4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
41 #include <linux/kernel_stat.h>
43 #include <linux/hugetlb.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/highmem.h>
47 #include <linux/pagemap.h>
48 #include <linux/ksm.h>
49 #include <linux/rmap.h>
50 #include <linux/module.h>
51 #include <linux/delayacct.h>
52 #include <linux/init.h>
53 #include <linux/writeback.h>
54 #include <linux/memcontrol.h>
55 #include <linux/mmu_notifier.h>
56 #include <linux/kallsyms.h>
57 #include <linux/swapops.h>
58 #include <linux/elf.h>
61 #include <asm/pgalloc.h>
62 #include <asm/uaccess.h>
64 #include <asm/tlbflush.h>
65 #include <asm/pgtable.h>
69 #ifndef CONFIG_NEED_MULTIPLE_NODES
70 /* use the per-pgdat data instead for discontigmem - mbligh */
71 unsigned long max_mapnr;
74 EXPORT_SYMBOL(max_mapnr);
75 EXPORT_SYMBOL(mem_map);
78 unsigned long num_physpages;
80 * A number of key systems in x86 including ioremap() rely on the assumption
81 * that high_memory defines the upper bound on direct map memory, then end
82 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
83 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
88 EXPORT_SYMBOL(num_physpages);
89 EXPORT_SYMBOL(high_memory);
92 * Randomize the address space (stacks, mmaps, brk, etc.).
94 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
95 * as ancient (libc5 based) binaries can segfault. )
97 int randomize_va_space __read_mostly =
98 #ifdef CONFIG_COMPAT_BRK
104 static int __init disable_randmaps(char *s)
106 randomize_va_space = 0;
109 __setup("norandmaps", disable_randmaps);
113 * If a p?d_bad entry is found while walking page tables, report
114 * the error, before resetting entry to p?d_none. Usually (but
115 * very seldom) called out from the p?d_none_or_clear_bad macros.
118 void pgd_clear_bad(pgd_t *pgd)
124 void pud_clear_bad(pud_t *pud)
130 void pmd_clear_bad(pmd_t *pmd)
137 * Note: this doesn't free the actual pages themselves. That
138 * has been handled earlier when unmapping all the memory regions.
140 static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
143 pgtable_t token = pmd_pgtable(*pmd);
145 pte_free_tlb(tlb, token, addr);
149 static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
150 unsigned long addr, unsigned long end,
151 unsigned long floor, unsigned long ceiling)
158 pmd = pmd_offset(pud, addr);
160 next = pmd_addr_end(addr, end);
161 if (pmd_none_or_clear_bad(pmd))
163 free_pte_range(tlb, pmd, addr);
164 } while (pmd++, addr = next, addr != end);
174 if (end - 1 > ceiling - 1)
177 pmd = pmd_offset(pud, start);
179 pmd_free_tlb(tlb, pmd, start);
182 static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
183 unsigned long addr, unsigned long end,
184 unsigned long floor, unsigned long ceiling)
191 pud = pud_offset(pgd, addr);
193 next = pud_addr_end(addr, end);
194 if (pud_none_or_clear_bad(pud))
196 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
197 } while (pud++, addr = next, addr != end);
203 ceiling &= PGDIR_MASK;
207 if (end - 1 > ceiling - 1)
210 pud = pud_offset(pgd, start);
212 pud_free_tlb(tlb, pud, start);
216 * This function frees user-level page tables of a process.
218 * Must be called with pagetable lock held.
220 void free_pgd_range(struct mmu_gather *tlb,
221 unsigned long addr, unsigned long end,
222 unsigned long floor, unsigned long ceiling)
229 * The next few lines have given us lots of grief...
231 * Why are we testing PMD* at this top level? Because often
232 * there will be no work to do at all, and we'd prefer not to
233 * go all the way down to the bottom just to discover that.
235 * Why all these "- 1"s? Because 0 represents both the bottom
236 * of the address space and the top of it (using -1 for the
237 * top wouldn't help much: the masks would do the wrong thing).
238 * The rule is that addr 0 and floor 0 refer to the bottom of
239 * the address space, but end 0 and ceiling 0 refer to the top
240 * Comparisons need to use "end - 1" and "ceiling - 1" (though
241 * that end 0 case should be mythical).
243 * Wherever addr is brought up or ceiling brought down, we must
244 * be careful to reject "the opposite 0" before it confuses the
245 * subsequent tests. But what about where end is brought down
246 * by PMD_SIZE below? no, end can't go down to 0 there.
248 * Whereas we round start (addr) and ceiling down, by different
249 * masks at different levels, in order to test whether a table
250 * now has no other vmas using it, so can be freed, we don't
251 * bother to round floor or end up - the tests don't need that.
265 if (end - 1 > ceiling - 1)
271 pgd = pgd_offset(tlb->mm, addr);
273 next = pgd_addr_end(addr, end);
274 if (pgd_none_or_clear_bad(pgd))
276 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
277 } while (pgd++, addr = next, addr != end);
280 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
281 unsigned long floor, unsigned long ceiling)
284 struct vm_area_struct *next = vma->vm_next;
285 unsigned long addr = vma->vm_start;
288 * Hide vma from rmap and vmtruncate before freeing pgtables
290 anon_vma_unlink(vma);
291 unlink_file_vma(vma);
293 if (is_vm_hugetlb_page(vma)) {
294 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
295 floor, next? next->vm_start: ceiling);
298 * Optimization: gather nearby vmas into one call down
300 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
301 && !is_vm_hugetlb_page(next)) {
304 anon_vma_unlink(vma);
305 unlink_file_vma(vma);
307 free_pgd_range(tlb, addr, vma->vm_end,
308 floor, next? next->vm_start: ceiling);
314 int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
316 pgtable_t new = pte_alloc_one(mm, address);
321 * Ensure all pte setup (eg. pte page lock and page clearing) are
322 * visible before the pte is made visible to other CPUs by being
323 * put into page tables.
325 * The other side of the story is the pointer chasing in the page
326 * table walking code (when walking the page table without locking;
327 * ie. most of the time). Fortunately, these data accesses consist
328 * of a chain of data-dependent loads, meaning most CPUs (alpha
329 * being the notable exception) will already guarantee loads are
330 * seen in-order. See the alpha page table accessors for the
331 * smp_read_barrier_depends() barriers in page table walking code.
333 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
335 spin_lock(&mm->page_table_lock);
336 if (!pmd_present(*pmd)) { /* Has another populated it ? */
338 pmd_populate(mm, pmd, new);
341 spin_unlock(&mm->page_table_lock);
347 int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
349 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
353 smp_wmb(); /* See comment in __pte_alloc */
355 spin_lock(&init_mm.page_table_lock);
356 if (!pmd_present(*pmd)) { /* Has another populated it ? */
357 pmd_populate_kernel(&init_mm, pmd, new);
360 spin_unlock(&init_mm.page_table_lock);
362 pte_free_kernel(&init_mm, new);
366 static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
369 add_mm_counter(mm, file_rss, file_rss);
371 add_mm_counter(mm, anon_rss, anon_rss);
375 * This function is called to print an error when a bad pte
376 * is found. For example, we might have a PFN-mapped pte in
377 * a region that doesn't allow it.
379 * The calling function must still handle the error.
381 static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
382 pte_t pte, struct page *page)
384 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
385 pud_t *pud = pud_offset(pgd, addr);
386 pmd_t *pmd = pmd_offset(pud, addr);
387 struct address_space *mapping;
389 static unsigned long resume;
390 static unsigned long nr_shown;
391 static unsigned long nr_unshown;
394 * Allow a burst of 60 reports, then keep quiet for that minute;
395 * or allow a steady drip of one report per second.
397 if (nr_shown == 60) {
398 if (time_before(jiffies, resume)) {
404 "BUG: Bad page map: %lu messages suppressed\n",
411 resume = jiffies + 60 * HZ;
413 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
414 index = linear_page_index(vma, addr);
417 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
419 (long long)pte_val(pte), (long long)pmd_val(*pmd));
422 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
423 page, (void *)page->flags, page_count(page),
424 page_mapcount(page), page->mapping, page->index);
427 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
428 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
430 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
433 print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n",
434 (unsigned long)vma->vm_ops->fault);
435 if (vma->vm_file && vma->vm_file->f_op)
436 print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n",
437 (unsigned long)vma->vm_file->f_op->mmap);
439 add_taint(TAINT_BAD_PAGE);
442 static inline int is_cow_mapping(unsigned int flags)
444 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
448 * vm_normal_page -- This function gets the "struct page" associated with a pte.
450 * "Special" mappings do not wish to be associated with a "struct page" (either
451 * it doesn't exist, or it exists but they don't want to touch it). In this
452 * case, NULL is returned here. "Normal" mappings do have a struct page.
454 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
455 * pte bit, in which case this function is trivial. Secondly, an architecture
456 * may not have a spare pte bit, which requires a more complicated scheme,
459 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
460 * special mapping (even if there are underlying and valid "struct pages").
461 * COWed pages of a VM_PFNMAP are always normal.
463 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
464 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
465 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
466 * mapping will always honor the rule
468 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
470 * And for normal mappings this is false.
472 * This restricts such mappings to be a linear translation from virtual address
473 * to pfn. To get around this restriction, we allow arbitrary mappings so long
474 * as the vma is not a COW mapping; in that case, we know that all ptes are
475 * special (because none can have been COWed).
478 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
480 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
481 * page" backing, however the difference is that _all_ pages with a struct
482 * page (that is, those where pfn_valid is true) are refcounted and considered
483 * normal pages by the VM. The disadvantage is that pages are refcounted
484 * (which can be slower and simply not an option for some PFNMAP users). The
485 * advantage is that we don't have to follow the strict linearity rule of
486 * PFNMAP mappings in order to support COWable mappings.
489 #ifdef __HAVE_ARCH_PTE_SPECIAL
490 # define HAVE_PTE_SPECIAL 1
492 # define HAVE_PTE_SPECIAL 0
494 struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
497 unsigned long pfn = pte_pfn(pte);
499 if (HAVE_PTE_SPECIAL) {
500 if (likely(!pte_special(pte)))
502 if (!(vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)))
503 print_bad_pte(vma, addr, pte, NULL);
507 /* !HAVE_PTE_SPECIAL case follows: */
509 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
510 if (vma->vm_flags & VM_MIXEDMAP) {
516 off = (addr - vma->vm_start) >> PAGE_SHIFT;
517 if (pfn == vma->vm_pgoff + off)
519 if (!is_cow_mapping(vma->vm_flags))
525 if (unlikely(pfn > highest_memmap_pfn)) {
526 print_bad_pte(vma, addr, pte, NULL);
531 * NOTE! We still have PageReserved() pages in the page tables.
532 * eg. VDSO mappings can cause them to exist.
535 return pfn_to_page(pfn);
539 * copy one vm_area from one task to the other. Assumes the page tables
540 * already present in the new task to be cleared in the whole range
541 * covered by this vma.
545 copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
546 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
547 unsigned long addr, int *rss)
549 unsigned long vm_flags = vma->vm_flags;
550 pte_t pte = *src_pte;
553 /* pte contains position in swap or file, so copy. */
554 if (unlikely(!pte_present(pte))) {
555 if (!pte_file(pte)) {
556 swp_entry_t entry = pte_to_swp_entry(pte);
558 swap_duplicate(entry);
559 /* make sure dst_mm is on swapoff's mmlist. */
560 if (unlikely(list_empty(&dst_mm->mmlist))) {
561 spin_lock(&mmlist_lock);
562 if (list_empty(&dst_mm->mmlist))
563 list_add(&dst_mm->mmlist,
565 spin_unlock(&mmlist_lock);
567 if (is_write_migration_entry(entry) &&
568 is_cow_mapping(vm_flags)) {
570 * COW mappings require pages in both parent
571 * and child to be set to read.
573 make_migration_entry_read(&entry);
574 pte = swp_entry_to_pte(entry);
575 set_pte_at(src_mm, addr, src_pte, pte);
582 * If it's a COW mapping, write protect it both
583 * in the parent and the child
585 if (is_cow_mapping(vm_flags)) {
586 ptep_set_wrprotect(src_mm, addr, src_pte);
587 pte = pte_wrprotect(pte);
591 * If it's a shared mapping, mark it clean in
594 if (vm_flags & VM_SHARED)
595 pte = pte_mkclean(pte);
596 pte = pte_mkold(pte);
598 page = vm_normal_page(vma, addr, pte);
602 rss[PageAnon(page)]++;
606 set_pte_at(dst_mm, addr, dst_pte, pte);
609 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
610 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
611 unsigned long addr, unsigned long end)
613 pte_t *src_pte, *dst_pte;
614 spinlock_t *src_ptl, *dst_ptl;
620 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
623 src_pte = pte_offset_map_nested(src_pmd, addr);
624 src_ptl = pte_lockptr(src_mm, src_pmd);
625 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
626 arch_enter_lazy_mmu_mode();
630 * We are holding two locks at this point - either of them
631 * could generate latencies in another task on another CPU.
633 if (progress >= 32) {
635 if (need_resched() ||
636 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
639 if (pte_none(*src_pte)) {
643 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
645 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
647 arch_leave_lazy_mmu_mode();
648 spin_unlock(src_ptl);
649 pte_unmap_nested(src_pte - 1);
650 add_mm_rss(dst_mm, rss[0], rss[1]);
651 pte_unmap_unlock(dst_pte - 1, dst_ptl);
658 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
659 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
660 unsigned long addr, unsigned long end)
662 pmd_t *src_pmd, *dst_pmd;
665 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
668 src_pmd = pmd_offset(src_pud, addr);
670 next = pmd_addr_end(addr, end);
671 if (pmd_none_or_clear_bad(src_pmd))
673 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
676 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
680 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
681 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
682 unsigned long addr, unsigned long end)
684 pud_t *src_pud, *dst_pud;
687 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
690 src_pud = pud_offset(src_pgd, addr);
692 next = pud_addr_end(addr, end);
693 if (pud_none_or_clear_bad(src_pud))
695 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
698 } while (dst_pud++, src_pud++, addr = next, addr != end);
702 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
703 struct vm_area_struct *vma)
705 pgd_t *src_pgd, *dst_pgd;
707 unsigned long addr = vma->vm_start;
708 unsigned long end = vma->vm_end;
712 * Don't copy ptes where a page fault will fill them correctly.
713 * Fork becomes much lighter when there are big shared or private
714 * readonly mappings. The tradeoff is that copy_page_range is more
715 * efficient than faulting.
717 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
722 if (is_vm_hugetlb_page(vma))
723 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
725 if (unlikely(is_pfn_mapping(vma))) {
727 * We do not free on error cases below as remove_vma
728 * gets called on error from higher level routine
730 ret = track_pfn_vma_copy(vma);
736 * We need to invalidate the secondary MMU mappings only when
737 * there could be a permission downgrade on the ptes of the
738 * parent mm. And a permission downgrade will only happen if
739 * is_cow_mapping() returns true.
741 if (is_cow_mapping(vma->vm_flags))
742 mmu_notifier_invalidate_range_start(src_mm, addr, end);
745 dst_pgd = pgd_offset(dst_mm, addr);
746 src_pgd = pgd_offset(src_mm, addr);
748 next = pgd_addr_end(addr, end);
749 if (pgd_none_or_clear_bad(src_pgd))
751 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
756 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
758 if (is_cow_mapping(vma->vm_flags))
759 mmu_notifier_invalidate_range_end(src_mm,
764 static unsigned long zap_pte_range(struct mmu_gather *tlb,
765 struct vm_area_struct *vma, pmd_t *pmd,
766 unsigned long addr, unsigned long end,
767 long *zap_work, struct zap_details *details)
769 struct mm_struct *mm = tlb->mm;
775 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
776 arch_enter_lazy_mmu_mode();
779 if (pte_none(ptent)) {
784 (*zap_work) -= PAGE_SIZE;
786 if (pte_present(ptent)) {
789 page = vm_normal_page(vma, addr, ptent);
790 if (unlikely(details) && page) {
792 * unmap_shared_mapping_pages() wants to
793 * invalidate cache without truncating:
794 * unmap shared but keep private pages.
796 if (details->check_mapping &&
797 details->check_mapping != page->mapping)
800 * Each page->index must be checked when
801 * invalidating or truncating nonlinear.
803 if (details->nonlinear_vma &&
804 (page->index < details->first_index ||
805 page->index > details->last_index))
808 ptent = ptep_get_and_clear_full(mm, addr, pte,
810 tlb_remove_tlb_entry(tlb, pte, addr);
813 if (unlikely(details) && details->nonlinear_vma
814 && linear_page_index(details->nonlinear_vma,
815 addr) != page->index)
816 set_pte_at(mm, addr, pte,
817 pgoff_to_pte(page->index));
821 if (pte_dirty(ptent))
822 set_page_dirty(page);
823 if (pte_young(ptent) &&
824 likely(!VM_SequentialReadHint(vma)))
825 mark_page_accessed(page);
828 page_remove_rmap(page);
829 if (unlikely(page_mapcount(page) < 0))
830 print_bad_pte(vma, addr, ptent, page);
831 tlb_remove_page(tlb, page);
835 * If details->check_mapping, we leave swap entries;
836 * if details->nonlinear_vma, we leave file entries.
838 if (unlikely(details))
840 if (pte_file(ptent)) {
841 if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))
842 print_bad_pte(vma, addr, ptent, NULL);
844 (unlikely(!free_swap_and_cache(pte_to_swp_entry(ptent))))
845 print_bad_pte(vma, addr, ptent, NULL);
846 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
847 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
849 add_mm_rss(mm, file_rss, anon_rss);
850 arch_leave_lazy_mmu_mode();
851 pte_unmap_unlock(pte - 1, ptl);
856 static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
857 struct vm_area_struct *vma, pud_t *pud,
858 unsigned long addr, unsigned long end,
859 long *zap_work, struct zap_details *details)
864 pmd = pmd_offset(pud, addr);
866 next = pmd_addr_end(addr, end);
867 if (pmd_none_or_clear_bad(pmd)) {
871 next = zap_pte_range(tlb, vma, pmd, addr, next,
873 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
878 static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
879 struct vm_area_struct *vma, pgd_t *pgd,
880 unsigned long addr, unsigned long end,
881 long *zap_work, struct zap_details *details)
886 pud = pud_offset(pgd, addr);
888 next = pud_addr_end(addr, end);
889 if (pud_none_or_clear_bad(pud)) {
893 next = zap_pmd_range(tlb, vma, pud, addr, next,
895 } while (pud++, addr = next, (addr != end && *zap_work > 0));
900 static unsigned long unmap_page_range(struct mmu_gather *tlb,
901 struct vm_area_struct *vma,
902 unsigned long addr, unsigned long end,
903 long *zap_work, struct zap_details *details)
908 if (details && !details->check_mapping && !details->nonlinear_vma)
912 tlb_start_vma(tlb, vma);
913 pgd = pgd_offset(vma->vm_mm, addr);
915 next = pgd_addr_end(addr, end);
916 if (pgd_none_or_clear_bad(pgd)) {
920 next = zap_pud_range(tlb, vma, pgd, addr, next,
922 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
923 tlb_end_vma(tlb, vma);
928 #ifdef CONFIG_PREEMPT
929 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
931 /* No preempt: go for improved straight-line efficiency */
932 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
936 * unmap_vmas - unmap a range of memory covered by a list of vma's
937 * @tlbp: address of the caller's struct mmu_gather
938 * @vma: the starting vma
939 * @start_addr: virtual address at which to start unmapping
940 * @end_addr: virtual address at which to end unmapping
941 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
942 * @details: details of nonlinear truncation or shared cache invalidation
944 * Returns the end address of the unmapping (restart addr if interrupted).
946 * Unmap all pages in the vma list.
948 * We aim to not hold locks for too long (for scheduling latency reasons).
949 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
950 * return the ending mmu_gather to the caller.
952 * Only addresses between `start' and `end' will be unmapped.
954 * The VMA list must be sorted in ascending virtual address order.
956 * unmap_vmas() assumes that the caller will flush the whole unmapped address
957 * range after unmap_vmas() returns. So the only responsibility here is to
958 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
959 * drops the lock and schedules.
961 unsigned long unmap_vmas(struct mmu_gather **tlbp,
962 struct vm_area_struct *vma, unsigned long start_addr,
963 unsigned long end_addr, unsigned long *nr_accounted,
964 struct zap_details *details)
966 long zap_work = ZAP_BLOCK_SIZE;
967 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
968 int tlb_start_valid = 0;
969 unsigned long start = start_addr;
970 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
971 int fullmm = (*tlbp)->fullmm;
972 struct mm_struct *mm = vma->vm_mm;
974 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
975 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
978 start = max(vma->vm_start, start_addr);
979 if (start >= vma->vm_end)
981 end = min(vma->vm_end, end_addr);
982 if (end <= vma->vm_start)
985 if (vma->vm_flags & VM_ACCOUNT)
986 *nr_accounted += (end - start) >> PAGE_SHIFT;
988 if (unlikely(is_pfn_mapping(vma)))
989 untrack_pfn_vma(vma, 0, 0);
991 while (start != end) {
992 if (!tlb_start_valid) {
997 if (unlikely(is_vm_hugetlb_page(vma))) {
999 * It is undesirable to test vma->vm_file as it
1000 * should be non-null for valid hugetlb area.
1001 * However, vm_file will be NULL in the error
1002 * cleanup path of do_mmap_pgoff. When
1003 * hugetlbfs ->mmap method fails,
1004 * do_mmap_pgoff() nullifies vma->vm_file
1005 * before calling this function to clean up.
1006 * Since no pte has actually been setup, it is
1007 * safe to do nothing in this case.
1010 unmap_hugepage_range(vma, start, end, NULL);
1011 zap_work -= (end - start) /
1012 pages_per_huge_page(hstate_vma(vma));
1017 start = unmap_page_range(*tlbp, vma,
1018 start, end, &zap_work, details);
1021 BUG_ON(start != end);
1025 tlb_finish_mmu(*tlbp, tlb_start, start);
1027 if (need_resched() ||
1028 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
1036 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
1037 tlb_start_valid = 0;
1038 zap_work = ZAP_BLOCK_SIZE;
1042 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
1043 return start; /* which is now the end (or restart) address */
1047 * zap_page_range - remove user pages in a given range
1048 * @vma: vm_area_struct holding the applicable pages
1049 * @address: starting address of pages to zap
1050 * @size: number of bytes to zap
1051 * @details: details of nonlinear truncation or shared cache invalidation
1053 unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
1054 unsigned long size, struct zap_details *details)
1056 struct mm_struct *mm = vma->vm_mm;
1057 struct mmu_gather *tlb;
1058 unsigned long end = address + size;
1059 unsigned long nr_accounted = 0;
1062 tlb = tlb_gather_mmu(mm, 0);
1063 update_hiwater_rss(mm);
1064 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
1066 tlb_finish_mmu(tlb, address, end);
1071 * zap_vma_ptes - remove ptes mapping the vma
1072 * @vma: vm_area_struct holding ptes to be zapped
1073 * @address: starting address of pages to zap
1074 * @size: number of bytes to zap
1076 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1078 * The entire address range must be fully contained within the vma.
1080 * Returns 0 if successful.
1082 int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1085 if (address < vma->vm_start || address + size > vma->vm_end ||
1086 !(vma->vm_flags & VM_PFNMAP))
1088 zap_page_range(vma, address, size, NULL);
1091 EXPORT_SYMBOL_GPL(zap_vma_ptes);
1094 * Do a quick page-table lookup for a single page.
1096 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1105 struct mm_struct *mm = vma->vm_mm;
1107 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
1108 if (!IS_ERR(page)) {
1109 BUG_ON(flags & FOLL_GET);
1114 pgd = pgd_offset(mm, address);
1115 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
1118 pud = pud_offset(pgd, address);
1121 if (pud_huge(*pud)) {
1122 BUG_ON(flags & FOLL_GET);
1123 page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
1126 if (unlikely(pud_bad(*pud)))
1129 pmd = pmd_offset(pud, address);
1132 if (pmd_huge(*pmd)) {
1133 BUG_ON(flags & FOLL_GET);
1134 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
1137 if (unlikely(pmd_bad(*pmd)))
1140 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
1143 if (!pte_present(pte))
1145 if ((flags & FOLL_WRITE) && !pte_write(pte))
1147 page = vm_normal_page(vma, address, pte);
1148 if (unlikely(!page))
1151 if (flags & FOLL_GET)
1153 if (flags & FOLL_TOUCH) {
1154 if ((flags & FOLL_WRITE) &&
1155 !pte_dirty(pte) && !PageDirty(page))
1156 set_page_dirty(page);
1158 * pte_mkyoung() would be more correct here, but atomic care
1159 * is needed to avoid losing the dirty bit: it is easier to use
1160 * mark_page_accessed().
1162 mark_page_accessed(page);
1165 pte_unmap_unlock(ptep, ptl);
1170 pte_unmap_unlock(ptep, ptl);
1171 return ERR_PTR(-EFAULT);
1174 pte_unmap_unlock(ptep, ptl);
1177 /* Fall through to ZERO_PAGE handling */
1180 * When core dumping an enormous anonymous area that nobody
1181 * has touched so far, we don't want to allocate page tables.
1183 if (flags & FOLL_ANON) {
1184 page = ZERO_PAGE(0);
1185 if (flags & FOLL_GET)
1187 BUG_ON(flags & FOLL_WRITE);
1192 /* Can we do the FOLL_ANON optimization? */
1193 static inline int use_zero_page(struct vm_area_struct *vma)
1196 * We don't want to optimize FOLL_ANON for make_pages_present()
1197 * when it tries to page in a VM_LOCKED region. As to VM_SHARED,
1198 * we want to get the page from the page tables to make sure
1199 * that we serialize and update with any other user of that
1202 if (vma->vm_flags & (VM_LOCKED | VM_SHARED))
1205 * And if we have a fault routine, it's not an anonymous region.
1207 return !vma->vm_ops || !vma->vm_ops->fault;
1212 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1213 unsigned long start, int nr_pages, int flags,
1214 struct page **pages, struct vm_area_struct **vmas)
1217 unsigned int vm_flags = 0;
1218 int write = !!(flags & GUP_FLAGS_WRITE);
1219 int force = !!(flags & GUP_FLAGS_FORCE);
1224 * Require read or write permissions.
1225 * If 'force' is set, we only require the "MAY" flags.
1227 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1228 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
1232 struct vm_area_struct *vma;
1233 unsigned int foll_flags;
1235 vma = find_extend_vma(mm, start);
1236 if (!vma && in_gate_area(tsk, start)) {
1237 unsigned long pg = start & PAGE_MASK;
1238 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1244 /* user gate pages are read-only */
1246 return i ? : -EFAULT;
1248 pgd = pgd_offset_k(pg);
1250 pgd = pgd_offset_gate(mm, pg);
1251 BUG_ON(pgd_none(*pgd));
1252 pud = pud_offset(pgd, pg);
1253 BUG_ON(pud_none(*pud));
1254 pmd = pmd_offset(pud, pg);
1256 return i ? : -EFAULT;
1257 pte = pte_offset_map(pmd, pg);
1258 if (pte_none(*pte)) {
1260 return i ? : -EFAULT;
1263 struct page *page = vm_normal_page(gate_vma, start, *pte);
1278 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
1279 !(vm_flags & vma->vm_flags))
1280 return i ? : -EFAULT;
1282 if (is_vm_hugetlb_page(vma)) {
1283 i = follow_hugetlb_page(mm, vma, pages, vmas,
1284 &start, &nr_pages, i, write);
1288 foll_flags = FOLL_TOUCH;
1290 foll_flags |= FOLL_GET;
1291 if (!write && use_zero_page(vma))
1292 foll_flags |= FOLL_ANON;
1298 * If we have a pending SIGKILL, don't keep faulting
1299 * pages and potentially allocating memory.
1301 if (unlikely(fatal_signal_pending(current)))
1302 return i ? i : -ERESTARTSYS;
1305 foll_flags |= FOLL_WRITE;
1308 while (!(page = follow_page(vma, start, foll_flags))) {
1311 ret = handle_mm_fault(mm, vma, start,
1312 (foll_flags & FOLL_WRITE) ?
1313 FAULT_FLAG_WRITE : 0);
1315 if (ret & VM_FAULT_ERROR) {
1316 if (ret & VM_FAULT_OOM)
1317 return i ? i : -ENOMEM;
1318 else if (ret & VM_FAULT_SIGBUS)
1319 return i ? i : -EFAULT;
1322 if (ret & VM_FAULT_MAJOR)
1328 * The VM_FAULT_WRITE bit tells us that
1329 * do_wp_page has broken COW when necessary,
1330 * even if maybe_mkwrite decided not to set
1331 * pte_write. We can thus safely do subsequent
1332 * page lookups as if they were reads. But only
1333 * do so when looping for pte_write is futile:
1334 * in some cases userspace may also be wanting
1335 * to write to the gotten user page, which a
1336 * read fault here might prevent (a readonly
1337 * page might get reCOWed by userspace write).
1339 if ((ret & VM_FAULT_WRITE) &&
1340 !(vma->vm_flags & VM_WRITE))
1341 foll_flags &= ~FOLL_WRITE;
1346 return i ? i : PTR_ERR(page);
1350 flush_anon_page(vma, page, start);
1351 flush_dcache_page(page);
1358 } while (nr_pages && start < vma->vm_end);
1364 * get_user_pages() - pin user pages in memory
1365 * @tsk: task_struct of target task
1366 * @mm: mm_struct of target mm
1367 * @start: starting user address
1368 * @nr_pages: number of pages from start to pin
1369 * @write: whether pages will be written to by the caller
1370 * @force: whether to force write access even if user mapping is
1371 * readonly. This will result in the page being COWed even
1372 * in MAP_SHARED mappings. You do not want this.
1373 * @pages: array that receives pointers to the pages pinned.
1374 * Should be at least nr_pages long. Or NULL, if caller
1375 * only intends to ensure the pages are faulted in.
1376 * @vmas: array of pointers to vmas corresponding to each page.
1377 * Or NULL if the caller does not require them.
1379 * Returns number of pages pinned. This may be fewer than the number
1380 * requested. If nr_pages is 0 or negative, returns 0. If no pages
1381 * were pinned, returns -errno. Each page returned must be released
1382 * with a put_page() call when it is finished with. vmas will only
1383 * remain valid while mmap_sem is held.
1385 * Must be called with mmap_sem held for read or write.
1387 * get_user_pages walks a process's page tables and takes a reference to
1388 * each struct page that each user address corresponds to at a given
1389 * instant. That is, it takes the page that would be accessed if a user
1390 * thread accesses the given user virtual address at that instant.
1392 * This does not guarantee that the page exists in the user mappings when
1393 * get_user_pages returns, and there may even be a completely different
1394 * page there in some cases (eg. if mmapped pagecache has been invalidated
1395 * and subsequently re faulted). However it does guarantee that the page
1396 * won't be freed completely. And mostly callers simply care that the page
1397 * contains data that was valid *at some point in time*. Typically, an IO
1398 * or similar operation cannot guarantee anything stronger anyway because
1399 * locks can't be held over the syscall boundary.
1401 * If write=0, the page must not be written to. If the page is written to,
1402 * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
1403 * after the page is finished with, and before put_page is called.
1405 * get_user_pages is typically used for fewer-copy IO operations, to get a
1406 * handle on the memory by some means other than accesses via the user virtual
1407 * addresses. The pages may be submitted for DMA to devices or accessed via
1408 * their kernel linear mapping (via the kmap APIs). Care should be taken to
1409 * use the correct cache flushing APIs.
1411 * See also get_user_pages_fast, for performance critical applications.
1413 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1414 unsigned long start, int nr_pages, int write, int force,
1415 struct page **pages, struct vm_area_struct **vmas)
1420 flags |= GUP_FLAGS_WRITE;
1422 flags |= GUP_FLAGS_FORCE;
1424 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
1426 EXPORT_SYMBOL(get_user_pages);
1429 * get_dump_page() - pin user page in memory while writing it to core dump
1430 * @addr: user address
1432 * Returns struct page pointer of user page pinned for dump,
1433 * to be freed afterwards by page_cache_release() or put_page().
1435 * Returns NULL on any kind of failure - a hole must then be inserted into
1436 * the corefile, to preserve alignment with its headers; and also returns
1437 * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
1438 * allowing a hole to be left in the corefile to save diskspace.
1440 * Called without mmap_sem, but after all other threads have been killed.
1442 #ifdef CONFIG_ELF_CORE
1443 struct page *get_dump_page(unsigned long addr)
1445 struct vm_area_struct *vma;
1448 if (__get_user_pages(current, current->mm, addr, 1,
1449 GUP_FLAGS_FORCE, &page, &vma) < 1)
1451 if (page == ZERO_PAGE(0)) {
1452 page_cache_release(page);
1455 flush_cache_page(vma, addr, page_to_pfn(page));
1458 #endif /* CONFIG_ELF_CORE */
1460 pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1463 pgd_t * pgd = pgd_offset(mm, addr);
1464 pud_t * pud = pud_alloc(mm, pgd, addr);
1466 pmd_t * pmd = pmd_alloc(mm, pud, addr);
1468 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1474 * This is the old fallback for page remapping.
1476 * For historical reasons, it only allows reserved pages. Only
1477 * old drivers should use this, and they needed to mark their
1478 * pages reserved for the old functions anyway.
1480 static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1481 struct page *page, pgprot_t prot)
1483 struct mm_struct *mm = vma->vm_mm;
1492 flush_dcache_page(page);
1493 pte = get_locked_pte(mm, addr, &ptl);
1497 if (!pte_none(*pte))
1500 /* Ok, finally just insert the thing.. */
1502 inc_mm_counter(mm, file_rss);
1503 page_add_file_rmap(page);
1504 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1507 pte_unmap_unlock(pte, ptl);
1510 pte_unmap_unlock(pte, ptl);
1516 * vm_insert_page - insert single page into user vma
1517 * @vma: user vma to map to
1518 * @addr: target user address of this page
1519 * @page: source kernel page
1521 * This allows drivers to insert individual pages they've allocated
1524 * The page has to be a nice clean _individual_ kernel allocation.
1525 * If you allocate a compound page, you need to have marked it as
1526 * such (__GFP_COMP), or manually just split the page up yourself
1527 * (see split_page()).
1529 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1530 * took an arbitrary page protection parameter. This doesn't allow
1531 * that. Your vma protection will have to be set up correctly, which
1532 * means that if you want a shared writable mapping, you'd better
1533 * ask for a shared writable mapping!
1535 * The page does not need to be reserved.
1537 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1540 if (addr < vma->vm_start || addr >= vma->vm_end)
1542 if (!page_count(page))
1544 vma->vm_flags |= VM_INSERTPAGE;
1545 return insert_page(vma, addr, page, vma->vm_page_prot);
1547 EXPORT_SYMBOL(vm_insert_page);
1549 static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1550 unsigned long pfn, pgprot_t prot)
1552 struct mm_struct *mm = vma->vm_mm;
1558 pte = get_locked_pte(mm, addr, &ptl);
1562 if (!pte_none(*pte))
1565 /* Ok, finally just insert the thing.. */
1566 entry = pte_mkspecial(pfn_pte(pfn, prot));
1567 set_pte_at(mm, addr, pte, entry);
1568 update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */
1572 pte_unmap_unlock(pte, ptl);
1578 * vm_insert_pfn - insert single pfn into user vma
1579 * @vma: user vma to map to
1580 * @addr: target user address of this page
1581 * @pfn: source kernel pfn
1583 * Similar to vm_inert_page, this allows drivers to insert individual pages
1584 * they've allocated into a user vma. Same comments apply.
1586 * This function should only be called from a vm_ops->fault handler, and
1587 * in that case the handler should return NULL.
1589 * vma cannot be a COW mapping.
1591 * As this is called only for pages that do not currently exist, we
1592 * do not need to flush old virtual caches or the TLB.
1594 int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1598 pgprot_t pgprot = vma->vm_page_prot;
1600 * Technically, architectures with pte_special can avoid all these
1601 * restrictions (same for remap_pfn_range). However we would like
1602 * consistency in testing and feature parity among all, so we should
1603 * try to keep these invariants in place for everybody.
1605 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1606 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1607 (VM_PFNMAP|VM_MIXEDMAP));
1608 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1609 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
1611 if (addr < vma->vm_start || addr >= vma->vm_end)
1613 if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE))
1616 ret = insert_pfn(vma, addr, pfn, pgprot);
1619 untrack_pfn_vma(vma, pfn, PAGE_SIZE);
1623 EXPORT_SYMBOL(vm_insert_pfn);
1625 int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1628 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1630 if (addr < vma->vm_start || addr >= vma->vm_end)
1634 * If we don't have pte special, then we have to use the pfn_valid()
1635 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1636 * refcount the page if pfn_valid is true (hence insert_page rather
1639 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1642 page = pfn_to_page(pfn);
1643 return insert_page(vma, addr, page, vma->vm_page_prot);
1645 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1647 EXPORT_SYMBOL(vm_insert_mixed);
1650 * maps a range of physical memory into the requested pages. the old
1651 * mappings are removed. any references to nonexistent pages results
1652 * in null mappings (currently treated as "copy-on-access")
1654 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1655 unsigned long addr, unsigned long end,
1656 unsigned long pfn, pgprot_t prot)
1661 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1664 arch_enter_lazy_mmu_mode();
1666 BUG_ON(!pte_none(*pte));
1667 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
1669 } while (pte++, addr += PAGE_SIZE, addr != end);
1670 arch_leave_lazy_mmu_mode();
1671 pte_unmap_unlock(pte - 1, ptl);
1675 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1676 unsigned long addr, unsigned long end,
1677 unsigned long pfn, pgprot_t prot)
1682 pfn -= addr >> PAGE_SHIFT;
1683 pmd = pmd_alloc(mm, pud, addr);
1687 next = pmd_addr_end(addr, end);
1688 if (remap_pte_range(mm, pmd, addr, next,
1689 pfn + (addr >> PAGE_SHIFT), prot))
1691 } while (pmd++, addr = next, addr != end);
1695 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1696 unsigned long addr, unsigned long end,
1697 unsigned long pfn, pgprot_t prot)
1702 pfn -= addr >> PAGE_SHIFT;
1703 pud = pud_alloc(mm, pgd, addr);
1707 next = pud_addr_end(addr, end);
1708 if (remap_pmd_range(mm, pud, addr, next,
1709 pfn + (addr >> PAGE_SHIFT), prot))
1711 } while (pud++, addr = next, addr != end);
1716 * remap_pfn_range - remap kernel memory to userspace
1717 * @vma: user vma to map to
1718 * @addr: target user address to start at
1719 * @pfn: physical address of kernel memory
1720 * @size: size of map area
1721 * @prot: page protection flags for this mapping
1723 * Note: this is only safe if the mm semaphore is held when called.
1725 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1726 unsigned long pfn, unsigned long size, pgprot_t prot)
1730 unsigned long end = addr + PAGE_ALIGN(size);
1731 struct mm_struct *mm = vma->vm_mm;
1735 * Physically remapped pages are special. Tell the
1736 * rest of the world about it:
1737 * VM_IO tells people not to look at these pages
1738 * (accesses can have side effects).
1739 * VM_RESERVED is specified all over the place, because
1740 * in 2.4 it kept swapout's vma scan off this vma; but
1741 * in 2.6 the LRU scan won't even find its pages, so this
1742 * flag means no more than count its pages in reserved_vm,
1743 * and omit it from core dump, even when VM_IO turned off.
1744 * VM_PFNMAP tells the core MM that the base pages are just
1745 * raw PFN mappings, and do not have a "struct page" associated
1748 * There's a horrible special case to handle copy-on-write
1749 * behaviour that some programs depend on. We mark the "original"
1750 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1752 if (addr == vma->vm_start && end == vma->vm_end) {
1753 vma->vm_pgoff = pfn;
1754 vma->vm_flags |= VM_PFN_AT_MMAP;
1755 } else if (is_cow_mapping(vma->vm_flags))
1758 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
1760 err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size));
1763 * To indicate that track_pfn related cleanup is not
1764 * needed from higher level routine calling unmap_vmas
1766 vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP);
1767 vma->vm_flags &= ~VM_PFN_AT_MMAP;
1771 BUG_ON(addr >= end);
1772 pfn -= addr >> PAGE_SHIFT;
1773 pgd = pgd_offset(mm, addr);
1774 flush_cache_range(vma, addr, end);
1776 next = pgd_addr_end(addr, end);
1777 err = remap_pud_range(mm, pgd, addr, next,
1778 pfn + (addr >> PAGE_SHIFT), prot);
1781 } while (pgd++, addr = next, addr != end);
1784 untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size));
1788 EXPORT_SYMBOL(remap_pfn_range);
1790 static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1791 unsigned long addr, unsigned long end,
1792 pte_fn_t fn, void *data)
1797 spinlock_t *uninitialized_var(ptl);
1799 pte = (mm == &init_mm) ?
1800 pte_alloc_kernel(pmd, addr) :
1801 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1805 BUG_ON(pmd_huge(*pmd));
1807 arch_enter_lazy_mmu_mode();
1809 token = pmd_pgtable(*pmd);
1812 err = fn(pte, token, addr, data);
1815 } while (pte++, addr += PAGE_SIZE, addr != end);
1817 arch_leave_lazy_mmu_mode();
1820 pte_unmap_unlock(pte-1, ptl);
1824 static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1825 unsigned long addr, unsigned long end,
1826 pte_fn_t fn, void *data)
1832 BUG_ON(pud_huge(*pud));
1834 pmd = pmd_alloc(mm, pud, addr);
1838 next = pmd_addr_end(addr, end);
1839 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1842 } while (pmd++, addr = next, addr != end);
1846 static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1847 unsigned long addr, unsigned long end,
1848 pte_fn_t fn, void *data)
1854 pud = pud_alloc(mm, pgd, addr);
1858 next = pud_addr_end(addr, end);
1859 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1862 } while (pud++, addr = next, addr != end);
1867 * Scan a region of virtual memory, filling in page tables as necessary
1868 * and calling a provided function on each leaf page table.
1870 int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1871 unsigned long size, pte_fn_t fn, void *data)
1875 unsigned long start = addr, end = addr + size;
1878 BUG_ON(addr >= end);
1879 mmu_notifier_invalidate_range_start(mm, start, end);
1880 pgd = pgd_offset(mm, addr);
1882 next = pgd_addr_end(addr, end);
1883 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1886 } while (pgd++, addr = next, addr != end);
1887 mmu_notifier_invalidate_range_end(mm, start, end);
1890 EXPORT_SYMBOL_GPL(apply_to_page_range);
1893 * handle_pte_fault chooses page fault handler according to an entry
1894 * which was read non-atomically. Before making any commitment, on
1895 * those architectures or configurations (e.g. i386 with PAE) which
1896 * might give a mix of unmatched parts, do_swap_page and do_file_page
1897 * must check under lock before unmapping the pte and proceeding
1898 * (but do_wp_page is only called after already making such a check;
1899 * and do_anonymous_page and do_no_page can safely check later on).
1901 static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
1902 pte_t *page_table, pte_t orig_pte)
1905 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1906 if (sizeof(pte_t) > sizeof(unsigned long)) {
1907 spinlock_t *ptl = pte_lockptr(mm, pmd);
1909 same = pte_same(*page_table, orig_pte);
1913 pte_unmap(page_table);
1918 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1919 * servicing faults for write access. In the normal case, do always want
1920 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1921 * that do not have writing enabled, when used by access_process_vm.
1923 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1925 if (likely(vma->vm_flags & VM_WRITE))
1926 pte = pte_mkwrite(pte);
1930 static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
1933 * If the source page was a PFN mapping, we don't have
1934 * a "struct page" for it. We do a best-effort copy by
1935 * just copying from the original user address. If that
1936 * fails, we just zero-fill it. Live with it.
1938 if (unlikely(!src)) {
1939 void *kaddr = kmap_atomic(dst, KM_USER0);
1940 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1943 * This really shouldn't fail, because the page is there
1944 * in the page tables. But it might just be unreadable,
1945 * in which case we just give up and fill the result with
1948 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
1949 memset(kaddr, 0, PAGE_SIZE);
1950 kunmap_atomic(kaddr, KM_USER0);
1951 flush_dcache_page(dst);
1953 copy_user_highpage(dst, src, va, vma);
1957 * This routine handles present pages, when users try to write
1958 * to a shared page. It is done by copying the page to a new address
1959 * and decrementing the shared-page counter for the old page.
1961 * Note that this routine assumes that the protection checks have been
1962 * done by the caller (the low-level page fault routine in most cases).
1963 * Thus we can safely just mark it writable once we've done any necessary
1966 * We also mark the page dirty at this point even though the page will
1967 * change only once the write actually happens. This avoids a few races,
1968 * and potentially makes it more efficient.
1970 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1971 * but allow concurrent faults), with pte both mapped and locked.
1972 * We return with mmap_sem still held, but pte unmapped and unlocked.
1974 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1975 unsigned long address, pte_t *page_table, pmd_t *pmd,
1976 spinlock_t *ptl, pte_t orig_pte)
1978 struct page *old_page, *new_page;
1980 int reuse = 0, ret = 0;
1981 int page_mkwrite = 0;
1982 struct page *dirty_page = NULL;
1984 old_page = vm_normal_page(vma, address, orig_pte);
1987 * VM_MIXEDMAP !pfn_valid() case
1989 * We should not cow pages in a shared writeable mapping.
1990 * Just mark the pages writable as we can't do any dirty
1991 * accounting on raw pfn maps.
1993 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
1994 (VM_WRITE|VM_SHARED))
2000 * Take out anonymous pages first, anonymous shared vmas are
2001 * not dirty accountable.
2003 if (PageAnon(old_page) && !PageKsm(old_page)) {
2004 if (!trylock_page(old_page)) {
2005 page_cache_get(old_page);
2006 pte_unmap_unlock(page_table, ptl);
2007 lock_page(old_page);
2008 page_table = pte_offset_map_lock(mm, pmd, address,
2010 if (!pte_same(*page_table, orig_pte)) {
2011 unlock_page(old_page);
2012 page_cache_release(old_page);
2015 page_cache_release(old_page);
2017 reuse = reuse_swap_page(old_page);
2018 unlock_page(old_page);
2019 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2020 (VM_WRITE|VM_SHARED))) {
2022 * Only catch write-faults on shared writable pages,
2023 * read-only shared pages can get COWed by
2024 * get_user_pages(.write=1, .force=1).
2026 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
2027 struct vm_fault vmf;
2030 vmf.virtual_address = (void __user *)(address &
2032 vmf.pgoff = old_page->index;
2033 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2034 vmf.page = old_page;
2037 * Notify the address space that the page is about to
2038 * become writable so that it can prohibit this or wait
2039 * for the page to get into an appropriate state.
2041 * We do this without the lock held, so that it can
2042 * sleep if it needs to.
2044 page_cache_get(old_page);
2045 pte_unmap_unlock(page_table, ptl);
2047 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2049 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2051 goto unwritable_page;
2053 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2054 lock_page(old_page);
2055 if (!old_page->mapping) {
2056 ret = 0; /* retry the fault */
2057 unlock_page(old_page);
2058 goto unwritable_page;
2061 VM_BUG_ON(!PageLocked(old_page));
2064 * Since we dropped the lock we need to revalidate
2065 * the PTE as someone else may have changed it. If
2066 * they did, we just return, as we can count on the
2067 * MMU to tell us if they didn't also make it writable.
2069 page_table = pte_offset_map_lock(mm, pmd, address,
2071 if (!pte_same(*page_table, orig_pte)) {
2072 unlock_page(old_page);
2073 page_cache_release(old_page);
2079 dirty_page = old_page;
2080 get_page(dirty_page);
2086 flush_cache_page(vma, address, pte_pfn(orig_pte));
2087 entry = pte_mkyoung(orig_pte);
2088 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2089 if (ptep_set_access_flags(vma, address, page_table, entry,1))
2090 update_mmu_cache(vma, address, entry);
2091 ret |= VM_FAULT_WRITE;
2096 * Ok, we need to copy. Oh, well..
2098 page_cache_get(old_page);
2100 pte_unmap_unlock(page_table, ptl);
2102 if (unlikely(anon_vma_prepare(vma)))
2104 VM_BUG_ON(old_page == ZERO_PAGE(0));
2105 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2109 * Don't let another task, with possibly unlocked vma,
2110 * keep the mlocked page.
2112 if ((vma->vm_flags & VM_LOCKED) && old_page) {
2113 lock_page(old_page); /* for LRU manipulation */
2114 clear_page_mlock(old_page);
2115 unlock_page(old_page);
2117 cow_user_page(new_page, old_page, address, vma);
2118 __SetPageUptodate(new_page);
2120 if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
2124 * Re-check the pte - we dropped the lock
2126 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2127 if (likely(pte_same(*page_table, orig_pte))) {
2129 if (!PageAnon(old_page)) {
2130 dec_mm_counter(mm, file_rss);
2131 inc_mm_counter(mm, anon_rss);
2134 inc_mm_counter(mm, anon_rss);
2135 flush_cache_page(vma, address, pte_pfn(orig_pte));
2136 entry = mk_pte(new_page, vma->vm_page_prot);
2137 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2139 * Clear the pte entry and flush it first, before updating the
2140 * pte with the new entry. This will avoid a race condition
2141 * seen in the presence of one thread doing SMC and another
2144 ptep_clear_flush(vma, address, page_table);
2145 page_add_new_anon_rmap(new_page, vma, address);
2147 * We call the notify macro here because, when using secondary
2148 * mmu page tables (such as kvm shadow page tables), we want the
2149 * new page to be mapped directly into the secondary page table.
2151 set_pte_at_notify(mm, address, page_table, entry);
2152 update_mmu_cache(vma, address, entry);
2155 * Only after switching the pte to the new page may
2156 * we remove the mapcount here. Otherwise another
2157 * process may come and find the rmap count decremented
2158 * before the pte is switched to the new page, and
2159 * "reuse" the old page writing into it while our pte
2160 * here still points into it and can be read by other
2163 * The critical issue is to order this
2164 * page_remove_rmap with the ptp_clear_flush above.
2165 * Those stores are ordered by (if nothing else,)
2166 * the barrier present in the atomic_add_negative
2167 * in page_remove_rmap.
2169 * Then the TLB flush in ptep_clear_flush ensures that
2170 * no process can access the old page before the
2171 * decremented mapcount is visible. And the old page
2172 * cannot be reused until after the decremented
2173 * mapcount is visible. So transitively, TLBs to
2174 * old page will be flushed before it can be reused.
2176 page_remove_rmap(old_page);
2179 /* Free the old page.. */
2180 new_page = old_page;
2181 ret |= VM_FAULT_WRITE;
2183 mem_cgroup_uncharge_page(new_page);
2186 page_cache_release(new_page);
2188 page_cache_release(old_page);
2190 pte_unmap_unlock(page_table, ptl);
2193 * Yes, Virginia, this is actually required to prevent a race
2194 * with clear_page_dirty_for_io() from clearing the page dirty
2195 * bit after it clear all dirty ptes, but before a racing
2196 * do_wp_page installs a dirty pte.
2198 * do_no_page is protected similarly.
2200 if (!page_mkwrite) {
2201 wait_on_page_locked(dirty_page);
2202 set_page_dirty_balance(dirty_page, page_mkwrite);
2204 put_page(dirty_page);
2206 struct address_space *mapping = dirty_page->mapping;
2208 set_page_dirty(dirty_page);
2209 unlock_page(dirty_page);
2210 page_cache_release(dirty_page);
2213 * Some device drivers do not set page.mapping
2214 * but still dirty their pages
2216 balance_dirty_pages_ratelimited(mapping);
2220 /* file_update_time outside page_lock */
2222 file_update_time(vma->vm_file);
2226 page_cache_release(new_page);
2230 unlock_page(old_page);
2231 page_cache_release(old_page);
2233 page_cache_release(old_page);
2235 return VM_FAULT_OOM;
2238 page_cache_release(old_page);
2243 * Helper functions for unmap_mapping_range().
2245 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
2247 * We have to restart searching the prio_tree whenever we drop the lock,
2248 * since the iterator is only valid while the lock is held, and anyway
2249 * a later vma might be split and reinserted earlier while lock dropped.
2251 * The list of nonlinear vmas could be handled more efficiently, using
2252 * a placeholder, but handle it in the same way until a need is shown.
2253 * It is important to search the prio_tree before nonlinear list: a vma
2254 * may become nonlinear and be shifted from prio_tree to nonlinear list
2255 * while the lock is dropped; but never shifted from list to prio_tree.
2257 * In order to make forward progress despite restarting the search,
2258 * vm_truncate_count is used to mark a vma as now dealt with, so we can
2259 * quickly skip it next time around. Since the prio_tree search only
2260 * shows us those vmas affected by unmapping the range in question, we
2261 * can't efficiently keep all vmas in step with mapping->truncate_count:
2262 * so instead reset them all whenever it wraps back to 0 (then go to 1).
2263 * mapping->truncate_count and vma->vm_truncate_count are protected by
2266 * In order to make forward progress despite repeatedly restarting some
2267 * large vma, note the restart_addr from unmap_vmas when it breaks out:
2268 * and restart from that address when we reach that vma again. It might
2269 * have been split or merged, shrunk or extended, but never shifted: so
2270 * restart_addr remains valid so long as it remains in the vma's range.
2271 * unmap_mapping_range forces truncate_count to leap over page-aligned
2272 * values so we can save vma's restart_addr in its truncate_count field.
2274 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
2276 static void reset_vma_truncate_counts(struct address_space *mapping)
2278 struct vm_area_struct *vma;
2279 struct prio_tree_iter iter;
2281 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
2282 vma->vm_truncate_count = 0;
2283 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
2284 vma->vm_truncate_count = 0;
2287 static int unmap_mapping_range_vma(struct vm_area_struct *vma,
2288 unsigned long start_addr, unsigned long end_addr,
2289 struct zap_details *details)
2291 unsigned long restart_addr;
2295 * files that support invalidating or truncating portions of the
2296 * file from under mmaped areas must have their ->fault function
2297 * return a locked page (and set VM_FAULT_LOCKED in the return).
2298 * This provides synchronisation against concurrent unmapping here.
2302 restart_addr = vma->vm_truncate_count;
2303 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
2304 start_addr = restart_addr;
2305 if (start_addr >= end_addr) {
2306 /* Top of vma has been split off since last time */
2307 vma->vm_truncate_count = details->truncate_count;
2312 restart_addr = zap_page_range(vma, start_addr,
2313 end_addr - start_addr, details);
2314 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
2316 if (restart_addr >= end_addr) {
2317 /* We have now completed this vma: mark it so */
2318 vma->vm_truncate_count = details->truncate_count;
2322 /* Note restart_addr in vma's truncate_count field */
2323 vma->vm_truncate_count = restart_addr;
2328 spin_unlock(details->i_mmap_lock);
2330 spin_lock(details->i_mmap_lock);
2334 static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
2335 struct zap_details *details)
2337 struct vm_area_struct *vma;
2338 struct prio_tree_iter iter;
2339 pgoff_t vba, vea, zba, zea;
2342 vma_prio_tree_foreach(vma, &iter, root,
2343 details->first_index, details->last_index) {
2344 /* Skip quickly over those we have already dealt with */
2345 if (vma->vm_truncate_count == details->truncate_count)
2348 vba = vma->vm_pgoff;
2349 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
2350 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2351 zba = details->first_index;
2354 zea = details->last_index;
2358 if (unmap_mapping_range_vma(vma,
2359 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2360 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2366 static inline void unmap_mapping_range_list(struct list_head *head,
2367 struct zap_details *details)
2369 struct vm_area_struct *vma;
2372 * In nonlinear VMAs there is no correspondence between virtual address
2373 * offset and file offset. So we must perform an exhaustive search
2374 * across *all* the pages in each nonlinear VMA, not just the pages
2375 * whose virtual address lies outside the file truncation point.
2378 list_for_each_entry(vma, head, shared.vm_set.list) {
2379 /* Skip quickly over those we have already dealt with */
2380 if (vma->vm_truncate_count == details->truncate_count)
2382 details->nonlinear_vma = vma;
2383 if (unmap_mapping_range_vma(vma, vma->vm_start,
2384 vma->vm_end, details) < 0)
2390 * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
2391 * @mapping: the address space containing mmaps to be unmapped.
2392 * @holebegin: byte in first page to unmap, relative to the start of
2393 * the underlying file. This will be rounded down to a PAGE_SIZE
2394 * boundary. Note that this is different from vmtruncate(), which
2395 * must keep the partial page. In contrast, we must get rid of
2397 * @holelen: size of prospective hole in bytes. This will be rounded
2398 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2400 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2401 * but 0 when invalidating pagecache, don't throw away private data.
2403 void unmap_mapping_range(struct address_space *mapping,
2404 loff_t const holebegin, loff_t const holelen, int even_cows)
2406 struct zap_details details;
2407 pgoff_t hba = holebegin >> PAGE_SHIFT;
2408 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2410 /* Check for overflow. */
2411 if (sizeof(holelen) > sizeof(hlen)) {
2413 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2414 if (holeend & ~(long long)ULONG_MAX)
2415 hlen = ULONG_MAX - hba + 1;
2418 details.check_mapping = even_cows? NULL: mapping;
2419 details.nonlinear_vma = NULL;
2420 details.first_index = hba;
2421 details.last_index = hba + hlen - 1;
2422 if (details.last_index < details.first_index)
2423 details.last_index = ULONG_MAX;
2424 details.i_mmap_lock = &mapping->i_mmap_lock;
2426 spin_lock(&mapping->i_mmap_lock);
2428 /* Protect against endless unmapping loops */
2429 mapping->truncate_count++;
2430 if (unlikely(is_restart_addr(mapping->truncate_count))) {
2431 if (mapping->truncate_count == 0)
2432 reset_vma_truncate_counts(mapping);
2433 mapping->truncate_count++;
2435 details.truncate_count = mapping->truncate_count;
2437 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2438 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2439 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2440 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2441 spin_unlock(&mapping->i_mmap_lock);
2443 EXPORT_SYMBOL(unmap_mapping_range);
2446 * vmtruncate - unmap mappings "freed" by truncate() syscall
2447 * @inode: inode of the file used
2448 * @offset: file offset to start truncating
2450 * NOTE! We have to be ready to update the memory sharing
2451 * between the file and the memory map for a potential last
2452 * incomplete page. Ugly, but necessary.
2454 int vmtruncate(struct inode * inode, loff_t offset)
2456 if (inode->i_size < offset) {
2457 unsigned long limit;
2459 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2460 if (limit != RLIM_INFINITY && offset > limit)
2462 if (offset > inode->i_sb->s_maxbytes)
2464 i_size_write(inode, offset);
2466 struct address_space *mapping = inode->i_mapping;
2469 * truncation of in-use swapfiles is disallowed - it would
2470 * cause subsequent swapout to scribble on the now-freed
2473 if (IS_SWAPFILE(inode))
2475 i_size_write(inode, offset);
2478 * unmap_mapping_range is called twice, first simply for
2479 * efficiency so that truncate_inode_pages does fewer
2480 * single-page unmaps. However after this first call, and
2481 * before truncate_inode_pages finishes, it is possible for
2482 * private pages to be COWed, which remain after
2483 * truncate_inode_pages finishes, hence the second
2484 * unmap_mapping_range call must be made for correctness.
2486 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2487 truncate_inode_pages(mapping, offset);
2488 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2491 if (inode->i_op->truncate)
2492 inode->i_op->truncate(inode);
2496 send_sig(SIGXFSZ, current, 0);
2500 EXPORT_SYMBOL(vmtruncate);
2502 int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2504 struct address_space *mapping = inode->i_mapping;
2507 * If the underlying filesystem is not going to provide
2508 * a way to truncate a range of blocks (punch a hole) -
2509 * we should return failure right now.
2511 if (!inode->i_op->truncate_range)
2514 mutex_lock(&inode->i_mutex);
2515 down_write(&inode->i_alloc_sem);
2516 unmap_mapping_range(mapping, offset, (end - offset), 1);
2517 truncate_inode_pages_range(mapping, offset, end);
2518 unmap_mapping_range(mapping, offset, (end - offset), 1);
2519 inode->i_op->truncate_range(inode, offset, end);
2520 up_write(&inode->i_alloc_sem);
2521 mutex_unlock(&inode->i_mutex);
2527 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2528 * but allow concurrent faults), and pte mapped but not yet locked.
2529 * We return with mmap_sem still held, but pte unmapped and unlocked.
2531 static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2532 unsigned long address, pte_t *page_table, pmd_t *pmd,
2533 unsigned int flags, pte_t orig_pte)
2539 struct mem_cgroup *ptr = NULL;
2542 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2545 entry = pte_to_swp_entry(orig_pte);
2546 if (is_migration_entry(entry)) {
2547 migration_entry_wait(mm, pmd, address);
2550 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
2551 page = lookup_swap_cache(entry);
2553 grab_swap_token(mm); /* Contend for token _before_ read-in */
2554 page = swapin_readahead(entry,
2555 GFP_HIGHUSER_MOVABLE, vma, address);
2558 * Back out if somebody else faulted in this pte
2559 * while we released the pte lock.
2561 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2562 if (likely(pte_same(*page_table, orig_pte)))
2564 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2568 /* Had to read the page from swap area: Major fault */
2569 ret = VM_FAULT_MAJOR;
2570 count_vm_event(PGMAJFAULT);
2574 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2576 if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
2582 * Back out if somebody else already faulted in this pte.
2584 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2585 if (unlikely(!pte_same(*page_table, orig_pte)))
2588 if (unlikely(!PageUptodate(page))) {
2589 ret = VM_FAULT_SIGBUS;
2594 * The page isn't present yet, go ahead with the fault.
2596 * Be careful about the sequence of operations here.
2597 * To get its accounting right, reuse_swap_page() must be called
2598 * while the page is counted on swap but not yet in mapcount i.e.
2599 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2600 * must be called after the swap_free(), or it will never succeed.
2601 * Because delete_from_swap_page() may be called by reuse_swap_page(),
2602 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
2603 * in page->private. In this case, a record in swap_cgroup is silently
2604 * discarded at swap_free().
2607 inc_mm_counter(mm, anon_rss);
2608 pte = mk_pte(page, vma->vm_page_prot);
2609 if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
2610 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2611 flags &= ~FAULT_FLAG_WRITE;
2613 flush_icache_page(vma, page);
2614 set_pte_at(mm, address, page_table, pte);
2615 page_add_anon_rmap(page, vma, address);
2616 /* It's better to call commit-charge after rmap is established */
2617 mem_cgroup_commit_charge_swapin(page, ptr);
2620 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
2621 try_to_free_swap(page);
2624 if (flags & FAULT_FLAG_WRITE) {
2625 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2626 if (ret & VM_FAULT_ERROR)
2627 ret &= VM_FAULT_ERROR;
2631 /* No need to invalidate - it was non-present before */
2632 update_mmu_cache(vma, address, pte);
2634 pte_unmap_unlock(page_table, ptl);
2638 mem_cgroup_cancel_charge_swapin(ptr);
2639 pte_unmap_unlock(page_table, ptl);
2642 page_cache_release(page);
2647 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2648 * but allow concurrent faults), and pte mapped but not yet locked.
2649 * We return with mmap_sem still held, but pte unmapped and unlocked.
2651 static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2652 unsigned long address, pte_t *page_table, pmd_t *pmd,
2659 /* Allocate our own private page. */
2660 pte_unmap(page_table);
2662 if (unlikely(anon_vma_prepare(vma)))
2664 page = alloc_zeroed_user_highpage_movable(vma, address);
2667 __SetPageUptodate(page);
2669 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))
2672 entry = mk_pte(page, vma->vm_page_prot);
2673 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2675 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2676 if (!pte_none(*page_table))
2679 inc_mm_counter(mm, anon_rss);
2680 page_add_new_anon_rmap(page, vma, address);
2681 set_pte_at(mm, address, page_table, entry);
2683 /* No need to invalidate - it was non-present before */
2684 update_mmu_cache(vma, address, entry);
2686 pte_unmap_unlock(page_table, ptl);
2689 mem_cgroup_uncharge_page(page);
2690 page_cache_release(page);
2693 page_cache_release(page);
2695 return VM_FAULT_OOM;
2699 * __do_fault() tries to create a new page mapping. It aggressively
2700 * tries to share with existing pages, but makes a separate copy if
2701 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2702 * the next page fault.
2704 * As this is called only for pages that do not currently exist, we
2705 * do not need to flush old virtual caches or the TLB.
2707 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2708 * but allow concurrent faults), and pte neither mapped nor locked.
2709 * We return with mmap_sem still held, but pte unmapped and unlocked.
2711 static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2712 unsigned long address, pmd_t *pmd,
2713 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
2721 struct page *dirty_page = NULL;
2722 struct vm_fault vmf;
2724 int page_mkwrite = 0;
2726 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2731 ret = vma->vm_ops->fault(vma, &vmf);
2732 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2736 * For consistency in subsequent calls, make the faulted page always
2739 if (unlikely(!(ret & VM_FAULT_LOCKED)))
2740 lock_page(vmf.page);
2742 VM_BUG_ON(!PageLocked(vmf.page));
2745 * Should we do an early C-O-W break?
2748 if (flags & FAULT_FLAG_WRITE) {
2749 if (!(vma->vm_flags & VM_SHARED)) {
2751 if (unlikely(anon_vma_prepare(vma))) {
2755 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2761 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) {
2763 page_cache_release(page);
2768 * Don't let another task, with possibly unlocked vma,
2769 * keep the mlocked page.
2771 if (vma->vm_flags & VM_LOCKED)
2772 clear_page_mlock(vmf.page);
2773 copy_user_highpage(page, vmf.page, address, vma);
2774 __SetPageUptodate(page);
2777 * If the page will be shareable, see if the backing
2778 * address space wants to know that the page is about
2779 * to become writable
2781 if (vma->vm_ops->page_mkwrite) {
2785 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2786 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2788 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2790 goto unwritable_page;
2792 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2794 if (!page->mapping) {
2795 ret = 0; /* retry the fault */
2797 goto unwritable_page;
2800 VM_BUG_ON(!PageLocked(page));
2807 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2810 * This silly early PAGE_DIRTY setting removes a race
2811 * due to the bad i386 page protection. But it's valid
2812 * for other architectures too.
2814 * Note that if FAULT_FLAG_WRITE is set, we either now have
2815 * an exclusive copy of the page, or this is a shared mapping,
2816 * so we can make it writable and dirty to avoid having to
2817 * handle that later.
2819 /* Only go through if we didn't race with anybody else... */
2820 if (likely(pte_same(*page_table, orig_pte))) {
2821 flush_icache_page(vma, page);
2822 entry = mk_pte(page, vma->vm_page_prot);
2823 if (flags & FAULT_FLAG_WRITE)
2824 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2826 inc_mm_counter(mm, anon_rss);
2827 page_add_new_anon_rmap(page, vma, address);
2829 inc_mm_counter(mm, file_rss);
2830 page_add_file_rmap(page);
2831 if (flags & FAULT_FLAG_WRITE) {
2833 get_page(dirty_page);
2836 set_pte_at(mm, address, page_table, entry);
2838 /* no need to invalidate: a not-present page won't be cached */
2839 update_mmu_cache(vma, address, entry);
2842 mem_cgroup_uncharge_page(page);
2844 page_cache_release(page);
2846 anon = 1; /* no anon but release faulted_page */
2849 pte_unmap_unlock(page_table, ptl);
2853 struct address_space *mapping = page->mapping;
2855 if (set_page_dirty(dirty_page))
2857 unlock_page(dirty_page);
2858 put_page(dirty_page);
2859 if (page_mkwrite && mapping) {
2861 * Some device drivers do not set page.mapping but still
2864 balance_dirty_pages_ratelimited(mapping);
2867 /* file_update_time outside page_lock */
2869 file_update_time(vma->vm_file);
2871 unlock_page(vmf.page);
2873 page_cache_release(vmf.page);
2879 page_cache_release(page);
2883 static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2884 unsigned long address, pte_t *page_table, pmd_t *pmd,
2885 unsigned int flags, pte_t orig_pte)
2887 pgoff_t pgoff = (((address & PAGE_MASK)
2888 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2890 pte_unmap(page_table);
2891 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
2895 * Fault of a previously existing named mapping. Repopulate the pte
2896 * from the encoded file_pte if possible. This enables swappable
2899 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2900 * but allow concurrent faults), and pte mapped but not yet locked.
2901 * We return with mmap_sem still held, but pte unmapped and unlocked.
2903 static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2904 unsigned long address, pte_t *page_table, pmd_t *pmd,
2905 unsigned int flags, pte_t orig_pte)
2909 flags |= FAULT_FLAG_NONLINEAR;
2911 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2914 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
2916 * Page table corrupted: show pte and kill process.
2918 print_bad_pte(vma, address, orig_pte, NULL);
2919 return VM_FAULT_OOM;
2922 pgoff = pte_to_pgoff(orig_pte);
2923 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
2927 * These routines also need to handle stuff like marking pages dirty
2928 * and/or accessed for architectures that don't do it in hardware (most
2929 * RISC architectures). The early dirtying is also good on the i386.
2931 * There is also a hook called "update_mmu_cache()" that architectures
2932 * with external mmu caches can use to update those (ie the Sparc or
2933 * PowerPC hashed page tables that act as extended TLBs).
2935 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2936 * but allow concurrent faults), and pte mapped but not yet locked.
2937 * We return with mmap_sem still held, but pte unmapped and unlocked.
2939 static inline int handle_pte_fault(struct mm_struct *mm,
2940 struct vm_area_struct *vma, unsigned long address,
2941 pte_t *pte, pmd_t *pmd, unsigned int flags)
2947 if (!pte_present(entry)) {
2948 if (pte_none(entry)) {
2950 if (likely(vma->vm_ops->fault))
2951 return do_linear_fault(mm, vma, address,
2952 pte, pmd, flags, entry);
2954 return do_anonymous_page(mm, vma, address,
2957 if (pte_file(entry))
2958 return do_nonlinear_fault(mm, vma, address,
2959 pte, pmd, flags, entry);
2960 return do_swap_page(mm, vma, address,
2961 pte, pmd, flags, entry);
2964 ptl = pte_lockptr(mm, pmd);
2966 if (unlikely(!pte_same(*pte, entry)))
2968 if (flags & FAULT_FLAG_WRITE) {
2969 if (!pte_write(entry))
2970 return do_wp_page(mm, vma, address,
2971 pte, pmd, ptl, entry);
2972 entry = pte_mkdirty(entry);
2974 entry = pte_mkyoung(entry);
2975 if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
2976 update_mmu_cache(vma, address, entry);
2979 * This is needed only for protection faults but the arch code
2980 * is not yet telling us if this is a protection fault or not.
2981 * This still avoids useless tlb flushes for .text page faults
2984 if (flags & FAULT_FLAG_WRITE)
2985 flush_tlb_page(vma, address);
2988 pte_unmap_unlock(pte, ptl);
2993 * By the time we get here, we already hold the mm semaphore
2995 int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2996 unsigned long address, unsigned int flags)
3003 __set_current_state(TASK_RUNNING);
3005 count_vm_event(PGFAULT);
3007 if (unlikely(is_vm_hugetlb_page(vma)))
3008 return hugetlb_fault(mm, vma, address, flags);
3010 pgd = pgd_offset(mm, address);
3011 pud = pud_alloc(mm, pgd, address);
3013 return VM_FAULT_OOM;
3014 pmd = pmd_alloc(mm, pud, address);
3016 return VM_FAULT_OOM;
3017 pte = pte_alloc_map(mm, pmd, address);
3019 return VM_FAULT_OOM;
3021 return handle_pte_fault(mm, vma, address, pte, pmd, flags);
3024 #ifndef __PAGETABLE_PUD_FOLDED
3026 * Allocate page upper directory.
3027 * We've already handled the fast-path in-line.
3029 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
3031 pud_t *new = pud_alloc_one(mm, address);
3035 smp_wmb(); /* See comment in __pte_alloc */
3037 spin_lock(&mm->page_table_lock);
3038 if (pgd_present(*pgd)) /* Another has populated it */
3041 pgd_populate(mm, pgd, new);
3042 spin_unlock(&mm->page_table_lock);
3045 #endif /* __PAGETABLE_PUD_FOLDED */
3047 #ifndef __PAGETABLE_PMD_FOLDED
3049 * Allocate page middle directory.
3050 * We've already handled the fast-path in-line.
3052 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
3054 pmd_t *new = pmd_alloc_one(mm, address);
3058 smp_wmb(); /* See comment in __pte_alloc */
3060 spin_lock(&mm->page_table_lock);
3061 #ifndef __ARCH_HAS_4LEVEL_HACK
3062 if (pud_present(*pud)) /* Another has populated it */
3065 pud_populate(mm, pud, new);
3067 if (pgd_present(*pud)) /* Another has populated it */
3070 pgd_populate(mm, pud, new);
3071 #endif /* __ARCH_HAS_4LEVEL_HACK */
3072 spin_unlock(&mm->page_table_lock);
3075 #endif /* __PAGETABLE_PMD_FOLDED */
3077 int make_pages_present(unsigned long addr, unsigned long end)
3079 int ret, len, write;
3080 struct vm_area_struct * vma;
3082 vma = find_vma(current->mm, addr);
3085 write = (vma->vm_flags & VM_WRITE) != 0;
3086 BUG_ON(addr >= end);
3087 BUG_ON(end > vma->vm_end);
3088 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
3089 ret = get_user_pages(current, current->mm, addr,
3090 len, write, 0, NULL, NULL);
3093 return ret == len ? 0 : -EFAULT;
3096 #if !defined(__HAVE_ARCH_GATE_AREA)
3098 #if defined(AT_SYSINFO_EHDR)
3099 static struct vm_area_struct gate_vma;
3101 static int __init gate_vma_init(void)
3103 gate_vma.vm_mm = NULL;
3104 gate_vma.vm_start = FIXADDR_USER_START;
3105 gate_vma.vm_end = FIXADDR_USER_END;
3106 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
3107 gate_vma.vm_page_prot = __P101;
3109 * Make sure the vDSO gets into every core dump.
3110 * Dumping its contents makes post-mortem fully interpretable later
3111 * without matching up the same kernel and hardware config to see
3112 * what PC values meant.
3114 gate_vma.vm_flags |= VM_ALWAYSDUMP;
3117 __initcall(gate_vma_init);
3120 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
3122 #ifdef AT_SYSINFO_EHDR
3129 int in_gate_area_no_task(unsigned long addr)
3131 #ifdef AT_SYSINFO_EHDR
3132 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
3138 #endif /* __HAVE_ARCH_GATE_AREA */
3140 static int follow_pte(struct mm_struct *mm, unsigned long address,
3141 pte_t **ptepp, spinlock_t **ptlp)
3148 pgd = pgd_offset(mm, address);
3149 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3152 pud = pud_offset(pgd, address);
3153 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3156 pmd = pmd_offset(pud, address);
3157 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3160 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3164 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3167 if (!pte_present(*ptep))
3172 pte_unmap_unlock(ptep, *ptlp);
3178 * follow_pfn - look up PFN at a user virtual address
3179 * @vma: memory mapping
3180 * @address: user virtual address
3181 * @pfn: location to store found PFN
3183 * Only IO mappings and raw PFN mappings are allowed.
3185 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3187 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3194 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3197 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3200 *pfn = pte_pfn(*ptep);
3201 pte_unmap_unlock(ptep, ptl);
3204 EXPORT_SYMBOL(follow_pfn);
3206 #ifdef CONFIG_HAVE_IOREMAP_PROT
3207 int follow_phys(struct vm_area_struct *vma,
3208 unsigned long address, unsigned int flags,
3209 unsigned long *prot, resource_size_t *phys)
3215 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3218 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
3222 if ((flags & FOLL_WRITE) && !pte_write(pte))
3225 *prot = pgprot_val(pte_pgprot(pte));
3226 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
3230 pte_unmap_unlock(ptep, ptl);
3235 int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3236 void *buf, int len, int write)
3238 resource_size_t phys_addr;
3239 unsigned long prot = 0;
3240 void __iomem *maddr;
3241 int offset = addr & (PAGE_SIZE-1);
3243 if (follow_phys(vma, addr, write, &prot, &phys_addr))
3246 maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
3248 memcpy_toio(maddr + offset, buf, len);
3250 memcpy_fromio(buf, maddr + offset, len);
3258 * Access another process' address space.
3259 * Source/target buffer must be kernel space,
3260 * Do not walk the page table directly, use get_user_pages
3262 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
3264 struct mm_struct *mm;
3265 struct vm_area_struct *vma;
3266 void *old_buf = buf;
3268 mm = get_task_mm(tsk);
3272 down_read(&mm->mmap_sem);
3273 /* ignore errors, just check how much was successfully transferred */
3275 int bytes, ret, offset;
3277 struct page *page = NULL;
3279 ret = get_user_pages(tsk, mm, addr, 1,
3280 write, 1, &page, &vma);
3283 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3284 * we can access using slightly different code.
3286 #ifdef CONFIG_HAVE_IOREMAP_PROT
3287 vma = find_vma(mm, addr);
3290 if (vma->vm_ops && vma->vm_ops->access)
3291 ret = vma->vm_ops->access(vma, addr, buf,
3299 offset = addr & (PAGE_SIZE-1);
3300 if (bytes > PAGE_SIZE-offset)
3301 bytes = PAGE_SIZE-offset;
3305 copy_to_user_page(vma, page, addr,
3306 maddr + offset, buf, bytes);
3307 set_page_dirty_lock(page);
3309 copy_from_user_page(vma, page, addr,
3310 buf, maddr + offset, bytes);
3313 page_cache_release(page);
3319 up_read(&mm->mmap_sem);
3322 return buf - old_buf;
3326 * Print the name of a VMA.
3328 void print_vma_addr(char *prefix, unsigned long ip)
3330 struct mm_struct *mm = current->mm;
3331 struct vm_area_struct *vma;
3334 * Do not print if we are in atomic
3335 * contexts (in exception stacks, etc.):
3337 if (preempt_count())
3340 down_read(&mm->mmap_sem);
3341 vma = find_vma(mm, ip);
3342 if (vma && vma->vm_file) {
3343 struct file *f = vma->vm_file;
3344 char *buf = (char *)__get_free_page(GFP_KERNEL);
3348 p = d_path(&f->f_path, buf, PAGE_SIZE);
3351 s = strrchr(p, '/');
3354 printk("%s%s[%lx+%lx]", prefix, p,
3356 vma->vm_end - vma->vm_start);
3357 free_page((unsigned long)buf);
3360 up_read(¤t->mm->mmap_sem);
3363 #ifdef CONFIG_PROVE_LOCKING
3364 void might_fault(void)
3367 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3368 * holding the mmap_sem, this is safe because kernel memory doesn't
3369 * get paged out, therefore we'll never actually fault, and the
3370 * below annotations will generate false positives.
3372 if (segment_eq(get_fs(), KERNEL_DS))
3377 * it would be nicer only to annotate paths which are not under
3378 * pagefault_disable, however that requires a larger audit and
3379 * providing helpers like get_user_atomic.
3381 if (!in_atomic() && current->mm)
3382 might_lock_read(¤t->mm->mmap_sem);
3384 EXPORT_SYMBOL(might_fault);