]>
git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blob - mm/memory.c
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
);
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
);
335 static inline void add_mm_rss(struct mm_struct
*mm
, int file_rss
, int anon_rss
)
338 add_mm_counter(mm
, file_rss
, file_rss
);
340 add_mm_counter(mm
, anon_rss
, anon_rss
);
343 #define NO_RSS 2 /* Increment neither file_rss nor anon_rss */
346 * This function is called to print an error when a pte in a
347 * !VM_RESERVED region is found pointing to an invalid pfn (which
350 * The calling function must still handle the error.
352 void print_bad_pte(struct vm_area_struct
*vma
, pte_t pte
, unsigned long vaddr
)
354 printk(KERN_ERR
"Bad pte = %08llx, process = %s, "
355 "vm_flags = %lx, vaddr = %lx\n",
356 (long long)pte_val(pte
),
357 (vma
->vm_mm
== current
->mm
? current
->comm
: "???"),
358 vma
->vm_flags
, vaddr
);
363 * copy one vm_area from one task to the other. Assumes the page tables
364 * already present in the new task to be cleared in the whole range
365 * covered by this vma.
367 * dst->page_table_lock is held on entry and exit,
368 * but may be dropped within p[mg]d_alloc() and pte_alloc_map().
372 copy_one_pte(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
373 pte_t
*dst_pte
, pte_t
*src_pte
, struct vm_area_struct
*vma
,
376 unsigned long vm_flags
= vma
->vm_flags
;
377 pte_t pte
= *src_pte
;
382 /* pte contains position in swap or file, so copy. */
383 if (unlikely(!pte_present(pte
))) {
384 if (!pte_file(pte
)) {
385 swap_duplicate(pte_to_swp_entry(pte
));
386 /* make sure dst_mm is on swapoff's mmlist. */
387 if (unlikely(list_empty(&dst_mm
->mmlist
))) {
388 spin_lock(&mmlist_lock
);
389 list_add(&dst_mm
->mmlist
, &src_mm
->mmlist
);
390 spin_unlock(&mmlist_lock
);
396 /* If the region is VM_RESERVED, the mapping is not
397 * mapped via rmap - duplicate the pte as is.
399 if (vm_flags
& VM_RESERVED
)
403 /* If the pte points outside of valid memory but
404 * the region is not VM_RESERVED, we have a problem.
406 if (unlikely(!pfn_valid(pfn
))) {
407 print_bad_pte(vma
, pte
, addr
);
408 goto out_set_pte
; /* try to do something sane */
411 page
= pfn_to_page(pfn
);
414 * If it's a COW mapping, write protect it both
415 * in the parent and the child
417 if ((vm_flags
& (VM_SHARED
| VM_MAYWRITE
)) == VM_MAYWRITE
) {
418 ptep_set_wrprotect(src_mm
, addr
, src_pte
);
423 * If it's a shared mapping, mark it clean in
426 if (vm_flags
& VM_SHARED
)
427 pte
= pte_mkclean(pte
);
428 pte
= pte_mkold(pte
);
431 anon
= !!PageAnon(page
);
434 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
438 static int copy_pte_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
439 pmd_t
*dst_pmd
, pmd_t
*src_pmd
, struct vm_area_struct
*vma
,
440 unsigned long addr
, unsigned long end
)
442 pte_t
*src_pte
, *dst_pte
;
444 int rss
[NO_RSS
+1], anon
;
448 dst_pte
= pte_alloc_map(dst_mm
, dst_pmd
, addr
);
451 src_pte
= pte_offset_map_nested(src_pmd
, addr
);
453 spin_lock(&src_mm
->page_table_lock
);
456 * We are holding two locks at this point - either of them
457 * could generate latencies in another task on another CPU.
459 if (progress
>= 32) {
461 if (need_resched() ||
462 need_lockbreak(&src_mm
->page_table_lock
) ||
463 need_lockbreak(&dst_mm
->page_table_lock
))
466 if (pte_none(*src_pte
)) {
470 anon
= copy_one_pte(dst_mm
, src_mm
, dst_pte
, src_pte
, vma
,addr
);
473 } while (dst_pte
++, src_pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
474 spin_unlock(&src_mm
->page_table_lock
);
476 pte_unmap_nested(src_pte
- 1);
477 pte_unmap(dst_pte
- 1);
478 add_mm_rss(dst_mm
, rss
[0], rss
[1]);
479 cond_resched_lock(&dst_mm
->page_table_lock
);
485 static inline int copy_pmd_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
486 pud_t
*dst_pud
, pud_t
*src_pud
, struct vm_area_struct
*vma
,
487 unsigned long addr
, unsigned long end
)
489 pmd_t
*src_pmd
, *dst_pmd
;
492 dst_pmd
= pmd_alloc(dst_mm
, dst_pud
, addr
);
495 src_pmd
= pmd_offset(src_pud
, addr
);
497 next
= pmd_addr_end(addr
, end
);
498 if (pmd_none_or_clear_bad(src_pmd
))
500 if (copy_pte_range(dst_mm
, src_mm
, dst_pmd
, src_pmd
,
503 } while (dst_pmd
++, src_pmd
++, addr
= next
, addr
!= end
);
507 static inline int copy_pud_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
508 pgd_t
*dst_pgd
, pgd_t
*src_pgd
, struct vm_area_struct
*vma
,
509 unsigned long addr
, unsigned long end
)
511 pud_t
*src_pud
, *dst_pud
;
514 dst_pud
= pud_alloc(dst_mm
, dst_pgd
, addr
);
517 src_pud
= pud_offset(src_pgd
, addr
);
519 next
= pud_addr_end(addr
, end
);
520 if (pud_none_or_clear_bad(src_pud
))
522 if (copy_pmd_range(dst_mm
, src_mm
, dst_pud
, src_pud
,
525 } while (dst_pud
++, src_pud
++, addr
= next
, addr
!= end
);
529 int copy_page_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
530 struct vm_area_struct
*vma
)
532 pgd_t
*src_pgd
, *dst_pgd
;
534 unsigned long addr
= vma
->vm_start
;
535 unsigned long end
= vma
->vm_end
;
538 * Don't copy ptes where a page fault will fill them correctly.
539 * Fork becomes much lighter when there are big shared or private
540 * readonly mappings. The tradeoff is that copy_page_range is more
541 * efficient than faulting.
543 if (!(vma
->vm_flags
& (VM_HUGETLB
|VM_NONLINEAR
|VM_RESERVED
))) {
548 if (is_vm_hugetlb_page(vma
))
549 return copy_hugetlb_page_range(dst_mm
, src_mm
, vma
);
551 dst_pgd
= pgd_offset(dst_mm
, addr
);
552 src_pgd
= pgd_offset(src_mm
, addr
);
554 next
= pgd_addr_end(addr
, end
);
555 if (pgd_none_or_clear_bad(src_pgd
))
557 if (copy_pud_range(dst_mm
, src_mm
, dst_pgd
, src_pgd
,
560 } while (dst_pgd
++, src_pgd
++, addr
= next
, addr
!= end
);
564 static void zap_pte_range(struct mmu_gather
*tlb
,
565 struct vm_area_struct
*vma
, pmd_t
*pmd
,
566 unsigned long addr
, unsigned long end
,
567 struct zap_details
*details
)
569 struct mm_struct
*mm
= tlb
->mm
;
574 pte
= pte_offset_map(pmd
, addr
);
579 if (pte_present(ptent
)) {
580 struct page
*page
= NULL
;
581 if (!(vma
->vm_flags
& VM_RESERVED
)) {
582 unsigned long pfn
= pte_pfn(ptent
);
583 if (unlikely(!pfn_valid(pfn
)))
584 print_bad_pte(vma
, ptent
, addr
);
586 page
= pfn_to_page(pfn
);
588 if (unlikely(details
) && page
) {
590 * unmap_shared_mapping_pages() wants to
591 * invalidate cache without truncating:
592 * unmap shared but keep private pages.
594 if (details
->check_mapping
&&
595 details
->check_mapping
!= page
->mapping
)
598 * Each page->index must be checked when
599 * invalidating or truncating nonlinear.
601 if (details
->nonlinear_vma
&&
602 (page
->index
< details
->first_index
||
603 page
->index
> details
->last_index
))
606 ptent
= ptep_get_and_clear_full(mm
, addr
, pte
,
608 tlb_remove_tlb_entry(tlb
, pte
, addr
);
611 if (unlikely(details
) && details
->nonlinear_vma
612 && linear_page_index(details
->nonlinear_vma
,
613 addr
) != page
->index
)
614 set_pte_at(mm
, addr
, pte
,
615 pgoff_to_pte(page
->index
));
619 if (pte_dirty(ptent
))
620 set_page_dirty(page
);
621 if (pte_young(ptent
))
622 mark_page_accessed(page
);
625 page_remove_rmap(page
);
626 tlb_remove_page(tlb
, page
);
630 * If details->check_mapping, we leave swap entries;
631 * if details->nonlinear_vma, we leave file entries.
633 if (unlikely(details
))
635 if (!pte_file(ptent
))
636 free_swap_and_cache(pte_to_swp_entry(ptent
));
637 pte_clear_full(mm
, addr
, pte
, tlb
->fullmm
);
638 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
640 add_mm_rss(mm
, -file_rss
, -anon_rss
);
644 static inline void zap_pmd_range(struct mmu_gather
*tlb
,
645 struct vm_area_struct
*vma
, pud_t
*pud
,
646 unsigned long addr
, unsigned long end
,
647 struct zap_details
*details
)
652 pmd
= pmd_offset(pud
, addr
);
654 next
= pmd_addr_end(addr
, end
);
655 if (pmd_none_or_clear_bad(pmd
))
657 zap_pte_range(tlb
, vma
, pmd
, addr
, next
, details
);
658 } while (pmd
++, addr
= next
, addr
!= end
);
661 static inline void zap_pud_range(struct mmu_gather
*tlb
,
662 struct vm_area_struct
*vma
, pgd_t
*pgd
,
663 unsigned long addr
, unsigned long end
,
664 struct zap_details
*details
)
669 pud
= pud_offset(pgd
, addr
);
671 next
= pud_addr_end(addr
, end
);
672 if (pud_none_or_clear_bad(pud
))
674 zap_pmd_range(tlb
, vma
, pud
, addr
, next
, details
);
675 } while (pud
++, addr
= next
, addr
!= end
);
678 static void unmap_page_range(struct mmu_gather
*tlb
, struct vm_area_struct
*vma
,
679 unsigned long addr
, unsigned long end
,
680 struct zap_details
*details
)
685 if (details
&& !details
->check_mapping
&& !details
->nonlinear_vma
)
689 tlb_start_vma(tlb
, vma
);
690 pgd
= pgd_offset(vma
->vm_mm
, addr
);
692 next
= pgd_addr_end(addr
, end
);
693 if (pgd_none_or_clear_bad(pgd
))
695 zap_pud_range(tlb
, vma
, pgd
, addr
, next
, details
);
696 } while (pgd
++, addr
= next
, addr
!= end
);
697 tlb_end_vma(tlb
, vma
);
700 #ifdef CONFIG_PREEMPT
701 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
703 /* No preempt: go for improved straight-line efficiency */
704 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
708 * unmap_vmas - unmap a range of memory covered by a list of vma's
709 * @tlbp: address of the caller's struct mmu_gather
710 * @mm: the controlling mm_struct
711 * @vma: the starting vma
712 * @start_addr: virtual address at which to start unmapping
713 * @end_addr: virtual address at which to end unmapping
714 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
715 * @details: details of nonlinear truncation or shared cache invalidation
717 * Returns the end address of the unmapping (restart addr if interrupted).
719 * Unmap all pages in the vma list. Called under page_table_lock.
721 * We aim to not hold page_table_lock for too long (for scheduling latency
722 * reasons). So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
723 * return the ending mmu_gather to the caller.
725 * Only addresses between `start' and `end' will be unmapped.
727 * The VMA list must be sorted in ascending virtual address order.
729 * unmap_vmas() assumes that the caller will flush the whole unmapped address
730 * range after unmap_vmas() returns. So the only responsibility here is to
731 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
732 * drops the lock and schedules.
734 unsigned long unmap_vmas(struct mmu_gather
**tlbp
, struct mm_struct
*mm
,
735 struct vm_area_struct
*vma
, unsigned long start_addr
,
736 unsigned long end_addr
, unsigned long *nr_accounted
,
737 struct zap_details
*details
)
739 unsigned long zap_bytes
= ZAP_BLOCK_SIZE
;
740 unsigned long tlb_start
= 0; /* For tlb_finish_mmu */
741 int tlb_start_valid
= 0;
742 unsigned long start
= start_addr
;
743 spinlock_t
*i_mmap_lock
= details
? details
->i_mmap_lock
: NULL
;
744 int fullmm
= (*tlbp
)->fullmm
;
746 for ( ; vma
&& vma
->vm_start
< end_addr
; vma
= vma
->vm_next
) {
749 start
= max(vma
->vm_start
, start_addr
);
750 if (start
>= vma
->vm_end
)
752 end
= min(vma
->vm_end
, end_addr
);
753 if (end
<= vma
->vm_start
)
756 if (vma
->vm_flags
& VM_ACCOUNT
)
757 *nr_accounted
+= (end
- start
) >> PAGE_SHIFT
;
759 while (start
!= end
) {
762 if (!tlb_start_valid
) {
767 if (is_vm_hugetlb_page(vma
)) {
769 unmap_hugepage_range(vma
, start
, end
);
771 block
= min(zap_bytes
, end
- start
);
772 unmap_page_range(*tlbp
, vma
, start
,
773 start
+ block
, details
);
778 if ((long)zap_bytes
> 0)
781 tlb_finish_mmu(*tlbp
, tlb_start
, start
);
783 if (need_resched() ||
784 need_lockbreak(&mm
->page_table_lock
) ||
785 (i_mmap_lock
&& need_lockbreak(i_mmap_lock
))) {
787 /* must reset count of rss freed */
788 *tlbp
= tlb_gather_mmu(mm
, fullmm
);
791 spin_unlock(&mm
->page_table_lock
);
793 spin_lock(&mm
->page_table_lock
);
796 *tlbp
= tlb_gather_mmu(mm
, fullmm
);
798 zap_bytes
= ZAP_BLOCK_SIZE
;
802 return start
; /* which is now the end (or restart) address */
806 * zap_page_range - remove user pages in a given range
807 * @vma: vm_area_struct holding the applicable pages
808 * @address: starting address of pages to zap
809 * @size: number of bytes to zap
810 * @details: details of nonlinear truncation or shared cache invalidation
812 unsigned long zap_page_range(struct vm_area_struct
*vma
, unsigned long address
,
813 unsigned long size
, struct zap_details
*details
)
815 struct mm_struct
*mm
= vma
->vm_mm
;
816 struct mmu_gather
*tlb
;
817 unsigned long end
= address
+ size
;
818 unsigned long nr_accounted
= 0;
820 if (is_vm_hugetlb_page(vma
)) {
821 zap_hugepage_range(vma
, address
, size
);
826 spin_lock(&mm
->page_table_lock
);
827 tlb
= tlb_gather_mmu(mm
, 0);
828 end
= unmap_vmas(&tlb
, mm
, vma
, address
, end
, &nr_accounted
, details
);
829 tlb_finish_mmu(tlb
, address
, end
);
830 spin_unlock(&mm
->page_table_lock
);
835 * Do a quick page-table lookup for a single page.
836 * mm->page_table_lock must be held.
838 static struct page
*__follow_page(struct mm_struct
*mm
, unsigned long address
,
839 int read
, int write
, int accessed
)
848 page
= follow_huge_addr(mm
, address
, write
);
852 pgd
= pgd_offset(mm
, address
);
853 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
856 pud
= pud_offset(pgd
, address
);
857 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
860 pmd
= pmd_offset(pud
, address
);
861 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
864 return follow_huge_pmd(mm
, address
, pmd
, write
);
866 ptep
= pte_offset_map(pmd
, address
);
872 if (pte_present(pte
)) {
873 if (write
&& !pte_write(pte
))
875 if (read
&& !pte_read(pte
))
878 if (pfn_valid(pfn
)) {
879 page
= pfn_to_page(pfn
);
881 if (write
&& !pte_dirty(pte
) &&!PageDirty(page
))
882 set_page_dirty(page
);
883 mark_page_accessed(page
);
894 follow_page(struct mm_struct
*mm
, unsigned long address
, int write
)
896 return __follow_page(mm
, address
, 0, write
, 1);
900 * check_user_page_readable() can be called frm niterrupt context by oprofile,
901 * so we need to avoid taking any non-irq-safe locks
903 int check_user_page_readable(struct mm_struct
*mm
, unsigned long address
)
905 return __follow_page(mm
, address
, 1, 0, 0) != NULL
;
907 EXPORT_SYMBOL(check_user_page_readable
);
910 untouched_anonymous_page(struct mm_struct
* mm
, struct vm_area_struct
*vma
,
911 unsigned long address
)
917 /* Check if the vma is for an anonymous mapping. */
918 if (vma
->vm_ops
&& vma
->vm_ops
->nopage
)
921 /* Check if page directory entry exists. */
922 pgd
= pgd_offset(mm
, address
);
923 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
926 pud
= pud_offset(pgd
, address
);
927 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
930 /* Check if page middle directory entry exists. */
931 pmd
= pmd_offset(pud
, address
);
932 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
935 /* There is a pte slot for 'address' in 'mm'. */
939 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
940 unsigned long start
, int len
, int write
, int force
,
941 struct page
**pages
, struct vm_area_struct
**vmas
)
947 * Require read or write permissions.
948 * If 'force' is set, we only require the "MAY" flags.
950 flags
= write
? (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
951 flags
&= force
? (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
955 struct vm_area_struct
* vma
;
957 vma
= find_extend_vma(mm
, start
);
958 if (!vma
&& in_gate_area(tsk
, start
)) {
959 unsigned long pg
= start
& PAGE_MASK
;
960 struct vm_area_struct
*gate_vma
= get_gate_vma(tsk
);
965 if (write
) /* user gate pages are read-only */
966 return i
? : -EFAULT
;
968 pgd
= pgd_offset_k(pg
);
970 pgd
= pgd_offset_gate(mm
, pg
);
971 BUG_ON(pgd_none(*pgd
));
972 pud
= pud_offset(pgd
, pg
);
973 BUG_ON(pud_none(*pud
));
974 pmd
= pmd_offset(pud
, pg
);
976 return i
? : -EFAULT
;
977 pte
= pte_offset_map(pmd
, pg
);
978 if (pte_none(*pte
)) {
980 return i
? : -EFAULT
;
983 pages
[i
] = pte_page(*pte
);
995 if (!vma
|| (vma
->vm_flags
& (VM_IO
| VM_RESERVED
))
996 || !(flags
& vma
->vm_flags
))
997 return i
? : -EFAULT
;
999 if (is_vm_hugetlb_page(vma
)) {
1000 i
= follow_hugetlb_page(mm
, vma
, pages
, vmas
,
1004 spin_lock(&mm
->page_table_lock
);
1006 int write_access
= write
;
1009 cond_resched_lock(&mm
->page_table_lock
);
1010 while (!(page
= follow_page(mm
, start
, write_access
))) {
1014 * Shortcut for anonymous pages. We don't want
1015 * to force the creation of pages tables for
1016 * insanely big anonymously mapped areas that
1017 * nobody touched so far. This is important
1018 * for doing a core dump for these mappings.
1020 if (!write
&& untouched_anonymous_page(mm
,vma
,start
)) {
1021 page
= ZERO_PAGE(start
);
1024 spin_unlock(&mm
->page_table_lock
);
1025 ret
= __handle_mm_fault(mm
, vma
, start
, write_access
);
1028 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1029 * broken COW when necessary, even if maybe_mkwrite
1030 * decided not to set pte_write. We can thus safely do
1031 * subsequent page lookups as if they were reads.
1033 if (ret
& VM_FAULT_WRITE
)
1036 switch (ret
& ~VM_FAULT_WRITE
) {
1037 case VM_FAULT_MINOR
:
1040 case VM_FAULT_MAJOR
:
1043 case VM_FAULT_SIGBUS
:
1044 return i
? i
: -EFAULT
;
1046 return i
? i
: -ENOMEM
;
1050 spin_lock(&mm
->page_table_lock
);
1054 flush_dcache_page(page
);
1055 page_cache_get(page
);
1062 } while (len
&& start
< vma
->vm_end
);
1063 spin_unlock(&mm
->page_table_lock
);
1067 EXPORT_SYMBOL(get_user_pages
);
1069 static int zeromap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1070 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1074 pte
= pte_alloc_map(mm
, pmd
, addr
);
1078 struct page
*page
= ZERO_PAGE(addr
);
1079 pte_t zero_pte
= pte_wrprotect(mk_pte(page
, prot
));
1080 page_cache_get(page
);
1081 page_add_file_rmap(page
);
1082 inc_mm_counter(mm
, file_rss
);
1083 BUG_ON(!pte_none(*pte
));
1084 set_pte_at(mm
, addr
, pte
, zero_pte
);
1085 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1090 static inline int zeromap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1091 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1096 pmd
= pmd_alloc(mm
, pud
, addr
);
1100 next
= pmd_addr_end(addr
, end
);
1101 if (zeromap_pte_range(mm
, pmd
, addr
, next
, prot
))
1103 } while (pmd
++, addr
= next
, addr
!= end
);
1107 static inline int zeromap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1108 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1113 pud
= pud_alloc(mm
, pgd
, addr
);
1117 next
= pud_addr_end(addr
, end
);
1118 if (zeromap_pmd_range(mm
, pud
, addr
, next
, prot
))
1120 } while (pud
++, addr
= next
, addr
!= end
);
1124 int zeromap_page_range(struct vm_area_struct
*vma
,
1125 unsigned long addr
, unsigned long size
, pgprot_t prot
)
1129 unsigned long end
= addr
+ size
;
1130 struct mm_struct
*mm
= vma
->vm_mm
;
1133 BUG_ON(addr
>= end
);
1134 pgd
= pgd_offset(mm
, addr
);
1135 flush_cache_range(vma
, addr
, end
);
1136 spin_lock(&mm
->page_table_lock
);
1138 next
= pgd_addr_end(addr
, end
);
1139 err
= zeromap_pud_range(mm
, pgd
, addr
, next
, prot
);
1142 } while (pgd
++, addr
= next
, addr
!= end
);
1143 spin_unlock(&mm
->page_table_lock
);
1148 * maps a range of physical memory into the requested pages. the old
1149 * mappings are removed. any references to nonexistent pages results
1150 * in null mappings (currently treated as "copy-on-access")
1152 static int remap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1153 unsigned long addr
, unsigned long end
,
1154 unsigned long pfn
, pgprot_t prot
)
1158 pte
= pte_alloc_map(mm
, pmd
, addr
);
1162 BUG_ON(!pte_none(*pte
));
1163 set_pte_at(mm
, addr
, pte
, pfn_pte(pfn
, prot
));
1165 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1170 static inline int remap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1171 unsigned long addr
, unsigned long end
,
1172 unsigned long pfn
, pgprot_t prot
)
1177 pfn
-= addr
>> PAGE_SHIFT
;
1178 pmd
= pmd_alloc(mm
, pud
, addr
);
1182 next
= pmd_addr_end(addr
, end
);
1183 if (remap_pte_range(mm
, pmd
, addr
, next
,
1184 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1186 } while (pmd
++, addr
= next
, addr
!= end
);
1190 static inline int remap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1191 unsigned long addr
, unsigned long end
,
1192 unsigned long pfn
, pgprot_t prot
)
1197 pfn
-= addr
>> PAGE_SHIFT
;
1198 pud
= pud_alloc(mm
, pgd
, addr
);
1202 next
= pud_addr_end(addr
, end
);
1203 if (remap_pmd_range(mm
, pud
, addr
, next
,
1204 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1206 } while (pud
++, addr
= next
, addr
!= end
);
1210 /* Note: this is only safe if the mm semaphore is held when called. */
1211 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1212 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1216 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1217 struct mm_struct
*mm
= vma
->vm_mm
;
1221 * Physically remapped pages are special. Tell the
1222 * rest of the world about it:
1223 * VM_IO tells people not to look at these pages
1224 * (accesses can have side effects).
1225 * VM_RESERVED tells the core MM not to "manage" these pages
1226 * (e.g. refcount, mapcount, try to swap them out).
1228 vma
->vm_flags
|= VM_IO
| VM_RESERVED
;
1230 BUG_ON(addr
>= end
);
1231 pfn
-= addr
>> PAGE_SHIFT
;
1232 pgd
= pgd_offset(mm
, addr
);
1233 flush_cache_range(vma
, addr
, end
);
1234 spin_lock(&mm
->page_table_lock
);
1236 next
= pgd_addr_end(addr
, end
);
1237 err
= remap_pud_range(mm
, pgd
, addr
, next
,
1238 pfn
+ (addr
>> PAGE_SHIFT
), prot
);
1241 } while (pgd
++, addr
= next
, addr
!= end
);
1242 spin_unlock(&mm
->page_table_lock
);
1245 EXPORT_SYMBOL(remap_pfn_range
);
1248 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1249 * servicing faults for write access. In the normal case, do always want
1250 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1251 * that do not have writing enabled, when used by access_process_vm.
1253 static inline pte_t
maybe_mkwrite(pte_t pte
, struct vm_area_struct
*vma
)
1255 if (likely(vma
->vm_flags
& VM_WRITE
))
1256 pte
= pte_mkwrite(pte
);
1261 * This routine handles present pages, when users try to write
1262 * to a shared page. It is done by copying the page to a new address
1263 * and decrementing the shared-page counter for the old page.
1265 * Note that this routine assumes that the protection checks have been
1266 * done by the caller (the low-level page fault routine in most cases).
1267 * Thus we can safely just mark it writable once we've done any necessary
1270 * We also mark the page dirty at this point even though the page will
1271 * change only once the write actually happens. This avoids a few races,
1272 * and potentially makes it more efficient.
1274 * We hold the mm semaphore and the page_table_lock on entry and exit
1275 * with the page_table_lock released.
1277 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1278 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1281 struct page
*old_page
, *new_page
;
1282 unsigned long pfn
= pte_pfn(orig_pte
);
1284 int ret
= VM_FAULT_MINOR
;
1286 BUG_ON(vma
->vm_flags
& VM_RESERVED
);
1288 if (unlikely(!pfn_valid(pfn
))) {
1290 * Page table corrupted: show pte and kill process.
1292 print_bad_pte(vma
, orig_pte
, address
);
1296 old_page
= pfn_to_page(pfn
);
1298 if (PageAnon(old_page
) && !TestSetPageLocked(old_page
)) {
1299 int reuse
= can_share_swap_page(old_page
);
1300 unlock_page(old_page
);
1302 flush_cache_page(vma
, address
, pfn
);
1303 entry
= pte_mkyoung(orig_pte
);
1304 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1305 ptep_set_access_flags(vma
, address
, page_table
, entry
, 1);
1306 update_mmu_cache(vma
, address
, entry
);
1307 lazy_mmu_prot_update(entry
);
1308 ret
|= VM_FAULT_WRITE
;
1314 * Ok, we need to copy. Oh, well..
1316 page_cache_get(old_page
);
1317 pte_unmap(page_table
);
1318 spin_unlock(&mm
->page_table_lock
);
1320 if (unlikely(anon_vma_prepare(vma
)))
1322 if (old_page
== ZERO_PAGE(address
)) {
1323 new_page
= alloc_zeroed_user_highpage(vma
, address
);
1327 new_page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
1330 copy_user_highpage(new_page
, old_page
, address
);
1334 * Re-check the pte - we dropped the lock
1336 spin_lock(&mm
->page_table_lock
);
1337 page_table
= pte_offset_map(pmd
, address
);
1338 if (likely(pte_same(*page_table
, orig_pte
))) {
1339 page_remove_rmap(old_page
);
1340 if (!PageAnon(old_page
)) {
1341 inc_mm_counter(mm
, anon_rss
);
1342 dec_mm_counter(mm
, file_rss
);
1344 flush_cache_page(vma
, address
, pfn
);
1345 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
1346 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1347 ptep_establish(vma
, address
, page_table
, entry
);
1348 update_mmu_cache(vma
, address
, entry
);
1349 lazy_mmu_prot_update(entry
);
1351 lru_cache_add_active(new_page
);
1352 page_add_anon_rmap(new_page
, vma
, address
);
1354 /* Free the old page.. */
1355 new_page
= old_page
;
1356 ret
|= VM_FAULT_WRITE
;
1358 page_cache_release(new_page
);
1359 page_cache_release(old_page
);
1361 pte_unmap(page_table
);
1362 spin_unlock(&mm
->page_table_lock
);
1365 page_cache_release(old_page
);
1366 return VM_FAULT_OOM
;
1370 * Helper functions for unmap_mapping_range().
1372 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1374 * We have to restart searching the prio_tree whenever we drop the lock,
1375 * since the iterator is only valid while the lock is held, and anyway
1376 * a later vma might be split and reinserted earlier while lock dropped.
1378 * The list of nonlinear vmas could be handled more efficiently, using
1379 * a placeholder, but handle it in the same way until a need is shown.
1380 * It is important to search the prio_tree before nonlinear list: a vma
1381 * may become nonlinear and be shifted from prio_tree to nonlinear list
1382 * while the lock is dropped; but never shifted from list to prio_tree.
1384 * In order to make forward progress despite restarting the search,
1385 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1386 * quickly skip it next time around. Since the prio_tree search only
1387 * shows us those vmas affected by unmapping the range in question, we
1388 * can't efficiently keep all vmas in step with mapping->truncate_count:
1389 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1390 * mapping->truncate_count and vma->vm_truncate_count are protected by
1393 * In order to make forward progress despite repeatedly restarting some
1394 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1395 * and restart from that address when we reach that vma again. It might
1396 * have been split or merged, shrunk or extended, but never shifted: so
1397 * restart_addr remains valid so long as it remains in the vma's range.
1398 * unmap_mapping_range forces truncate_count to leap over page-aligned
1399 * values so we can save vma's restart_addr in its truncate_count field.
1401 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1403 static void reset_vma_truncate_counts(struct address_space
*mapping
)
1405 struct vm_area_struct
*vma
;
1406 struct prio_tree_iter iter
;
1408 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, 0, ULONG_MAX
)
1409 vma
->vm_truncate_count
= 0;
1410 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
, shared
.vm_set
.list
)
1411 vma
->vm_truncate_count
= 0;
1414 static int unmap_mapping_range_vma(struct vm_area_struct
*vma
,
1415 unsigned long start_addr
, unsigned long end_addr
,
1416 struct zap_details
*details
)
1418 unsigned long restart_addr
;
1422 restart_addr
= vma
->vm_truncate_count
;
1423 if (is_restart_addr(restart_addr
) && start_addr
< restart_addr
) {
1424 start_addr
= restart_addr
;
1425 if (start_addr
>= end_addr
) {
1426 /* Top of vma has been split off since last time */
1427 vma
->vm_truncate_count
= details
->truncate_count
;
1432 restart_addr
= zap_page_range(vma
, start_addr
,
1433 end_addr
- start_addr
, details
);
1436 * We cannot rely on the break test in unmap_vmas:
1437 * on the one hand, we don't want to restart our loop
1438 * just because that broke out for the page_table_lock;
1439 * on the other hand, it does no test when vma is small.
1441 need_break
= need_resched() ||
1442 need_lockbreak(details
->i_mmap_lock
);
1444 if (restart_addr
>= end_addr
) {
1445 /* We have now completed this vma: mark it so */
1446 vma
->vm_truncate_count
= details
->truncate_count
;
1450 /* Note restart_addr in vma's truncate_count field */
1451 vma
->vm_truncate_count
= restart_addr
;
1456 spin_unlock(details
->i_mmap_lock
);
1458 spin_lock(details
->i_mmap_lock
);
1462 static inline void unmap_mapping_range_tree(struct prio_tree_root
*root
,
1463 struct zap_details
*details
)
1465 struct vm_area_struct
*vma
;
1466 struct prio_tree_iter iter
;
1467 pgoff_t vba
, vea
, zba
, zea
;
1470 vma_prio_tree_foreach(vma
, &iter
, root
,
1471 details
->first_index
, details
->last_index
) {
1472 /* Skip quickly over those we have already dealt with */
1473 if (vma
->vm_truncate_count
== details
->truncate_count
)
1476 vba
= vma
->vm_pgoff
;
1477 vea
= vba
+ ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
) - 1;
1478 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1479 zba
= details
->first_index
;
1482 zea
= details
->last_index
;
1486 if (unmap_mapping_range_vma(vma
,
1487 ((zba
- vba
) << PAGE_SHIFT
) + vma
->vm_start
,
1488 ((zea
- vba
+ 1) << PAGE_SHIFT
) + vma
->vm_start
,
1494 static inline void unmap_mapping_range_list(struct list_head
*head
,
1495 struct zap_details
*details
)
1497 struct vm_area_struct
*vma
;
1500 * In nonlinear VMAs there is no correspondence between virtual address
1501 * offset and file offset. So we must perform an exhaustive search
1502 * across *all* the pages in each nonlinear VMA, not just the pages
1503 * whose virtual address lies outside the file truncation point.
1506 list_for_each_entry(vma
, head
, shared
.vm_set
.list
) {
1507 /* Skip quickly over those we have already dealt with */
1508 if (vma
->vm_truncate_count
== details
->truncate_count
)
1510 details
->nonlinear_vma
= vma
;
1511 if (unmap_mapping_range_vma(vma
, vma
->vm_start
,
1512 vma
->vm_end
, details
) < 0)
1518 * unmap_mapping_range - unmap the portion of all mmaps
1519 * in the specified address_space corresponding to the specified
1520 * page range in the underlying file.
1521 * @mapping: the address space containing mmaps to be unmapped.
1522 * @holebegin: byte in first page to unmap, relative to the start of
1523 * the underlying file. This will be rounded down to a PAGE_SIZE
1524 * boundary. Note that this is different from vmtruncate(), which
1525 * must keep the partial page. In contrast, we must get rid of
1527 * @holelen: size of prospective hole in bytes. This will be rounded
1528 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1530 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1531 * but 0 when invalidating pagecache, don't throw away private data.
1533 void unmap_mapping_range(struct address_space
*mapping
,
1534 loff_t
const holebegin
, loff_t
const holelen
, int even_cows
)
1536 struct zap_details details
;
1537 pgoff_t hba
= holebegin
>> PAGE_SHIFT
;
1538 pgoff_t hlen
= (holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1540 /* Check for overflow. */
1541 if (sizeof(holelen
) > sizeof(hlen
)) {
1543 (holebegin
+ holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1544 if (holeend
& ~(long long)ULONG_MAX
)
1545 hlen
= ULONG_MAX
- hba
+ 1;
1548 details
.check_mapping
= even_cows
? NULL
: mapping
;
1549 details
.nonlinear_vma
= NULL
;
1550 details
.first_index
= hba
;
1551 details
.last_index
= hba
+ hlen
- 1;
1552 if (details
.last_index
< details
.first_index
)
1553 details
.last_index
= ULONG_MAX
;
1554 details
.i_mmap_lock
= &mapping
->i_mmap_lock
;
1556 spin_lock(&mapping
->i_mmap_lock
);
1558 /* serialize i_size write against truncate_count write */
1560 /* Protect against page faults, and endless unmapping loops */
1561 mapping
->truncate_count
++;
1563 * For archs where spin_lock has inclusive semantics like ia64
1564 * this smp_mb() will prevent to read pagetable contents
1565 * before the truncate_count increment is visible to
1569 if (unlikely(is_restart_addr(mapping
->truncate_count
))) {
1570 if (mapping
->truncate_count
== 0)
1571 reset_vma_truncate_counts(mapping
);
1572 mapping
->truncate_count
++;
1574 details
.truncate_count
= mapping
->truncate_count
;
1576 if (unlikely(!prio_tree_empty(&mapping
->i_mmap
)))
1577 unmap_mapping_range_tree(&mapping
->i_mmap
, &details
);
1578 if (unlikely(!list_empty(&mapping
->i_mmap_nonlinear
)))
1579 unmap_mapping_range_list(&mapping
->i_mmap_nonlinear
, &details
);
1580 spin_unlock(&mapping
->i_mmap_lock
);
1582 EXPORT_SYMBOL(unmap_mapping_range
);
1585 * Handle all mappings that got truncated by a "truncate()"
1588 * NOTE! We have to be ready to update the memory sharing
1589 * between the file and the memory map for a potential last
1590 * incomplete page. Ugly, but necessary.
1592 int vmtruncate(struct inode
* inode
, loff_t offset
)
1594 struct address_space
*mapping
= inode
->i_mapping
;
1595 unsigned long limit
;
1597 if (inode
->i_size
< offset
)
1600 * truncation of in-use swapfiles is disallowed - it would cause
1601 * subsequent swapout to scribble on the now-freed blocks.
1603 if (IS_SWAPFILE(inode
))
1605 i_size_write(inode
, offset
);
1606 unmap_mapping_range(mapping
, offset
+ PAGE_SIZE
- 1, 0, 1);
1607 truncate_inode_pages(mapping
, offset
);
1611 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1612 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
1614 if (offset
> inode
->i_sb
->s_maxbytes
)
1616 i_size_write(inode
, offset
);
1619 if (inode
->i_op
&& inode
->i_op
->truncate
)
1620 inode
->i_op
->truncate(inode
);
1623 send_sig(SIGXFSZ
, current
, 0);
1630 EXPORT_SYMBOL(vmtruncate
);
1633 * Primitive swap readahead code. We simply read an aligned block of
1634 * (1 << page_cluster) entries in the swap area. This method is chosen
1635 * because it doesn't cost us any seek time. We also make sure to queue
1636 * the 'original' request together with the readahead ones...
1638 * This has been extended to use the NUMA policies from the mm triggering
1641 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1643 void swapin_readahead(swp_entry_t entry
, unsigned long addr
,struct vm_area_struct
*vma
)
1646 struct vm_area_struct
*next_vma
= vma
? vma
->vm_next
: NULL
;
1649 struct page
*new_page
;
1650 unsigned long offset
;
1653 * Get the number of handles we should do readahead io to.
1655 num
= valid_swaphandles(entry
, &offset
);
1656 for (i
= 0; i
< num
; offset
++, i
++) {
1657 /* Ok, do the async read-ahead now */
1658 new_page
= read_swap_cache_async(swp_entry(swp_type(entry
),
1659 offset
), vma
, addr
);
1662 page_cache_release(new_page
);
1665 * Find the next applicable VMA for the NUMA policy.
1671 if (addr
>= vma
->vm_end
) {
1673 next_vma
= vma
? vma
->vm_next
: NULL
;
1675 if (vma
&& addr
< vma
->vm_start
)
1678 if (next_vma
&& addr
>= next_vma
->vm_start
) {
1680 next_vma
= vma
->vm_next
;
1685 lru_add_drain(); /* Push any new pages onto the LRU now */
1689 * We hold the mm semaphore and the page_table_lock on entry and
1690 * should release the pagetable lock on exit..
1692 static int do_swap_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1693 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1694 int write_access
, pte_t orig_pte
)
1699 int ret
= VM_FAULT_MINOR
;
1701 pte_unmap(page_table
);
1702 spin_unlock(&mm
->page_table_lock
);
1704 entry
= pte_to_swp_entry(orig_pte
);
1705 page
= lookup_swap_cache(entry
);
1707 swapin_readahead(entry
, address
, vma
);
1708 page
= read_swap_cache_async(entry
, vma
, address
);
1711 * Back out if somebody else faulted in this pte while
1712 * we released the page table lock.
1714 spin_lock(&mm
->page_table_lock
);
1715 page_table
= pte_offset_map(pmd
, address
);
1716 if (likely(pte_same(*page_table
, orig_pte
)))
1721 /* Had to read the page from swap area: Major fault */
1722 ret
= VM_FAULT_MAJOR
;
1723 inc_page_state(pgmajfault
);
1727 mark_page_accessed(page
);
1731 * Back out if somebody else faulted in this pte while we
1732 * released the page table lock.
1734 spin_lock(&mm
->page_table_lock
);
1735 page_table
= pte_offset_map(pmd
, address
);
1736 if (unlikely(!pte_same(*page_table
, orig_pte
))) {
1737 ret
= VM_FAULT_MINOR
;
1741 if (unlikely(!PageUptodate(page
))) {
1742 ret
= VM_FAULT_SIGBUS
;
1746 /* The page isn't present yet, go ahead with the fault. */
1748 inc_mm_counter(mm
, anon_rss
);
1749 pte
= mk_pte(page
, vma
->vm_page_prot
);
1750 if (write_access
&& can_share_swap_page(page
)) {
1751 pte
= maybe_mkwrite(pte_mkdirty(pte
), vma
);
1755 flush_icache_page(vma
, page
);
1756 set_pte_at(mm
, address
, page_table
, pte
);
1757 page_add_anon_rmap(page
, vma
, address
);
1761 remove_exclusive_swap_page(page
);
1765 if (do_wp_page(mm
, vma
, address
,
1766 page_table
, pmd
, pte
) == VM_FAULT_OOM
)
1771 /* No need to invalidate - it was non-present before */
1772 update_mmu_cache(vma
, address
, pte
);
1773 lazy_mmu_prot_update(pte
);
1775 pte_unmap(page_table
);
1776 spin_unlock(&mm
->page_table_lock
);
1780 pte_unmap(page_table
);
1781 spin_unlock(&mm
->page_table_lock
);
1783 page_cache_release(page
);
1788 * We are called with the MM semaphore and page_table_lock
1789 * spinlock held to protect against concurrent faults in
1790 * multithreaded programs.
1792 static int do_anonymous_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1793 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1796 struct page
*page
= ZERO_PAGE(addr
);
1799 /* Mapping of ZERO_PAGE - vm_page_prot is readonly */
1800 entry
= mk_pte(page
, vma
->vm_page_prot
);
1803 /* Allocate our own private page. */
1804 pte_unmap(page_table
);
1805 spin_unlock(&mm
->page_table_lock
);
1807 if (unlikely(anon_vma_prepare(vma
)))
1809 page
= alloc_zeroed_user_highpage(vma
, address
);
1813 spin_lock(&mm
->page_table_lock
);
1814 page_table
= pte_offset_map(pmd
, address
);
1816 if (!pte_none(*page_table
)) {
1817 page_cache_release(page
);
1820 inc_mm_counter(mm
, anon_rss
);
1821 entry
= mk_pte(page
, vma
->vm_page_prot
);
1822 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1823 lru_cache_add_active(page
);
1824 SetPageReferenced(page
);
1825 page_add_anon_rmap(page
, vma
, address
);
1827 inc_mm_counter(mm
, file_rss
);
1828 page_add_file_rmap(page
);
1829 page_cache_get(page
);
1832 set_pte_at(mm
, address
, page_table
, entry
);
1834 /* No need to invalidate - it was non-present before */
1835 update_mmu_cache(vma
, address
, entry
);
1836 lazy_mmu_prot_update(entry
);
1838 pte_unmap(page_table
);
1839 spin_unlock(&mm
->page_table_lock
);
1840 return VM_FAULT_MINOR
;
1842 return VM_FAULT_OOM
;
1846 * do_no_page() tries to create a new page mapping. It aggressively
1847 * tries to share with existing pages, but makes a separate copy if
1848 * the "write_access" parameter is true in order to avoid the next
1851 * As this is called only for pages that do not currently exist, we
1852 * do not need to flush old virtual caches or the TLB.
1854 * This is called with the MM semaphore held and the page table
1855 * spinlock held. Exit with the spinlock released.
1857 static int do_no_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1858 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1861 struct page
*new_page
;
1862 struct address_space
*mapping
= NULL
;
1864 unsigned int sequence
= 0;
1865 int ret
= VM_FAULT_MINOR
;
1868 pte_unmap(page_table
);
1869 spin_unlock(&mm
->page_table_lock
);
1872 mapping
= vma
->vm_file
->f_mapping
;
1873 sequence
= mapping
->truncate_count
;
1874 smp_rmb(); /* serializes i_size against truncate_count */
1877 new_page
= vma
->vm_ops
->nopage(vma
, address
& PAGE_MASK
, &ret
);
1879 * No smp_rmb is needed here as long as there's a full
1880 * spin_lock/unlock sequence inside the ->nopage callback
1881 * (for the pagecache lookup) that acts as an implicit
1882 * smp_mb() and prevents the i_size read to happen
1883 * after the next truncate_count read.
1886 /* no page was available -- either SIGBUS or OOM */
1887 if (new_page
== NOPAGE_SIGBUS
)
1888 return VM_FAULT_SIGBUS
;
1889 if (new_page
== NOPAGE_OOM
)
1890 return VM_FAULT_OOM
;
1893 * Should we do an early C-O-W break?
1895 if (write_access
&& !(vma
->vm_flags
& VM_SHARED
)) {
1898 if (unlikely(anon_vma_prepare(vma
)))
1900 page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
1903 copy_user_highpage(page
, new_page
, address
);
1904 page_cache_release(new_page
);
1909 spin_lock(&mm
->page_table_lock
);
1911 * For a file-backed vma, someone could have truncated or otherwise
1912 * invalidated this page. If unmap_mapping_range got called,
1913 * retry getting the page.
1915 if (mapping
&& unlikely(sequence
!= mapping
->truncate_count
)) {
1916 spin_unlock(&mm
->page_table_lock
);
1917 page_cache_release(new_page
);
1919 sequence
= mapping
->truncate_count
;
1923 page_table
= pte_offset_map(pmd
, address
);
1926 * This silly early PAGE_DIRTY setting removes a race
1927 * due to the bad i386 page protection. But it's valid
1928 * for other architectures too.
1930 * Note that if write_access is true, we either now have
1931 * an exclusive copy of the page, or this is a shared mapping,
1932 * so we can make it writable and dirty to avoid having to
1933 * handle that later.
1935 /* Only go through if we didn't race with anybody else... */
1936 if (pte_none(*page_table
)) {
1937 flush_icache_page(vma
, new_page
);
1938 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
1940 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1941 set_pte_at(mm
, address
, page_table
, entry
);
1943 inc_mm_counter(mm
, anon_rss
);
1944 lru_cache_add_active(new_page
);
1945 page_add_anon_rmap(new_page
, vma
, address
);
1946 } else if (!(vma
->vm_flags
& VM_RESERVED
)) {
1947 inc_mm_counter(mm
, file_rss
);
1948 page_add_file_rmap(new_page
);
1951 /* One of our sibling threads was faster, back out. */
1952 page_cache_release(new_page
);
1956 /* no need to invalidate: a not-present page shouldn't be cached */
1957 update_mmu_cache(vma
, address
, entry
);
1958 lazy_mmu_prot_update(entry
);
1960 pte_unmap(page_table
);
1961 spin_unlock(&mm
->page_table_lock
);
1964 page_cache_release(new_page
);
1965 return VM_FAULT_OOM
;
1969 * Fault of a previously existing named mapping. Repopulate the pte
1970 * from the encoded file_pte if possible. This enables swappable
1973 static int do_file_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1974 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1975 int write_access
, pte_t orig_pte
)
1980 pte_unmap(page_table
);
1981 spin_unlock(&mm
->page_table_lock
);
1983 if (unlikely(!(vma
->vm_flags
& VM_NONLINEAR
))) {
1985 * Page table corrupted: show pte and kill process.
1987 print_bad_pte(vma
, orig_pte
, address
);
1988 return VM_FAULT_OOM
;
1990 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
1992 pgoff
= pte_to_pgoff(orig_pte
);
1993 err
= vma
->vm_ops
->populate(vma
, address
& PAGE_MASK
, PAGE_SIZE
,
1994 vma
->vm_page_prot
, pgoff
, 0);
1996 return VM_FAULT_OOM
;
1998 return VM_FAULT_SIGBUS
;
1999 return VM_FAULT_MAJOR
;
2003 * These routines also need to handle stuff like marking pages dirty
2004 * and/or accessed for architectures that don't do it in hardware (most
2005 * RISC architectures). The early dirtying is also good on the i386.
2007 * There is also a hook called "update_mmu_cache()" that architectures
2008 * with external mmu caches can use to update those (ie the Sparc or
2009 * PowerPC hashed page tables that act as extended TLBs).
2011 * Note the "page_table_lock". It is to protect against kswapd removing
2012 * pages from under us. Note that kswapd only ever _removes_ pages, never
2013 * adds them. As such, once we have noticed that the page is not present,
2014 * we can drop the lock early.
2016 * The adding of pages is protected by the MM semaphore (which we hold),
2017 * so we don't need to worry about a page being suddenly been added into
2020 * We enter with the pagetable spinlock held, we are supposed to
2021 * release it when done.
2023 static inline int handle_pte_fault(struct mm_struct
*mm
,
2024 struct vm_area_struct
*vma
, unsigned long address
,
2025 pte_t
*pte
, pmd_t
*pmd
, int write_access
)
2030 if (!pte_present(entry
)) {
2031 if (pte_none(entry
)) {
2032 if (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
)
2033 return do_anonymous_page(mm
, vma
, address
,
2034 pte
, pmd
, write_access
);
2035 return do_no_page(mm
, vma
, address
,
2036 pte
, pmd
, write_access
);
2038 if (pte_file(entry
))
2039 return do_file_page(mm
, vma
, address
,
2040 pte
, pmd
, write_access
, entry
);
2041 return do_swap_page(mm
, vma
, address
,
2042 pte
, pmd
, write_access
, entry
);
2046 if (!pte_write(entry
))
2047 return do_wp_page(mm
, vma
, address
, pte
, pmd
, entry
);
2048 entry
= pte_mkdirty(entry
);
2050 entry
= pte_mkyoung(entry
);
2051 ptep_set_access_flags(vma
, address
, pte
, entry
, write_access
);
2052 update_mmu_cache(vma
, address
, entry
);
2053 lazy_mmu_prot_update(entry
);
2055 spin_unlock(&mm
->page_table_lock
);
2056 return VM_FAULT_MINOR
;
2060 * By the time we get here, we already hold the mm semaphore
2062 int __handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2063 unsigned long address
, int write_access
)
2070 __set_current_state(TASK_RUNNING
);
2072 inc_page_state(pgfault
);
2074 if (unlikely(is_vm_hugetlb_page(vma
)))
2075 return hugetlb_fault(mm
, vma
, address
, write_access
);
2078 * We need the page table lock to synchronize with kswapd
2079 * and the SMP-safe atomic PTE updates.
2081 pgd
= pgd_offset(mm
, address
);
2082 spin_lock(&mm
->page_table_lock
);
2084 pud
= pud_alloc(mm
, pgd
, address
);
2088 pmd
= pmd_alloc(mm
, pud
, address
);
2092 pte
= pte_alloc_map(mm
, pmd
, address
);
2096 return handle_pte_fault(mm
, vma
, address
, pte
, pmd
, write_access
);
2099 spin_unlock(&mm
->page_table_lock
);
2100 return VM_FAULT_OOM
;
2103 #ifndef __PAGETABLE_PUD_FOLDED
2105 * Allocate page upper directory.
2107 * We've already handled the fast-path in-line, and we own the
2110 pud_t fastcall
*__pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
2114 spin_unlock(&mm
->page_table_lock
);
2115 new = pud_alloc_one(mm
, address
);
2116 spin_lock(&mm
->page_table_lock
);
2121 * Because we dropped the lock, we should re-check the
2122 * entry, as somebody else could have populated it..
2124 if (pgd_present(*pgd
)) {
2128 pgd_populate(mm
, pgd
, new);
2130 return pud_offset(pgd
, address
);
2132 #endif /* __PAGETABLE_PUD_FOLDED */
2134 #ifndef __PAGETABLE_PMD_FOLDED
2136 * Allocate page middle directory.
2138 * We've already handled the fast-path in-line, and we own the
2141 pmd_t fastcall
*__pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
2145 spin_unlock(&mm
->page_table_lock
);
2146 new = pmd_alloc_one(mm
, address
);
2147 spin_lock(&mm
->page_table_lock
);
2152 * Because we dropped the lock, we should re-check the
2153 * entry, as somebody else could have populated it..
2155 #ifndef __ARCH_HAS_4LEVEL_HACK
2156 if (pud_present(*pud
)) {
2160 pud_populate(mm
, pud
, new);
2162 if (pgd_present(*pud
)) {
2166 pgd_populate(mm
, pud
, new);
2167 #endif /* __ARCH_HAS_4LEVEL_HACK */
2170 return pmd_offset(pud
, address
);
2172 #endif /* __PAGETABLE_PMD_FOLDED */
2174 int make_pages_present(unsigned long addr
, unsigned long end
)
2176 int ret
, len
, write
;
2177 struct vm_area_struct
* vma
;
2179 vma
= find_vma(current
->mm
, addr
);
2182 write
= (vma
->vm_flags
& VM_WRITE
) != 0;
2185 if (end
> vma
->vm_end
)
2187 len
= (end
+PAGE_SIZE
-1)/PAGE_SIZE
-addr
/PAGE_SIZE
;
2188 ret
= get_user_pages(current
, current
->mm
, addr
,
2189 len
, write
, 0, NULL
, NULL
);
2192 return ret
== len
? 0 : -1;
2196 * Map a vmalloc()-space virtual address to the physical page.
2198 struct page
* vmalloc_to_page(void * vmalloc_addr
)
2200 unsigned long addr
= (unsigned long) vmalloc_addr
;
2201 struct page
*page
= NULL
;
2202 pgd_t
*pgd
= pgd_offset_k(addr
);
2207 if (!pgd_none(*pgd
)) {
2208 pud
= pud_offset(pgd
, addr
);
2209 if (!pud_none(*pud
)) {
2210 pmd
= pmd_offset(pud
, addr
);
2211 if (!pmd_none(*pmd
)) {
2212 ptep
= pte_offset_map(pmd
, addr
);
2214 if (pte_present(pte
))
2215 page
= pte_page(pte
);
2223 EXPORT_SYMBOL(vmalloc_to_page
);
2226 * Map a vmalloc()-space virtual address to the physical page frame number.
2228 unsigned long vmalloc_to_pfn(void * vmalloc_addr
)
2230 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
2233 EXPORT_SYMBOL(vmalloc_to_pfn
);
2236 * update_mem_hiwater
2237 * - update per process rss and vm high water data
2239 void update_mem_hiwater(struct task_struct
*tsk
)
2242 unsigned long rss
= get_mm_rss(tsk
->mm
);
2244 if (tsk
->mm
->hiwater_rss
< rss
)
2245 tsk
->mm
->hiwater_rss
= rss
;
2246 if (tsk
->mm
->hiwater_vm
< tsk
->mm
->total_vm
)
2247 tsk
->mm
->hiwater_vm
= tsk
->mm
->total_vm
;
2251 #if !defined(__HAVE_ARCH_GATE_AREA)
2253 #if defined(AT_SYSINFO_EHDR)
2254 static struct vm_area_struct gate_vma
;
2256 static int __init
gate_vma_init(void)
2258 gate_vma
.vm_mm
= NULL
;
2259 gate_vma
.vm_start
= FIXADDR_USER_START
;
2260 gate_vma
.vm_end
= FIXADDR_USER_END
;
2261 gate_vma
.vm_page_prot
= PAGE_READONLY
;
2262 gate_vma
.vm_flags
= VM_RESERVED
;
2265 __initcall(gate_vma_init
);
2268 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
2270 #ifdef AT_SYSINFO_EHDR
2277 int in_gate_area_no_task(unsigned long addr
)
2279 #ifdef AT_SYSINFO_EHDR
2280 if ((addr
>= FIXADDR_USER_START
) && (addr
< FIXADDR_USER_END
))
2286 #endif /* __HAVE_ARCH_GATE_AREA */