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/rmap.h>
49 #include <linux/module.h>
50 #include <linux/delayacct.h>
51 #include <linux/init.h>
52 #include <linux/writeback.h>
53 #include <linux/memcontrol.h>
55 #include <asm/pgalloc.h>
56 #include <asm/uaccess.h>
58 #include <asm/tlbflush.h>
59 #include <asm/pgtable.h>
61 #include <linux/swapops.h>
62 #include <linux/elf.h>
64 #ifndef CONFIG_NEED_MULTIPLE_NODES
65 /* use the per-pgdat data instead for discontigmem - mbligh */
66 unsigned long max_mapnr
;
69 EXPORT_SYMBOL(max_mapnr
);
70 EXPORT_SYMBOL(mem_map
);
73 unsigned long num_physpages
;
75 * A number of key systems in x86 including ioremap() rely on the assumption
76 * that high_memory defines the upper bound on direct map memory, then end
77 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
78 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
83 EXPORT_SYMBOL(num_physpages
);
84 EXPORT_SYMBOL(high_memory
);
87 * Randomize the address space (stacks, mmaps, brk, etc.).
89 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
90 * as ancient (libc5 based) binaries can segfault. )
92 int randomize_va_space __read_mostly
=
93 #ifdef CONFIG_COMPAT_BRK
99 static int __init
disable_randmaps(char *s
)
101 randomize_va_space
= 0;
104 __setup("norandmaps", disable_randmaps
);
108 * If a p?d_bad entry is found while walking page tables, report
109 * the error, before resetting entry to p?d_none. Usually (but
110 * very seldom) called out from the p?d_none_or_clear_bad macros.
113 void pgd_clear_bad(pgd_t
*pgd
)
119 void pud_clear_bad(pud_t
*pud
)
125 void pmd_clear_bad(pmd_t
*pmd
)
132 * Note: this doesn't free the actual pages themselves. That
133 * has been handled earlier when unmapping all the memory regions.
135 static void free_pte_range(struct mmu_gather
*tlb
, pmd_t
*pmd
)
137 pgtable_t token
= pmd_pgtable(*pmd
);
139 pte_free_tlb(tlb
, token
);
143 static inline void free_pmd_range(struct mmu_gather
*tlb
, pud_t
*pud
,
144 unsigned long addr
, unsigned long end
,
145 unsigned long floor
, unsigned long ceiling
)
152 pmd
= pmd_offset(pud
, addr
);
154 next
= pmd_addr_end(addr
, end
);
155 if (pmd_none_or_clear_bad(pmd
))
157 free_pte_range(tlb
, pmd
);
158 } while (pmd
++, addr
= next
, addr
!= end
);
168 if (end
- 1 > ceiling
- 1)
171 pmd
= pmd_offset(pud
, start
);
173 pmd_free_tlb(tlb
, pmd
);
176 static inline void free_pud_range(struct mmu_gather
*tlb
, pgd_t
*pgd
,
177 unsigned long addr
, unsigned long end
,
178 unsigned long floor
, unsigned long ceiling
)
185 pud
= pud_offset(pgd
, addr
);
187 next
= pud_addr_end(addr
, end
);
188 if (pud_none_or_clear_bad(pud
))
190 free_pmd_range(tlb
, pud
, addr
, next
, floor
, ceiling
);
191 } while (pud
++, addr
= next
, addr
!= end
);
197 ceiling
&= PGDIR_MASK
;
201 if (end
- 1 > ceiling
- 1)
204 pud
= pud_offset(pgd
, start
);
206 pud_free_tlb(tlb
, pud
);
210 * This function frees user-level page tables of a process.
212 * Must be called with pagetable lock held.
214 void free_pgd_range(struct mmu_gather
**tlb
,
215 unsigned long addr
, unsigned long end
,
216 unsigned long floor
, unsigned long ceiling
)
223 * The next few lines have given us lots of grief...
225 * Why are we testing PMD* at this top level? Because often
226 * there will be no work to do at all, and we'd prefer not to
227 * go all the way down to the bottom just to discover that.
229 * Why all these "- 1"s? Because 0 represents both the bottom
230 * of the address space and the top of it (using -1 for the
231 * top wouldn't help much: the masks would do the wrong thing).
232 * The rule is that addr 0 and floor 0 refer to the bottom of
233 * the address space, but end 0 and ceiling 0 refer to the top
234 * Comparisons need to use "end - 1" and "ceiling - 1" (though
235 * that end 0 case should be mythical).
237 * Wherever addr is brought up or ceiling brought down, we must
238 * be careful to reject "the opposite 0" before it confuses the
239 * subsequent tests. But what about where end is brought down
240 * by PMD_SIZE below? no, end can't go down to 0 there.
242 * Whereas we round start (addr) and ceiling down, by different
243 * masks at different levels, in order to test whether a table
244 * now has no other vmas using it, so can be freed, we don't
245 * bother to round floor or end up - the tests don't need that.
259 if (end
- 1 > ceiling
- 1)
265 pgd
= pgd_offset((*tlb
)->mm
, addr
);
267 next
= pgd_addr_end(addr
, end
);
268 if (pgd_none_or_clear_bad(pgd
))
270 free_pud_range(*tlb
, pgd
, addr
, next
, floor
, ceiling
);
271 } while (pgd
++, addr
= next
, addr
!= end
);
274 void free_pgtables(struct mmu_gather
**tlb
, struct vm_area_struct
*vma
,
275 unsigned long floor
, unsigned long ceiling
)
278 struct vm_area_struct
*next
= vma
->vm_next
;
279 unsigned long addr
= vma
->vm_start
;
282 * Hide vma from rmap and vmtruncate before freeing pgtables
284 anon_vma_unlink(vma
);
285 unlink_file_vma(vma
);
287 if (is_vm_hugetlb_page(vma
)) {
288 hugetlb_free_pgd_range(tlb
, addr
, vma
->vm_end
,
289 floor
, next
? next
->vm_start
: ceiling
);
292 * Optimization: gather nearby vmas into one call down
294 while (next
&& next
->vm_start
<= vma
->vm_end
+ PMD_SIZE
295 && !is_vm_hugetlb_page(next
)) {
298 anon_vma_unlink(vma
);
299 unlink_file_vma(vma
);
301 free_pgd_range(tlb
, addr
, vma
->vm_end
,
302 floor
, next
? next
->vm_start
: ceiling
);
308 int __pte_alloc(struct mm_struct
*mm
, pmd_t
*pmd
, unsigned long address
)
310 pgtable_t
new = pte_alloc_one(mm
, address
);
314 spin_lock(&mm
->page_table_lock
);
315 if (!pmd_present(*pmd
)) { /* Has another populated it ? */
317 pmd_populate(mm
, pmd
, new);
320 spin_unlock(&mm
->page_table_lock
);
326 int __pte_alloc_kernel(pmd_t
*pmd
, unsigned long address
)
328 pte_t
*new = pte_alloc_one_kernel(&init_mm
, address
);
332 spin_lock(&init_mm
.page_table_lock
);
333 if (!pmd_present(*pmd
)) { /* Has another populated it ? */
334 pmd_populate_kernel(&init_mm
, pmd
, new);
337 spin_unlock(&init_mm
.page_table_lock
);
339 pte_free_kernel(&init_mm
, new);
343 static inline void add_mm_rss(struct mm_struct
*mm
, int file_rss
, int anon_rss
)
346 add_mm_counter(mm
, file_rss
, file_rss
);
348 add_mm_counter(mm
, anon_rss
, anon_rss
);
352 * This function is called to print an error when a bad pte
353 * is found. For example, we might have a PFN-mapped pte in
354 * a region that doesn't allow it.
356 * The calling function must still handle the error.
358 void print_bad_pte(struct vm_area_struct
*vma
, pte_t pte
, unsigned long vaddr
)
360 printk(KERN_ERR
"Bad pte = %08llx, process = %s, "
361 "vm_flags = %lx, vaddr = %lx\n",
362 (long long)pte_val(pte
),
363 (vma
->vm_mm
== current
->mm
? current
->comm
: "???"),
364 vma
->vm_flags
, vaddr
);
368 static inline int is_cow_mapping(unsigned int flags
)
370 return (flags
& (VM_SHARED
| VM_MAYWRITE
)) == VM_MAYWRITE
;
374 * This function gets the "struct page" associated with a pte or returns
375 * NULL if no "struct page" is associated with the pte.
377 * A raw VM_PFNMAP mapping (ie. one that is not COWed) may not have any "struct
378 * page" backing, and even if they do, they are not refcounted. COWed pages of
379 * a VM_PFNMAP do always have a struct page, and they are normally refcounted
380 * (they are _normal_ pages).
382 * So a raw PFNMAP mapping will have each page table entry just pointing
383 * to a page frame number, and as far as the VM layer is concerned, those do
384 * not have pages associated with them - even if the PFN might point to memory
385 * that otherwise is perfectly fine and has a "struct page".
387 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
388 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
389 * set, and the vm_pgoff will point to the first PFN mapped: thus every
390 * page that is a raw mapping will always honor the rule
392 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
394 * A call to vm_normal_page() will return NULL for such a page.
396 * If the page doesn't follow the "remap_pfn_range()" rule in a VM_PFNMAP
397 * then the page has been COW'ed. A COW'ed page _does_ have a "struct page"
398 * associated with it even if it is in a VM_PFNMAP range. Calling
399 * vm_normal_page() on such a page will therefore return the "struct page".
402 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
403 * page" backing, however the difference is that _all_ pages with a struct
404 * page (that is, those where pfn_valid is true) are refcounted and considered
405 * normal pages by the VM. The disadvantage is that pages are refcounted
406 * (which can be slower and simply not an option for some PFNMAP users). The
407 * advantage is that we don't have to follow the strict linearity rule of
408 * PFNMAP mappings in order to support COWable mappings.
410 * A call to vm_normal_page() with a VM_MIXEDMAP mapping will return the
411 * associated "struct page" or NULL for memory not backed by a "struct page".
414 * All other mappings should have a valid struct page, which will be
415 * returned by a call to vm_normal_page().
417 struct page
*vm_normal_page(struct vm_area_struct
*vma
, unsigned long addr
, pte_t pte
)
419 unsigned long pfn
= pte_pfn(pte
);
421 if (unlikely(vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
))) {
422 if (vma
->vm_flags
& VM_MIXEDMAP
) {
427 unsigned long off
= (addr
-vma
->vm_start
) >> PAGE_SHIFT
;
428 if (pfn
== vma
->vm_pgoff
+ off
)
430 if (!is_cow_mapping(vma
->vm_flags
))
435 #ifdef CONFIG_DEBUG_VM
437 * Add some anal sanity checks for now. Eventually,
438 * we should just do "return pfn_to_page(pfn)", but
439 * in the meantime we check that we get a valid pfn,
440 * and that the resulting page looks ok.
442 if (unlikely(!pfn_valid(pfn
))) {
443 print_bad_pte(vma
, pte
, addr
);
449 * NOTE! We still have PageReserved() pages in the page
452 * The PAGE_ZERO() pages and various VDSO mappings can
453 * cause them to exist.
456 return pfn_to_page(pfn
);
460 * copy one vm_area from one task to the other. Assumes the page tables
461 * already present in the new task to be cleared in the whole range
462 * covered by this vma.
466 copy_one_pte(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
467 pte_t
*dst_pte
, pte_t
*src_pte
, struct vm_area_struct
*vma
,
468 unsigned long addr
, int *rss
)
470 unsigned long vm_flags
= vma
->vm_flags
;
471 pte_t pte
= *src_pte
;
474 /* pte contains position in swap or file, so copy. */
475 if (unlikely(!pte_present(pte
))) {
476 if (!pte_file(pte
)) {
477 swp_entry_t entry
= pte_to_swp_entry(pte
);
479 swap_duplicate(entry
);
480 /* make sure dst_mm is on swapoff's mmlist. */
481 if (unlikely(list_empty(&dst_mm
->mmlist
))) {
482 spin_lock(&mmlist_lock
);
483 if (list_empty(&dst_mm
->mmlist
))
484 list_add(&dst_mm
->mmlist
,
486 spin_unlock(&mmlist_lock
);
488 if (is_write_migration_entry(entry
) &&
489 is_cow_mapping(vm_flags
)) {
491 * COW mappings require pages in both parent
492 * and child to be set to read.
494 make_migration_entry_read(&entry
);
495 pte
= swp_entry_to_pte(entry
);
496 set_pte_at(src_mm
, addr
, src_pte
, pte
);
503 * If it's a COW mapping, write protect it both
504 * in the parent and the child
506 if (is_cow_mapping(vm_flags
)) {
507 ptep_set_wrprotect(src_mm
, addr
, src_pte
);
508 pte
= pte_wrprotect(pte
);
512 * If it's a shared mapping, mark it clean in
515 if (vm_flags
& VM_SHARED
)
516 pte
= pte_mkclean(pte
);
517 pte
= pte_mkold(pte
);
519 page
= vm_normal_page(vma
, addr
, pte
);
522 page_dup_rmap(page
, vma
, addr
);
523 rss
[!!PageAnon(page
)]++;
527 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
530 static int copy_pte_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
531 pmd_t
*dst_pmd
, pmd_t
*src_pmd
, struct vm_area_struct
*vma
,
532 unsigned long addr
, unsigned long end
)
534 pte_t
*src_pte
, *dst_pte
;
535 spinlock_t
*src_ptl
, *dst_ptl
;
541 dst_pte
= pte_alloc_map_lock(dst_mm
, dst_pmd
, addr
, &dst_ptl
);
544 src_pte
= pte_offset_map_nested(src_pmd
, addr
);
545 src_ptl
= pte_lockptr(src_mm
, src_pmd
);
546 spin_lock_nested(src_ptl
, SINGLE_DEPTH_NESTING
);
547 arch_enter_lazy_mmu_mode();
551 * We are holding two locks at this point - either of them
552 * could generate latencies in another task on another CPU.
554 if (progress
>= 32) {
556 if (need_resched() ||
557 spin_needbreak(src_ptl
) || spin_needbreak(dst_ptl
))
560 if (pte_none(*src_pte
)) {
564 copy_one_pte(dst_mm
, src_mm
, dst_pte
, src_pte
, vma
, addr
, rss
);
566 } while (dst_pte
++, src_pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
568 arch_leave_lazy_mmu_mode();
569 spin_unlock(src_ptl
);
570 pte_unmap_nested(src_pte
- 1);
571 add_mm_rss(dst_mm
, rss
[0], rss
[1]);
572 pte_unmap_unlock(dst_pte
- 1, dst_ptl
);
579 static inline int copy_pmd_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
580 pud_t
*dst_pud
, pud_t
*src_pud
, struct vm_area_struct
*vma
,
581 unsigned long addr
, unsigned long end
)
583 pmd_t
*src_pmd
, *dst_pmd
;
586 dst_pmd
= pmd_alloc(dst_mm
, dst_pud
, addr
);
589 src_pmd
= pmd_offset(src_pud
, addr
);
591 next
= pmd_addr_end(addr
, end
);
592 if (pmd_none_or_clear_bad(src_pmd
))
594 if (copy_pte_range(dst_mm
, src_mm
, dst_pmd
, src_pmd
,
597 } while (dst_pmd
++, src_pmd
++, addr
= next
, addr
!= end
);
601 static inline int copy_pud_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
602 pgd_t
*dst_pgd
, pgd_t
*src_pgd
, struct vm_area_struct
*vma
,
603 unsigned long addr
, unsigned long end
)
605 pud_t
*src_pud
, *dst_pud
;
608 dst_pud
= pud_alloc(dst_mm
, dst_pgd
, addr
);
611 src_pud
= pud_offset(src_pgd
, addr
);
613 next
= pud_addr_end(addr
, end
);
614 if (pud_none_or_clear_bad(src_pud
))
616 if (copy_pmd_range(dst_mm
, src_mm
, dst_pud
, src_pud
,
619 } while (dst_pud
++, src_pud
++, addr
= next
, addr
!= end
);
623 int copy_page_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
624 struct vm_area_struct
*vma
)
626 pgd_t
*src_pgd
, *dst_pgd
;
628 unsigned long addr
= vma
->vm_start
;
629 unsigned long end
= vma
->vm_end
;
632 * Don't copy ptes where a page fault will fill them correctly.
633 * Fork becomes much lighter when there are big shared or private
634 * readonly mappings. The tradeoff is that copy_page_range is more
635 * efficient than faulting.
637 if (!(vma
->vm_flags
& (VM_HUGETLB
|VM_NONLINEAR
|VM_PFNMAP
|VM_INSERTPAGE
))) {
642 if (is_vm_hugetlb_page(vma
))
643 return copy_hugetlb_page_range(dst_mm
, src_mm
, vma
);
645 dst_pgd
= pgd_offset(dst_mm
, addr
);
646 src_pgd
= pgd_offset(src_mm
, addr
);
648 next
= pgd_addr_end(addr
, end
);
649 if (pgd_none_or_clear_bad(src_pgd
))
651 if (copy_pud_range(dst_mm
, src_mm
, dst_pgd
, src_pgd
,
654 } while (dst_pgd
++, src_pgd
++, addr
= next
, addr
!= end
);
658 static unsigned long zap_pte_range(struct mmu_gather
*tlb
,
659 struct vm_area_struct
*vma
, pmd_t
*pmd
,
660 unsigned long addr
, unsigned long end
,
661 long *zap_work
, struct zap_details
*details
)
663 struct mm_struct
*mm
= tlb
->mm
;
669 pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
670 arch_enter_lazy_mmu_mode();
673 if (pte_none(ptent
)) {
678 (*zap_work
) -= PAGE_SIZE
;
680 if (pte_present(ptent
)) {
683 page
= vm_normal_page(vma
, addr
, ptent
);
684 if (unlikely(details
) && page
) {
686 * unmap_shared_mapping_pages() wants to
687 * invalidate cache without truncating:
688 * unmap shared but keep private pages.
690 if (details
->check_mapping
&&
691 details
->check_mapping
!= page
->mapping
)
694 * Each page->index must be checked when
695 * invalidating or truncating nonlinear.
697 if (details
->nonlinear_vma
&&
698 (page
->index
< details
->first_index
||
699 page
->index
> details
->last_index
))
702 ptent
= ptep_get_and_clear_full(mm
, addr
, pte
,
704 tlb_remove_tlb_entry(tlb
, pte
, addr
);
707 if (unlikely(details
) && details
->nonlinear_vma
708 && linear_page_index(details
->nonlinear_vma
,
709 addr
) != page
->index
)
710 set_pte_at(mm
, addr
, pte
,
711 pgoff_to_pte(page
->index
));
715 if (pte_dirty(ptent
))
716 set_page_dirty(page
);
717 if (pte_young(ptent
))
718 SetPageReferenced(page
);
721 page_remove_rmap(page
, vma
);
722 tlb_remove_page(tlb
, page
);
726 * If details->check_mapping, we leave swap entries;
727 * if details->nonlinear_vma, we leave file entries.
729 if (unlikely(details
))
731 if (!pte_file(ptent
))
732 free_swap_and_cache(pte_to_swp_entry(ptent
));
733 pte_clear_not_present_full(mm
, addr
, pte
, tlb
->fullmm
);
734 } while (pte
++, addr
+= PAGE_SIZE
, (addr
!= end
&& *zap_work
> 0));
736 add_mm_rss(mm
, file_rss
, anon_rss
);
737 arch_leave_lazy_mmu_mode();
738 pte_unmap_unlock(pte
- 1, ptl
);
743 static inline unsigned long zap_pmd_range(struct mmu_gather
*tlb
,
744 struct vm_area_struct
*vma
, pud_t
*pud
,
745 unsigned long addr
, unsigned long end
,
746 long *zap_work
, struct zap_details
*details
)
751 pmd
= pmd_offset(pud
, addr
);
753 next
= pmd_addr_end(addr
, end
);
754 if (pmd_none_or_clear_bad(pmd
)) {
758 next
= zap_pte_range(tlb
, vma
, pmd
, addr
, next
,
760 } while (pmd
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
765 static inline unsigned long zap_pud_range(struct mmu_gather
*tlb
,
766 struct vm_area_struct
*vma
, pgd_t
*pgd
,
767 unsigned long addr
, unsigned long end
,
768 long *zap_work
, struct zap_details
*details
)
773 pud
= pud_offset(pgd
, addr
);
775 next
= pud_addr_end(addr
, end
);
776 if (pud_none_or_clear_bad(pud
)) {
780 next
= zap_pmd_range(tlb
, vma
, pud
, addr
, next
,
782 } while (pud
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
787 static unsigned long unmap_page_range(struct mmu_gather
*tlb
,
788 struct vm_area_struct
*vma
,
789 unsigned long addr
, unsigned long end
,
790 long *zap_work
, struct zap_details
*details
)
795 if (details
&& !details
->check_mapping
&& !details
->nonlinear_vma
)
799 tlb_start_vma(tlb
, vma
);
800 pgd
= pgd_offset(vma
->vm_mm
, addr
);
802 next
= pgd_addr_end(addr
, end
);
803 if (pgd_none_or_clear_bad(pgd
)) {
807 next
= zap_pud_range(tlb
, vma
, pgd
, addr
, next
,
809 } while (pgd
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
810 tlb_end_vma(tlb
, vma
);
815 #ifdef CONFIG_PREEMPT
816 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
818 /* No preempt: go for improved straight-line efficiency */
819 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
823 * unmap_vmas - unmap a range of memory covered by a list of vma's
824 * @tlbp: address of the caller's struct mmu_gather
825 * @vma: the starting vma
826 * @start_addr: virtual address at which to start unmapping
827 * @end_addr: virtual address at which to end unmapping
828 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
829 * @details: details of nonlinear truncation or shared cache invalidation
831 * Returns the end address of the unmapping (restart addr if interrupted).
833 * Unmap all pages in the vma list.
835 * We aim to not hold locks for too long (for scheduling latency reasons).
836 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
837 * return the ending mmu_gather to the caller.
839 * Only addresses between `start' and `end' will be unmapped.
841 * The VMA list must be sorted in ascending virtual address order.
843 * unmap_vmas() assumes that the caller will flush the whole unmapped address
844 * range after unmap_vmas() returns. So the only responsibility here is to
845 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
846 * drops the lock and schedules.
848 unsigned long unmap_vmas(struct mmu_gather
**tlbp
,
849 struct vm_area_struct
*vma
, unsigned long start_addr
,
850 unsigned long end_addr
, unsigned long *nr_accounted
,
851 struct zap_details
*details
)
853 long zap_work
= ZAP_BLOCK_SIZE
;
854 unsigned long tlb_start
= 0; /* For tlb_finish_mmu */
855 int tlb_start_valid
= 0;
856 unsigned long start
= start_addr
;
857 spinlock_t
*i_mmap_lock
= details
? details
->i_mmap_lock
: NULL
;
858 int fullmm
= (*tlbp
)->fullmm
;
860 for ( ; vma
&& vma
->vm_start
< end_addr
; vma
= vma
->vm_next
) {
863 start
= max(vma
->vm_start
, start_addr
);
864 if (start
>= vma
->vm_end
)
866 end
= min(vma
->vm_end
, end_addr
);
867 if (end
<= vma
->vm_start
)
870 if (vma
->vm_flags
& VM_ACCOUNT
)
871 *nr_accounted
+= (end
- start
) >> PAGE_SHIFT
;
873 while (start
!= end
) {
874 if (!tlb_start_valid
) {
879 if (unlikely(is_vm_hugetlb_page(vma
))) {
880 unmap_hugepage_range(vma
, start
, end
);
881 zap_work
-= (end
- start
) /
882 (HPAGE_SIZE
/ PAGE_SIZE
);
885 start
= unmap_page_range(*tlbp
, vma
,
886 start
, end
, &zap_work
, details
);
889 BUG_ON(start
!= end
);
893 tlb_finish_mmu(*tlbp
, tlb_start
, start
);
895 if (need_resched() ||
896 (i_mmap_lock
&& spin_needbreak(i_mmap_lock
))) {
904 *tlbp
= tlb_gather_mmu(vma
->vm_mm
, fullmm
);
906 zap_work
= ZAP_BLOCK_SIZE
;
910 return start
; /* which is now the end (or restart) address */
914 * zap_page_range - remove user pages in a given range
915 * @vma: vm_area_struct holding the applicable pages
916 * @address: starting address of pages to zap
917 * @size: number of bytes to zap
918 * @details: details of nonlinear truncation or shared cache invalidation
920 unsigned long zap_page_range(struct vm_area_struct
*vma
, unsigned long address
,
921 unsigned long size
, struct zap_details
*details
)
923 struct mm_struct
*mm
= vma
->vm_mm
;
924 struct mmu_gather
*tlb
;
925 unsigned long end
= address
+ size
;
926 unsigned long nr_accounted
= 0;
929 tlb
= tlb_gather_mmu(mm
, 0);
930 update_hiwater_rss(mm
);
931 end
= unmap_vmas(&tlb
, vma
, address
, end
, &nr_accounted
, details
);
933 tlb_finish_mmu(tlb
, address
, end
);
938 * Do a quick page-table lookup for a single page.
940 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
949 struct mm_struct
*mm
= vma
->vm_mm
;
951 page
= follow_huge_addr(mm
, address
, flags
& FOLL_WRITE
);
953 BUG_ON(flags
& FOLL_GET
);
958 pgd
= pgd_offset(mm
, address
);
959 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
962 pud
= pud_offset(pgd
, address
);
963 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
966 pmd
= pmd_offset(pud
, address
);
967 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
970 if (pmd_huge(*pmd
)) {
971 BUG_ON(flags
& FOLL_GET
);
972 page
= follow_huge_pmd(mm
, address
, pmd
, flags
& FOLL_WRITE
);
976 ptep
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
981 if (!pte_present(pte
))
983 if ((flags
& FOLL_WRITE
) && !pte_write(pte
))
985 page
= vm_normal_page(vma
, address
, pte
);
989 if (flags
& FOLL_GET
)
991 if (flags
& FOLL_TOUCH
) {
992 if ((flags
& FOLL_WRITE
) &&
993 !pte_dirty(pte
) && !PageDirty(page
))
994 set_page_dirty(page
);
995 mark_page_accessed(page
);
998 pte_unmap_unlock(ptep
, ptl
);
1004 * When core dumping an enormous anonymous area that nobody
1005 * has touched so far, we don't want to allocate page tables.
1007 if (flags
& FOLL_ANON
) {
1008 page
= ZERO_PAGE(0);
1009 if (flags
& FOLL_GET
)
1011 BUG_ON(flags
& FOLL_WRITE
);
1016 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
1017 unsigned long start
, int len
, int write
, int force
,
1018 struct page
**pages
, struct vm_area_struct
**vmas
)
1021 unsigned int vm_flags
;
1026 * Require read or write permissions.
1027 * If 'force' is set, we only require the "MAY" flags.
1029 vm_flags
= write
? (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
1030 vm_flags
&= force
? (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
1034 struct vm_area_struct
*vma
;
1035 unsigned int foll_flags
;
1037 vma
= find_extend_vma(mm
, start
);
1038 if (!vma
&& in_gate_area(tsk
, start
)) {
1039 unsigned long pg
= start
& PAGE_MASK
;
1040 struct vm_area_struct
*gate_vma
= get_gate_vma(tsk
);
1045 if (write
) /* user gate pages are read-only */
1046 return i
? : -EFAULT
;
1048 pgd
= pgd_offset_k(pg
);
1050 pgd
= pgd_offset_gate(mm
, pg
);
1051 BUG_ON(pgd_none(*pgd
));
1052 pud
= pud_offset(pgd
, pg
);
1053 BUG_ON(pud_none(*pud
));
1054 pmd
= pmd_offset(pud
, pg
);
1056 return i
? : -EFAULT
;
1057 pte
= pte_offset_map(pmd
, pg
);
1058 if (pte_none(*pte
)) {
1060 return i
? : -EFAULT
;
1063 struct page
*page
= vm_normal_page(gate_vma
, start
, *pte
);
1077 if (!vma
|| (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
))
1078 || !(vm_flags
& vma
->vm_flags
))
1079 return i
? : -EFAULT
;
1081 if (is_vm_hugetlb_page(vma
)) {
1082 i
= follow_hugetlb_page(mm
, vma
, pages
, vmas
,
1083 &start
, &len
, i
, write
);
1087 foll_flags
= FOLL_TOUCH
;
1089 foll_flags
|= FOLL_GET
;
1090 if (!write
&& !(vma
->vm_flags
& VM_LOCKED
) &&
1091 (!vma
->vm_ops
|| !vma
->vm_ops
->fault
))
1092 foll_flags
|= FOLL_ANON
;
1098 * If tsk is ooming, cut off its access to large memory
1099 * allocations. It has a pending SIGKILL, but it can't
1100 * be processed until returning to user space.
1102 if (unlikely(test_tsk_thread_flag(tsk
, TIF_MEMDIE
)))
1106 foll_flags
|= FOLL_WRITE
;
1109 while (!(page
= follow_page(vma
, start
, foll_flags
))) {
1111 ret
= handle_mm_fault(mm
, vma
, start
,
1112 foll_flags
& FOLL_WRITE
);
1113 if (ret
& VM_FAULT_ERROR
) {
1114 if (ret
& VM_FAULT_OOM
)
1115 return i
? i
: -ENOMEM
;
1116 else if (ret
& VM_FAULT_SIGBUS
)
1117 return i
? i
: -EFAULT
;
1120 if (ret
& VM_FAULT_MAJOR
)
1126 * The VM_FAULT_WRITE bit tells us that
1127 * do_wp_page has broken COW when necessary,
1128 * even if maybe_mkwrite decided not to set
1129 * pte_write. We can thus safely do subsequent
1130 * page lookups as if they were reads.
1132 if (ret
& VM_FAULT_WRITE
)
1133 foll_flags
&= ~FOLL_WRITE
;
1140 flush_anon_page(vma
, page
, start
);
1141 flush_dcache_page(page
);
1148 } while (len
&& start
< vma
->vm_end
);
1152 EXPORT_SYMBOL(get_user_pages
);
1154 pte_t
*get_locked_pte(struct mm_struct
*mm
, unsigned long addr
,
1157 pgd_t
* pgd
= pgd_offset(mm
, addr
);
1158 pud_t
* pud
= pud_alloc(mm
, pgd
, addr
);
1160 pmd_t
* pmd
= pmd_alloc(mm
, pud
, addr
);
1162 return pte_alloc_map_lock(mm
, pmd
, addr
, ptl
);
1168 * This is the old fallback for page remapping.
1170 * For historical reasons, it only allows reserved pages. Only
1171 * old drivers should use this, and they needed to mark their
1172 * pages reserved for the old functions anyway.
1174 static int insert_page(struct mm_struct
*mm
, unsigned long addr
, struct page
*page
, pgprot_t prot
)
1180 retval
= mem_cgroup_charge(page
, mm
, GFP_KERNEL
);
1188 flush_dcache_page(page
);
1189 pte
= get_locked_pte(mm
, addr
, &ptl
);
1193 if (!pte_none(*pte
))
1196 /* Ok, finally just insert the thing.. */
1198 inc_mm_counter(mm
, file_rss
);
1199 page_add_file_rmap(page
);
1200 set_pte_at(mm
, addr
, pte
, mk_pte(page
, prot
));
1203 pte_unmap_unlock(pte
, ptl
);
1206 pte_unmap_unlock(pte
, ptl
);
1208 mem_cgroup_uncharge_page(page
);
1214 * vm_insert_page - insert single page into user vma
1215 * @vma: user vma to map to
1216 * @addr: target user address of this page
1217 * @page: source kernel page
1219 * This allows drivers to insert individual pages they've allocated
1222 * The page has to be a nice clean _individual_ kernel allocation.
1223 * If you allocate a compound page, you need to have marked it as
1224 * such (__GFP_COMP), or manually just split the page up yourself
1225 * (see split_page()).
1227 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1228 * took an arbitrary page protection parameter. This doesn't allow
1229 * that. Your vma protection will have to be set up correctly, which
1230 * means that if you want a shared writable mapping, you'd better
1231 * ask for a shared writable mapping!
1233 * The page does not need to be reserved.
1235 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
, struct page
*page
)
1237 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
1239 if (!page_count(page
))
1241 vma
->vm_flags
|= VM_INSERTPAGE
;
1242 return insert_page(vma
->vm_mm
, addr
, page
, vma
->vm_page_prot
);
1244 EXPORT_SYMBOL(vm_insert_page
);
1247 * vm_insert_pfn - insert single pfn into user vma
1248 * @vma: user vma to map to
1249 * @addr: target user address of this page
1250 * @pfn: source kernel pfn
1252 * Similar to vm_inert_page, this allows drivers to insert individual pages
1253 * they've allocated into a user vma. Same comments apply.
1255 * This function should only be called from a vm_ops->fault handler, and
1256 * in that case the handler should return NULL.
1258 int vm_insert_pfn(struct vm_area_struct
*vma
, unsigned long addr
,
1261 struct mm_struct
*mm
= vma
->vm_mm
;
1266 BUG_ON(!(vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
)));
1267 BUG_ON((vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
)) ==
1268 (VM_PFNMAP
|VM_MIXEDMAP
));
1269 BUG_ON((vma
->vm_flags
& VM_PFNMAP
) && is_cow_mapping(vma
->vm_flags
));
1270 BUG_ON((vma
->vm_flags
& VM_MIXEDMAP
) && pfn_valid(pfn
));
1273 pte
= get_locked_pte(mm
, addr
, &ptl
);
1277 if (!pte_none(*pte
))
1280 /* Ok, finally just insert the thing.. */
1281 entry
= pfn_pte(pfn
, vma
->vm_page_prot
);
1282 set_pte_at(mm
, addr
, pte
, entry
);
1283 update_mmu_cache(vma
, addr
, entry
);
1287 pte_unmap_unlock(pte
, ptl
);
1292 EXPORT_SYMBOL(vm_insert_pfn
);
1295 * maps a range of physical memory into the requested pages. the old
1296 * mappings are removed. any references to nonexistent pages results
1297 * in null mappings (currently treated as "copy-on-access")
1299 static int remap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1300 unsigned long addr
, unsigned long end
,
1301 unsigned long pfn
, pgprot_t prot
)
1306 pte
= pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1309 arch_enter_lazy_mmu_mode();
1311 BUG_ON(!pte_none(*pte
));
1312 set_pte_at(mm
, addr
, pte
, pfn_pte(pfn
, prot
));
1314 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1315 arch_leave_lazy_mmu_mode();
1316 pte_unmap_unlock(pte
- 1, ptl
);
1320 static inline int remap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1321 unsigned long addr
, unsigned long end
,
1322 unsigned long pfn
, pgprot_t prot
)
1327 pfn
-= addr
>> PAGE_SHIFT
;
1328 pmd
= pmd_alloc(mm
, pud
, addr
);
1332 next
= pmd_addr_end(addr
, end
);
1333 if (remap_pte_range(mm
, pmd
, addr
, next
,
1334 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1336 } while (pmd
++, addr
= next
, addr
!= end
);
1340 static inline int remap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1341 unsigned long addr
, unsigned long end
,
1342 unsigned long pfn
, pgprot_t prot
)
1347 pfn
-= addr
>> PAGE_SHIFT
;
1348 pud
= pud_alloc(mm
, pgd
, addr
);
1352 next
= pud_addr_end(addr
, end
);
1353 if (remap_pmd_range(mm
, pud
, addr
, next
,
1354 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1356 } while (pud
++, addr
= next
, addr
!= end
);
1361 * remap_pfn_range - remap kernel memory to userspace
1362 * @vma: user vma to map to
1363 * @addr: target user address to start at
1364 * @pfn: physical address of kernel memory
1365 * @size: size of map area
1366 * @prot: page protection flags for this mapping
1368 * Note: this is only safe if the mm semaphore is held when called.
1370 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1371 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1375 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1376 struct mm_struct
*mm
= vma
->vm_mm
;
1380 * Physically remapped pages are special. Tell the
1381 * rest of the world about it:
1382 * VM_IO tells people not to look at these pages
1383 * (accesses can have side effects).
1384 * VM_RESERVED is specified all over the place, because
1385 * in 2.4 it kept swapout's vma scan off this vma; but
1386 * in 2.6 the LRU scan won't even find its pages, so this
1387 * flag means no more than count its pages in reserved_vm,
1388 * and omit it from core dump, even when VM_IO turned off.
1389 * VM_PFNMAP tells the core MM that the base pages are just
1390 * raw PFN mappings, and do not have a "struct page" associated
1393 * There's a horrible special case to handle copy-on-write
1394 * behaviour that some programs depend on. We mark the "original"
1395 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1397 if (is_cow_mapping(vma
->vm_flags
)) {
1398 if (addr
!= vma
->vm_start
|| end
!= vma
->vm_end
)
1400 vma
->vm_pgoff
= pfn
;
1403 vma
->vm_flags
|= VM_IO
| VM_RESERVED
| VM_PFNMAP
;
1405 BUG_ON(addr
>= end
);
1406 pfn
-= addr
>> PAGE_SHIFT
;
1407 pgd
= pgd_offset(mm
, addr
);
1408 flush_cache_range(vma
, addr
, end
);
1410 next
= pgd_addr_end(addr
, end
);
1411 err
= remap_pud_range(mm
, pgd
, addr
, next
,
1412 pfn
+ (addr
>> PAGE_SHIFT
), prot
);
1415 } while (pgd
++, addr
= next
, addr
!= end
);
1418 EXPORT_SYMBOL(remap_pfn_range
);
1420 static int apply_to_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1421 unsigned long addr
, unsigned long end
,
1422 pte_fn_t fn
, void *data
)
1427 spinlock_t
*uninitialized_var(ptl
);
1429 pte
= (mm
== &init_mm
) ?
1430 pte_alloc_kernel(pmd
, addr
) :
1431 pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1435 BUG_ON(pmd_huge(*pmd
));
1437 token
= pmd_pgtable(*pmd
);
1440 err
= fn(pte
, token
, addr
, data
);
1443 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1446 pte_unmap_unlock(pte
-1, ptl
);
1450 static int apply_to_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1451 unsigned long addr
, unsigned long end
,
1452 pte_fn_t fn
, void *data
)
1458 pmd
= pmd_alloc(mm
, pud
, addr
);
1462 next
= pmd_addr_end(addr
, end
);
1463 err
= apply_to_pte_range(mm
, pmd
, addr
, next
, fn
, data
);
1466 } while (pmd
++, addr
= next
, addr
!= end
);
1470 static int apply_to_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1471 unsigned long addr
, unsigned long end
,
1472 pte_fn_t fn
, void *data
)
1478 pud
= pud_alloc(mm
, pgd
, addr
);
1482 next
= pud_addr_end(addr
, end
);
1483 err
= apply_to_pmd_range(mm
, pud
, addr
, next
, fn
, data
);
1486 } while (pud
++, addr
= next
, addr
!= end
);
1491 * Scan a region of virtual memory, filling in page tables as necessary
1492 * and calling a provided function on each leaf page table.
1494 int apply_to_page_range(struct mm_struct
*mm
, unsigned long addr
,
1495 unsigned long size
, pte_fn_t fn
, void *data
)
1499 unsigned long end
= addr
+ size
;
1502 BUG_ON(addr
>= end
);
1503 pgd
= pgd_offset(mm
, addr
);
1505 next
= pgd_addr_end(addr
, end
);
1506 err
= apply_to_pud_range(mm
, pgd
, addr
, next
, fn
, data
);
1509 } while (pgd
++, addr
= next
, addr
!= end
);
1512 EXPORT_SYMBOL_GPL(apply_to_page_range
);
1515 * handle_pte_fault chooses page fault handler according to an entry
1516 * which was read non-atomically. Before making any commitment, on
1517 * those architectures or configurations (e.g. i386 with PAE) which
1518 * might give a mix of unmatched parts, do_swap_page and do_file_page
1519 * must check under lock before unmapping the pte and proceeding
1520 * (but do_wp_page is only called after already making such a check;
1521 * and do_anonymous_page and do_no_page can safely check later on).
1523 static inline int pte_unmap_same(struct mm_struct
*mm
, pmd_t
*pmd
,
1524 pte_t
*page_table
, pte_t orig_pte
)
1527 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1528 if (sizeof(pte_t
) > sizeof(unsigned long)) {
1529 spinlock_t
*ptl
= pte_lockptr(mm
, pmd
);
1531 same
= pte_same(*page_table
, orig_pte
);
1535 pte_unmap(page_table
);
1540 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1541 * servicing faults for write access. In the normal case, do always want
1542 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1543 * that do not have writing enabled, when used by access_process_vm.
1545 static inline pte_t
maybe_mkwrite(pte_t pte
, struct vm_area_struct
*vma
)
1547 if (likely(vma
->vm_flags
& VM_WRITE
))
1548 pte
= pte_mkwrite(pte
);
1552 static inline void cow_user_page(struct page
*dst
, struct page
*src
, unsigned long va
, struct vm_area_struct
*vma
)
1555 * If the source page was a PFN mapping, we don't have
1556 * a "struct page" for it. We do a best-effort copy by
1557 * just copying from the original user address. If that
1558 * fails, we just zero-fill it. Live with it.
1560 if (unlikely(!src
)) {
1561 void *kaddr
= kmap_atomic(dst
, KM_USER0
);
1562 void __user
*uaddr
= (void __user
*)(va
& PAGE_MASK
);
1565 * This really shouldn't fail, because the page is there
1566 * in the page tables. But it might just be unreadable,
1567 * in which case we just give up and fill the result with
1570 if (__copy_from_user_inatomic(kaddr
, uaddr
, PAGE_SIZE
))
1571 memset(kaddr
, 0, PAGE_SIZE
);
1572 kunmap_atomic(kaddr
, KM_USER0
);
1573 flush_dcache_page(dst
);
1575 copy_user_highpage(dst
, src
, va
, vma
);
1579 * This routine handles present pages, when users try to write
1580 * to a shared page. It is done by copying the page to a new address
1581 * and decrementing the shared-page counter for the old page.
1583 * Note that this routine assumes that the protection checks have been
1584 * done by the caller (the low-level page fault routine in most cases).
1585 * Thus we can safely just mark it writable once we've done any necessary
1588 * We also mark the page dirty at this point even though the page will
1589 * change only once the write actually happens. This avoids a few races,
1590 * and potentially makes it more efficient.
1592 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1593 * but allow concurrent faults), with pte both mapped and locked.
1594 * We return with mmap_sem still held, but pte unmapped and unlocked.
1596 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1597 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1598 spinlock_t
*ptl
, pte_t orig_pte
)
1600 struct page
*old_page
, *new_page
;
1602 int reuse
= 0, ret
= 0;
1603 int page_mkwrite
= 0;
1604 struct page
*dirty_page
= NULL
;
1606 old_page
= vm_normal_page(vma
, address
, orig_pte
);
1611 * Take out anonymous pages first, anonymous shared vmas are
1612 * not dirty accountable.
1614 if (PageAnon(old_page
)) {
1615 if (!TestSetPageLocked(old_page
)) {
1616 reuse
= can_share_swap_page(old_page
);
1617 unlock_page(old_page
);
1619 } else if (unlikely((vma
->vm_flags
& (VM_WRITE
|VM_SHARED
)) ==
1620 (VM_WRITE
|VM_SHARED
))) {
1622 * Only catch write-faults on shared writable pages,
1623 * read-only shared pages can get COWed by
1624 * get_user_pages(.write=1, .force=1).
1626 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
) {
1628 * Notify the address space that the page is about to
1629 * become writable so that it can prohibit this or wait
1630 * for the page to get into an appropriate state.
1632 * We do this without the lock held, so that it can
1633 * sleep if it needs to.
1635 page_cache_get(old_page
);
1636 pte_unmap_unlock(page_table
, ptl
);
1638 if (vma
->vm_ops
->page_mkwrite(vma
, old_page
) < 0)
1639 goto unwritable_page
;
1642 * Since we dropped the lock we need to revalidate
1643 * the PTE as someone else may have changed it. If
1644 * they did, we just return, as we can count on the
1645 * MMU to tell us if they didn't also make it writable.
1647 page_table
= pte_offset_map_lock(mm
, pmd
, address
,
1649 page_cache_release(old_page
);
1650 if (!pte_same(*page_table
, orig_pte
))
1655 dirty_page
= old_page
;
1656 get_page(dirty_page
);
1661 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
1662 entry
= pte_mkyoung(orig_pte
);
1663 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1664 if (ptep_set_access_flags(vma
, address
, page_table
, entry
,1))
1665 update_mmu_cache(vma
, address
, entry
);
1666 ret
|= VM_FAULT_WRITE
;
1671 * Ok, we need to copy. Oh, well..
1673 page_cache_get(old_page
);
1675 pte_unmap_unlock(page_table
, ptl
);
1677 if (unlikely(anon_vma_prepare(vma
)))
1679 VM_BUG_ON(old_page
== ZERO_PAGE(0));
1680 new_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, address
);
1683 cow_user_page(new_page
, old_page
, address
, vma
);
1684 __SetPageUptodate(new_page
);
1686 if (mem_cgroup_charge(new_page
, mm
, GFP_KERNEL
))
1690 * Re-check the pte - we dropped the lock
1692 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1693 if (likely(pte_same(*page_table
, orig_pte
))) {
1695 page_remove_rmap(old_page
, vma
);
1696 if (!PageAnon(old_page
)) {
1697 dec_mm_counter(mm
, file_rss
);
1698 inc_mm_counter(mm
, anon_rss
);
1701 inc_mm_counter(mm
, anon_rss
);
1702 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
1703 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
1704 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1706 * Clear the pte entry and flush it first, before updating the
1707 * pte with the new entry. This will avoid a race condition
1708 * seen in the presence of one thread doing SMC and another
1711 ptep_clear_flush(vma
, address
, page_table
);
1712 set_pte_at(mm
, address
, page_table
, entry
);
1713 update_mmu_cache(vma
, address
, entry
);
1714 lru_cache_add_active(new_page
);
1715 page_add_new_anon_rmap(new_page
, vma
, address
);
1717 /* Free the old page.. */
1718 new_page
= old_page
;
1719 ret
|= VM_FAULT_WRITE
;
1721 mem_cgroup_uncharge_page(new_page
);
1724 page_cache_release(new_page
);
1726 page_cache_release(old_page
);
1728 pte_unmap_unlock(page_table
, ptl
);
1731 file_update_time(vma
->vm_file
);
1734 * Yes, Virginia, this is actually required to prevent a race
1735 * with clear_page_dirty_for_io() from clearing the page dirty
1736 * bit after it clear all dirty ptes, but before a racing
1737 * do_wp_page installs a dirty pte.
1739 * do_no_page is protected similarly.
1741 wait_on_page_locked(dirty_page
);
1742 set_page_dirty_balance(dirty_page
, page_mkwrite
);
1743 put_page(dirty_page
);
1747 page_cache_release(new_page
);
1750 page_cache_release(old_page
);
1751 return VM_FAULT_OOM
;
1754 page_cache_release(old_page
);
1755 return VM_FAULT_SIGBUS
;
1759 * Helper functions for unmap_mapping_range().
1761 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1763 * We have to restart searching the prio_tree whenever we drop the lock,
1764 * since the iterator is only valid while the lock is held, and anyway
1765 * a later vma might be split and reinserted earlier while lock dropped.
1767 * The list of nonlinear vmas could be handled more efficiently, using
1768 * a placeholder, but handle it in the same way until a need is shown.
1769 * It is important to search the prio_tree before nonlinear list: a vma
1770 * may become nonlinear and be shifted from prio_tree to nonlinear list
1771 * while the lock is dropped; but never shifted from list to prio_tree.
1773 * In order to make forward progress despite restarting the search,
1774 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1775 * quickly skip it next time around. Since the prio_tree search only
1776 * shows us those vmas affected by unmapping the range in question, we
1777 * can't efficiently keep all vmas in step with mapping->truncate_count:
1778 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1779 * mapping->truncate_count and vma->vm_truncate_count are protected by
1782 * In order to make forward progress despite repeatedly restarting some
1783 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1784 * and restart from that address when we reach that vma again. It might
1785 * have been split or merged, shrunk or extended, but never shifted: so
1786 * restart_addr remains valid so long as it remains in the vma's range.
1787 * unmap_mapping_range forces truncate_count to leap over page-aligned
1788 * values so we can save vma's restart_addr in its truncate_count field.
1790 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1792 static void reset_vma_truncate_counts(struct address_space
*mapping
)
1794 struct vm_area_struct
*vma
;
1795 struct prio_tree_iter iter
;
1797 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, 0, ULONG_MAX
)
1798 vma
->vm_truncate_count
= 0;
1799 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
, shared
.vm_set
.list
)
1800 vma
->vm_truncate_count
= 0;
1803 static int unmap_mapping_range_vma(struct vm_area_struct
*vma
,
1804 unsigned long start_addr
, unsigned long end_addr
,
1805 struct zap_details
*details
)
1807 unsigned long restart_addr
;
1811 * files that support invalidating or truncating portions of the
1812 * file from under mmaped areas must have their ->fault function
1813 * return a locked page (and set VM_FAULT_LOCKED in the return).
1814 * This provides synchronisation against concurrent unmapping here.
1818 restart_addr
= vma
->vm_truncate_count
;
1819 if (is_restart_addr(restart_addr
) && start_addr
< restart_addr
) {
1820 start_addr
= restart_addr
;
1821 if (start_addr
>= end_addr
) {
1822 /* Top of vma has been split off since last time */
1823 vma
->vm_truncate_count
= details
->truncate_count
;
1828 restart_addr
= zap_page_range(vma
, start_addr
,
1829 end_addr
- start_addr
, details
);
1830 need_break
= need_resched() || spin_needbreak(details
->i_mmap_lock
);
1832 if (restart_addr
>= end_addr
) {
1833 /* We have now completed this vma: mark it so */
1834 vma
->vm_truncate_count
= details
->truncate_count
;
1838 /* Note restart_addr in vma's truncate_count field */
1839 vma
->vm_truncate_count
= restart_addr
;
1844 spin_unlock(details
->i_mmap_lock
);
1846 spin_lock(details
->i_mmap_lock
);
1850 static inline void unmap_mapping_range_tree(struct prio_tree_root
*root
,
1851 struct zap_details
*details
)
1853 struct vm_area_struct
*vma
;
1854 struct prio_tree_iter iter
;
1855 pgoff_t vba
, vea
, zba
, zea
;
1858 vma_prio_tree_foreach(vma
, &iter
, root
,
1859 details
->first_index
, details
->last_index
) {
1860 /* Skip quickly over those we have already dealt with */
1861 if (vma
->vm_truncate_count
== details
->truncate_count
)
1864 vba
= vma
->vm_pgoff
;
1865 vea
= vba
+ ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
) - 1;
1866 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1867 zba
= details
->first_index
;
1870 zea
= details
->last_index
;
1874 if (unmap_mapping_range_vma(vma
,
1875 ((zba
- vba
) << PAGE_SHIFT
) + vma
->vm_start
,
1876 ((zea
- vba
+ 1) << PAGE_SHIFT
) + vma
->vm_start
,
1882 static inline void unmap_mapping_range_list(struct list_head
*head
,
1883 struct zap_details
*details
)
1885 struct vm_area_struct
*vma
;
1888 * In nonlinear VMAs there is no correspondence between virtual address
1889 * offset and file offset. So we must perform an exhaustive search
1890 * across *all* the pages in each nonlinear VMA, not just the pages
1891 * whose virtual address lies outside the file truncation point.
1894 list_for_each_entry(vma
, head
, shared
.vm_set
.list
) {
1895 /* Skip quickly over those we have already dealt with */
1896 if (vma
->vm_truncate_count
== details
->truncate_count
)
1898 details
->nonlinear_vma
= vma
;
1899 if (unmap_mapping_range_vma(vma
, vma
->vm_start
,
1900 vma
->vm_end
, details
) < 0)
1906 * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
1907 * @mapping: the address space containing mmaps to be unmapped.
1908 * @holebegin: byte in first page to unmap, relative to the start of
1909 * the underlying file. This will be rounded down to a PAGE_SIZE
1910 * boundary. Note that this is different from vmtruncate(), which
1911 * must keep the partial page. In contrast, we must get rid of
1913 * @holelen: size of prospective hole in bytes. This will be rounded
1914 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1916 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1917 * but 0 when invalidating pagecache, don't throw away private data.
1919 void unmap_mapping_range(struct address_space
*mapping
,
1920 loff_t
const holebegin
, loff_t
const holelen
, int even_cows
)
1922 struct zap_details details
;
1923 pgoff_t hba
= holebegin
>> PAGE_SHIFT
;
1924 pgoff_t hlen
= (holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1926 /* Check for overflow. */
1927 if (sizeof(holelen
) > sizeof(hlen
)) {
1929 (holebegin
+ holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1930 if (holeend
& ~(long long)ULONG_MAX
)
1931 hlen
= ULONG_MAX
- hba
+ 1;
1934 details
.check_mapping
= even_cows
? NULL
: mapping
;
1935 details
.nonlinear_vma
= NULL
;
1936 details
.first_index
= hba
;
1937 details
.last_index
= hba
+ hlen
- 1;
1938 if (details
.last_index
< details
.first_index
)
1939 details
.last_index
= ULONG_MAX
;
1940 details
.i_mmap_lock
= &mapping
->i_mmap_lock
;
1942 spin_lock(&mapping
->i_mmap_lock
);
1944 /* Protect against endless unmapping loops */
1945 mapping
->truncate_count
++;
1946 if (unlikely(is_restart_addr(mapping
->truncate_count
))) {
1947 if (mapping
->truncate_count
== 0)
1948 reset_vma_truncate_counts(mapping
);
1949 mapping
->truncate_count
++;
1951 details
.truncate_count
= mapping
->truncate_count
;
1953 if (unlikely(!prio_tree_empty(&mapping
->i_mmap
)))
1954 unmap_mapping_range_tree(&mapping
->i_mmap
, &details
);
1955 if (unlikely(!list_empty(&mapping
->i_mmap_nonlinear
)))
1956 unmap_mapping_range_list(&mapping
->i_mmap_nonlinear
, &details
);
1957 spin_unlock(&mapping
->i_mmap_lock
);
1959 EXPORT_SYMBOL(unmap_mapping_range
);
1962 * vmtruncate - unmap mappings "freed" by truncate() syscall
1963 * @inode: inode of the file used
1964 * @offset: file offset to start truncating
1966 * NOTE! We have to be ready to update the memory sharing
1967 * between the file and the memory map for a potential last
1968 * incomplete page. Ugly, but necessary.
1970 int vmtruncate(struct inode
* inode
, loff_t offset
)
1972 if (inode
->i_size
< offset
) {
1973 unsigned long limit
;
1975 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1976 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
1978 if (offset
> inode
->i_sb
->s_maxbytes
)
1980 i_size_write(inode
, offset
);
1982 struct address_space
*mapping
= inode
->i_mapping
;
1985 * truncation of in-use swapfiles is disallowed - it would
1986 * cause subsequent swapout to scribble on the now-freed
1989 if (IS_SWAPFILE(inode
))
1991 i_size_write(inode
, offset
);
1994 * unmap_mapping_range is called twice, first simply for
1995 * efficiency so that truncate_inode_pages does fewer
1996 * single-page unmaps. However after this first call, and
1997 * before truncate_inode_pages finishes, it is possible for
1998 * private pages to be COWed, which remain after
1999 * truncate_inode_pages finishes, hence the second
2000 * unmap_mapping_range call must be made for correctness.
2002 unmap_mapping_range(mapping
, offset
+ PAGE_SIZE
- 1, 0, 1);
2003 truncate_inode_pages(mapping
, offset
);
2004 unmap_mapping_range(mapping
, offset
+ PAGE_SIZE
- 1, 0, 1);
2007 if (inode
->i_op
&& inode
->i_op
->truncate
)
2008 inode
->i_op
->truncate(inode
);
2012 send_sig(SIGXFSZ
, current
, 0);
2016 EXPORT_SYMBOL(vmtruncate
);
2018 int vmtruncate_range(struct inode
*inode
, loff_t offset
, loff_t end
)
2020 struct address_space
*mapping
= inode
->i_mapping
;
2023 * If the underlying filesystem is not going to provide
2024 * a way to truncate a range of blocks (punch a hole) -
2025 * we should return failure right now.
2027 if (!inode
->i_op
|| !inode
->i_op
->truncate_range
)
2030 mutex_lock(&inode
->i_mutex
);
2031 down_write(&inode
->i_alloc_sem
);
2032 unmap_mapping_range(mapping
, offset
, (end
- offset
), 1);
2033 truncate_inode_pages_range(mapping
, offset
, end
);
2034 unmap_mapping_range(mapping
, offset
, (end
- offset
), 1);
2035 inode
->i_op
->truncate_range(inode
, offset
, end
);
2036 up_write(&inode
->i_alloc_sem
);
2037 mutex_unlock(&inode
->i_mutex
);
2043 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2044 * but allow concurrent faults), and pte mapped but not yet locked.
2045 * We return with mmap_sem still held, but pte unmapped and unlocked.
2047 static int do_swap_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2048 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2049 int write_access
, pte_t orig_pte
)
2057 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
2060 entry
= pte_to_swp_entry(orig_pte
);
2061 if (is_migration_entry(entry
)) {
2062 migration_entry_wait(mm
, pmd
, address
);
2065 delayacct_set_flag(DELAYACCT_PF_SWAPIN
);
2066 page
= lookup_swap_cache(entry
);
2068 grab_swap_token(); /* Contend for token _before_ read-in */
2069 page
= swapin_readahead(entry
,
2070 GFP_HIGHUSER_MOVABLE
, vma
, address
);
2073 * Back out if somebody else faulted in this pte
2074 * while we released the pte lock.
2076 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2077 if (likely(pte_same(*page_table
, orig_pte
)))
2079 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2083 /* Had to read the page from swap area: Major fault */
2084 ret
= VM_FAULT_MAJOR
;
2085 count_vm_event(PGMAJFAULT
);
2088 if (mem_cgroup_charge(page
, mm
, GFP_KERNEL
)) {
2089 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2094 mark_page_accessed(page
);
2096 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2099 * Back out if somebody else already faulted in this pte.
2101 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2102 if (unlikely(!pte_same(*page_table
, orig_pte
)))
2105 if (unlikely(!PageUptodate(page
))) {
2106 ret
= VM_FAULT_SIGBUS
;
2110 /* The page isn't present yet, go ahead with the fault. */
2112 inc_mm_counter(mm
, anon_rss
);
2113 pte
= mk_pte(page
, vma
->vm_page_prot
);
2114 if (write_access
&& can_share_swap_page(page
)) {
2115 pte
= maybe_mkwrite(pte_mkdirty(pte
), vma
);
2119 flush_icache_page(vma
, page
);
2120 set_pte_at(mm
, address
, page_table
, pte
);
2121 page_add_anon_rmap(page
, vma
, address
);
2125 remove_exclusive_swap_page(page
);
2129 ret
|= do_wp_page(mm
, vma
, address
, page_table
, pmd
, ptl
, pte
);
2130 if (ret
& VM_FAULT_ERROR
)
2131 ret
&= VM_FAULT_ERROR
;
2135 /* No need to invalidate - it was non-present before */
2136 update_mmu_cache(vma
, address
, pte
);
2138 pte_unmap_unlock(page_table
, ptl
);
2142 mem_cgroup_uncharge_page(page
);
2143 pte_unmap_unlock(page_table
, ptl
);
2145 page_cache_release(page
);
2150 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2151 * but allow concurrent faults), and pte mapped but not yet locked.
2152 * We return with mmap_sem still held, but pte unmapped and unlocked.
2154 static int do_anonymous_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2155 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2162 /* Allocate our own private page. */
2163 pte_unmap(page_table
);
2165 if (unlikely(anon_vma_prepare(vma
)))
2167 page
= alloc_zeroed_user_highpage_movable(vma
, address
);
2170 __SetPageUptodate(page
);
2172 if (mem_cgroup_charge(page
, mm
, GFP_KERNEL
))
2175 entry
= mk_pte(page
, vma
->vm_page_prot
);
2176 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2178 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2179 if (!pte_none(*page_table
))
2181 inc_mm_counter(mm
, anon_rss
);
2182 lru_cache_add_active(page
);
2183 page_add_new_anon_rmap(page
, vma
, address
);
2184 set_pte_at(mm
, address
, page_table
, entry
);
2186 /* No need to invalidate - it was non-present before */
2187 update_mmu_cache(vma
, address
, entry
);
2189 pte_unmap_unlock(page_table
, ptl
);
2192 mem_cgroup_uncharge_page(page
);
2193 page_cache_release(page
);
2196 page_cache_release(page
);
2198 return VM_FAULT_OOM
;
2202 * __do_fault() tries to create a new page mapping. It aggressively
2203 * tries to share with existing pages, but makes a separate copy if
2204 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2205 * the next page fault.
2207 * As this is called only for pages that do not currently exist, we
2208 * do not need to flush old virtual caches or the TLB.
2210 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2211 * but allow concurrent faults), and pte neither mapped nor locked.
2212 * We return with mmap_sem still held, but pte unmapped and unlocked.
2214 static int __do_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2215 unsigned long address
, pmd_t
*pmd
,
2216 pgoff_t pgoff
, unsigned int flags
, pte_t orig_pte
)
2223 struct page
*dirty_page
= NULL
;
2224 struct vm_fault vmf
;
2226 int page_mkwrite
= 0;
2228 vmf
.virtual_address
= (void __user
*)(address
& PAGE_MASK
);
2233 BUG_ON(vma
->vm_flags
& VM_PFNMAP
);
2235 ret
= vma
->vm_ops
->fault(vma
, &vmf
);
2236 if (unlikely(ret
& (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
)))
2240 * For consistency in subsequent calls, make the faulted page always
2243 if (unlikely(!(ret
& VM_FAULT_LOCKED
)))
2244 lock_page(vmf
.page
);
2246 VM_BUG_ON(!PageLocked(vmf
.page
));
2249 * Should we do an early C-O-W break?
2252 if (flags
& FAULT_FLAG_WRITE
) {
2253 if (!(vma
->vm_flags
& VM_SHARED
)) {
2255 if (unlikely(anon_vma_prepare(vma
))) {
2259 page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
,
2265 copy_user_highpage(page
, vmf
.page
, address
, vma
);
2266 __SetPageUptodate(page
);
2269 * If the page will be shareable, see if the backing
2270 * address space wants to know that the page is about
2271 * to become writable
2273 if (vma
->vm_ops
->page_mkwrite
) {
2275 if (vma
->vm_ops
->page_mkwrite(vma
, page
) < 0) {
2276 ret
= VM_FAULT_SIGBUS
;
2277 anon
= 1; /* no anon but release vmf.page */
2282 * XXX: this is not quite right (racy vs
2283 * invalidate) to unlock and relock the page
2284 * like this, however a better fix requires
2285 * reworking page_mkwrite locking API, which
2286 * is better done later.
2288 if (!page
->mapping
) {
2290 anon
= 1; /* no anon but release vmf.page */
2299 if (mem_cgroup_charge(page
, mm
, GFP_KERNEL
)) {
2304 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2307 * This silly early PAGE_DIRTY setting removes a race
2308 * due to the bad i386 page protection. But it's valid
2309 * for other architectures too.
2311 * Note that if write_access is true, we either now have
2312 * an exclusive copy of the page, or this is a shared mapping,
2313 * so we can make it writable and dirty to avoid having to
2314 * handle that later.
2316 /* Only go through if we didn't race with anybody else... */
2317 if (likely(pte_same(*page_table
, orig_pte
))) {
2318 flush_icache_page(vma
, page
);
2319 entry
= mk_pte(page
, vma
->vm_page_prot
);
2320 if (flags
& FAULT_FLAG_WRITE
)
2321 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2322 set_pte_at(mm
, address
, page_table
, entry
);
2324 inc_mm_counter(mm
, anon_rss
);
2325 lru_cache_add_active(page
);
2326 page_add_new_anon_rmap(page
, vma
, address
);
2328 inc_mm_counter(mm
, file_rss
);
2329 page_add_file_rmap(page
);
2330 if (flags
& FAULT_FLAG_WRITE
) {
2332 get_page(dirty_page
);
2336 /* no need to invalidate: a not-present page won't be cached */
2337 update_mmu_cache(vma
, address
, entry
);
2339 mem_cgroup_uncharge_page(page
);
2341 page_cache_release(page
);
2343 anon
= 1; /* no anon but release faulted_page */
2346 pte_unmap_unlock(page_table
, ptl
);
2349 unlock_page(vmf
.page
);
2352 page_cache_release(vmf
.page
);
2353 else if (dirty_page
) {
2355 file_update_time(vma
->vm_file
);
2357 set_page_dirty_balance(dirty_page
, page_mkwrite
);
2358 put_page(dirty_page
);
2364 static int do_linear_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2365 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2366 int write_access
, pte_t orig_pte
)
2368 pgoff_t pgoff
= (((address
& PAGE_MASK
)
2369 - vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
;
2370 unsigned int flags
= (write_access
? FAULT_FLAG_WRITE
: 0);
2372 pte_unmap(page_table
);
2373 return __do_fault(mm
, vma
, address
, pmd
, pgoff
, flags
, orig_pte
);
2378 * do_no_pfn() tries to create a new page mapping for a page without
2379 * a struct_page backing it
2381 * As this is called only for pages that do not currently exist, we
2382 * do not need to flush old virtual caches or the TLB.
2384 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2385 * but allow concurrent faults), and pte mapped but not yet locked.
2386 * We return with mmap_sem still held, but pte unmapped and unlocked.
2388 * It is expected that the ->nopfn handler always returns the same pfn
2389 * for a given virtual mapping.
2391 * Mark this `noinline' to prevent it from bloating the main pagefault code.
2393 static noinline
int do_no_pfn(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2394 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2401 pte_unmap(page_table
);
2402 BUG_ON(!(vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
)));
2403 BUG_ON((vma
->vm_flags
& VM_PFNMAP
) && is_cow_mapping(vma
->vm_flags
));
2405 pfn
= vma
->vm_ops
->nopfn(vma
, address
& PAGE_MASK
);
2407 BUG_ON((vma
->vm_flags
& VM_MIXEDMAP
) && pfn_valid(pfn
));
2409 if (unlikely(pfn
== NOPFN_OOM
))
2410 return VM_FAULT_OOM
;
2411 else if (unlikely(pfn
== NOPFN_SIGBUS
))
2412 return VM_FAULT_SIGBUS
;
2413 else if (unlikely(pfn
== NOPFN_REFAULT
))
2416 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2418 /* Only go through if we didn't race with anybody else... */
2419 if (pte_none(*page_table
)) {
2420 entry
= pfn_pte(pfn
, vma
->vm_page_prot
);
2422 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2423 set_pte_at(mm
, address
, page_table
, entry
);
2425 pte_unmap_unlock(page_table
, ptl
);
2430 * Fault of a previously existing named mapping. Repopulate the pte
2431 * from the encoded file_pte if possible. This enables swappable
2434 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2435 * but allow concurrent faults), and pte mapped but not yet locked.
2436 * We return with mmap_sem still held, but pte unmapped and unlocked.
2438 static int do_nonlinear_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2439 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2440 int write_access
, pte_t orig_pte
)
2442 unsigned int flags
= FAULT_FLAG_NONLINEAR
|
2443 (write_access
? FAULT_FLAG_WRITE
: 0);
2446 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
2449 if (unlikely(!(vma
->vm_flags
& VM_NONLINEAR
) ||
2450 !(vma
->vm_flags
& VM_CAN_NONLINEAR
))) {
2452 * Page table corrupted: show pte and kill process.
2454 print_bad_pte(vma
, orig_pte
, address
);
2455 return VM_FAULT_OOM
;
2458 pgoff
= pte_to_pgoff(orig_pte
);
2459 return __do_fault(mm
, vma
, address
, pmd
, pgoff
, flags
, orig_pte
);
2463 * These routines also need to handle stuff like marking pages dirty
2464 * and/or accessed for architectures that don't do it in hardware (most
2465 * RISC architectures). The early dirtying is also good on the i386.
2467 * There is also a hook called "update_mmu_cache()" that architectures
2468 * with external mmu caches can use to update those (ie the Sparc or
2469 * PowerPC hashed page tables that act as extended TLBs).
2471 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2472 * but allow concurrent faults), and pte mapped but not yet locked.
2473 * We return with mmap_sem still held, but pte unmapped and unlocked.
2475 static inline int handle_pte_fault(struct mm_struct
*mm
,
2476 struct vm_area_struct
*vma
, unsigned long address
,
2477 pte_t
*pte
, pmd_t
*pmd
, int write_access
)
2483 if (!pte_present(entry
)) {
2484 if (pte_none(entry
)) {
2486 if (likely(vma
->vm_ops
->fault
))
2487 return do_linear_fault(mm
, vma
, address
,
2488 pte
, pmd
, write_access
, entry
);
2489 if (unlikely(vma
->vm_ops
->nopfn
))
2490 return do_no_pfn(mm
, vma
, address
, pte
,
2493 return do_anonymous_page(mm
, vma
, address
,
2494 pte
, pmd
, write_access
);
2496 if (pte_file(entry
))
2497 return do_nonlinear_fault(mm
, vma
, address
,
2498 pte
, pmd
, write_access
, entry
);
2499 return do_swap_page(mm
, vma
, address
,
2500 pte
, pmd
, write_access
, entry
);
2503 ptl
= pte_lockptr(mm
, pmd
);
2505 if (unlikely(!pte_same(*pte
, entry
)))
2508 if (!pte_write(entry
))
2509 return do_wp_page(mm
, vma
, address
,
2510 pte
, pmd
, ptl
, entry
);
2511 entry
= pte_mkdirty(entry
);
2513 entry
= pte_mkyoung(entry
);
2514 if (ptep_set_access_flags(vma
, address
, pte
, entry
, write_access
)) {
2515 update_mmu_cache(vma
, address
, entry
);
2518 * This is needed only for protection faults but the arch code
2519 * is not yet telling us if this is a protection fault or not.
2520 * This still avoids useless tlb flushes for .text page faults
2524 flush_tlb_page(vma
, address
);
2527 pte_unmap_unlock(pte
, ptl
);
2532 * By the time we get here, we already hold the mm semaphore
2534 int handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2535 unsigned long address
, int write_access
)
2542 __set_current_state(TASK_RUNNING
);
2544 count_vm_event(PGFAULT
);
2546 if (unlikely(is_vm_hugetlb_page(vma
)))
2547 return hugetlb_fault(mm
, vma
, address
, write_access
);
2549 pgd
= pgd_offset(mm
, address
);
2550 pud
= pud_alloc(mm
, pgd
, address
);
2552 return VM_FAULT_OOM
;
2553 pmd
= pmd_alloc(mm
, pud
, address
);
2555 return VM_FAULT_OOM
;
2556 pte
= pte_alloc_map(mm
, pmd
, address
);
2558 return VM_FAULT_OOM
;
2560 return handle_pte_fault(mm
, vma
, address
, pte
, pmd
, write_access
);
2563 #ifndef __PAGETABLE_PUD_FOLDED
2565 * Allocate page upper directory.
2566 * We've already handled the fast-path in-line.
2568 int __pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
2570 pud_t
*new = pud_alloc_one(mm
, address
);
2574 spin_lock(&mm
->page_table_lock
);
2575 if (pgd_present(*pgd
)) /* Another has populated it */
2578 pgd_populate(mm
, pgd
, new);
2579 spin_unlock(&mm
->page_table_lock
);
2582 #endif /* __PAGETABLE_PUD_FOLDED */
2584 #ifndef __PAGETABLE_PMD_FOLDED
2586 * Allocate page middle directory.
2587 * We've already handled the fast-path in-line.
2589 int __pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
2591 pmd_t
*new = pmd_alloc_one(mm
, address
);
2595 spin_lock(&mm
->page_table_lock
);
2596 #ifndef __ARCH_HAS_4LEVEL_HACK
2597 if (pud_present(*pud
)) /* Another has populated it */
2600 pud_populate(mm
, pud
, new);
2602 if (pgd_present(*pud
)) /* Another has populated it */
2605 pgd_populate(mm
, pud
, new);
2606 #endif /* __ARCH_HAS_4LEVEL_HACK */
2607 spin_unlock(&mm
->page_table_lock
);
2610 #endif /* __PAGETABLE_PMD_FOLDED */
2612 int make_pages_present(unsigned long addr
, unsigned long end
)
2614 int ret
, len
, write
;
2615 struct vm_area_struct
* vma
;
2617 vma
= find_vma(current
->mm
, addr
);
2620 write
= (vma
->vm_flags
& VM_WRITE
) != 0;
2621 BUG_ON(addr
>= end
);
2622 BUG_ON(end
> vma
->vm_end
);
2623 len
= DIV_ROUND_UP(end
, PAGE_SIZE
) - addr
/PAGE_SIZE
;
2624 ret
= get_user_pages(current
, current
->mm
, addr
,
2625 len
, write
, 0, NULL
, NULL
);
2628 return ret
== len
? 0 : -1;
2631 #if !defined(__HAVE_ARCH_GATE_AREA)
2633 #if defined(AT_SYSINFO_EHDR)
2634 static struct vm_area_struct gate_vma
;
2636 static int __init
gate_vma_init(void)
2638 gate_vma
.vm_mm
= NULL
;
2639 gate_vma
.vm_start
= FIXADDR_USER_START
;
2640 gate_vma
.vm_end
= FIXADDR_USER_END
;
2641 gate_vma
.vm_flags
= VM_READ
| VM_MAYREAD
| VM_EXEC
| VM_MAYEXEC
;
2642 gate_vma
.vm_page_prot
= __P101
;
2644 * Make sure the vDSO gets into every core dump.
2645 * Dumping its contents makes post-mortem fully interpretable later
2646 * without matching up the same kernel and hardware config to see
2647 * what PC values meant.
2649 gate_vma
.vm_flags
|= VM_ALWAYSDUMP
;
2652 __initcall(gate_vma_init
);
2655 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
2657 #ifdef AT_SYSINFO_EHDR
2664 int in_gate_area_no_task(unsigned long addr
)
2666 #ifdef AT_SYSINFO_EHDR
2667 if ((addr
>= FIXADDR_USER_START
) && (addr
< FIXADDR_USER_END
))
2673 #endif /* __HAVE_ARCH_GATE_AREA */
2676 * Access another process' address space.
2677 * Source/target buffer must be kernel space,
2678 * Do not walk the page table directly, use get_user_pages
2680 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
, void *buf
, int len
, int write
)
2682 struct mm_struct
*mm
;
2683 struct vm_area_struct
*vma
;
2685 void *old_buf
= buf
;
2687 mm
= get_task_mm(tsk
);
2691 down_read(&mm
->mmap_sem
);
2692 /* ignore errors, just check how much was successfully transferred */
2694 int bytes
, ret
, offset
;
2697 ret
= get_user_pages(tsk
, mm
, addr
, 1,
2698 write
, 1, &page
, &vma
);
2703 offset
= addr
& (PAGE_SIZE
-1);
2704 if (bytes
> PAGE_SIZE
-offset
)
2705 bytes
= PAGE_SIZE
-offset
;
2709 copy_to_user_page(vma
, page
, addr
,
2710 maddr
+ offset
, buf
, bytes
);
2711 set_page_dirty_lock(page
);
2713 copy_from_user_page(vma
, page
, addr
,
2714 buf
, maddr
+ offset
, bytes
);
2717 page_cache_release(page
);
2722 up_read(&mm
->mmap_sem
);
2725 return buf
- old_buf
;
2729 * Print the name of a VMA.
2731 void print_vma_addr(char *prefix
, unsigned long ip
)
2733 struct mm_struct
*mm
= current
->mm
;
2734 struct vm_area_struct
*vma
;
2737 * Do not print if we are in atomic
2738 * contexts (in exception stacks, etc.):
2740 if (preempt_count())
2743 down_read(&mm
->mmap_sem
);
2744 vma
= find_vma(mm
, ip
);
2745 if (vma
&& vma
->vm_file
) {
2746 struct file
*f
= vma
->vm_file
;
2747 char *buf
= (char *)__get_free_page(GFP_KERNEL
);
2751 p
= d_path(&f
->f_path
, buf
, PAGE_SIZE
);
2754 s
= strrchr(p
, '/');
2757 printk("%s%s[%lx+%lx]", prefix
, p
,
2759 vma
->vm_end
- vma
->vm_start
);
2760 free_page((unsigned long)buf
);
2763 up_read(¤t
->mm
->mmap_sem
);