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/module.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>
61 #include <asm/pgalloc.h>
62 #include <asm/uaccess.h>
64 #include <asm/tlbflush.h>
65 #include <asm/pgtable.h>
69 #ifndef CONFIG_NEED_MULTIPLE_NODES
70 /* use the per-pgdat data instead for discontigmem - mbligh */
71 unsigned long max_mapnr
;
74 EXPORT_SYMBOL(max_mapnr
);
75 EXPORT_SYMBOL(mem_map
);
78 unsigned long num_physpages
;
80 * A number of key systems in x86 including ioremap() rely on the assumption
81 * that high_memory defines the upper bound on direct map memory, then end
82 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
83 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
88 EXPORT_SYMBOL(num_physpages
);
89 EXPORT_SYMBOL(high_memory
);
92 * Randomize the address space (stacks, mmaps, brk, etc.).
94 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
95 * as ancient (libc5 based) binaries can segfault. )
97 int randomize_va_space __read_mostly
=
98 #ifdef CONFIG_COMPAT_BRK
104 static int __init
disable_randmaps(char *s
)
106 randomize_va_space
= 0;
109 __setup("norandmaps", disable_randmaps
);
111 static unsigned long zero_pfn __read_mostly
;
114 * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
116 static int __init
init_zero_pfn(void)
118 zero_pfn
= page_to_pfn(ZERO_PAGE(0));
121 core_initcall(init_zero_pfn
);
124 * If a p?d_bad entry is found while walking page tables, report
125 * the error, before resetting entry to p?d_none. Usually (but
126 * very seldom) called out from the p?d_none_or_clear_bad macros.
129 void pgd_clear_bad(pgd_t
*pgd
)
135 void pud_clear_bad(pud_t
*pud
)
141 void pmd_clear_bad(pmd_t
*pmd
)
148 * Note: this doesn't free the actual pages themselves. That
149 * has been handled earlier when unmapping all the memory regions.
151 static void free_pte_range(struct mmu_gather
*tlb
, pmd_t
*pmd
,
154 pgtable_t token
= pmd_pgtable(*pmd
);
156 pte_free_tlb(tlb
, token
, addr
);
160 static inline void free_pmd_range(struct mmu_gather
*tlb
, pud_t
*pud
,
161 unsigned long addr
, unsigned long end
,
162 unsigned long floor
, unsigned long ceiling
)
169 pmd
= pmd_offset(pud
, addr
);
171 next
= pmd_addr_end(addr
, end
);
172 if (pmd_none_or_clear_bad(pmd
))
174 free_pte_range(tlb
, pmd
, addr
);
175 } while (pmd
++, addr
= next
, addr
!= end
);
185 if (end
- 1 > ceiling
- 1)
188 pmd
= pmd_offset(pud
, start
);
190 pmd_free_tlb(tlb
, pmd
, start
);
193 static inline void free_pud_range(struct mmu_gather
*tlb
, pgd_t
*pgd
,
194 unsigned long addr
, unsigned long end
,
195 unsigned long floor
, unsigned long ceiling
)
202 pud
= pud_offset(pgd
, addr
);
204 next
= pud_addr_end(addr
, end
);
205 if (pud_none_or_clear_bad(pud
))
207 free_pmd_range(tlb
, pud
, addr
, next
, floor
, ceiling
);
208 } while (pud
++, addr
= next
, addr
!= end
);
214 ceiling
&= PGDIR_MASK
;
218 if (end
- 1 > ceiling
- 1)
221 pud
= pud_offset(pgd
, start
);
223 pud_free_tlb(tlb
, pud
, start
);
227 * This function frees user-level page tables of a process.
229 * Must be called with pagetable lock held.
231 void free_pgd_range(struct mmu_gather
*tlb
,
232 unsigned long addr
, unsigned long end
,
233 unsigned long floor
, unsigned long ceiling
)
240 * The next few lines have given us lots of grief...
242 * Why are we testing PMD* at this top level? Because often
243 * there will be no work to do at all, and we'd prefer not to
244 * go all the way down to the bottom just to discover that.
246 * Why all these "- 1"s? Because 0 represents both the bottom
247 * of the address space and the top of it (using -1 for the
248 * top wouldn't help much: the masks would do the wrong thing).
249 * The rule is that addr 0 and floor 0 refer to the bottom of
250 * the address space, but end 0 and ceiling 0 refer to the top
251 * Comparisons need to use "end - 1" and "ceiling - 1" (though
252 * that end 0 case should be mythical).
254 * Wherever addr is brought up or ceiling brought down, we must
255 * be careful to reject "the opposite 0" before it confuses the
256 * subsequent tests. But what about where end is brought down
257 * by PMD_SIZE below? no, end can't go down to 0 there.
259 * Whereas we round start (addr) and ceiling down, by different
260 * masks at different levels, in order to test whether a table
261 * now has no other vmas using it, so can be freed, we don't
262 * bother to round floor or end up - the tests don't need that.
276 if (end
- 1 > ceiling
- 1)
282 pgd
= pgd_offset(tlb
->mm
, addr
);
284 next
= pgd_addr_end(addr
, end
);
285 if (pgd_none_or_clear_bad(pgd
))
287 free_pud_range(tlb
, pgd
, addr
, next
, floor
, ceiling
);
288 } while (pgd
++, addr
= next
, addr
!= end
);
291 void free_pgtables(struct mmu_gather
*tlb
, struct vm_area_struct
*vma
,
292 unsigned long floor
, unsigned long ceiling
)
295 struct vm_area_struct
*next
= vma
->vm_next
;
296 unsigned long addr
= vma
->vm_start
;
299 * Hide vma from rmap and vmtruncate before freeing pgtables
301 anon_vma_unlink(vma
);
302 unlink_file_vma(vma
);
304 if (is_vm_hugetlb_page(vma
)) {
305 hugetlb_free_pgd_range(tlb
, addr
, vma
->vm_end
,
306 floor
, next
? next
->vm_start
: ceiling
);
309 * Optimization: gather nearby vmas into one call down
311 while (next
&& next
->vm_start
<= vma
->vm_end
+ PMD_SIZE
312 && !is_vm_hugetlb_page(next
)) {
315 anon_vma_unlink(vma
);
316 unlink_file_vma(vma
);
318 free_pgd_range(tlb
, addr
, vma
->vm_end
,
319 floor
, next
? next
->vm_start
: ceiling
);
325 int __pte_alloc(struct mm_struct
*mm
, pmd_t
*pmd
, unsigned long address
)
327 pgtable_t
new = pte_alloc_one(mm
, address
);
332 * Ensure all pte setup (eg. pte page lock and page clearing) are
333 * visible before the pte is made visible to other CPUs by being
334 * put into page tables.
336 * The other side of the story is the pointer chasing in the page
337 * table walking code (when walking the page table without locking;
338 * ie. most of the time). Fortunately, these data accesses consist
339 * of a chain of data-dependent loads, meaning most CPUs (alpha
340 * being the notable exception) will already guarantee loads are
341 * seen in-order. See the alpha page table accessors for the
342 * smp_read_barrier_depends() barriers in page table walking code.
344 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
346 spin_lock(&mm
->page_table_lock
);
347 if (!pmd_present(*pmd
)) { /* Has another populated it ? */
349 pmd_populate(mm
, pmd
, new);
352 spin_unlock(&mm
->page_table_lock
);
358 int __pte_alloc_kernel(pmd_t
*pmd
, unsigned long address
)
360 pte_t
*new = pte_alloc_one_kernel(&init_mm
, address
);
364 smp_wmb(); /* See comment in __pte_alloc */
366 spin_lock(&init_mm
.page_table_lock
);
367 if (!pmd_present(*pmd
)) { /* Has another populated it ? */
368 pmd_populate_kernel(&init_mm
, pmd
, new);
371 spin_unlock(&init_mm
.page_table_lock
);
373 pte_free_kernel(&init_mm
, new);
377 static inline void add_mm_rss(struct mm_struct
*mm
, int file_rss
, int anon_rss
)
380 add_mm_counter(mm
, file_rss
, file_rss
);
382 add_mm_counter(mm
, anon_rss
, anon_rss
);
386 * This function is called to print an error when a bad pte
387 * is found. For example, we might have a PFN-mapped pte in
388 * a region that doesn't allow it.
390 * The calling function must still handle the error.
392 static void print_bad_pte(struct vm_area_struct
*vma
, unsigned long addr
,
393 pte_t pte
, struct page
*page
)
395 pgd_t
*pgd
= pgd_offset(vma
->vm_mm
, addr
);
396 pud_t
*pud
= pud_offset(pgd
, addr
);
397 pmd_t
*pmd
= pmd_offset(pud
, addr
);
398 struct address_space
*mapping
;
400 static unsigned long resume
;
401 static unsigned long nr_shown
;
402 static unsigned long nr_unshown
;
405 * Allow a burst of 60 reports, then keep quiet for that minute;
406 * or allow a steady drip of one report per second.
408 if (nr_shown
== 60) {
409 if (time_before(jiffies
, resume
)) {
415 "BUG: Bad page map: %lu messages suppressed\n",
422 resume
= jiffies
+ 60 * HZ
;
424 mapping
= vma
->vm_file
? vma
->vm_file
->f_mapping
: NULL
;
425 index
= linear_page_index(vma
, addr
);
428 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
430 (long long)pte_val(pte
), (long long)pmd_val(*pmd
));
433 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
434 page
, (void *)page
->flags
, page_count(page
),
435 page_mapcount(page
), page
->mapping
, page
->index
);
438 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
439 (void *)addr
, vma
->vm_flags
, vma
->anon_vma
, mapping
, index
);
441 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
444 print_symbol(KERN_ALERT
"vma->vm_ops->fault: %s\n",
445 (unsigned long)vma
->vm_ops
->fault
);
446 if (vma
->vm_file
&& vma
->vm_file
->f_op
)
447 print_symbol(KERN_ALERT
"vma->vm_file->f_op->mmap: %s\n",
448 (unsigned long)vma
->vm_file
->f_op
->mmap
);
450 add_taint(TAINT_BAD_PAGE
);
453 static inline int is_cow_mapping(unsigned int flags
)
455 return (flags
& (VM_SHARED
| VM_MAYWRITE
)) == VM_MAYWRITE
;
459 * vm_normal_page -- This function gets the "struct page" associated with a pte.
461 * "Special" mappings do not wish to be associated with a "struct page" (either
462 * it doesn't exist, or it exists but they don't want to touch it). In this
463 * case, NULL is returned here. "Normal" mappings do have a struct page.
465 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
466 * pte bit, in which case this function is trivial. Secondly, an architecture
467 * may not have a spare pte bit, which requires a more complicated scheme,
470 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
471 * special mapping (even if there are underlying and valid "struct pages").
472 * COWed pages of a VM_PFNMAP are always normal.
474 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
475 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
476 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
477 * mapping will always honor the rule
479 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
481 * And for normal mappings this is false.
483 * This restricts such mappings to be a linear translation from virtual address
484 * to pfn. To get around this restriction, we allow arbitrary mappings so long
485 * as the vma is not a COW mapping; in that case, we know that all ptes are
486 * special (because none can have been COWed).
489 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
491 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
492 * page" backing, however the difference is that _all_ pages with a struct
493 * page (that is, those where pfn_valid is true) are refcounted and considered
494 * normal pages by the VM. The disadvantage is that pages are refcounted
495 * (which can be slower and simply not an option for some PFNMAP users). The
496 * advantage is that we don't have to follow the strict linearity rule of
497 * PFNMAP mappings in order to support COWable mappings.
500 #ifdef __HAVE_ARCH_PTE_SPECIAL
501 # define HAVE_PTE_SPECIAL 1
503 # define HAVE_PTE_SPECIAL 0
505 struct page
*vm_normal_page(struct vm_area_struct
*vma
, unsigned long addr
,
508 unsigned long pfn
= pte_pfn(pte
);
510 if (HAVE_PTE_SPECIAL
) {
511 if (likely(!pte_special(pte
)))
513 if (vma
->vm_flags
& (VM_PFNMAP
| VM_MIXEDMAP
))
516 print_bad_pte(vma
, addr
, pte
, NULL
);
520 /* !HAVE_PTE_SPECIAL case follows: */
522 if (unlikely(vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
))) {
523 if (vma
->vm_flags
& VM_MIXEDMAP
) {
529 off
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
530 if (pfn
== vma
->vm_pgoff
+ off
)
532 if (!is_cow_mapping(vma
->vm_flags
))
538 if (unlikely(pfn
> highest_memmap_pfn
)) {
539 print_bad_pte(vma
, addr
, pte
, NULL
);
544 * NOTE! We still have PageReserved() pages in the page tables.
545 * eg. VDSO mappings can cause them to exist.
548 return pfn_to_page(pfn
);
552 * copy one vm_area from one task to the other. Assumes the page tables
553 * already present in the new task to be cleared in the whole range
554 * covered by this vma.
558 copy_one_pte(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
559 pte_t
*dst_pte
, pte_t
*src_pte
, struct vm_area_struct
*vma
,
560 unsigned long addr
, int *rss
)
562 unsigned long vm_flags
= vma
->vm_flags
;
563 pte_t pte
= *src_pte
;
566 /* pte contains position in swap or file, so copy. */
567 if (unlikely(!pte_present(pte
))) {
568 if (!pte_file(pte
)) {
569 swp_entry_t entry
= pte_to_swp_entry(pte
);
571 swap_duplicate(entry
);
572 /* make sure dst_mm is on swapoff's mmlist. */
573 if (unlikely(list_empty(&dst_mm
->mmlist
))) {
574 spin_lock(&mmlist_lock
);
575 if (list_empty(&dst_mm
->mmlist
))
576 list_add(&dst_mm
->mmlist
,
578 spin_unlock(&mmlist_lock
);
580 if (is_write_migration_entry(entry
) &&
581 is_cow_mapping(vm_flags
)) {
583 * COW mappings require pages in both parent
584 * and child to be set to read.
586 make_migration_entry_read(&entry
);
587 pte
= swp_entry_to_pte(entry
);
588 set_pte_at(src_mm
, addr
, src_pte
, pte
);
595 * If it's a COW mapping, write protect it both
596 * in the parent and the child
598 if (is_cow_mapping(vm_flags
)) {
599 ptep_set_wrprotect(src_mm
, addr
, src_pte
);
600 pte
= pte_wrprotect(pte
);
604 * If it's a shared mapping, mark it clean in
607 if (vm_flags
& VM_SHARED
)
608 pte
= pte_mkclean(pte
);
609 pte
= pte_mkold(pte
);
611 page
= vm_normal_page(vma
, addr
, pte
);
615 rss
[PageAnon(page
)]++;
619 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
622 static int copy_pte_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
623 pmd_t
*dst_pmd
, pmd_t
*src_pmd
, struct vm_area_struct
*vma
,
624 unsigned long addr
, unsigned long end
)
626 pte_t
*src_pte
, *dst_pte
;
627 spinlock_t
*src_ptl
, *dst_ptl
;
633 dst_pte
= pte_alloc_map_lock(dst_mm
, dst_pmd
, addr
, &dst_ptl
);
636 src_pte
= pte_offset_map_nested(src_pmd
, addr
);
637 src_ptl
= pte_lockptr(src_mm
, src_pmd
);
638 spin_lock_nested(src_ptl
, SINGLE_DEPTH_NESTING
);
639 arch_enter_lazy_mmu_mode();
643 * We are holding two locks at this point - either of them
644 * could generate latencies in another task on another CPU.
646 if (progress
>= 32) {
648 if (need_resched() ||
649 spin_needbreak(src_ptl
) || spin_needbreak(dst_ptl
))
652 if (pte_none(*src_pte
)) {
656 copy_one_pte(dst_mm
, src_mm
, dst_pte
, src_pte
, vma
, addr
, rss
);
658 } while (dst_pte
++, src_pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
660 arch_leave_lazy_mmu_mode();
661 spin_unlock(src_ptl
);
662 pte_unmap_nested(src_pte
- 1);
663 add_mm_rss(dst_mm
, rss
[0], rss
[1]);
664 pte_unmap_unlock(dst_pte
- 1, dst_ptl
);
671 static inline int copy_pmd_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
672 pud_t
*dst_pud
, pud_t
*src_pud
, struct vm_area_struct
*vma
,
673 unsigned long addr
, unsigned long end
)
675 pmd_t
*src_pmd
, *dst_pmd
;
678 dst_pmd
= pmd_alloc(dst_mm
, dst_pud
, addr
);
681 src_pmd
= pmd_offset(src_pud
, addr
);
683 next
= pmd_addr_end(addr
, end
);
684 if (pmd_none_or_clear_bad(src_pmd
))
686 if (copy_pte_range(dst_mm
, src_mm
, dst_pmd
, src_pmd
,
689 } while (dst_pmd
++, src_pmd
++, addr
= next
, addr
!= end
);
693 static inline int copy_pud_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
694 pgd_t
*dst_pgd
, pgd_t
*src_pgd
, struct vm_area_struct
*vma
,
695 unsigned long addr
, unsigned long end
)
697 pud_t
*src_pud
, *dst_pud
;
700 dst_pud
= pud_alloc(dst_mm
, dst_pgd
, addr
);
703 src_pud
= pud_offset(src_pgd
, addr
);
705 next
= pud_addr_end(addr
, end
);
706 if (pud_none_or_clear_bad(src_pud
))
708 if (copy_pmd_range(dst_mm
, src_mm
, dst_pud
, src_pud
,
711 } while (dst_pud
++, src_pud
++, addr
= next
, addr
!= end
);
715 int copy_page_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
716 struct vm_area_struct
*vma
)
718 pgd_t
*src_pgd
, *dst_pgd
;
720 unsigned long addr
= vma
->vm_start
;
721 unsigned long end
= vma
->vm_end
;
725 * Don't copy ptes where a page fault will fill them correctly.
726 * Fork becomes much lighter when there are big shared or private
727 * readonly mappings. The tradeoff is that copy_page_range is more
728 * efficient than faulting.
730 if (!(vma
->vm_flags
& (VM_HUGETLB
|VM_NONLINEAR
|VM_PFNMAP
|VM_INSERTPAGE
))) {
735 if (is_vm_hugetlb_page(vma
))
736 return copy_hugetlb_page_range(dst_mm
, src_mm
, vma
);
738 if (unlikely(is_pfn_mapping(vma
))) {
740 * We do not free on error cases below as remove_vma
741 * gets called on error from higher level routine
743 ret
= track_pfn_vma_copy(vma
);
749 * We need to invalidate the secondary MMU mappings only when
750 * there could be a permission downgrade on the ptes of the
751 * parent mm. And a permission downgrade will only happen if
752 * is_cow_mapping() returns true.
754 if (is_cow_mapping(vma
->vm_flags
))
755 mmu_notifier_invalidate_range_start(src_mm
, addr
, end
);
758 dst_pgd
= pgd_offset(dst_mm
, addr
);
759 src_pgd
= pgd_offset(src_mm
, addr
);
761 next
= pgd_addr_end(addr
, end
);
762 if (pgd_none_or_clear_bad(src_pgd
))
764 if (unlikely(copy_pud_range(dst_mm
, src_mm
, dst_pgd
, src_pgd
,
769 } while (dst_pgd
++, src_pgd
++, addr
= next
, addr
!= end
);
771 if (is_cow_mapping(vma
->vm_flags
))
772 mmu_notifier_invalidate_range_end(src_mm
,
777 static unsigned long zap_pte_range(struct mmu_gather
*tlb
,
778 struct vm_area_struct
*vma
, pmd_t
*pmd
,
779 unsigned long addr
, unsigned long end
,
780 long *zap_work
, struct zap_details
*details
)
782 struct mm_struct
*mm
= tlb
->mm
;
788 pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
789 arch_enter_lazy_mmu_mode();
792 if (pte_none(ptent
)) {
797 (*zap_work
) -= PAGE_SIZE
;
799 if (pte_present(ptent
)) {
802 page
= vm_normal_page(vma
, addr
, ptent
);
803 if (unlikely(details
) && page
) {
805 * unmap_shared_mapping_pages() wants to
806 * invalidate cache without truncating:
807 * unmap shared but keep private pages.
809 if (details
->check_mapping
&&
810 details
->check_mapping
!= page
->mapping
)
813 * Each page->index must be checked when
814 * invalidating or truncating nonlinear.
816 if (details
->nonlinear_vma
&&
817 (page
->index
< details
->first_index
||
818 page
->index
> details
->last_index
))
821 ptent
= ptep_get_and_clear_full(mm
, addr
, pte
,
823 tlb_remove_tlb_entry(tlb
, pte
, addr
);
826 if (unlikely(details
) && details
->nonlinear_vma
827 && linear_page_index(details
->nonlinear_vma
,
828 addr
) != page
->index
)
829 set_pte_at(mm
, addr
, pte
,
830 pgoff_to_pte(page
->index
));
834 if (pte_dirty(ptent
))
835 set_page_dirty(page
);
836 if (pte_young(ptent
) &&
837 likely(!VM_SequentialReadHint(vma
)))
838 mark_page_accessed(page
);
841 page_remove_rmap(page
);
842 if (unlikely(page_mapcount(page
) < 0))
843 print_bad_pte(vma
, addr
, ptent
, page
);
844 tlb_remove_page(tlb
, page
);
848 * If details->check_mapping, we leave swap entries;
849 * if details->nonlinear_vma, we leave file entries.
851 if (unlikely(details
))
853 if (pte_file(ptent
)) {
854 if (unlikely(!(vma
->vm_flags
& VM_NONLINEAR
)))
855 print_bad_pte(vma
, addr
, ptent
, NULL
);
857 (unlikely(!free_swap_and_cache(pte_to_swp_entry(ptent
))))
858 print_bad_pte(vma
, addr
, ptent
, NULL
);
859 pte_clear_not_present_full(mm
, addr
, pte
, tlb
->fullmm
);
860 } while (pte
++, addr
+= PAGE_SIZE
, (addr
!= end
&& *zap_work
> 0));
862 add_mm_rss(mm
, file_rss
, anon_rss
);
863 arch_leave_lazy_mmu_mode();
864 pte_unmap_unlock(pte
- 1, ptl
);
869 static inline unsigned long zap_pmd_range(struct mmu_gather
*tlb
,
870 struct vm_area_struct
*vma
, pud_t
*pud
,
871 unsigned long addr
, unsigned long end
,
872 long *zap_work
, struct zap_details
*details
)
877 pmd
= pmd_offset(pud
, addr
);
879 next
= pmd_addr_end(addr
, end
);
880 if (pmd_none_or_clear_bad(pmd
)) {
884 next
= zap_pte_range(tlb
, vma
, pmd
, addr
, next
,
886 } while (pmd
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
891 static inline unsigned long zap_pud_range(struct mmu_gather
*tlb
,
892 struct vm_area_struct
*vma
, pgd_t
*pgd
,
893 unsigned long addr
, unsigned long end
,
894 long *zap_work
, struct zap_details
*details
)
899 pud
= pud_offset(pgd
, addr
);
901 next
= pud_addr_end(addr
, end
);
902 if (pud_none_or_clear_bad(pud
)) {
906 next
= zap_pmd_range(tlb
, vma
, pud
, addr
, next
,
908 } while (pud
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
913 static unsigned long unmap_page_range(struct mmu_gather
*tlb
,
914 struct vm_area_struct
*vma
,
915 unsigned long addr
, unsigned long end
,
916 long *zap_work
, struct zap_details
*details
)
921 if (details
&& !details
->check_mapping
&& !details
->nonlinear_vma
)
925 tlb_start_vma(tlb
, vma
);
926 pgd
= pgd_offset(vma
->vm_mm
, addr
);
928 next
= pgd_addr_end(addr
, end
);
929 if (pgd_none_or_clear_bad(pgd
)) {
933 next
= zap_pud_range(tlb
, vma
, pgd
, addr
, next
,
935 } while (pgd
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
936 tlb_end_vma(tlb
, vma
);
941 #ifdef CONFIG_PREEMPT
942 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
944 /* No preempt: go for improved straight-line efficiency */
945 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
949 * unmap_vmas - unmap a range of memory covered by a list of vma's
950 * @tlbp: address of the caller's struct mmu_gather
951 * @vma: the starting vma
952 * @start_addr: virtual address at which to start unmapping
953 * @end_addr: virtual address at which to end unmapping
954 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
955 * @details: details of nonlinear truncation or shared cache invalidation
957 * Returns the end address of the unmapping (restart addr if interrupted).
959 * Unmap all pages in the vma list.
961 * We aim to not hold locks for too long (for scheduling latency reasons).
962 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
963 * return the ending mmu_gather to the caller.
965 * Only addresses between `start' and `end' will be unmapped.
967 * The VMA list must be sorted in ascending virtual address order.
969 * unmap_vmas() assumes that the caller will flush the whole unmapped address
970 * range after unmap_vmas() returns. So the only responsibility here is to
971 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
972 * drops the lock and schedules.
974 unsigned long unmap_vmas(struct mmu_gather
**tlbp
,
975 struct vm_area_struct
*vma
, unsigned long start_addr
,
976 unsigned long end_addr
, unsigned long *nr_accounted
,
977 struct zap_details
*details
)
979 long zap_work
= ZAP_BLOCK_SIZE
;
980 unsigned long tlb_start
= 0; /* For tlb_finish_mmu */
981 int tlb_start_valid
= 0;
982 unsigned long start
= start_addr
;
983 spinlock_t
*i_mmap_lock
= details
? details
->i_mmap_lock
: NULL
;
984 int fullmm
= (*tlbp
)->fullmm
;
985 struct mm_struct
*mm
= vma
->vm_mm
;
987 mmu_notifier_invalidate_range_start(mm
, start_addr
, end_addr
);
988 for ( ; vma
&& vma
->vm_start
< end_addr
; vma
= vma
->vm_next
) {
991 start
= max(vma
->vm_start
, start_addr
);
992 if (start
>= vma
->vm_end
)
994 end
= min(vma
->vm_end
, end_addr
);
995 if (end
<= vma
->vm_start
)
998 if (vma
->vm_flags
& VM_ACCOUNT
)
999 *nr_accounted
+= (end
- start
) >> PAGE_SHIFT
;
1001 if (unlikely(is_pfn_mapping(vma
)))
1002 untrack_pfn_vma(vma
, 0, 0);
1004 while (start
!= end
) {
1005 if (!tlb_start_valid
) {
1007 tlb_start_valid
= 1;
1010 if (unlikely(is_vm_hugetlb_page(vma
))) {
1012 * It is undesirable to test vma->vm_file as it
1013 * should be non-null for valid hugetlb area.
1014 * However, vm_file will be NULL in the error
1015 * cleanup path of do_mmap_pgoff. When
1016 * hugetlbfs ->mmap method fails,
1017 * do_mmap_pgoff() nullifies vma->vm_file
1018 * before calling this function to clean up.
1019 * Since no pte has actually been setup, it is
1020 * safe to do nothing in this case.
1023 unmap_hugepage_range(vma
, start
, end
, NULL
);
1024 zap_work
-= (end
- start
) /
1025 pages_per_huge_page(hstate_vma(vma
));
1030 start
= unmap_page_range(*tlbp
, vma
,
1031 start
, end
, &zap_work
, details
);
1034 BUG_ON(start
!= end
);
1038 tlb_finish_mmu(*tlbp
, tlb_start
, start
);
1040 if (need_resched() ||
1041 (i_mmap_lock
&& spin_needbreak(i_mmap_lock
))) {
1049 *tlbp
= tlb_gather_mmu(vma
->vm_mm
, fullmm
);
1050 tlb_start_valid
= 0;
1051 zap_work
= ZAP_BLOCK_SIZE
;
1055 mmu_notifier_invalidate_range_end(mm
, start_addr
, end_addr
);
1056 return start
; /* which is now the end (or restart) address */
1060 * zap_page_range - remove user pages in a given range
1061 * @vma: vm_area_struct holding the applicable pages
1062 * @address: starting address of pages to zap
1063 * @size: number of bytes to zap
1064 * @details: details of nonlinear truncation or shared cache invalidation
1066 unsigned long zap_page_range(struct vm_area_struct
*vma
, unsigned long address
,
1067 unsigned long size
, struct zap_details
*details
)
1069 struct mm_struct
*mm
= vma
->vm_mm
;
1070 struct mmu_gather
*tlb
;
1071 unsigned long end
= address
+ size
;
1072 unsigned long nr_accounted
= 0;
1075 tlb
= tlb_gather_mmu(mm
, 0);
1076 update_hiwater_rss(mm
);
1077 end
= unmap_vmas(&tlb
, vma
, address
, end
, &nr_accounted
, details
);
1079 tlb_finish_mmu(tlb
, address
, end
);
1084 * zap_vma_ptes - remove ptes mapping the vma
1085 * @vma: vm_area_struct holding ptes to be zapped
1086 * @address: starting address of pages to zap
1087 * @size: number of bytes to zap
1089 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1091 * The entire address range must be fully contained within the vma.
1093 * Returns 0 if successful.
1095 int zap_vma_ptes(struct vm_area_struct
*vma
, unsigned long address
,
1098 if (address
< vma
->vm_start
|| address
+ size
> vma
->vm_end
||
1099 !(vma
->vm_flags
& VM_PFNMAP
))
1101 zap_page_range(vma
, address
, size
, NULL
);
1104 EXPORT_SYMBOL_GPL(zap_vma_ptes
);
1107 * Do a quick page-table lookup for a single page.
1109 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
1118 struct mm_struct
*mm
= vma
->vm_mm
;
1120 page
= follow_huge_addr(mm
, address
, flags
& FOLL_WRITE
);
1121 if (!IS_ERR(page
)) {
1122 BUG_ON(flags
& FOLL_GET
);
1127 pgd
= pgd_offset(mm
, address
);
1128 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
1131 pud
= pud_offset(pgd
, address
);
1134 if (pud_huge(*pud
)) {
1135 BUG_ON(flags
& FOLL_GET
);
1136 page
= follow_huge_pud(mm
, address
, pud
, flags
& FOLL_WRITE
);
1139 if (unlikely(pud_bad(*pud
)))
1142 pmd
= pmd_offset(pud
, address
);
1145 if (pmd_huge(*pmd
)) {
1146 BUG_ON(flags
& FOLL_GET
);
1147 page
= follow_huge_pmd(mm
, address
, pmd
, flags
& FOLL_WRITE
);
1150 if (unlikely(pmd_bad(*pmd
)))
1153 ptep
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1156 if (!pte_present(pte
))
1158 if ((flags
& FOLL_WRITE
) && !pte_write(pte
))
1161 page
= vm_normal_page(vma
, address
, pte
);
1162 if (unlikely(!page
)) {
1163 if ((flags
& FOLL_DUMP
) ||
1164 pte_pfn(pte
) != zero_pfn
)
1166 page
= pte_page(pte
);
1169 if (flags
& FOLL_GET
)
1171 if (flags
& FOLL_TOUCH
) {
1172 if ((flags
& FOLL_WRITE
) &&
1173 !pte_dirty(pte
) && !PageDirty(page
))
1174 set_page_dirty(page
);
1176 * pte_mkyoung() would be more correct here, but atomic care
1177 * is needed to avoid losing the dirty bit: it is easier to use
1178 * mark_page_accessed().
1180 mark_page_accessed(page
);
1183 pte_unmap_unlock(ptep
, ptl
);
1188 pte_unmap_unlock(ptep
, ptl
);
1189 return ERR_PTR(-EFAULT
);
1192 pte_unmap_unlock(ptep
, ptl
);
1198 * When core dumping an enormous anonymous area that nobody
1199 * has touched so far, we don't want to allocate unnecessary pages or
1200 * page tables. Return error instead of NULL to skip handle_mm_fault,
1201 * then get_dump_page() will return NULL to leave a hole in the dump.
1202 * But we can only make this optimization where a hole would surely
1203 * be zero-filled if handle_mm_fault() actually did handle it.
1205 if ((flags
& FOLL_DUMP
) &&
1206 (!vma
->vm_ops
|| !vma
->vm_ops
->fault
))
1207 return ERR_PTR(-EFAULT
);
1211 int __get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
1212 unsigned long start
, int nr_pages
, int flags
,
1213 struct page
**pages
, struct vm_area_struct
**vmas
)
1216 unsigned int vm_flags
= 0;
1217 int write
= !!(flags
& GUP_FLAGS_WRITE
);
1218 int force
= !!(flags
& GUP_FLAGS_FORCE
);
1223 * Require read or write permissions.
1224 * If 'force' is set, we only require the "MAY" flags.
1226 vm_flags
= write
? (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
1227 vm_flags
&= force
? (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
1231 struct vm_area_struct
*vma
;
1232 unsigned int foll_flags
;
1234 vma
= find_extend_vma(mm
, start
);
1235 if (!vma
&& in_gate_area(tsk
, start
)) {
1236 unsigned long pg
= start
& PAGE_MASK
;
1237 struct vm_area_struct
*gate_vma
= get_gate_vma(tsk
);
1243 /* user gate pages are read-only */
1245 return i
? : -EFAULT
;
1247 pgd
= pgd_offset_k(pg
);
1249 pgd
= pgd_offset_gate(mm
, pg
);
1250 BUG_ON(pgd_none(*pgd
));
1251 pud
= pud_offset(pgd
, pg
);
1252 BUG_ON(pud_none(*pud
));
1253 pmd
= pmd_offset(pud
, pg
);
1255 return i
? : -EFAULT
;
1256 pte
= pte_offset_map(pmd
, pg
);
1257 if (pte_none(*pte
)) {
1259 return i
? : -EFAULT
;
1262 struct page
*page
= vm_normal_page(gate_vma
, start
, *pte
);
1277 (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) ||
1278 !(vm_flags
& vma
->vm_flags
))
1279 return i
? : -EFAULT
;
1281 foll_flags
= FOLL_TOUCH
;
1283 foll_flags
|= FOLL_GET
;
1284 if (flags
& GUP_FLAGS_DUMP
)
1285 foll_flags
|= FOLL_DUMP
;
1287 foll_flags
|= FOLL_WRITE
;
1289 if (is_vm_hugetlb_page(vma
)) {
1290 i
= follow_hugetlb_page(mm
, vma
, pages
, vmas
,
1291 &start
, &nr_pages
, i
, foll_flags
);
1299 * If we have a pending SIGKILL, don't keep faulting
1300 * pages and potentially allocating memory.
1302 if (unlikely(fatal_signal_pending(current
)))
1303 return i
? i
: -ERESTARTSYS
;
1306 foll_flags
|= FOLL_WRITE
;
1309 while (!(page
= follow_page(vma
, start
, foll_flags
))) {
1312 ret
= handle_mm_fault(mm
, vma
, start
,
1313 (foll_flags
& FOLL_WRITE
) ?
1314 FAULT_FLAG_WRITE
: 0);
1316 if (ret
& VM_FAULT_ERROR
) {
1317 if (ret
& VM_FAULT_OOM
)
1318 return i
? i
: -ENOMEM
;
1319 else if (ret
& VM_FAULT_SIGBUS
)
1320 return i
? i
: -EFAULT
;
1323 if (ret
& VM_FAULT_MAJOR
)
1329 * The VM_FAULT_WRITE bit tells us that
1330 * do_wp_page has broken COW when necessary,
1331 * even if maybe_mkwrite decided not to set
1332 * pte_write. We can thus safely do subsequent
1333 * page lookups as if they were reads. But only
1334 * do so when looping for pte_write is futile:
1335 * in some cases userspace may also be wanting
1336 * to write to the gotten user page, which a
1337 * read fault here might prevent (a readonly
1338 * page might get reCOWed by userspace write).
1340 if ((ret
& VM_FAULT_WRITE
) &&
1341 !(vma
->vm_flags
& VM_WRITE
))
1342 foll_flags
&= ~FOLL_WRITE
;
1347 return i
? i
: PTR_ERR(page
);
1351 flush_anon_page(vma
, page
, start
);
1352 flush_dcache_page(page
);
1359 } while (nr_pages
&& start
< vma
->vm_end
);
1365 * get_user_pages() - pin user pages in memory
1366 * @tsk: task_struct of target task
1367 * @mm: mm_struct of target mm
1368 * @start: starting user address
1369 * @nr_pages: number of pages from start to pin
1370 * @write: whether pages will be written to by the caller
1371 * @force: whether to force write access even if user mapping is
1372 * readonly. This will result in the page being COWed even
1373 * in MAP_SHARED mappings. You do not want this.
1374 * @pages: array that receives pointers to the pages pinned.
1375 * Should be at least nr_pages long. Or NULL, if caller
1376 * only intends to ensure the pages are faulted in.
1377 * @vmas: array of pointers to vmas corresponding to each page.
1378 * Or NULL if the caller does not require them.
1380 * Returns number of pages pinned. This may be fewer than the number
1381 * requested. If nr_pages is 0 or negative, returns 0. If no pages
1382 * were pinned, returns -errno. Each page returned must be released
1383 * with a put_page() call when it is finished with. vmas will only
1384 * remain valid while mmap_sem is held.
1386 * Must be called with mmap_sem held for read or write.
1388 * get_user_pages walks a process's page tables and takes a reference to
1389 * each struct page that each user address corresponds to at a given
1390 * instant. That is, it takes the page that would be accessed if a user
1391 * thread accesses the given user virtual address at that instant.
1393 * This does not guarantee that the page exists in the user mappings when
1394 * get_user_pages returns, and there may even be a completely different
1395 * page there in some cases (eg. if mmapped pagecache has been invalidated
1396 * and subsequently re faulted). However it does guarantee that the page
1397 * won't be freed completely. And mostly callers simply care that the page
1398 * contains data that was valid *at some point in time*. Typically, an IO
1399 * or similar operation cannot guarantee anything stronger anyway because
1400 * locks can't be held over the syscall boundary.
1402 * If write=0, the page must not be written to. If the page is written to,
1403 * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
1404 * after the page is finished with, and before put_page is called.
1406 * get_user_pages is typically used for fewer-copy IO operations, to get a
1407 * handle on the memory by some means other than accesses via the user virtual
1408 * addresses. The pages may be submitted for DMA to devices or accessed via
1409 * their kernel linear mapping (via the kmap APIs). Care should be taken to
1410 * use the correct cache flushing APIs.
1412 * See also get_user_pages_fast, for performance critical applications.
1414 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
1415 unsigned long start
, int nr_pages
, int write
, int force
,
1416 struct page
**pages
, struct vm_area_struct
**vmas
)
1421 flags
|= GUP_FLAGS_WRITE
;
1423 flags
|= GUP_FLAGS_FORCE
;
1425 return __get_user_pages(tsk
, mm
, start
, nr_pages
, flags
, pages
, vmas
);
1427 EXPORT_SYMBOL(get_user_pages
);
1430 * get_dump_page() - pin user page in memory while writing it to core dump
1431 * @addr: user address
1433 * Returns struct page pointer of user page pinned for dump,
1434 * to be freed afterwards by page_cache_release() or put_page().
1436 * Returns NULL on any kind of failure - a hole must then be inserted into
1437 * the corefile, to preserve alignment with its headers; and also returns
1438 * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
1439 * allowing a hole to be left in the corefile to save diskspace.
1441 * Called without mmap_sem, but after all other threads have been killed.
1443 #ifdef CONFIG_ELF_CORE
1444 struct page
*get_dump_page(unsigned long addr
)
1446 struct vm_area_struct
*vma
;
1449 if (__get_user_pages(current
, current
->mm
, addr
, 1,
1450 GUP_FLAGS_FORCE
| GUP_FLAGS_DUMP
, &page
, &vma
) < 1)
1452 if (page
== ZERO_PAGE(0)) {
1453 page_cache_release(page
);
1456 flush_cache_page(vma
, addr
, page_to_pfn(page
));
1459 #endif /* CONFIG_ELF_CORE */
1461 pte_t
*get_locked_pte(struct mm_struct
*mm
, unsigned long addr
,
1464 pgd_t
* pgd
= pgd_offset(mm
, addr
);
1465 pud_t
* pud
= pud_alloc(mm
, pgd
, addr
);
1467 pmd_t
* pmd
= pmd_alloc(mm
, pud
, addr
);
1469 return pte_alloc_map_lock(mm
, pmd
, addr
, ptl
);
1475 * This is the old fallback for page remapping.
1477 * For historical reasons, it only allows reserved pages. Only
1478 * old drivers should use this, and they needed to mark their
1479 * pages reserved for the old functions anyway.
1481 static int insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
1482 struct page
*page
, pgprot_t prot
)
1484 struct mm_struct
*mm
= vma
->vm_mm
;
1493 flush_dcache_page(page
);
1494 pte
= get_locked_pte(mm
, addr
, &ptl
);
1498 if (!pte_none(*pte
))
1501 /* Ok, finally just insert the thing.. */
1503 inc_mm_counter(mm
, file_rss
);
1504 page_add_file_rmap(page
);
1505 set_pte_at(mm
, addr
, pte
, mk_pte(page
, prot
));
1508 pte_unmap_unlock(pte
, ptl
);
1511 pte_unmap_unlock(pte
, ptl
);
1517 * vm_insert_page - insert single page into user vma
1518 * @vma: user vma to map to
1519 * @addr: target user address of this page
1520 * @page: source kernel page
1522 * This allows drivers to insert individual pages they've allocated
1525 * The page has to be a nice clean _individual_ kernel allocation.
1526 * If you allocate a compound page, you need to have marked it as
1527 * such (__GFP_COMP), or manually just split the page up yourself
1528 * (see split_page()).
1530 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1531 * took an arbitrary page protection parameter. This doesn't allow
1532 * that. Your vma protection will have to be set up correctly, which
1533 * means that if you want a shared writable mapping, you'd better
1534 * ask for a shared writable mapping!
1536 * The page does not need to be reserved.
1538 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
1541 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
1543 if (!page_count(page
))
1545 vma
->vm_flags
|= VM_INSERTPAGE
;
1546 return insert_page(vma
, addr
, page
, vma
->vm_page_prot
);
1548 EXPORT_SYMBOL(vm_insert_page
);
1550 static int insert_pfn(struct vm_area_struct
*vma
, unsigned long addr
,
1551 unsigned long pfn
, pgprot_t prot
)
1553 struct mm_struct
*mm
= vma
->vm_mm
;
1559 pte
= get_locked_pte(mm
, addr
, &ptl
);
1563 if (!pte_none(*pte
))
1566 /* Ok, finally just insert the thing.. */
1567 entry
= pte_mkspecial(pfn_pte(pfn
, prot
));
1568 set_pte_at(mm
, addr
, pte
, entry
);
1569 update_mmu_cache(vma
, addr
, entry
); /* XXX: why not for insert_page? */
1573 pte_unmap_unlock(pte
, ptl
);
1579 * vm_insert_pfn - insert single pfn into user vma
1580 * @vma: user vma to map to
1581 * @addr: target user address of this page
1582 * @pfn: source kernel pfn
1584 * Similar to vm_inert_page, this allows drivers to insert individual pages
1585 * they've allocated into a user vma. Same comments apply.
1587 * This function should only be called from a vm_ops->fault handler, and
1588 * in that case the handler should return NULL.
1590 * vma cannot be a COW mapping.
1592 * As this is called only for pages that do not currently exist, we
1593 * do not need to flush old virtual caches or the TLB.
1595 int vm_insert_pfn(struct vm_area_struct
*vma
, unsigned long addr
,
1599 pgprot_t pgprot
= vma
->vm_page_prot
;
1601 * Technically, architectures with pte_special can avoid all these
1602 * restrictions (same for remap_pfn_range). However we would like
1603 * consistency in testing and feature parity among all, so we should
1604 * try to keep these invariants in place for everybody.
1606 BUG_ON(!(vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
)));
1607 BUG_ON((vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
)) ==
1608 (VM_PFNMAP
|VM_MIXEDMAP
));
1609 BUG_ON((vma
->vm_flags
& VM_PFNMAP
) && is_cow_mapping(vma
->vm_flags
));
1610 BUG_ON((vma
->vm_flags
& VM_MIXEDMAP
) && pfn_valid(pfn
));
1612 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
1614 if (track_pfn_vma_new(vma
, &pgprot
, pfn
, PAGE_SIZE
))
1617 ret
= insert_pfn(vma
, addr
, pfn
, pgprot
);
1620 untrack_pfn_vma(vma
, pfn
, PAGE_SIZE
);
1624 EXPORT_SYMBOL(vm_insert_pfn
);
1626 int vm_insert_mixed(struct vm_area_struct
*vma
, unsigned long addr
,
1629 BUG_ON(!(vma
->vm_flags
& VM_MIXEDMAP
));
1631 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
1635 * If we don't have pte special, then we have to use the pfn_valid()
1636 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1637 * refcount the page if pfn_valid is true (hence insert_page rather
1640 if (!HAVE_PTE_SPECIAL
&& pfn_valid(pfn
)) {
1643 page
= pfn_to_page(pfn
);
1644 return insert_page(vma
, addr
, page
, vma
->vm_page_prot
);
1646 return insert_pfn(vma
, addr
, pfn
, vma
->vm_page_prot
);
1648 EXPORT_SYMBOL(vm_insert_mixed
);
1651 * maps a range of physical memory into the requested pages. the old
1652 * mappings are removed. any references to nonexistent pages results
1653 * in null mappings (currently treated as "copy-on-access")
1655 static int remap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1656 unsigned long addr
, unsigned long end
,
1657 unsigned long pfn
, pgprot_t prot
)
1662 pte
= pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1665 arch_enter_lazy_mmu_mode();
1667 BUG_ON(!pte_none(*pte
));
1668 set_pte_at(mm
, addr
, pte
, pte_mkspecial(pfn_pte(pfn
, prot
)));
1670 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1671 arch_leave_lazy_mmu_mode();
1672 pte_unmap_unlock(pte
- 1, ptl
);
1676 static inline int remap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1677 unsigned long addr
, unsigned long end
,
1678 unsigned long pfn
, pgprot_t prot
)
1683 pfn
-= addr
>> PAGE_SHIFT
;
1684 pmd
= pmd_alloc(mm
, pud
, addr
);
1688 next
= pmd_addr_end(addr
, end
);
1689 if (remap_pte_range(mm
, pmd
, addr
, next
,
1690 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1692 } while (pmd
++, addr
= next
, addr
!= end
);
1696 static inline int remap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1697 unsigned long addr
, unsigned long end
,
1698 unsigned long pfn
, pgprot_t prot
)
1703 pfn
-= addr
>> PAGE_SHIFT
;
1704 pud
= pud_alloc(mm
, pgd
, addr
);
1708 next
= pud_addr_end(addr
, end
);
1709 if (remap_pmd_range(mm
, pud
, addr
, next
,
1710 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1712 } while (pud
++, addr
= next
, addr
!= end
);
1717 * remap_pfn_range - remap kernel memory to userspace
1718 * @vma: user vma to map to
1719 * @addr: target user address to start at
1720 * @pfn: physical address of kernel memory
1721 * @size: size of map area
1722 * @prot: page protection flags for this mapping
1724 * Note: this is only safe if the mm semaphore is held when called.
1726 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1727 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1731 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1732 struct mm_struct
*mm
= vma
->vm_mm
;
1736 * Physically remapped pages are special. Tell the
1737 * rest of the world about it:
1738 * VM_IO tells people not to look at these pages
1739 * (accesses can have side effects).
1740 * VM_RESERVED is specified all over the place, because
1741 * in 2.4 it kept swapout's vma scan off this vma; but
1742 * in 2.6 the LRU scan won't even find its pages, so this
1743 * flag means no more than count its pages in reserved_vm,
1744 * and omit it from core dump, even when VM_IO turned off.
1745 * VM_PFNMAP tells the core MM that the base pages are just
1746 * raw PFN mappings, and do not have a "struct page" associated
1749 * There's a horrible special case to handle copy-on-write
1750 * behaviour that some programs depend on. We mark the "original"
1751 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1753 if (addr
== vma
->vm_start
&& end
== vma
->vm_end
) {
1754 vma
->vm_pgoff
= pfn
;
1755 vma
->vm_flags
|= VM_PFN_AT_MMAP
;
1756 } else if (is_cow_mapping(vma
->vm_flags
))
1759 vma
->vm_flags
|= VM_IO
| VM_RESERVED
| VM_PFNMAP
;
1761 err
= track_pfn_vma_new(vma
, &prot
, pfn
, PAGE_ALIGN(size
));
1764 * To indicate that track_pfn related cleanup is not
1765 * needed from higher level routine calling unmap_vmas
1767 vma
->vm_flags
&= ~(VM_IO
| VM_RESERVED
| VM_PFNMAP
);
1768 vma
->vm_flags
&= ~VM_PFN_AT_MMAP
;
1772 BUG_ON(addr
>= end
);
1773 pfn
-= addr
>> PAGE_SHIFT
;
1774 pgd
= pgd_offset(mm
, addr
);
1775 flush_cache_range(vma
, addr
, end
);
1777 next
= pgd_addr_end(addr
, end
);
1778 err
= remap_pud_range(mm
, pgd
, addr
, next
,
1779 pfn
+ (addr
>> PAGE_SHIFT
), prot
);
1782 } while (pgd
++, addr
= next
, addr
!= end
);
1785 untrack_pfn_vma(vma
, pfn
, PAGE_ALIGN(size
));
1789 EXPORT_SYMBOL(remap_pfn_range
);
1791 static int apply_to_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1792 unsigned long addr
, unsigned long end
,
1793 pte_fn_t fn
, void *data
)
1798 spinlock_t
*uninitialized_var(ptl
);
1800 pte
= (mm
== &init_mm
) ?
1801 pte_alloc_kernel(pmd
, addr
) :
1802 pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1806 BUG_ON(pmd_huge(*pmd
));
1808 arch_enter_lazy_mmu_mode();
1810 token
= pmd_pgtable(*pmd
);
1813 err
= fn(pte
, token
, addr
, data
);
1816 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1818 arch_leave_lazy_mmu_mode();
1821 pte_unmap_unlock(pte
-1, ptl
);
1825 static int apply_to_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1826 unsigned long addr
, unsigned long end
,
1827 pte_fn_t fn
, void *data
)
1833 BUG_ON(pud_huge(*pud
));
1835 pmd
= pmd_alloc(mm
, pud
, addr
);
1839 next
= pmd_addr_end(addr
, end
);
1840 err
= apply_to_pte_range(mm
, pmd
, addr
, next
, fn
, data
);
1843 } while (pmd
++, addr
= next
, addr
!= end
);
1847 static int apply_to_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1848 unsigned long addr
, unsigned long end
,
1849 pte_fn_t fn
, void *data
)
1855 pud
= pud_alloc(mm
, pgd
, addr
);
1859 next
= pud_addr_end(addr
, end
);
1860 err
= apply_to_pmd_range(mm
, pud
, addr
, next
, fn
, data
);
1863 } while (pud
++, addr
= next
, addr
!= end
);
1868 * Scan a region of virtual memory, filling in page tables as necessary
1869 * and calling a provided function on each leaf page table.
1871 int apply_to_page_range(struct mm_struct
*mm
, unsigned long addr
,
1872 unsigned long size
, pte_fn_t fn
, void *data
)
1876 unsigned long start
= addr
, end
= addr
+ size
;
1879 BUG_ON(addr
>= end
);
1880 mmu_notifier_invalidate_range_start(mm
, start
, end
);
1881 pgd
= pgd_offset(mm
, addr
);
1883 next
= pgd_addr_end(addr
, end
);
1884 err
= apply_to_pud_range(mm
, pgd
, addr
, next
, fn
, data
);
1887 } while (pgd
++, addr
= next
, addr
!= end
);
1888 mmu_notifier_invalidate_range_end(mm
, start
, end
);
1891 EXPORT_SYMBOL_GPL(apply_to_page_range
);
1894 * handle_pte_fault chooses page fault handler according to an entry
1895 * which was read non-atomically. Before making any commitment, on
1896 * those architectures or configurations (e.g. i386 with PAE) which
1897 * might give a mix of unmatched parts, do_swap_page and do_file_page
1898 * must check under lock before unmapping the pte and proceeding
1899 * (but do_wp_page is only called after already making such a check;
1900 * and do_anonymous_page and do_no_page can safely check later on).
1902 static inline int pte_unmap_same(struct mm_struct
*mm
, pmd_t
*pmd
,
1903 pte_t
*page_table
, pte_t orig_pte
)
1906 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1907 if (sizeof(pte_t
) > sizeof(unsigned long)) {
1908 spinlock_t
*ptl
= pte_lockptr(mm
, pmd
);
1910 same
= pte_same(*page_table
, orig_pte
);
1914 pte_unmap(page_table
);
1919 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1920 * servicing faults for write access. In the normal case, do always want
1921 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1922 * that do not have writing enabled, when used by access_process_vm.
1924 static inline pte_t
maybe_mkwrite(pte_t pte
, struct vm_area_struct
*vma
)
1926 if (likely(vma
->vm_flags
& VM_WRITE
))
1927 pte
= pte_mkwrite(pte
);
1931 static inline void cow_user_page(struct page
*dst
, struct page
*src
, unsigned long va
, struct vm_area_struct
*vma
)
1934 * If the source page was a PFN mapping, we don't have
1935 * a "struct page" for it. We do a best-effort copy by
1936 * just copying from the original user address. If that
1937 * fails, we just zero-fill it. Live with it.
1939 if (unlikely(!src
)) {
1940 void *kaddr
= kmap_atomic(dst
, KM_USER0
);
1941 void __user
*uaddr
= (void __user
*)(va
& PAGE_MASK
);
1944 * This really shouldn't fail, because the page is there
1945 * in the page tables. But it might just be unreadable,
1946 * in which case we just give up and fill the result with
1949 if (__copy_from_user_inatomic(kaddr
, uaddr
, PAGE_SIZE
))
1950 memset(kaddr
, 0, PAGE_SIZE
);
1951 kunmap_atomic(kaddr
, KM_USER0
);
1952 flush_dcache_page(dst
);
1954 copy_user_highpage(dst
, src
, va
, vma
);
1958 * This routine handles present pages, when users try to write
1959 * to a shared page. It is done by copying the page to a new address
1960 * and decrementing the shared-page counter for the old page.
1962 * Note that this routine assumes that the protection checks have been
1963 * done by the caller (the low-level page fault routine in most cases).
1964 * Thus we can safely just mark it writable once we've done any necessary
1967 * We also mark the page dirty at this point even though the page will
1968 * change only once the write actually happens. This avoids a few races,
1969 * and potentially makes it more efficient.
1971 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1972 * but allow concurrent faults), with pte both mapped and locked.
1973 * We return with mmap_sem still held, but pte unmapped and unlocked.
1975 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1976 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1977 spinlock_t
*ptl
, pte_t orig_pte
)
1979 struct page
*old_page
, *new_page
;
1981 int reuse
= 0, ret
= 0;
1982 int page_mkwrite
= 0;
1983 struct page
*dirty_page
= NULL
;
1985 old_page
= vm_normal_page(vma
, address
, orig_pte
);
1988 * VM_MIXEDMAP !pfn_valid() case
1990 * We should not cow pages in a shared writeable mapping.
1991 * Just mark the pages writable as we can't do any dirty
1992 * accounting on raw pfn maps.
1994 if ((vma
->vm_flags
& (VM_WRITE
|VM_SHARED
)) ==
1995 (VM_WRITE
|VM_SHARED
))
2001 * Take out anonymous pages first, anonymous shared vmas are
2002 * not dirty accountable.
2004 if (PageAnon(old_page
) && !PageKsm(old_page
)) {
2005 if (!trylock_page(old_page
)) {
2006 page_cache_get(old_page
);
2007 pte_unmap_unlock(page_table
, ptl
);
2008 lock_page(old_page
);
2009 page_table
= pte_offset_map_lock(mm
, pmd
, address
,
2011 if (!pte_same(*page_table
, orig_pte
)) {
2012 unlock_page(old_page
);
2013 page_cache_release(old_page
);
2016 page_cache_release(old_page
);
2018 reuse
= reuse_swap_page(old_page
);
2019 unlock_page(old_page
);
2020 } else if (unlikely((vma
->vm_flags
& (VM_WRITE
|VM_SHARED
)) ==
2021 (VM_WRITE
|VM_SHARED
))) {
2023 * Only catch write-faults on shared writable pages,
2024 * read-only shared pages can get COWed by
2025 * get_user_pages(.write=1, .force=1).
2027 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
) {
2028 struct vm_fault vmf
;
2031 vmf
.virtual_address
= (void __user
*)(address
&
2033 vmf
.pgoff
= old_page
->index
;
2034 vmf
.flags
= FAULT_FLAG_WRITE
|FAULT_FLAG_MKWRITE
;
2035 vmf
.page
= old_page
;
2038 * Notify the address space that the page is about to
2039 * become writable so that it can prohibit this or wait
2040 * for the page to get into an appropriate state.
2042 * We do this without the lock held, so that it can
2043 * sleep if it needs to.
2045 page_cache_get(old_page
);
2046 pte_unmap_unlock(page_table
, ptl
);
2048 tmp
= vma
->vm_ops
->page_mkwrite(vma
, &vmf
);
2050 (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
))) {
2052 goto unwritable_page
;
2054 if (unlikely(!(tmp
& VM_FAULT_LOCKED
))) {
2055 lock_page(old_page
);
2056 if (!old_page
->mapping
) {
2057 ret
= 0; /* retry the fault */
2058 unlock_page(old_page
);
2059 goto unwritable_page
;
2062 VM_BUG_ON(!PageLocked(old_page
));
2065 * Since we dropped the lock we need to revalidate
2066 * the PTE as someone else may have changed it. If
2067 * they did, we just return, as we can count on the
2068 * MMU to tell us if they didn't also make it writable.
2070 page_table
= pte_offset_map_lock(mm
, pmd
, address
,
2072 if (!pte_same(*page_table
, orig_pte
)) {
2073 unlock_page(old_page
);
2074 page_cache_release(old_page
);
2080 dirty_page
= old_page
;
2081 get_page(dirty_page
);
2087 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
2088 entry
= pte_mkyoung(orig_pte
);
2089 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2090 if (ptep_set_access_flags(vma
, address
, page_table
, entry
,1))
2091 update_mmu_cache(vma
, address
, entry
);
2092 ret
|= VM_FAULT_WRITE
;
2097 * Ok, we need to copy. Oh, well..
2099 page_cache_get(old_page
);
2101 pte_unmap_unlock(page_table
, ptl
);
2103 if (unlikely(anon_vma_prepare(vma
)))
2106 if (pte_pfn(orig_pte
) == zero_pfn
) {
2107 new_page
= alloc_zeroed_user_highpage_movable(vma
, address
);
2111 new_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, address
);
2114 cow_user_page(new_page
, old_page
, address
, vma
);
2116 __SetPageUptodate(new_page
);
2119 * Don't let another task, with possibly unlocked vma,
2120 * keep the mlocked page.
2122 if ((vma
->vm_flags
& VM_LOCKED
) && old_page
) {
2123 lock_page(old_page
); /* for LRU manipulation */
2124 clear_page_mlock(old_page
);
2125 unlock_page(old_page
);
2128 if (mem_cgroup_newpage_charge(new_page
, mm
, GFP_KERNEL
))
2132 * Re-check the pte - we dropped the lock
2134 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2135 if (likely(pte_same(*page_table
, orig_pte
))) {
2137 if (!PageAnon(old_page
)) {
2138 dec_mm_counter(mm
, file_rss
);
2139 inc_mm_counter(mm
, anon_rss
);
2142 inc_mm_counter(mm
, anon_rss
);
2143 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
2144 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
2145 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2147 * Clear the pte entry and flush it first, before updating the
2148 * pte with the new entry. This will avoid a race condition
2149 * seen in the presence of one thread doing SMC and another
2152 ptep_clear_flush(vma
, address
, page_table
);
2153 page_add_new_anon_rmap(new_page
, vma
, address
);
2155 * We call the notify macro here because, when using secondary
2156 * mmu page tables (such as kvm shadow page tables), we want the
2157 * new page to be mapped directly into the secondary page table.
2159 set_pte_at_notify(mm
, address
, page_table
, entry
);
2160 update_mmu_cache(vma
, address
, entry
);
2163 * Only after switching the pte to the new page may
2164 * we remove the mapcount here. Otherwise another
2165 * process may come and find the rmap count decremented
2166 * before the pte is switched to the new page, and
2167 * "reuse" the old page writing into it while our pte
2168 * here still points into it and can be read by other
2171 * The critical issue is to order this
2172 * page_remove_rmap with the ptp_clear_flush above.
2173 * Those stores are ordered by (if nothing else,)
2174 * the barrier present in the atomic_add_negative
2175 * in page_remove_rmap.
2177 * Then the TLB flush in ptep_clear_flush ensures that
2178 * no process can access the old page before the
2179 * decremented mapcount is visible. And the old page
2180 * cannot be reused until after the decremented
2181 * mapcount is visible. So transitively, TLBs to
2182 * old page will be flushed before it can be reused.
2184 page_remove_rmap(old_page
);
2187 /* Free the old page.. */
2188 new_page
= old_page
;
2189 ret
|= VM_FAULT_WRITE
;
2191 mem_cgroup_uncharge_page(new_page
);
2194 page_cache_release(new_page
);
2196 page_cache_release(old_page
);
2198 pte_unmap_unlock(page_table
, ptl
);
2201 * Yes, Virginia, this is actually required to prevent a race
2202 * with clear_page_dirty_for_io() from clearing the page dirty
2203 * bit after it clear all dirty ptes, but before a racing
2204 * do_wp_page installs a dirty pte.
2206 * do_no_page is protected similarly.
2208 if (!page_mkwrite
) {
2209 wait_on_page_locked(dirty_page
);
2210 set_page_dirty_balance(dirty_page
, page_mkwrite
);
2212 put_page(dirty_page
);
2214 struct address_space
*mapping
= dirty_page
->mapping
;
2216 set_page_dirty(dirty_page
);
2217 unlock_page(dirty_page
);
2218 page_cache_release(dirty_page
);
2221 * Some device drivers do not set page.mapping
2222 * but still dirty their pages
2224 balance_dirty_pages_ratelimited(mapping
);
2228 /* file_update_time outside page_lock */
2230 file_update_time(vma
->vm_file
);
2234 page_cache_release(new_page
);
2238 unlock_page(old_page
);
2239 page_cache_release(old_page
);
2241 page_cache_release(old_page
);
2243 return VM_FAULT_OOM
;
2246 page_cache_release(old_page
);
2251 * Helper functions for unmap_mapping_range().
2253 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
2255 * We have to restart searching the prio_tree whenever we drop the lock,
2256 * since the iterator is only valid while the lock is held, and anyway
2257 * a later vma might be split and reinserted earlier while lock dropped.
2259 * The list of nonlinear vmas could be handled more efficiently, using
2260 * a placeholder, but handle it in the same way until a need is shown.
2261 * It is important to search the prio_tree before nonlinear list: a vma
2262 * may become nonlinear and be shifted from prio_tree to nonlinear list
2263 * while the lock is dropped; but never shifted from list to prio_tree.
2265 * In order to make forward progress despite restarting the search,
2266 * vm_truncate_count is used to mark a vma as now dealt with, so we can
2267 * quickly skip it next time around. Since the prio_tree search only
2268 * shows us those vmas affected by unmapping the range in question, we
2269 * can't efficiently keep all vmas in step with mapping->truncate_count:
2270 * so instead reset them all whenever it wraps back to 0 (then go to 1).
2271 * mapping->truncate_count and vma->vm_truncate_count are protected by
2274 * In order to make forward progress despite repeatedly restarting some
2275 * large vma, note the restart_addr from unmap_vmas when it breaks out:
2276 * and restart from that address when we reach that vma again. It might
2277 * have been split or merged, shrunk or extended, but never shifted: so
2278 * restart_addr remains valid so long as it remains in the vma's range.
2279 * unmap_mapping_range forces truncate_count to leap over page-aligned
2280 * values so we can save vma's restart_addr in its truncate_count field.
2282 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
2284 static void reset_vma_truncate_counts(struct address_space
*mapping
)
2286 struct vm_area_struct
*vma
;
2287 struct prio_tree_iter iter
;
2289 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, 0, ULONG_MAX
)
2290 vma
->vm_truncate_count
= 0;
2291 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
, shared
.vm_set
.list
)
2292 vma
->vm_truncate_count
= 0;
2295 static int unmap_mapping_range_vma(struct vm_area_struct
*vma
,
2296 unsigned long start_addr
, unsigned long end_addr
,
2297 struct zap_details
*details
)
2299 unsigned long restart_addr
;
2303 * files that support invalidating or truncating portions of the
2304 * file from under mmaped areas must have their ->fault function
2305 * return a locked page (and set VM_FAULT_LOCKED in the return).
2306 * This provides synchronisation against concurrent unmapping here.
2310 restart_addr
= vma
->vm_truncate_count
;
2311 if (is_restart_addr(restart_addr
) && start_addr
< restart_addr
) {
2312 start_addr
= restart_addr
;
2313 if (start_addr
>= end_addr
) {
2314 /* Top of vma has been split off since last time */
2315 vma
->vm_truncate_count
= details
->truncate_count
;
2320 restart_addr
= zap_page_range(vma
, start_addr
,
2321 end_addr
- start_addr
, details
);
2322 need_break
= need_resched() || spin_needbreak(details
->i_mmap_lock
);
2324 if (restart_addr
>= end_addr
) {
2325 /* We have now completed this vma: mark it so */
2326 vma
->vm_truncate_count
= details
->truncate_count
;
2330 /* Note restart_addr in vma's truncate_count field */
2331 vma
->vm_truncate_count
= restart_addr
;
2336 spin_unlock(details
->i_mmap_lock
);
2338 spin_lock(details
->i_mmap_lock
);
2342 static inline void unmap_mapping_range_tree(struct prio_tree_root
*root
,
2343 struct zap_details
*details
)
2345 struct vm_area_struct
*vma
;
2346 struct prio_tree_iter iter
;
2347 pgoff_t vba
, vea
, zba
, zea
;
2350 vma_prio_tree_foreach(vma
, &iter
, root
,
2351 details
->first_index
, details
->last_index
) {
2352 /* Skip quickly over those we have already dealt with */
2353 if (vma
->vm_truncate_count
== details
->truncate_count
)
2356 vba
= vma
->vm_pgoff
;
2357 vea
= vba
+ ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
) - 1;
2358 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2359 zba
= details
->first_index
;
2362 zea
= details
->last_index
;
2366 if (unmap_mapping_range_vma(vma
,
2367 ((zba
- vba
) << PAGE_SHIFT
) + vma
->vm_start
,
2368 ((zea
- vba
+ 1) << PAGE_SHIFT
) + vma
->vm_start
,
2374 static inline void unmap_mapping_range_list(struct list_head
*head
,
2375 struct zap_details
*details
)
2377 struct vm_area_struct
*vma
;
2380 * In nonlinear VMAs there is no correspondence between virtual address
2381 * offset and file offset. So we must perform an exhaustive search
2382 * across *all* the pages in each nonlinear VMA, not just the pages
2383 * whose virtual address lies outside the file truncation point.
2386 list_for_each_entry(vma
, head
, shared
.vm_set
.list
) {
2387 /* Skip quickly over those we have already dealt with */
2388 if (vma
->vm_truncate_count
== details
->truncate_count
)
2390 details
->nonlinear_vma
= vma
;
2391 if (unmap_mapping_range_vma(vma
, vma
->vm_start
,
2392 vma
->vm_end
, details
) < 0)
2398 * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
2399 * @mapping: the address space containing mmaps to be unmapped.
2400 * @holebegin: byte in first page to unmap, relative to the start of
2401 * the underlying file. This will be rounded down to a PAGE_SIZE
2402 * boundary. Note that this is different from vmtruncate(), which
2403 * must keep the partial page. In contrast, we must get rid of
2405 * @holelen: size of prospective hole in bytes. This will be rounded
2406 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2408 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2409 * but 0 when invalidating pagecache, don't throw away private data.
2411 void unmap_mapping_range(struct address_space
*mapping
,
2412 loff_t
const holebegin
, loff_t
const holelen
, int even_cows
)
2414 struct zap_details details
;
2415 pgoff_t hba
= holebegin
>> PAGE_SHIFT
;
2416 pgoff_t hlen
= (holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2418 /* Check for overflow. */
2419 if (sizeof(holelen
) > sizeof(hlen
)) {
2421 (holebegin
+ holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2422 if (holeend
& ~(long long)ULONG_MAX
)
2423 hlen
= ULONG_MAX
- hba
+ 1;
2426 details
.check_mapping
= even_cows
? NULL
: mapping
;
2427 details
.nonlinear_vma
= NULL
;
2428 details
.first_index
= hba
;
2429 details
.last_index
= hba
+ hlen
- 1;
2430 if (details
.last_index
< details
.first_index
)
2431 details
.last_index
= ULONG_MAX
;
2432 details
.i_mmap_lock
= &mapping
->i_mmap_lock
;
2434 spin_lock(&mapping
->i_mmap_lock
);
2436 /* Protect against endless unmapping loops */
2437 mapping
->truncate_count
++;
2438 if (unlikely(is_restart_addr(mapping
->truncate_count
))) {
2439 if (mapping
->truncate_count
== 0)
2440 reset_vma_truncate_counts(mapping
);
2441 mapping
->truncate_count
++;
2443 details
.truncate_count
= mapping
->truncate_count
;
2445 if (unlikely(!prio_tree_empty(&mapping
->i_mmap
)))
2446 unmap_mapping_range_tree(&mapping
->i_mmap
, &details
);
2447 if (unlikely(!list_empty(&mapping
->i_mmap_nonlinear
)))
2448 unmap_mapping_range_list(&mapping
->i_mmap_nonlinear
, &details
);
2449 spin_unlock(&mapping
->i_mmap_lock
);
2451 EXPORT_SYMBOL(unmap_mapping_range
);
2454 * vmtruncate - unmap mappings "freed" by truncate() syscall
2455 * @inode: inode of the file used
2456 * @offset: file offset to start truncating
2458 * NOTE! We have to be ready to update the memory sharing
2459 * between the file and the memory map for a potential last
2460 * incomplete page. Ugly, but necessary.
2462 int vmtruncate(struct inode
* inode
, loff_t offset
)
2464 if (inode
->i_size
< offset
) {
2465 unsigned long limit
;
2467 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
2468 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
2470 if (offset
> inode
->i_sb
->s_maxbytes
)
2472 i_size_write(inode
, offset
);
2474 struct address_space
*mapping
= inode
->i_mapping
;
2477 * truncation of in-use swapfiles is disallowed - it would
2478 * cause subsequent swapout to scribble on the now-freed
2481 if (IS_SWAPFILE(inode
))
2483 i_size_write(inode
, offset
);
2486 * unmap_mapping_range is called twice, first simply for
2487 * efficiency so that truncate_inode_pages does fewer
2488 * single-page unmaps. However after this first call, and
2489 * before truncate_inode_pages finishes, it is possible for
2490 * private pages to be COWed, which remain after
2491 * truncate_inode_pages finishes, hence the second
2492 * unmap_mapping_range call must be made for correctness.
2494 unmap_mapping_range(mapping
, offset
+ PAGE_SIZE
- 1, 0, 1);
2495 truncate_inode_pages(mapping
, offset
);
2496 unmap_mapping_range(mapping
, offset
+ PAGE_SIZE
- 1, 0, 1);
2499 if (inode
->i_op
->truncate
)
2500 inode
->i_op
->truncate(inode
);
2504 send_sig(SIGXFSZ
, current
, 0);
2508 EXPORT_SYMBOL(vmtruncate
);
2510 int vmtruncate_range(struct inode
*inode
, loff_t offset
, loff_t end
)
2512 struct address_space
*mapping
= inode
->i_mapping
;
2515 * If the underlying filesystem is not going to provide
2516 * a way to truncate a range of blocks (punch a hole) -
2517 * we should return failure right now.
2519 if (!inode
->i_op
->truncate_range
)
2522 mutex_lock(&inode
->i_mutex
);
2523 down_write(&inode
->i_alloc_sem
);
2524 unmap_mapping_range(mapping
, offset
, (end
- offset
), 1);
2525 truncate_inode_pages_range(mapping
, offset
, end
);
2526 unmap_mapping_range(mapping
, offset
, (end
- offset
), 1);
2527 inode
->i_op
->truncate_range(inode
, offset
, end
);
2528 up_write(&inode
->i_alloc_sem
);
2529 mutex_unlock(&inode
->i_mutex
);
2535 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2536 * but allow concurrent faults), and pte mapped but not yet locked.
2537 * We return with mmap_sem still held, but pte unmapped and unlocked.
2539 static int do_swap_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2540 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2541 unsigned int flags
, pte_t orig_pte
)
2547 struct mem_cgroup
*ptr
= NULL
;
2550 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
2553 entry
= pte_to_swp_entry(orig_pte
);
2554 if (is_migration_entry(entry
)) {
2555 migration_entry_wait(mm
, pmd
, address
);
2558 delayacct_set_flag(DELAYACCT_PF_SWAPIN
);
2559 page
= lookup_swap_cache(entry
);
2561 grab_swap_token(mm
); /* Contend for token _before_ read-in */
2562 page
= swapin_readahead(entry
,
2563 GFP_HIGHUSER_MOVABLE
, vma
, address
);
2566 * Back out if somebody else faulted in this pte
2567 * while we released the pte lock.
2569 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2570 if (likely(pte_same(*page_table
, orig_pte
)))
2572 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2576 /* Had to read the page from swap area: Major fault */
2577 ret
= VM_FAULT_MAJOR
;
2578 count_vm_event(PGMAJFAULT
);
2582 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2584 if (mem_cgroup_try_charge_swapin(mm
, page
, GFP_KERNEL
, &ptr
)) {
2590 * Back out if somebody else already faulted in this pte.
2592 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2593 if (unlikely(!pte_same(*page_table
, orig_pte
)))
2596 if (unlikely(!PageUptodate(page
))) {
2597 ret
= VM_FAULT_SIGBUS
;
2602 * The page isn't present yet, go ahead with the fault.
2604 * Be careful about the sequence of operations here.
2605 * To get its accounting right, reuse_swap_page() must be called
2606 * while the page is counted on swap but not yet in mapcount i.e.
2607 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2608 * must be called after the swap_free(), or it will never succeed.
2609 * Because delete_from_swap_page() may be called by reuse_swap_page(),
2610 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
2611 * in page->private. In this case, a record in swap_cgroup is silently
2612 * discarded at swap_free().
2615 inc_mm_counter(mm
, anon_rss
);
2616 pte
= mk_pte(page
, vma
->vm_page_prot
);
2617 if ((flags
& FAULT_FLAG_WRITE
) && reuse_swap_page(page
)) {
2618 pte
= maybe_mkwrite(pte_mkdirty(pte
), vma
);
2619 flags
&= ~FAULT_FLAG_WRITE
;
2621 flush_icache_page(vma
, page
);
2622 set_pte_at(mm
, address
, page_table
, pte
);
2623 page_add_anon_rmap(page
, vma
, address
);
2624 /* It's better to call commit-charge after rmap is established */
2625 mem_cgroup_commit_charge_swapin(page
, ptr
);
2628 if (vm_swap_full() || (vma
->vm_flags
& VM_LOCKED
) || PageMlocked(page
))
2629 try_to_free_swap(page
);
2632 if (flags
& FAULT_FLAG_WRITE
) {
2633 ret
|= do_wp_page(mm
, vma
, address
, page_table
, pmd
, ptl
, pte
);
2634 if (ret
& VM_FAULT_ERROR
)
2635 ret
&= VM_FAULT_ERROR
;
2639 /* No need to invalidate - it was non-present before */
2640 update_mmu_cache(vma
, address
, pte
);
2642 pte_unmap_unlock(page_table
, ptl
);
2646 mem_cgroup_cancel_charge_swapin(ptr
);
2647 pte_unmap_unlock(page_table
, ptl
);
2650 page_cache_release(page
);
2655 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2656 * but allow concurrent faults), and pte mapped but not yet locked.
2657 * We return with mmap_sem still held, but pte unmapped and unlocked.
2659 static int do_anonymous_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2660 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2667 if (HAVE_PTE_SPECIAL
&& !(flags
& FAULT_FLAG_WRITE
)) {
2668 entry
= pte_mkspecial(pfn_pte(zero_pfn
, vma
->vm_page_prot
));
2669 ptl
= pte_lockptr(mm
, pmd
);
2671 if (!pte_none(*page_table
))
2676 /* Allocate our own private page. */
2677 pte_unmap(page_table
);
2679 if (unlikely(anon_vma_prepare(vma
)))
2681 page
= alloc_zeroed_user_highpage_movable(vma
, address
);
2684 __SetPageUptodate(page
);
2686 if (mem_cgroup_newpage_charge(page
, mm
, GFP_KERNEL
))
2689 entry
= mk_pte(page
, vma
->vm_page_prot
);
2690 if (vma
->vm_flags
& VM_WRITE
)
2691 entry
= pte_mkwrite(pte_mkdirty(entry
));
2693 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2694 if (!pte_none(*page_table
))
2697 inc_mm_counter(mm
, anon_rss
);
2698 page_add_new_anon_rmap(page
, vma
, address
);
2700 set_pte_at(mm
, address
, page_table
, entry
);
2702 /* No need to invalidate - it was non-present before */
2703 update_mmu_cache(vma
, address
, entry
);
2705 pte_unmap_unlock(page_table
, ptl
);
2708 mem_cgroup_uncharge_page(page
);
2709 page_cache_release(page
);
2712 page_cache_release(page
);
2714 return VM_FAULT_OOM
;
2718 * __do_fault() tries to create a new page mapping. It aggressively
2719 * tries to share with existing pages, but makes a separate copy if
2720 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2721 * the next page fault.
2723 * As this is called only for pages that do not currently exist, we
2724 * do not need to flush old virtual caches or the TLB.
2726 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2727 * but allow concurrent faults), and pte neither mapped nor locked.
2728 * We return with mmap_sem still held, but pte unmapped and unlocked.
2730 static int __do_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2731 unsigned long address
, pmd_t
*pmd
,
2732 pgoff_t pgoff
, unsigned int flags
, pte_t orig_pte
)
2740 struct page
*dirty_page
= NULL
;
2741 struct vm_fault vmf
;
2743 int page_mkwrite
= 0;
2745 vmf
.virtual_address
= (void __user
*)(address
& PAGE_MASK
);
2750 ret
= vma
->vm_ops
->fault(vma
, &vmf
);
2751 if (unlikely(ret
& (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
)))
2755 * For consistency in subsequent calls, make the faulted page always
2758 if (unlikely(!(ret
& VM_FAULT_LOCKED
)))
2759 lock_page(vmf
.page
);
2761 VM_BUG_ON(!PageLocked(vmf
.page
));
2764 * Should we do an early C-O-W break?
2767 if (flags
& FAULT_FLAG_WRITE
) {
2768 if (!(vma
->vm_flags
& VM_SHARED
)) {
2770 if (unlikely(anon_vma_prepare(vma
))) {
2774 page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
,
2780 if (mem_cgroup_newpage_charge(page
, mm
, GFP_KERNEL
)) {
2782 page_cache_release(page
);
2787 * Don't let another task, with possibly unlocked vma,
2788 * keep the mlocked page.
2790 if (vma
->vm_flags
& VM_LOCKED
)
2791 clear_page_mlock(vmf
.page
);
2792 copy_user_highpage(page
, vmf
.page
, address
, vma
);
2793 __SetPageUptodate(page
);
2796 * If the page will be shareable, see if the backing
2797 * address space wants to know that the page is about
2798 * to become writable
2800 if (vma
->vm_ops
->page_mkwrite
) {
2804 vmf
.flags
= FAULT_FLAG_WRITE
|FAULT_FLAG_MKWRITE
;
2805 tmp
= vma
->vm_ops
->page_mkwrite(vma
, &vmf
);
2807 (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
))) {
2809 goto unwritable_page
;
2811 if (unlikely(!(tmp
& VM_FAULT_LOCKED
))) {
2813 if (!page
->mapping
) {
2814 ret
= 0; /* retry the fault */
2816 goto unwritable_page
;
2819 VM_BUG_ON(!PageLocked(page
));
2826 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2829 * This silly early PAGE_DIRTY setting removes a race
2830 * due to the bad i386 page protection. But it's valid
2831 * for other architectures too.
2833 * Note that if FAULT_FLAG_WRITE is set, we either now have
2834 * an exclusive copy of the page, or this is a shared mapping,
2835 * so we can make it writable and dirty to avoid having to
2836 * handle that later.
2838 /* Only go through if we didn't race with anybody else... */
2839 if (likely(pte_same(*page_table
, orig_pte
))) {
2840 flush_icache_page(vma
, page
);
2841 entry
= mk_pte(page
, vma
->vm_page_prot
);
2842 if (flags
& FAULT_FLAG_WRITE
)
2843 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2845 inc_mm_counter(mm
, anon_rss
);
2846 page_add_new_anon_rmap(page
, vma
, address
);
2848 inc_mm_counter(mm
, file_rss
);
2849 page_add_file_rmap(page
);
2850 if (flags
& FAULT_FLAG_WRITE
) {
2852 get_page(dirty_page
);
2855 set_pte_at(mm
, address
, page_table
, entry
);
2857 /* no need to invalidate: a not-present page won't be cached */
2858 update_mmu_cache(vma
, address
, entry
);
2861 mem_cgroup_uncharge_page(page
);
2863 page_cache_release(page
);
2865 anon
= 1; /* no anon but release faulted_page */
2868 pte_unmap_unlock(page_table
, ptl
);
2872 struct address_space
*mapping
= page
->mapping
;
2874 if (set_page_dirty(dirty_page
))
2876 unlock_page(dirty_page
);
2877 put_page(dirty_page
);
2878 if (page_mkwrite
&& mapping
) {
2880 * Some device drivers do not set page.mapping but still
2883 balance_dirty_pages_ratelimited(mapping
);
2886 /* file_update_time outside page_lock */
2888 file_update_time(vma
->vm_file
);
2890 unlock_page(vmf
.page
);
2892 page_cache_release(vmf
.page
);
2898 page_cache_release(page
);
2902 static int do_linear_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2903 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2904 unsigned int flags
, pte_t orig_pte
)
2906 pgoff_t pgoff
= (((address
& PAGE_MASK
)
2907 - vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2909 pte_unmap(page_table
);
2910 return __do_fault(mm
, vma
, address
, pmd
, pgoff
, flags
, orig_pte
);
2914 * Fault of a previously existing named mapping. Repopulate the pte
2915 * from the encoded file_pte if possible. This enables swappable
2918 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2919 * but allow concurrent faults), and pte mapped but not yet locked.
2920 * We return with mmap_sem still held, but pte unmapped and unlocked.
2922 static int do_nonlinear_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2923 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2924 unsigned int flags
, pte_t orig_pte
)
2928 flags
|= FAULT_FLAG_NONLINEAR
;
2930 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
2933 if (unlikely(!(vma
->vm_flags
& VM_NONLINEAR
))) {
2935 * Page table corrupted: show pte and kill process.
2937 print_bad_pte(vma
, address
, orig_pte
, NULL
);
2938 return VM_FAULT_OOM
;
2941 pgoff
= pte_to_pgoff(orig_pte
);
2942 return __do_fault(mm
, vma
, address
, pmd
, pgoff
, flags
, orig_pte
);
2946 * These routines also need to handle stuff like marking pages dirty
2947 * and/or accessed for architectures that don't do it in hardware (most
2948 * RISC architectures). The early dirtying is also good on the i386.
2950 * There is also a hook called "update_mmu_cache()" that architectures
2951 * with external mmu caches can use to update those (ie the Sparc or
2952 * PowerPC hashed page tables that act as extended TLBs).
2954 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2955 * but allow concurrent faults), and pte mapped but not yet locked.
2956 * We return with mmap_sem still held, but pte unmapped and unlocked.
2958 static inline int handle_pte_fault(struct mm_struct
*mm
,
2959 struct vm_area_struct
*vma
, unsigned long address
,
2960 pte_t
*pte
, pmd_t
*pmd
, unsigned int flags
)
2966 if (!pte_present(entry
)) {
2967 if (pte_none(entry
)) {
2969 if (likely(vma
->vm_ops
->fault
))
2970 return do_linear_fault(mm
, vma
, address
,
2971 pte
, pmd
, flags
, entry
);
2973 return do_anonymous_page(mm
, vma
, address
,
2976 if (pte_file(entry
))
2977 return do_nonlinear_fault(mm
, vma
, address
,
2978 pte
, pmd
, flags
, entry
);
2979 return do_swap_page(mm
, vma
, address
,
2980 pte
, pmd
, flags
, entry
);
2983 ptl
= pte_lockptr(mm
, pmd
);
2985 if (unlikely(!pte_same(*pte
, entry
)))
2987 if (flags
& FAULT_FLAG_WRITE
) {
2988 if (!pte_write(entry
))
2989 return do_wp_page(mm
, vma
, address
,
2990 pte
, pmd
, ptl
, entry
);
2991 entry
= pte_mkdirty(entry
);
2993 entry
= pte_mkyoung(entry
);
2994 if (ptep_set_access_flags(vma
, address
, pte
, entry
, flags
& FAULT_FLAG_WRITE
)) {
2995 update_mmu_cache(vma
, address
, entry
);
2998 * This is needed only for protection faults but the arch code
2999 * is not yet telling us if this is a protection fault or not.
3000 * This still avoids useless tlb flushes for .text page faults
3003 if (flags
& FAULT_FLAG_WRITE
)
3004 flush_tlb_page(vma
, address
);
3007 pte_unmap_unlock(pte
, ptl
);
3012 * By the time we get here, we already hold the mm semaphore
3014 int handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3015 unsigned long address
, unsigned int flags
)
3022 __set_current_state(TASK_RUNNING
);
3024 count_vm_event(PGFAULT
);
3026 if (unlikely(is_vm_hugetlb_page(vma
)))
3027 return hugetlb_fault(mm
, vma
, address
, flags
);
3029 pgd
= pgd_offset(mm
, address
);
3030 pud
= pud_alloc(mm
, pgd
, address
);
3032 return VM_FAULT_OOM
;
3033 pmd
= pmd_alloc(mm
, pud
, address
);
3035 return VM_FAULT_OOM
;
3036 pte
= pte_alloc_map(mm
, pmd
, address
);
3038 return VM_FAULT_OOM
;
3040 return handle_pte_fault(mm
, vma
, address
, pte
, pmd
, flags
);
3043 #ifndef __PAGETABLE_PUD_FOLDED
3045 * Allocate page upper directory.
3046 * We've already handled the fast-path in-line.
3048 int __pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
3050 pud_t
*new = pud_alloc_one(mm
, address
);
3054 smp_wmb(); /* See comment in __pte_alloc */
3056 spin_lock(&mm
->page_table_lock
);
3057 if (pgd_present(*pgd
)) /* Another has populated it */
3060 pgd_populate(mm
, pgd
, new);
3061 spin_unlock(&mm
->page_table_lock
);
3064 #endif /* __PAGETABLE_PUD_FOLDED */
3066 #ifndef __PAGETABLE_PMD_FOLDED
3068 * Allocate page middle directory.
3069 * We've already handled the fast-path in-line.
3071 int __pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
3073 pmd_t
*new = pmd_alloc_one(mm
, address
);
3077 smp_wmb(); /* See comment in __pte_alloc */
3079 spin_lock(&mm
->page_table_lock
);
3080 #ifndef __ARCH_HAS_4LEVEL_HACK
3081 if (pud_present(*pud
)) /* Another has populated it */
3084 pud_populate(mm
, pud
, new);
3086 if (pgd_present(*pud
)) /* Another has populated it */
3089 pgd_populate(mm
, pud
, new);
3090 #endif /* __ARCH_HAS_4LEVEL_HACK */
3091 spin_unlock(&mm
->page_table_lock
);
3094 #endif /* __PAGETABLE_PMD_FOLDED */
3096 int make_pages_present(unsigned long addr
, unsigned long end
)
3098 int ret
, len
, write
;
3099 struct vm_area_struct
* vma
;
3101 vma
= find_vma(current
->mm
, addr
);
3104 write
= (vma
->vm_flags
& VM_WRITE
) != 0;
3105 BUG_ON(addr
>= end
);
3106 BUG_ON(end
> vma
->vm_end
);
3107 len
= DIV_ROUND_UP(end
, PAGE_SIZE
) - addr
/PAGE_SIZE
;
3108 ret
= get_user_pages(current
, current
->mm
, addr
,
3109 len
, write
, 0, NULL
, NULL
);
3112 return ret
== len
? 0 : -EFAULT
;
3115 #if !defined(__HAVE_ARCH_GATE_AREA)
3117 #if defined(AT_SYSINFO_EHDR)
3118 static struct vm_area_struct gate_vma
;
3120 static int __init
gate_vma_init(void)
3122 gate_vma
.vm_mm
= NULL
;
3123 gate_vma
.vm_start
= FIXADDR_USER_START
;
3124 gate_vma
.vm_end
= FIXADDR_USER_END
;
3125 gate_vma
.vm_flags
= VM_READ
| VM_MAYREAD
| VM_EXEC
| VM_MAYEXEC
;
3126 gate_vma
.vm_page_prot
= __P101
;
3128 * Make sure the vDSO gets into every core dump.
3129 * Dumping its contents makes post-mortem fully interpretable later
3130 * without matching up the same kernel and hardware config to see
3131 * what PC values meant.
3133 gate_vma
.vm_flags
|= VM_ALWAYSDUMP
;
3136 __initcall(gate_vma_init
);
3139 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
3141 #ifdef AT_SYSINFO_EHDR
3148 int in_gate_area_no_task(unsigned long addr
)
3150 #ifdef AT_SYSINFO_EHDR
3151 if ((addr
>= FIXADDR_USER_START
) && (addr
< FIXADDR_USER_END
))
3157 #endif /* __HAVE_ARCH_GATE_AREA */
3159 static int follow_pte(struct mm_struct
*mm
, unsigned long address
,
3160 pte_t
**ptepp
, spinlock_t
**ptlp
)
3167 pgd
= pgd_offset(mm
, address
);
3168 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
3171 pud
= pud_offset(pgd
, address
);
3172 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
3175 pmd
= pmd_offset(pud
, address
);
3176 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
3179 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3183 ptep
= pte_offset_map_lock(mm
, pmd
, address
, ptlp
);
3186 if (!pte_present(*ptep
))
3191 pte_unmap_unlock(ptep
, *ptlp
);
3197 * follow_pfn - look up PFN at a user virtual address
3198 * @vma: memory mapping
3199 * @address: user virtual address
3200 * @pfn: location to store found PFN
3202 * Only IO mappings and raw PFN mappings are allowed.
3204 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3206 int follow_pfn(struct vm_area_struct
*vma
, unsigned long address
,
3213 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
3216 ret
= follow_pte(vma
->vm_mm
, address
, &ptep
, &ptl
);
3219 *pfn
= pte_pfn(*ptep
);
3220 pte_unmap_unlock(ptep
, ptl
);
3223 EXPORT_SYMBOL(follow_pfn
);
3225 #ifdef CONFIG_HAVE_IOREMAP_PROT
3226 int follow_phys(struct vm_area_struct
*vma
,
3227 unsigned long address
, unsigned int flags
,
3228 unsigned long *prot
, resource_size_t
*phys
)
3234 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
3237 if (follow_pte(vma
->vm_mm
, address
, &ptep
, &ptl
))
3241 if ((flags
& FOLL_WRITE
) && !pte_write(pte
))
3244 *prot
= pgprot_val(pte_pgprot(pte
));
3245 *phys
= (resource_size_t
)pte_pfn(pte
) << PAGE_SHIFT
;
3249 pte_unmap_unlock(ptep
, ptl
);
3254 int generic_access_phys(struct vm_area_struct
*vma
, unsigned long addr
,
3255 void *buf
, int len
, int write
)
3257 resource_size_t phys_addr
;
3258 unsigned long prot
= 0;
3259 void __iomem
*maddr
;
3260 int offset
= addr
& (PAGE_SIZE
-1);
3262 if (follow_phys(vma
, addr
, write
, &prot
, &phys_addr
))
3265 maddr
= ioremap_prot(phys_addr
, PAGE_SIZE
, prot
);
3267 memcpy_toio(maddr
+ offset
, buf
, len
);
3269 memcpy_fromio(buf
, maddr
+ offset
, len
);
3277 * Access another process' address space.
3278 * Source/target buffer must be kernel space,
3279 * Do not walk the page table directly, use get_user_pages
3281 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
, int write
)
3283 struct mm_struct
*mm
;
3284 struct vm_area_struct
*vma
;
3285 void *old_buf
= buf
;
3287 mm
= get_task_mm(tsk
);
3291 down_read(&mm
->mmap_sem
);
3292 /* ignore errors, just check how much was successfully transferred */
3294 int bytes
, ret
, offset
;
3296 struct page
*page
= NULL
;
3298 ret
= get_user_pages(tsk
, mm
, addr
, 1,
3299 write
, 1, &page
, &vma
);
3302 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3303 * we can access using slightly different code.
3305 #ifdef CONFIG_HAVE_IOREMAP_PROT
3306 vma
= find_vma(mm
, addr
);
3309 if (vma
->vm_ops
&& vma
->vm_ops
->access
)
3310 ret
= vma
->vm_ops
->access(vma
, addr
, buf
,
3318 offset
= addr
& (PAGE_SIZE
-1);
3319 if (bytes
> PAGE_SIZE
-offset
)
3320 bytes
= PAGE_SIZE
-offset
;
3324 copy_to_user_page(vma
, page
, addr
,
3325 maddr
+ offset
, buf
, bytes
);
3326 set_page_dirty_lock(page
);
3328 copy_from_user_page(vma
, page
, addr
,
3329 buf
, maddr
+ offset
, bytes
);
3332 page_cache_release(page
);
3338 up_read(&mm
->mmap_sem
);
3341 return buf
- old_buf
;
3345 * Print the name of a VMA.
3347 void print_vma_addr(char *prefix
, unsigned long ip
)
3349 struct mm_struct
*mm
= current
->mm
;
3350 struct vm_area_struct
*vma
;
3353 * Do not print if we are in atomic
3354 * contexts (in exception stacks, etc.):
3356 if (preempt_count())
3359 down_read(&mm
->mmap_sem
);
3360 vma
= find_vma(mm
, ip
);
3361 if (vma
&& vma
->vm_file
) {
3362 struct file
*f
= vma
->vm_file
;
3363 char *buf
= (char *)__get_free_page(GFP_KERNEL
);
3367 p
= d_path(&f
->f_path
, buf
, PAGE_SIZE
);
3370 s
= strrchr(p
, '/');
3373 printk("%s%s[%lx+%lx]", prefix
, p
,
3375 vma
->vm_end
- vma
->vm_start
);
3376 free_page((unsigned long)buf
);
3379 up_read(¤t
->mm
->mmap_sem
);
3382 #ifdef CONFIG_PROVE_LOCKING
3383 void might_fault(void)
3386 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3387 * holding the mmap_sem, this is safe because kernel memory doesn't
3388 * get paged out, therefore we'll never actually fault, and the
3389 * below annotations will generate false positives.
3391 if (segment_eq(get_fs(), KERNEL_DS
))
3396 * it would be nicer only to annotate paths which are not under
3397 * pagefault_disable, however that requires a larger audit and
3398 * providing helpers like get_user_atomic.
3400 if (!in_atomic() && current
->mm
)
3401 might_lock_read(¤t
->mm
->mmap_sem
);
3403 EXPORT_SYMBOL(might_fault
);