4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
41 #include <linux/kernel_stat.h>
43 #include <linux/hugetlb.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/highmem.h>
47 #include <linux/pagemap.h>
48 #include <linux/ksm.h>
49 #include <linux/rmap.h>
50 #include <linux/export.h>
51 #include <linux/delayacct.h>
52 #include <linux/init.h>
53 #include <linux/writeback.h>
54 #include <linux/memcontrol.h>
55 #include <linux/mmu_notifier.h>
56 #include <linux/kallsyms.h>
57 #include <linux/swapops.h>
58 #include <linux/elf.h>
59 #include <linux/gfp.h>
62 #include <asm/pgalloc.h>
63 #include <asm/uaccess.h>
65 #include <asm/tlbflush.h>
66 #include <asm/pgtable.h>
70 #ifndef CONFIG_NEED_MULTIPLE_NODES
71 /* use the per-pgdat data instead for discontigmem - mbligh */
72 unsigned long max_mapnr
;
75 EXPORT_SYMBOL(max_mapnr
);
76 EXPORT_SYMBOL(mem_map
);
79 unsigned long num_physpages
;
81 * A number of key systems in x86 including ioremap() rely on the assumption
82 * that high_memory defines the upper bound on direct map memory, then end
83 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
84 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
89 EXPORT_SYMBOL(num_physpages
);
90 EXPORT_SYMBOL(high_memory
);
93 * Randomize the address space (stacks, mmaps, brk, etc.).
95 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
96 * as ancient (libc5 based) binaries can segfault. )
98 int randomize_va_space __read_mostly
=
99 #ifdef CONFIG_COMPAT_BRK
105 static int __init
disable_randmaps(char *s
)
107 randomize_va_space
= 0;
110 __setup("norandmaps", disable_randmaps
);
112 unsigned long zero_pfn __read_mostly
;
113 unsigned long highest_memmap_pfn __read_mostly
;
116 * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
118 static int __init
init_zero_pfn(void)
120 zero_pfn
= page_to_pfn(ZERO_PAGE(0));
123 core_initcall(init_zero_pfn
);
126 #if defined(SPLIT_RSS_COUNTING)
128 void sync_mm_rss(struct mm_struct
*mm
)
132 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
133 if (current
->rss_stat
.count
[i
]) {
134 add_mm_counter(mm
, i
, current
->rss_stat
.count
[i
]);
135 current
->rss_stat
.count
[i
] = 0;
138 current
->rss_stat
.events
= 0;
141 static void add_mm_counter_fast(struct mm_struct
*mm
, int member
, int val
)
143 struct task_struct
*task
= current
;
145 if (likely(task
->mm
== mm
))
146 task
->rss_stat
.count
[member
] += val
;
148 add_mm_counter(mm
, member
, val
);
150 #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
151 #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
153 /* sync counter once per 64 page faults */
154 #define TASK_RSS_EVENTS_THRESH (64)
155 static void check_sync_rss_stat(struct task_struct
*task
)
157 if (unlikely(task
!= current
))
159 if (unlikely(task
->rss_stat
.events
++ > TASK_RSS_EVENTS_THRESH
))
160 sync_mm_rss(task
->mm
);
162 #else /* SPLIT_RSS_COUNTING */
164 #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
165 #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
167 static void check_sync_rss_stat(struct task_struct
*task
)
171 #endif /* SPLIT_RSS_COUNTING */
173 #ifdef HAVE_GENERIC_MMU_GATHER
175 static int tlb_next_batch(struct mmu_gather
*tlb
)
177 struct mmu_gather_batch
*batch
;
181 tlb
->active
= batch
->next
;
185 batch
= (void *)__get_free_pages(GFP_NOWAIT
| __GFP_NOWARN
, 0);
191 batch
->max
= MAX_GATHER_BATCH
;
193 tlb
->active
->next
= batch
;
200 * Called to initialize an (on-stack) mmu_gather structure for page-table
201 * tear-down from @mm. The @fullmm argument is used when @mm is without
202 * users and we're going to destroy the full address space (exit/execve).
204 void tlb_gather_mmu(struct mmu_gather
*tlb
, struct mm_struct
*mm
, bool fullmm
)
208 tlb
->fullmm
= fullmm
;
210 tlb
->fast_mode
= (num_possible_cpus() == 1);
211 tlb
->local
.next
= NULL
;
213 tlb
->local
.max
= ARRAY_SIZE(tlb
->__pages
);
214 tlb
->active
= &tlb
->local
;
216 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
221 void tlb_flush_mmu(struct mmu_gather
*tlb
)
223 struct mmu_gather_batch
*batch
;
225 if (!tlb
->need_flush
)
229 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
230 tlb_table_flush(tlb
);
233 if (tlb_fast_mode(tlb
))
236 for (batch
= &tlb
->local
; batch
; batch
= batch
->next
) {
237 free_pages_and_swap_cache(batch
->pages
, batch
->nr
);
240 tlb
->active
= &tlb
->local
;
244 * Called at the end of the shootdown operation to free up any resources
245 * that were required.
247 void tlb_finish_mmu(struct mmu_gather
*tlb
, unsigned long start
, unsigned long end
)
249 struct mmu_gather_batch
*batch
, *next
;
253 /* keep the page table cache within bounds */
256 for (batch
= tlb
->local
.next
; batch
; batch
= next
) {
258 free_pages((unsigned long)batch
, 0);
260 tlb
->local
.next
= NULL
;
264 * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
265 * handling the additional races in SMP caused by other CPUs caching valid
266 * mappings in their TLBs. Returns the number of free page slots left.
267 * When out of page slots we must call tlb_flush_mmu().
269 int __tlb_remove_page(struct mmu_gather
*tlb
, struct page
*page
)
271 struct mmu_gather_batch
*batch
;
273 VM_BUG_ON(!tlb
->need_flush
);
275 if (tlb_fast_mode(tlb
)) {
276 free_page_and_swap_cache(page
);
277 return 1; /* avoid calling tlb_flush_mmu() */
281 batch
->pages
[batch
->nr
++] = page
;
282 if (batch
->nr
== batch
->max
) {
283 if (!tlb_next_batch(tlb
))
287 VM_BUG_ON(batch
->nr
> batch
->max
);
289 return batch
->max
- batch
->nr
;
292 #endif /* HAVE_GENERIC_MMU_GATHER */
294 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
297 * See the comment near struct mmu_table_batch.
300 static void tlb_remove_table_smp_sync(void *arg
)
302 /* Simply deliver the interrupt */
305 static void tlb_remove_table_one(void *table
)
308 * This isn't an RCU grace period and hence the page-tables cannot be
309 * assumed to be actually RCU-freed.
311 * It is however sufficient for software page-table walkers that rely on
312 * IRQ disabling. See the comment near struct mmu_table_batch.
314 smp_call_function(tlb_remove_table_smp_sync
, NULL
, 1);
315 __tlb_remove_table(table
);
318 static void tlb_remove_table_rcu(struct rcu_head
*head
)
320 struct mmu_table_batch
*batch
;
323 batch
= container_of(head
, struct mmu_table_batch
, rcu
);
325 for (i
= 0; i
< batch
->nr
; i
++)
326 __tlb_remove_table(batch
->tables
[i
]);
328 free_page((unsigned long)batch
);
331 void tlb_table_flush(struct mmu_gather
*tlb
)
333 struct mmu_table_batch
**batch
= &tlb
->batch
;
336 call_rcu_sched(&(*batch
)->rcu
, tlb_remove_table_rcu
);
341 void tlb_remove_table(struct mmu_gather
*tlb
, void *table
)
343 struct mmu_table_batch
**batch
= &tlb
->batch
;
348 * When there's less then two users of this mm there cannot be a
349 * concurrent page-table walk.
351 if (atomic_read(&tlb
->mm
->mm_users
) < 2) {
352 __tlb_remove_table(table
);
356 if (*batch
== NULL
) {
357 *batch
= (struct mmu_table_batch
*)__get_free_page(GFP_NOWAIT
| __GFP_NOWARN
);
358 if (*batch
== NULL
) {
359 tlb_remove_table_one(table
);
364 (*batch
)->tables
[(*batch
)->nr
++] = table
;
365 if ((*batch
)->nr
== MAX_TABLE_BATCH
)
366 tlb_table_flush(tlb
);
369 #endif /* CONFIG_HAVE_RCU_TABLE_FREE */
372 * If a p?d_bad entry is found while walking page tables, report
373 * the error, before resetting entry to p?d_none. Usually (but
374 * very seldom) called out from the p?d_none_or_clear_bad macros.
377 void pgd_clear_bad(pgd_t
*pgd
)
383 void pud_clear_bad(pud_t
*pud
)
389 void pmd_clear_bad(pmd_t
*pmd
)
396 * Note: this doesn't free the actual pages themselves. That
397 * has been handled earlier when unmapping all the memory regions.
399 static void free_pte_range(struct mmu_gather
*tlb
, pmd_t
*pmd
,
402 pgtable_t token
= pmd_pgtable(*pmd
);
404 pte_free_tlb(tlb
, token
, addr
);
408 static inline void free_pmd_range(struct mmu_gather
*tlb
, pud_t
*pud
,
409 unsigned long addr
, unsigned long end
,
410 unsigned long floor
, unsigned long ceiling
)
417 pmd
= pmd_offset(pud
, addr
);
419 next
= pmd_addr_end(addr
, end
);
420 if (pmd_none_or_clear_bad(pmd
))
422 free_pte_range(tlb
, pmd
, addr
);
423 } while (pmd
++, addr
= next
, addr
!= end
);
433 if (end
- 1 > ceiling
- 1)
436 pmd
= pmd_offset(pud
, start
);
438 pmd_free_tlb(tlb
, pmd
, start
);
441 static inline void free_pud_range(struct mmu_gather
*tlb
, pgd_t
*pgd
,
442 unsigned long addr
, unsigned long end
,
443 unsigned long floor
, unsigned long ceiling
)
450 pud
= pud_offset(pgd
, addr
);
452 next
= pud_addr_end(addr
, end
);
453 if (pud_none_or_clear_bad(pud
))
455 free_pmd_range(tlb
, pud
, addr
, next
, floor
, ceiling
);
456 } while (pud
++, addr
= next
, addr
!= end
);
462 ceiling
&= PGDIR_MASK
;
466 if (end
- 1 > ceiling
- 1)
469 pud
= pud_offset(pgd
, start
);
471 pud_free_tlb(tlb
, pud
, start
);
475 * This function frees user-level page tables of a process.
477 * Must be called with pagetable lock held.
479 void free_pgd_range(struct mmu_gather
*tlb
,
480 unsigned long addr
, unsigned long end
,
481 unsigned long floor
, unsigned long ceiling
)
487 * The next few lines have given us lots of grief...
489 * Why are we testing PMD* at this top level? Because often
490 * there will be no work to do at all, and we'd prefer not to
491 * go all the way down to the bottom just to discover that.
493 * Why all these "- 1"s? Because 0 represents both the bottom
494 * of the address space and the top of it (using -1 for the
495 * top wouldn't help much: the masks would do the wrong thing).
496 * The rule is that addr 0 and floor 0 refer to the bottom of
497 * the address space, but end 0 and ceiling 0 refer to the top
498 * Comparisons need to use "end - 1" and "ceiling - 1" (though
499 * that end 0 case should be mythical).
501 * Wherever addr is brought up or ceiling brought down, we must
502 * be careful to reject "the opposite 0" before it confuses the
503 * subsequent tests. But what about where end is brought down
504 * by PMD_SIZE below? no, end can't go down to 0 there.
506 * Whereas we round start (addr) and ceiling down, by different
507 * masks at different levels, in order to test whether a table
508 * now has no other vmas using it, so can be freed, we don't
509 * bother to round floor or end up - the tests don't need that.
523 if (end
- 1 > ceiling
- 1)
528 pgd
= pgd_offset(tlb
->mm
, addr
);
530 next
= pgd_addr_end(addr
, end
);
531 if (pgd_none_or_clear_bad(pgd
))
533 free_pud_range(tlb
, pgd
, addr
, next
, floor
, ceiling
);
534 } while (pgd
++, addr
= next
, addr
!= end
);
537 void free_pgtables(struct mmu_gather
*tlb
, struct vm_area_struct
*vma
,
538 unsigned long floor
, unsigned long ceiling
)
541 struct vm_area_struct
*next
= vma
->vm_next
;
542 unsigned long addr
= vma
->vm_start
;
545 * Hide vma from rmap and truncate_pagecache before freeing
548 unlink_anon_vmas(vma
);
549 unlink_file_vma(vma
);
551 if (is_vm_hugetlb_page(vma
)) {
552 hugetlb_free_pgd_range(tlb
, addr
, vma
->vm_end
,
553 floor
, next
? next
->vm_start
: ceiling
);
556 * Optimization: gather nearby vmas into one call down
558 while (next
&& next
->vm_start
<= vma
->vm_end
+ PMD_SIZE
559 && !is_vm_hugetlb_page(next
)) {
562 unlink_anon_vmas(vma
);
563 unlink_file_vma(vma
);
565 free_pgd_range(tlb
, addr
, vma
->vm_end
,
566 floor
, next
? next
->vm_start
: ceiling
);
572 int __pte_alloc(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
573 pmd_t
*pmd
, unsigned long address
)
575 pgtable_t
new = pte_alloc_one(mm
, address
);
576 int wait_split_huge_page
;
581 * Ensure all pte setup (eg. pte page lock and page clearing) are
582 * visible before the pte is made visible to other CPUs by being
583 * put into page tables.
585 * The other side of the story is the pointer chasing in the page
586 * table walking code (when walking the page table without locking;
587 * ie. most of the time). Fortunately, these data accesses consist
588 * of a chain of data-dependent loads, meaning most CPUs (alpha
589 * being the notable exception) will already guarantee loads are
590 * seen in-order. See the alpha page table accessors for the
591 * smp_read_barrier_depends() barriers in page table walking code.
593 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
595 spin_lock(&mm
->page_table_lock
);
596 wait_split_huge_page
= 0;
597 if (likely(pmd_none(*pmd
))) { /* Has another populated it ? */
599 pmd_populate(mm
, pmd
, new);
601 } else if (unlikely(pmd_trans_splitting(*pmd
)))
602 wait_split_huge_page
= 1;
603 spin_unlock(&mm
->page_table_lock
);
606 if (wait_split_huge_page
)
607 wait_split_huge_page(vma
->anon_vma
, pmd
);
611 int __pte_alloc_kernel(pmd_t
*pmd
, unsigned long address
)
613 pte_t
*new = pte_alloc_one_kernel(&init_mm
, address
);
617 smp_wmb(); /* See comment in __pte_alloc */
619 spin_lock(&init_mm
.page_table_lock
);
620 if (likely(pmd_none(*pmd
))) { /* Has another populated it ? */
621 pmd_populate_kernel(&init_mm
, pmd
, new);
624 VM_BUG_ON(pmd_trans_splitting(*pmd
));
625 spin_unlock(&init_mm
.page_table_lock
);
627 pte_free_kernel(&init_mm
, new);
631 static inline void init_rss_vec(int *rss
)
633 memset(rss
, 0, sizeof(int) * NR_MM_COUNTERS
);
636 static inline void add_mm_rss_vec(struct mm_struct
*mm
, int *rss
)
640 if (current
->mm
== mm
)
642 for (i
= 0; i
< NR_MM_COUNTERS
; i
++)
644 add_mm_counter(mm
, i
, rss
[i
]);
648 * This function is called to print an error when a bad pte
649 * is found. For example, we might have a PFN-mapped pte in
650 * a region that doesn't allow it.
652 * The calling function must still handle the error.
654 static void print_bad_pte(struct vm_area_struct
*vma
, unsigned long addr
,
655 pte_t pte
, struct page
*page
)
657 pgd_t
*pgd
= pgd_offset(vma
->vm_mm
, addr
);
658 pud_t
*pud
= pud_offset(pgd
, addr
);
659 pmd_t
*pmd
= pmd_offset(pud
, addr
);
660 struct address_space
*mapping
;
662 static unsigned long resume
;
663 static unsigned long nr_shown
;
664 static unsigned long nr_unshown
;
667 * Allow a burst of 60 reports, then keep quiet for that minute;
668 * or allow a steady drip of one report per second.
670 if (nr_shown
== 60) {
671 if (time_before(jiffies
, resume
)) {
677 "BUG: Bad page map: %lu messages suppressed\n",
684 resume
= jiffies
+ 60 * HZ
;
686 mapping
= vma
->vm_file
? vma
->vm_file
->f_mapping
: NULL
;
687 index
= linear_page_index(vma
, addr
);
690 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
692 (long long)pte_val(pte
), (long long)pmd_val(*pmd
));
696 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
697 (void *)addr
, vma
->vm_flags
, vma
->anon_vma
, mapping
, index
);
699 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
702 print_symbol(KERN_ALERT
"vma->vm_ops->fault: %s\n",
703 (unsigned long)vma
->vm_ops
->fault
);
704 if (vma
->vm_file
&& vma
->vm_file
->f_op
)
705 print_symbol(KERN_ALERT
"vma->vm_file->f_op->mmap: %s\n",
706 (unsigned long)vma
->vm_file
->f_op
->mmap
);
708 add_taint(TAINT_BAD_PAGE
);
711 static inline int is_cow_mapping(vm_flags_t flags
)
713 return (flags
& (VM_SHARED
| VM_MAYWRITE
)) == VM_MAYWRITE
;
717 static inline int is_zero_pfn(unsigned long pfn
)
719 return pfn
== zero_pfn
;
724 static inline unsigned long my_zero_pfn(unsigned long addr
)
731 * vm_normal_page -- This function gets the "struct page" associated with a pte.
733 * "Special" mappings do not wish to be associated with a "struct page" (either
734 * it doesn't exist, or it exists but they don't want to touch it). In this
735 * case, NULL is returned here. "Normal" mappings do have a struct page.
737 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
738 * pte bit, in which case this function is trivial. Secondly, an architecture
739 * may not have a spare pte bit, which requires a more complicated scheme,
742 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
743 * special mapping (even if there are underlying and valid "struct pages").
744 * COWed pages of a VM_PFNMAP are always normal.
746 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
747 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
748 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
749 * mapping will always honor the rule
751 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
753 * And for normal mappings this is false.
755 * This restricts such mappings to be a linear translation from virtual address
756 * to pfn. To get around this restriction, we allow arbitrary mappings so long
757 * as the vma is not a COW mapping; in that case, we know that all ptes are
758 * special (because none can have been COWed).
761 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
763 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
764 * page" backing, however the difference is that _all_ pages with a struct
765 * page (that is, those where pfn_valid is true) are refcounted and considered
766 * normal pages by the VM. The disadvantage is that pages are refcounted
767 * (which can be slower and simply not an option for some PFNMAP users). The
768 * advantage is that we don't have to follow the strict linearity rule of
769 * PFNMAP mappings in order to support COWable mappings.
772 #ifdef __HAVE_ARCH_PTE_SPECIAL
773 # define HAVE_PTE_SPECIAL 1
775 # define HAVE_PTE_SPECIAL 0
777 struct page
*vm_normal_page(struct vm_area_struct
*vma
, unsigned long addr
,
780 unsigned long pfn
= pte_pfn(pte
);
782 if (HAVE_PTE_SPECIAL
) {
783 if (likely(!pte_special(pte
)))
785 if (vma
->vm_flags
& (VM_PFNMAP
| VM_MIXEDMAP
))
787 if (!is_zero_pfn(pfn
))
788 print_bad_pte(vma
, addr
, pte
, NULL
);
792 /* !HAVE_PTE_SPECIAL case follows: */
794 if (unlikely(vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
))) {
795 if (vma
->vm_flags
& VM_MIXEDMAP
) {
801 off
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
802 if (pfn
== vma
->vm_pgoff
+ off
)
804 if (!is_cow_mapping(vma
->vm_flags
))
809 if (is_zero_pfn(pfn
))
812 if (unlikely(pfn
> highest_memmap_pfn
)) {
813 print_bad_pte(vma
, addr
, pte
, NULL
);
818 * NOTE! We still have PageReserved() pages in the page tables.
819 * eg. VDSO mappings can cause them to exist.
822 return pfn_to_page(pfn
);
824 EXPORT_SYMBOL(vm_normal_page
);
827 * copy one vm_area from one task to the other. Assumes the page tables
828 * already present in the new task to be cleared in the whole range
829 * covered by this vma.
832 static inline unsigned long
833 copy_one_pte(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
834 pte_t
*dst_pte
, pte_t
*src_pte
, struct vm_area_struct
*vma
,
835 unsigned long addr
, int *rss
)
837 unsigned long vm_flags
= vma
->vm_flags
;
838 pte_t pte
= *src_pte
;
841 /* pte contains position in swap or file, so copy. */
842 if (unlikely(!pte_present(pte
))) {
843 if (!pte_file(pte
)) {
844 swp_entry_t entry
= pte_to_swp_entry(pte
);
846 if (swap_duplicate(entry
) < 0)
849 /* make sure dst_mm is on swapoff's mmlist. */
850 if (unlikely(list_empty(&dst_mm
->mmlist
))) {
851 spin_lock(&mmlist_lock
);
852 if (list_empty(&dst_mm
->mmlist
))
853 list_add(&dst_mm
->mmlist
,
855 spin_unlock(&mmlist_lock
);
857 if (likely(!non_swap_entry(entry
)))
859 else if (is_migration_entry(entry
)) {
860 page
= migration_entry_to_page(entry
);
867 if (is_write_migration_entry(entry
) &&
868 is_cow_mapping(vm_flags
)) {
870 * COW mappings require pages in both
871 * parent and child to be set to read.
873 make_migration_entry_read(&entry
);
874 pte
= swp_entry_to_pte(entry
);
875 set_pte_at(src_mm
, addr
, src_pte
, pte
);
883 * If it's a COW mapping, write protect it both
884 * in the parent and the child
886 if (is_cow_mapping(vm_flags
)) {
887 ptep_set_wrprotect(src_mm
, addr
, src_pte
);
888 pte
= pte_wrprotect(pte
);
892 * If it's a shared mapping, mark it clean in
895 if (vm_flags
& VM_SHARED
)
896 pte
= pte_mkclean(pte
);
897 pte
= pte_mkold(pte
);
899 page
= vm_normal_page(vma
, addr
, pte
);
910 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
914 int copy_pte_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
915 pmd_t
*dst_pmd
, pmd_t
*src_pmd
, struct vm_area_struct
*vma
,
916 unsigned long addr
, unsigned long end
)
918 pte_t
*orig_src_pte
, *orig_dst_pte
;
919 pte_t
*src_pte
, *dst_pte
;
920 spinlock_t
*src_ptl
, *dst_ptl
;
922 int rss
[NR_MM_COUNTERS
];
923 swp_entry_t entry
= (swp_entry_t
){0};
928 dst_pte
= pte_alloc_map_lock(dst_mm
, dst_pmd
, addr
, &dst_ptl
);
931 src_pte
= pte_offset_map(src_pmd
, addr
);
932 src_ptl
= pte_lockptr(src_mm
, src_pmd
);
933 spin_lock_nested(src_ptl
, SINGLE_DEPTH_NESTING
);
934 orig_src_pte
= src_pte
;
935 orig_dst_pte
= dst_pte
;
936 arch_enter_lazy_mmu_mode();
940 * We are holding two locks at this point - either of them
941 * could generate latencies in another task on another CPU.
943 if (progress
>= 32) {
945 if (need_resched() ||
946 spin_needbreak(src_ptl
) || spin_needbreak(dst_ptl
))
949 if (pte_none(*src_pte
)) {
953 entry
.val
= copy_one_pte(dst_mm
, src_mm
, dst_pte
, src_pte
,
958 } while (dst_pte
++, src_pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
960 arch_leave_lazy_mmu_mode();
961 spin_unlock(src_ptl
);
962 pte_unmap(orig_src_pte
);
963 add_mm_rss_vec(dst_mm
, rss
);
964 pte_unmap_unlock(orig_dst_pte
, dst_ptl
);
968 if (add_swap_count_continuation(entry
, GFP_KERNEL
) < 0)
977 static inline int copy_pmd_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
978 pud_t
*dst_pud
, pud_t
*src_pud
, struct vm_area_struct
*vma
,
979 unsigned long addr
, unsigned long end
)
981 pmd_t
*src_pmd
, *dst_pmd
;
984 dst_pmd
= pmd_alloc(dst_mm
, dst_pud
, addr
);
987 src_pmd
= pmd_offset(src_pud
, addr
);
989 next
= pmd_addr_end(addr
, end
);
990 if (pmd_trans_huge(*src_pmd
)) {
992 VM_BUG_ON(next
-addr
!= HPAGE_PMD_SIZE
);
993 err
= copy_huge_pmd(dst_mm
, src_mm
,
994 dst_pmd
, src_pmd
, addr
, vma
);
1001 if (pmd_none_or_clear_bad(src_pmd
))
1003 if (copy_pte_range(dst_mm
, src_mm
, dst_pmd
, src_pmd
,
1006 } while (dst_pmd
++, src_pmd
++, addr
= next
, addr
!= end
);
1010 static inline int copy_pud_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
1011 pgd_t
*dst_pgd
, pgd_t
*src_pgd
, struct vm_area_struct
*vma
,
1012 unsigned long addr
, unsigned long end
)
1014 pud_t
*src_pud
, *dst_pud
;
1017 dst_pud
= pud_alloc(dst_mm
, dst_pgd
, addr
);
1020 src_pud
= pud_offset(src_pgd
, addr
);
1022 next
= pud_addr_end(addr
, end
);
1023 if (pud_none_or_clear_bad(src_pud
))
1025 if (copy_pmd_range(dst_mm
, src_mm
, dst_pud
, src_pud
,
1028 } while (dst_pud
++, src_pud
++, addr
= next
, addr
!= end
);
1032 int copy_page_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
1033 struct vm_area_struct
*vma
)
1035 pgd_t
*src_pgd
, *dst_pgd
;
1037 unsigned long addr
= vma
->vm_start
;
1038 unsigned long end
= vma
->vm_end
;
1042 * Don't copy ptes where a page fault will fill them correctly.
1043 * Fork becomes much lighter when there are big shared or private
1044 * readonly mappings. The tradeoff is that copy_page_range is more
1045 * efficient than faulting.
1047 if (!(vma
->vm_flags
& (VM_HUGETLB
|VM_NONLINEAR
|VM_PFNMAP
|VM_INSERTPAGE
))) {
1052 if (is_vm_hugetlb_page(vma
))
1053 return copy_hugetlb_page_range(dst_mm
, src_mm
, vma
);
1055 if (unlikely(is_pfn_mapping(vma
))) {
1057 * We do not free on error cases below as remove_vma
1058 * gets called on error from higher level routine
1060 ret
= track_pfn_vma_copy(vma
);
1066 * We need to invalidate the secondary MMU mappings only when
1067 * there could be a permission downgrade on the ptes of the
1068 * parent mm. And a permission downgrade will only happen if
1069 * is_cow_mapping() returns true.
1071 if (is_cow_mapping(vma
->vm_flags
))
1072 mmu_notifier_invalidate_range_start(src_mm
, addr
, end
);
1075 dst_pgd
= pgd_offset(dst_mm
, addr
);
1076 src_pgd
= pgd_offset(src_mm
, addr
);
1078 next
= pgd_addr_end(addr
, end
);
1079 if (pgd_none_or_clear_bad(src_pgd
))
1081 if (unlikely(copy_pud_range(dst_mm
, src_mm
, dst_pgd
, src_pgd
,
1082 vma
, addr
, next
))) {
1086 } while (dst_pgd
++, src_pgd
++, addr
= next
, addr
!= end
);
1088 if (is_cow_mapping(vma
->vm_flags
))
1089 mmu_notifier_invalidate_range_end(src_mm
,
1090 vma
->vm_start
, end
);
1094 static unsigned long zap_pte_range(struct mmu_gather
*tlb
,
1095 struct vm_area_struct
*vma
, pmd_t
*pmd
,
1096 unsigned long addr
, unsigned long end
,
1097 struct zap_details
*details
)
1099 struct mm_struct
*mm
= tlb
->mm
;
1100 int force_flush
= 0;
1101 int rss
[NR_MM_COUNTERS
];
1108 start_pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
1110 arch_enter_lazy_mmu_mode();
1113 if (pte_none(ptent
)) {
1117 if (pte_present(ptent
)) {
1120 page
= vm_normal_page(vma
, addr
, ptent
);
1121 if (unlikely(details
) && page
) {
1123 * unmap_shared_mapping_pages() wants to
1124 * invalidate cache without truncating:
1125 * unmap shared but keep private pages.
1127 if (details
->check_mapping
&&
1128 details
->check_mapping
!= page
->mapping
)
1131 * Each page->index must be checked when
1132 * invalidating or truncating nonlinear.
1134 if (details
->nonlinear_vma
&&
1135 (page
->index
< details
->first_index
||
1136 page
->index
> details
->last_index
))
1139 ptent
= ptep_get_and_clear_full(mm
, addr
, pte
,
1141 tlb_remove_tlb_entry(tlb
, pte
, addr
);
1142 if (unlikely(!page
))
1144 if (unlikely(details
) && details
->nonlinear_vma
1145 && linear_page_index(details
->nonlinear_vma
,
1146 addr
) != page
->index
)
1147 set_pte_at(mm
, addr
, pte
,
1148 pgoff_to_pte(page
->index
));
1150 rss
[MM_ANONPAGES
]--;
1152 if (pte_dirty(ptent
))
1153 set_page_dirty(page
);
1154 if (pte_young(ptent
) &&
1155 likely(!VM_SequentialReadHint(vma
)))
1156 mark_page_accessed(page
);
1157 rss
[MM_FILEPAGES
]--;
1159 page_remove_rmap(page
);
1160 if (unlikely(page_mapcount(page
) < 0))
1161 print_bad_pte(vma
, addr
, ptent
, page
);
1162 force_flush
= !__tlb_remove_page(tlb
, page
);
1168 * If details->check_mapping, we leave swap entries;
1169 * if details->nonlinear_vma, we leave file entries.
1171 if (unlikely(details
))
1173 if (pte_file(ptent
)) {
1174 if (unlikely(!(vma
->vm_flags
& VM_NONLINEAR
)))
1175 print_bad_pte(vma
, addr
, ptent
, NULL
);
1177 swp_entry_t entry
= pte_to_swp_entry(ptent
);
1179 if (!non_swap_entry(entry
))
1181 else if (is_migration_entry(entry
)) {
1184 page
= migration_entry_to_page(entry
);
1187 rss
[MM_ANONPAGES
]--;
1189 rss
[MM_FILEPAGES
]--;
1191 if (unlikely(!free_swap_and_cache(entry
)))
1192 print_bad_pte(vma
, addr
, ptent
, NULL
);
1194 pte_clear_not_present_full(mm
, addr
, pte
, tlb
->fullmm
);
1195 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1197 add_mm_rss_vec(mm
, rss
);
1198 arch_leave_lazy_mmu_mode();
1199 pte_unmap_unlock(start_pte
, ptl
);
1202 * mmu_gather ran out of room to batch pages, we break out of
1203 * the PTE lock to avoid doing the potential expensive TLB invalidate
1204 * and page-free while holding it.
1216 static inline unsigned long zap_pmd_range(struct mmu_gather
*tlb
,
1217 struct vm_area_struct
*vma
, pud_t
*pud
,
1218 unsigned long addr
, unsigned long end
,
1219 struct zap_details
*details
)
1224 pmd
= pmd_offset(pud
, addr
);
1226 next
= pmd_addr_end(addr
, end
);
1227 if (pmd_trans_huge(*pmd
)) {
1228 if (next
- addr
!= HPAGE_PMD_SIZE
) {
1229 #ifdef CONFIG_DEBUG_VM
1230 if (!rwsem_is_locked(&tlb
->mm
->mmap_sem
)) {
1231 pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n",
1232 __func__
, addr
, end
,
1238 split_huge_page_pmd(vma
->vm_mm
, pmd
);
1239 } else if (zap_huge_pmd(tlb
, vma
, pmd
, addr
))
1244 * Here there can be other concurrent MADV_DONTNEED or
1245 * trans huge page faults running, and if the pmd is
1246 * none or trans huge it can change under us. This is
1247 * because MADV_DONTNEED holds the mmap_sem in read
1250 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
1252 next
= zap_pte_range(tlb
, vma
, pmd
, addr
, next
, details
);
1255 } while (pmd
++, addr
= next
, addr
!= end
);
1260 static inline unsigned long zap_pud_range(struct mmu_gather
*tlb
,
1261 struct vm_area_struct
*vma
, pgd_t
*pgd
,
1262 unsigned long addr
, unsigned long end
,
1263 struct zap_details
*details
)
1268 pud
= pud_offset(pgd
, addr
);
1270 next
= pud_addr_end(addr
, end
);
1271 if (pud_none_or_clear_bad(pud
))
1273 next
= zap_pmd_range(tlb
, vma
, pud
, addr
, next
, details
);
1274 } while (pud
++, addr
= next
, addr
!= end
);
1279 static void unmap_page_range(struct mmu_gather
*tlb
,
1280 struct vm_area_struct
*vma
,
1281 unsigned long addr
, unsigned long end
,
1282 struct zap_details
*details
)
1287 if (details
&& !details
->check_mapping
&& !details
->nonlinear_vma
)
1290 BUG_ON(addr
>= end
);
1291 mem_cgroup_uncharge_start();
1292 tlb_start_vma(tlb
, vma
);
1293 pgd
= pgd_offset(vma
->vm_mm
, addr
);
1295 next
= pgd_addr_end(addr
, end
);
1296 if (pgd_none_or_clear_bad(pgd
))
1298 next
= zap_pud_range(tlb
, vma
, pgd
, addr
, next
, details
);
1299 } while (pgd
++, addr
= next
, addr
!= end
);
1300 tlb_end_vma(tlb
, vma
);
1301 mem_cgroup_uncharge_end();
1305 static void unmap_single_vma(struct mmu_gather
*tlb
,
1306 struct vm_area_struct
*vma
, unsigned long start_addr
,
1307 unsigned long end_addr
,
1308 struct zap_details
*details
)
1310 unsigned long start
= max(vma
->vm_start
, start_addr
);
1313 if (start
>= vma
->vm_end
)
1315 end
= min(vma
->vm_end
, end_addr
);
1316 if (end
<= vma
->vm_start
)
1320 uprobe_munmap(vma
, start
, end
);
1322 if (unlikely(is_pfn_mapping(vma
)))
1323 untrack_pfn_vma(vma
, 0, 0);
1326 if (unlikely(is_vm_hugetlb_page(vma
))) {
1328 * It is undesirable to test vma->vm_file as it
1329 * should be non-null for valid hugetlb area.
1330 * However, vm_file will be NULL in the error
1331 * cleanup path of do_mmap_pgoff. When
1332 * hugetlbfs ->mmap method fails,
1333 * do_mmap_pgoff() nullifies vma->vm_file
1334 * before calling this function to clean up.
1335 * Since no pte has actually been setup, it is
1336 * safe to do nothing in this case.
1339 unmap_hugepage_range(vma
, start
, end
, NULL
);
1341 unmap_page_range(tlb
, vma
, start
, end
, details
);
1346 * unmap_vmas - unmap a range of memory covered by a list of vma's
1347 * @tlb: address of the caller's struct mmu_gather
1348 * @vma: the starting vma
1349 * @start_addr: virtual address at which to start unmapping
1350 * @end_addr: virtual address at which to end unmapping
1352 * Unmap all pages in the vma list.
1354 * Only addresses between `start' and `end' will be unmapped.
1356 * The VMA list must be sorted in ascending virtual address order.
1358 * unmap_vmas() assumes that the caller will flush the whole unmapped address
1359 * range after unmap_vmas() returns. So the only responsibility here is to
1360 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1361 * drops the lock and schedules.
1363 void unmap_vmas(struct mmu_gather
*tlb
,
1364 struct vm_area_struct
*vma
, unsigned long start_addr
,
1365 unsigned long end_addr
)
1367 struct mm_struct
*mm
= vma
->vm_mm
;
1369 mmu_notifier_invalidate_range_start(mm
, start_addr
, end_addr
);
1370 for ( ; vma
&& vma
->vm_start
< end_addr
; vma
= vma
->vm_next
)
1371 unmap_single_vma(tlb
, vma
, start_addr
, end_addr
, NULL
);
1372 mmu_notifier_invalidate_range_end(mm
, start_addr
, end_addr
);
1376 * zap_page_range - remove user pages in a given range
1377 * @vma: vm_area_struct holding the applicable pages
1378 * @start: starting address of pages to zap
1379 * @size: number of bytes to zap
1380 * @details: details of nonlinear truncation or shared cache invalidation
1382 * Caller must protect the VMA list
1384 void zap_page_range(struct vm_area_struct
*vma
, unsigned long start
,
1385 unsigned long size
, struct zap_details
*details
)
1387 struct mm_struct
*mm
= vma
->vm_mm
;
1388 struct mmu_gather tlb
;
1389 unsigned long end
= start
+ size
;
1392 tlb_gather_mmu(&tlb
, mm
, 0);
1393 update_hiwater_rss(mm
);
1394 mmu_notifier_invalidate_range_start(mm
, start
, end
);
1395 for ( ; vma
&& vma
->vm_start
< end
; vma
= vma
->vm_next
)
1396 unmap_single_vma(&tlb
, vma
, start
, end
, details
);
1397 mmu_notifier_invalidate_range_end(mm
, start
, end
);
1398 tlb_finish_mmu(&tlb
, start
, end
);
1402 * zap_page_range_single - remove user pages in a given range
1403 * @vma: vm_area_struct holding the applicable pages
1404 * @address: starting address of pages to zap
1405 * @size: number of bytes to zap
1406 * @details: details of nonlinear truncation or shared cache invalidation
1408 * The range must fit into one VMA.
1410 static void zap_page_range_single(struct vm_area_struct
*vma
, unsigned long address
,
1411 unsigned long size
, struct zap_details
*details
)
1413 struct mm_struct
*mm
= vma
->vm_mm
;
1414 struct mmu_gather tlb
;
1415 unsigned long end
= address
+ size
;
1418 tlb_gather_mmu(&tlb
, mm
, 0);
1419 update_hiwater_rss(mm
);
1420 mmu_notifier_invalidate_range_start(mm
, address
, end
);
1421 unmap_single_vma(&tlb
, vma
, address
, end
, details
);
1422 mmu_notifier_invalidate_range_end(mm
, address
, end
);
1423 tlb_finish_mmu(&tlb
, address
, end
);
1427 * zap_vma_ptes - remove ptes mapping the vma
1428 * @vma: vm_area_struct holding ptes to be zapped
1429 * @address: starting address of pages to zap
1430 * @size: number of bytes to zap
1432 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1434 * The entire address range must be fully contained within the vma.
1436 * Returns 0 if successful.
1438 int zap_vma_ptes(struct vm_area_struct
*vma
, unsigned long address
,
1441 if (address
< vma
->vm_start
|| address
+ size
> vma
->vm_end
||
1442 !(vma
->vm_flags
& VM_PFNMAP
))
1444 zap_page_range_single(vma
, address
, size
, NULL
);
1447 EXPORT_SYMBOL_GPL(zap_vma_ptes
);
1450 * follow_page - look up a page descriptor from a user-virtual address
1451 * @vma: vm_area_struct mapping @address
1452 * @address: virtual address to look up
1453 * @flags: flags modifying lookup behaviour
1455 * @flags can have FOLL_ flags set, defined in <linux/mm.h>
1457 * Returns the mapped (struct page *), %NULL if no mapping exists, or
1458 * an error pointer if there is a mapping to something not represented
1459 * by a page descriptor (see also vm_normal_page()).
1461 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
1470 struct mm_struct
*mm
= vma
->vm_mm
;
1472 page
= follow_huge_addr(mm
, address
, flags
& FOLL_WRITE
);
1473 if (!IS_ERR(page
)) {
1474 BUG_ON(flags
& FOLL_GET
);
1479 pgd
= pgd_offset(mm
, address
);
1480 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
1483 pud
= pud_offset(pgd
, address
);
1486 if (pud_huge(*pud
) && vma
->vm_flags
& VM_HUGETLB
) {
1487 BUG_ON(flags
& FOLL_GET
);
1488 page
= follow_huge_pud(mm
, address
, pud
, flags
& FOLL_WRITE
);
1491 if (unlikely(pud_bad(*pud
)))
1494 pmd
= pmd_offset(pud
, address
);
1497 if (pmd_huge(*pmd
) && vma
->vm_flags
& VM_HUGETLB
) {
1498 BUG_ON(flags
& FOLL_GET
);
1499 page
= follow_huge_pmd(mm
, address
, pmd
, flags
& FOLL_WRITE
);
1502 if (pmd_trans_huge(*pmd
)) {
1503 if (flags
& FOLL_SPLIT
) {
1504 split_huge_page_pmd(mm
, pmd
);
1505 goto split_fallthrough
;
1507 spin_lock(&mm
->page_table_lock
);
1508 if (likely(pmd_trans_huge(*pmd
))) {
1509 if (unlikely(pmd_trans_splitting(*pmd
))) {
1510 spin_unlock(&mm
->page_table_lock
);
1511 wait_split_huge_page(vma
->anon_vma
, pmd
);
1513 page
= follow_trans_huge_pmd(mm
, address
,
1515 spin_unlock(&mm
->page_table_lock
);
1519 spin_unlock(&mm
->page_table_lock
);
1523 if (unlikely(pmd_bad(*pmd
)))
1526 ptep
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1529 if (!pte_present(pte
))
1531 if ((flags
& FOLL_WRITE
) && !pte_write(pte
))
1534 page
= vm_normal_page(vma
, address
, pte
);
1535 if (unlikely(!page
)) {
1536 if ((flags
& FOLL_DUMP
) ||
1537 !is_zero_pfn(pte_pfn(pte
)))
1539 page
= pte_page(pte
);
1542 if (flags
& FOLL_GET
)
1543 get_page_foll(page
);
1544 if (flags
& FOLL_TOUCH
) {
1545 if ((flags
& FOLL_WRITE
) &&
1546 !pte_dirty(pte
) && !PageDirty(page
))
1547 set_page_dirty(page
);
1549 * pte_mkyoung() would be more correct here, but atomic care
1550 * is needed to avoid losing the dirty bit: it is easier to use
1551 * mark_page_accessed().
1553 mark_page_accessed(page
);
1555 if ((flags
& FOLL_MLOCK
) && (vma
->vm_flags
& VM_LOCKED
)) {
1557 * The preliminary mapping check is mainly to avoid the
1558 * pointless overhead of lock_page on the ZERO_PAGE
1559 * which might bounce very badly if there is contention.
1561 * If the page is already locked, we don't need to
1562 * handle it now - vmscan will handle it later if and
1563 * when it attempts to reclaim the page.
1565 if (page
->mapping
&& trylock_page(page
)) {
1566 lru_add_drain(); /* push cached pages to LRU */
1568 * Because we lock page here and migration is
1569 * blocked by the pte's page reference, we need
1570 * only check for file-cache page truncation.
1573 mlock_vma_page(page
);
1578 pte_unmap_unlock(ptep
, ptl
);
1583 pte_unmap_unlock(ptep
, ptl
);
1584 return ERR_PTR(-EFAULT
);
1587 pte_unmap_unlock(ptep
, ptl
);
1593 * When core dumping an enormous anonymous area that nobody
1594 * has touched so far, we don't want to allocate unnecessary pages or
1595 * page tables. Return error instead of NULL to skip handle_mm_fault,
1596 * then get_dump_page() will return NULL to leave a hole in the dump.
1597 * But we can only make this optimization where a hole would surely
1598 * be zero-filled if handle_mm_fault() actually did handle it.
1600 if ((flags
& FOLL_DUMP
) &&
1601 (!vma
->vm_ops
|| !vma
->vm_ops
->fault
))
1602 return ERR_PTR(-EFAULT
);
1605 EXPORT_SYMBOL(follow_page
);
1607 static inline int stack_guard_page(struct vm_area_struct
*vma
, unsigned long addr
)
1609 return stack_guard_page_start(vma
, addr
) ||
1610 stack_guard_page_end(vma
, addr
+PAGE_SIZE
);
1614 * __get_user_pages() - pin user pages in memory
1615 * @tsk: task_struct of target task
1616 * @mm: mm_struct of target mm
1617 * @start: starting user address
1618 * @nr_pages: number of pages from start to pin
1619 * @gup_flags: flags modifying pin behaviour
1620 * @pages: array that receives pointers to the pages pinned.
1621 * Should be at least nr_pages long. Or NULL, if caller
1622 * only intends to ensure the pages are faulted in.
1623 * @vmas: array of pointers to vmas corresponding to each page.
1624 * Or NULL if the caller does not require them.
1625 * @nonblocking: whether waiting for disk IO or mmap_sem contention
1627 * Returns number of pages pinned. This may be fewer than the number
1628 * requested. If nr_pages is 0 or negative, returns 0. If no pages
1629 * were pinned, returns -errno. Each page returned must be released
1630 * with a put_page() call when it is finished with. vmas will only
1631 * remain valid while mmap_sem is held.
1633 * Must be called with mmap_sem held for read or write.
1635 * __get_user_pages walks a process's page tables and takes a reference to
1636 * each struct page that each user address corresponds to at a given
1637 * instant. That is, it takes the page that would be accessed if a user
1638 * thread accesses the given user virtual address at that instant.
1640 * This does not guarantee that the page exists in the user mappings when
1641 * __get_user_pages returns, and there may even be a completely different
1642 * page there in some cases (eg. if mmapped pagecache has been invalidated
1643 * and subsequently re faulted). However it does guarantee that the page
1644 * won't be freed completely. And mostly callers simply care that the page
1645 * contains data that was valid *at some point in time*. Typically, an IO
1646 * or similar operation cannot guarantee anything stronger anyway because
1647 * locks can't be held over the syscall boundary.
1649 * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If
1650 * the page is written to, set_page_dirty (or set_page_dirty_lock, as
1651 * appropriate) must be called after the page is finished with, and
1652 * before put_page is called.
1654 * If @nonblocking != NULL, __get_user_pages will not wait for disk IO
1655 * or mmap_sem contention, and if waiting is needed to pin all pages,
1656 * *@nonblocking will be set to 0.
1658 * In most cases, get_user_pages or get_user_pages_fast should be used
1659 * instead of __get_user_pages. __get_user_pages should be used only if
1660 * you need some special @gup_flags.
1662 int __get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
1663 unsigned long start
, int nr_pages
, unsigned int gup_flags
,
1664 struct page
**pages
, struct vm_area_struct
**vmas
,
1668 unsigned long vm_flags
;
1673 VM_BUG_ON(!!pages
!= !!(gup_flags
& FOLL_GET
));
1676 * Require read or write permissions.
1677 * If FOLL_FORCE is set, we only require the "MAY" flags.
1679 vm_flags
= (gup_flags
& FOLL_WRITE
) ?
1680 (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
1681 vm_flags
&= (gup_flags
& FOLL_FORCE
) ?
1682 (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
1686 struct vm_area_struct
*vma
;
1688 vma
= find_extend_vma(mm
, start
);
1689 if (!vma
&& in_gate_area(mm
, start
)) {
1690 unsigned long pg
= start
& PAGE_MASK
;
1696 /* user gate pages are read-only */
1697 if (gup_flags
& FOLL_WRITE
)
1698 return i
? : -EFAULT
;
1700 pgd
= pgd_offset_k(pg
);
1702 pgd
= pgd_offset_gate(mm
, pg
);
1703 BUG_ON(pgd_none(*pgd
));
1704 pud
= pud_offset(pgd
, pg
);
1705 BUG_ON(pud_none(*pud
));
1706 pmd
= pmd_offset(pud
, pg
);
1708 return i
? : -EFAULT
;
1709 VM_BUG_ON(pmd_trans_huge(*pmd
));
1710 pte
= pte_offset_map(pmd
, pg
);
1711 if (pte_none(*pte
)) {
1713 return i
? : -EFAULT
;
1715 vma
= get_gate_vma(mm
);
1719 page
= vm_normal_page(vma
, start
, *pte
);
1721 if (!(gup_flags
& FOLL_DUMP
) &&
1722 is_zero_pfn(pte_pfn(*pte
)))
1723 page
= pte_page(*pte
);
1726 return i
? : -EFAULT
;
1737 (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) ||
1738 !(vm_flags
& vma
->vm_flags
))
1739 return i
? : -EFAULT
;
1741 if (is_vm_hugetlb_page(vma
)) {
1742 i
= follow_hugetlb_page(mm
, vma
, pages
, vmas
,
1743 &start
, &nr_pages
, i
, gup_flags
);
1749 unsigned int foll_flags
= gup_flags
;
1752 * If we have a pending SIGKILL, don't keep faulting
1753 * pages and potentially allocating memory.
1755 if (unlikely(fatal_signal_pending(current
)))
1756 return i
? i
: -ERESTARTSYS
;
1759 while (!(page
= follow_page(vma
, start
, foll_flags
))) {
1761 unsigned int fault_flags
= 0;
1763 /* For mlock, just skip the stack guard page. */
1764 if (foll_flags
& FOLL_MLOCK
) {
1765 if (stack_guard_page(vma
, start
))
1768 if (foll_flags
& FOLL_WRITE
)
1769 fault_flags
|= FAULT_FLAG_WRITE
;
1771 fault_flags
|= FAULT_FLAG_ALLOW_RETRY
;
1772 if (foll_flags
& FOLL_NOWAIT
)
1773 fault_flags
|= (FAULT_FLAG_ALLOW_RETRY
| FAULT_FLAG_RETRY_NOWAIT
);
1775 ret
= handle_mm_fault(mm
, vma
, start
,
1778 if (ret
& VM_FAULT_ERROR
) {
1779 if (ret
& VM_FAULT_OOM
)
1780 return i
? i
: -ENOMEM
;
1781 if (ret
& (VM_FAULT_HWPOISON
|
1782 VM_FAULT_HWPOISON_LARGE
)) {
1785 else if (gup_flags
& FOLL_HWPOISON
)
1790 if (ret
& VM_FAULT_SIGBUS
)
1791 return i
? i
: -EFAULT
;
1796 if (ret
& VM_FAULT_MAJOR
)
1802 if (ret
& VM_FAULT_RETRY
) {
1809 * The VM_FAULT_WRITE bit tells us that
1810 * do_wp_page has broken COW when necessary,
1811 * even if maybe_mkwrite decided not to set
1812 * pte_write. We can thus safely do subsequent
1813 * page lookups as if they were reads. But only
1814 * do so when looping for pte_write is futile:
1815 * in some cases userspace may also be wanting
1816 * to write to the gotten user page, which a
1817 * read fault here might prevent (a readonly
1818 * page might get reCOWed by userspace write).
1820 if ((ret
& VM_FAULT_WRITE
) &&
1821 !(vma
->vm_flags
& VM_WRITE
))
1822 foll_flags
&= ~FOLL_WRITE
;
1827 return i
? i
: PTR_ERR(page
);
1831 flush_anon_page(vma
, page
, start
);
1832 flush_dcache_page(page
);
1840 } while (nr_pages
&& start
< vma
->vm_end
);
1844 EXPORT_SYMBOL(__get_user_pages
);
1847 * fixup_user_fault() - manually resolve a user page fault
1848 * @tsk: the task_struct to use for page fault accounting, or
1849 * NULL if faults are not to be recorded.
1850 * @mm: mm_struct of target mm
1851 * @address: user address
1852 * @fault_flags:flags to pass down to handle_mm_fault()
1854 * This is meant to be called in the specific scenario where for locking reasons
1855 * we try to access user memory in atomic context (within a pagefault_disable()
1856 * section), this returns -EFAULT, and we want to resolve the user fault before
1859 * Typically this is meant to be used by the futex code.
1861 * The main difference with get_user_pages() is that this function will
1862 * unconditionally call handle_mm_fault() which will in turn perform all the
1863 * necessary SW fixup of the dirty and young bits in the PTE, while
1864 * handle_mm_fault() only guarantees to update these in the struct page.
1866 * This is important for some architectures where those bits also gate the
1867 * access permission to the page because they are maintained in software. On
1868 * such architectures, gup() will not be enough to make a subsequent access
1871 * This should be called with the mm_sem held for read.
1873 int fixup_user_fault(struct task_struct
*tsk
, struct mm_struct
*mm
,
1874 unsigned long address
, unsigned int fault_flags
)
1876 struct vm_area_struct
*vma
;
1879 vma
= find_extend_vma(mm
, address
);
1880 if (!vma
|| address
< vma
->vm_start
)
1883 ret
= handle_mm_fault(mm
, vma
, address
, fault_flags
);
1884 if (ret
& VM_FAULT_ERROR
) {
1885 if (ret
& VM_FAULT_OOM
)
1887 if (ret
& (VM_FAULT_HWPOISON
| VM_FAULT_HWPOISON_LARGE
))
1889 if (ret
& VM_FAULT_SIGBUS
)
1894 if (ret
& VM_FAULT_MAJOR
)
1903 * get_user_pages() - pin user pages in memory
1904 * @tsk: the task_struct to use for page fault accounting, or
1905 * NULL if faults are not to be recorded.
1906 * @mm: mm_struct of target mm
1907 * @start: starting user address
1908 * @nr_pages: number of pages from start to pin
1909 * @write: whether pages will be written to by the caller
1910 * @force: whether to force write access even if user mapping is
1911 * readonly. This will result in the page being COWed even
1912 * in MAP_SHARED mappings. You do not want this.
1913 * @pages: array that receives pointers to the pages pinned.
1914 * Should be at least nr_pages long. Or NULL, if caller
1915 * only intends to ensure the pages are faulted in.
1916 * @vmas: array of pointers to vmas corresponding to each page.
1917 * Or NULL if the caller does not require them.
1919 * Returns number of pages pinned. This may be fewer than the number
1920 * requested. If nr_pages is 0 or negative, returns 0. If no pages
1921 * were pinned, returns -errno. Each page returned must be released
1922 * with a put_page() call when it is finished with. vmas will only
1923 * remain valid while mmap_sem is held.
1925 * Must be called with mmap_sem held for read or write.
1927 * get_user_pages walks a process's page tables and takes a reference to
1928 * each struct page that each user address corresponds to at a given
1929 * instant. That is, it takes the page that would be accessed if a user
1930 * thread accesses the given user virtual address at that instant.
1932 * This does not guarantee that the page exists in the user mappings when
1933 * get_user_pages returns, and there may even be a completely different
1934 * page there in some cases (eg. if mmapped pagecache has been invalidated
1935 * and subsequently re faulted). However it does guarantee that the page
1936 * won't be freed completely. And mostly callers simply care that the page
1937 * contains data that was valid *at some point in time*. Typically, an IO
1938 * or similar operation cannot guarantee anything stronger anyway because
1939 * locks can't be held over the syscall boundary.
1941 * If write=0, the page must not be written to. If the page is written to,
1942 * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
1943 * after the page is finished with, and before put_page is called.
1945 * get_user_pages is typically used for fewer-copy IO operations, to get a
1946 * handle on the memory by some means other than accesses via the user virtual
1947 * addresses. The pages may be submitted for DMA to devices or accessed via
1948 * their kernel linear mapping (via the kmap APIs). Care should be taken to
1949 * use the correct cache flushing APIs.
1951 * See also get_user_pages_fast, for performance critical applications.
1953 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
1954 unsigned long start
, int nr_pages
, int write
, int force
,
1955 struct page
**pages
, struct vm_area_struct
**vmas
)
1957 int flags
= FOLL_TOUCH
;
1962 flags
|= FOLL_WRITE
;
1964 flags
|= FOLL_FORCE
;
1966 return __get_user_pages(tsk
, mm
, start
, nr_pages
, flags
, pages
, vmas
,
1969 EXPORT_SYMBOL(get_user_pages
);
1972 * get_dump_page() - pin user page in memory while writing it to core dump
1973 * @addr: user address
1975 * Returns struct page pointer of user page pinned for dump,
1976 * to be freed afterwards by page_cache_release() or put_page().
1978 * Returns NULL on any kind of failure - a hole must then be inserted into
1979 * the corefile, to preserve alignment with its headers; and also returns
1980 * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
1981 * allowing a hole to be left in the corefile to save diskspace.
1983 * Called without mmap_sem, but after all other threads have been killed.
1985 #ifdef CONFIG_ELF_CORE
1986 struct page
*get_dump_page(unsigned long addr
)
1988 struct vm_area_struct
*vma
;
1991 if (__get_user_pages(current
, current
->mm
, addr
, 1,
1992 FOLL_FORCE
| FOLL_DUMP
| FOLL_GET
, &page
, &vma
,
1995 flush_cache_page(vma
, addr
, page_to_pfn(page
));
1998 #endif /* CONFIG_ELF_CORE */
2000 pte_t
*__get_locked_pte(struct mm_struct
*mm
, unsigned long addr
,
2003 pgd_t
* pgd
= pgd_offset(mm
, addr
);
2004 pud_t
* pud
= pud_alloc(mm
, pgd
, addr
);
2006 pmd_t
* pmd
= pmd_alloc(mm
, pud
, addr
);
2008 VM_BUG_ON(pmd_trans_huge(*pmd
));
2009 return pte_alloc_map_lock(mm
, pmd
, addr
, ptl
);
2016 * This is the old fallback for page remapping.
2018 * For historical reasons, it only allows reserved pages. Only
2019 * old drivers should use this, and they needed to mark their
2020 * pages reserved for the old functions anyway.
2022 static int insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
2023 struct page
*page
, pgprot_t prot
)
2025 struct mm_struct
*mm
= vma
->vm_mm
;
2034 flush_dcache_page(page
);
2035 pte
= get_locked_pte(mm
, addr
, &ptl
);
2039 if (!pte_none(*pte
))
2042 /* Ok, finally just insert the thing.. */
2044 inc_mm_counter_fast(mm
, MM_FILEPAGES
);
2045 page_add_file_rmap(page
);
2046 set_pte_at(mm
, addr
, pte
, mk_pte(page
, prot
));
2049 pte_unmap_unlock(pte
, ptl
);
2052 pte_unmap_unlock(pte
, ptl
);
2058 * vm_insert_page - insert single page into user vma
2059 * @vma: user vma to map to
2060 * @addr: target user address of this page
2061 * @page: source kernel page
2063 * This allows drivers to insert individual pages they've allocated
2066 * The page has to be a nice clean _individual_ kernel allocation.
2067 * If you allocate a compound page, you need to have marked it as
2068 * such (__GFP_COMP), or manually just split the page up yourself
2069 * (see split_page()).
2071 * NOTE! Traditionally this was done with "remap_pfn_range()" which
2072 * took an arbitrary page protection parameter. This doesn't allow
2073 * that. Your vma protection will have to be set up correctly, which
2074 * means that if you want a shared writable mapping, you'd better
2075 * ask for a shared writable mapping!
2077 * The page does not need to be reserved.
2079 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
2082 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
2084 if (!page_count(page
))
2086 vma
->vm_flags
|= VM_INSERTPAGE
;
2087 return insert_page(vma
, addr
, page
, vma
->vm_page_prot
);
2089 EXPORT_SYMBOL(vm_insert_page
);
2091 static int insert_pfn(struct vm_area_struct
*vma
, unsigned long addr
,
2092 unsigned long pfn
, pgprot_t prot
)
2094 struct mm_struct
*mm
= vma
->vm_mm
;
2100 pte
= get_locked_pte(mm
, addr
, &ptl
);
2104 if (!pte_none(*pte
))
2107 /* Ok, finally just insert the thing.. */
2108 entry
= pte_mkspecial(pfn_pte(pfn
, prot
));
2109 set_pte_at(mm
, addr
, pte
, entry
);
2110 update_mmu_cache(vma
, addr
, pte
); /* XXX: why not for insert_page? */
2114 pte_unmap_unlock(pte
, ptl
);
2120 * vm_insert_pfn - insert single pfn into user vma
2121 * @vma: user vma to map to
2122 * @addr: target user address of this page
2123 * @pfn: source kernel pfn
2125 * Similar to vm_inert_page, this allows drivers to insert individual pages
2126 * they've allocated into a user vma. Same comments apply.
2128 * This function should only be called from a vm_ops->fault handler, and
2129 * in that case the handler should return NULL.
2131 * vma cannot be a COW mapping.
2133 * As this is called only for pages that do not currently exist, we
2134 * do not need to flush old virtual caches or the TLB.
2136 int vm_insert_pfn(struct vm_area_struct
*vma
, unsigned long addr
,
2140 pgprot_t pgprot
= vma
->vm_page_prot
;
2142 * Technically, architectures with pte_special can avoid all these
2143 * restrictions (same for remap_pfn_range). However we would like
2144 * consistency in testing and feature parity among all, so we should
2145 * try to keep these invariants in place for everybody.
2147 BUG_ON(!(vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
)));
2148 BUG_ON((vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
)) ==
2149 (VM_PFNMAP
|VM_MIXEDMAP
));
2150 BUG_ON((vma
->vm_flags
& VM_PFNMAP
) && is_cow_mapping(vma
->vm_flags
));
2151 BUG_ON((vma
->vm_flags
& VM_MIXEDMAP
) && pfn_valid(pfn
));
2153 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
2155 if (track_pfn_vma_new(vma
, &pgprot
, pfn
, PAGE_SIZE
))
2158 ret
= insert_pfn(vma
, addr
, pfn
, pgprot
);
2161 untrack_pfn_vma(vma
, pfn
, PAGE_SIZE
);
2165 EXPORT_SYMBOL(vm_insert_pfn
);
2167 int vm_insert_mixed(struct vm_area_struct
*vma
, unsigned long addr
,
2170 BUG_ON(!(vma
->vm_flags
& VM_MIXEDMAP
));
2172 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
2176 * If we don't have pte special, then we have to use the pfn_valid()
2177 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
2178 * refcount the page if pfn_valid is true (hence insert_page rather
2179 * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
2180 * without pte special, it would there be refcounted as a normal page.
2182 if (!HAVE_PTE_SPECIAL
&& pfn_valid(pfn
)) {
2185 page
= pfn_to_page(pfn
);
2186 return insert_page(vma
, addr
, page
, vma
->vm_page_prot
);
2188 return insert_pfn(vma
, addr
, pfn
, vma
->vm_page_prot
);
2190 EXPORT_SYMBOL(vm_insert_mixed
);
2193 * maps a range of physical memory into the requested pages. the old
2194 * mappings are removed. any references to nonexistent pages results
2195 * in null mappings (currently treated as "copy-on-access")
2197 static int remap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
2198 unsigned long addr
, unsigned long end
,
2199 unsigned long pfn
, pgprot_t prot
)
2204 pte
= pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
2207 arch_enter_lazy_mmu_mode();
2209 BUG_ON(!pte_none(*pte
));
2210 set_pte_at(mm
, addr
, pte
, pte_mkspecial(pfn_pte(pfn
, prot
)));
2212 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
2213 arch_leave_lazy_mmu_mode();
2214 pte_unmap_unlock(pte
- 1, ptl
);
2218 static inline int remap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
2219 unsigned long addr
, unsigned long end
,
2220 unsigned long pfn
, pgprot_t prot
)
2225 pfn
-= addr
>> PAGE_SHIFT
;
2226 pmd
= pmd_alloc(mm
, pud
, addr
);
2229 VM_BUG_ON(pmd_trans_huge(*pmd
));
2231 next
= pmd_addr_end(addr
, end
);
2232 if (remap_pte_range(mm
, pmd
, addr
, next
,
2233 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
2235 } while (pmd
++, addr
= next
, addr
!= end
);
2239 static inline int remap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
2240 unsigned long addr
, unsigned long end
,
2241 unsigned long pfn
, pgprot_t prot
)
2246 pfn
-= addr
>> PAGE_SHIFT
;
2247 pud
= pud_alloc(mm
, pgd
, addr
);
2251 next
= pud_addr_end(addr
, end
);
2252 if (remap_pmd_range(mm
, pud
, addr
, next
,
2253 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
2255 } while (pud
++, addr
= next
, addr
!= end
);
2260 * remap_pfn_range - remap kernel memory to userspace
2261 * @vma: user vma to map to
2262 * @addr: target user address to start at
2263 * @pfn: physical address of kernel memory
2264 * @size: size of map area
2265 * @prot: page protection flags for this mapping
2267 * Note: this is only safe if the mm semaphore is held when called.
2269 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
2270 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
2274 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2275 struct mm_struct
*mm
= vma
->vm_mm
;
2279 * Physically remapped pages are special. Tell the
2280 * rest of the world about it:
2281 * VM_IO tells people not to look at these pages
2282 * (accesses can have side effects).
2283 * VM_RESERVED is specified all over the place, because
2284 * in 2.4 it kept swapout's vma scan off this vma; but
2285 * in 2.6 the LRU scan won't even find its pages, so this
2286 * flag means no more than count its pages in reserved_vm,
2287 * and omit it from core dump, even when VM_IO turned off.
2288 * VM_PFNMAP tells the core MM that the base pages are just
2289 * raw PFN mappings, and do not have a "struct page" associated
2292 * There's a horrible special case to handle copy-on-write
2293 * behaviour that some programs depend on. We mark the "original"
2294 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
2296 if (addr
== vma
->vm_start
&& end
== vma
->vm_end
) {
2297 vma
->vm_pgoff
= pfn
;
2298 vma
->vm_flags
|= VM_PFN_AT_MMAP
;
2299 } else if (is_cow_mapping(vma
->vm_flags
))
2302 vma
->vm_flags
|= VM_IO
| VM_RESERVED
| VM_PFNMAP
;
2304 err
= track_pfn_vma_new(vma
, &prot
, pfn
, PAGE_ALIGN(size
));
2307 * To indicate that track_pfn related cleanup is not
2308 * needed from higher level routine calling unmap_vmas
2310 vma
->vm_flags
&= ~(VM_IO
| VM_RESERVED
| VM_PFNMAP
);
2311 vma
->vm_flags
&= ~VM_PFN_AT_MMAP
;
2315 BUG_ON(addr
>= end
);
2316 pfn
-= addr
>> PAGE_SHIFT
;
2317 pgd
= pgd_offset(mm
, addr
);
2318 flush_cache_range(vma
, addr
, end
);
2320 next
= pgd_addr_end(addr
, end
);
2321 err
= remap_pud_range(mm
, pgd
, addr
, next
,
2322 pfn
+ (addr
>> PAGE_SHIFT
), prot
);
2325 } while (pgd
++, addr
= next
, addr
!= end
);
2328 untrack_pfn_vma(vma
, pfn
, PAGE_ALIGN(size
));
2332 EXPORT_SYMBOL(remap_pfn_range
);
2334 static int apply_to_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
2335 unsigned long addr
, unsigned long end
,
2336 pte_fn_t fn
, void *data
)
2341 spinlock_t
*uninitialized_var(ptl
);
2343 pte
= (mm
== &init_mm
) ?
2344 pte_alloc_kernel(pmd
, addr
) :
2345 pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
2349 BUG_ON(pmd_huge(*pmd
));
2351 arch_enter_lazy_mmu_mode();
2353 token
= pmd_pgtable(*pmd
);
2356 err
= fn(pte
++, token
, addr
, data
);
2359 } while (addr
+= PAGE_SIZE
, addr
!= end
);
2361 arch_leave_lazy_mmu_mode();
2364 pte_unmap_unlock(pte
-1, ptl
);
2368 static int apply_to_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
2369 unsigned long addr
, unsigned long end
,
2370 pte_fn_t fn
, void *data
)
2376 BUG_ON(pud_huge(*pud
));
2378 pmd
= pmd_alloc(mm
, pud
, addr
);
2382 next
= pmd_addr_end(addr
, end
);
2383 err
= apply_to_pte_range(mm
, pmd
, addr
, next
, fn
, data
);
2386 } while (pmd
++, addr
= next
, addr
!= end
);
2390 static int apply_to_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
2391 unsigned long addr
, unsigned long end
,
2392 pte_fn_t fn
, void *data
)
2398 pud
= pud_alloc(mm
, pgd
, addr
);
2402 next
= pud_addr_end(addr
, end
);
2403 err
= apply_to_pmd_range(mm
, pud
, addr
, next
, fn
, data
);
2406 } while (pud
++, addr
= next
, addr
!= end
);
2411 * Scan a region of virtual memory, filling in page tables as necessary
2412 * and calling a provided function on each leaf page table.
2414 int apply_to_page_range(struct mm_struct
*mm
, unsigned long addr
,
2415 unsigned long size
, pte_fn_t fn
, void *data
)
2419 unsigned long end
= addr
+ size
;
2422 BUG_ON(addr
>= end
);
2423 pgd
= pgd_offset(mm
, addr
);
2425 next
= pgd_addr_end(addr
, end
);
2426 err
= apply_to_pud_range(mm
, pgd
, addr
, next
, fn
, data
);
2429 } while (pgd
++, addr
= next
, addr
!= end
);
2433 EXPORT_SYMBOL_GPL(apply_to_page_range
);
2436 * handle_pte_fault chooses page fault handler according to an entry
2437 * which was read non-atomically. Before making any commitment, on
2438 * those architectures or configurations (e.g. i386 with PAE) which
2439 * might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault
2440 * must check under lock before unmapping the pte and proceeding
2441 * (but do_wp_page is only called after already making such a check;
2442 * and do_anonymous_page can safely check later on).
2444 static inline int pte_unmap_same(struct mm_struct
*mm
, pmd_t
*pmd
,
2445 pte_t
*page_table
, pte_t orig_pte
)
2448 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
2449 if (sizeof(pte_t
) > sizeof(unsigned long)) {
2450 spinlock_t
*ptl
= pte_lockptr(mm
, pmd
);
2452 same
= pte_same(*page_table
, orig_pte
);
2456 pte_unmap(page_table
);
2460 static inline void cow_user_page(struct page
*dst
, struct page
*src
, unsigned long va
, struct vm_area_struct
*vma
)
2463 * If the source page was a PFN mapping, we don't have
2464 * a "struct page" for it. We do a best-effort copy by
2465 * just copying from the original user address. If that
2466 * fails, we just zero-fill it. Live with it.
2468 if (unlikely(!src
)) {
2469 void *kaddr
= kmap_atomic(dst
);
2470 void __user
*uaddr
= (void __user
*)(va
& PAGE_MASK
);
2473 * This really shouldn't fail, because the page is there
2474 * in the page tables. But it might just be unreadable,
2475 * in which case we just give up and fill the result with
2478 if (__copy_from_user_inatomic(kaddr
, uaddr
, PAGE_SIZE
))
2480 kunmap_atomic(kaddr
);
2481 flush_dcache_page(dst
);
2483 copy_user_highpage(dst
, src
, va
, vma
);
2487 * This routine handles present pages, when users try to write
2488 * to a shared page. It is done by copying the page to a new address
2489 * and decrementing the shared-page counter for the old page.
2491 * Note that this routine assumes that the protection checks have been
2492 * done by the caller (the low-level page fault routine in most cases).
2493 * Thus we can safely just mark it writable once we've done any necessary
2496 * We also mark the page dirty at this point even though the page will
2497 * change only once the write actually happens. This avoids a few races,
2498 * and potentially makes it more efficient.
2500 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2501 * but allow concurrent faults), with pte both mapped and locked.
2502 * We return with mmap_sem still held, but pte unmapped and unlocked.
2504 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2505 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2506 spinlock_t
*ptl
, pte_t orig_pte
)
2509 struct page
*old_page
, *new_page
;
2512 int page_mkwrite
= 0;
2513 struct page
*dirty_page
= NULL
;
2515 old_page
= vm_normal_page(vma
, address
, orig_pte
);
2518 * VM_MIXEDMAP !pfn_valid() case
2520 * We should not cow pages in a shared writeable mapping.
2521 * Just mark the pages writable as we can't do any dirty
2522 * accounting on raw pfn maps.
2524 if ((vma
->vm_flags
& (VM_WRITE
|VM_SHARED
)) ==
2525 (VM_WRITE
|VM_SHARED
))
2531 * Take out anonymous pages first, anonymous shared vmas are
2532 * not dirty accountable.
2534 if (PageAnon(old_page
) && !PageKsm(old_page
)) {
2535 if (!trylock_page(old_page
)) {
2536 page_cache_get(old_page
);
2537 pte_unmap_unlock(page_table
, ptl
);
2538 lock_page(old_page
);
2539 page_table
= pte_offset_map_lock(mm
, pmd
, address
,
2541 if (!pte_same(*page_table
, orig_pte
)) {
2542 unlock_page(old_page
);
2545 page_cache_release(old_page
);
2547 if (reuse_swap_page(old_page
)) {
2549 * The page is all ours. Move it to our anon_vma so
2550 * the rmap code will not search our parent or siblings.
2551 * Protected against the rmap code by the page lock.
2553 page_move_anon_rmap(old_page
, vma
, address
);
2554 unlock_page(old_page
);
2557 unlock_page(old_page
);
2558 } else if (unlikely((vma
->vm_flags
& (VM_WRITE
|VM_SHARED
)) ==
2559 (VM_WRITE
|VM_SHARED
))) {
2561 * Only catch write-faults on shared writable pages,
2562 * read-only shared pages can get COWed by
2563 * get_user_pages(.write=1, .force=1).
2565 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
) {
2566 struct vm_fault vmf
;
2569 vmf
.virtual_address
= (void __user
*)(address
&
2571 vmf
.pgoff
= old_page
->index
;
2572 vmf
.flags
= FAULT_FLAG_WRITE
|FAULT_FLAG_MKWRITE
;
2573 vmf
.page
= old_page
;
2576 * Notify the address space that the page is about to
2577 * become writable so that it can prohibit this or wait
2578 * for the page to get into an appropriate state.
2580 * We do this without the lock held, so that it can
2581 * sleep if it needs to.
2583 page_cache_get(old_page
);
2584 pte_unmap_unlock(page_table
, ptl
);
2586 tmp
= vma
->vm_ops
->page_mkwrite(vma
, &vmf
);
2588 (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
))) {
2590 goto unwritable_page
;
2592 if (unlikely(!(tmp
& VM_FAULT_LOCKED
))) {
2593 lock_page(old_page
);
2594 if (!old_page
->mapping
) {
2595 ret
= 0; /* retry the fault */
2596 unlock_page(old_page
);
2597 goto unwritable_page
;
2600 VM_BUG_ON(!PageLocked(old_page
));
2603 * Since we dropped the lock we need to revalidate
2604 * the PTE as someone else may have changed it. If
2605 * they did, we just return, as we can count on the
2606 * MMU to tell us if they didn't also make it writable.
2608 page_table
= pte_offset_map_lock(mm
, pmd
, address
,
2610 if (!pte_same(*page_table
, orig_pte
)) {
2611 unlock_page(old_page
);
2617 dirty_page
= old_page
;
2618 get_page(dirty_page
);
2621 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
2622 entry
= pte_mkyoung(orig_pte
);
2623 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2624 if (ptep_set_access_flags(vma
, address
, page_table
, entry
,1))
2625 update_mmu_cache(vma
, address
, page_table
);
2626 pte_unmap_unlock(page_table
, ptl
);
2627 ret
|= VM_FAULT_WRITE
;
2633 * Yes, Virginia, this is actually required to prevent a race
2634 * with clear_page_dirty_for_io() from clearing the page dirty
2635 * bit after it clear all dirty ptes, but before a racing
2636 * do_wp_page installs a dirty pte.
2638 * __do_fault is protected similarly.
2640 if (!page_mkwrite
) {
2641 wait_on_page_locked(dirty_page
);
2642 set_page_dirty_balance(dirty_page
, page_mkwrite
);
2644 put_page(dirty_page
);
2646 struct address_space
*mapping
= dirty_page
->mapping
;
2648 set_page_dirty(dirty_page
);
2649 unlock_page(dirty_page
);
2650 page_cache_release(dirty_page
);
2653 * Some device drivers do not set page.mapping
2654 * but still dirty their pages
2656 balance_dirty_pages_ratelimited(mapping
);
2660 /* file_update_time outside page_lock */
2662 file_update_time(vma
->vm_file
);
2668 * Ok, we need to copy. Oh, well..
2670 page_cache_get(old_page
);
2672 pte_unmap_unlock(page_table
, ptl
);
2674 if (unlikely(anon_vma_prepare(vma
)))
2677 if (is_zero_pfn(pte_pfn(orig_pte
))) {
2678 new_page
= alloc_zeroed_user_highpage_movable(vma
, address
);
2682 new_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, address
);
2685 cow_user_page(new_page
, old_page
, address
, vma
);
2687 __SetPageUptodate(new_page
);
2689 if (mem_cgroup_newpage_charge(new_page
, mm
, GFP_KERNEL
))
2693 * Re-check the pte - we dropped the lock
2695 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2696 if (likely(pte_same(*page_table
, orig_pte
))) {
2698 if (!PageAnon(old_page
)) {
2699 dec_mm_counter_fast(mm
, MM_FILEPAGES
);
2700 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
2703 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
2704 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
2705 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
2706 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2708 * Clear the pte entry and flush it first, before updating the
2709 * pte with the new entry. This will avoid a race condition
2710 * seen in the presence of one thread doing SMC and another
2713 ptep_clear_flush(vma
, address
, page_table
);
2714 page_add_new_anon_rmap(new_page
, vma
, address
);
2716 * We call the notify macro here because, when using secondary
2717 * mmu page tables (such as kvm shadow page tables), we want the
2718 * new page to be mapped directly into the secondary page table.
2720 set_pte_at_notify(mm
, address
, page_table
, entry
);
2721 update_mmu_cache(vma
, address
, page_table
);
2724 * Only after switching the pte to the new page may
2725 * we remove the mapcount here. Otherwise another
2726 * process may come and find the rmap count decremented
2727 * before the pte is switched to the new page, and
2728 * "reuse" the old page writing into it while our pte
2729 * here still points into it and can be read by other
2732 * The critical issue is to order this
2733 * page_remove_rmap with the ptp_clear_flush above.
2734 * Those stores are ordered by (if nothing else,)
2735 * the barrier present in the atomic_add_negative
2736 * in page_remove_rmap.
2738 * Then the TLB flush in ptep_clear_flush ensures that
2739 * no process can access the old page before the
2740 * decremented mapcount is visible. And the old page
2741 * cannot be reused until after the decremented
2742 * mapcount is visible. So transitively, TLBs to
2743 * old page will be flushed before it can be reused.
2745 page_remove_rmap(old_page
);
2748 /* Free the old page.. */
2749 new_page
= old_page
;
2750 ret
|= VM_FAULT_WRITE
;
2752 mem_cgroup_uncharge_page(new_page
);
2755 page_cache_release(new_page
);
2757 pte_unmap_unlock(page_table
, ptl
);
2760 * Don't let another task, with possibly unlocked vma,
2761 * keep the mlocked page.
2763 if ((ret
& VM_FAULT_WRITE
) && (vma
->vm_flags
& VM_LOCKED
)) {
2764 lock_page(old_page
); /* LRU manipulation */
2765 munlock_vma_page(old_page
);
2766 unlock_page(old_page
);
2768 page_cache_release(old_page
);
2772 page_cache_release(new_page
);
2776 unlock_page(old_page
);
2777 page_cache_release(old_page
);
2779 page_cache_release(old_page
);
2781 return VM_FAULT_OOM
;
2784 page_cache_release(old_page
);
2788 static void unmap_mapping_range_vma(struct vm_area_struct
*vma
,
2789 unsigned long start_addr
, unsigned long end_addr
,
2790 struct zap_details
*details
)
2792 zap_page_range_single(vma
, start_addr
, end_addr
- start_addr
, details
);
2795 static inline void unmap_mapping_range_tree(struct prio_tree_root
*root
,
2796 struct zap_details
*details
)
2798 struct vm_area_struct
*vma
;
2799 struct prio_tree_iter iter
;
2800 pgoff_t vba
, vea
, zba
, zea
;
2802 vma_prio_tree_foreach(vma
, &iter
, root
,
2803 details
->first_index
, details
->last_index
) {
2805 vba
= vma
->vm_pgoff
;
2806 vea
= vba
+ ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
) - 1;
2807 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2808 zba
= details
->first_index
;
2811 zea
= details
->last_index
;
2815 unmap_mapping_range_vma(vma
,
2816 ((zba
- vba
) << PAGE_SHIFT
) + vma
->vm_start
,
2817 ((zea
- vba
+ 1) << PAGE_SHIFT
) + vma
->vm_start
,
2822 static inline void unmap_mapping_range_list(struct list_head
*head
,
2823 struct zap_details
*details
)
2825 struct vm_area_struct
*vma
;
2828 * In nonlinear VMAs there is no correspondence between virtual address
2829 * offset and file offset. So we must perform an exhaustive search
2830 * across *all* the pages in each nonlinear VMA, not just the pages
2831 * whose virtual address lies outside the file truncation point.
2833 list_for_each_entry(vma
, head
, shared
.vm_set
.list
) {
2834 details
->nonlinear_vma
= vma
;
2835 unmap_mapping_range_vma(vma
, vma
->vm_start
, vma
->vm_end
, details
);
2840 * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
2841 * @mapping: the address space containing mmaps to be unmapped.
2842 * @holebegin: byte in first page to unmap, relative to the start of
2843 * the underlying file. This will be rounded down to a PAGE_SIZE
2844 * boundary. Note that this is different from truncate_pagecache(), which
2845 * must keep the partial page. In contrast, we must get rid of
2847 * @holelen: size of prospective hole in bytes. This will be rounded
2848 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2850 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2851 * but 0 when invalidating pagecache, don't throw away private data.
2853 void unmap_mapping_range(struct address_space
*mapping
,
2854 loff_t
const holebegin
, loff_t
const holelen
, int even_cows
)
2856 struct zap_details details
;
2857 pgoff_t hba
= holebegin
>> PAGE_SHIFT
;
2858 pgoff_t hlen
= (holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2860 /* Check for overflow. */
2861 if (sizeof(holelen
) > sizeof(hlen
)) {
2863 (holebegin
+ holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2864 if (holeend
& ~(long long)ULONG_MAX
)
2865 hlen
= ULONG_MAX
- hba
+ 1;
2868 details
.check_mapping
= even_cows
? NULL
: mapping
;
2869 details
.nonlinear_vma
= NULL
;
2870 details
.first_index
= hba
;
2871 details
.last_index
= hba
+ hlen
- 1;
2872 if (details
.last_index
< details
.first_index
)
2873 details
.last_index
= ULONG_MAX
;
2876 mutex_lock(&mapping
->i_mmap_mutex
);
2877 if (unlikely(!prio_tree_empty(&mapping
->i_mmap
)))
2878 unmap_mapping_range_tree(&mapping
->i_mmap
, &details
);
2879 if (unlikely(!list_empty(&mapping
->i_mmap_nonlinear
)))
2880 unmap_mapping_range_list(&mapping
->i_mmap_nonlinear
, &details
);
2881 mutex_unlock(&mapping
->i_mmap_mutex
);
2883 EXPORT_SYMBOL(unmap_mapping_range
);
2886 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2887 * but allow concurrent faults), and pte mapped but not yet locked.
2888 * We return with mmap_sem still held, but pte unmapped and unlocked.
2890 static int do_swap_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2891 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2892 unsigned int flags
, pte_t orig_pte
)
2895 struct page
*page
, *swapcache
= NULL
;
2899 struct mem_cgroup
*ptr
;
2903 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
2906 entry
= pte_to_swp_entry(orig_pte
);
2907 if (unlikely(non_swap_entry(entry
))) {
2908 if (is_migration_entry(entry
)) {
2909 migration_entry_wait(mm
, pmd
, address
);
2910 } else if (is_hwpoison_entry(entry
)) {
2911 ret
= VM_FAULT_HWPOISON
;
2913 print_bad_pte(vma
, address
, orig_pte
, NULL
);
2914 ret
= VM_FAULT_SIGBUS
;
2918 delayacct_set_flag(DELAYACCT_PF_SWAPIN
);
2919 page
= lookup_swap_cache(entry
);
2921 page
= swapin_readahead(entry
,
2922 GFP_HIGHUSER_MOVABLE
, vma
, address
);
2925 * Back out if somebody else faulted in this pte
2926 * while we released the pte lock.
2928 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2929 if (likely(pte_same(*page_table
, orig_pte
)))
2931 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2935 /* Had to read the page from swap area: Major fault */
2936 ret
= VM_FAULT_MAJOR
;
2937 count_vm_event(PGMAJFAULT
);
2938 mem_cgroup_count_vm_event(mm
, PGMAJFAULT
);
2939 } else if (PageHWPoison(page
)) {
2941 * hwpoisoned dirty swapcache pages are kept for killing
2942 * owner processes (which may be unknown at hwpoison time)
2944 ret
= VM_FAULT_HWPOISON
;
2945 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2949 locked
= lock_page_or_retry(page
, mm
, flags
);
2951 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2953 ret
|= VM_FAULT_RETRY
;
2958 * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
2959 * release the swapcache from under us. The page pin, and pte_same
2960 * test below, are not enough to exclude that. Even if it is still
2961 * swapcache, we need to check that the page's swap has not changed.
2963 if (unlikely(!PageSwapCache(page
) || page_private(page
) != entry
.val
))
2966 if (ksm_might_need_to_copy(page
, vma
, address
)) {
2968 page
= ksm_does_need_to_copy(page
, vma
, address
);
2970 if (unlikely(!page
)) {
2978 if (mem_cgroup_try_charge_swapin(mm
, page
, GFP_KERNEL
, &ptr
)) {
2984 * Back out if somebody else already faulted in this pte.
2986 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2987 if (unlikely(!pte_same(*page_table
, orig_pte
)))
2990 if (unlikely(!PageUptodate(page
))) {
2991 ret
= VM_FAULT_SIGBUS
;
2996 * The page isn't present yet, go ahead with the fault.
2998 * Be careful about the sequence of operations here.
2999 * To get its accounting right, reuse_swap_page() must be called
3000 * while the page is counted on swap but not yet in mapcount i.e.
3001 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
3002 * must be called after the swap_free(), or it will never succeed.
3003 * Because delete_from_swap_page() may be called by reuse_swap_page(),
3004 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
3005 * in page->private. In this case, a record in swap_cgroup is silently
3006 * discarded at swap_free().
3009 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
3010 dec_mm_counter_fast(mm
, MM_SWAPENTS
);
3011 pte
= mk_pte(page
, vma
->vm_page_prot
);
3012 if ((flags
& FAULT_FLAG_WRITE
) && reuse_swap_page(page
)) {
3013 pte
= maybe_mkwrite(pte_mkdirty(pte
), vma
);
3014 flags
&= ~FAULT_FLAG_WRITE
;
3015 ret
|= VM_FAULT_WRITE
;
3018 flush_icache_page(vma
, page
);
3019 set_pte_at(mm
, address
, page_table
, pte
);
3020 do_page_add_anon_rmap(page
, vma
, address
, exclusive
);
3021 /* It's better to call commit-charge after rmap is established */
3022 mem_cgroup_commit_charge_swapin(page
, ptr
);
3025 if (vm_swap_full() || (vma
->vm_flags
& VM_LOCKED
) || PageMlocked(page
))
3026 try_to_free_swap(page
);
3030 * Hold the lock to avoid the swap entry to be reused
3031 * until we take the PT lock for the pte_same() check
3032 * (to avoid false positives from pte_same). For
3033 * further safety release the lock after the swap_free
3034 * so that the swap count won't change under a
3035 * parallel locked swapcache.
3037 unlock_page(swapcache
);
3038 page_cache_release(swapcache
);
3041 if (flags
& FAULT_FLAG_WRITE
) {
3042 ret
|= do_wp_page(mm
, vma
, address
, page_table
, pmd
, ptl
, pte
);
3043 if (ret
& VM_FAULT_ERROR
)
3044 ret
&= VM_FAULT_ERROR
;
3048 /* No need to invalidate - it was non-present before */
3049 update_mmu_cache(vma
, address
, page_table
);
3051 pte_unmap_unlock(page_table
, ptl
);
3055 mem_cgroup_cancel_charge_swapin(ptr
);
3056 pte_unmap_unlock(page_table
, ptl
);
3060 page_cache_release(page
);
3062 unlock_page(swapcache
);
3063 page_cache_release(swapcache
);
3069 * This is like a special single-page "expand_{down|up}wards()",
3070 * except we must first make sure that 'address{-|+}PAGE_SIZE'
3071 * doesn't hit another vma.
3073 static inline int check_stack_guard_page(struct vm_area_struct
*vma
, unsigned long address
)
3075 address
&= PAGE_MASK
;
3076 if ((vma
->vm_flags
& VM_GROWSDOWN
) && address
== vma
->vm_start
) {
3077 struct vm_area_struct
*prev
= vma
->vm_prev
;
3080 * Is there a mapping abutting this one below?
3082 * That's only ok if it's the same stack mapping
3083 * that has gotten split..
3085 if (prev
&& prev
->vm_end
== address
)
3086 return prev
->vm_flags
& VM_GROWSDOWN
? 0 : -ENOMEM
;
3088 expand_downwards(vma
, address
- PAGE_SIZE
);
3090 if ((vma
->vm_flags
& VM_GROWSUP
) && address
+ PAGE_SIZE
== vma
->vm_end
) {
3091 struct vm_area_struct
*next
= vma
->vm_next
;
3093 /* As VM_GROWSDOWN but s/below/above/ */
3094 if (next
&& next
->vm_start
== address
+ PAGE_SIZE
)
3095 return next
->vm_flags
& VM_GROWSUP
? 0 : -ENOMEM
;
3097 expand_upwards(vma
, address
+ PAGE_SIZE
);
3103 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3104 * but allow concurrent faults), and pte mapped but not yet locked.
3105 * We return with mmap_sem still held, but pte unmapped and unlocked.
3107 static int do_anonymous_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3108 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
3115 pte_unmap(page_table
);
3117 /* Check if we need to add a guard page to the stack */
3118 if (check_stack_guard_page(vma
, address
) < 0)
3119 return VM_FAULT_SIGBUS
;
3121 /* Use the zero-page for reads */
3122 if (!(flags
& FAULT_FLAG_WRITE
)) {
3123 entry
= pte_mkspecial(pfn_pte(my_zero_pfn(address
),
3124 vma
->vm_page_prot
));
3125 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
3126 if (!pte_none(*page_table
))
3131 /* Allocate our own private page. */
3132 if (unlikely(anon_vma_prepare(vma
)))
3134 page
= alloc_zeroed_user_highpage_movable(vma
, address
);
3137 __SetPageUptodate(page
);
3139 if (mem_cgroup_newpage_charge(page
, mm
, GFP_KERNEL
))
3142 entry
= mk_pte(page
, vma
->vm_page_prot
);
3143 if (vma
->vm_flags
& VM_WRITE
)
3144 entry
= pte_mkwrite(pte_mkdirty(entry
));
3146 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
3147 if (!pte_none(*page_table
))
3150 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
3151 page_add_new_anon_rmap(page
, vma
, address
);
3153 set_pte_at(mm
, address
, page_table
, entry
);
3155 /* No need to invalidate - it was non-present before */
3156 update_mmu_cache(vma
, address
, page_table
);
3158 pte_unmap_unlock(page_table
, ptl
);
3161 mem_cgroup_uncharge_page(page
);
3162 page_cache_release(page
);
3165 page_cache_release(page
);
3167 return VM_FAULT_OOM
;
3171 * __do_fault() tries to create a new page mapping. It aggressively
3172 * tries to share with existing pages, but makes a separate copy if
3173 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
3174 * the next page fault.
3176 * As this is called only for pages that do not currently exist, we
3177 * do not need to flush old virtual caches or the TLB.
3179 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3180 * but allow concurrent faults), and pte neither mapped nor locked.
3181 * We return with mmap_sem still held, but pte unmapped and unlocked.
3183 static int __do_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3184 unsigned long address
, pmd_t
*pmd
,
3185 pgoff_t pgoff
, unsigned int flags
, pte_t orig_pte
)
3190 struct page
*cow_page
;
3193 struct page
*dirty_page
= NULL
;
3194 struct vm_fault vmf
;
3196 int page_mkwrite
= 0;
3199 * If we do COW later, allocate page befor taking lock_page()
3200 * on the file cache page. This will reduce lock holding time.
3202 if ((flags
& FAULT_FLAG_WRITE
) && !(vma
->vm_flags
& VM_SHARED
)) {
3204 if (unlikely(anon_vma_prepare(vma
)))
3205 return VM_FAULT_OOM
;
3207 cow_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, address
);
3209 return VM_FAULT_OOM
;
3211 if (mem_cgroup_newpage_charge(cow_page
, mm
, GFP_KERNEL
)) {
3212 page_cache_release(cow_page
);
3213 return VM_FAULT_OOM
;
3218 vmf
.virtual_address
= (void __user
*)(address
& PAGE_MASK
);
3223 ret
= vma
->vm_ops
->fault(vma
, &vmf
);
3224 if (unlikely(ret
& (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
|
3228 if (unlikely(PageHWPoison(vmf
.page
))) {
3229 if (ret
& VM_FAULT_LOCKED
)
3230 unlock_page(vmf
.page
);
3231 ret
= VM_FAULT_HWPOISON
;
3236 * For consistency in subsequent calls, make the faulted page always
3239 if (unlikely(!(ret
& VM_FAULT_LOCKED
)))
3240 lock_page(vmf
.page
);
3242 VM_BUG_ON(!PageLocked(vmf
.page
));
3245 * Should we do an early C-O-W break?
3248 if (flags
& FAULT_FLAG_WRITE
) {
3249 if (!(vma
->vm_flags
& VM_SHARED
)) {
3252 copy_user_highpage(page
, vmf
.page
, address
, vma
);
3253 __SetPageUptodate(page
);
3256 * If the page will be shareable, see if the backing
3257 * address space wants to know that the page is about
3258 * to become writable
3260 if (vma
->vm_ops
->page_mkwrite
) {
3264 vmf
.flags
= FAULT_FLAG_WRITE
|FAULT_FLAG_MKWRITE
;
3265 tmp
= vma
->vm_ops
->page_mkwrite(vma
, &vmf
);
3267 (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
))) {
3269 goto unwritable_page
;
3271 if (unlikely(!(tmp
& VM_FAULT_LOCKED
))) {
3273 if (!page
->mapping
) {
3274 ret
= 0; /* retry the fault */
3276 goto unwritable_page
;
3279 VM_BUG_ON(!PageLocked(page
));
3286 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
3289 * This silly early PAGE_DIRTY setting removes a race
3290 * due to the bad i386 page protection. But it's valid
3291 * for other architectures too.
3293 * Note that if FAULT_FLAG_WRITE is set, we either now have
3294 * an exclusive copy of the page, or this is a shared mapping,
3295 * so we can make it writable and dirty to avoid having to
3296 * handle that later.
3298 /* Only go through if we didn't race with anybody else... */
3299 if (likely(pte_same(*page_table
, orig_pte
))) {
3300 flush_icache_page(vma
, page
);
3301 entry
= mk_pte(page
, vma
->vm_page_prot
);
3302 if (flags
& FAULT_FLAG_WRITE
)
3303 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
3305 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
3306 page_add_new_anon_rmap(page
, vma
, address
);
3308 inc_mm_counter_fast(mm
, MM_FILEPAGES
);
3309 page_add_file_rmap(page
);
3310 if (flags
& FAULT_FLAG_WRITE
) {
3312 get_page(dirty_page
);
3315 set_pte_at(mm
, address
, page_table
, entry
);
3317 /* no need to invalidate: a not-present page won't be cached */
3318 update_mmu_cache(vma
, address
, page_table
);
3321 mem_cgroup_uncharge_page(cow_page
);
3323 page_cache_release(page
);
3325 anon
= 1; /* no anon but release faulted_page */
3328 pte_unmap_unlock(page_table
, ptl
);
3331 struct address_space
*mapping
= page
->mapping
;
3333 if (set_page_dirty(dirty_page
))
3335 unlock_page(dirty_page
);
3336 put_page(dirty_page
);
3337 if (page_mkwrite
&& mapping
) {
3339 * Some device drivers do not set page.mapping but still
3342 balance_dirty_pages_ratelimited(mapping
);
3345 /* file_update_time outside page_lock */
3347 file_update_time(vma
->vm_file
);
3349 unlock_page(vmf
.page
);
3351 page_cache_release(vmf
.page
);
3357 page_cache_release(page
);
3360 /* fs's fault handler get error */
3362 mem_cgroup_uncharge_page(cow_page
);
3363 page_cache_release(cow_page
);
3368 static int do_linear_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3369 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
3370 unsigned int flags
, pte_t orig_pte
)
3372 pgoff_t pgoff
= (((address
& PAGE_MASK
)
3373 - vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
3375 pte_unmap(page_table
);
3376 return __do_fault(mm
, vma
, address
, pmd
, pgoff
, flags
, orig_pte
);
3380 * Fault of a previously existing named mapping. Repopulate the pte
3381 * from the encoded file_pte if possible. This enables swappable
3384 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3385 * but allow concurrent faults), and pte mapped but not yet locked.
3386 * We return with mmap_sem still held, but pte unmapped and unlocked.
3388 static int do_nonlinear_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3389 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
3390 unsigned int flags
, pte_t orig_pte
)
3394 flags
|= FAULT_FLAG_NONLINEAR
;
3396 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
3399 if (unlikely(!(vma
->vm_flags
& VM_NONLINEAR
))) {
3401 * Page table corrupted: show pte and kill process.
3403 print_bad_pte(vma
, address
, orig_pte
, NULL
);
3404 return VM_FAULT_SIGBUS
;
3407 pgoff
= pte_to_pgoff(orig_pte
);
3408 return __do_fault(mm
, vma
, address
, pmd
, pgoff
, flags
, orig_pte
);
3412 * These routines also need to handle stuff like marking pages dirty
3413 * and/or accessed for architectures that don't do it in hardware (most
3414 * RISC architectures). The early dirtying is also good on the i386.
3416 * There is also a hook called "update_mmu_cache()" that architectures
3417 * with external mmu caches can use to update those (ie the Sparc or
3418 * PowerPC hashed page tables that act as extended TLBs).
3420 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3421 * but allow concurrent faults), and pte mapped but not yet locked.
3422 * We return with mmap_sem still held, but pte unmapped and unlocked.
3424 int handle_pte_fault(struct mm_struct
*mm
,
3425 struct vm_area_struct
*vma
, unsigned long address
,
3426 pte_t
*pte
, pmd_t
*pmd
, unsigned int flags
)
3432 if (!pte_present(entry
)) {
3433 if (pte_none(entry
)) {
3435 if (likely(vma
->vm_ops
->fault
))
3436 return do_linear_fault(mm
, vma
, address
,
3437 pte
, pmd
, flags
, entry
);
3439 return do_anonymous_page(mm
, vma
, address
,
3442 if (pte_file(entry
))
3443 return do_nonlinear_fault(mm
, vma
, address
,
3444 pte
, pmd
, flags
, entry
);
3445 return do_swap_page(mm
, vma
, address
,
3446 pte
, pmd
, flags
, entry
);
3449 ptl
= pte_lockptr(mm
, pmd
);
3451 if (unlikely(!pte_same(*pte
, entry
)))
3453 if (flags
& FAULT_FLAG_WRITE
) {
3454 if (!pte_write(entry
))
3455 return do_wp_page(mm
, vma
, address
,
3456 pte
, pmd
, ptl
, entry
);
3457 entry
= pte_mkdirty(entry
);
3459 entry
= pte_mkyoung(entry
);
3460 if (ptep_set_access_flags(vma
, address
, pte
, entry
, flags
& FAULT_FLAG_WRITE
)) {
3461 update_mmu_cache(vma
, address
, pte
);
3464 * This is needed only for protection faults but the arch code
3465 * is not yet telling us if this is a protection fault or not.
3466 * This still avoids useless tlb flushes for .text page faults
3469 if (flags
& FAULT_FLAG_WRITE
)
3470 flush_tlb_fix_spurious_fault(vma
, address
);
3473 pte_unmap_unlock(pte
, ptl
);
3478 * By the time we get here, we already hold the mm semaphore
3480 int handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3481 unsigned long address
, unsigned int flags
)
3488 __set_current_state(TASK_RUNNING
);
3490 count_vm_event(PGFAULT
);
3491 mem_cgroup_count_vm_event(mm
, PGFAULT
);
3493 /* do counter updates before entering really critical section. */
3494 check_sync_rss_stat(current
);
3496 if (unlikely(is_vm_hugetlb_page(vma
)))
3497 return hugetlb_fault(mm
, vma
, address
, flags
);
3500 pgd
= pgd_offset(mm
, address
);
3501 pud
= pud_alloc(mm
, pgd
, address
);
3503 return VM_FAULT_OOM
;
3504 pmd
= pmd_alloc(mm
, pud
, address
);
3506 return VM_FAULT_OOM
;
3507 if (pmd_none(*pmd
) && transparent_hugepage_enabled(vma
)) {
3509 return do_huge_pmd_anonymous_page(mm
, vma
, address
,
3512 pmd_t orig_pmd
= *pmd
;
3516 if (pmd_trans_huge(orig_pmd
)) {
3517 if (flags
& FAULT_FLAG_WRITE
&&
3518 !pmd_write(orig_pmd
) &&
3519 !pmd_trans_splitting(orig_pmd
)) {
3520 ret
= do_huge_pmd_wp_page(mm
, vma
, address
, pmd
,
3523 * If COW results in an oom, the huge pmd will
3524 * have been split, so retry the fault on the
3525 * pte for a smaller charge.
3527 if (unlikely(ret
& VM_FAULT_OOM
))
3536 * Use __pte_alloc instead of pte_alloc_map, because we can't
3537 * run pte_offset_map on the pmd, if an huge pmd could
3538 * materialize from under us from a different thread.
3540 if (unlikely(pmd_none(*pmd
)) && __pte_alloc(mm
, vma
, pmd
, address
))
3541 return VM_FAULT_OOM
;
3542 /* if an huge pmd materialized from under us just retry later */
3543 if (unlikely(pmd_trans_huge(*pmd
)))
3546 * A regular pmd is established and it can't morph into a huge pmd
3547 * from under us anymore at this point because we hold the mmap_sem
3548 * read mode and khugepaged takes it in write mode. So now it's
3549 * safe to run pte_offset_map().
3551 pte
= pte_offset_map(pmd
, address
);
3553 return handle_pte_fault(mm
, vma
, address
, pte
, pmd
, flags
);
3555 EXPORT_SYMBOL(handle_mm_fault
);
3557 #ifndef __PAGETABLE_PUD_FOLDED
3559 * Allocate page upper directory.
3560 * We've already handled the fast-path in-line.
3562 int __pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
3564 pud_t
*new = pud_alloc_one(mm
, address
);
3568 smp_wmb(); /* See comment in __pte_alloc */
3570 spin_lock(&mm
->page_table_lock
);
3571 if (pgd_present(*pgd
)) /* Another has populated it */
3574 pgd_populate(mm
, pgd
, new);
3575 spin_unlock(&mm
->page_table_lock
);
3578 #endif /* __PAGETABLE_PUD_FOLDED */
3580 #ifndef __PAGETABLE_PMD_FOLDED
3582 * Allocate page middle directory.
3583 * We've already handled the fast-path in-line.
3585 int __pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
3587 pmd_t
*new = pmd_alloc_one(mm
, address
);
3591 smp_wmb(); /* See comment in __pte_alloc */
3593 spin_lock(&mm
->page_table_lock
);
3594 #ifndef __ARCH_HAS_4LEVEL_HACK
3595 if (pud_present(*pud
)) /* Another has populated it */
3598 pud_populate(mm
, pud
, new);
3600 if (pgd_present(*pud
)) /* Another has populated it */
3603 pgd_populate(mm
, pud
, new);
3604 #endif /* __ARCH_HAS_4LEVEL_HACK */
3605 spin_unlock(&mm
->page_table_lock
);
3608 #endif /* __PAGETABLE_PMD_FOLDED */
3610 int make_pages_present(unsigned long addr
, unsigned long end
)
3612 int ret
, len
, write
;
3613 struct vm_area_struct
* vma
;
3615 vma
= find_vma(current
->mm
, addr
);
3619 * We want to touch writable mappings with a write fault in order
3620 * to break COW, except for shared mappings because these don't COW
3621 * and we would not want to dirty them for nothing.
3623 write
= (vma
->vm_flags
& (VM_WRITE
| VM_SHARED
)) == VM_WRITE
;
3624 BUG_ON(addr
>= end
);
3625 BUG_ON(end
> vma
->vm_end
);
3626 len
= DIV_ROUND_UP(end
, PAGE_SIZE
) - addr
/PAGE_SIZE
;
3627 ret
= get_user_pages(current
, current
->mm
, addr
,
3628 len
, write
, 0, NULL
, NULL
);
3631 return ret
== len
? 0 : -EFAULT
;
3634 #if !defined(__HAVE_ARCH_GATE_AREA)
3636 #if defined(AT_SYSINFO_EHDR)
3637 static struct vm_area_struct gate_vma
;
3639 static int __init
gate_vma_init(void)
3641 gate_vma
.vm_mm
= NULL
;
3642 gate_vma
.vm_start
= FIXADDR_USER_START
;
3643 gate_vma
.vm_end
= FIXADDR_USER_END
;
3644 gate_vma
.vm_flags
= VM_READ
| VM_MAYREAD
| VM_EXEC
| VM_MAYEXEC
;
3645 gate_vma
.vm_page_prot
= __P101
;
3649 __initcall(gate_vma_init
);
3652 struct vm_area_struct
*get_gate_vma(struct mm_struct
*mm
)
3654 #ifdef AT_SYSINFO_EHDR
3660 EXPORT_SYMBOL(get_gate_vma
);
3662 int in_gate_area_no_mm(unsigned long addr
)
3664 #ifdef AT_SYSINFO_EHDR
3665 if ((addr
>= FIXADDR_USER_START
) && (addr
< FIXADDR_USER_END
))
3670 EXPORT_SYMBOL(in_gate_area_no_mm
);
3672 #endif /* __HAVE_ARCH_GATE_AREA */
3674 static int __follow_pte(struct mm_struct
*mm
, unsigned long address
,
3675 pte_t
**ptepp
, spinlock_t
**ptlp
)
3682 pgd
= pgd_offset(mm
, address
);
3683 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
3686 pud
= pud_offset(pgd
, address
);
3687 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
3690 pmd
= pmd_offset(pud
, address
);
3691 VM_BUG_ON(pmd_trans_huge(*pmd
));
3692 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
3695 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3699 ptep
= pte_offset_map_lock(mm
, pmd
, address
, ptlp
);
3702 if (!pte_present(*ptep
))
3707 pte_unmap_unlock(ptep
, *ptlp
);
3712 static inline int follow_pte(struct mm_struct
*mm
, unsigned long address
,
3713 pte_t
**ptepp
, spinlock_t
**ptlp
)
3717 /* (void) is needed to make gcc happy */
3718 (void) __cond_lock(*ptlp
,
3719 !(res
= __follow_pte(mm
, address
, ptepp
, ptlp
)));
3724 * follow_pfn - look up PFN at a user virtual address
3725 * @vma: memory mapping
3726 * @address: user virtual address
3727 * @pfn: location to store found PFN
3729 * Only IO mappings and raw PFN mappings are allowed.
3731 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3733 int follow_pfn(struct vm_area_struct
*vma
, unsigned long address
,
3740 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
3743 ret
= follow_pte(vma
->vm_mm
, address
, &ptep
, &ptl
);
3746 *pfn
= pte_pfn(*ptep
);
3747 pte_unmap_unlock(ptep
, ptl
);
3750 EXPORT_SYMBOL(follow_pfn
);
3752 #ifdef CONFIG_HAVE_IOREMAP_PROT
3753 int follow_phys(struct vm_area_struct
*vma
,
3754 unsigned long address
, unsigned int flags
,
3755 unsigned long *prot
, resource_size_t
*phys
)
3761 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
3764 if (follow_pte(vma
->vm_mm
, address
, &ptep
, &ptl
))
3768 if ((flags
& FOLL_WRITE
) && !pte_write(pte
))
3771 *prot
= pgprot_val(pte_pgprot(pte
));
3772 *phys
= (resource_size_t
)pte_pfn(pte
) << PAGE_SHIFT
;
3776 pte_unmap_unlock(ptep
, ptl
);
3781 int generic_access_phys(struct vm_area_struct
*vma
, unsigned long addr
,
3782 void *buf
, int len
, int write
)
3784 resource_size_t phys_addr
;
3785 unsigned long prot
= 0;
3786 void __iomem
*maddr
;
3787 int offset
= addr
& (PAGE_SIZE
-1);
3789 if (follow_phys(vma
, addr
, write
, &prot
, &phys_addr
))
3792 maddr
= ioremap_prot(phys_addr
, PAGE_SIZE
, prot
);
3794 memcpy_toio(maddr
+ offset
, buf
, len
);
3796 memcpy_fromio(buf
, maddr
+ offset
, len
);
3804 * Access another process' address space as given in mm. If non-NULL, use the
3805 * given task for page fault accounting.
3807 static int __access_remote_vm(struct task_struct
*tsk
, struct mm_struct
*mm
,
3808 unsigned long addr
, void *buf
, int len
, int write
)
3810 struct vm_area_struct
*vma
;
3811 void *old_buf
= buf
;
3813 down_read(&mm
->mmap_sem
);
3814 /* ignore errors, just check how much was successfully transferred */
3816 int bytes
, ret
, offset
;
3818 struct page
*page
= NULL
;
3820 ret
= get_user_pages(tsk
, mm
, addr
, 1,
3821 write
, 1, &page
, &vma
);
3824 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3825 * we can access using slightly different code.
3827 #ifdef CONFIG_HAVE_IOREMAP_PROT
3828 vma
= find_vma(mm
, addr
);
3829 if (!vma
|| vma
->vm_start
> addr
)
3831 if (vma
->vm_ops
&& vma
->vm_ops
->access
)
3832 ret
= vma
->vm_ops
->access(vma
, addr
, buf
,
3840 offset
= addr
& (PAGE_SIZE
-1);
3841 if (bytes
> PAGE_SIZE
-offset
)
3842 bytes
= PAGE_SIZE
-offset
;
3846 copy_to_user_page(vma
, page
, addr
,
3847 maddr
+ offset
, buf
, bytes
);
3848 set_page_dirty_lock(page
);
3850 copy_from_user_page(vma
, page
, addr
,
3851 buf
, maddr
+ offset
, bytes
);
3854 page_cache_release(page
);
3860 up_read(&mm
->mmap_sem
);
3862 return buf
- old_buf
;
3866 * access_remote_vm - access another process' address space
3867 * @mm: the mm_struct of the target address space
3868 * @addr: start address to access
3869 * @buf: source or destination buffer
3870 * @len: number of bytes to transfer
3871 * @write: whether the access is a write
3873 * The caller must hold a reference on @mm.
3875 int access_remote_vm(struct mm_struct
*mm
, unsigned long addr
,
3876 void *buf
, int len
, int write
)
3878 return __access_remote_vm(NULL
, mm
, addr
, buf
, len
, write
);
3882 * Access another process' address space.
3883 * Source/target buffer must be kernel space,
3884 * Do not walk the page table directly, use get_user_pages
3886 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
,
3887 void *buf
, int len
, int write
)
3889 struct mm_struct
*mm
;
3892 mm
= get_task_mm(tsk
);
3896 ret
= __access_remote_vm(tsk
, mm
, addr
, buf
, len
, write
);
3903 * Print the name of a VMA.
3905 void print_vma_addr(char *prefix
, unsigned long ip
)
3907 struct mm_struct
*mm
= current
->mm
;
3908 struct vm_area_struct
*vma
;
3911 * Do not print if we are in atomic
3912 * contexts (in exception stacks, etc.):
3914 if (preempt_count())
3917 down_read(&mm
->mmap_sem
);
3918 vma
= find_vma(mm
, ip
);
3919 if (vma
&& vma
->vm_file
) {
3920 struct file
*f
= vma
->vm_file
;
3921 char *buf
= (char *)__get_free_page(GFP_KERNEL
);
3925 p
= d_path(&f
->f_path
, buf
, PAGE_SIZE
);
3928 s
= strrchr(p
, '/');
3931 printk("%s%s[%lx+%lx]", prefix
, p
,
3933 vma
->vm_end
- vma
->vm_start
);
3934 free_page((unsigned long)buf
);
3937 up_read(¤t
->mm
->mmap_sem
);
3940 #ifdef CONFIG_PROVE_LOCKING
3941 void might_fault(void)
3944 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3945 * holding the mmap_sem, this is safe because kernel memory doesn't
3946 * get paged out, therefore we'll never actually fault, and the
3947 * below annotations will generate false positives.
3949 if (segment_eq(get_fs(), KERNEL_DS
))
3954 * it would be nicer only to annotate paths which are not under
3955 * pagefault_disable, however that requires a larger audit and
3956 * providing helpers like get_user_atomic.
3958 if (!in_atomic() && current
->mm
)
3959 might_lock_read(¤t
->mm
->mmap_sem
);
3961 EXPORT_SYMBOL(might_fault
);
3964 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
3965 static void clear_gigantic_page(struct page
*page
,
3967 unsigned int pages_per_huge_page
)
3970 struct page
*p
= page
;
3973 for (i
= 0; i
< pages_per_huge_page
;
3974 i
++, p
= mem_map_next(p
, page
, i
)) {
3976 clear_user_highpage(p
, addr
+ i
* PAGE_SIZE
);
3979 void clear_huge_page(struct page
*page
,
3980 unsigned long addr
, unsigned int pages_per_huge_page
)
3984 if (unlikely(pages_per_huge_page
> MAX_ORDER_NR_PAGES
)) {
3985 clear_gigantic_page(page
, addr
, pages_per_huge_page
);
3990 for (i
= 0; i
< pages_per_huge_page
; i
++) {
3992 clear_user_highpage(page
+ i
, addr
+ i
* PAGE_SIZE
);
3996 static void copy_user_gigantic_page(struct page
*dst
, struct page
*src
,
3998 struct vm_area_struct
*vma
,
3999 unsigned int pages_per_huge_page
)
4002 struct page
*dst_base
= dst
;
4003 struct page
*src_base
= src
;
4005 for (i
= 0; i
< pages_per_huge_page
; ) {
4007 copy_user_highpage(dst
, src
, addr
+ i
*PAGE_SIZE
, vma
);
4010 dst
= mem_map_next(dst
, dst_base
, i
);
4011 src
= mem_map_next(src
, src_base
, i
);
4015 void copy_user_huge_page(struct page
*dst
, struct page
*src
,
4016 unsigned long addr
, struct vm_area_struct
*vma
,
4017 unsigned int pages_per_huge_page
)
4021 if (unlikely(pages_per_huge_page
> MAX_ORDER_NR_PAGES
)) {
4022 copy_user_gigantic_page(dst
, src
, addr
, vma
,
4023 pages_per_huge_page
);
4028 for (i
= 0; i
< pages_per_huge_page
; i
++) {
4030 copy_user_highpage(dst
+ i
, src
+ i
, addr
+ i
*PAGE_SIZE
, vma
);
4033 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */