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
;
212 tlb
->fast_mode
= (num_possible_cpus() == 1);
213 tlb
->local
.next
= NULL
;
215 tlb
->local
.max
= ARRAY_SIZE(tlb
->__pages
);
216 tlb
->active
= &tlb
->local
;
218 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
223 void tlb_flush_mmu(struct mmu_gather
*tlb
)
225 struct mmu_gather_batch
*batch
;
227 if (!tlb
->need_flush
)
231 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
232 tlb_table_flush(tlb
);
235 if (tlb_fast_mode(tlb
))
238 for (batch
= &tlb
->local
; batch
; batch
= batch
->next
) {
239 free_pages_and_swap_cache(batch
->pages
, batch
->nr
);
242 tlb
->active
= &tlb
->local
;
246 * Called at the end of the shootdown operation to free up any resources
247 * that were required.
249 void tlb_finish_mmu(struct mmu_gather
*tlb
, unsigned long start
, unsigned long end
)
251 struct mmu_gather_batch
*batch
, *next
;
257 /* keep the page table cache within bounds */
260 for (batch
= tlb
->local
.next
; batch
; batch
= next
) {
262 free_pages((unsigned long)batch
, 0);
264 tlb
->local
.next
= NULL
;
268 * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
269 * handling the additional races in SMP caused by other CPUs caching valid
270 * mappings in their TLBs. Returns the number of free page slots left.
271 * When out of page slots we must call tlb_flush_mmu().
273 int __tlb_remove_page(struct mmu_gather
*tlb
, struct page
*page
)
275 struct mmu_gather_batch
*batch
;
277 VM_BUG_ON(!tlb
->need_flush
);
279 if (tlb_fast_mode(tlb
)) {
280 free_page_and_swap_cache(page
);
281 return 1; /* avoid calling tlb_flush_mmu() */
285 batch
->pages
[batch
->nr
++] = page
;
286 if (batch
->nr
== batch
->max
) {
287 if (!tlb_next_batch(tlb
))
291 VM_BUG_ON(batch
->nr
> batch
->max
);
293 return batch
->max
- batch
->nr
;
296 #endif /* HAVE_GENERIC_MMU_GATHER */
298 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
301 * See the comment near struct mmu_table_batch.
304 static void tlb_remove_table_smp_sync(void *arg
)
306 /* Simply deliver the interrupt */
309 static void tlb_remove_table_one(void *table
)
312 * This isn't an RCU grace period and hence the page-tables cannot be
313 * assumed to be actually RCU-freed.
315 * It is however sufficient for software page-table walkers that rely on
316 * IRQ disabling. See the comment near struct mmu_table_batch.
318 smp_call_function(tlb_remove_table_smp_sync
, NULL
, 1);
319 __tlb_remove_table(table
);
322 static void tlb_remove_table_rcu(struct rcu_head
*head
)
324 struct mmu_table_batch
*batch
;
327 batch
= container_of(head
, struct mmu_table_batch
, rcu
);
329 for (i
= 0; i
< batch
->nr
; i
++)
330 __tlb_remove_table(batch
->tables
[i
]);
332 free_page((unsigned long)batch
);
335 void tlb_table_flush(struct mmu_gather
*tlb
)
337 struct mmu_table_batch
**batch
= &tlb
->batch
;
340 call_rcu_sched(&(*batch
)->rcu
, tlb_remove_table_rcu
);
345 void tlb_remove_table(struct mmu_gather
*tlb
, void *table
)
347 struct mmu_table_batch
**batch
= &tlb
->batch
;
352 * When there's less then two users of this mm there cannot be a
353 * concurrent page-table walk.
355 if (atomic_read(&tlb
->mm
->mm_users
) < 2) {
356 __tlb_remove_table(table
);
360 if (*batch
== NULL
) {
361 *batch
= (struct mmu_table_batch
*)__get_free_page(GFP_NOWAIT
| __GFP_NOWARN
);
362 if (*batch
== NULL
) {
363 tlb_remove_table_one(table
);
368 (*batch
)->tables
[(*batch
)->nr
++] = table
;
369 if ((*batch
)->nr
== MAX_TABLE_BATCH
)
370 tlb_table_flush(tlb
);
373 #endif /* CONFIG_HAVE_RCU_TABLE_FREE */
376 * If a p?d_bad entry is found while walking page tables, report
377 * the error, before resetting entry to p?d_none. Usually (but
378 * very seldom) called out from the p?d_none_or_clear_bad macros.
381 void pgd_clear_bad(pgd_t
*pgd
)
387 void pud_clear_bad(pud_t
*pud
)
393 void pmd_clear_bad(pmd_t
*pmd
)
400 * Note: this doesn't free the actual pages themselves. That
401 * has been handled earlier when unmapping all the memory regions.
403 static void free_pte_range(struct mmu_gather
*tlb
, pmd_t
*pmd
,
406 pgtable_t token
= pmd_pgtable(*pmd
);
408 pte_free_tlb(tlb
, token
, addr
);
412 static inline void free_pmd_range(struct mmu_gather
*tlb
, pud_t
*pud
,
413 unsigned long addr
, unsigned long end
,
414 unsigned long floor
, unsigned long ceiling
)
421 pmd
= pmd_offset(pud
, addr
);
423 next
= pmd_addr_end(addr
, end
);
424 if (pmd_none_or_clear_bad(pmd
))
426 free_pte_range(tlb
, pmd
, addr
);
427 } while (pmd
++, addr
= next
, addr
!= end
);
437 if (end
- 1 > ceiling
- 1)
440 pmd
= pmd_offset(pud
, start
);
442 pmd_free_tlb(tlb
, pmd
, start
);
445 static inline void free_pud_range(struct mmu_gather
*tlb
, pgd_t
*pgd
,
446 unsigned long addr
, unsigned long end
,
447 unsigned long floor
, unsigned long ceiling
)
454 pud
= pud_offset(pgd
, addr
);
456 next
= pud_addr_end(addr
, end
);
457 if (pud_none_or_clear_bad(pud
))
459 free_pmd_range(tlb
, pud
, addr
, next
, floor
, ceiling
);
460 } while (pud
++, addr
= next
, addr
!= end
);
466 ceiling
&= PGDIR_MASK
;
470 if (end
- 1 > ceiling
- 1)
473 pud
= pud_offset(pgd
, start
);
475 pud_free_tlb(tlb
, pud
, start
);
479 * This function frees user-level page tables of a process.
481 * Must be called with pagetable lock held.
483 void free_pgd_range(struct mmu_gather
*tlb
,
484 unsigned long addr
, unsigned long end
,
485 unsigned long floor
, unsigned long ceiling
)
491 * The next few lines have given us lots of grief...
493 * Why are we testing PMD* at this top level? Because often
494 * there will be no work to do at all, and we'd prefer not to
495 * go all the way down to the bottom just to discover that.
497 * Why all these "- 1"s? Because 0 represents both the bottom
498 * of the address space and the top of it (using -1 for the
499 * top wouldn't help much: the masks would do the wrong thing).
500 * The rule is that addr 0 and floor 0 refer to the bottom of
501 * the address space, but end 0 and ceiling 0 refer to the top
502 * Comparisons need to use "end - 1" and "ceiling - 1" (though
503 * that end 0 case should be mythical).
505 * Wherever addr is brought up or ceiling brought down, we must
506 * be careful to reject "the opposite 0" before it confuses the
507 * subsequent tests. But what about where end is brought down
508 * by PMD_SIZE below? no, end can't go down to 0 there.
510 * Whereas we round start (addr) and ceiling down, by different
511 * masks at different levels, in order to test whether a table
512 * now has no other vmas using it, so can be freed, we don't
513 * bother to round floor or end up - the tests don't need that.
527 if (end
- 1 > ceiling
- 1)
532 pgd
= pgd_offset(tlb
->mm
, addr
);
534 next
= pgd_addr_end(addr
, end
);
535 if (pgd_none_or_clear_bad(pgd
))
537 free_pud_range(tlb
, pgd
, addr
, next
, floor
, ceiling
);
538 } while (pgd
++, addr
= next
, addr
!= end
);
541 void free_pgtables(struct mmu_gather
*tlb
, struct vm_area_struct
*vma
,
542 unsigned long floor
, unsigned long ceiling
)
545 struct vm_area_struct
*next
= vma
->vm_next
;
546 unsigned long addr
= vma
->vm_start
;
549 * Hide vma from rmap and truncate_pagecache before freeing
552 unlink_anon_vmas(vma
);
553 unlink_file_vma(vma
);
555 if (is_vm_hugetlb_page(vma
)) {
556 hugetlb_free_pgd_range(tlb
, addr
, vma
->vm_end
,
557 floor
, next
? next
->vm_start
: ceiling
);
560 * Optimization: gather nearby vmas into one call down
562 while (next
&& next
->vm_start
<= vma
->vm_end
+ PMD_SIZE
563 && !is_vm_hugetlb_page(next
)) {
566 unlink_anon_vmas(vma
);
567 unlink_file_vma(vma
);
569 free_pgd_range(tlb
, addr
, vma
->vm_end
,
570 floor
, next
? next
->vm_start
: ceiling
);
576 int __pte_alloc(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
577 pmd_t
*pmd
, unsigned long address
)
579 pgtable_t
new = pte_alloc_one(mm
, address
);
580 int wait_split_huge_page
;
585 * Ensure all pte setup (eg. pte page lock and page clearing) are
586 * visible before the pte is made visible to other CPUs by being
587 * put into page tables.
589 * The other side of the story is the pointer chasing in the page
590 * table walking code (when walking the page table without locking;
591 * ie. most of the time). Fortunately, these data accesses consist
592 * of a chain of data-dependent loads, meaning most CPUs (alpha
593 * being the notable exception) will already guarantee loads are
594 * seen in-order. See the alpha page table accessors for the
595 * smp_read_barrier_depends() barriers in page table walking code.
597 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
599 spin_lock(&mm
->page_table_lock
);
600 wait_split_huge_page
= 0;
601 if (likely(pmd_none(*pmd
))) { /* Has another populated it ? */
603 pmd_populate(mm
, pmd
, new);
605 } else if (unlikely(pmd_trans_splitting(*pmd
)))
606 wait_split_huge_page
= 1;
607 spin_unlock(&mm
->page_table_lock
);
610 if (wait_split_huge_page
)
611 wait_split_huge_page(vma
->anon_vma
, pmd
);
615 int __pte_alloc_kernel(pmd_t
*pmd
, unsigned long address
)
617 pte_t
*new = pte_alloc_one_kernel(&init_mm
, address
);
621 smp_wmb(); /* See comment in __pte_alloc */
623 spin_lock(&init_mm
.page_table_lock
);
624 if (likely(pmd_none(*pmd
))) { /* Has another populated it ? */
625 pmd_populate_kernel(&init_mm
, pmd
, new);
628 VM_BUG_ON(pmd_trans_splitting(*pmd
));
629 spin_unlock(&init_mm
.page_table_lock
);
631 pte_free_kernel(&init_mm
, new);
635 static inline void init_rss_vec(int *rss
)
637 memset(rss
, 0, sizeof(int) * NR_MM_COUNTERS
);
640 static inline void add_mm_rss_vec(struct mm_struct
*mm
, int *rss
)
644 if (current
->mm
== mm
)
646 for (i
= 0; i
< NR_MM_COUNTERS
; i
++)
648 add_mm_counter(mm
, i
, rss
[i
]);
652 * This function is called to print an error when a bad pte
653 * is found. For example, we might have a PFN-mapped pte in
654 * a region that doesn't allow it.
656 * The calling function must still handle the error.
658 static void print_bad_pte(struct vm_area_struct
*vma
, unsigned long addr
,
659 pte_t pte
, struct page
*page
)
661 pgd_t
*pgd
= pgd_offset(vma
->vm_mm
, addr
);
662 pud_t
*pud
= pud_offset(pgd
, addr
);
663 pmd_t
*pmd
= pmd_offset(pud
, addr
);
664 struct address_space
*mapping
;
666 static unsigned long resume
;
667 static unsigned long nr_shown
;
668 static unsigned long nr_unshown
;
671 * Allow a burst of 60 reports, then keep quiet for that minute;
672 * or allow a steady drip of one report per second.
674 if (nr_shown
== 60) {
675 if (time_before(jiffies
, resume
)) {
681 "BUG: Bad page map: %lu messages suppressed\n",
688 resume
= jiffies
+ 60 * HZ
;
690 mapping
= vma
->vm_file
? vma
->vm_file
->f_mapping
: NULL
;
691 index
= linear_page_index(vma
, addr
);
694 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
696 (long long)pte_val(pte
), (long long)pmd_val(*pmd
));
700 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
701 (void *)addr
, vma
->vm_flags
, vma
->anon_vma
, mapping
, index
);
703 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
706 print_symbol(KERN_ALERT
"vma->vm_ops->fault: %s\n",
707 (unsigned long)vma
->vm_ops
->fault
);
708 if (vma
->vm_file
&& vma
->vm_file
->f_op
)
709 print_symbol(KERN_ALERT
"vma->vm_file->f_op->mmap: %s\n",
710 (unsigned long)vma
->vm_file
->f_op
->mmap
);
712 add_taint(TAINT_BAD_PAGE
);
715 static inline int is_cow_mapping(vm_flags_t flags
)
717 return (flags
& (VM_SHARED
| VM_MAYWRITE
)) == VM_MAYWRITE
;
721 static inline int is_zero_pfn(unsigned long pfn
)
723 return pfn
== zero_pfn
;
728 static inline unsigned long my_zero_pfn(unsigned long addr
)
735 * vm_normal_page -- This function gets the "struct page" associated with a pte.
737 * "Special" mappings do not wish to be associated with a "struct page" (either
738 * it doesn't exist, or it exists but they don't want to touch it). In this
739 * case, NULL is returned here. "Normal" mappings do have a struct page.
741 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
742 * pte bit, in which case this function is trivial. Secondly, an architecture
743 * may not have a spare pte bit, which requires a more complicated scheme,
746 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
747 * special mapping (even if there are underlying and valid "struct pages").
748 * COWed pages of a VM_PFNMAP are always normal.
750 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
751 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
752 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
753 * mapping will always honor the rule
755 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
757 * And for normal mappings this is false.
759 * This restricts such mappings to be a linear translation from virtual address
760 * to pfn. To get around this restriction, we allow arbitrary mappings so long
761 * as the vma is not a COW mapping; in that case, we know that all ptes are
762 * special (because none can have been COWed).
765 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
767 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
768 * page" backing, however the difference is that _all_ pages with a struct
769 * page (that is, those where pfn_valid is true) are refcounted and considered
770 * normal pages by the VM. The disadvantage is that pages are refcounted
771 * (which can be slower and simply not an option for some PFNMAP users). The
772 * advantage is that we don't have to follow the strict linearity rule of
773 * PFNMAP mappings in order to support COWable mappings.
776 #ifdef __HAVE_ARCH_PTE_SPECIAL
777 # define HAVE_PTE_SPECIAL 1
779 # define HAVE_PTE_SPECIAL 0
781 struct page
*vm_normal_page(struct vm_area_struct
*vma
, unsigned long addr
,
784 unsigned long pfn
= pte_pfn(pte
);
786 if (HAVE_PTE_SPECIAL
) {
787 if (likely(!pte_special(pte
)))
789 if (vma
->vm_flags
& (VM_PFNMAP
| VM_MIXEDMAP
))
791 if (!is_zero_pfn(pfn
))
792 print_bad_pte(vma
, addr
, pte
, NULL
);
796 /* !HAVE_PTE_SPECIAL case follows: */
798 if (unlikely(vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
))) {
799 if (vma
->vm_flags
& VM_MIXEDMAP
) {
805 off
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
806 if (pfn
== vma
->vm_pgoff
+ off
)
808 if (!is_cow_mapping(vma
->vm_flags
))
813 if (is_zero_pfn(pfn
))
816 if (unlikely(pfn
> highest_memmap_pfn
)) {
817 print_bad_pte(vma
, addr
, pte
, NULL
);
822 * NOTE! We still have PageReserved() pages in the page tables.
823 * eg. VDSO mappings can cause them to exist.
826 return pfn_to_page(pfn
);
830 * copy one vm_area from one task to the other. Assumes the page tables
831 * already present in the new task to be cleared in the whole range
832 * covered by this vma.
835 static inline unsigned long
836 copy_one_pte(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
837 pte_t
*dst_pte
, pte_t
*src_pte
, struct vm_area_struct
*vma
,
838 unsigned long addr
, int *rss
)
840 unsigned long vm_flags
= vma
->vm_flags
;
841 pte_t pte
= *src_pte
;
844 /* pte contains position in swap or file, so copy. */
845 if (unlikely(!pte_present(pte
))) {
846 if (!pte_file(pte
)) {
847 swp_entry_t entry
= pte_to_swp_entry(pte
);
849 if (swap_duplicate(entry
) < 0)
852 /* make sure dst_mm is on swapoff's mmlist. */
853 if (unlikely(list_empty(&dst_mm
->mmlist
))) {
854 spin_lock(&mmlist_lock
);
855 if (list_empty(&dst_mm
->mmlist
))
856 list_add(&dst_mm
->mmlist
,
858 spin_unlock(&mmlist_lock
);
860 if (likely(!non_swap_entry(entry
)))
862 else if (is_migration_entry(entry
)) {
863 page
= migration_entry_to_page(entry
);
870 if (is_write_migration_entry(entry
) &&
871 is_cow_mapping(vm_flags
)) {
873 * COW mappings require pages in both
874 * parent and child to be set to read.
876 make_migration_entry_read(&entry
);
877 pte
= swp_entry_to_pte(entry
);
878 set_pte_at(src_mm
, addr
, src_pte
, pte
);
886 * If it's a COW mapping, write protect it both
887 * in the parent and the child
889 if (is_cow_mapping(vm_flags
)) {
890 ptep_set_wrprotect(src_mm
, addr
, src_pte
);
891 pte
= pte_wrprotect(pte
);
895 * If it's a shared mapping, mark it clean in
898 if (vm_flags
& VM_SHARED
)
899 pte
= pte_mkclean(pte
);
900 pte
= pte_mkold(pte
);
902 page
= vm_normal_page(vma
, addr
, pte
);
913 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
917 int copy_pte_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
918 pmd_t
*dst_pmd
, pmd_t
*src_pmd
, struct vm_area_struct
*vma
,
919 unsigned long addr
, unsigned long end
)
921 pte_t
*orig_src_pte
, *orig_dst_pte
;
922 pte_t
*src_pte
, *dst_pte
;
923 spinlock_t
*src_ptl
, *dst_ptl
;
925 int rss
[NR_MM_COUNTERS
];
926 swp_entry_t entry
= (swp_entry_t
){0};
931 dst_pte
= pte_alloc_map_lock(dst_mm
, dst_pmd
, addr
, &dst_ptl
);
934 src_pte
= pte_offset_map(src_pmd
, addr
);
935 src_ptl
= pte_lockptr(src_mm
, src_pmd
);
936 spin_lock_nested(src_ptl
, SINGLE_DEPTH_NESTING
);
937 orig_src_pte
= src_pte
;
938 orig_dst_pte
= dst_pte
;
939 arch_enter_lazy_mmu_mode();
943 * We are holding two locks at this point - either of them
944 * could generate latencies in another task on another CPU.
946 if (progress
>= 32) {
948 if (need_resched() ||
949 spin_needbreak(src_ptl
) || spin_needbreak(dst_ptl
))
952 if (pte_none(*src_pte
)) {
956 entry
.val
= copy_one_pte(dst_mm
, src_mm
, dst_pte
, src_pte
,
961 } while (dst_pte
++, src_pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
963 arch_leave_lazy_mmu_mode();
964 spin_unlock(src_ptl
);
965 pte_unmap(orig_src_pte
);
966 add_mm_rss_vec(dst_mm
, rss
);
967 pte_unmap_unlock(orig_dst_pte
, dst_ptl
);
971 if (add_swap_count_continuation(entry
, GFP_KERNEL
) < 0)
980 static inline int copy_pmd_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
981 pud_t
*dst_pud
, pud_t
*src_pud
, struct vm_area_struct
*vma
,
982 unsigned long addr
, unsigned long end
)
984 pmd_t
*src_pmd
, *dst_pmd
;
987 dst_pmd
= pmd_alloc(dst_mm
, dst_pud
, addr
);
990 src_pmd
= pmd_offset(src_pud
, addr
);
992 next
= pmd_addr_end(addr
, end
);
993 if (pmd_trans_huge(*src_pmd
)) {
995 VM_BUG_ON(next
-addr
!= HPAGE_PMD_SIZE
);
996 err
= copy_huge_pmd(dst_mm
, src_mm
,
997 dst_pmd
, src_pmd
, addr
, vma
);
1004 if (pmd_none_or_clear_bad(src_pmd
))
1006 if (copy_pte_range(dst_mm
, src_mm
, dst_pmd
, src_pmd
,
1009 } while (dst_pmd
++, src_pmd
++, addr
= next
, addr
!= end
);
1013 static inline int copy_pud_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
1014 pgd_t
*dst_pgd
, pgd_t
*src_pgd
, struct vm_area_struct
*vma
,
1015 unsigned long addr
, unsigned long end
)
1017 pud_t
*src_pud
, *dst_pud
;
1020 dst_pud
= pud_alloc(dst_mm
, dst_pgd
, addr
);
1023 src_pud
= pud_offset(src_pgd
, addr
);
1025 next
= pud_addr_end(addr
, end
);
1026 if (pud_none_or_clear_bad(src_pud
))
1028 if (copy_pmd_range(dst_mm
, src_mm
, dst_pud
, src_pud
,
1031 } while (dst_pud
++, src_pud
++, addr
= next
, addr
!= end
);
1035 int copy_page_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
1036 struct vm_area_struct
*vma
)
1038 pgd_t
*src_pgd
, *dst_pgd
;
1040 unsigned long addr
= vma
->vm_start
;
1041 unsigned long end
= vma
->vm_end
;
1045 * Don't copy ptes where a page fault will fill them correctly.
1046 * Fork becomes much lighter when there are big shared or private
1047 * readonly mappings. The tradeoff is that copy_page_range is more
1048 * efficient than faulting.
1050 if (!(vma
->vm_flags
& (VM_HUGETLB
| VM_NONLINEAR
|
1051 VM_PFNMAP
| VM_MIXEDMAP
))) {
1056 if (is_vm_hugetlb_page(vma
))
1057 return copy_hugetlb_page_range(dst_mm
, src_mm
, vma
);
1059 if (unlikely(vma
->vm_flags
& VM_PFNMAP
)) {
1061 * We do not free on error cases below as remove_vma
1062 * gets called on error from higher level routine
1064 ret
= track_pfn_copy(vma
);
1070 * We need to invalidate the secondary MMU mappings only when
1071 * there could be a permission downgrade on the ptes of the
1072 * parent mm. And a permission downgrade will only happen if
1073 * is_cow_mapping() returns true.
1075 if (is_cow_mapping(vma
->vm_flags
))
1076 mmu_notifier_invalidate_range_start(src_mm
, addr
, end
);
1079 dst_pgd
= pgd_offset(dst_mm
, addr
);
1080 src_pgd
= pgd_offset(src_mm
, addr
);
1082 next
= pgd_addr_end(addr
, end
);
1083 if (pgd_none_or_clear_bad(src_pgd
))
1085 if (unlikely(copy_pud_range(dst_mm
, src_mm
, dst_pgd
, src_pgd
,
1086 vma
, addr
, next
))) {
1090 } while (dst_pgd
++, src_pgd
++, addr
= next
, addr
!= end
);
1092 if (is_cow_mapping(vma
->vm_flags
))
1093 mmu_notifier_invalidate_range_end(src_mm
,
1094 vma
->vm_start
, end
);
1098 static unsigned long zap_pte_range(struct mmu_gather
*tlb
,
1099 struct vm_area_struct
*vma
, pmd_t
*pmd
,
1100 unsigned long addr
, unsigned long end
,
1101 struct zap_details
*details
)
1103 struct mm_struct
*mm
= tlb
->mm
;
1104 int force_flush
= 0;
1105 int rss
[NR_MM_COUNTERS
];
1112 start_pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
1114 arch_enter_lazy_mmu_mode();
1117 if (pte_none(ptent
)) {
1121 if (pte_present(ptent
)) {
1124 page
= vm_normal_page(vma
, addr
, ptent
);
1125 if (unlikely(details
) && page
) {
1127 * unmap_shared_mapping_pages() wants to
1128 * invalidate cache without truncating:
1129 * unmap shared but keep private pages.
1131 if (details
->check_mapping
&&
1132 details
->check_mapping
!= page
->mapping
)
1135 * Each page->index must be checked when
1136 * invalidating or truncating nonlinear.
1138 if (details
->nonlinear_vma
&&
1139 (page
->index
< details
->first_index
||
1140 page
->index
> details
->last_index
))
1143 ptent
= ptep_get_and_clear_full(mm
, addr
, pte
,
1145 tlb_remove_tlb_entry(tlb
, pte
, addr
);
1146 if (unlikely(!page
))
1148 if (unlikely(details
) && details
->nonlinear_vma
1149 && linear_page_index(details
->nonlinear_vma
,
1150 addr
) != page
->index
)
1151 set_pte_at(mm
, addr
, pte
,
1152 pgoff_to_pte(page
->index
));
1154 rss
[MM_ANONPAGES
]--;
1156 if (pte_dirty(ptent
))
1157 set_page_dirty(page
);
1158 if (pte_young(ptent
) &&
1159 likely(!VM_SequentialReadHint(vma
)))
1160 mark_page_accessed(page
);
1161 rss
[MM_FILEPAGES
]--;
1163 page_remove_rmap(page
);
1164 if (unlikely(page_mapcount(page
) < 0))
1165 print_bad_pte(vma
, addr
, ptent
, page
);
1166 force_flush
= !__tlb_remove_page(tlb
, page
);
1172 * If details->check_mapping, we leave swap entries;
1173 * if details->nonlinear_vma, we leave file entries.
1175 if (unlikely(details
))
1177 if (pte_file(ptent
)) {
1178 if (unlikely(!(vma
->vm_flags
& VM_NONLINEAR
)))
1179 print_bad_pte(vma
, addr
, ptent
, NULL
);
1181 swp_entry_t entry
= pte_to_swp_entry(ptent
);
1183 if (!non_swap_entry(entry
))
1185 else if (is_migration_entry(entry
)) {
1188 page
= migration_entry_to_page(entry
);
1191 rss
[MM_ANONPAGES
]--;
1193 rss
[MM_FILEPAGES
]--;
1195 if (unlikely(!free_swap_and_cache(entry
)))
1196 print_bad_pte(vma
, addr
, ptent
, NULL
);
1198 pte_clear_not_present_full(mm
, addr
, pte
, tlb
->fullmm
);
1199 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1201 add_mm_rss_vec(mm
, rss
);
1202 arch_leave_lazy_mmu_mode();
1203 pte_unmap_unlock(start_pte
, ptl
);
1206 * mmu_gather ran out of room to batch pages, we break out of
1207 * the PTE lock to avoid doing the potential expensive TLB invalidate
1208 * and page-free while holding it.
1213 #ifdef HAVE_GENERIC_MMU_GATHER
1225 static inline unsigned long zap_pmd_range(struct mmu_gather
*tlb
,
1226 struct vm_area_struct
*vma
, pud_t
*pud
,
1227 unsigned long addr
, unsigned long end
,
1228 struct zap_details
*details
)
1233 pmd
= pmd_offset(pud
, addr
);
1235 next
= pmd_addr_end(addr
, end
);
1236 if (pmd_trans_huge(*pmd
)) {
1237 if (next
- addr
!= HPAGE_PMD_SIZE
) {
1238 #ifdef CONFIG_DEBUG_VM
1239 if (!rwsem_is_locked(&tlb
->mm
->mmap_sem
)) {
1240 pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n",
1241 __func__
, addr
, end
,
1247 split_huge_page_pmd(vma
->vm_mm
, pmd
);
1248 } else if (zap_huge_pmd(tlb
, vma
, pmd
, addr
))
1253 * Here there can be other concurrent MADV_DONTNEED or
1254 * trans huge page faults running, and if the pmd is
1255 * none or trans huge it can change under us. This is
1256 * because MADV_DONTNEED holds the mmap_sem in read
1259 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
1261 next
= zap_pte_range(tlb
, vma
, pmd
, addr
, next
, details
);
1264 } while (pmd
++, addr
= next
, addr
!= end
);
1269 static inline unsigned long zap_pud_range(struct mmu_gather
*tlb
,
1270 struct vm_area_struct
*vma
, pgd_t
*pgd
,
1271 unsigned long addr
, unsigned long end
,
1272 struct zap_details
*details
)
1277 pud
= pud_offset(pgd
, addr
);
1279 next
= pud_addr_end(addr
, end
);
1280 if (pud_none_or_clear_bad(pud
))
1282 next
= zap_pmd_range(tlb
, vma
, pud
, addr
, next
, details
);
1283 } while (pud
++, addr
= next
, addr
!= end
);
1288 static void unmap_page_range(struct mmu_gather
*tlb
,
1289 struct vm_area_struct
*vma
,
1290 unsigned long addr
, unsigned long end
,
1291 struct zap_details
*details
)
1296 if (details
&& !details
->check_mapping
&& !details
->nonlinear_vma
)
1299 BUG_ON(addr
>= end
);
1300 mem_cgroup_uncharge_start();
1301 tlb_start_vma(tlb
, vma
);
1302 pgd
= pgd_offset(vma
->vm_mm
, addr
);
1304 next
= pgd_addr_end(addr
, end
);
1305 if (pgd_none_or_clear_bad(pgd
))
1307 next
= zap_pud_range(tlb
, vma
, pgd
, addr
, next
, details
);
1308 } while (pgd
++, addr
= next
, addr
!= end
);
1309 tlb_end_vma(tlb
, vma
);
1310 mem_cgroup_uncharge_end();
1314 static void unmap_single_vma(struct mmu_gather
*tlb
,
1315 struct vm_area_struct
*vma
, unsigned long start_addr
,
1316 unsigned long end_addr
,
1317 struct zap_details
*details
)
1319 unsigned long start
= max(vma
->vm_start
, start_addr
);
1322 if (start
>= vma
->vm_end
)
1324 end
= min(vma
->vm_end
, end_addr
);
1325 if (end
<= vma
->vm_start
)
1329 uprobe_munmap(vma
, start
, end
);
1331 if (unlikely(vma
->vm_flags
& VM_PFNMAP
))
1332 untrack_pfn(vma
, 0, 0);
1335 if (unlikely(is_vm_hugetlb_page(vma
))) {
1337 * It is undesirable to test vma->vm_file as it
1338 * should be non-null for valid hugetlb area.
1339 * However, vm_file will be NULL in the error
1340 * cleanup path of do_mmap_pgoff. When
1341 * hugetlbfs ->mmap method fails,
1342 * do_mmap_pgoff() nullifies vma->vm_file
1343 * before calling this function to clean up.
1344 * Since no pte has actually been setup, it is
1345 * safe to do nothing in this case.
1348 mutex_lock(&vma
->vm_file
->f_mapping
->i_mmap_mutex
);
1349 __unmap_hugepage_range_final(tlb
, vma
, start
, end
, NULL
);
1350 mutex_unlock(&vma
->vm_file
->f_mapping
->i_mmap_mutex
);
1353 unmap_page_range(tlb
, vma
, start
, end
, details
);
1358 * unmap_vmas - unmap a range of memory covered by a list of vma's
1359 * @tlb: address of the caller's struct mmu_gather
1360 * @vma: the starting vma
1361 * @start_addr: virtual address at which to start unmapping
1362 * @end_addr: virtual address at which to end unmapping
1364 * Unmap all pages in the vma list.
1366 * Only addresses between `start' and `end' will be unmapped.
1368 * The VMA list must be sorted in ascending virtual address order.
1370 * unmap_vmas() assumes that the caller will flush the whole unmapped address
1371 * range after unmap_vmas() returns. So the only responsibility here is to
1372 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1373 * drops the lock and schedules.
1375 void unmap_vmas(struct mmu_gather
*tlb
,
1376 struct vm_area_struct
*vma
, unsigned long start_addr
,
1377 unsigned long end_addr
)
1379 struct mm_struct
*mm
= vma
->vm_mm
;
1381 mmu_notifier_invalidate_range_start(mm
, start_addr
, end_addr
);
1382 for ( ; vma
&& vma
->vm_start
< end_addr
; vma
= vma
->vm_next
)
1383 unmap_single_vma(tlb
, vma
, start_addr
, end_addr
, NULL
);
1384 mmu_notifier_invalidate_range_end(mm
, start_addr
, end_addr
);
1388 * zap_page_range - remove user pages in a given range
1389 * @vma: vm_area_struct holding the applicable pages
1390 * @start: starting address of pages to zap
1391 * @size: number of bytes to zap
1392 * @details: details of nonlinear truncation or shared cache invalidation
1394 * Caller must protect the VMA list
1396 void zap_page_range(struct vm_area_struct
*vma
, unsigned long start
,
1397 unsigned long size
, struct zap_details
*details
)
1399 struct mm_struct
*mm
= vma
->vm_mm
;
1400 struct mmu_gather tlb
;
1401 unsigned long end
= start
+ size
;
1404 tlb_gather_mmu(&tlb
, mm
, 0);
1405 update_hiwater_rss(mm
);
1406 mmu_notifier_invalidate_range_start(mm
, start
, end
);
1407 for ( ; vma
&& vma
->vm_start
< end
; vma
= vma
->vm_next
)
1408 unmap_single_vma(&tlb
, vma
, start
, end
, details
);
1409 mmu_notifier_invalidate_range_end(mm
, start
, end
);
1410 tlb_finish_mmu(&tlb
, start
, end
);
1414 * zap_page_range_single - remove user pages in a given range
1415 * @vma: vm_area_struct holding the applicable pages
1416 * @address: starting address of pages to zap
1417 * @size: number of bytes to zap
1418 * @details: details of nonlinear truncation or shared cache invalidation
1420 * The range must fit into one VMA.
1422 static void zap_page_range_single(struct vm_area_struct
*vma
, unsigned long address
,
1423 unsigned long size
, struct zap_details
*details
)
1425 struct mm_struct
*mm
= vma
->vm_mm
;
1426 struct mmu_gather tlb
;
1427 unsigned long end
= address
+ size
;
1430 tlb_gather_mmu(&tlb
, mm
, 0);
1431 update_hiwater_rss(mm
);
1432 mmu_notifier_invalidate_range_start(mm
, address
, end
);
1433 unmap_single_vma(&tlb
, vma
, address
, end
, details
);
1434 mmu_notifier_invalidate_range_end(mm
, address
, end
);
1435 tlb_finish_mmu(&tlb
, address
, end
);
1439 * zap_vma_ptes - remove ptes mapping the vma
1440 * @vma: vm_area_struct holding ptes to be zapped
1441 * @address: starting address of pages to zap
1442 * @size: number of bytes to zap
1444 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1446 * The entire address range must be fully contained within the vma.
1448 * Returns 0 if successful.
1450 int zap_vma_ptes(struct vm_area_struct
*vma
, unsigned long address
,
1453 if (address
< vma
->vm_start
|| address
+ size
> vma
->vm_end
||
1454 !(vma
->vm_flags
& VM_PFNMAP
))
1456 zap_page_range_single(vma
, address
, size
, NULL
);
1459 EXPORT_SYMBOL_GPL(zap_vma_ptes
);
1462 * follow_page - look up a page descriptor from a user-virtual address
1463 * @vma: vm_area_struct mapping @address
1464 * @address: virtual address to look up
1465 * @flags: flags modifying lookup behaviour
1467 * @flags can have FOLL_ flags set, defined in <linux/mm.h>
1469 * Returns the mapped (struct page *), %NULL if no mapping exists, or
1470 * an error pointer if there is a mapping to something not represented
1471 * by a page descriptor (see also vm_normal_page()).
1473 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
1482 struct mm_struct
*mm
= vma
->vm_mm
;
1484 page
= follow_huge_addr(mm
, address
, flags
& FOLL_WRITE
);
1485 if (!IS_ERR(page
)) {
1486 BUG_ON(flags
& FOLL_GET
);
1491 pgd
= pgd_offset(mm
, address
);
1492 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
1495 pud
= pud_offset(pgd
, address
);
1498 if (pud_huge(*pud
) && vma
->vm_flags
& VM_HUGETLB
) {
1499 BUG_ON(flags
& FOLL_GET
);
1500 page
= follow_huge_pud(mm
, address
, pud
, flags
& FOLL_WRITE
);
1503 if (unlikely(pud_bad(*pud
)))
1506 pmd
= pmd_offset(pud
, address
);
1509 if (pmd_huge(*pmd
) && vma
->vm_flags
& VM_HUGETLB
) {
1510 BUG_ON(flags
& FOLL_GET
);
1511 page
= follow_huge_pmd(mm
, address
, pmd
, flags
& FOLL_WRITE
);
1514 if (pmd_trans_huge(*pmd
)) {
1515 if (flags
& FOLL_SPLIT
) {
1516 split_huge_page_pmd(mm
, pmd
);
1517 goto split_fallthrough
;
1519 spin_lock(&mm
->page_table_lock
);
1520 if (likely(pmd_trans_huge(*pmd
))) {
1521 if (unlikely(pmd_trans_splitting(*pmd
))) {
1522 spin_unlock(&mm
->page_table_lock
);
1523 wait_split_huge_page(vma
->anon_vma
, pmd
);
1525 page
= follow_trans_huge_pmd(mm
, address
,
1527 spin_unlock(&mm
->page_table_lock
);
1531 spin_unlock(&mm
->page_table_lock
);
1535 if (unlikely(pmd_bad(*pmd
)))
1538 ptep
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1541 if (!pte_present(pte
))
1543 if ((flags
& FOLL_WRITE
) && !pte_write(pte
))
1546 page
= vm_normal_page(vma
, address
, pte
);
1547 if (unlikely(!page
)) {
1548 if ((flags
& FOLL_DUMP
) ||
1549 !is_zero_pfn(pte_pfn(pte
)))
1551 page
= pte_page(pte
);
1554 if (flags
& FOLL_GET
)
1555 get_page_foll(page
);
1556 if (flags
& FOLL_TOUCH
) {
1557 if ((flags
& FOLL_WRITE
) &&
1558 !pte_dirty(pte
) && !PageDirty(page
))
1559 set_page_dirty(page
);
1561 * pte_mkyoung() would be more correct here, but atomic care
1562 * is needed to avoid losing the dirty bit: it is easier to use
1563 * mark_page_accessed().
1565 mark_page_accessed(page
);
1567 if ((flags
& FOLL_MLOCK
) && (vma
->vm_flags
& VM_LOCKED
)) {
1569 * The preliminary mapping check is mainly to avoid the
1570 * pointless overhead of lock_page on the ZERO_PAGE
1571 * which might bounce very badly if there is contention.
1573 * If the page is already locked, we don't need to
1574 * handle it now - vmscan will handle it later if and
1575 * when it attempts to reclaim the page.
1577 if (page
->mapping
&& trylock_page(page
)) {
1578 lru_add_drain(); /* push cached pages to LRU */
1580 * Because we lock page here and migration is
1581 * blocked by the pte's page reference, we need
1582 * only check for file-cache page truncation.
1585 mlock_vma_page(page
);
1590 pte_unmap_unlock(ptep
, ptl
);
1595 pte_unmap_unlock(ptep
, ptl
);
1596 return ERR_PTR(-EFAULT
);
1599 pte_unmap_unlock(ptep
, ptl
);
1605 * When core dumping an enormous anonymous area that nobody
1606 * has touched so far, we don't want to allocate unnecessary pages or
1607 * page tables. Return error instead of NULL to skip handle_mm_fault,
1608 * then get_dump_page() will return NULL to leave a hole in the dump.
1609 * But we can only make this optimization where a hole would surely
1610 * be zero-filled if handle_mm_fault() actually did handle it.
1612 if ((flags
& FOLL_DUMP
) &&
1613 (!vma
->vm_ops
|| !vma
->vm_ops
->fault
))
1614 return ERR_PTR(-EFAULT
);
1618 static inline int stack_guard_page(struct vm_area_struct
*vma
, unsigned long addr
)
1620 return stack_guard_page_start(vma
, addr
) ||
1621 stack_guard_page_end(vma
, addr
+PAGE_SIZE
);
1625 * __get_user_pages() - pin user pages in memory
1626 * @tsk: task_struct of target task
1627 * @mm: mm_struct of target mm
1628 * @start: starting user address
1629 * @nr_pages: number of pages from start to pin
1630 * @gup_flags: flags modifying pin behaviour
1631 * @pages: array that receives pointers to the pages pinned.
1632 * Should be at least nr_pages long. Or NULL, if caller
1633 * only intends to ensure the pages are faulted in.
1634 * @vmas: array of pointers to vmas corresponding to each page.
1635 * Or NULL if the caller does not require them.
1636 * @nonblocking: whether waiting for disk IO or mmap_sem contention
1638 * Returns number of pages pinned. This may be fewer than the number
1639 * requested. If nr_pages is 0 or negative, returns 0. If no pages
1640 * were pinned, returns -errno. Each page returned must be released
1641 * with a put_page() call when it is finished with. vmas will only
1642 * remain valid while mmap_sem is held.
1644 * Must be called with mmap_sem held for read or write.
1646 * __get_user_pages walks a process's page tables and takes a reference to
1647 * each struct page that each user address corresponds to at a given
1648 * instant. That is, it takes the page that would be accessed if a user
1649 * thread accesses the given user virtual address at that instant.
1651 * This does not guarantee that the page exists in the user mappings when
1652 * __get_user_pages returns, and there may even be a completely different
1653 * page there in some cases (eg. if mmapped pagecache has been invalidated
1654 * and subsequently re faulted). However it does guarantee that the page
1655 * won't be freed completely. And mostly callers simply care that the page
1656 * contains data that was valid *at some point in time*. Typically, an IO
1657 * or similar operation cannot guarantee anything stronger anyway because
1658 * locks can't be held over the syscall boundary.
1660 * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If
1661 * the page is written to, set_page_dirty (or set_page_dirty_lock, as
1662 * appropriate) must be called after the page is finished with, and
1663 * before put_page is called.
1665 * If @nonblocking != NULL, __get_user_pages will not wait for disk IO
1666 * or mmap_sem contention, and if waiting is needed to pin all pages,
1667 * *@nonblocking will be set to 0.
1669 * In most cases, get_user_pages or get_user_pages_fast should be used
1670 * instead of __get_user_pages. __get_user_pages should be used only if
1671 * you need some special @gup_flags.
1673 int __get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
1674 unsigned long start
, int nr_pages
, unsigned int gup_flags
,
1675 struct page
**pages
, struct vm_area_struct
**vmas
,
1679 unsigned long vm_flags
;
1684 VM_BUG_ON(!!pages
!= !!(gup_flags
& FOLL_GET
));
1687 * Require read or write permissions.
1688 * If FOLL_FORCE is set, we only require the "MAY" flags.
1690 vm_flags
= (gup_flags
& FOLL_WRITE
) ?
1691 (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
1692 vm_flags
&= (gup_flags
& FOLL_FORCE
) ?
1693 (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
1697 struct vm_area_struct
*vma
;
1699 vma
= find_extend_vma(mm
, start
);
1700 if (!vma
&& in_gate_area(mm
, start
)) {
1701 unsigned long pg
= start
& PAGE_MASK
;
1707 /* user gate pages are read-only */
1708 if (gup_flags
& FOLL_WRITE
)
1709 return i
? : -EFAULT
;
1711 pgd
= pgd_offset_k(pg
);
1713 pgd
= pgd_offset_gate(mm
, pg
);
1714 BUG_ON(pgd_none(*pgd
));
1715 pud
= pud_offset(pgd
, pg
);
1716 BUG_ON(pud_none(*pud
));
1717 pmd
= pmd_offset(pud
, pg
);
1719 return i
? : -EFAULT
;
1720 VM_BUG_ON(pmd_trans_huge(*pmd
));
1721 pte
= pte_offset_map(pmd
, pg
);
1722 if (pte_none(*pte
)) {
1724 return i
? : -EFAULT
;
1726 vma
= get_gate_vma(mm
);
1730 page
= vm_normal_page(vma
, start
, *pte
);
1732 if (!(gup_flags
& FOLL_DUMP
) &&
1733 is_zero_pfn(pte_pfn(*pte
)))
1734 page
= pte_page(*pte
);
1737 return i
? : -EFAULT
;
1748 (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) ||
1749 !(vm_flags
& vma
->vm_flags
))
1750 return i
? : -EFAULT
;
1752 if (is_vm_hugetlb_page(vma
)) {
1753 i
= follow_hugetlb_page(mm
, vma
, pages
, vmas
,
1754 &start
, &nr_pages
, i
, gup_flags
);
1760 unsigned int foll_flags
= gup_flags
;
1763 * If we have a pending SIGKILL, don't keep faulting
1764 * pages and potentially allocating memory.
1766 if (unlikely(fatal_signal_pending(current
)))
1767 return i
? i
: -ERESTARTSYS
;
1770 while (!(page
= follow_page(vma
, start
, foll_flags
))) {
1772 unsigned int fault_flags
= 0;
1774 /* For mlock, just skip the stack guard page. */
1775 if (foll_flags
& FOLL_MLOCK
) {
1776 if (stack_guard_page(vma
, start
))
1779 if (foll_flags
& FOLL_WRITE
)
1780 fault_flags
|= FAULT_FLAG_WRITE
;
1782 fault_flags
|= FAULT_FLAG_ALLOW_RETRY
;
1783 if (foll_flags
& FOLL_NOWAIT
)
1784 fault_flags
|= (FAULT_FLAG_ALLOW_RETRY
| FAULT_FLAG_RETRY_NOWAIT
);
1786 ret
= handle_mm_fault(mm
, vma
, start
,
1789 if (ret
& VM_FAULT_ERROR
) {
1790 if (ret
& VM_FAULT_OOM
)
1791 return i
? i
: -ENOMEM
;
1792 if (ret
& (VM_FAULT_HWPOISON
|
1793 VM_FAULT_HWPOISON_LARGE
)) {
1796 else if (gup_flags
& FOLL_HWPOISON
)
1801 if (ret
& VM_FAULT_SIGBUS
)
1802 return i
? i
: -EFAULT
;
1807 if (ret
& VM_FAULT_MAJOR
)
1813 if (ret
& VM_FAULT_RETRY
) {
1820 * The VM_FAULT_WRITE bit tells us that
1821 * do_wp_page has broken COW when necessary,
1822 * even if maybe_mkwrite decided not to set
1823 * pte_write. We can thus safely do subsequent
1824 * page lookups as if they were reads. But only
1825 * do so when looping for pte_write is futile:
1826 * in some cases userspace may also be wanting
1827 * to write to the gotten user page, which a
1828 * read fault here might prevent (a readonly
1829 * page might get reCOWed by userspace write).
1831 if ((ret
& VM_FAULT_WRITE
) &&
1832 !(vma
->vm_flags
& VM_WRITE
))
1833 foll_flags
&= ~FOLL_WRITE
;
1838 return i
? i
: PTR_ERR(page
);
1842 flush_anon_page(vma
, page
, start
);
1843 flush_dcache_page(page
);
1851 } while (nr_pages
&& start
< vma
->vm_end
);
1855 EXPORT_SYMBOL(__get_user_pages
);
1858 * fixup_user_fault() - manually resolve a user page fault
1859 * @tsk: the task_struct to use for page fault accounting, or
1860 * NULL if faults are not to be recorded.
1861 * @mm: mm_struct of target mm
1862 * @address: user address
1863 * @fault_flags:flags to pass down to handle_mm_fault()
1865 * This is meant to be called in the specific scenario where for locking reasons
1866 * we try to access user memory in atomic context (within a pagefault_disable()
1867 * section), this returns -EFAULT, and we want to resolve the user fault before
1870 * Typically this is meant to be used by the futex code.
1872 * The main difference with get_user_pages() is that this function will
1873 * unconditionally call handle_mm_fault() which will in turn perform all the
1874 * necessary SW fixup of the dirty and young bits in the PTE, while
1875 * handle_mm_fault() only guarantees to update these in the struct page.
1877 * This is important for some architectures where those bits also gate the
1878 * access permission to the page because they are maintained in software. On
1879 * such architectures, gup() will not be enough to make a subsequent access
1882 * This should be called with the mm_sem held for read.
1884 int fixup_user_fault(struct task_struct
*tsk
, struct mm_struct
*mm
,
1885 unsigned long address
, unsigned int fault_flags
)
1887 struct vm_area_struct
*vma
;
1890 vma
= find_extend_vma(mm
, address
);
1891 if (!vma
|| address
< vma
->vm_start
)
1894 ret
= handle_mm_fault(mm
, vma
, address
, fault_flags
);
1895 if (ret
& VM_FAULT_ERROR
) {
1896 if (ret
& VM_FAULT_OOM
)
1898 if (ret
& (VM_FAULT_HWPOISON
| VM_FAULT_HWPOISON_LARGE
))
1900 if (ret
& VM_FAULT_SIGBUS
)
1905 if (ret
& VM_FAULT_MAJOR
)
1914 * get_user_pages() - pin user pages in memory
1915 * @tsk: the task_struct to use for page fault accounting, or
1916 * NULL if faults are not to be recorded.
1917 * @mm: mm_struct of target mm
1918 * @start: starting user address
1919 * @nr_pages: number of pages from start to pin
1920 * @write: whether pages will be written to by the caller
1921 * @force: whether to force write access even if user mapping is
1922 * readonly. This will result in the page being COWed even
1923 * in MAP_SHARED mappings. You do not want this.
1924 * @pages: array that receives pointers to the pages pinned.
1925 * Should be at least nr_pages long. Or NULL, if caller
1926 * only intends to ensure the pages are faulted in.
1927 * @vmas: array of pointers to vmas corresponding to each page.
1928 * Or NULL if the caller does not require them.
1930 * Returns number of pages pinned. This may be fewer than the number
1931 * requested. If nr_pages is 0 or negative, returns 0. If no pages
1932 * were pinned, returns -errno. Each page returned must be released
1933 * with a put_page() call when it is finished with. vmas will only
1934 * remain valid while mmap_sem is held.
1936 * Must be called with mmap_sem held for read or write.
1938 * get_user_pages walks a process's page tables and takes a reference to
1939 * each struct page that each user address corresponds to at a given
1940 * instant. That is, it takes the page that would be accessed if a user
1941 * thread accesses the given user virtual address at that instant.
1943 * This does not guarantee that the page exists in the user mappings when
1944 * get_user_pages returns, and there may even be a completely different
1945 * page there in some cases (eg. if mmapped pagecache has been invalidated
1946 * and subsequently re faulted). However it does guarantee that the page
1947 * won't be freed completely. And mostly callers simply care that the page
1948 * contains data that was valid *at some point in time*. Typically, an IO
1949 * or similar operation cannot guarantee anything stronger anyway because
1950 * locks can't be held over the syscall boundary.
1952 * If write=0, the page must not be written to. If the page is written to,
1953 * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
1954 * after the page is finished with, and before put_page is called.
1956 * get_user_pages is typically used for fewer-copy IO operations, to get a
1957 * handle on the memory by some means other than accesses via the user virtual
1958 * addresses. The pages may be submitted for DMA to devices or accessed via
1959 * their kernel linear mapping (via the kmap APIs). Care should be taken to
1960 * use the correct cache flushing APIs.
1962 * See also get_user_pages_fast, for performance critical applications.
1964 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
1965 unsigned long start
, int nr_pages
, int write
, int force
,
1966 struct page
**pages
, struct vm_area_struct
**vmas
)
1968 int flags
= FOLL_TOUCH
;
1973 flags
|= FOLL_WRITE
;
1975 flags
|= FOLL_FORCE
;
1977 return __get_user_pages(tsk
, mm
, start
, nr_pages
, flags
, pages
, vmas
,
1980 EXPORT_SYMBOL(get_user_pages
);
1983 * get_dump_page() - pin user page in memory while writing it to core dump
1984 * @addr: user address
1986 * Returns struct page pointer of user page pinned for dump,
1987 * to be freed afterwards by page_cache_release() or put_page().
1989 * Returns NULL on any kind of failure - a hole must then be inserted into
1990 * the corefile, to preserve alignment with its headers; and also returns
1991 * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
1992 * allowing a hole to be left in the corefile to save diskspace.
1994 * Called without mmap_sem, but after all other threads have been killed.
1996 #ifdef CONFIG_ELF_CORE
1997 struct page
*get_dump_page(unsigned long addr
)
1999 struct vm_area_struct
*vma
;
2002 if (__get_user_pages(current
, current
->mm
, addr
, 1,
2003 FOLL_FORCE
| FOLL_DUMP
| FOLL_GET
, &page
, &vma
,
2006 flush_cache_page(vma
, addr
, page_to_pfn(page
));
2009 #endif /* CONFIG_ELF_CORE */
2011 pte_t
*__get_locked_pte(struct mm_struct
*mm
, unsigned long addr
,
2014 pgd_t
* pgd
= pgd_offset(mm
, addr
);
2015 pud_t
* pud
= pud_alloc(mm
, pgd
, addr
);
2017 pmd_t
* pmd
= pmd_alloc(mm
, pud
, addr
);
2019 VM_BUG_ON(pmd_trans_huge(*pmd
));
2020 return pte_alloc_map_lock(mm
, pmd
, addr
, ptl
);
2027 * This is the old fallback for page remapping.
2029 * For historical reasons, it only allows reserved pages. Only
2030 * old drivers should use this, and they needed to mark their
2031 * pages reserved for the old functions anyway.
2033 static int insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
2034 struct page
*page
, pgprot_t prot
)
2036 struct mm_struct
*mm
= vma
->vm_mm
;
2045 flush_dcache_page(page
);
2046 pte
= get_locked_pte(mm
, addr
, &ptl
);
2050 if (!pte_none(*pte
))
2053 /* Ok, finally just insert the thing.. */
2055 inc_mm_counter_fast(mm
, MM_FILEPAGES
);
2056 page_add_file_rmap(page
);
2057 set_pte_at(mm
, addr
, pte
, mk_pte(page
, prot
));
2060 pte_unmap_unlock(pte
, ptl
);
2063 pte_unmap_unlock(pte
, ptl
);
2069 * vm_insert_page - insert single page into user vma
2070 * @vma: user vma to map to
2071 * @addr: target user address of this page
2072 * @page: source kernel page
2074 * This allows drivers to insert individual pages they've allocated
2077 * The page has to be a nice clean _individual_ kernel allocation.
2078 * If you allocate a compound page, you need to have marked it as
2079 * such (__GFP_COMP), or manually just split the page up yourself
2080 * (see split_page()).
2082 * NOTE! Traditionally this was done with "remap_pfn_range()" which
2083 * took an arbitrary page protection parameter. This doesn't allow
2084 * that. Your vma protection will have to be set up correctly, which
2085 * means that if you want a shared writable mapping, you'd better
2086 * ask for a shared writable mapping!
2088 * The page does not need to be reserved.
2090 * Usually this function is called from f_op->mmap() handler
2091 * under mm->mmap_sem write-lock, so it can change vma->vm_flags.
2092 * Caller must set VM_MIXEDMAP on vma if it wants to call this
2093 * function from other places, for example from page-fault handler.
2095 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
2098 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
2100 if (!page_count(page
))
2102 if (!(vma
->vm_flags
& VM_MIXEDMAP
)) {
2103 BUG_ON(down_read_trylock(&vma
->vm_mm
->mmap_sem
));
2104 BUG_ON(vma
->vm_flags
& VM_PFNMAP
);
2105 vma
->vm_flags
|= VM_MIXEDMAP
;
2107 return insert_page(vma
, addr
, page
, vma
->vm_page_prot
);
2109 EXPORT_SYMBOL(vm_insert_page
);
2111 static int insert_pfn(struct vm_area_struct
*vma
, unsigned long addr
,
2112 unsigned long pfn
, pgprot_t prot
)
2114 struct mm_struct
*mm
= vma
->vm_mm
;
2120 pte
= get_locked_pte(mm
, addr
, &ptl
);
2124 if (!pte_none(*pte
))
2127 /* Ok, finally just insert the thing.. */
2128 entry
= pte_mkspecial(pfn_pte(pfn
, prot
));
2129 set_pte_at(mm
, addr
, pte
, entry
);
2130 update_mmu_cache(vma
, addr
, pte
); /* XXX: why not for insert_page? */
2134 pte_unmap_unlock(pte
, ptl
);
2140 * vm_insert_pfn - insert single pfn into user vma
2141 * @vma: user vma to map to
2142 * @addr: target user address of this page
2143 * @pfn: source kernel pfn
2145 * Similar to vm_inert_page, this allows drivers to insert individual pages
2146 * they've allocated into a user vma. Same comments apply.
2148 * This function should only be called from a vm_ops->fault handler, and
2149 * in that case the handler should return NULL.
2151 * vma cannot be a COW mapping.
2153 * As this is called only for pages that do not currently exist, we
2154 * do not need to flush old virtual caches or the TLB.
2156 int vm_insert_pfn(struct vm_area_struct
*vma
, unsigned long addr
,
2160 pgprot_t pgprot
= vma
->vm_page_prot
;
2162 * Technically, architectures with pte_special can avoid all these
2163 * restrictions (same for remap_pfn_range). However we would like
2164 * consistency in testing and feature parity among all, so we should
2165 * try to keep these invariants in place for everybody.
2167 BUG_ON(!(vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
)));
2168 BUG_ON((vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
)) ==
2169 (VM_PFNMAP
|VM_MIXEDMAP
));
2170 BUG_ON((vma
->vm_flags
& VM_PFNMAP
) && is_cow_mapping(vma
->vm_flags
));
2171 BUG_ON((vma
->vm_flags
& VM_MIXEDMAP
) && pfn_valid(pfn
));
2173 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
2175 if (track_pfn_insert(vma
, &pgprot
, pfn
))
2178 ret
= insert_pfn(vma
, addr
, pfn
, pgprot
);
2182 EXPORT_SYMBOL(vm_insert_pfn
);
2184 int vm_insert_mixed(struct vm_area_struct
*vma
, unsigned long addr
,
2187 BUG_ON(!(vma
->vm_flags
& VM_MIXEDMAP
));
2189 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
2193 * If we don't have pte special, then we have to use the pfn_valid()
2194 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
2195 * refcount the page if pfn_valid is true (hence insert_page rather
2196 * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
2197 * without pte special, it would there be refcounted as a normal page.
2199 if (!HAVE_PTE_SPECIAL
&& pfn_valid(pfn
)) {
2202 page
= pfn_to_page(pfn
);
2203 return insert_page(vma
, addr
, page
, vma
->vm_page_prot
);
2205 return insert_pfn(vma
, addr
, pfn
, vma
->vm_page_prot
);
2207 EXPORT_SYMBOL(vm_insert_mixed
);
2210 * maps a range of physical memory into the requested pages. the old
2211 * mappings are removed. any references to nonexistent pages results
2212 * in null mappings (currently treated as "copy-on-access")
2214 static int remap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
2215 unsigned long addr
, unsigned long end
,
2216 unsigned long pfn
, pgprot_t prot
)
2221 pte
= pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
2224 arch_enter_lazy_mmu_mode();
2226 BUG_ON(!pte_none(*pte
));
2227 set_pte_at(mm
, addr
, pte
, pte_mkspecial(pfn_pte(pfn
, prot
)));
2229 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
2230 arch_leave_lazy_mmu_mode();
2231 pte_unmap_unlock(pte
- 1, ptl
);
2235 static inline int remap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
2236 unsigned long addr
, unsigned long end
,
2237 unsigned long pfn
, pgprot_t prot
)
2242 pfn
-= addr
>> PAGE_SHIFT
;
2243 pmd
= pmd_alloc(mm
, pud
, addr
);
2246 VM_BUG_ON(pmd_trans_huge(*pmd
));
2248 next
= pmd_addr_end(addr
, end
);
2249 if (remap_pte_range(mm
, pmd
, addr
, next
,
2250 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
2252 } while (pmd
++, addr
= next
, addr
!= end
);
2256 static inline int remap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
2257 unsigned long addr
, unsigned long end
,
2258 unsigned long pfn
, pgprot_t prot
)
2263 pfn
-= addr
>> PAGE_SHIFT
;
2264 pud
= pud_alloc(mm
, pgd
, addr
);
2268 next
= pud_addr_end(addr
, end
);
2269 if (remap_pmd_range(mm
, pud
, addr
, next
,
2270 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
2272 } while (pud
++, addr
= next
, addr
!= end
);
2277 * remap_pfn_range - remap kernel memory to userspace
2278 * @vma: user vma to map to
2279 * @addr: target user address to start at
2280 * @pfn: physical address of kernel memory
2281 * @size: size of map area
2282 * @prot: page protection flags for this mapping
2284 * Note: this is only safe if the mm semaphore is held when called.
2286 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
2287 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
2291 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2292 struct mm_struct
*mm
= vma
->vm_mm
;
2296 * Physically remapped pages are special. Tell the
2297 * rest of the world about it:
2298 * VM_IO tells people not to look at these pages
2299 * (accesses can have side effects).
2300 * VM_PFNMAP tells the core MM that the base pages are just
2301 * raw PFN mappings, and do not have a "struct page" associated
2304 * Disable vma merging and expanding with mremap().
2306 * Omit vma from core dump, even when VM_IO turned off.
2308 * There's a horrible special case to handle copy-on-write
2309 * behaviour that some programs depend on. We mark the "original"
2310 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
2311 * See vm_normal_page() for details.
2313 if (is_cow_mapping(vma
->vm_flags
)) {
2314 if (addr
!= vma
->vm_start
|| end
!= vma
->vm_end
)
2316 vma
->vm_pgoff
= pfn
;
2319 err
= track_pfn_remap(vma
, &prot
, pfn
, addr
, PAGE_ALIGN(size
));
2323 vma
->vm_flags
|= VM_IO
| VM_PFNMAP
| VM_DONTEXPAND
| VM_DONTDUMP
;
2325 BUG_ON(addr
>= end
);
2326 pfn
-= addr
>> PAGE_SHIFT
;
2327 pgd
= pgd_offset(mm
, addr
);
2328 flush_cache_range(vma
, addr
, end
);
2330 next
= pgd_addr_end(addr
, end
);
2331 err
= remap_pud_range(mm
, pgd
, addr
, next
,
2332 pfn
+ (addr
>> PAGE_SHIFT
), prot
);
2335 } while (pgd
++, addr
= next
, addr
!= end
);
2338 untrack_pfn(vma
, pfn
, PAGE_ALIGN(size
));
2342 EXPORT_SYMBOL(remap_pfn_range
);
2344 static int apply_to_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
2345 unsigned long addr
, unsigned long end
,
2346 pte_fn_t fn
, void *data
)
2351 spinlock_t
*uninitialized_var(ptl
);
2353 pte
= (mm
== &init_mm
) ?
2354 pte_alloc_kernel(pmd
, addr
) :
2355 pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
2359 BUG_ON(pmd_huge(*pmd
));
2361 arch_enter_lazy_mmu_mode();
2363 token
= pmd_pgtable(*pmd
);
2366 err
= fn(pte
++, token
, addr
, data
);
2369 } while (addr
+= PAGE_SIZE
, addr
!= end
);
2371 arch_leave_lazy_mmu_mode();
2374 pte_unmap_unlock(pte
-1, ptl
);
2378 static int apply_to_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
2379 unsigned long addr
, unsigned long end
,
2380 pte_fn_t fn
, void *data
)
2386 BUG_ON(pud_huge(*pud
));
2388 pmd
= pmd_alloc(mm
, pud
, addr
);
2392 next
= pmd_addr_end(addr
, end
);
2393 err
= apply_to_pte_range(mm
, pmd
, addr
, next
, fn
, data
);
2396 } while (pmd
++, addr
= next
, addr
!= end
);
2400 static int apply_to_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
2401 unsigned long addr
, unsigned long end
,
2402 pte_fn_t fn
, void *data
)
2408 pud
= pud_alloc(mm
, pgd
, addr
);
2412 next
= pud_addr_end(addr
, end
);
2413 err
= apply_to_pmd_range(mm
, pud
, addr
, next
, fn
, data
);
2416 } while (pud
++, addr
= next
, addr
!= end
);
2421 * Scan a region of virtual memory, filling in page tables as necessary
2422 * and calling a provided function on each leaf page table.
2424 int apply_to_page_range(struct mm_struct
*mm
, unsigned long addr
,
2425 unsigned long size
, pte_fn_t fn
, void *data
)
2429 unsigned long end
= addr
+ size
;
2432 BUG_ON(addr
>= end
);
2433 pgd
= pgd_offset(mm
, addr
);
2435 next
= pgd_addr_end(addr
, end
);
2436 err
= apply_to_pud_range(mm
, pgd
, addr
, next
, fn
, data
);
2439 } while (pgd
++, addr
= next
, addr
!= end
);
2443 EXPORT_SYMBOL_GPL(apply_to_page_range
);
2446 * handle_pte_fault chooses page fault handler according to an entry
2447 * which was read non-atomically. Before making any commitment, on
2448 * those architectures or configurations (e.g. i386 with PAE) which
2449 * might give a mix of unmatched parts, do_swap_page and do_nonlinear_fault
2450 * must check under lock before unmapping the pte and proceeding
2451 * (but do_wp_page is only called after already making such a check;
2452 * and do_anonymous_page can safely check later on).
2454 static inline int pte_unmap_same(struct mm_struct
*mm
, pmd_t
*pmd
,
2455 pte_t
*page_table
, pte_t orig_pte
)
2458 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
2459 if (sizeof(pte_t
) > sizeof(unsigned long)) {
2460 spinlock_t
*ptl
= pte_lockptr(mm
, pmd
);
2462 same
= pte_same(*page_table
, orig_pte
);
2466 pte_unmap(page_table
);
2470 static inline void cow_user_page(struct page
*dst
, struct page
*src
, unsigned long va
, struct vm_area_struct
*vma
)
2473 * If the source page was a PFN mapping, we don't have
2474 * a "struct page" for it. We do a best-effort copy by
2475 * just copying from the original user address. If that
2476 * fails, we just zero-fill it. Live with it.
2478 if (unlikely(!src
)) {
2479 void *kaddr
= kmap_atomic(dst
);
2480 void __user
*uaddr
= (void __user
*)(va
& PAGE_MASK
);
2483 * This really shouldn't fail, because the page is there
2484 * in the page tables. But it might just be unreadable,
2485 * in which case we just give up and fill the result with
2488 if (__copy_from_user_inatomic(kaddr
, uaddr
, PAGE_SIZE
))
2490 kunmap_atomic(kaddr
);
2491 flush_dcache_page(dst
);
2493 copy_user_highpage(dst
, src
, va
, vma
);
2497 * This routine handles present pages, when users try to write
2498 * to a shared page. It is done by copying the page to a new address
2499 * and decrementing the shared-page counter for the old page.
2501 * Note that this routine assumes that the protection checks have been
2502 * done by the caller (the low-level page fault routine in most cases).
2503 * Thus we can safely just mark it writable once we've done any necessary
2506 * We also mark the page dirty at this point even though the page will
2507 * change only once the write actually happens. This avoids a few races,
2508 * and potentially makes it more efficient.
2510 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2511 * but allow concurrent faults), with pte both mapped and locked.
2512 * We return with mmap_sem still held, but pte unmapped and unlocked.
2514 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2515 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2516 spinlock_t
*ptl
, pte_t orig_pte
)
2519 struct page
*old_page
, *new_page
;
2522 int page_mkwrite
= 0;
2523 struct page
*dirty_page
= NULL
;
2525 old_page
= vm_normal_page(vma
, address
, orig_pte
);
2528 * VM_MIXEDMAP !pfn_valid() case
2530 * We should not cow pages in a shared writeable mapping.
2531 * Just mark the pages writable as we can't do any dirty
2532 * accounting on raw pfn maps.
2534 if ((vma
->vm_flags
& (VM_WRITE
|VM_SHARED
)) ==
2535 (VM_WRITE
|VM_SHARED
))
2541 * Take out anonymous pages first, anonymous shared vmas are
2542 * not dirty accountable.
2544 if (PageAnon(old_page
) && !PageKsm(old_page
)) {
2545 if (!trylock_page(old_page
)) {
2546 page_cache_get(old_page
);
2547 pte_unmap_unlock(page_table
, ptl
);
2548 lock_page(old_page
);
2549 page_table
= pte_offset_map_lock(mm
, pmd
, address
,
2551 if (!pte_same(*page_table
, orig_pte
)) {
2552 unlock_page(old_page
);
2555 page_cache_release(old_page
);
2557 if (reuse_swap_page(old_page
)) {
2559 * The page is all ours. Move it to our anon_vma so
2560 * the rmap code will not search our parent or siblings.
2561 * Protected against the rmap code by the page lock.
2563 page_move_anon_rmap(old_page
, vma
, address
);
2564 unlock_page(old_page
);
2567 unlock_page(old_page
);
2568 } else if (unlikely((vma
->vm_flags
& (VM_WRITE
|VM_SHARED
)) ==
2569 (VM_WRITE
|VM_SHARED
))) {
2571 * Only catch write-faults on shared writable pages,
2572 * read-only shared pages can get COWed by
2573 * get_user_pages(.write=1, .force=1).
2575 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
) {
2576 struct vm_fault vmf
;
2579 vmf
.virtual_address
= (void __user
*)(address
&
2581 vmf
.pgoff
= old_page
->index
;
2582 vmf
.flags
= FAULT_FLAG_WRITE
|FAULT_FLAG_MKWRITE
;
2583 vmf
.page
= old_page
;
2586 * Notify the address space that the page is about to
2587 * become writable so that it can prohibit this or wait
2588 * for the page to get into an appropriate state.
2590 * We do this without the lock held, so that it can
2591 * sleep if it needs to.
2593 page_cache_get(old_page
);
2594 pte_unmap_unlock(page_table
, ptl
);
2596 tmp
= vma
->vm_ops
->page_mkwrite(vma
, &vmf
);
2598 (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
))) {
2600 goto unwritable_page
;
2602 if (unlikely(!(tmp
& VM_FAULT_LOCKED
))) {
2603 lock_page(old_page
);
2604 if (!old_page
->mapping
) {
2605 ret
= 0; /* retry the fault */
2606 unlock_page(old_page
);
2607 goto unwritable_page
;
2610 VM_BUG_ON(!PageLocked(old_page
));
2613 * Since we dropped the lock we need to revalidate
2614 * the PTE as someone else may have changed it. If
2615 * they did, we just return, as we can count on the
2616 * MMU to tell us if they didn't also make it writable.
2618 page_table
= pte_offset_map_lock(mm
, pmd
, address
,
2620 if (!pte_same(*page_table
, orig_pte
)) {
2621 unlock_page(old_page
);
2627 dirty_page
= old_page
;
2628 get_page(dirty_page
);
2631 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
2632 entry
= pte_mkyoung(orig_pte
);
2633 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2634 if (ptep_set_access_flags(vma
, address
, page_table
, entry
,1))
2635 update_mmu_cache(vma
, address
, page_table
);
2636 pte_unmap_unlock(page_table
, ptl
);
2637 ret
|= VM_FAULT_WRITE
;
2643 * Yes, Virginia, this is actually required to prevent a race
2644 * with clear_page_dirty_for_io() from clearing the page dirty
2645 * bit after it clear all dirty ptes, but before a racing
2646 * do_wp_page installs a dirty pte.
2648 * __do_fault is protected similarly.
2650 if (!page_mkwrite
) {
2651 wait_on_page_locked(dirty_page
);
2652 set_page_dirty_balance(dirty_page
, page_mkwrite
);
2653 /* file_update_time outside page_lock */
2655 file_update_time(vma
->vm_file
);
2657 put_page(dirty_page
);
2659 struct address_space
*mapping
= dirty_page
->mapping
;
2661 set_page_dirty(dirty_page
);
2662 unlock_page(dirty_page
);
2663 page_cache_release(dirty_page
);
2666 * Some device drivers do not set page.mapping
2667 * but still dirty their pages
2669 balance_dirty_pages_ratelimited(mapping
);
2677 * Ok, we need to copy. Oh, well..
2679 page_cache_get(old_page
);
2681 pte_unmap_unlock(page_table
, ptl
);
2683 if (unlikely(anon_vma_prepare(vma
)))
2686 if (is_zero_pfn(pte_pfn(orig_pte
))) {
2687 new_page
= alloc_zeroed_user_highpage_movable(vma
, address
);
2691 new_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, address
);
2694 cow_user_page(new_page
, old_page
, address
, vma
);
2696 __SetPageUptodate(new_page
);
2698 if (mem_cgroup_newpage_charge(new_page
, mm
, GFP_KERNEL
))
2702 * Re-check the pte - we dropped the lock
2704 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2705 if (likely(pte_same(*page_table
, orig_pte
))) {
2707 if (!PageAnon(old_page
)) {
2708 dec_mm_counter_fast(mm
, MM_FILEPAGES
);
2709 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
2712 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
2713 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
2714 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
2715 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2717 * Clear the pte entry and flush it first, before updating the
2718 * pte with the new entry. This will avoid a race condition
2719 * seen in the presence of one thread doing SMC and another
2722 ptep_clear_flush(vma
, address
, page_table
);
2723 page_add_new_anon_rmap(new_page
, vma
, address
);
2725 * We call the notify macro here because, when using secondary
2726 * mmu page tables (such as kvm shadow page tables), we want the
2727 * new page to be mapped directly into the secondary page table.
2729 set_pte_at_notify(mm
, address
, page_table
, entry
);
2730 update_mmu_cache(vma
, address
, page_table
);
2733 * Only after switching the pte to the new page may
2734 * we remove the mapcount here. Otherwise another
2735 * process may come and find the rmap count decremented
2736 * before the pte is switched to the new page, and
2737 * "reuse" the old page writing into it while our pte
2738 * here still points into it and can be read by other
2741 * The critical issue is to order this
2742 * page_remove_rmap with the ptp_clear_flush above.
2743 * Those stores are ordered by (if nothing else,)
2744 * the barrier present in the atomic_add_negative
2745 * in page_remove_rmap.
2747 * Then the TLB flush in ptep_clear_flush ensures that
2748 * no process can access the old page before the
2749 * decremented mapcount is visible. And the old page
2750 * cannot be reused until after the decremented
2751 * mapcount is visible. So transitively, TLBs to
2752 * old page will be flushed before it can be reused.
2754 page_remove_rmap(old_page
);
2757 /* Free the old page.. */
2758 new_page
= old_page
;
2759 ret
|= VM_FAULT_WRITE
;
2761 mem_cgroup_uncharge_page(new_page
);
2764 page_cache_release(new_page
);
2766 pte_unmap_unlock(page_table
, ptl
);
2769 * Don't let another task, with possibly unlocked vma,
2770 * keep the mlocked page.
2772 if ((ret
& VM_FAULT_WRITE
) && (vma
->vm_flags
& VM_LOCKED
)) {
2773 lock_page(old_page
); /* LRU manipulation */
2774 munlock_vma_page(old_page
);
2775 unlock_page(old_page
);
2777 page_cache_release(old_page
);
2781 page_cache_release(new_page
);
2785 unlock_page(old_page
);
2786 page_cache_release(old_page
);
2788 page_cache_release(old_page
);
2790 return VM_FAULT_OOM
;
2793 page_cache_release(old_page
);
2797 static void unmap_mapping_range_vma(struct vm_area_struct
*vma
,
2798 unsigned long start_addr
, unsigned long end_addr
,
2799 struct zap_details
*details
)
2801 zap_page_range_single(vma
, start_addr
, end_addr
- start_addr
, details
);
2804 static inline void unmap_mapping_range_tree(struct rb_root
*root
,
2805 struct zap_details
*details
)
2807 struct vm_area_struct
*vma
;
2808 pgoff_t vba
, vea
, zba
, zea
;
2810 vma_interval_tree_foreach(vma
, root
,
2811 details
->first_index
, details
->last_index
) {
2813 vba
= vma
->vm_pgoff
;
2814 vea
= vba
+ ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
) - 1;
2815 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2816 zba
= details
->first_index
;
2819 zea
= details
->last_index
;
2823 unmap_mapping_range_vma(vma
,
2824 ((zba
- vba
) << PAGE_SHIFT
) + vma
->vm_start
,
2825 ((zea
- vba
+ 1) << PAGE_SHIFT
) + vma
->vm_start
,
2830 static inline void unmap_mapping_range_list(struct list_head
*head
,
2831 struct zap_details
*details
)
2833 struct vm_area_struct
*vma
;
2836 * In nonlinear VMAs there is no correspondence between virtual address
2837 * offset and file offset. So we must perform an exhaustive search
2838 * across *all* the pages in each nonlinear VMA, not just the pages
2839 * whose virtual address lies outside the file truncation point.
2841 list_for_each_entry(vma
, head
, shared
.nonlinear
) {
2842 details
->nonlinear_vma
= vma
;
2843 unmap_mapping_range_vma(vma
, vma
->vm_start
, vma
->vm_end
, details
);
2848 * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
2849 * @mapping: the address space containing mmaps to be unmapped.
2850 * @holebegin: byte in first page to unmap, relative to the start of
2851 * the underlying file. This will be rounded down to a PAGE_SIZE
2852 * boundary. Note that this is different from truncate_pagecache(), which
2853 * must keep the partial page. In contrast, we must get rid of
2855 * @holelen: size of prospective hole in bytes. This will be rounded
2856 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2858 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2859 * but 0 when invalidating pagecache, don't throw away private data.
2861 void unmap_mapping_range(struct address_space
*mapping
,
2862 loff_t
const holebegin
, loff_t
const holelen
, int even_cows
)
2864 struct zap_details details
;
2865 pgoff_t hba
= holebegin
>> PAGE_SHIFT
;
2866 pgoff_t hlen
= (holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2868 /* Check for overflow. */
2869 if (sizeof(holelen
) > sizeof(hlen
)) {
2871 (holebegin
+ holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2872 if (holeend
& ~(long long)ULONG_MAX
)
2873 hlen
= ULONG_MAX
- hba
+ 1;
2876 details
.check_mapping
= even_cows
? NULL
: mapping
;
2877 details
.nonlinear_vma
= NULL
;
2878 details
.first_index
= hba
;
2879 details
.last_index
= hba
+ hlen
- 1;
2880 if (details
.last_index
< details
.first_index
)
2881 details
.last_index
= ULONG_MAX
;
2884 mutex_lock(&mapping
->i_mmap_mutex
);
2885 if (unlikely(!RB_EMPTY_ROOT(&mapping
->i_mmap
)))
2886 unmap_mapping_range_tree(&mapping
->i_mmap
, &details
);
2887 if (unlikely(!list_empty(&mapping
->i_mmap_nonlinear
)))
2888 unmap_mapping_range_list(&mapping
->i_mmap_nonlinear
, &details
);
2889 mutex_unlock(&mapping
->i_mmap_mutex
);
2891 EXPORT_SYMBOL(unmap_mapping_range
);
2894 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2895 * but allow concurrent faults), and pte mapped but not yet locked.
2896 * We return with mmap_sem still held, but pte unmapped and unlocked.
2898 static int do_swap_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2899 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2900 unsigned int flags
, pte_t orig_pte
)
2903 struct page
*page
, *swapcache
= NULL
;
2907 struct mem_cgroup
*ptr
;
2911 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
2914 entry
= pte_to_swp_entry(orig_pte
);
2915 if (unlikely(non_swap_entry(entry
))) {
2916 if (is_migration_entry(entry
)) {
2917 migration_entry_wait(mm
, pmd
, address
);
2918 } else if (is_hwpoison_entry(entry
)) {
2919 ret
= VM_FAULT_HWPOISON
;
2921 print_bad_pte(vma
, address
, orig_pte
, NULL
);
2922 ret
= VM_FAULT_SIGBUS
;
2926 delayacct_set_flag(DELAYACCT_PF_SWAPIN
);
2927 page
= lookup_swap_cache(entry
);
2929 page
= swapin_readahead(entry
,
2930 GFP_HIGHUSER_MOVABLE
, vma
, address
);
2933 * Back out if somebody else faulted in this pte
2934 * while we released the pte lock.
2936 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2937 if (likely(pte_same(*page_table
, orig_pte
)))
2939 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2943 /* Had to read the page from swap area: Major fault */
2944 ret
= VM_FAULT_MAJOR
;
2945 count_vm_event(PGMAJFAULT
);
2946 mem_cgroup_count_vm_event(mm
, PGMAJFAULT
);
2947 } else if (PageHWPoison(page
)) {
2949 * hwpoisoned dirty swapcache pages are kept for killing
2950 * owner processes (which may be unknown at hwpoison time)
2952 ret
= VM_FAULT_HWPOISON
;
2953 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2957 locked
= lock_page_or_retry(page
, mm
, flags
);
2959 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2961 ret
|= VM_FAULT_RETRY
;
2966 * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
2967 * release the swapcache from under us. The page pin, and pte_same
2968 * test below, are not enough to exclude that. Even if it is still
2969 * swapcache, we need to check that the page's swap has not changed.
2971 if (unlikely(!PageSwapCache(page
) || page_private(page
) != entry
.val
))
2974 if (ksm_might_need_to_copy(page
, vma
, address
)) {
2976 page
= ksm_does_need_to_copy(page
, vma
, address
);
2978 if (unlikely(!page
)) {
2986 if (mem_cgroup_try_charge_swapin(mm
, page
, GFP_KERNEL
, &ptr
)) {
2992 * Back out if somebody else already faulted in this pte.
2994 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2995 if (unlikely(!pte_same(*page_table
, orig_pte
)))
2998 if (unlikely(!PageUptodate(page
))) {
2999 ret
= VM_FAULT_SIGBUS
;
3004 * The page isn't present yet, go ahead with the fault.
3006 * Be careful about the sequence of operations here.
3007 * To get its accounting right, reuse_swap_page() must be called
3008 * while the page is counted on swap but not yet in mapcount i.e.
3009 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
3010 * must be called after the swap_free(), or it will never succeed.
3011 * Because delete_from_swap_page() may be called by reuse_swap_page(),
3012 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
3013 * in page->private. In this case, a record in swap_cgroup is silently
3014 * discarded at swap_free().
3017 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
3018 dec_mm_counter_fast(mm
, MM_SWAPENTS
);
3019 pte
= mk_pte(page
, vma
->vm_page_prot
);
3020 if ((flags
& FAULT_FLAG_WRITE
) && reuse_swap_page(page
)) {
3021 pte
= maybe_mkwrite(pte_mkdirty(pte
), vma
);
3022 flags
&= ~FAULT_FLAG_WRITE
;
3023 ret
|= VM_FAULT_WRITE
;
3026 flush_icache_page(vma
, page
);
3027 set_pte_at(mm
, address
, page_table
, pte
);
3028 do_page_add_anon_rmap(page
, vma
, address
, exclusive
);
3029 /* It's better to call commit-charge after rmap is established */
3030 mem_cgroup_commit_charge_swapin(page
, ptr
);
3033 if (vm_swap_full() || (vma
->vm_flags
& VM_LOCKED
) || PageMlocked(page
))
3034 try_to_free_swap(page
);
3038 * Hold the lock to avoid the swap entry to be reused
3039 * until we take the PT lock for the pte_same() check
3040 * (to avoid false positives from pte_same). For
3041 * further safety release the lock after the swap_free
3042 * so that the swap count won't change under a
3043 * parallel locked swapcache.
3045 unlock_page(swapcache
);
3046 page_cache_release(swapcache
);
3049 if (flags
& FAULT_FLAG_WRITE
) {
3050 ret
|= do_wp_page(mm
, vma
, address
, page_table
, pmd
, ptl
, pte
);
3051 if (ret
& VM_FAULT_ERROR
)
3052 ret
&= VM_FAULT_ERROR
;
3056 /* No need to invalidate - it was non-present before */
3057 update_mmu_cache(vma
, address
, page_table
);
3059 pte_unmap_unlock(page_table
, ptl
);
3063 mem_cgroup_cancel_charge_swapin(ptr
);
3064 pte_unmap_unlock(page_table
, ptl
);
3068 page_cache_release(page
);
3070 unlock_page(swapcache
);
3071 page_cache_release(swapcache
);
3077 * This is like a special single-page "expand_{down|up}wards()",
3078 * except we must first make sure that 'address{-|+}PAGE_SIZE'
3079 * doesn't hit another vma.
3081 static inline int check_stack_guard_page(struct vm_area_struct
*vma
, unsigned long address
)
3083 address
&= PAGE_MASK
;
3084 if ((vma
->vm_flags
& VM_GROWSDOWN
) && address
== vma
->vm_start
) {
3085 struct vm_area_struct
*prev
= vma
->vm_prev
;
3088 * Is there a mapping abutting this one below?
3090 * That's only ok if it's the same stack mapping
3091 * that has gotten split..
3093 if (prev
&& prev
->vm_end
== address
)
3094 return prev
->vm_flags
& VM_GROWSDOWN
? 0 : -ENOMEM
;
3096 expand_downwards(vma
, address
- PAGE_SIZE
);
3098 if ((vma
->vm_flags
& VM_GROWSUP
) && address
+ PAGE_SIZE
== vma
->vm_end
) {
3099 struct vm_area_struct
*next
= vma
->vm_next
;
3101 /* As VM_GROWSDOWN but s/below/above/ */
3102 if (next
&& next
->vm_start
== address
+ PAGE_SIZE
)
3103 return next
->vm_flags
& VM_GROWSUP
? 0 : -ENOMEM
;
3105 expand_upwards(vma
, address
+ PAGE_SIZE
);
3111 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3112 * but allow concurrent faults), and pte mapped but not yet locked.
3113 * We return with mmap_sem still held, but pte unmapped and unlocked.
3115 static int do_anonymous_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3116 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
3123 pte_unmap(page_table
);
3125 /* Check if we need to add a guard page to the stack */
3126 if (check_stack_guard_page(vma
, address
) < 0)
3127 return VM_FAULT_SIGBUS
;
3129 /* Use the zero-page for reads */
3130 if (!(flags
& FAULT_FLAG_WRITE
)) {
3131 entry
= pte_mkspecial(pfn_pte(my_zero_pfn(address
),
3132 vma
->vm_page_prot
));
3133 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
3134 if (!pte_none(*page_table
))
3139 /* Allocate our own private page. */
3140 if (unlikely(anon_vma_prepare(vma
)))
3142 page
= alloc_zeroed_user_highpage_movable(vma
, address
);
3145 __SetPageUptodate(page
);
3147 if (mem_cgroup_newpage_charge(page
, mm
, GFP_KERNEL
))
3150 entry
= mk_pte(page
, vma
->vm_page_prot
);
3151 if (vma
->vm_flags
& VM_WRITE
)
3152 entry
= pte_mkwrite(pte_mkdirty(entry
));
3154 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
3155 if (!pte_none(*page_table
))
3158 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
3159 page_add_new_anon_rmap(page
, vma
, address
);
3161 set_pte_at(mm
, address
, page_table
, entry
);
3163 /* No need to invalidate - it was non-present before */
3164 update_mmu_cache(vma
, address
, page_table
);
3166 pte_unmap_unlock(page_table
, ptl
);
3169 mem_cgroup_uncharge_page(page
);
3170 page_cache_release(page
);
3173 page_cache_release(page
);
3175 return VM_FAULT_OOM
;
3179 * __do_fault() tries to create a new page mapping. It aggressively
3180 * tries to share with existing pages, but makes a separate copy if
3181 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
3182 * the next page fault.
3184 * As this is called only for pages that do not currently exist, we
3185 * do not need to flush old virtual caches or the TLB.
3187 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3188 * but allow concurrent faults), and pte neither mapped nor locked.
3189 * We return with mmap_sem still held, but pte unmapped and unlocked.
3191 static int __do_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3192 unsigned long address
, pmd_t
*pmd
,
3193 pgoff_t pgoff
, unsigned int flags
, pte_t orig_pte
)
3198 struct page
*cow_page
;
3201 struct page
*dirty_page
= NULL
;
3202 struct vm_fault vmf
;
3204 int page_mkwrite
= 0;
3207 * If we do COW later, allocate page befor taking lock_page()
3208 * on the file cache page. This will reduce lock holding time.
3210 if ((flags
& FAULT_FLAG_WRITE
) && !(vma
->vm_flags
& VM_SHARED
)) {
3212 if (unlikely(anon_vma_prepare(vma
)))
3213 return VM_FAULT_OOM
;
3215 cow_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, address
);
3217 return VM_FAULT_OOM
;
3219 if (mem_cgroup_newpage_charge(cow_page
, mm
, GFP_KERNEL
)) {
3220 page_cache_release(cow_page
);
3221 return VM_FAULT_OOM
;
3226 vmf
.virtual_address
= (void __user
*)(address
& PAGE_MASK
);
3231 ret
= vma
->vm_ops
->fault(vma
, &vmf
);
3232 if (unlikely(ret
& (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
|
3236 if (unlikely(PageHWPoison(vmf
.page
))) {
3237 if (ret
& VM_FAULT_LOCKED
)
3238 unlock_page(vmf
.page
);
3239 ret
= VM_FAULT_HWPOISON
;
3244 * For consistency in subsequent calls, make the faulted page always
3247 if (unlikely(!(ret
& VM_FAULT_LOCKED
)))
3248 lock_page(vmf
.page
);
3250 VM_BUG_ON(!PageLocked(vmf
.page
));
3253 * Should we do an early C-O-W break?
3256 if (flags
& FAULT_FLAG_WRITE
) {
3257 if (!(vma
->vm_flags
& VM_SHARED
)) {
3260 copy_user_highpage(page
, vmf
.page
, address
, vma
);
3261 __SetPageUptodate(page
);
3264 * If the page will be shareable, see if the backing
3265 * address space wants to know that the page is about
3266 * to become writable
3268 if (vma
->vm_ops
->page_mkwrite
) {
3272 vmf
.flags
= FAULT_FLAG_WRITE
|FAULT_FLAG_MKWRITE
;
3273 tmp
= vma
->vm_ops
->page_mkwrite(vma
, &vmf
);
3275 (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
))) {
3277 goto unwritable_page
;
3279 if (unlikely(!(tmp
& VM_FAULT_LOCKED
))) {
3281 if (!page
->mapping
) {
3282 ret
= 0; /* retry the fault */
3284 goto unwritable_page
;
3287 VM_BUG_ON(!PageLocked(page
));
3294 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
3297 * This silly early PAGE_DIRTY setting removes a race
3298 * due to the bad i386 page protection. But it's valid
3299 * for other architectures too.
3301 * Note that if FAULT_FLAG_WRITE is set, we either now have
3302 * an exclusive copy of the page, or this is a shared mapping,
3303 * so we can make it writable and dirty to avoid having to
3304 * handle that later.
3306 /* Only go through if we didn't race with anybody else... */
3307 if (likely(pte_same(*page_table
, orig_pte
))) {
3308 flush_icache_page(vma
, page
);
3309 entry
= mk_pte(page
, vma
->vm_page_prot
);
3310 if (flags
& FAULT_FLAG_WRITE
)
3311 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
3313 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
3314 page_add_new_anon_rmap(page
, vma
, address
);
3316 inc_mm_counter_fast(mm
, MM_FILEPAGES
);
3317 page_add_file_rmap(page
);
3318 if (flags
& FAULT_FLAG_WRITE
) {
3320 get_page(dirty_page
);
3323 set_pte_at(mm
, address
, page_table
, entry
);
3325 /* no need to invalidate: a not-present page won't be cached */
3326 update_mmu_cache(vma
, address
, page_table
);
3329 mem_cgroup_uncharge_page(cow_page
);
3331 page_cache_release(page
);
3333 anon
= 1; /* no anon but release faulted_page */
3336 pte_unmap_unlock(page_table
, ptl
);
3339 struct address_space
*mapping
= page
->mapping
;
3342 if (set_page_dirty(dirty_page
))
3344 unlock_page(dirty_page
);
3345 put_page(dirty_page
);
3346 if ((dirtied
|| page_mkwrite
) && mapping
) {
3348 * Some device drivers do not set page.mapping but still
3351 balance_dirty_pages_ratelimited(mapping
);
3354 /* file_update_time outside page_lock */
3355 if (vma
->vm_file
&& !page_mkwrite
)
3356 file_update_time(vma
->vm_file
);
3358 unlock_page(vmf
.page
);
3360 page_cache_release(vmf
.page
);
3366 page_cache_release(page
);
3369 /* fs's fault handler get error */
3371 mem_cgroup_uncharge_page(cow_page
);
3372 page_cache_release(cow_page
);
3377 static int do_linear_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3378 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
3379 unsigned int flags
, pte_t orig_pte
)
3381 pgoff_t pgoff
= (((address
& PAGE_MASK
)
3382 - vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
3384 pte_unmap(page_table
);
3385 return __do_fault(mm
, vma
, address
, pmd
, pgoff
, flags
, orig_pte
);
3389 * Fault of a previously existing named mapping. Repopulate the pte
3390 * from the encoded file_pte if possible. This enables swappable
3393 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3394 * but allow concurrent faults), and pte mapped but not yet locked.
3395 * We return with mmap_sem still held, but pte unmapped and unlocked.
3397 static int do_nonlinear_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3398 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
3399 unsigned int flags
, pte_t orig_pte
)
3403 flags
|= FAULT_FLAG_NONLINEAR
;
3405 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
3408 if (unlikely(!(vma
->vm_flags
& VM_NONLINEAR
))) {
3410 * Page table corrupted: show pte and kill process.
3412 print_bad_pte(vma
, address
, orig_pte
, NULL
);
3413 return VM_FAULT_SIGBUS
;
3416 pgoff
= pte_to_pgoff(orig_pte
);
3417 return __do_fault(mm
, vma
, address
, pmd
, pgoff
, flags
, orig_pte
);
3421 * These routines also need to handle stuff like marking pages dirty
3422 * and/or accessed for architectures that don't do it in hardware (most
3423 * RISC architectures). The early dirtying is also good on the i386.
3425 * There is also a hook called "update_mmu_cache()" that architectures
3426 * with external mmu caches can use to update those (ie the Sparc or
3427 * PowerPC hashed page tables that act as extended TLBs).
3429 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3430 * but allow concurrent faults), and pte mapped but not yet locked.
3431 * We return with mmap_sem still held, but pte unmapped and unlocked.
3433 int handle_pte_fault(struct mm_struct
*mm
,
3434 struct vm_area_struct
*vma
, unsigned long address
,
3435 pte_t
*pte
, pmd_t
*pmd
, unsigned int flags
)
3441 if (!pte_present(entry
)) {
3442 if (pte_none(entry
)) {
3444 if (likely(vma
->vm_ops
->fault
))
3445 return do_linear_fault(mm
, vma
, address
,
3446 pte
, pmd
, flags
, entry
);
3448 return do_anonymous_page(mm
, vma
, address
,
3451 if (pte_file(entry
))
3452 return do_nonlinear_fault(mm
, vma
, address
,
3453 pte
, pmd
, flags
, entry
);
3454 return do_swap_page(mm
, vma
, address
,
3455 pte
, pmd
, flags
, entry
);
3458 ptl
= pte_lockptr(mm
, pmd
);
3460 if (unlikely(!pte_same(*pte
, entry
)))
3462 if (flags
& FAULT_FLAG_WRITE
) {
3463 if (!pte_write(entry
))
3464 return do_wp_page(mm
, vma
, address
,
3465 pte
, pmd
, ptl
, entry
);
3466 entry
= pte_mkdirty(entry
);
3468 entry
= pte_mkyoung(entry
);
3469 if (ptep_set_access_flags(vma
, address
, pte
, entry
, flags
& FAULT_FLAG_WRITE
)) {
3470 update_mmu_cache(vma
, address
, pte
);
3473 * This is needed only for protection faults but the arch code
3474 * is not yet telling us if this is a protection fault or not.
3475 * This still avoids useless tlb flushes for .text page faults
3478 if (flags
& FAULT_FLAG_WRITE
)
3479 flush_tlb_fix_spurious_fault(vma
, address
);
3482 pte_unmap_unlock(pte
, ptl
);
3487 * By the time we get here, we already hold the mm semaphore
3489 int handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3490 unsigned long address
, unsigned int flags
)
3497 __set_current_state(TASK_RUNNING
);
3499 count_vm_event(PGFAULT
);
3500 mem_cgroup_count_vm_event(mm
, PGFAULT
);
3502 /* do counter updates before entering really critical section. */
3503 check_sync_rss_stat(current
);
3505 if (unlikely(is_vm_hugetlb_page(vma
)))
3506 return hugetlb_fault(mm
, vma
, address
, flags
);
3509 pgd
= pgd_offset(mm
, address
);
3510 pud
= pud_alloc(mm
, pgd
, address
);
3512 return VM_FAULT_OOM
;
3513 pmd
= pmd_alloc(mm
, pud
, address
);
3515 return VM_FAULT_OOM
;
3516 if (pmd_none(*pmd
) && transparent_hugepage_enabled(vma
)) {
3518 return do_huge_pmd_anonymous_page(mm
, vma
, address
,
3521 pmd_t orig_pmd
= *pmd
;
3525 if (pmd_trans_huge(orig_pmd
)) {
3526 if (flags
& FAULT_FLAG_WRITE
&&
3527 !pmd_write(orig_pmd
) &&
3528 !pmd_trans_splitting(orig_pmd
)) {
3529 ret
= do_huge_pmd_wp_page(mm
, vma
, address
, pmd
,
3532 * If COW results in an oom, the huge pmd will
3533 * have been split, so retry the fault on the
3534 * pte for a smaller charge.
3536 if (unlikely(ret
& VM_FAULT_OOM
))
3545 * Use __pte_alloc instead of pte_alloc_map, because we can't
3546 * run pte_offset_map on the pmd, if an huge pmd could
3547 * materialize from under us from a different thread.
3549 if (unlikely(pmd_none(*pmd
)) && __pte_alloc(mm
, vma
, pmd
, address
))
3550 return VM_FAULT_OOM
;
3551 /* if an huge pmd materialized from under us just retry later */
3552 if (unlikely(pmd_trans_huge(*pmd
)))
3555 * A regular pmd is established and it can't morph into a huge pmd
3556 * from under us anymore at this point because we hold the mmap_sem
3557 * read mode and khugepaged takes it in write mode. So now it's
3558 * safe to run pte_offset_map().
3560 pte
= pte_offset_map(pmd
, address
);
3562 return handle_pte_fault(mm
, vma
, address
, pte
, pmd
, flags
);
3565 #ifndef __PAGETABLE_PUD_FOLDED
3567 * Allocate page upper directory.
3568 * We've already handled the fast-path in-line.
3570 int __pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
3572 pud_t
*new = pud_alloc_one(mm
, address
);
3576 smp_wmb(); /* See comment in __pte_alloc */
3578 spin_lock(&mm
->page_table_lock
);
3579 if (pgd_present(*pgd
)) /* Another has populated it */
3582 pgd_populate(mm
, pgd
, new);
3583 spin_unlock(&mm
->page_table_lock
);
3586 #endif /* __PAGETABLE_PUD_FOLDED */
3588 #ifndef __PAGETABLE_PMD_FOLDED
3590 * Allocate page middle directory.
3591 * We've already handled the fast-path in-line.
3593 int __pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
3595 pmd_t
*new = pmd_alloc_one(mm
, address
);
3599 smp_wmb(); /* See comment in __pte_alloc */
3601 spin_lock(&mm
->page_table_lock
);
3602 #ifndef __ARCH_HAS_4LEVEL_HACK
3603 if (pud_present(*pud
)) /* Another has populated it */
3606 pud_populate(mm
, pud
, new);
3608 if (pgd_present(*pud
)) /* Another has populated it */
3611 pgd_populate(mm
, pud
, new);
3612 #endif /* __ARCH_HAS_4LEVEL_HACK */
3613 spin_unlock(&mm
->page_table_lock
);
3616 #endif /* __PAGETABLE_PMD_FOLDED */
3618 int make_pages_present(unsigned long addr
, unsigned long end
)
3620 int ret
, len
, write
;
3621 struct vm_area_struct
* vma
;
3623 vma
= find_vma(current
->mm
, addr
);
3627 * We want to touch writable mappings with a write fault in order
3628 * to break COW, except for shared mappings because these don't COW
3629 * and we would not want to dirty them for nothing.
3631 write
= (vma
->vm_flags
& (VM_WRITE
| VM_SHARED
)) == VM_WRITE
;
3632 BUG_ON(addr
>= end
);
3633 BUG_ON(end
> vma
->vm_end
);
3634 len
= DIV_ROUND_UP(end
, PAGE_SIZE
) - addr
/PAGE_SIZE
;
3635 ret
= get_user_pages(current
, current
->mm
, addr
,
3636 len
, write
, 0, NULL
, NULL
);
3639 return ret
== len
? 0 : -EFAULT
;
3642 #if !defined(__HAVE_ARCH_GATE_AREA)
3644 #if defined(AT_SYSINFO_EHDR)
3645 static struct vm_area_struct gate_vma
;
3647 static int __init
gate_vma_init(void)
3649 gate_vma
.vm_mm
= NULL
;
3650 gate_vma
.vm_start
= FIXADDR_USER_START
;
3651 gate_vma
.vm_end
= FIXADDR_USER_END
;
3652 gate_vma
.vm_flags
= VM_READ
| VM_MAYREAD
| VM_EXEC
| VM_MAYEXEC
;
3653 gate_vma
.vm_page_prot
= __P101
;
3657 __initcall(gate_vma_init
);
3660 struct vm_area_struct
*get_gate_vma(struct mm_struct
*mm
)
3662 #ifdef AT_SYSINFO_EHDR
3669 int in_gate_area_no_mm(unsigned long addr
)
3671 #ifdef AT_SYSINFO_EHDR
3672 if ((addr
>= FIXADDR_USER_START
) && (addr
< FIXADDR_USER_END
))
3678 #endif /* __HAVE_ARCH_GATE_AREA */
3680 static int __follow_pte(struct mm_struct
*mm
, unsigned long address
,
3681 pte_t
**ptepp
, spinlock_t
**ptlp
)
3688 pgd
= pgd_offset(mm
, address
);
3689 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
3692 pud
= pud_offset(pgd
, address
);
3693 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
3696 pmd
= pmd_offset(pud
, address
);
3697 VM_BUG_ON(pmd_trans_huge(*pmd
));
3698 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
3701 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3705 ptep
= pte_offset_map_lock(mm
, pmd
, address
, ptlp
);
3708 if (!pte_present(*ptep
))
3713 pte_unmap_unlock(ptep
, *ptlp
);
3718 static inline int follow_pte(struct mm_struct
*mm
, unsigned long address
,
3719 pte_t
**ptepp
, spinlock_t
**ptlp
)
3723 /* (void) is needed to make gcc happy */
3724 (void) __cond_lock(*ptlp
,
3725 !(res
= __follow_pte(mm
, address
, ptepp
, ptlp
)));
3730 * follow_pfn - look up PFN at a user virtual address
3731 * @vma: memory mapping
3732 * @address: user virtual address
3733 * @pfn: location to store found PFN
3735 * Only IO mappings and raw PFN mappings are allowed.
3737 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3739 int follow_pfn(struct vm_area_struct
*vma
, unsigned long address
,
3746 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
3749 ret
= follow_pte(vma
->vm_mm
, address
, &ptep
, &ptl
);
3752 *pfn
= pte_pfn(*ptep
);
3753 pte_unmap_unlock(ptep
, ptl
);
3756 EXPORT_SYMBOL(follow_pfn
);
3758 #ifdef CONFIG_HAVE_IOREMAP_PROT
3759 int follow_phys(struct vm_area_struct
*vma
,
3760 unsigned long address
, unsigned int flags
,
3761 unsigned long *prot
, resource_size_t
*phys
)
3767 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
3770 if (follow_pte(vma
->vm_mm
, address
, &ptep
, &ptl
))
3774 if ((flags
& FOLL_WRITE
) && !pte_write(pte
))
3777 *prot
= pgprot_val(pte_pgprot(pte
));
3778 *phys
= (resource_size_t
)pte_pfn(pte
) << PAGE_SHIFT
;
3782 pte_unmap_unlock(ptep
, ptl
);
3787 int generic_access_phys(struct vm_area_struct
*vma
, unsigned long addr
,
3788 void *buf
, int len
, int write
)
3790 resource_size_t phys_addr
;
3791 unsigned long prot
= 0;
3792 void __iomem
*maddr
;
3793 int offset
= addr
& (PAGE_SIZE
-1);
3795 if (follow_phys(vma
, addr
, write
, &prot
, &phys_addr
))
3798 maddr
= ioremap_prot(phys_addr
, PAGE_SIZE
, prot
);
3800 memcpy_toio(maddr
+ offset
, buf
, len
);
3802 memcpy_fromio(buf
, maddr
+ offset
, len
);
3810 * Access another process' address space as given in mm. If non-NULL, use the
3811 * given task for page fault accounting.
3813 static int __access_remote_vm(struct task_struct
*tsk
, struct mm_struct
*mm
,
3814 unsigned long addr
, void *buf
, int len
, int write
)
3816 struct vm_area_struct
*vma
;
3817 void *old_buf
= buf
;
3819 down_read(&mm
->mmap_sem
);
3820 /* ignore errors, just check how much was successfully transferred */
3822 int bytes
, ret
, offset
;
3824 struct page
*page
= NULL
;
3826 ret
= get_user_pages(tsk
, mm
, addr
, 1,
3827 write
, 1, &page
, &vma
);
3830 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3831 * we can access using slightly different code.
3833 #ifdef CONFIG_HAVE_IOREMAP_PROT
3834 vma
= find_vma(mm
, addr
);
3835 if (!vma
|| vma
->vm_start
> addr
)
3837 if (vma
->vm_ops
&& vma
->vm_ops
->access
)
3838 ret
= vma
->vm_ops
->access(vma
, addr
, buf
,
3846 offset
= addr
& (PAGE_SIZE
-1);
3847 if (bytes
> PAGE_SIZE
-offset
)
3848 bytes
= PAGE_SIZE
-offset
;
3852 copy_to_user_page(vma
, page
, addr
,
3853 maddr
+ offset
, buf
, bytes
);
3854 set_page_dirty_lock(page
);
3856 copy_from_user_page(vma
, page
, addr
,
3857 buf
, maddr
+ offset
, bytes
);
3860 page_cache_release(page
);
3866 up_read(&mm
->mmap_sem
);
3868 return buf
- old_buf
;
3872 * access_remote_vm - access another process' address space
3873 * @mm: the mm_struct of the target address space
3874 * @addr: start address to access
3875 * @buf: source or destination buffer
3876 * @len: number of bytes to transfer
3877 * @write: whether the access is a write
3879 * The caller must hold a reference on @mm.
3881 int access_remote_vm(struct mm_struct
*mm
, unsigned long addr
,
3882 void *buf
, int len
, int write
)
3884 return __access_remote_vm(NULL
, mm
, addr
, buf
, len
, write
);
3888 * Access another process' address space.
3889 * Source/target buffer must be kernel space,
3890 * Do not walk the page table directly, use get_user_pages
3892 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
,
3893 void *buf
, int len
, int write
)
3895 struct mm_struct
*mm
;
3898 mm
= get_task_mm(tsk
);
3902 ret
= __access_remote_vm(tsk
, mm
, addr
, buf
, len
, write
);
3909 * Print the name of a VMA.
3911 void print_vma_addr(char *prefix
, unsigned long ip
)
3913 struct mm_struct
*mm
= current
->mm
;
3914 struct vm_area_struct
*vma
;
3917 * Do not print if we are in atomic
3918 * contexts (in exception stacks, etc.):
3920 if (preempt_count())
3923 down_read(&mm
->mmap_sem
);
3924 vma
= find_vma(mm
, ip
);
3925 if (vma
&& vma
->vm_file
) {
3926 struct file
*f
= vma
->vm_file
;
3927 char *buf
= (char *)__get_free_page(GFP_KERNEL
);
3931 p
= d_path(&f
->f_path
, buf
, PAGE_SIZE
);
3934 s
= strrchr(p
, '/');
3937 printk("%s%s[%lx+%lx]", prefix
, p
,
3939 vma
->vm_end
- vma
->vm_start
);
3940 free_page((unsigned long)buf
);
3943 up_read(&mm
->mmap_sem
);
3946 #ifdef CONFIG_PROVE_LOCKING
3947 void might_fault(void)
3950 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3951 * holding the mmap_sem, this is safe because kernel memory doesn't
3952 * get paged out, therefore we'll never actually fault, and the
3953 * below annotations will generate false positives.
3955 if (segment_eq(get_fs(), KERNEL_DS
))
3960 * it would be nicer only to annotate paths which are not under
3961 * pagefault_disable, however that requires a larger audit and
3962 * providing helpers like get_user_atomic.
3964 if (!in_atomic() && current
->mm
)
3965 might_lock_read(¤t
->mm
->mmap_sem
);
3967 EXPORT_SYMBOL(might_fault
);
3970 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
3971 static void clear_gigantic_page(struct page
*page
,
3973 unsigned int pages_per_huge_page
)
3976 struct page
*p
= page
;
3979 for (i
= 0; i
< pages_per_huge_page
;
3980 i
++, p
= mem_map_next(p
, page
, i
)) {
3982 clear_user_highpage(p
, addr
+ i
* PAGE_SIZE
);
3985 void clear_huge_page(struct page
*page
,
3986 unsigned long addr
, unsigned int pages_per_huge_page
)
3990 if (unlikely(pages_per_huge_page
> MAX_ORDER_NR_PAGES
)) {
3991 clear_gigantic_page(page
, addr
, pages_per_huge_page
);
3996 for (i
= 0; i
< pages_per_huge_page
; i
++) {
3998 clear_user_highpage(page
+ i
, addr
+ i
* PAGE_SIZE
);
4002 static void copy_user_gigantic_page(struct page
*dst
, struct page
*src
,
4004 struct vm_area_struct
*vma
,
4005 unsigned int pages_per_huge_page
)
4008 struct page
*dst_base
= dst
;
4009 struct page
*src_base
= src
;
4011 for (i
= 0; i
< pages_per_huge_page
; ) {
4013 copy_user_highpage(dst
, src
, addr
+ i
*PAGE_SIZE
, vma
);
4016 dst
= mem_map_next(dst
, dst_base
, i
);
4017 src
= mem_map_next(src
, src_base
, i
);
4021 void copy_user_huge_page(struct page
*dst
, struct page
*src
,
4022 unsigned long addr
, struct vm_area_struct
*vma
,
4023 unsigned int pages_per_huge_page
)
4027 if (unlikely(pages_per_huge_page
> MAX_ORDER_NR_PAGES
)) {
4028 copy_user_gigantic_page(dst
, src
, addr
, vma
,
4029 pages_per_huge_page
);
4034 for (i
= 0; i
< pages_per_huge_page
; i
++) {
4036 copy_user_highpage(dst
+ i
, src
+ i
, addr
+ i
*PAGE_SIZE
, vma
);
4039 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */