2002/**
2003 * is_zero_folio - Query if a folio is a zero page
2004 * @folio: The folio to query
2005 *
2006 * This returns true if @folio is one of the permanent zero pages.
2007 */
2008static inline bool is_zero_folio(const struct folio *folio)
2009{
2010 return is_zero_page(page: &folio->page);
2011}
2012
2013/* MIGRATE_CMA and ZONE_MOVABLE do not allow pin folios */
2014#ifdef CONFIG_MIGRATION
2015static inline bool folio_is_longterm_pinnable(struct folio *folio)
2016{
2017#ifdef CONFIG_CMA
2018 int mt = folio_migratetype(folio);
2019
2020 if (mt == MIGRATE_CMA || mt == MIGRATE_ISOLATE)
2021 return false;
2022#endif
2023 /* The zero page can be "pinned" but gets special handling. */
2024 if (is_zero_folio(folio))
2025 return true;
2026
2027 /* Coherent device memory must always allow eviction. */
2028 if (folio_is_device_coherent(folio))
2029 return false;
2030
2031 /* Otherwise, non-movable zone folios can be pinned. */
2032 return !folio_is_zone_movable(folio);
2033
2034}
2035#else
2036static inline bool folio_is_longterm_pinnable(struct folio *folio)
2037{
2038 return true;
2039}
2040#endif
2041
2042static inline void set_page_zone(struct page *page, enum zone_type zone)
2043{
2044 page->flags &= ~(ZONES_MASK << ZONES_PGSHIFT);
2045 page->flags |= (zone & ZONES_MASK) << ZONES_PGSHIFT;
2046}
2047
2048static inline void set_page_node(struct page *page, unsigned long node)
2049{
2050 page->flags &= ~(NODES_MASK << NODES_PGSHIFT);
2051 page->flags |= (node & NODES_MASK) << NODES_PGSHIFT;
2052}
2053
2054static inline void set_page_links(struct page *page, enum zone_type zone,
2055 unsigned long node, unsigned long pfn)
2056{
2057 set_page_zone(page, zone);
2058 set_page_node(page, node);
2059#ifdef SECTION_IN_PAGE_FLAGS
2060 set_page_section(page, pfn_to_section_nr(pfn));
2061#endif
2062}
2063
2064/**
2065 * folio_nr_pages - The number of pages in the folio.
2066 * @folio: The folio.
2067 *
2068 * Return: A positive power of two.
2069 */
2070static inline long folio_nr_pages(struct folio *folio)
2071{
2072 if (!folio_test_large(folio))
2073 return 1;
2074#ifdef CONFIG_64BIT
2075 return folio->_folio_nr_pages;
2076#else
2077 return 1L << (folio->_flags_1 & 0xff);
2078#endif
2079}
2080
2081/* Only hugetlbfs can allocate folios larger than MAX_ORDER */
2082#ifdef CONFIG_ARCH_HAS_GIGANTIC_PAGE
2083#define MAX_FOLIO_NR_PAGES (1UL << PUD_ORDER)
2084#else
2085#define MAX_FOLIO_NR_PAGES MAX_ORDER_NR_PAGES
2086#endif
2087
2088/*
2089 * compound_nr() returns the number of pages in this potentially compound
2090 * page. compound_nr() can be called on a tail page, and is defined to
2091 * return 1 in that case.
2092 */
2093static inline unsigned long compound_nr(struct page *page)
2094{
2095 struct folio *folio = (struct folio *)page;
2096
2097 if (!test_bit(PG_head, &folio->flags))
2098 return 1;
2099#ifdef CONFIG_64BIT
2100 return folio->_folio_nr_pages;
2101#else
2102 return 1L << (folio->_flags_1 & 0xff);
2103#endif
2104}
2105
2106/**
2107 * thp_nr_pages - The number of regular pages in this huge page.
2108 * @page: The head page of a huge page.
2109 */
2110static inline int thp_nr_pages(struct page *page)
2111{
2112 return folio_nr_pages(folio: (struct folio *)page);
2113}
2114
2115/**
2116 * folio_next - Move to the next physical folio.
2117 * @folio: The folio we're currently operating on.
2118 *
2119 * If you have physically contiguous memory which may span more than
2120 * one folio (eg a &struct bio_vec), use this function to move from one
2121 * folio to the next. Do not use it if the memory is only virtually
2122 * contiguous as the folios are almost certainly not adjacent to each
2123 * other. This is the folio equivalent to writing ``page++``.
2124 *
2125 * Context: We assume that the folios are refcounted and/or locked at a
2126 * higher level and do not adjust the reference counts.
2127 * Return: The next struct folio.
2128 */
2129static inline struct folio *folio_next(struct folio *folio)
2130{
2131 return (struct folio *)folio_page(folio, folio_nr_pages(folio));
2132}
2133
2134/**
2135 * folio_shift - The size of the memory described by this folio.
2136 * @folio: The folio.
2137 *
2138 * A folio represents a number of bytes which is a power-of-two in size.
2139 * This function tells you which power-of-two the folio is. See also
2140 * folio_size() and folio_order().
2141 *
2142 * Context: The caller should have a reference on the folio to prevent
2143 * it from being split. It is not necessary for the folio to be locked.
2144 * Return: The base-2 logarithm of the size of this folio.
2145 */
2146static inline unsigned int folio_shift(struct folio *folio)
2147{
2148 return PAGE_SHIFT + folio_order(folio);
2149}
2150
2151/**
2152 * folio_size - The number of bytes in a folio.
2153 * @folio: The folio.
2154 *
2155 * Context: The caller should have a reference on the folio to prevent
2156 * it from being split. It is not necessary for the folio to be locked.
2157 * Return: The number of bytes in this folio.
2158 */
2159static inline size_t folio_size(struct folio *folio)
2160{
2161 return PAGE_SIZE << folio_order(folio);
2162}
2163
2164/**
2165 * folio_estimated_sharers - Estimate the number of sharers of a folio.
2166 * @folio: The folio.
2167 *
2168 * folio_estimated_sharers() aims to serve as a function to efficiently
2169 * estimate the number of processes sharing a folio. This is done by
2170 * looking at the precise mapcount of the first subpage in the folio, and
2171 * assuming the other subpages are the same. This may not be true for large
2172 * folios. If you want exact mapcounts for exact calculations, look at
2173 * page_mapcount() or folio_total_mapcount().
2174 *
2175 * Return: The estimated number of processes sharing a folio.
2176 */
2177static inline int folio_estimated_sharers(struct folio *folio)
2178{
2179 return page_mapcount(folio_page(folio, 0));
2180}
2181
2182#ifndef HAVE_ARCH_MAKE_PAGE_ACCESSIBLE
2183static inline int arch_make_page_accessible(struct page *page)
2184{
2185 return 0;
2186}
2187#endif
2188
2189#ifndef HAVE_ARCH_MAKE_FOLIO_ACCESSIBLE
2190static inline int arch_make_folio_accessible(struct folio *folio)
2191{
2192 int ret;
2193 long i, nr = folio_nr_pages(folio);
2194
2195 for (i = 0; i < nr; i++) {
2196 ret = arch_make_page_accessible(folio_page(folio, i));
2197 if (ret)
2198 break;
2199 }
2200
2201 return ret;
2202}
2203#endif
2204
2205/*
2206 * Some inline functions in vmstat.h depend on page_zone()
2207 */
2208#include <linux/vmstat.h>
2209
2210#if defined(CONFIG_HIGHMEM) && !defined(WANT_PAGE_VIRTUAL)
2211#define HASHED_PAGE_VIRTUAL
2212#endif
2213
2214#if defined(WANT_PAGE_VIRTUAL)
2215static inline void *page_address(const struct page *page)
2216{
2217 return page->virtual;
2218}
2219static inline void set_page_address(struct page *page, void *address)
2220{
2221 page->virtual = address;
2222}
2223#define page_address_init() do { } while(0)
2224#endif
2225
2226#if defined(HASHED_PAGE_VIRTUAL)
2227void *page_address(const struct page *page);
2228void set_page_address(struct page *page, void *virtual);
2229void page_address_init(void);
2230#endif
2231
2232static __always_inline void *lowmem_page_address(const struct page *page)
2233{
2234 return page_to_virt(page);
2235}
2236
2237#if !defined(HASHED_PAGE_VIRTUAL) && !defined(WANT_PAGE_VIRTUAL)
2238#define page_address(page) lowmem_page_address(page)
2239#define set_page_address(page, address) do { } while(0)
2240#define page_address_init() do { } while(0)
2241#endif
2242
2243static inline void *folio_address(const struct folio *folio)
2244{
2245 return page_address(&folio->page);
2246}
2247
2248extern pgoff_t __page_file_index(struct page *page);
2249
2250/*
2251 * Return the pagecache index of the passed page. Regular pagecache pages
2252 * use ->index whereas swapcache pages use swp_offset(->private)
2253 */
2254static inline pgoff_t page_index(struct page *page)
2255{
2256 if (unlikely(PageSwapCache(page)))
2257 return __page_file_index(page);
2258 return page->index;
2259}
2260
2261/*
2262 * Return true only if the page has been allocated with
2263 * ALLOC_NO_WATERMARKS and the low watermark was not
2264 * met implying that the system is under some pressure.
2265 */
2266static inline bool page_is_pfmemalloc(const struct page *page)
2267{
2268 /*
2269 * lru.next has bit 1 set if the page is allocated from the
2270 * pfmemalloc reserves. Callers may simply overwrite it if
2271 * they do not need to preserve that information.
2272 */
2273 return (uintptr_t)page->lru.next & BIT(1);
2274}
2275
2276/*
2277 * Return true only if the folio has been allocated with
2278 * ALLOC_NO_WATERMARKS and the low watermark was not
2279 * met implying that the system is under some pressure.
2280 */
2281static inline bool folio_is_pfmemalloc(const struct folio *folio)
2282{
2283 /*
2284 * lru.next has bit 1 set if the page is allocated from the
2285 * pfmemalloc reserves. Callers may simply overwrite it if
2286 * they do not need to preserve that information.
2287 */
2288 return (uintptr_t)folio->lru.next & BIT(1);
2289}
2290
2291/*
2292 * Only to be called by the page allocator on a freshly allocated
2293 * page.
2294 */
2295static inline void set_page_pfmemalloc(struct page *page)
2296{
2297 page->lru.next = (void *)BIT(1);
2298}
2299
2300static inline void clear_page_pfmemalloc(struct page *page)
2301{
2302 page->lru.next = NULL;
2303}
2304
2305/*
2306 * Can be called by the pagefault handler when it gets a VM_FAULT_OOM.
2307 */
2308extern void pagefault_out_of_memory(void);
2309
2310#define offset_in_page(p) ((unsigned long)(p) & ~PAGE_MASK)
2311#define offset_in_thp(page, p) ((unsigned long)(p) & (thp_size(page) - 1))
2312#define offset_in_folio(folio, p) ((unsigned long)(p) & (folio_size(folio) - 1))
2313
2314/*
2315 * Parameter block passed down to zap_pte_range in exceptional cases.
2316 */
2317struct zap_details {
2318 struct folio *single_folio; /* Locked folio to be unmapped */
2319 bool even_cows; /* Zap COWed private pages too? */
2320 zap_flags_t zap_flags; /* Extra flags for zapping */
2321};
2322
2323/*
2324 * Whether to drop the pte markers, for example, the uffd-wp information for
2325 * file-backed memory. This should only be specified when we will completely
2326 * drop the page in the mm, either by truncation or unmapping of the vma. By
2327 * default, the flag is not set.
2328 */
2329#define ZAP_FLAG_DROP_MARKER ((__force zap_flags_t) BIT(0))
2330/* Set in unmap_vmas() to indicate a final unmap call. Only used by hugetlb */
2331#define ZAP_FLAG_UNMAP ((__force zap_flags_t) BIT(1))
2332
2333#ifdef CONFIG_SCHED_MM_CID
2334void sched_mm_cid_before_execve(struct task_struct *t);
2335void sched_mm_cid_after_execve(struct task_struct *t);
2336void sched_mm_cid_fork(struct task_struct *t);
2337void sched_mm_cid_exit_signals(struct task_struct *t);
2338static inline int task_mm_cid(struct task_struct *t)
2339{
2340 return t->mm_cid;
2341}
2342#else
2343static inline void sched_mm_cid_before_execve(struct task_struct *t) { }
2344static inline void sched_mm_cid_after_execve(struct task_struct *t) { }
2345static inline void sched_mm_cid_fork(struct task_struct *t) { }
2346static inline void sched_mm_cid_exit_signals(struct task_struct *t) { }
2347static inline int task_mm_cid(struct task_struct *t)
2348{
2349 /*
2350 * Use the processor id as a fall-back when the mm cid feature is
2351 * disabled. This provides functional per-cpu data structure accesses
2352 * in user-space, althrough it won't provide the memory usage benefits.
2353 */
2354 return raw_smp_processor_id();
2355}
2356#endif
2357
2358#ifdef CONFIG_MMU
2359extern bool can_do_mlock(void);
2360#else
2361static inline bool can_do_mlock(void) { return false; }
2362#endif
2363extern int user_shm_lock(size_t, struct ucounts *);
2364extern void user_shm_unlock(size_t, struct ucounts *);
2365
2366struct folio *vm_normal_folio(struct vm_area_struct *vma, unsigned long addr,
2367 pte_t pte);
2368struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
2369 pte_t pte);
2370struct folio *vm_normal_folio_pmd(struct vm_area_struct *vma,
2371 unsigned long addr, pmd_t pmd);
2372struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr,
2373 pmd_t pmd);
2374
2375void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
2376 unsigned long size);
2377void zap_page_range_single(struct vm_area_struct *vma, unsigned long address,
2378 unsigned long size, struct zap_details *details);
2379static inline void zap_vma_pages(struct vm_area_struct *vma)
2380{
2381 zap_page_range_single(vma, address: vma->vm_start,
2382 size: vma->vm_end - vma->vm_start, NULL);
2383}
2384void unmap_vmas(struct mmu_gather *tlb, struct ma_state *mas,
2385 struct vm_area_struct *start_vma, unsigned long start,
2386 unsigned long end, unsigned long tree_end, bool mm_wr_locked);
2387
2388struct mmu_notifier_range;
2389
2390void free_pgd_range(struct mmu_gather *tlb, unsigned long addr,
2391 unsigned long end, unsigned long floor, unsigned long ceiling);
2392int
2393copy_page_range(struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma);
2394int follow_pte(struct mm_struct *mm, unsigned long address,
2395 pte_t **ptepp, spinlock_t **ptlp);
2396int follow_pfn(struct vm_area_struct *vma, unsigned long address,
2397 unsigned long *pfn);
2398int follow_phys(struct vm_area_struct *vma, unsigned long address,
2399 unsigned int flags, unsigned long *prot, resource_size_t *phys);
2400int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
2401 void *buf, int len, int write);
2402
2403extern void truncate_pagecache(struct inode *inode, loff_t new);
2404extern void truncate_setsize(struct inode *inode, loff_t newsize);
2405void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to);
2406void truncate_pagecache_range(struct inode *inode, loff_t offset, loff_t end);
2407int generic_error_remove_folio(struct address_space *mapping,
2408 struct folio *folio);
2409
2410struct vm_area_struct *lock_mm_and_find_vma(struct mm_struct *mm,
2411 unsigned long address, struct pt_regs *regs);
2412
2413#ifdef CONFIG_MMU
2414extern vm_fault_t handle_mm_fault(struct vm_area_struct *vma,
2415 unsigned long address, unsigned int flags,
2416 struct pt_regs *regs);
2417extern int fixup_user_fault(struct mm_struct *mm,
2418 unsigned long address, unsigned int fault_flags,
2419 bool *unlocked);
2420void unmap_mapping_pages(struct address_space *mapping,
2421 pgoff_t start, pgoff_t nr, bool even_cows);
2422void unmap_mapping_range(struct address_space *mapping,
2423 loff_t const holebegin, loff_t const holelen, int even_cows);
2424#else
2425static inline vm_fault_t handle_mm_fault(struct vm_area_struct *vma,
2426 unsigned long address, unsigned int flags,
2427 struct pt_regs *regs)
2428{
2429 /* should never happen if there's no MMU */
2430 BUG();
2431 return VM_FAULT_SIGBUS;
2432}
2433static inline int fixup_user_fault(struct mm_struct *mm, unsigned long address,
2434 unsigned int fault_flags, bool *unlocked)
2435{
2436 /* should never happen if there's no MMU */
2437 BUG();
2438 return -EFAULT;
2439}
2440static inline void unmap_mapping_pages(struct address_space *mapping,
2441 pgoff_t start, pgoff_t nr, bool even_cows) { }
2442static inline void unmap_mapping_range(struct address_space *mapping,
2443 loff_t const holebegin, loff_t const holelen, int even_cows) { }
2444#endif
2445
2446static inline void unmap_shared_mapping_range(struct address_space *mapping,
2447 loff_t const holebegin, loff_t const holelen)
2448{
2449 unmap_mapping_range(mapping, holebegin, holelen, even_cows: 0);
2450}
2451
2452static inline struct vm_area_struct *vma_lookup(struct mm_struct *mm,
2453 unsigned long addr);
2454
2455extern int access_process_vm(struct task_struct *tsk, unsigned long addr,
2456 void *buf, int len, unsigned int gup_flags);
2457extern int access_remote_vm(struct mm_struct *mm, unsigned long addr,
2458 void *buf, int len, unsigned int gup_flags);
2459
2460long get_user_pages_remote(struct mm_struct *mm,
2461 unsigned long start, unsigned long nr_pages,
2462 unsigned int gup_flags, struct page **pages,
2463 int *locked);
2464long pin_user_pages_remote(struct mm_struct *mm,
2465 unsigned long start, unsigned long nr_pages,
2466 unsigned int gup_flags, struct page **pages,
2467 int *locked);
2468
2469/*
2470 * Retrieves a single page alongside its VMA. Does not support FOLL_NOWAIT.
2471 */
2472static inline struct page *get_user_page_vma_remote(struct mm_struct *mm,
2473 unsigned long addr,
2474 int gup_flags,
2475 struct vm_area_struct **vmap)
2476{
2477 struct page *page;
2478 struct vm_area_struct *vma;
2479 int got;
2480
2481 if (WARN_ON_ONCE(unlikely(gup_flags & FOLL_NOWAIT)))
2482 return ERR_PTR(error: -EINVAL);
2483
2484 got = get_user_pages_remote(mm, start: addr, nr_pages: 1, gup_flags, pages: &page, NULL);
2485
2486 if (got < 0)
2487 return ERR_PTR(error: got);
2488
2489 vma = vma_lookup(mm, addr);
2490 if (WARN_ON_ONCE(!vma)) {
2491 put_page(page);
2492 return ERR_PTR(error: -EINVAL);
2493 }
2494
2495 *vmap = vma;
2496 return page;
2497}
2498
2499long get_user_pages(unsigned long start, unsigned long nr_pages,
2500 unsigned int gup_flags, struct page **pages);
2501long pin_user_pages(unsigned long start, unsigned long nr_pages,
2502 unsigned int gup_flags, struct page **pages);
2503long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
2504 struct page **pages, unsigned int gup_flags);
2505long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
2506 struct page **pages, unsigned int gup_flags);
2507
2508int get_user_pages_fast(unsigned long start, int nr_pages,
2509 unsigned int gup_flags, struct page **pages);
2510int pin_user_pages_fast(unsigned long start, int nr_pages,
2511 unsigned int gup_flags, struct page **pages);
2512void folio_add_pin(struct folio *folio);
2513
2514int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc);
2515int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc,
2516 struct task_struct *task, bool bypass_rlim);
2517
2518struct kvec;
2519struct page *get_dump_page(unsigned long addr);
2520
2521bool folio_mark_dirty(struct folio *folio);
2522bool set_page_dirty(struct page *page);
2523int set_page_dirty_lock(struct page *page);
2524
2525int get_cmdline(struct task_struct *task, char *buffer, int buflen);
2526
2527extern unsigned long move_page_tables(struct vm_area_struct *vma,
2528 unsigned long old_addr, struct vm_area_struct *new_vma,
2529 unsigned long new_addr, unsigned long len,
2530 bool need_rmap_locks, bool for_stack);
2531
2532/*
2533 * Flags used by change_protection(). For now we make it a bitmap so
2534 * that we can pass in multiple flags just like parameters. However
2535 * for now all the callers are only use one of the flags at the same
2536 * time.
2537 */
2538/*
2539 * Whether we should manually check if we can map individual PTEs writable,
2540 * because something (e.g., COW, uffd-wp) blocks that from happening for all
2541 * PTEs automatically in a writable mapping.
2542 */
2543#define MM_CP_TRY_CHANGE_WRITABLE (1UL << 0)
2544/* Whether this protection change is for NUMA hints */
2545#define MM_CP_PROT_NUMA (1UL << 1)
2546/* Whether this change is for write protecting */
2547#define MM_CP_UFFD_WP (1UL << 2) /* do wp */
2548#define MM_CP_UFFD_WP_RESOLVE (1UL << 3) /* Resolve wp */
2549#define MM_CP_UFFD_WP_ALL (MM_CP_UFFD_WP | \
2550 MM_CP_UFFD_WP_RESOLVE)
2551
2552bool vma_needs_dirty_tracking(struct vm_area_struct *vma);
2553int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot);
2554static inline bool vma_wants_manual_pte_write_upgrade(struct vm_area_struct *vma)
2555{
2556 /*
2557 * We want to check manually if we can change individual PTEs writable
2558 * if we can't do that automatically for all PTEs in a mapping. For
2559 * private mappings, that's always the case when we have write
2560 * permissions as we properly have to handle COW.
2561 */
2562 if (vma->vm_flags & VM_SHARED)
2563 return vma_wants_writenotify(vma, vm_page_prot: vma->vm_page_prot);
2564 return !!(vma->vm_flags & VM_WRITE);
2565
2566}
2567bool can_change_pte_writable(struct vm_area_struct *vma, unsigned long addr,
2568 pte_t pte);
2569extern long change_protection(struct mmu_gather *tlb,
2570 struct vm_area_struct *vma, unsigned long start,
2571 unsigned long end, unsigned long cp_flags);
2572extern int mprotect_fixup(struct vma_iterator *vmi, struct mmu_gather *tlb,
2573 struct vm_area_struct *vma, struct vm_area_struct **pprev,
2574 unsigned long start, unsigned long end, unsigned long newflags);
2575
2576/*
2577 * doesn't attempt to fault and will return short.
2578 */
2579int get_user_pages_fast_only(unsigned long start, int nr_pages,
2580 unsigned int gup_flags, struct page **pages);
2581
2582static inline bool get_user_page_fast_only(unsigned long addr,
2583 unsigned int gup_flags, struct page **pagep)
2584{
2585 return get_user_pages_fast_only(start: addr, nr_pages: 1, gup_flags, pages: pagep) == 1;
2586}
2587/*
2588 * per-process(per-mm_struct) statistics.
2589 */
2590static inline unsigned long get_mm_counter(struct mm_struct *mm, int member)
2591{
2592 return percpu_counter_read_positive(fbc: &mm->rss_stat[member]);
2593}
2594
2595void mm_trace_rss_stat(struct mm_struct *mm, int member);
2596
2597static inline void add_mm_counter(struct mm_struct *mm, int member, long value)
2598{
2599 percpu_counter_add(fbc: &mm->rss_stat[member], amount: value);
2600
2601 mm_trace_rss_stat(mm, member);
2602}
2603
2604static inline void inc_mm_counter(struct mm_struct *mm, int member)
2605{
2606 percpu_counter_inc(fbc: &mm->rss_stat[member]);
2607
2608 mm_trace_rss_stat(mm, member);
2609}
2610
2611static inline void dec_mm_counter(struct mm_struct *mm, int member)
2612{
2613 percpu_counter_dec(fbc: &mm->rss_stat[member]);
2614
2615 mm_trace_rss_stat(mm, member);
2616}
2617
2618/* Optimized variant when folio is already known not to be anon */
2619static inline int mm_counter_file(struct folio *folio)
2620{
2621 if (folio_test_swapbacked(folio))
2622 return MM_SHMEMPAGES;
2623 return MM_FILEPAGES;
2624}
2625
2626static inline int mm_counter(struct folio *folio)
2627{
2628 if (folio_test_anon(folio))
2629 return MM_ANONPAGES;
2630 return mm_counter_file(folio);
2631}
2632
2633static inline unsigned long get_mm_rss(struct mm_struct *mm)
2634{
2635 return get_mm_counter(mm, member: MM_FILEPAGES) +
2636 get_mm_counter(mm, member: MM_ANONPAGES) +
2637 get_mm_counter(mm, member: MM_SHMEMPAGES);
2638}
2639
2640static inline unsigned long get_mm_hiwater_rss(struct mm_struct *mm)
2641{
2642 return max(mm->hiwater_rss, get_mm_rss(mm));
2643}
2644
2645static inline unsigned long get_mm_hiwater_vm(struct mm_struct *mm)
2646{
2647 return max(mm->hiwater_vm, mm->total_vm);
2648}
2649
2650static inline void update_hiwater_rss(struct mm_struct *mm)
2651{
2652 unsigned long _rss = get_mm_rss(mm);
2653
2654 if ((mm)->hiwater_rss < _rss)
2655 (mm)->hiwater_rss = _rss;
2656}
2657
2658static inline void update_hiwater_vm(struct mm_struct *mm)
2659{
2660 if (mm->hiwater_vm < mm->total_vm)
2661 mm->hiwater_vm = mm->total_vm;
2662}
2663
2664static inline void reset_mm_hiwater_rss(struct mm_struct *mm)
2665{
2666 mm->hiwater_rss = get_mm_rss(mm);
2667}
2668
2669static inline void setmax_mm_hiwater_rss(unsigned long *maxrss,
2670 struct mm_struct *mm)
2671{
2672 unsigned long hiwater_rss = get_mm_hiwater_rss(mm);
2673
2674 if (*maxrss < hiwater_rss)
2675 *maxrss = hiwater_rss;
2676}
2677
2678#ifndef CONFIG_ARCH_HAS_PTE_SPECIAL
2679static inline int pte_special(pte_t pte)
2680{
2681 return 0;
2682}
2683
2684static inline pte_t pte_mkspecial(pte_t pte)
2685{
2686 return pte;
2687}
2688#endif
2689
2690#ifndef CONFIG_ARCH_HAS_PTE_DEVMAP
2691static inline int pte_devmap(pte_t pte)
2692{
2693 return 0;
2694}
2695#endif
2696
2697extern pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
2698 spinlock_t **ptl);
2699static inline pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
2700 spinlock_t **ptl)
2701{
2702 pte_t *ptep;
2703 __cond_lock(*ptl, ptep = __get_locked_pte(mm, addr, ptl));
2704 return ptep;
2705}
2706
2707#ifdef __PAGETABLE_P4D_FOLDED
2708static inline int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd,
2709 unsigned long address)
2710{
2711 return 0;
2712}
2713#else
2714int __p4d_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address);
2715#endif
2716
2717#if defined(__PAGETABLE_PUD_FOLDED) || !defined(CONFIG_MMU)
2718static inline int __pud_alloc(struct mm_struct *mm, p4d_t *p4d,
2719 unsigned long address)
2720{
2721 return 0;
2722}
2723static inline void mm_inc_nr_puds(struct mm_struct *mm) {}
2724static inline void mm_dec_nr_puds(struct mm_struct *mm) {}
2725
2726#else
2727int __pud_alloc(struct mm_struct *mm, p4d_t *p4d, unsigned long address);
2728
2729static inline void mm_inc_nr_puds(struct mm_struct *mm)
2730{
2731 if (mm_pud_folded(mm))
2732 return;
2733 atomic_long_add(PTRS_PER_PUD * sizeof(pud_t), v: &mm->pgtables_bytes);
2734}
2735
2736static inline void mm_dec_nr_puds(struct mm_struct *mm)
2737{
2738 if (mm_pud_folded(mm))
2739 return;
2740 atomic_long_sub(PTRS_PER_PUD * sizeof(pud_t), v: &mm->pgtables_bytes);
2741}
2742#endif
2743
2744#if defined(__PAGETABLE_PMD_FOLDED) || !defined(CONFIG_MMU)
2745static inline int __pmd_alloc(struct mm_struct *mm, pud_t *pud,
2746 unsigned long address)
2747{
2748 return 0;
2749}
2750
2751static inline void mm_inc_nr_pmds(struct mm_struct *mm) {}
2752static inline void mm_dec_nr_pmds(struct mm_struct *mm) {}
2753
2754#else
2755int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address);
2756
2757static inline void mm_inc_nr_pmds(struct mm_struct *mm)
2758{
2759 if (mm_pmd_folded(mm))
2760 return;
2761 atomic_long_add(PTRS_PER_PMD * sizeof(pmd_t), v: &mm->pgtables_bytes);
2762}
2763
2764static inline void mm_dec_nr_pmds(struct mm_struct *mm)
2765{
2766 if (mm_pmd_folded(mm))
2767 return;
2768 atomic_long_sub(PTRS_PER_PMD * sizeof(pmd_t), v: &mm->pgtables_bytes);
2769}
2770#endif
2771
2772#ifdef CONFIG_MMU
2773static inline void mm_pgtables_bytes_init(struct mm_struct *mm)
2774{
2775 atomic_long_set(v: &mm->pgtables_bytes, i: 0);
2776}
2777
2778static inline unsigned long mm_pgtables_bytes(const struct mm_struct *mm)
2779{
2780 return atomic_long_read(v: &mm->pgtables_bytes);
2781}
2782
2783static inline void mm_inc_nr_ptes(struct mm_struct *mm)
2784{
2785 atomic_long_add(PTRS_PER_PTE * sizeof(pte_t), v: &mm->pgtables_bytes);
2786}
2787
2788static inline void mm_dec_nr_ptes(struct mm_struct *mm)
2789{
2790 atomic_long_sub(PTRS_PER_PTE * sizeof(pte_t), v: &mm->pgtables_bytes);
2791}
2792#else
2793
2794static inline void mm_pgtables_bytes_init(struct mm_struct *mm) {}
2795static inline unsigned long mm_pgtables_bytes(const struct mm_struct *mm)
2796{
2797 return 0;
2798}
2799
2800static inline void mm_inc_nr_ptes(struct mm_struct *mm) {}
2801static inline void mm_dec_nr_ptes(struct mm_struct *mm) {}
2802#endif
2803
2804int __pte_alloc(struct mm_struct *mm, pmd_t *pmd);
2805int __pte_alloc_kernel(pmd_t *pmd);
2806
2807#if defined(CONFIG_MMU)
2808
2809static inline p4d_t *p4d_alloc(struct mm_struct *mm, pgd_t *pgd,
2810 unsigned long address)
2811{
2812 return (unlikely(pgd_none(*pgd)) && __p4d_alloc(mm, pgd, address)) ?
2813 NULL : p4d_offset(pgd, address);
2814}
2815
2816static inline pud_t *pud_alloc(struct mm_struct *mm, p4d_t *p4d,
2817 unsigned long address)
2818{
2819 return (unlikely(p4d_none(*p4d)) && __pud_alloc(mm, p4d, address)) ?
2820 NULL : pud_offset(p4d, address);
2821}
2822
2823static inline pmd_t *pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2824{
2825 return (unlikely(pud_none(*pud)) && __pmd_alloc(mm, pud, address))?
2826 NULL: pmd_offset(pud, address);
2827}
2828#endif /* CONFIG_MMU */
2829
2830static inline struct ptdesc *virt_to_ptdesc(const void *x)
2831{
2832 return page_ptdesc(virt_to_page(x));
2833}
2834
2835static inline void *ptdesc_to_virt(const struct ptdesc *pt)
2836{
2837 return page_to_virt(ptdesc_page(pt));
2838}
2839
2840static inline void *ptdesc_address(const struct ptdesc *pt)
2841{
2842 return folio_address(ptdesc_folio(pt));
2843}
2844
2845static inline bool pagetable_is_reserved(struct ptdesc *pt)
2846{
2847 return folio_test_reserved(ptdesc_folio(pt));
2848}
2849
2850/**
2851 * pagetable_alloc - Allocate pagetables
2852 * @gfp: GFP flags
2853 * @order: desired pagetable order
2854 *
2855 * pagetable_alloc allocates memory for page tables as well as a page table
2856 * descriptor to describe that memory.
2857 *
2858 * Return: The ptdesc describing the allocated page tables.
2859 */
2860static inline struct ptdesc *pagetable_alloc(gfp_t gfp, unsigned int order)
2861{
2862 struct page *page = alloc_pages(gfp: gfp | __GFP_COMP, order);
2863
2864 return page_ptdesc(page);
2865}
2866
2867/**
2868 * pagetable_free - Free pagetables
2869 * @pt: The page table descriptor
2870 *
2871 * pagetable_free frees the memory of all page tables described by a page
2872 * table descriptor and the memory for the descriptor itself.
2873 */
2874static inline void pagetable_free(struct ptdesc *pt)
2875{
2876 struct page *page = ptdesc_page(pt);
2877
2878 __free_pages(page, order: compound_order(page));
2879}
2880
2881#if USE_SPLIT_PTE_PTLOCKS
2882#if ALLOC_SPLIT_PTLOCKS
2883void __init ptlock_cache_init(void);
2884bool ptlock_alloc(struct ptdesc *ptdesc);
2885void ptlock_free(struct ptdesc *ptdesc);
2886
2887static inline spinlock_t *ptlock_ptr(struct ptdesc *ptdesc)
2888{
2889 return ptdesc->ptl;
2890}
2891#else /* ALLOC_SPLIT_PTLOCKS */
2892static inline void ptlock_cache_init(void)
2893{
2894}
2895
2896static inline bool ptlock_alloc(struct ptdesc *ptdesc)
2897{
2898 return true;
2899}
2900
2901static inline void ptlock_free(struct ptdesc *ptdesc)
2902{
2903}
2904
2905static inline spinlock_t *ptlock_ptr(struct ptdesc *ptdesc)
2906{
2907 return &ptdesc->ptl;
2908}
2909#endif /* ALLOC_SPLIT_PTLOCKS */
2910
2911static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd)
2912{
2913 return ptlock_ptr(page_ptdesc(pmd_page(*pmd)));
2914}
2915
2916static inline bool ptlock_init(struct ptdesc *ptdesc)
2917{
2918 /*
2919 * prep_new_page() initialize page->private (and therefore page->ptl)
2920 * with 0. Make sure nobody took it in use in between.
2921 *
2922 * It can happen if arch try to use slab for page table allocation:
2923 * slab code uses page->slab_cache, which share storage with page->ptl.
2924 */
2925 VM_BUG_ON_PAGE(*(unsigned long *)&ptdesc->ptl, ptdesc_page(ptdesc));
2926 if (!ptlock_alloc(ptdesc))
2927 return false;
2928 spin_lock_init(ptlock_ptr(ptdesc));
2929 return true;
2930}
2931
2932#else /* !USE_SPLIT_PTE_PTLOCKS */
2933/*
2934 * We use mm->page_table_lock to guard all pagetable pages of the mm.
2935 */
2936static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd)
2937{
2938 return &mm->page_table_lock;
2939}
2940static inline void ptlock_cache_init(void) {}
2941static inline bool ptlock_init(struct ptdesc *ptdesc) { return true; }
2942static inline void ptlock_free(struct ptdesc *ptdesc) {}
2943#endif /* USE_SPLIT_PTE_PTLOCKS */
2944
2945static inline bool pagetable_pte_ctor(struct ptdesc *ptdesc)
2946{
2947 struct folio *folio = ptdesc_folio(ptdesc);
2948
2949 if (!ptlock_init(ptdesc))
2950 return false;
2951 __folio_set_pgtable(folio);
2952 lruvec_stat_add_folio(folio, idx: NR_PAGETABLE);
2953 return true;
2954}
2955
2956static inline void pagetable_pte_dtor(struct ptdesc *ptdesc)
2957{
2958 struct folio *folio = ptdesc_folio(ptdesc);
2959
2960 ptlock_free(ptdesc);
2961 __folio_clear_pgtable(folio);
2962 lruvec_stat_sub_folio(folio, idx: NR_PAGETABLE);
2963}
2964
2965pte_t *__pte_offset_map(pmd_t *pmd, unsigned long addr, pmd_t *pmdvalp);
2966static inline pte_t *pte_offset_map(pmd_t *pmd, unsigned long addr)
2967{
2968 return __pte_offset_map(pmd, addr, NULL);
2969}
2970
2971pte_t *__pte_offset_map_lock(struct mm_struct *mm, pmd_t *pmd,
2972 unsigned long addr, spinlock_t **ptlp);
2973static inline pte_t *pte_offset_map_lock(struct mm_struct *mm, pmd_t *pmd,
2974 unsigned long addr, spinlock_t **ptlp)
2975{
2976 pte_t *pte;
2977
2978 __cond_lock(*ptlp, pte = __pte_offset_map_lock(mm, pmd, addr, ptlp));
2979 return pte;
2980}
2981
2982pte_t *pte_offset_map_nolock(struct mm_struct *mm, pmd_t *pmd,
2983 unsigned long addr, spinlock_t **ptlp);
2984
2985#define pte_unmap_unlock(pte, ptl) do { \
2986 spin_unlock(ptl); \
2987 pte_unmap(pte); \
2988} while (0)
2989
2990#define pte_alloc(mm, pmd) (unlikely(pmd_none(*(pmd))) && __pte_alloc(mm, pmd))
2991
2992#define pte_alloc_map(mm, pmd, address) \
2993 (pte_alloc(mm, pmd) ? NULL : pte_offset_map(pmd, address))
2994
2995#define pte_alloc_map_lock(mm, pmd, address, ptlp) \
2996 (pte_alloc(mm, pmd) ? \
2997 NULL : pte_offset_map_lock(mm, pmd, address, ptlp))
2998
2999#define pte_alloc_kernel(pmd, address) \
3000 ((unlikely(pmd_none(*(pmd))) && __pte_alloc_kernel(pmd))? \
3001 NULL: pte_offset_kernel(pmd, address))
3002
3003#if USE_SPLIT_PMD_PTLOCKS
3004
3005static inline struct page *pmd_pgtable_page(pmd_t *pmd)
3006{
3007 unsigned long mask = ~(PTRS_PER_PMD * sizeof(pmd_t) - 1);
3008 return virt_to_page((void *)((unsigned long) pmd & mask));
3009}
3010
3011static inline struct ptdesc *pmd_ptdesc(pmd_t *pmd)
3012{
3013 return page_ptdesc(pmd_pgtable_page(pmd));
3014}
3015
3016static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd)
3017{
3018 return ptlock_ptr(ptdesc: pmd_ptdesc(pmd));
3019}
3020
3021static inline bool pmd_ptlock_init(struct ptdesc *ptdesc)
3022{
3023#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3024 ptdesc->pmd_huge_pte = NULL;
3025#endif
3026 return ptlock_init(ptdesc);
3027}
3028
3029static inline void pmd_ptlock_free(struct ptdesc *ptdesc)
3030{
3031#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3032 VM_BUG_ON_PAGE(ptdesc->pmd_huge_pte, ptdesc_page(ptdesc));
3033#endif
3034 ptlock_free(ptdesc);
3035}
3036
3037#define pmd_huge_pte(mm, pmd) (pmd_ptdesc(pmd)->pmd_huge_pte)
3038
3039#else
3040
3041static inline spinlock_t *pmd_lockptr(struct mm_struct *mm, pmd_t *pmd)
3042{
3043 return &mm->page_table_lock;
3044}
3045
3046static inline bool pmd_ptlock_init(struct ptdesc *ptdesc) { return true; }
3047static inline void pmd_ptlock_free(struct ptdesc *ptdesc) {}
3048
3049#define pmd_huge_pte(mm, pmd) ((mm)->pmd_huge_pte)
3050
3051#endif
3052
3053static inline spinlock_t *pmd_lock(struct mm_struct *mm, pmd_t *pmd)
3054{
3055 spinlock_t *ptl = pmd_lockptr(mm, pmd);
3056 spin_lock(lock: ptl);
3057 return ptl;
3058}
3059
3060static inline bool pagetable_pmd_ctor(struct ptdesc *ptdesc)
3061{
3062 struct folio *folio = ptdesc_folio(ptdesc);
3063
3064 if (!pmd_ptlock_init(ptdesc))
3065 return false;
3066 __folio_set_pgtable(folio);
3067 lruvec_stat_add_folio(folio, idx: NR_PAGETABLE);
3068 return true;
3069}
3070
3071static inline void pagetable_pmd_dtor(struct ptdesc *ptdesc)
3072{
3073 struct folio *folio = ptdesc_folio(ptdesc);
3074
3075 pmd_ptlock_free(ptdesc);
3076 __folio_clear_pgtable(folio);
3077 lruvec_stat_sub_folio(folio, idx: NR_PAGETABLE);
3078}
3079
3080/*
3081 * No scalability reason to split PUD locks yet, but follow the same pattern
3082 * as the PMD locks to make it easier if we decide to. The VM should not be
3083 * considered ready to switch to split PUD locks yet; there may be places
3084 * which need to be converted from page_table_lock.
3085 */
3086static inline spinlock_t *pud_lockptr(struct mm_struct *mm, pud_t *pud)
3087{
3088 return &mm->page_table_lock;
3089}
3090
3091static inline spinlock_t *pud_lock(struct mm_struct *mm, pud_t *pud)
3092{
3093 spinlock_t *ptl = pud_lockptr(mm, pud);
3094
3095 spin_lock(lock: ptl);
3096 return ptl;
3097}
3098
3099static inline void pagetable_pud_ctor(struct ptdesc *ptdesc)
3100{
3101 struct folio *folio = ptdesc_folio(ptdesc);
3102
3103 __folio_set_pgtable(folio);
3104 lruvec_stat_add_folio(folio, idx: NR_PAGETABLE);
3105}
3106
3107static inline void pagetable_pud_dtor(struct ptdesc *ptdesc)
3108{
3109 struct folio *folio = ptdesc_folio(ptdesc);
3110
3111 __folio_clear_pgtable(folio);
3112 lruvec_stat_sub_folio(folio, idx: NR_PAGETABLE);
3113}
3114
3115extern void __init pagecache_init(void);
3116extern void free_initmem(void);
3117
3118/*
3119 * Free reserved pages within range [PAGE_ALIGN(start), end & PAGE_MASK)
3120 * into the buddy system. The freed pages will be poisoned with pattern
3121 * "poison" if it's within range [0, UCHAR_MAX].
3122 * Return pages freed into the buddy system.
3123 */
3124extern unsigned long free_reserved_area(void *start, void *end,
3125 int poison, const char *s);
3126
3127extern void adjust_managed_page_count(struct page *page, long count);
3128
3129extern void reserve_bootmem_region(phys_addr_t start,
3130 phys_addr_t end, int nid);
3131
3132/* Free the reserved page into the buddy system, so it gets managed. */
3133static inline void free_reserved_page(struct page *page)
3134{
3135 ClearPageReserved(page);
3136 init_page_count(page);
3137 __free_page(page);
3138 adjust_managed_page_count(page, count: 1);
3139}
3140#define free_highmem_page(page) free_reserved_page(page)
3141
3142static inline void mark_page_reserved(struct page *page)
3143{
3144 SetPageReserved(page);
3145 adjust_managed_page_count(page, count: -1);
3146}
3147
3148static inline void free_reserved_ptdesc(struct ptdesc *pt)
3149{
3150 free_reserved_page(ptdesc_page(pt));
3151}
3152
3153/*
3154 * Default method to free all the __init memory into the buddy system.
3155 * The freed pages will be poisoned with pattern "poison" if it's within
3156 * range [0, UCHAR_MAX].
3157 * Return pages freed into the buddy system.
3158 */
3159static inline unsigned long free_initmem_default(int poison)
3160{
3161 extern char __init_begin[], __init_end[];
3162
3163 return free_reserved_area(start: &__init_begin, end: &__init_end,
3164 poison, s: "unused kernel image (initmem)");
3165}
3166
3167static inline unsigned long get_num_physpages(void)
3168{
3169 int nid;
3170 unsigned long phys_pages = 0;
3171
3172 for_each_online_node(nid)
3173 phys_pages += node_present_pages(nid);
3174
3175 return phys_pages;
3176}
3177
3178/*
3179 * Using memblock node mappings, an architecture may initialise its
3180 * zones, allocate the backing mem_map and account for memory holes in an
3181 * architecture independent manner.
3182 *
3183 * An architecture is expected to register range of page frames backed by
3184 * physical memory with memblock_add[_node]() before calling
3185 * free_area_init() passing in the PFN each zone ends at. At a basic
3186 * usage, an architecture is expected to do something like
3187 *
3188 * unsigned long max_zone_pfns[MAX_NR_ZONES] = {max_dma, max_normal_pfn,
3189 * max_highmem_pfn};
3190 * for_each_valid_physical_page_range()
3191 * memblock_add_node(base, size, nid, MEMBLOCK_NONE)
3192 * free_area_init(max_zone_pfns);
3193 */
3194void free_area_init(unsigned long *max_zone_pfn);
3195unsigned long node_map_pfn_alignment(void);
3196unsigned long __absent_pages_in_range(int nid, unsigned long start_pfn,
3197 unsigned long end_pfn);
3198extern unsigned long absent_pages_in_range(unsigned long start_pfn,
3199 unsigned long end_pfn);
3200extern void get_pfn_range_for_nid(unsigned int nid,
3201 unsigned long *start_pfn, unsigned long *end_pfn);
3202
3203#ifndef CONFIG_NUMA
3204static inline int early_pfn_to_nid(unsigned long pfn)
3205{
3206 return 0;
3207}
3208#else
3209/* please see mm/page_alloc.c */
3210extern int __meminit early_pfn_to_nid(unsigned long pfn);
3211#endif
3212
3213extern void set_dma_reserve(unsigned long new_dma_reserve);
3214extern void mem_init(void);
3215extern void __init mmap_init(void);
3216
3217extern void __show_mem(unsigned int flags, nodemask_t *nodemask, int max_zone_idx);
3218static inline void show_mem(void)
3219{
3220 __show_mem(flags: 0, NULL, max_zone_idx: MAX_NR_ZONES - 1);
3221}
3222extern long si_mem_available(void);
3223extern void si_meminfo(struct sysinfo * val);
3224extern void si_meminfo_node(struct sysinfo *val, int nid);
3225#ifdef __HAVE_ARCH_RESERVED_KERNEL_PAGES
3226extern unsigned long arch_reserved_kernel_pages(void);
3227#endif
3228
3229extern __printf(3, 4)
3230void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...);
3231
3232extern void setup_per_cpu_pageset(void);
3233
3234/* nommu.c */
3235extern atomic_long_t mmap_pages_allocated;
3236extern int nommu_shrink_inode_mappings(struct inode *, size_t, size_t);
3237
3238/* interval_tree.c */
3239void vma_interval_tree_insert(struct vm_area_struct *node,
3240 struct rb_root_cached *root);
3241void vma_interval_tree_insert_after(struct vm_area_struct *node,
3242 struct vm_area_struct *prev,
3243 struct rb_root_cached *root);
3244void vma_interval_tree_remove(struct vm_area_struct *node,
3245 struct rb_root_cached *root);
3246struct vm_area_struct *vma_interval_tree_iter_first(struct rb_root_cached *root,
3247 unsigned long start, unsigned long last);
3248struct vm_area_struct *vma_interval_tree_iter_next(struct vm_area_struct *node,
3249 unsigned long start, unsigned long last);
3250
3251#define vma_interval_tree_foreach(vma, root, start, last) \
3252 for (vma = vma_interval_tree_iter_first(root, start, last); \
3253 vma; vma = vma_interval_tree_iter_next(vma, start, last))
3254
3255void anon_vma_interval_tree_insert(struct anon_vma_chain *node,
3256 struct rb_root_cached *root);
3257void anon_vma_interval_tree_remove(struct anon_vma_chain *node,
3258 struct rb_root_cached *root);
3259struct anon_vma_chain *
3260anon_vma_interval_tree_iter_first(struct rb_root_cached *root,
3261 unsigned long start, unsigned long last);
3262struct anon_vma_chain *anon_vma_interval_tree_iter_next(
3263 struct anon_vma_chain *node, unsigned long start, unsigned long last);
3264#ifdef CONFIG_DEBUG_VM_RB
3265void anon_vma_interval_tree_verify(struct anon_vma_chain *node);
3266#endif
3267
3268#define anon_vma_interval_tree_foreach(avc, root, start, last) \
3269 for (avc = anon_vma_interval_tree_iter_first(root, start, last); \
3270 avc; avc = anon_vma_interval_tree_iter_next(avc, start, last))
3271
3272/* mmap.c */
3273extern int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin);
3274extern int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma,
3275 unsigned long start, unsigned long end, pgoff_t pgoff,
3276 struct vm_area_struct *next);
3277extern int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
3278 unsigned long start, unsigned long end, pgoff_t pgoff);
3279extern struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *);
3280extern int insert_vm_struct(struct mm_struct *, struct vm_area_struct *);
3281extern void unlink_file_vma(struct vm_area_struct *);
3282extern struct vm_area_struct *copy_vma(struct vm_area_struct **,
3283 unsigned long addr, unsigned long len, pgoff_t pgoff,
3284 bool *need_rmap_locks);
3285extern void exit_mmap(struct mm_struct *);
3286struct vm_area_struct *vma_modify(struct vma_iterator *vmi,
3287 struct vm_area_struct *prev,
3288 struct vm_area_struct *vma,
3289 unsigned long start, unsigned long end,
3290 unsigned long vm_flags,
3291 struct mempolicy *policy,
3292 struct vm_userfaultfd_ctx uffd_ctx,
3293 struct anon_vma_name *anon_name);
3294
3295/* We are about to modify the VMA's flags. */
3296static inline struct vm_area_struct
3297*vma_modify_flags(struct vma_iterator *vmi,
3298 struct vm_area_struct *prev,
3299 struct vm_area_struct *vma,
3300 unsigned long start, unsigned long end,
3301 unsigned long new_flags)
3302{
3303 return vma_modify(vmi, prev, vma, start, end, vm_flags: new_flags,
3304 vma_policy(vma), uffd_ctx: vma->vm_userfaultfd_ctx,
3305 anon_name: anon_vma_name(vma));
3306}
3307
3308/* We are about to modify the VMA's flags and/or anon_name. */
3309static inline struct vm_area_struct
3310*vma_modify_flags_name(struct vma_iterator *vmi,
3311 struct vm_area_struct *prev,
3312 struct vm_area_struct *vma,
3313 unsigned long start,
3314 unsigned long end,
3315 unsigned long new_flags,
3316 struct anon_vma_name *new_name)
3317{
3318 return vma_modify(vmi, prev, vma, start, end, vm_flags: new_flags,
3319 vma_policy(vma), uffd_ctx: vma->vm_userfaultfd_ctx, anon_name: new_name);
3320}
3321
3322/* We are about to modify the VMA's memory policy. */
3323static inline struct vm_area_struct
3324*vma_modify_policy(struct vma_iterator *vmi,
3325 struct vm_area_struct *prev,
3326 struct vm_area_struct *vma,
3327 unsigned long start, unsigned long end,
3328 struct mempolicy *new_pol)
3329{
3330 return vma_modify(vmi, prev, vma, start, end, vm_flags: vma->vm_flags,
3331 policy: new_pol, uffd_ctx: vma->vm_userfaultfd_ctx, anon_name: anon_vma_name(vma));
3332}
3333
3334/* We are about to modify the VMA's flags and/or uffd context. */
3335static inline struct vm_area_struct
3336*vma_modify_flags_uffd(struct vma_iterator *vmi,
3337 struct vm_area_struct *prev,
3338 struct vm_area_struct *vma,
3339 unsigned long start, unsigned long end,
3340 unsigned long new_flags,
3341 struct vm_userfaultfd_ctx new_ctx)
3342{
3343 return vma_modify(vmi, prev, vma, start, end, vm_flags: new_flags,
3344 vma_policy(vma), uffd_ctx: new_ctx, anon_name: anon_vma_name(vma));
3345}
3346
3347static inline int check_data_rlimit(unsigned long rlim,
3348 unsigned long new,
3349 unsigned long start,
3350 unsigned long end_data,
3351 unsigned long start_data)
3352{
3353 if (rlim < RLIM_INFINITY) {
3354 if (((new - start) + (end_data - start_data)) > rlim)
3355 return -ENOSPC;
3356 }
3357
3358 return 0;
3359}
3360
3361extern int mm_take_all_locks(struct mm_struct *mm);
3362extern void mm_drop_all_locks(struct mm_struct *mm);
3363
3364extern int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file);
3365extern int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file);
3366extern struct file *get_mm_exe_file(struct mm_struct *mm);
3367extern struct file *get_task_exe_file(struct task_struct *task);
3368
3369extern bool may_expand_vm(struct mm_struct *, vm_flags_t, unsigned long npages);
3370extern void vm_stat_account(struct mm_struct *, vm_flags_t, long npages);
3371
3372extern bool vma_is_special_mapping(const struct vm_area_struct *vma,
3373 const struct vm_special_mapping *sm);
3374extern struct vm_area_struct *_install_special_mapping(struct mm_struct *mm,
3375 unsigned long addr, unsigned long len,
3376 unsigned long flags,
3377 const struct vm_special_mapping *spec);
3378/* This is an obsolete alternative to _install_special_mapping. */
3379extern int install_special_mapping(struct mm_struct *mm,
3380 unsigned long addr, unsigned long len,
3381 unsigned long flags, struct page **pages);
3382
3383unsigned long randomize_stack_top(unsigned long stack_top);
3384unsigned long randomize_page(unsigned long start, unsigned long range);
3385
3386extern unsigned long get_unmapped_area(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
3387
3388extern unsigned long mmap_region(struct file *file, unsigned long addr,
3389 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
3390 struct list_head *uf);
3391extern unsigned long do_mmap(struct file *file, unsigned long addr,
3392 unsigned long len, unsigned long prot, unsigned long flags,
3393 vm_flags_t vm_flags, unsigned long pgoff, unsigned long *populate,
3394 struct list_head *uf);
3395extern int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm,
3396 unsigned long start, size_t len, struct list_head *uf,
3397 bool unlock);
3398extern int do_munmap(struct mm_struct *, unsigned long, size_t,
3399 struct list_head *uf);
3400extern int do_madvise(struct mm_struct *mm, unsigned long start, size_t len_in, int behavior);
3401
3402#ifdef CONFIG_MMU
3403extern int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma,
3404 unsigned long start, unsigned long end,
3405 struct list_head *uf, bool unlock);
3406extern int __mm_populate(unsigned long addr, unsigned long len,
3407 int ignore_errors);
3408static inline void mm_populate(unsigned long addr, unsigned long len)
3409{
3410 /* Ignore errors */
3411 (void) __mm_populate(addr, len, ignore_errors: 1);
3412}
3413#else
3414static inline void mm_populate(unsigned long addr, unsigned long len) {}
3415#endif
3416
3417/* This takes the mm semaphore itself */
3418extern int __must_check vm_brk_flags(unsigned long, unsigned long, unsigned long);
3419extern int vm_munmap(unsigned long, size_t);
3420extern unsigned long __must_check vm_mmap(struct file *, unsigned long,
3421 unsigned long, unsigned long,
3422 unsigned long, unsigned long);
3423
3424struct vm_unmapped_area_info {
3425#define VM_UNMAPPED_AREA_TOPDOWN 1
3426 unsigned long flags;
3427 unsigned long length;
3428 unsigned long low_limit;
3429 unsigned long high_limit;
3430 unsigned long align_mask;
3431 unsigned long align_offset;
3432};
3433
3434extern unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info);
3435
3436/* truncate.c */
3437extern void truncate_inode_pages(struct address_space *, loff_t);
3438extern void truncate_inode_pages_range(struct address_space *,
3439 loff_t lstart, loff_t lend);
3440extern void truncate_inode_pages_final(struct address_space *);
3441
3442/* generic vm_area_ops exported for stackable file systems */
3443extern vm_fault_t filemap_fault(struct vm_fault *vmf);
3444extern vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3445 pgoff_t start_pgoff, pgoff_t end_pgoff);
3446extern vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf);
3447
3448extern unsigned long stack_guard_gap;
3449/* Generic expand stack which grows the stack according to GROWS{UP,DOWN} */
3450int expand_stack_locked(struct vm_area_struct *vma, unsigned long address);
3451struct vm_area_struct *expand_stack(struct mm_struct * mm, unsigned long addr);
3452
3453/* CONFIG_STACK_GROWSUP still needs to grow downwards at some places */
3454int expand_downwards(struct vm_area_struct *vma, unsigned long address);
3455
3456/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
3457extern struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr);
3458extern struct vm_area_struct * find_vma_prev(struct mm_struct * mm, unsigned long addr,
3459 struct vm_area_struct **pprev);
3460
3461/*
3462 * Look up the first VMA which intersects the interval [start_addr, end_addr)
3463 * NULL if none. Assume start_addr < end_addr.
3464 */
3465struct vm_area_struct *find_vma_intersection(struct mm_struct *mm,
3466 unsigned long start_addr, unsigned long end_addr);
3467
3468/**
3469 * vma_lookup() - Find a VMA at a specific address
3470 * @mm: The process address space.
3471 * @addr: The user address.
3472 *
3473 * Return: The vm_area_struct at the given address, %NULL otherwise.
3474 */
3475static inline
3476struct vm_area_struct *vma_lookup(struct mm_struct *mm, unsigned long addr)
3477{
3478 return mtree_load(mt: &mm->mm_mt, index: addr);
3479}
3480
3481static inline unsigned long stack_guard_start_gap(struct vm_area_struct *vma)
3482{
3483 if (vma->vm_flags & VM_GROWSDOWN)
3484 return stack_guard_gap;
3485
3486 /* See reasoning around the VM_SHADOW_STACK definition */
3487 if (vma->vm_flags & VM_SHADOW_STACK)
3488 return PAGE_SIZE;
3489
3490 return 0;
3491}
3492
3493static inline unsigned long vm_start_gap(struct vm_area_struct *vma)
3494{
3495 unsigned long gap = stack_guard_start_gap(vma);
3496 unsigned long vm_start = vma->vm_start;
3497
3498 vm_start -= gap;
3499 if (vm_start > vma->vm_start)
3500 vm_start = 0;
3501 return vm_start;
3502}
3503
3504static inline unsigned long vm_end_gap(struct vm_area_struct *vma)
3505{
3506 unsigned long vm_end = vma->vm_end;
3507
3508 if (vma->vm_flags & VM_GROWSUP) {
3509 vm_end += stack_guard_gap;
3510 if (vm_end < vma->vm_end)
3511 vm_end = -PAGE_SIZE;
3512 }
3513 return vm_end;
3514}
3515
3516static inline unsigned long vma_pages(struct vm_area_struct *vma)
3517{
3518 return (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
3519}
3520
3521/* Look up the first VMA which exactly match the interval vm_start ... vm_end */
3522static inline struct vm_area_struct *find_exact_vma(struct mm_struct *mm,
3523 unsigned long vm_start, unsigned long vm_end)
3524{
3525 struct vm_area_struct *vma = vma_lookup(mm, addr: vm_start);
3526
3527 if (vma && (vma->vm_start != vm_start || vma->vm_end != vm_end))
3528 vma = NULL;
3529
3530 return vma;
3531}
3532
3533static inline bool range_in_vma(struct vm_area_struct *vma,
3534 unsigned long start, unsigned long end)
3535{
3536 return (vma && vma->vm_start <= start && end <= vma->vm_end);
3537}
3538
3539#ifdef CONFIG_MMU
3540pgprot_t vm_get_page_prot(unsigned long vm_flags);
3541void vma_set_page_prot(struct vm_area_struct *vma);
3542#else
3543static inline pgprot_t vm_get_page_prot(unsigned long vm_flags)
3544{
3545 return __pgprot(0);
3546}
3547static inline void vma_set_page_prot(struct vm_area_struct *vma)
3548{
3549 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
3550}
3551#endif
3552
3553void vma_set_file(struct vm_area_struct *vma, struct file *file);
3554
3555#ifdef CONFIG_NUMA_BALANCING
3556unsigned long change_prot_numa(struct vm_area_struct *vma,
3557 unsigned long start, unsigned long end);
3558#endif
3559
3560struct vm_area_struct *find_extend_vma_locked(struct mm_struct *,
3561 unsigned long addr);
3562int remap_pfn_range(struct vm_area_struct *, unsigned long addr,
3563 unsigned long pfn, unsigned long size, pgprot_t);
3564int remap_pfn_range_notrack(struct vm_area_struct *vma, unsigned long addr,
3565 unsigned long pfn, unsigned long size, pgprot_t prot);
3566int vm_insert_page(struct vm_area_struct *, unsigned long addr, struct page *);
3567int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr,
3568 struct page **pages, unsigned long *num);
3569int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
3570 unsigned long num);
3571int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,
3572 unsigned long num);
3573vm_fault_t vmf_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
3574 unsigned long pfn);
3575vm_fault_t vmf_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr,
3576 unsigned long pfn, pgprot_t pgprot);
3577vm_fault_t vmf_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
3578 pfn_t pfn);
3579vm_fault_t vmf_insert_mixed_mkwrite(struct vm_area_struct *vma,
3580 unsigned long addr, pfn_t pfn);
3581int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len);
3582
3583static inline vm_fault_t vmf_insert_page(struct vm_area_struct *vma,
3584 unsigned long addr, struct page *page)
3585{
3586 int err = vm_insert_page(vma, addr, page);
3587
3588 if (err == -ENOMEM)
3589 return VM_FAULT_OOM;
3590 if (err < 0 && err != -EBUSY)
3591 return VM_FAULT_SIGBUS;
3592
3593 return VM_FAULT_NOPAGE;
3594}
3595
3596#ifndef io_remap_pfn_range
3597static inline int io_remap_pfn_range(struct vm_area_struct *vma,
3598 unsigned long addr, unsigned long pfn,
3599 unsigned long size, pgprot_t prot)
3600{
3601 return remap_pfn_range(vma, addr, pfn, size, pgprot_decrypted(prot));
3602}
3603#endif
3604
3605static inline vm_fault_t vmf_error(int err)
3606{
3607 if (err == -ENOMEM)
3608 return VM_FAULT_OOM;
3609 else if (err == -EHWPOISON)
3610 return VM_FAULT_HWPOISON;
3611 return VM_FAULT_SIGBUS;
3612}
3613
3614/*
3615 * Convert errno to return value for ->page_mkwrite() calls.
3616 *
3617 * This should eventually be merged with vmf_error() above, but will need a
3618 * careful audit of all vmf_error() callers.
3619 */
3620static inline vm_fault_t vmf_fs_error(int err)
3621{
3622 if (err == 0)
3623 return VM_FAULT_LOCKED;
3624 if (err == -EFAULT || err == -EAGAIN)
3625 return VM_FAULT_NOPAGE;
3626 if (err == -ENOMEM)
3627 return VM_FAULT_OOM;
3628 /* -ENOSPC, -EDQUOT, -EIO ... */
3629 return VM_FAULT_SIGBUS;
3630}
3631
3632struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
3633 unsigned int foll_flags);
3634
3635static inline int vm_fault_to_errno(vm_fault_t vm_fault, int foll_flags)
3636{
3637 if (vm_fault & VM_FAULT_OOM)
3638 return -ENOMEM;
3639 if (vm_fault & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE))
3640 return (foll_flags & FOLL_HWPOISON) ? -EHWPOISON : -EFAULT;
3641 if (vm_fault & (VM_FAULT_SIGBUS | VM_FAULT_SIGSEGV))
3642 return -EFAULT;
3643 return 0;
3644}
3645
3646/*
3647 * Indicates whether GUP can follow a PROT_NONE mapped page, or whether
3648 * a (NUMA hinting) fault is required.
3649 */
3650static inline bool gup_can_follow_protnone(struct vm_area_struct *vma,
3651 unsigned int flags)
3652{
3653 /*
3654 * If callers don't want to honor NUMA hinting faults, no need to
3655 * determine if we would actually have to trigger a NUMA hinting fault.
3656 */
3657 if (!(flags & FOLL_HONOR_NUMA_FAULT))
3658 return true;
3659
3660 /*
3661 * NUMA hinting faults don't apply in inaccessible (PROT_NONE) VMAs.
3662 *
3663 * Requiring a fault here even for inaccessible VMAs would mean that
3664 * FOLL_FORCE cannot make any progress, because handle_mm_fault()
3665 * refuses to process NUMA hinting faults in inaccessible VMAs.
3666 */
3667 return !vma_is_accessible(vma);
3668}
3669
3670typedef int (*pte_fn_t)(pte_t *pte, unsigned long addr, void *data);
3671extern int apply_to_page_range(struct mm_struct *mm, unsigned long address,
3672 unsigned long size, pte_fn_t fn, void *data);
3673extern int apply_to_existing_page_range(struct mm_struct *mm,
3674 unsigned long address, unsigned long size,
3675 pte_fn_t fn, void *data);
3676
3677#ifdef CONFIG_PAGE_POISONING
3678extern void __kernel_poison_pages(struct page *page, int numpages);
3679extern void __kernel_unpoison_pages(struct page *page, int numpages);
3680extern bool _page_poisoning_enabled_early;
3681DECLARE_STATIC_KEY_FALSE(_page_poisoning_enabled);
3682static inline bool page_poisoning_enabled(void)
3683{
3684 return _page_poisoning_enabled_early;
3685}
3686/*
3687 * For use in fast paths after init_mem_debugging() has run, or when a
3688 * false negative result is not harmful when called too early.
3689 */
3690static inline bool page_poisoning_enabled_static(void)
3691{
3692 return static_branch_unlikely(&_page_poisoning_enabled);
3693}
3694static inline void kernel_poison_pages(struct page *page, int numpages)
3695{
3696 if (page_poisoning_enabled_static())
3697 __kernel_poison_pages(page, numpages);
3698}
3699static inline void kernel_unpoison_pages(struct page *page, int numpages)
3700{
3701 if (page_poisoning_enabled_static())
3702 __kernel_unpoison_pages(page, numpages);
3703}
3704#else
3705static inline bool page_poisoning_enabled(void) { return false; }
3706static inline bool page_poisoning_enabled_static(void) { return false; }
3707static inline void __kernel_poison_pages(struct page *page, int nunmpages) { }
3708static inline void kernel_poison_pages(struct page *page, int numpages) { }
3709static inline void kernel_unpoison_pages(struct page *page, int numpages) { }
3710#endif
3711
3712DECLARE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc);
3713static inline bool want_init_on_alloc(gfp_t flags)
3714{
3715 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
3716 &init_on_alloc))
3717 return true;
3718 return flags & __GFP_ZERO;
3719}
3720
3721DECLARE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free);
3722static inline bool want_init_on_free(void)
3723{
3724 return static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
3725 &init_on_free);
3726}
3727
3728extern bool _debug_pagealloc_enabled_early;
3729DECLARE_STATIC_KEY_FALSE(_debug_pagealloc_enabled);
3730
3731static inline bool debug_pagealloc_enabled(void)
3732{
3733 return IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
3734 _debug_pagealloc_enabled_early;
3735}
3736
3737/*
3738 * For use in fast paths after mem_debugging_and_hardening_init() has run,
3739 * or when a false negative result is not harmful when called too early.
3740 */
3741static inline bool debug_pagealloc_enabled_static(void)
3742{
3743 if (!IS_ENABLED(CONFIG_DEBUG_PAGEALLOC))
3744 return false;
3745
3746 return static_branch_unlikely(&_debug_pagealloc_enabled);
3747}
3748
3749/*
3750 * To support DEBUG_PAGEALLOC architecture must ensure that
3751 * __kernel_map_pages() never fails
3752 */
3753extern void __kernel_map_pages(struct page *page, int numpages, int enable);
3754#ifdef CONFIG_DEBUG_PAGEALLOC
3755static inline void debug_pagealloc_map_pages(struct page *page, int numpages)
3756{
3757 if (debug_pagealloc_enabled_static())
3758 __kernel_map_pages(page, numpages, enable: 1);
3759}
3760
3761static inline void debug_pagealloc_unmap_pages(struct page *page, int numpages)
3762{
3763 if (debug_pagealloc_enabled_static())
3764 __kernel_map_pages(page, numpages, enable: 0);
3765}
3766
3767extern unsigned int _debug_guardpage_minorder;
3768DECLARE_STATIC_KEY_FALSE(_debug_guardpage_enabled);
3769
3770static inline unsigned int debug_guardpage_minorder(void)
3771{
3772 return _debug_guardpage_minorder;
3773}
3774
3775static inline bool debug_guardpage_enabled(void)
3776{
3777 return static_branch_unlikely(&_debug_guardpage_enabled);
3778}
3779
3780static inline bool page_is_guard(struct page *page)
3781{
3782 if (!debug_guardpage_enabled())
3783 return false;
3784
3785 return PageGuard(page);
3786}
3787
3788bool __set_page_guard(struct zone *zone, struct page *page, unsigned int order,
3789 int migratetype);
3790static inline bool set_page_guard(struct zone *zone, struct page *page,
3791 unsigned int order, int migratetype)
3792{
3793 if (!debug_guardpage_enabled())
3794 return false;
3795 return __set_page_guard(zone, page, order, migratetype);
3796}
3797
3798void __clear_page_guard(struct zone *zone, struct page *page, unsigned int order,
3799 int migratetype);
3800static inline void clear_page_guard(struct zone *zone, struct page *page,
3801 unsigned int order, int migratetype)
3802{
3803 if (!debug_guardpage_enabled())
3804 return;
3805 __clear_page_guard(zone, page, order, migratetype);
3806}
3807
3808#else /* CONFIG_DEBUG_PAGEALLOC */
3809static inline void debug_pagealloc_map_pages(struct page *page, int numpages) {}
3810static inline void debug_pagealloc_unmap_pages(struct page *page, int numpages) {}
3811static inline unsigned int debug_guardpage_minorder(void) { return 0; }
3812static inline bool debug_guardpage_enabled(void) { return false; }
3813static inline bool page_is_guard(struct page *page) { return false; }
3814static inline bool set_page_guard(struct zone *zone, struct page *page,
3815 unsigned int order, int migratetype) { return false; }
3816static inline void clear_page_guard(struct zone *zone, struct page *page,
3817 unsigned int order, int migratetype) {}
3818#endif /* CONFIG_DEBUG_PAGEALLOC */
3819
3820#ifdef __HAVE_ARCH_GATE_AREA
3821extern struct vm_area_struct *get_gate_vma(struct mm_struct *mm);
3822extern int in_gate_area_no_mm(unsigned long addr);
3823extern int in_gate_area(struct mm_struct *mm, unsigned long addr);
3824#else
3825static inline struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
3826{
3827 return NULL;
3828}
3829static inline int in_gate_area_no_mm(unsigned long addr) { return 0; }
3830static inline int in_gate_area(struct mm_struct *mm, unsigned long addr)
3831{
3832 return 0;
3833}
3834#endif /* __HAVE_ARCH_GATE_AREA */
3835
3836extern bool process_shares_mm(struct task_struct *p, struct mm_struct *mm);
3837
3838#ifdef CONFIG_SYSCTL
3839extern int sysctl_drop_caches;
3840int drop_caches_sysctl_handler(struct ctl_table *, int, void *, size_t *,
3841 loff_t *);
3842#endif
3843
3844void drop_slab(void);
3845
3846#ifndef CONFIG_MMU
3847#define randomize_va_space 0
3848#else
3849extern int randomize_va_space;
3850#endif
3851
3852const char * arch_vma_name(struct vm_area_struct *vma);
3853#ifdef CONFIG_MMU
3854void print_vma_addr(char *prefix, unsigned long rip);
3855#else
3856static inline void print_vma_addr(char *prefix, unsigned long rip)
3857{
3858}
3859#endif
3860
3861void *sparse_buffer_alloc(unsigned long size);
3862struct page * __populate_section_memmap(unsigned long pfn,
3863 unsigned long nr_pages, int nid, struct vmem_altmap *altmap,
3864 struct dev_pagemap *pgmap);
3865void pmd_init(void *addr);
3866void pud_init(void *addr);
3867pgd_t *vmemmap_pgd_populate(unsigned long addr, int node);
3868p4d_t *vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node);
3869pud_t *vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node);
3870pmd_t *vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node);
3871pte_t *vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node,
3872 struct vmem_altmap *altmap, struct page *reuse);
3873void *vmemmap_alloc_block(unsigned long size, int node);
3874struct vmem_altmap;
3875void *vmemmap_alloc_block_buf(unsigned long size, int node,
3876 struct vmem_altmap *altmap);
3877void vmemmap_verify(pte_t *, int, unsigned long, unsigned long);
3878void vmemmap_set_pmd(pmd_t *pmd, void *p, int node,
3879 unsigned long addr, unsigned long next);
3880int vmemmap_check_pmd(pmd_t *pmd, int node,
3881 unsigned long addr, unsigned long next);
3882int vmemmap_populate_basepages(unsigned long start, unsigned long end,
3883 int node, struct vmem_altmap *altmap);
3884int vmemmap_populate_hugepages(unsigned long start, unsigned long end,
3885 int node, struct vmem_altmap *altmap);
3886int vmemmap_populate(unsigned long start, unsigned long end, int node,
3887 struct vmem_altmap *altmap);
3888void vmemmap_populate_print_last(void);
3889#ifdef CONFIG_MEMORY_HOTPLUG
3890void vmemmap_free(unsigned long start, unsigned long end,
3891 struct vmem_altmap *altmap);
3892#endif
3893
3894#ifdef CONFIG_SPARSEMEM_VMEMMAP
3895static inline unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
3896{
3897 /* number of pfns from base where pfn_to_page() is valid */
3898 if (altmap)
3899 return altmap->reserve + altmap->free;
3900 return 0;
3901}
3902
3903static inline void vmem_altmap_free(struct vmem_altmap *altmap,
3904 unsigned long nr_pfns)
3905{
3906 altmap->alloc -= nr_pfns;
3907}
3908#else
3909static inline unsigned long vmem_altmap_offset(struct vmem_altmap *altmap)
3910{
3911 return 0;
3912}
3913
3914static inline void vmem_altmap_free(struct vmem_altmap *altmap,
3915 unsigned long nr_pfns)
3916{
3917}
3918#endif
3919
3920#define VMEMMAP_RESERVE_NR 2
3921#ifdef CONFIG_ARCH_WANT_OPTIMIZE_DAX_VMEMMAP
3922static inline bool __vmemmap_can_optimize(struct vmem_altmap *altmap,
3923 struct dev_pagemap *pgmap)
3924{
3925 unsigned long nr_pages;
3926 unsigned long nr_vmemmap_pages;
3927
3928 if (!pgmap || !is_power_of_2(n: sizeof(struct page)))
3929 return false;
3930
3931 nr_pages = pgmap_vmemmap_nr(pgmap);
3932 nr_vmemmap_pages = ((nr_pages * sizeof(struct page)) >> PAGE_SHIFT);
3933 /*
3934 * For vmemmap optimization with DAX we need minimum 2 vmemmap
3935 * pages. See layout diagram in Documentation/mm/vmemmap_dedup.rst
3936 */
3937 return !altmap && (nr_vmemmap_pages > VMEMMAP_RESERVE_NR);
3938}
3939/*
3940 * If we don't have an architecture override, use the generic rule
3941 */
3942#ifndef vmemmap_can_optimize
3943#define vmemmap_can_optimize __vmemmap_can_optimize
3944#endif
3945
3946#else
3947static inline bool vmemmap_can_optimize(struct vmem_altmap *altmap,
3948 struct dev_pagemap *pgmap)
3949{
3950 return false;
3951}
3952#endif
3953
3954void register_page_bootmem_memmap(unsigned long section_nr, struct page *map,
3955 unsigned long nr_pages);
3956
3957enum mf_flags {
3958 MF_COUNT_INCREASED = 1 << 0,
3959 MF_ACTION_REQUIRED = 1 << 1,
3960 MF_MUST_KILL = 1 << 2,
3961 MF_SOFT_OFFLINE = 1 << 3,
3962 MF_UNPOISON = 1 << 4,
3963 MF_SW_SIMULATED = 1 << 5,
3964 MF_NO_RETRY = 1 << 6,
3965 MF_MEM_PRE_REMOVE = 1 << 7,
3966};
3967int mf_dax_kill_procs(struct address_space *mapping, pgoff_t index,
3968 unsigned long count, int mf_flags);
3969extern int memory_failure(unsigned long pfn, int flags);
3970extern void memory_failure_queue_kick(int cpu);
3971extern int unpoison_memory(unsigned long pfn);
3972extern void shake_page(struct page *p);
3973extern atomic_long_t num_poisoned_pages __read_mostly;
3974extern int soft_offline_page(unsigned long pfn, int flags);
3975#ifdef CONFIG_MEMORY_FAILURE
3976/*
3977 * Sysfs entries for memory failure handling statistics.
3978 */
3979extern const struct attribute_group memory_failure_attr_group;
3980extern void memory_failure_queue(unsigned long pfn, int flags);
3981extern int __get_huge_page_for_hwpoison(unsigned long pfn, int flags,
3982 bool *migratable_cleared);
3983void num_poisoned_pages_inc(unsigned long pfn);
3984void num_poisoned_pages_sub(unsigned long pfn, long i);
3985struct task_struct *task_early_kill(struct task_struct *tsk, int force_early);
3986#else
3987static inline void memory_failure_queue(unsigned long pfn, int flags)
3988{
3989}
3990
3991static inline int __get_huge_page_for_hwpoison(unsigned long pfn, int flags,
3992 bool *migratable_cleared)
3993{
3994 return 0;
3995}
3996
3997static inline void num_poisoned_pages_inc(unsigned long pfn)
3998{
3999}
4000
4001static inline void num_poisoned_pages_sub(unsigned long pfn, long i)
4002{
4003}
4004#endif
4005
4006#if defined(CONFIG_MEMORY_FAILURE) && defined(CONFIG_KSM)
4007void add_to_kill_ksm(struct task_struct *tsk, struct page *p,
4008 struct vm_area_struct *vma, struct list_head *to_kill,
4009 unsigned long ksm_addr);
4010#endif
4011
4012#if defined(CONFIG_MEMORY_FAILURE) && defined(CONFIG_MEMORY_HOTPLUG)
4013extern void memblk_nr_poison_inc(unsigned long pfn);
4014extern void memblk_nr_poison_sub(unsigned long pfn, long i);
4015#else
4016static inline void memblk_nr_poison_inc(unsigned long pfn)
4017{
4018}
4019
4020static inline void memblk_nr_poison_sub(unsigned long pfn, long i)
4021{
4022}
4023#endif
4024
4025#ifndef arch_memory_failure
4026static inline int arch_memory_failure(unsigned long pfn, int flags)
4027{
4028 return -ENXIO;
4029}
4030#endif
4031
4032#ifndef arch_is_platform_page
4033static inline bool arch_is_platform_page(u64 paddr)
4034{
4035 return false;
4036}
4037#endif
4038
4039/*
4040 * Error handlers for various types of pages.
4041 */
4042enum mf_result {
4043 MF_IGNORED, /* Error: cannot be handled */
4044 MF_FAILED, /* Error: handling failed */
4045 MF_DELAYED, /* Will be handled later */
4046 MF_RECOVERED, /* Successfully recovered */
4047};
4048
4049enum mf_action_page_type {
4050 MF_MSG_KERNEL,
4051 MF_MSG_KERNEL_HIGH_ORDER,
4052 MF_MSG_SLAB,
4053 MF_MSG_DIFFERENT_COMPOUND,
4054 MF_MSG_HUGE,
4055 MF_MSG_FREE_HUGE,
4056 MF_MSG_UNMAP_FAILED,
4057 MF_MSG_DIRTY_SWAPCACHE,
4058 MF_MSG_CLEAN_SWAPCACHE,
4059 MF_MSG_DIRTY_MLOCKED_LRU,
4060 MF_MSG_CLEAN_MLOCKED_LRU,
4061 MF_MSG_DIRTY_UNEVICTABLE_LRU,
4062 MF_MSG_CLEAN_UNEVICTABLE_LRU,
4063 MF_MSG_DIRTY_LRU,
4064 MF_MSG_CLEAN_LRU,
4065 MF_MSG_TRUNCATED_LRU,
4066 MF_MSG_BUDDY,
4067 MF_MSG_DAX,
4068 MF_MSG_UNSPLIT_THP,
4069 MF_MSG_UNKNOWN,
4070};
4071
4072#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
4073extern void clear_huge_page(struct page *page,
4074 unsigned long addr_hint,
4075 unsigned int pages_per_huge_page);
4076int copy_user_large_folio(struct folio *dst, struct folio *src,
4077 unsigned long addr_hint,
4078 struct vm_area_struct *vma);
4079long copy_folio_from_user(struct folio *dst_folio,
4080 const void __user *usr_src,
4081 bool allow_pagefault);
4082
4083/**
4084 * vma_is_special_huge - Are transhuge page-table entries considered special?
4085 * @vma: Pointer to the struct vm_area_struct to consider
4086 *
4087 * Whether transhuge page-table entries are considered "special" following
4088 * the definition in vm_normal_page().
4089 *
4090 * Return: true if transhuge page-table entries should be considered special,
4091 * false otherwise.
4092 */
4093static inline bool vma_is_special_huge(const struct vm_area_struct *vma)
4094{
4095 return vma_is_dax(vma) || (vma->vm_file &&
4096 (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)));
4097}
4098
4099#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
4100
4101#if MAX_NUMNODES > 1
4102void __init setup_nr_node_ids(void);
4103#else
4104static inline void setup_nr_node_ids(void) {}
4105#endif
4106
4107extern int memcmp_pages(struct page *page1, struct page *page2);
4108
4109static inline int pages_identical(struct page *page1, struct page *page2)
4110{
4111 return !memcmp_pages(page1, page2);
4112}
4113
4114#ifdef CONFIG_MAPPING_DIRTY_HELPERS
4115unsigned long clean_record_shared_mapping_range(struct address_space *mapping,
4116 pgoff_t first_index, pgoff_t nr,
4117 pgoff_t bitmap_pgoff,
4118 unsigned long *bitmap,
4119 pgoff_t *start,
4120 pgoff_t *end);
4121
4122unsigned long wp_shared_mapping_range(struct address_space *mapping,
4123 pgoff_t first_index, pgoff_t nr);
4124#endif
4125
4126extern int sysctl_nr_trim_pages;
4127
4128#ifdef CONFIG_PRINTK
4129void mem_dump_obj(void *object);
4130#else
4131static inline void mem_dump_obj(void *object) {}
4132#endif
4133
4134/**
4135 * seal_check_write - Check for F_SEAL_WRITE or F_SEAL_FUTURE_WRITE flags and
4136 * handle them.
4137 * @seals: the seals to check
4138 * @vma: the vma to operate on
4139 *
4140 * Check whether F_SEAL_WRITE or F_SEAL_FUTURE_WRITE are set; if so, do proper
4141 * check/handling on the vma flags. Return 0 if check pass, or <0 for errors.
4142 */
4143static inline int seal_check_write(int seals, struct vm_area_struct *vma)
4144{
4145 if (seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
4146 /*
4147 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
4148 * write seals are active.
4149 */
4150 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE))
4151 return -EPERM;
4152
4153 /*
4154 * Since an F_SEAL_[FUTURE_]WRITE sealed memfd can be mapped as
4155 * MAP_SHARED and read-only, take care to not allow mprotect to
4156 * revert protections on such mappings. Do this only for shared
4157 * mappings. For private mappings, don't need to mask
4158 * VM_MAYWRITE as we still want them to be COW-writable.
4159 */
4160 if (vma->vm_flags & VM_SHARED)
4161 vm_flags_clear(vma, VM_MAYWRITE);
4162 }
4163
4164 return 0;
4165}
4166
4167#ifdef CONFIG_ANON_VMA_NAME
4168int madvise_set_anon_name(struct mm_struct *mm, unsigned long start,
4169 unsigned long len_in,
4170 struct anon_vma_name *anon_name);
4171#else
4172static inline int
4173madvise_set_anon_name(struct mm_struct *mm, unsigned long start,
4174 unsigned long len_in, struct anon_vma_name *anon_name) {
4175 return 0;
4176}
4177#endif
4178
4179#ifdef CONFIG_UNACCEPTED_MEMORY
4180
4181bool range_contains_unaccepted_memory(phys_addr_t start, phys_addr_t end);
4182void accept_memory(phys_addr_t start, phys_addr_t end);
4183
4184#else
4185
4186static inline bool range_contains_unaccepted_memory(phys_addr_t start,
4187 phys_addr_t end)
4188{
4189 return false;
4190}
4191
4192static inline void accept_memory(phys_addr_t start, phys_addr_t end)
4193{
4194}
4195
4196#endif
4197
4198static inline bool pfn_is_unaccepted_memory(unsigned long pfn)
4199{
4200 phys_addr_t paddr = pfn << PAGE_SHIFT;
4201
4202 return range_contains_unaccepted_memory(start: paddr, end: paddr + PAGE_SIZE);
4203}
4204
4205#endif /* _LINUX_MM_H */
4206

source code of linux/include/linux/mm.h