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/init.h>
52 #include <asm/pgalloc.h>
53 #include <asm/uaccess.h>
55 #include <asm/tlbflush.h>
56 #include <asm/pgtable.h>
58 #include <linux/swapops.h>
59 #include <linux/elf.h>
61 #ifndef CONFIG_NEED_MULTIPLE_NODES
62 /* use the per-pgdat data instead for discontigmem - mbligh */
63 unsigned long max_mapnr
;
66 EXPORT_SYMBOL(max_mapnr
);
67 EXPORT_SYMBOL(mem_map
);
70 unsigned long num_physpages
;
72 * A number of key systems in x86 including ioremap() rely on the assumption
73 * that high_memory defines the upper bound on direct map memory, then end
74 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
75 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
79 unsigned long vmalloc_earlyreserve
;
81 EXPORT_SYMBOL(num_physpages
);
82 EXPORT_SYMBOL(high_memory
);
83 EXPORT_SYMBOL(vmalloc_earlyreserve
);
86 * If a p?d_bad entry is found while walking page tables, report
87 * the error, before resetting entry to p?d_none. Usually (but
88 * very seldom) called out from the p?d_none_or_clear_bad macros.
91 void pgd_clear_bad(pgd_t
*pgd
)
97 void pud_clear_bad(pud_t
*pud
)
103 void pmd_clear_bad(pmd_t
*pmd
)
110 * Note: this doesn't free the actual pages themselves. That
111 * has been handled earlier when unmapping all the memory regions.
113 static void free_pte_range(struct mmu_gather
*tlb
, pmd_t
*pmd
)
115 struct page
*page
= pmd_page(*pmd
);
117 pte_free_tlb(tlb
, page
);
118 dec_page_state(nr_page_table_pages
);
122 static inline void free_pmd_range(struct mmu_gather
*tlb
, pud_t
*pud
,
123 unsigned long addr
, unsigned long end
,
124 unsigned long floor
, unsigned long ceiling
)
131 pmd
= pmd_offset(pud
, addr
);
133 next
= pmd_addr_end(addr
, end
);
134 if (pmd_none_or_clear_bad(pmd
))
136 free_pte_range(tlb
, pmd
);
137 } while (pmd
++, addr
= next
, addr
!= end
);
147 if (end
- 1 > ceiling
- 1)
150 pmd
= pmd_offset(pud
, start
);
152 pmd_free_tlb(tlb
, pmd
);
155 static inline void free_pud_range(struct mmu_gather
*tlb
, pgd_t
*pgd
,
156 unsigned long addr
, unsigned long end
,
157 unsigned long floor
, unsigned long ceiling
)
164 pud
= pud_offset(pgd
, addr
);
166 next
= pud_addr_end(addr
, end
);
167 if (pud_none_or_clear_bad(pud
))
169 free_pmd_range(tlb
, pud
, addr
, next
, floor
, ceiling
);
170 } while (pud
++, addr
= next
, addr
!= end
);
176 ceiling
&= PGDIR_MASK
;
180 if (end
- 1 > ceiling
- 1)
183 pud
= pud_offset(pgd
, start
);
185 pud_free_tlb(tlb
, pud
);
189 * This function frees user-level page tables of a process.
191 * Must be called with pagetable lock held.
193 void free_pgd_range(struct mmu_gather
**tlb
,
194 unsigned long addr
, unsigned long end
,
195 unsigned long floor
, unsigned long ceiling
)
202 * The next few lines have given us lots of grief...
204 * Why are we testing PMD* at this top level? Because often
205 * there will be no work to do at all, and we'd prefer not to
206 * go all the way down to the bottom just to discover that.
208 * Why all these "- 1"s? Because 0 represents both the bottom
209 * of the address space and the top of it (using -1 for the
210 * top wouldn't help much: the masks would do the wrong thing).
211 * The rule is that addr 0 and floor 0 refer to the bottom of
212 * the address space, but end 0 and ceiling 0 refer to the top
213 * Comparisons need to use "end - 1" and "ceiling - 1" (though
214 * that end 0 case should be mythical).
216 * Wherever addr is brought up or ceiling brought down, we must
217 * be careful to reject "the opposite 0" before it confuses the
218 * subsequent tests. But what about where end is brought down
219 * by PMD_SIZE below? no, end can't go down to 0 there.
221 * Whereas we round start (addr) and ceiling down, by different
222 * masks at different levels, in order to test whether a table
223 * now has no other vmas using it, so can be freed, we don't
224 * bother to round floor or end up - the tests don't need that.
238 if (end
- 1 > ceiling
- 1)
244 pgd
= pgd_offset((*tlb
)->mm
, addr
);
246 next
= pgd_addr_end(addr
, end
);
247 if (pgd_none_or_clear_bad(pgd
))
249 free_pud_range(*tlb
, pgd
, addr
, next
, floor
, ceiling
);
250 } while (pgd
++, addr
= next
, addr
!= end
);
252 if (!tlb_is_full_mm(*tlb
))
253 flush_tlb_pgtables((*tlb
)->mm
, start
, end
);
256 void free_pgtables(struct mmu_gather
**tlb
, struct vm_area_struct
*vma
,
257 unsigned long floor
, unsigned long ceiling
)
260 struct vm_area_struct
*next
= vma
->vm_next
;
261 unsigned long addr
= vma
->vm_start
;
263 if (is_hugepage_only_range(vma
->vm_mm
, addr
, HPAGE_SIZE
)) {
264 hugetlb_free_pgd_range(tlb
, addr
, vma
->vm_end
,
265 floor
, next
? next
->vm_start
: ceiling
);
268 * Optimization: gather nearby vmas into one call down
270 while (next
&& next
->vm_start
<= vma
->vm_end
+ PMD_SIZE
271 && !is_hugepage_only_range(vma
->vm_mm
, next
->vm_start
,
276 free_pgd_range(tlb
, addr
, vma
->vm_end
,
277 floor
, next
? next
->vm_start
: ceiling
);
283 pte_t fastcall
*pte_alloc_map(struct mm_struct
*mm
, pmd_t
*pmd
,
284 unsigned long address
)
286 if (!pmd_present(*pmd
)) {
289 spin_unlock(&mm
->page_table_lock
);
290 new = pte_alloc_one(mm
, address
);
291 spin_lock(&mm
->page_table_lock
);
295 * Because we dropped the lock, we should re-check the
296 * entry, as somebody else could have populated it..
298 if (pmd_present(*pmd
)) {
303 inc_page_state(nr_page_table_pages
);
304 pmd_populate(mm
, pmd
, new);
307 return pte_offset_map(pmd
, address
);
310 pte_t fastcall
* pte_alloc_kernel(struct mm_struct
*mm
, pmd_t
*pmd
, unsigned long address
)
312 if (!pmd_present(*pmd
)) {
315 spin_unlock(&mm
->page_table_lock
);
316 new = pte_alloc_one_kernel(mm
, address
);
317 spin_lock(&mm
->page_table_lock
);
322 * Because we dropped the lock, we should re-check the
323 * entry, as somebody else could have populated it..
325 if (pmd_present(*pmd
)) {
326 pte_free_kernel(new);
329 pmd_populate_kernel(mm
, pmd
, new);
332 return pte_offset_kernel(pmd
, address
);
336 * copy one vm_area from one task to the other. Assumes the page tables
337 * already present in the new task to be cleared in the whole range
338 * covered by this vma.
340 * dst->page_table_lock is held on entry and exit,
341 * but may be dropped within p[mg]d_alloc() and pte_alloc_map().
345 copy_one_pte(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
346 pte_t
*dst_pte
, pte_t
*src_pte
, unsigned long vm_flags
,
349 pte_t pte
= *src_pte
;
353 /* pte contains position in swap or file, so copy. */
354 if (unlikely(!pte_present(pte
))) {
355 if (!pte_file(pte
)) {
356 swap_duplicate(pte_to_swp_entry(pte
));
357 /* make sure dst_mm is on swapoff's mmlist. */
358 if (unlikely(list_empty(&dst_mm
->mmlist
))) {
359 spin_lock(&mmlist_lock
);
360 list_add(&dst_mm
->mmlist
, &src_mm
->mmlist
);
361 spin_unlock(&mmlist_lock
);
364 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
369 /* the pte points outside of valid memory, the
370 * mapping is assumed to be good, meaningful
371 * and not mapped via rmap - duplicate the
376 page
= pfn_to_page(pfn
);
378 if (!page
|| PageReserved(page
)) {
379 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
384 * If it's a COW mapping, write protect it both
385 * in the parent and the child
387 if ((vm_flags
& (VM_SHARED
| VM_MAYWRITE
)) == VM_MAYWRITE
) {
388 ptep_set_wrprotect(src_mm
, addr
, src_pte
);
393 * If it's a shared mapping, mark it clean in
396 if (vm_flags
& VM_SHARED
)
397 pte
= pte_mkclean(pte
);
398 pte
= pte_mkold(pte
);
400 inc_mm_counter(dst_mm
, rss
);
402 inc_mm_counter(dst_mm
, anon_rss
);
403 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
407 static int copy_pte_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
408 pmd_t
*dst_pmd
, pmd_t
*src_pmd
, struct vm_area_struct
*vma
,
409 unsigned long addr
, unsigned long end
)
411 pte_t
*src_pte
, *dst_pte
;
412 unsigned long vm_flags
= vma
->vm_flags
;
416 dst_pte
= pte_alloc_map(dst_mm
, dst_pmd
, addr
);
419 src_pte
= pte_offset_map_nested(src_pmd
, addr
);
421 spin_lock(&src_mm
->page_table_lock
);
424 * We are holding two locks at this point - either of them
425 * could generate latencies in another task on another CPU.
427 if (progress
>= 32) {
429 if (need_resched() ||
430 need_lockbreak(&src_mm
->page_table_lock
) ||
431 need_lockbreak(&dst_mm
->page_table_lock
))
434 if (pte_none(*src_pte
)) {
438 copy_one_pte(dst_mm
, src_mm
, dst_pte
, src_pte
, vm_flags
, addr
);
440 } while (dst_pte
++, src_pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
441 spin_unlock(&src_mm
->page_table_lock
);
443 pte_unmap_nested(src_pte
- 1);
444 pte_unmap(dst_pte
- 1);
445 cond_resched_lock(&dst_mm
->page_table_lock
);
451 static inline int copy_pmd_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
452 pud_t
*dst_pud
, pud_t
*src_pud
, struct vm_area_struct
*vma
,
453 unsigned long addr
, unsigned long end
)
455 pmd_t
*src_pmd
, *dst_pmd
;
458 dst_pmd
= pmd_alloc(dst_mm
, dst_pud
, addr
);
461 src_pmd
= pmd_offset(src_pud
, addr
);
463 next
= pmd_addr_end(addr
, end
);
464 if (pmd_none_or_clear_bad(src_pmd
))
466 if (copy_pte_range(dst_mm
, src_mm
, dst_pmd
, src_pmd
,
469 } while (dst_pmd
++, src_pmd
++, addr
= next
, addr
!= end
);
473 static inline int copy_pud_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
474 pgd_t
*dst_pgd
, pgd_t
*src_pgd
, struct vm_area_struct
*vma
,
475 unsigned long addr
, unsigned long end
)
477 pud_t
*src_pud
, *dst_pud
;
480 dst_pud
= pud_alloc(dst_mm
, dst_pgd
, addr
);
483 src_pud
= pud_offset(src_pgd
, addr
);
485 next
= pud_addr_end(addr
, end
);
486 if (pud_none_or_clear_bad(src_pud
))
488 if (copy_pmd_range(dst_mm
, src_mm
, dst_pud
, src_pud
,
491 } while (dst_pud
++, src_pud
++, addr
= next
, addr
!= end
);
495 int copy_page_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
496 struct vm_area_struct
*vma
)
498 pgd_t
*src_pgd
, *dst_pgd
;
500 unsigned long addr
= vma
->vm_start
;
501 unsigned long end
= vma
->vm_end
;
504 * Don't copy ptes where a page fault will fill them correctly.
505 * Fork becomes much lighter when there are big shared or private
506 * readonly mappings. The tradeoff is that copy_page_range is more
507 * efficient than faulting.
509 if (!(vma
->vm_flags
& (VM_HUGETLB
|VM_NONLINEAR
|VM_RESERVED
))) {
514 if (is_vm_hugetlb_page(vma
))
515 return copy_hugetlb_page_range(dst_mm
, src_mm
, vma
);
517 dst_pgd
= pgd_offset(dst_mm
, addr
);
518 src_pgd
= pgd_offset(src_mm
, addr
);
520 next
= pgd_addr_end(addr
, end
);
521 if (pgd_none_or_clear_bad(src_pgd
))
523 if (copy_pud_range(dst_mm
, src_mm
, dst_pgd
, src_pgd
,
526 } while (dst_pgd
++, src_pgd
++, addr
= next
, addr
!= end
);
530 static void zap_pte_range(struct mmu_gather
*tlb
, pmd_t
*pmd
,
531 unsigned long addr
, unsigned long end
,
532 struct zap_details
*details
)
536 pte
= pte_offset_map(pmd
, addr
);
541 if (pte_present(ptent
)) {
542 struct page
*page
= NULL
;
543 unsigned long pfn
= pte_pfn(ptent
);
544 if (pfn_valid(pfn
)) {
545 page
= pfn_to_page(pfn
);
546 if (PageReserved(page
))
549 if (unlikely(details
) && page
) {
551 * unmap_shared_mapping_pages() wants to
552 * invalidate cache without truncating:
553 * unmap shared but keep private pages.
555 if (details
->check_mapping
&&
556 details
->check_mapping
!= page
->mapping
)
559 * Each page->index must be checked when
560 * invalidating or truncating nonlinear.
562 if (details
->nonlinear_vma
&&
563 (page
->index
< details
->first_index
||
564 page
->index
> details
->last_index
))
567 ptent
= ptep_get_and_clear_full(tlb
->mm
, addr
, pte
,
569 tlb_remove_tlb_entry(tlb
, pte
, addr
);
572 if (unlikely(details
) && details
->nonlinear_vma
573 && linear_page_index(details
->nonlinear_vma
,
574 addr
) != page
->index
)
575 set_pte_at(tlb
->mm
, addr
, pte
,
576 pgoff_to_pte(page
->index
));
577 if (pte_dirty(ptent
))
578 set_page_dirty(page
);
580 dec_mm_counter(tlb
->mm
, anon_rss
);
581 else if (pte_young(ptent
))
582 mark_page_accessed(page
);
584 page_remove_rmap(page
);
585 tlb_remove_page(tlb
, page
);
589 * If details->check_mapping, we leave swap entries;
590 * if details->nonlinear_vma, we leave file entries.
592 if (unlikely(details
))
594 if (!pte_file(ptent
))
595 free_swap_and_cache(pte_to_swp_entry(ptent
));
596 pte_clear_full(tlb
->mm
, addr
, pte
, tlb
->fullmm
);
597 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
601 static inline void zap_pmd_range(struct mmu_gather
*tlb
, pud_t
*pud
,
602 unsigned long addr
, unsigned long end
,
603 struct zap_details
*details
)
608 pmd
= pmd_offset(pud
, addr
);
610 next
= pmd_addr_end(addr
, end
);
611 if (pmd_none_or_clear_bad(pmd
))
613 zap_pte_range(tlb
, pmd
, addr
, next
, details
);
614 } while (pmd
++, addr
= next
, addr
!= end
);
617 static inline void zap_pud_range(struct mmu_gather
*tlb
, pgd_t
*pgd
,
618 unsigned long addr
, unsigned long end
,
619 struct zap_details
*details
)
624 pud
= pud_offset(pgd
, addr
);
626 next
= pud_addr_end(addr
, end
);
627 if (pud_none_or_clear_bad(pud
))
629 zap_pmd_range(tlb
, pud
, addr
, next
, details
);
630 } while (pud
++, addr
= next
, addr
!= end
);
633 static void unmap_page_range(struct mmu_gather
*tlb
, struct vm_area_struct
*vma
,
634 unsigned long addr
, unsigned long end
,
635 struct zap_details
*details
)
640 if (details
&& !details
->check_mapping
&& !details
->nonlinear_vma
)
644 tlb_start_vma(tlb
, vma
);
645 pgd
= pgd_offset(vma
->vm_mm
, addr
);
647 next
= pgd_addr_end(addr
, end
);
648 if (pgd_none_or_clear_bad(pgd
))
650 zap_pud_range(tlb
, pgd
, addr
, next
, details
);
651 } while (pgd
++, addr
= next
, addr
!= end
);
652 tlb_end_vma(tlb
, vma
);
655 #ifdef CONFIG_PREEMPT
656 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
658 /* No preempt: go for improved straight-line efficiency */
659 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
663 * unmap_vmas - unmap a range of memory covered by a list of vma's
664 * @tlbp: address of the caller's struct mmu_gather
665 * @mm: the controlling mm_struct
666 * @vma: the starting vma
667 * @start_addr: virtual address at which to start unmapping
668 * @end_addr: virtual address at which to end unmapping
669 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
670 * @details: details of nonlinear truncation or shared cache invalidation
672 * Returns the end address of the unmapping (restart addr if interrupted).
674 * Unmap all pages in the vma list. Called under page_table_lock.
676 * We aim to not hold page_table_lock for too long (for scheduling latency
677 * reasons). So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
678 * return the ending mmu_gather to the caller.
680 * Only addresses between `start' and `end' will be unmapped.
682 * The VMA list must be sorted in ascending virtual address order.
684 * unmap_vmas() assumes that the caller will flush the whole unmapped address
685 * range after unmap_vmas() returns. So the only responsibility here is to
686 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
687 * drops the lock and schedules.
689 unsigned long unmap_vmas(struct mmu_gather
**tlbp
, struct mm_struct
*mm
,
690 struct vm_area_struct
*vma
, unsigned long start_addr
,
691 unsigned long end_addr
, unsigned long *nr_accounted
,
692 struct zap_details
*details
)
694 unsigned long zap_bytes
= ZAP_BLOCK_SIZE
;
695 unsigned long tlb_start
= 0; /* For tlb_finish_mmu */
696 int tlb_start_valid
= 0;
697 unsigned long start
= start_addr
;
698 spinlock_t
*i_mmap_lock
= details
? details
->i_mmap_lock
: NULL
;
699 int fullmm
= tlb_is_full_mm(*tlbp
);
701 for ( ; vma
&& vma
->vm_start
< end_addr
; vma
= vma
->vm_next
) {
704 start
= max(vma
->vm_start
, start_addr
);
705 if (start
>= vma
->vm_end
)
707 end
= min(vma
->vm_end
, end_addr
);
708 if (end
<= vma
->vm_start
)
711 if (vma
->vm_flags
& VM_ACCOUNT
)
712 *nr_accounted
+= (end
- start
) >> PAGE_SHIFT
;
714 while (start
!= end
) {
717 if (!tlb_start_valid
) {
722 if (is_vm_hugetlb_page(vma
)) {
724 unmap_hugepage_range(vma
, start
, end
);
726 block
= min(zap_bytes
, end
- start
);
727 unmap_page_range(*tlbp
, vma
, start
,
728 start
+ block
, details
);
733 if ((long)zap_bytes
> 0)
736 tlb_finish_mmu(*tlbp
, tlb_start
, start
);
738 if (need_resched() ||
739 need_lockbreak(&mm
->page_table_lock
) ||
740 (i_mmap_lock
&& need_lockbreak(i_mmap_lock
))) {
742 /* must reset count of rss freed */
743 *tlbp
= tlb_gather_mmu(mm
, fullmm
);
746 spin_unlock(&mm
->page_table_lock
);
748 spin_lock(&mm
->page_table_lock
);
751 *tlbp
= tlb_gather_mmu(mm
, fullmm
);
753 zap_bytes
= ZAP_BLOCK_SIZE
;
757 return start
; /* which is now the end (or restart) address */
761 * zap_page_range - remove user pages in a given range
762 * @vma: vm_area_struct holding the applicable pages
763 * @address: starting address of pages to zap
764 * @size: number of bytes to zap
765 * @details: details of nonlinear truncation or shared cache invalidation
767 unsigned long zap_page_range(struct vm_area_struct
*vma
, unsigned long address
,
768 unsigned long size
, struct zap_details
*details
)
770 struct mm_struct
*mm
= vma
->vm_mm
;
771 struct mmu_gather
*tlb
;
772 unsigned long end
= address
+ size
;
773 unsigned long nr_accounted
= 0;
775 if (is_vm_hugetlb_page(vma
)) {
776 zap_hugepage_range(vma
, address
, size
);
781 spin_lock(&mm
->page_table_lock
);
782 tlb
= tlb_gather_mmu(mm
, 0);
783 end
= unmap_vmas(&tlb
, mm
, vma
, address
, end
, &nr_accounted
, details
);
784 tlb_finish_mmu(tlb
, address
, end
);
785 spin_unlock(&mm
->page_table_lock
);
790 * Do a quick page-table lookup for a single page.
791 * mm->page_table_lock must be held.
793 static struct page
*__follow_page(struct mm_struct
*mm
, unsigned long address
,
794 int read
, int write
, int accessed
)
803 page
= follow_huge_addr(mm
, address
, write
);
807 pgd
= pgd_offset(mm
, address
);
808 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
811 pud
= pud_offset(pgd
, address
);
812 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
815 pmd
= pmd_offset(pud
, address
);
816 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
819 return follow_huge_pmd(mm
, address
, pmd
, write
);
821 ptep
= pte_offset_map(pmd
, address
);
827 if (pte_present(pte
)) {
828 if (write
&& !pte_write(pte
))
830 if (read
&& !pte_read(pte
))
833 if (pfn_valid(pfn
)) {
834 page
= pfn_to_page(pfn
);
836 if (write
&& !pte_dirty(pte
) &&!PageDirty(page
))
837 set_page_dirty(page
);
838 mark_page_accessed(page
);
849 follow_page(struct mm_struct
*mm
, unsigned long address
, int write
)
851 return __follow_page(mm
, address
, 0, write
, 1);
855 * check_user_page_readable() can be called frm niterrupt context by oprofile,
856 * so we need to avoid taking any non-irq-safe locks
858 int check_user_page_readable(struct mm_struct
*mm
, unsigned long address
)
860 return __follow_page(mm
, address
, 1, 0, 0) != NULL
;
862 EXPORT_SYMBOL(check_user_page_readable
);
865 untouched_anonymous_page(struct mm_struct
* mm
, struct vm_area_struct
*vma
,
866 unsigned long address
)
872 /* Check if the vma is for an anonymous mapping. */
873 if (vma
->vm_ops
&& vma
->vm_ops
->nopage
)
876 /* Check if page directory entry exists. */
877 pgd
= pgd_offset(mm
, address
);
878 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
881 pud
= pud_offset(pgd
, address
);
882 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
885 /* Check if page middle directory entry exists. */
886 pmd
= pmd_offset(pud
, address
);
887 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
890 /* There is a pte slot for 'address' in 'mm'. */
894 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
895 unsigned long start
, int len
, int write
, int force
,
896 struct page
**pages
, struct vm_area_struct
**vmas
)
902 * Require read or write permissions.
903 * If 'force' is set, we only require the "MAY" flags.
905 flags
= write
? (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
906 flags
&= force
? (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
910 struct vm_area_struct
* vma
;
912 vma
= find_extend_vma(mm
, start
);
913 if (!vma
&& in_gate_area(tsk
, start
)) {
914 unsigned long pg
= start
& PAGE_MASK
;
915 struct vm_area_struct
*gate_vma
= get_gate_vma(tsk
);
920 if (write
) /* user gate pages are read-only */
921 return i
? : -EFAULT
;
923 pgd
= pgd_offset_k(pg
);
925 pgd
= pgd_offset_gate(mm
, pg
);
926 BUG_ON(pgd_none(*pgd
));
927 pud
= pud_offset(pgd
, pg
);
928 BUG_ON(pud_none(*pud
));
929 pmd
= pmd_offset(pud
, pg
);
931 return i
? : -EFAULT
;
932 pte
= pte_offset_map(pmd
, pg
);
933 if (pte_none(*pte
)) {
935 return i
? : -EFAULT
;
938 pages
[i
] = pte_page(*pte
);
950 if (!vma
|| (vma
->vm_flags
& VM_IO
)
951 || !(flags
& vma
->vm_flags
))
952 return i
? : -EFAULT
;
954 if (is_vm_hugetlb_page(vma
)) {
955 i
= follow_hugetlb_page(mm
, vma
, pages
, vmas
,
959 spin_lock(&mm
->page_table_lock
);
961 int write_access
= write
;
964 cond_resched_lock(&mm
->page_table_lock
);
965 while (!(page
= follow_page(mm
, start
, write_access
))) {
969 * Shortcut for anonymous pages. We don't want
970 * to force the creation of pages tables for
971 * insanely big anonymously mapped areas that
972 * nobody touched so far. This is important
973 * for doing a core dump for these mappings.
975 if (!write
&& untouched_anonymous_page(mm
,vma
,start
)) {
976 page
= ZERO_PAGE(start
);
979 spin_unlock(&mm
->page_table_lock
);
980 ret
= __handle_mm_fault(mm
, vma
, start
, write_access
);
983 * The VM_FAULT_WRITE bit tells us that do_wp_page has
984 * broken COW when necessary, even if maybe_mkwrite
985 * decided not to set pte_write. We can thus safely do
986 * subsequent page lookups as if they were reads.
988 if (ret
& VM_FAULT_WRITE
)
991 switch (ret
& ~VM_FAULT_WRITE
) {
998 case VM_FAULT_SIGBUS
:
999 return i
? i
: -EFAULT
;
1001 return i
? i
: -ENOMEM
;
1005 spin_lock(&mm
->page_table_lock
);
1009 flush_dcache_page(page
);
1010 if (!PageReserved(page
))
1011 page_cache_get(page
);
1018 } while (len
&& start
< vma
->vm_end
);
1019 spin_unlock(&mm
->page_table_lock
);
1023 EXPORT_SYMBOL(get_user_pages
);
1025 static int zeromap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1026 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1030 pte
= pte_alloc_map(mm
, pmd
, addr
);
1034 pte_t zero_pte
= pte_wrprotect(mk_pte(ZERO_PAGE(addr
), prot
));
1035 BUG_ON(!pte_none(*pte
));
1036 set_pte_at(mm
, addr
, pte
, zero_pte
);
1037 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1042 static inline int zeromap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1043 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1048 pmd
= pmd_alloc(mm
, pud
, addr
);
1052 next
= pmd_addr_end(addr
, end
);
1053 if (zeromap_pte_range(mm
, pmd
, addr
, next
, prot
))
1055 } while (pmd
++, addr
= next
, addr
!= end
);
1059 static inline int zeromap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1060 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1065 pud
= pud_alloc(mm
, pgd
, addr
);
1069 next
= pud_addr_end(addr
, end
);
1070 if (zeromap_pmd_range(mm
, pud
, addr
, next
, prot
))
1072 } while (pud
++, addr
= next
, addr
!= end
);
1076 int zeromap_page_range(struct vm_area_struct
*vma
,
1077 unsigned long addr
, unsigned long size
, pgprot_t prot
)
1081 unsigned long end
= addr
+ size
;
1082 struct mm_struct
*mm
= vma
->vm_mm
;
1085 BUG_ON(addr
>= end
);
1086 pgd
= pgd_offset(mm
, addr
);
1087 flush_cache_range(vma
, addr
, end
);
1088 spin_lock(&mm
->page_table_lock
);
1090 next
= pgd_addr_end(addr
, end
);
1091 err
= zeromap_pud_range(mm
, pgd
, addr
, next
, prot
);
1094 } while (pgd
++, addr
= next
, addr
!= end
);
1095 spin_unlock(&mm
->page_table_lock
);
1100 * maps a range of physical memory into the requested pages. the old
1101 * mappings are removed. any references to nonexistent pages results
1102 * in null mappings (currently treated as "copy-on-access")
1104 static int remap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1105 unsigned long addr
, unsigned long end
,
1106 unsigned long pfn
, pgprot_t prot
)
1110 pte
= pte_alloc_map(mm
, pmd
, addr
);
1114 BUG_ON(!pte_none(*pte
));
1115 if (!pfn_valid(pfn
) || PageReserved(pfn_to_page(pfn
)))
1116 set_pte_at(mm
, addr
, pte
, pfn_pte(pfn
, prot
));
1118 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1123 static inline int remap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1124 unsigned long addr
, unsigned long end
,
1125 unsigned long pfn
, pgprot_t prot
)
1130 pfn
-= addr
>> PAGE_SHIFT
;
1131 pmd
= pmd_alloc(mm
, pud
, addr
);
1135 next
= pmd_addr_end(addr
, end
);
1136 if (remap_pte_range(mm
, pmd
, addr
, next
,
1137 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1139 } while (pmd
++, addr
= next
, addr
!= end
);
1143 static inline int remap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1144 unsigned long addr
, unsigned long end
,
1145 unsigned long pfn
, pgprot_t prot
)
1150 pfn
-= addr
>> PAGE_SHIFT
;
1151 pud
= pud_alloc(mm
, pgd
, addr
);
1155 next
= pud_addr_end(addr
, end
);
1156 if (remap_pmd_range(mm
, pud
, addr
, next
,
1157 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1159 } while (pud
++, addr
= next
, addr
!= end
);
1163 /* Note: this is only safe if the mm semaphore is held when called. */
1164 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1165 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1169 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1170 struct mm_struct
*mm
= vma
->vm_mm
;
1174 * Physically remapped pages are special. Tell the
1175 * rest of the world about it:
1176 * VM_IO tells people not to look at these pages
1177 * (accesses can have side effects).
1178 * VM_RESERVED tells swapout not to try to touch
1181 vma
->vm_flags
|= VM_IO
| VM_RESERVED
;
1183 BUG_ON(addr
>= end
);
1184 pfn
-= addr
>> PAGE_SHIFT
;
1185 pgd
= pgd_offset(mm
, addr
);
1186 flush_cache_range(vma
, addr
, end
);
1187 spin_lock(&mm
->page_table_lock
);
1189 next
= pgd_addr_end(addr
, end
);
1190 err
= remap_pud_range(mm
, pgd
, addr
, next
,
1191 pfn
+ (addr
>> PAGE_SHIFT
), prot
);
1194 } while (pgd
++, addr
= next
, addr
!= end
);
1195 spin_unlock(&mm
->page_table_lock
);
1198 EXPORT_SYMBOL(remap_pfn_range
);
1201 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1202 * servicing faults for write access. In the normal case, do always want
1203 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1204 * that do not have writing enabled, when used by access_process_vm.
1206 static inline pte_t
maybe_mkwrite(pte_t pte
, struct vm_area_struct
*vma
)
1208 if (likely(vma
->vm_flags
& VM_WRITE
))
1209 pte
= pte_mkwrite(pte
);
1214 * We hold the mm semaphore for reading and vma->vm_mm->page_table_lock
1216 static inline void break_cow(struct vm_area_struct
* vma
, struct page
* new_page
, unsigned long address
,
1221 entry
= maybe_mkwrite(pte_mkdirty(mk_pte(new_page
, vma
->vm_page_prot
)),
1223 ptep_establish(vma
, address
, page_table
, entry
);
1224 update_mmu_cache(vma
, address
, entry
);
1225 lazy_mmu_prot_update(entry
);
1229 * This routine handles present pages, when users try to write
1230 * to a shared page. It is done by copying the page to a new address
1231 * and decrementing the shared-page counter for the old page.
1233 * Goto-purists beware: the only reason for goto's here is that it results
1234 * in better assembly code.. The "default" path will see no jumps at all.
1236 * Note that this routine assumes that the protection checks have been
1237 * done by the caller (the low-level page fault routine in most cases).
1238 * Thus we can safely just mark it writable once we've done any necessary
1241 * We also mark the page dirty at this point even though the page will
1242 * change only once the write actually happens. This avoids a few races,
1243 * and potentially makes it more efficient.
1245 * We hold the mm semaphore and the page_table_lock on entry and exit
1246 * with the page_table_lock released.
1248 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
* vma
,
1249 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
, pte_t pte
)
1251 struct page
*old_page
, *new_page
;
1252 unsigned long pfn
= pte_pfn(pte
);
1256 if (unlikely(!pfn_valid(pfn
))) {
1258 * This should really halt the system so it can be debugged or
1259 * at least the kernel stops what it's doing before it corrupts
1260 * data, but for the moment just pretend this is OOM.
1262 pte_unmap(page_table
);
1263 printk(KERN_ERR
"do_wp_page: bogus page at address %08lx\n",
1265 spin_unlock(&mm
->page_table_lock
);
1266 return VM_FAULT_OOM
;
1268 old_page
= pfn_to_page(pfn
);
1270 if (PageAnon(old_page
) && !TestSetPageLocked(old_page
)) {
1271 int reuse
= can_share_swap_page(old_page
);
1272 unlock_page(old_page
);
1274 flush_cache_page(vma
, address
, pfn
);
1275 entry
= maybe_mkwrite(pte_mkyoung(pte_mkdirty(pte
)),
1277 ptep_set_access_flags(vma
, address
, page_table
, entry
, 1);
1278 update_mmu_cache(vma
, address
, entry
);
1279 lazy_mmu_prot_update(entry
);
1280 pte_unmap(page_table
);
1281 spin_unlock(&mm
->page_table_lock
);
1282 return VM_FAULT_MINOR
|VM_FAULT_WRITE
;
1285 pte_unmap(page_table
);
1288 * Ok, we need to copy. Oh, well..
1290 if (!PageReserved(old_page
))
1291 page_cache_get(old_page
);
1292 spin_unlock(&mm
->page_table_lock
);
1294 if (unlikely(anon_vma_prepare(vma
)))
1296 if (old_page
== ZERO_PAGE(address
)) {
1297 new_page
= alloc_zeroed_user_highpage(vma
, address
);
1301 new_page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
1304 copy_user_highpage(new_page
, old_page
, address
);
1307 * Re-check the pte - we dropped the lock
1309 ret
= VM_FAULT_MINOR
;
1310 spin_lock(&mm
->page_table_lock
);
1311 page_table
= pte_offset_map(pmd
, address
);
1312 if (likely(pte_same(*page_table
, pte
))) {
1313 if (PageAnon(old_page
))
1314 dec_mm_counter(mm
, anon_rss
);
1315 if (PageReserved(old_page
))
1316 inc_mm_counter(mm
, rss
);
1318 page_remove_rmap(old_page
);
1319 flush_cache_page(vma
, address
, pfn
);
1320 break_cow(vma
, new_page
, address
, page_table
);
1321 lru_cache_add_active(new_page
);
1322 page_add_anon_rmap(new_page
, vma
, address
);
1324 /* Free the old page.. */
1325 new_page
= old_page
;
1326 ret
|= VM_FAULT_WRITE
;
1328 pte_unmap(page_table
);
1329 page_cache_release(new_page
);
1330 page_cache_release(old_page
);
1331 spin_unlock(&mm
->page_table_lock
);
1335 page_cache_release(old_page
);
1336 return VM_FAULT_OOM
;
1340 * Helper functions for unmap_mapping_range().
1342 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1344 * We have to restart searching the prio_tree whenever we drop the lock,
1345 * since the iterator is only valid while the lock is held, and anyway
1346 * a later vma might be split and reinserted earlier while lock dropped.
1348 * The list of nonlinear vmas could be handled more efficiently, using
1349 * a placeholder, but handle it in the same way until a need is shown.
1350 * It is important to search the prio_tree before nonlinear list: a vma
1351 * may become nonlinear and be shifted from prio_tree to nonlinear list
1352 * while the lock is dropped; but never shifted from list to prio_tree.
1354 * In order to make forward progress despite restarting the search,
1355 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1356 * quickly skip it next time around. Since the prio_tree search only
1357 * shows us those vmas affected by unmapping the range in question, we
1358 * can't efficiently keep all vmas in step with mapping->truncate_count:
1359 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1360 * mapping->truncate_count and vma->vm_truncate_count are protected by
1363 * In order to make forward progress despite repeatedly restarting some
1364 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1365 * and restart from that address when we reach that vma again. It might
1366 * have been split or merged, shrunk or extended, but never shifted: so
1367 * restart_addr remains valid so long as it remains in the vma's range.
1368 * unmap_mapping_range forces truncate_count to leap over page-aligned
1369 * values so we can save vma's restart_addr in its truncate_count field.
1371 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1373 static void reset_vma_truncate_counts(struct address_space
*mapping
)
1375 struct vm_area_struct
*vma
;
1376 struct prio_tree_iter iter
;
1378 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, 0, ULONG_MAX
)
1379 vma
->vm_truncate_count
= 0;
1380 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
, shared
.vm_set
.list
)
1381 vma
->vm_truncate_count
= 0;
1384 static int unmap_mapping_range_vma(struct vm_area_struct
*vma
,
1385 unsigned long start_addr
, unsigned long end_addr
,
1386 struct zap_details
*details
)
1388 unsigned long restart_addr
;
1392 restart_addr
= vma
->vm_truncate_count
;
1393 if (is_restart_addr(restart_addr
) && start_addr
< restart_addr
) {
1394 start_addr
= restart_addr
;
1395 if (start_addr
>= end_addr
) {
1396 /* Top of vma has been split off since last time */
1397 vma
->vm_truncate_count
= details
->truncate_count
;
1402 restart_addr
= zap_page_range(vma
, start_addr
,
1403 end_addr
- start_addr
, details
);
1406 * We cannot rely on the break test in unmap_vmas:
1407 * on the one hand, we don't want to restart our loop
1408 * just because that broke out for the page_table_lock;
1409 * on the other hand, it does no test when vma is small.
1411 need_break
= need_resched() ||
1412 need_lockbreak(details
->i_mmap_lock
);
1414 if (restart_addr
>= end_addr
) {
1415 /* We have now completed this vma: mark it so */
1416 vma
->vm_truncate_count
= details
->truncate_count
;
1420 /* Note restart_addr in vma's truncate_count field */
1421 vma
->vm_truncate_count
= restart_addr
;
1426 spin_unlock(details
->i_mmap_lock
);
1428 spin_lock(details
->i_mmap_lock
);
1432 static inline void unmap_mapping_range_tree(struct prio_tree_root
*root
,
1433 struct zap_details
*details
)
1435 struct vm_area_struct
*vma
;
1436 struct prio_tree_iter iter
;
1437 pgoff_t vba
, vea
, zba
, zea
;
1440 vma_prio_tree_foreach(vma
, &iter
, root
,
1441 details
->first_index
, details
->last_index
) {
1442 /* Skip quickly over those we have already dealt with */
1443 if (vma
->vm_truncate_count
== details
->truncate_count
)
1446 vba
= vma
->vm_pgoff
;
1447 vea
= vba
+ ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
) - 1;
1448 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1449 zba
= details
->first_index
;
1452 zea
= details
->last_index
;
1456 if (unmap_mapping_range_vma(vma
,
1457 ((zba
- vba
) << PAGE_SHIFT
) + vma
->vm_start
,
1458 ((zea
- vba
+ 1) << PAGE_SHIFT
) + vma
->vm_start
,
1464 static inline void unmap_mapping_range_list(struct list_head
*head
,
1465 struct zap_details
*details
)
1467 struct vm_area_struct
*vma
;
1470 * In nonlinear VMAs there is no correspondence between virtual address
1471 * offset and file offset. So we must perform an exhaustive search
1472 * across *all* the pages in each nonlinear VMA, not just the pages
1473 * whose virtual address lies outside the file truncation point.
1476 list_for_each_entry(vma
, head
, shared
.vm_set
.list
) {
1477 /* Skip quickly over those we have already dealt with */
1478 if (vma
->vm_truncate_count
== details
->truncate_count
)
1480 details
->nonlinear_vma
= vma
;
1481 if (unmap_mapping_range_vma(vma
, vma
->vm_start
,
1482 vma
->vm_end
, details
) < 0)
1488 * unmap_mapping_range - unmap the portion of all mmaps
1489 * in the specified address_space corresponding to the specified
1490 * page range in the underlying file.
1491 * @mapping: the address space containing mmaps to be unmapped.
1492 * @holebegin: byte in first page to unmap, relative to the start of
1493 * the underlying file. This will be rounded down to a PAGE_SIZE
1494 * boundary. Note that this is different from vmtruncate(), which
1495 * must keep the partial page. In contrast, we must get rid of
1497 * @holelen: size of prospective hole in bytes. This will be rounded
1498 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1500 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1501 * but 0 when invalidating pagecache, don't throw away private data.
1503 void unmap_mapping_range(struct address_space
*mapping
,
1504 loff_t
const holebegin
, loff_t
const holelen
, int even_cows
)
1506 struct zap_details details
;
1507 pgoff_t hba
= holebegin
>> PAGE_SHIFT
;
1508 pgoff_t hlen
= (holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1510 /* Check for overflow. */
1511 if (sizeof(holelen
) > sizeof(hlen
)) {
1513 (holebegin
+ holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1514 if (holeend
& ~(long long)ULONG_MAX
)
1515 hlen
= ULONG_MAX
- hba
+ 1;
1518 details
.check_mapping
= even_cows
? NULL
: mapping
;
1519 details
.nonlinear_vma
= NULL
;
1520 details
.first_index
= hba
;
1521 details
.last_index
= hba
+ hlen
- 1;
1522 if (details
.last_index
< details
.first_index
)
1523 details
.last_index
= ULONG_MAX
;
1524 details
.i_mmap_lock
= &mapping
->i_mmap_lock
;
1526 spin_lock(&mapping
->i_mmap_lock
);
1528 /* serialize i_size write against truncate_count write */
1530 /* Protect against page faults, and endless unmapping loops */
1531 mapping
->truncate_count
++;
1533 * For archs where spin_lock has inclusive semantics like ia64
1534 * this smp_mb() will prevent to read pagetable contents
1535 * before the truncate_count increment is visible to
1539 if (unlikely(is_restart_addr(mapping
->truncate_count
))) {
1540 if (mapping
->truncate_count
== 0)
1541 reset_vma_truncate_counts(mapping
);
1542 mapping
->truncate_count
++;
1544 details
.truncate_count
= mapping
->truncate_count
;
1546 if (unlikely(!prio_tree_empty(&mapping
->i_mmap
)))
1547 unmap_mapping_range_tree(&mapping
->i_mmap
, &details
);
1548 if (unlikely(!list_empty(&mapping
->i_mmap_nonlinear
)))
1549 unmap_mapping_range_list(&mapping
->i_mmap_nonlinear
, &details
);
1550 spin_unlock(&mapping
->i_mmap_lock
);
1552 EXPORT_SYMBOL(unmap_mapping_range
);
1555 * Handle all mappings that got truncated by a "truncate()"
1558 * NOTE! We have to be ready to update the memory sharing
1559 * between the file and the memory map for a potential last
1560 * incomplete page. Ugly, but necessary.
1562 int vmtruncate(struct inode
* inode
, loff_t offset
)
1564 struct address_space
*mapping
= inode
->i_mapping
;
1565 unsigned long limit
;
1567 if (inode
->i_size
< offset
)
1570 * truncation of in-use swapfiles is disallowed - it would cause
1571 * subsequent swapout to scribble on the now-freed blocks.
1573 if (IS_SWAPFILE(inode
))
1575 i_size_write(inode
, offset
);
1576 unmap_mapping_range(mapping
, offset
+ PAGE_SIZE
- 1, 0, 1);
1577 truncate_inode_pages(mapping
, offset
);
1581 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1582 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
1584 if (offset
> inode
->i_sb
->s_maxbytes
)
1586 i_size_write(inode
, offset
);
1589 if (inode
->i_op
&& inode
->i_op
->truncate
)
1590 inode
->i_op
->truncate(inode
);
1593 send_sig(SIGXFSZ
, current
, 0);
1600 EXPORT_SYMBOL(vmtruncate
);
1603 * Primitive swap readahead code. We simply read an aligned block of
1604 * (1 << page_cluster) entries in the swap area. This method is chosen
1605 * because it doesn't cost us any seek time. We also make sure to queue
1606 * the 'original' request together with the readahead ones...
1608 * This has been extended to use the NUMA policies from the mm triggering
1611 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1613 void swapin_readahead(swp_entry_t entry
, unsigned long addr
,struct vm_area_struct
*vma
)
1616 struct vm_area_struct
*next_vma
= vma
? vma
->vm_next
: NULL
;
1619 struct page
*new_page
;
1620 unsigned long offset
;
1623 * Get the number of handles we should do readahead io to.
1625 num
= valid_swaphandles(entry
, &offset
);
1626 for (i
= 0; i
< num
; offset
++, i
++) {
1627 /* Ok, do the async read-ahead now */
1628 new_page
= read_swap_cache_async(swp_entry(swp_type(entry
),
1629 offset
), vma
, addr
);
1632 page_cache_release(new_page
);
1635 * Find the next applicable VMA for the NUMA policy.
1641 if (addr
>= vma
->vm_end
) {
1643 next_vma
= vma
? vma
->vm_next
: NULL
;
1645 if (vma
&& addr
< vma
->vm_start
)
1648 if (next_vma
&& addr
>= next_vma
->vm_start
) {
1650 next_vma
= vma
->vm_next
;
1655 lru_add_drain(); /* Push any new pages onto the LRU now */
1659 * We hold the mm semaphore and the page_table_lock on entry and
1660 * should release the pagetable lock on exit..
1662 static int do_swap_page(struct mm_struct
* mm
,
1663 struct vm_area_struct
* vma
, unsigned long address
,
1664 pte_t
*page_table
, pmd_t
*pmd
, pte_t orig_pte
, int write_access
)
1667 swp_entry_t entry
= pte_to_swp_entry(orig_pte
);
1669 int ret
= VM_FAULT_MINOR
;
1671 pte_unmap(page_table
);
1672 spin_unlock(&mm
->page_table_lock
);
1673 page
= lookup_swap_cache(entry
);
1675 swapin_readahead(entry
, address
, vma
);
1676 page
= read_swap_cache_async(entry
, vma
, address
);
1679 * Back out if somebody else faulted in this pte while
1680 * we released the page table lock.
1682 spin_lock(&mm
->page_table_lock
);
1683 page_table
= pte_offset_map(pmd
, address
);
1684 if (likely(pte_same(*page_table
, orig_pte
)))
1687 ret
= VM_FAULT_MINOR
;
1688 pte_unmap(page_table
);
1689 spin_unlock(&mm
->page_table_lock
);
1693 /* Had to read the page from swap area: Major fault */
1694 ret
= VM_FAULT_MAJOR
;
1695 inc_page_state(pgmajfault
);
1699 mark_page_accessed(page
);
1703 * Back out if somebody else faulted in this pte while we
1704 * released the page table lock.
1706 spin_lock(&mm
->page_table_lock
);
1707 page_table
= pte_offset_map(pmd
, address
);
1708 if (unlikely(!pte_same(*page_table
, orig_pte
))) {
1709 ret
= VM_FAULT_MINOR
;
1713 if (unlikely(!PageUptodate(page
))) {
1714 ret
= VM_FAULT_SIGBUS
;
1718 /* The page isn't present yet, go ahead with the fault. */
1720 inc_mm_counter(mm
, rss
);
1721 pte
= mk_pte(page
, vma
->vm_page_prot
);
1722 if (write_access
&& can_share_swap_page(page
)) {
1723 pte
= maybe_mkwrite(pte_mkdirty(pte
), vma
);
1727 flush_icache_page(vma
, page
);
1728 set_pte_at(mm
, address
, page_table
, pte
);
1729 page_add_anon_rmap(page
, vma
, address
);
1733 remove_exclusive_swap_page(page
);
1737 if (do_wp_page(mm
, vma
, address
,
1738 page_table
, pmd
, pte
) == VM_FAULT_OOM
)
1743 /* No need to invalidate - it was non-present before */
1744 update_mmu_cache(vma
, address
, pte
);
1745 lazy_mmu_prot_update(pte
);
1746 pte_unmap(page_table
);
1747 spin_unlock(&mm
->page_table_lock
);
1751 pte_unmap(page_table
);
1752 spin_unlock(&mm
->page_table_lock
);
1754 page_cache_release(page
);
1759 * We are called with the MM semaphore and page_table_lock
1760 * spinlock held to protect against concurrent faults in
1761 * multithreaded programs.
1764 do_anonymous_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1765 pte_t
*page_table
, pmd_t
*pmd
, int write_access
,
1769 struct page
* page
= ZERO_PAGE(addr
);
1771 /* Read-only mapping of ZERO_PAGE. */
1772 entry
= pte_wrprotect(mk_pte(ZERO_PAGE(addr
), vma
->vm_page_prot
));
1774 /* ..except if it's a write access */
1776 /* Allocate our own private page. */
1777 pte_unmap(page_table
);
1778 spin_unlock(&mm
->page_table_lock
);
1780 if (unlikely(anon_vma_prepare(vma
)))
1782 page
= alloc_zeroed_user_highpage(vma
, addr
);
1786 spin_lock(&mm
->page_table_lock
);
1787 page_table
= pte_offset_map(pmd
, addr
);
1789 if (!pte_none(*page_table
)) {
1790 pte_unmap(page_table
);
1791 page_cache_release(page
);
1792 spin_unlock(&mm
->page_table_lock
);
1795 inc_mm_counter(mm
, rss
);
1796 entry
= maybe_mkwrite(pte_mkdirty(mk_pte(page
,
1797 vma
->vm_page_prot
)),
1799 lru_cache_add_active(page
);
1800 SetPageReferenced(page
);
1801 page_add_anon_rmap(page
, vma
, addr
);
1804 set_pte_at(mm
, addr
, page_table
, entry
);
1805 pte_unmap(page_table
);
1807 /* No need to invalidate - it was non-present before */
1808 update_mmu_cache(vma
, addr
, entry
);
1809 lazy_mmu_prot_update(entry
);
1810 spin_unlock(&mm
->page_table_lock
);
1812 return VM_FAULT_MINOR
;
1814 return VM_FAULT_OOM
;
1818 * do_no_page() tries to create a new page mapping. It aggressively
1819 * tries to share with existing pages, but makes a separate copy if
1820 * the "write_access" parameter is true in order to avoid the next
1823 * As this is called only for pages that do not currently exist, we
1824 * do not need to flush old virtual caches or the TLB.
1826 * This is called with the MM semaphore held and the page table
1827 * spinlock held. Exit with the spinlock released.
1830 do_no_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1831 unsigned long address
, int write_access
, pte_t
*page_table
, pmd_t
*pmd
)
1833 struct page
* new_page
;
1834 struct address_space
*mapping
= NULL
;
1836 unsigned int sequence
= 0;
1837 int ret
= VM_FAULT_MINOR
;
1840 if (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
)
1841 return do_anonymous_page(mm
, vma
, page_table
,
1842 pmd
, write_access
, address
);
1843 pte_unmap(page_table
);
1844 spin_unlock(&mm
->page_table_lock
);
1847 mapping
= vma
->vm_file
->f_mapping
;
1848 sequence
= mapping
->truncate_count
;
1849 smp_rmb(); /* serializes i_size against truncate_count */
1853 new_page
= vma
->vm_ops
->nopage(vma
, address
& PAGE_MASK
, &ret
);
1855 * No smp_rmb is needed here as long as there's a full
1856 * spin_lock/unlock sequence inside the ->nopage callback
1857 * (for the pagecache lookup) that acts as an implicit
1858 * smp_mb() and prevents the i_size read to happen
1859 * after the next truncate_count read.
1862 /* no page was available -- either SIGBUS or OOM */
1863 if (new_page
== NOPAGE_SIGBUS
)
1864 return VM_FAULT_SIGBUS
;
1865 if (new_page
== NOPAGE_OOM
)
1866 return VM_FAULT_OOM
;
1869 * Should we do an early C-O-W break?
1871 if (write_access
&& !(vma
->vm_flags
& VM_SHARED
)) {
1874 if (unlikely(anon_vma_prepare(vma
)))
1876 page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
1879 copy_user_highpage(page
, new_page
, address
);
1880 page_cache_release(new_page
);
1885 spin_lock(&mm
->page_table_lock
);
1887 * For a file-backed vma, someone could have truncated or otherwise
1888 * invalidated this page. If unmap_mapping_range got called,
1889 * retry getting the page.
1891 if (mapping
&& unlikely(sequence
!= mapping
->truncate_count
)) {
1892 sequence
= mapping
->truncate_count
;
1893 spin_unlock(&mm
->page_table_lock
);
1894 page_cache_release(new_page
);
1897 page_table
= pte_offset_map(pmd
, address
);
1900 * This silly early PAGE_DIRTY setting removes a race
1901 * due to the bad i386 page protection. But it's valid
1902 * for other architectures too.
1904 * Note that if write_access is true, we either now have
1905 * an exclusive copy of the page, or this is a shared mapping,
1906 * so we can make it writable and dirty to avoid having to
1907 * handle that later.
1909 /* Only go through if we didn't race with anybody else... */
1910 if (pte_none(*page_table
)) {
1911 if (!PageReserved(new_page
))
1912 inc_mm_counter(mm
, rss
);
1914 flush_icache_page(vma
, new_page
);
1915 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
1917 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1918 set_pte_at(mm
, address
, page_table
, entry
);
1920 lru_cache_add_active(new_page
);
1921 page_add_anon_rmap(new_page
, vma
, address
);
1923 page_add_file_rmap(new_page
);
1924 pte_unmap(page_table
);
1926 /* One of our sibling threads was faster, back out. */
1927 pte_unmap(page_table
);
1928 page_cache_release(new_page
);
1929 spin_unlock(&mm
->page_table_lock
);
1933 /* no need to invalidate: a not-present page shouldn't be cached */
1934 update_mmu_cache(vma
, address
, entry
);
1935 lazy_mmu_prot_update(entry
);
1936 spin_unlock(&mm
->page_table_lock
);
1940 page_cache_release(new_page
);
1946 * Fault of a previously existing named mapping. Repopulate the pte
1947 * from the encoded file_pte if possible. This enables swappable
1950 static int do_file_page(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1951 unsigned long address
, int write_access
, pte_t
*pte
, pmd_t
*pmd
)
1953 unsigned long pgoff
;
1956 BUG_ON(!vma
->vm_ops
|| !vma
->vm_ops
->nopage
);
1958 * Fall back to the linear mapping if the fs does not support
1961 if (!vma
->vm_ops
->populate
||
1962 (write_access
&& !(vma
->vm_flags
& VM_SHARED
))) {
1963 pte_clear(mm
, address
, pte
);
1964 return do_no_page(mm
, vma
, address
, write_access
, pte
, pmd
);
1967 pgoff
= pte_to_pgoff(*pte
);
1970 spin_unlock(&mm
->page_table_lock
);
1972 err
= vma
->vm_ops
->populate(vma
, address
& PAGE_MASK
, PAGE_SIZE
, vma
->vm_page_prot
, pgoff
, 0);
1974 return VM_FAULT_OOM
;
1976 return VM_FAULT_SIGBUS
;
1977 return VM_FAULT_MAJOR
;
1981 * These routines also need to handle stuff like marking pages dirty
1982 * and/or accessed for architectures that don't do it in hardware (most
1983 * RISC architectures). The early dirtying is also good on the i386.
1985 * There is also a hook called "update_mmu_cache()" that architectures
1986 * with external mmu caches can use to update those (ie the Sparc or
1987 * PowerPC hashed page tables that act as extended TLBs).
1989 * Note the "page_table_lock". It is to protect against kswapd removing
1990 * pages from under us. Note that kswapd only ever _removes_ pages, never
1991 * adds them. As such, once we have noticed that the page is not present,
1992 * we can drop the lock early.
1994 * The adding of pages is protected by the MM semaphore (which we hold),
1995 * so we don't need to worry about a page being suddenly been added into
1998 * We enter with the pagetable spinlock held, we are supposed to
1999 * release it when done.
2001 static inline int handle_pte_fault(struct mm_struct
*mm
,
2002 struct vm_area_struct
* vma
, unsigned long address
,
2003 int write_access
, pte_t
*pte
, pmd_t
*pmd
)
2008 if (!pte_present(entry
)) {
2010 * If it truly wasn't present, we know that kswapd
2011 * and the PTE updates will not touch it later. So
2014 if (pte_none(entry
))
2015 return do_no_page(mm
, vma
, address
, write_access
, pte
, pmd
);
2016 if (pte_file(entry
))
2017 return do_file_page(mm
, vma
, address
, write_access
, pte
, pmd
);
2018 return do_swap_page(mm
, vma
, address
, pte
, pmd
, entry
, write_access
);
2022 if (!pte_write(entry
))
2023 return do_wp_page(mm
, vma
, address
, pte
, pmd
, entry
);
2024 entry
= pte_mkdirty(entry
);
2026 entry
= pte_mkyoung(entry
);
2027 ptep_set_access_flags(vma
, address
, pte
, entry
, write_access
);
2028 update_mmu_cache(vma
, address
, entry
);
2029 lazy_mmu_prot_update(entry
);
2031 spin_unlock(&mm
->page_table_lock
);
2032 return VM_FAULT_MINOR
;
2036 * By the time we get here, we already hold the mm semaphore
2038 int __handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
* vma
,
2039 unsigned long address
, int write_access
)
2046 __set_current_state(TASK_RUNNING
);
2048 inc_page_state(pgfault
);
2050 if (unlikely(is_vm_hugetlb_page(vma
)))
2051 return hugetlb_fault(mm
, vma
, address
, write_access
);
2054 * We need the page table lock to synchronize with kswapd
2055 * and the SMP-safe atomic PTE updates.
2057 pgd
= pgd_offset(mm
, address
);
2058 spin_lock(&mm
->page_table_lock
);
2060 pud
= pud_alloc(mm
, pgd
, address
);
2064 pmd
= pmd_alloc(mm
, pud
, address
);
2068 pte
= pte_alloc_map(mm
, pmd
, address
);
2072 return handle_pte_fault(mm
, vma
, address
, write_access
, pte
, pmd
);
2075 spin_unlock(&mm
->page_table_lock
);
2076 return VM_FAULT_OOM
;
2079 #ifndef __PAGETABLE_PUD_FOLDED
2081 * Allocate page upper directory.
2083 * We've already handled the fast-path in-line, and we own the
2086 pud_t fastcall
*__pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
2090 spin_unlock(&mm
->page_table_lock
);
2091 new = pud_alloc_one(mm
, address
);
2092 spin_lock(&mm
->page_table_lock
);
2097 * Because we dropped the lock, we should re-check the
2098 * entry, as somebody else could have populated it..
2100 if (pgd_present(*pgd
)) {
2104 pgd_populate(mm
, pgd
, new);
2106 return pud_offset(pgd
, address
);
2108 #endif /* __PAGETABLE_PUD_FOLDED */
2110 #ifndef __PAGETABLE_PMD_FOLDED
2112 * Allocate page middle directory.
2114 * We've already handled the fast-path in-line, and we own the
2117 pmd_t fastcall
*__pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
2121 spin_unlock(&mm
->page_table_lock
);
2122 new = pmd_alloc_one(mm
, address
);
2123 spin_lock(&mm
->page_table_lock
);
2128 * Because we dropped the lock, we should re-check the
2129 * entry, as somebody else could have populated it..
2131 #ifndef __ARCH_HAS_4LEVEL_HACK
2132 if (pud_present(*pud
)) {
2136 pud_populate(mm
, pud
, new);
2138 if (pgd_present(*pud
)) {
2142 pgd_populate(mm
, pud
, new);
2143 #endif /* __ARCH_HAS_4LEVEL_HACK */
2146 return pmd_offset(pud
, address
);
2148 #endif /* __PAGETABLE_PMD_FOLDED */
2150 int make_pages_present(unsigned long addr
, unsigned long end
)
2152 int ret
, len
, write
;
2153 struct vm_area_struct
* vma
;
2155 vma
= find_vma(current
->mm
, addr
);
2158 write
= (vma
->vm_flags
& VM_WRITE
) != 0;
2161 if (end
> vma
->vm_end
)
2163 len
= (end
+PAGE_SIZE
-1)/PAGE_SIZE
-addr
/PAGE_SIZE
;
2164 ret
= get_user_pages(current
, current
->mm
, addr
,
2165 len
, write
, 0, NULL
, NULL
);
2168 return ret
== len
? 0 : -1;
2172 * Map a vmalloc()-space virtual address to the physical page.
2174 struct page
* vmalloc_to_page(void * vmalloc_addr
)
2176 unsigned long addr
= (unsigned long) vmalloc_addr
;
2177 struct page
*page
= NULL
;
2178 pgd_t
*pgd
= pgd_offset_k(addr
);
2183 if (!pgd_none(*pgd
)) {
2184 pud
= pud_offset(pgd
, addr
);
2185 if (!pud_none(*pud
)) {
2186 pmd
= pmd_offset(pud
, addr
);
2187 if (!pmd_none(*pmd
)) {
2188 ptep
= pte_offset_map(pmd
, addr
);
2190 if (pte_present(pte
))
2191 page
= pte_page(pte
);
2199 EXPORT_SYMBOL(vmalloc_to_page
);
2202 * Map a vmalloc()-space virtual address to the physical page frame number.
2204 unsigned long vmalloc_to_pfn(void * vmalloc_addr
)
2206 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
2209 EXPORT_SYMBOL(vmalloc_to_pfn
);
2212 * update_mem_hiwater
2213 * - update per process rss and vm high water data
2215 void update_mem_hiwater(struct task_struct
*tsk
)
2218 unsigned long rss
= get_mm_counter(tsk
->mm
, rss
);
2220 if (tsk
->mm
->hiwater_rss
< rss
)
2221 tsk
->mm
->hiwater_rss
= rss
;
2222 if (tsk
->mm
->hiwater_vm
< tsk
->mm
->total_vm
)
2223 tsk
->mm
->hiwater_vm
= tsk
->mm
->total_vm
;
2227 #if !defined(__HAVE_ARCH_GATE_AREA)
2229 #if defined(AT_SYSINFO_EHDR)
2230 static struct vm_area_struct gate_vma
;
2232 static int __init
gate_vma_init(void)
2234 gate_vma
.vm_mm
= NULL
;
2235 gate_vma
.vm_start
= FIXADDR_USER_START
;
2236 gate_vma
.vm_end
= FIXADDR_USER_END
;
2237 gate_vma
.vm_page_prot
= PAGE_READONLY
;
2238 gate_vma
.vm_flags
= 0;
2241 __initcall(gate_vma_init
);
2244 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
2246 #ifdef AT_SYSINFO_EHDR
2253 int in_gate_area_no_task(unsigned long addr
)
2255 #ifdef AT_SYSINFO_EHDR
2256 if ((addr
>= FIXADDR_USER_START
) && (addr
< FIXADDR_USER_END
))
2262 #endif /* __HAVE_ARCH_GATE_AREA */