4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
41 #include <linux/kernel_stat.h>
43 #include <linux/hugetlb.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/highmem.h>
47 #include <linux/pagemap.h>
48 #include <linux/rmap.h>
49 #include <linux/module.h>
50 #include <linux/delayacct.h>
51 #include <linux/init.h>
52 #include <linux/writeback.h>
53 #include <linux/memcontrol.h>
55 #include <asm/pgalloc.h>
56 #include <asm/uaccess.h>
58 #include <asm/tlbflush.h>
59 #include <asm/pgtable.h>
61 #include <linux/swapops.h>
62 #include <linux/elf.h>
64 #ifndef CONFIG_NEED_MULTIPLE_NODES
65 /* use the per-pgdat data instead for discontigmem - mbligh */
66 unsigned long max_mapnr;
69 EXPORT_SYMBOL(max_mapnr);
70 EXPORT_SYMBOL(mem_map);
73 unsigned long num_physpages;
75 * A number of key systems in x86 including ioremap() rely on the assumption
76 * that high_memory defines the upper bound on direct map memory, then end
77 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
78 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
83 EXPORT_SYMBOL(num_physpages);
84 EXPORT_SYMBOL(high_memory);
87 * Randomize the address space (stacks, mmaps, brk, etc.).
89 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
90 * as ancient (libc5 based) binaries can segfault. )
92 int randomize_va_space __read_mostly =
93 #ifdef CONFIG_COMPAT_BRK
99 static int __init disable_randmaps(char *s)
101 randomize_va_space = 0;
104 __setup("norandmaps", disable_randmaps);
108 * If a p?d_bad entry is found while walking page tables, report
109 * the error, before resetting entry to p?d_none. Usually (but
110 * very seldom) called out from the p?d_none_or_clear_bad macros.
113 void pgd_clear_bad(pgd_t *pgd)
119 void pud_clear_bad(pud_t *pud)
125 void pmd_clear_bad(pmd_t *pmd)
132 * Note: this doesn't free the actual pages themselves. That
133 * has been handled earlier when unmapping all the memory regions.
135 static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
137 pgtable_t token = pmd_pgtable(*pmd);
139 pte_free_tlb(tlb, token);
143 static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
144 unsigned long addr, unsigned long end,
145 unsigned long floor, unsigned long ceiling)
152 pmd = pmd_offset(pud, addr);
154 next = pmd_addr_end(addr, end);
155 if (pmd_none_or_clear_bad(pmd))
157 free_pte_range(tlb, pmd);
158 } while (pmd++, addr = next, addr != end);
168 if (end - 1 > ceiling - 1)
171 pmd = pmd_offset(pud, start);
173 pmd_free_tlb(tlb, pmd);
176 static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
177 unsigned long addr, unsigned long end,
178 unsigned long floor, unsigned long ceiling)
185 pud = pud_offset(pgd, addr);
187 next = pud_addr_end(addr, end);
188 if (pud_none_or_clear_bad(pud))
190 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
191 } while (pud++, addr = next, addr != end);
197 ceiling &= PGDIR_MASK;
201 if (end - 1 > ceiling - 1)
204 pud = pud_offset(pgd, start);
206 pud_free_tlb(tlb, pud);
210 * This function frees user-level page tables of a process.
212 * Must be called with pagetable lock held.
214 void free_pgd_range(struct mmu_gather **tlb,
215 unsigned long addr, unsigned long end,
216 unsigned long floor, unsigned long ceiling)
223 * The next few lines have given us lots of grief...
225 * Why are we testing PMD* at this top level? Because often
226 * there will be no work to do at all, and we'd prefer not to
227 * go all the way down to the bottom just to discover that.
229 * Why all these "- 1"s? Because 0 represents both the bottom
230 * of the address space and the top of it (using -1 for the
231 * top wouldn't help much: the masks would do the wrong thing).
232 * The rule is that addr 0 and floor 0 refer to the bottom of
233 * the address space, but end 0 and ceiling 0 refer to the top
234 * Comparisons need to use "end - 1" and "ceiling - 1" (though
235 * that end 0 case should be mythical).
237 * Wherever addr is brought up or ceiling brought down, we must
238 * be careful to reject "the opposite 0" before it confuses the
239 * subsequent tests. But what about where end is brought down
240 * by PMD_SIZE below? no, end can't go down to 0 there.
242 * Whereas we round start (addr) and ceiling down, by different
243 * masks at different levels, in order to test whether a table
244 * now has no other vmas using it, so can be freed, we don't
245 * bother to round floor or end up - the tests don't need that.
259 if (end - 1 > ceiling - 1)
265 pgd = pgd_offset((*tlb)->mm, addr);
267 next = pgd_addr_end(addr, end);
268 if (pgd_none_or_clear_bad(pgd))
270 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
271 } while (pgd++, addr = next, addr != end);
274 void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
275 unsigned long floor, unsigned long ceiling)
278 struct vm_area_struct *next = vma->vm_next;
279 unsigned long addr = vma->vm_start;
282 * Hide vma from rmap and vmtruncate before freeing pgtables
284 anon_vma_unlink(vma);
285 unlink_file_vma(vma);
287 if (is_vm_hugetlb_page(vma)) {
288 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
289 floor, next? next->vm_start: ceiling);
292 * Optimization: gather nearby vmas into one call down
294 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
295 && !is_vm_hugetlb_page(next)) {
298 anon_vma_unlink(vma);
299 unlink_file_vma(vma);
301 free_pgd_range(tlb, addr, vma->vm_end,
302 floor, next? next->vm_start: ceiling);
308 int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
310 pgtable_t new = pte_alloc_one(mm, address);
314 spin_lock(&mm->page_table_lock);
315 if (!pmd_present(*pmd)) { /* Has another populated it ? */
317 pmd_populate(mm, pmd, new);
320 spin_unlock(&mm->page_table_lock);
326 int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
328 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
332 spin_lock(&init_mm.page_table_lock);
333 if (!pmd_present(*pmd)) { /* Has another populated it ? */
334 pmd_populate_kernel(&init_mm, pmd, new);
337 spin_unlock(&init_mm.page_table_lock);
339 pte_free_kernel(&init_mm, new);
343 static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
346 add_mm_counter(mm, file_rss, file_rss);
348 add_mm_counter(mm, anon_rss, anon_rss);
352 * This function is called to print an error when a bad pte
353 * is found. For example, we might have a PFN-mapped pte in
354 * a region that doesn't allow it.
356 * The calling function must still handle the error.
358 void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
360 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
361 "vm_flags = %lx, vaddr = %lx\n",
362 (long long)pte_val(pte),
363 (vma->vm_mm == current->mm ? current->comm : "???"),
364 vma->vm_flags, vaddr);
368 static inline int is_cow_mapping(unsigned int flags)
370 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
374 * This function gets the "struct page" associated with a pte.
376 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
377 * will have each page table entry just pointing to a raw page frame
378 * number, and as far as the VM layer is concerned, those do not have
379 * pages associated with them - even if the PFN might point to memory
380 * that otherwise is perfectly fine and has a "struct page".
382 * The way we recognize those mappings is through the rules set up
383 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
384 * and the vm_pgoff will point to the first PFN mapped: thus every
385 * page that is a raw mapping will always honor the rule
387 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
389 * and if that isn't true, the page has been COW'ed (in which case it
390 * _does_ have a "struct page" associated with it even if it is in a
393 struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
395 unsigned long pfn = pte_pfn(pte);
397 if (unlikely(vma->vm_flags & VM_PFNMAP)) {
398 unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT;
399 if (pfn == vma->vm_pgoff + off)
401 if (!is_cow_mapping(vma->vm_flags))
405 #ifdef CONFIG_DEBUG_VM
407 * Add some anal sanity checks for now. Eventually,
408 * we should just do "return pfn_to_page(pfn)", but
409 * in the meantime we check that we get a valid pfn,
410 * and that the resulting page looks ok.
412 if (unlikely(!pfn_valid(pfn))) {
413 print_bad_pte(vma, pte, addr);
419 * NOTE! We still have PageReserved() pages in the page
422 * The PAGE_ZERO() pages and various VDSO mappings can
423 * cause them to exist.
425 return pfn_to_page(pfn);
429 * copy one vm_area from one task to the other. Assumes the page tables
430 * already present in the new task to be cleared in the whole range
431 * covered by this vma.
435 copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
436 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
437 unsigned long addr, int *rss)
439 unsigned long vm_flags = vma->vm_flags;
440 pte_t pte = *src_pte;
443 /* pte contains position in swap or file, so copy. */
444 if (unlikely(!pte_present(pte))) {
445 if (!pte_file(pte)) {
446 swp_entry_t entry = pte_to_swp_entry(pte);
448 swap_duplicate(entry);
449 /* make sure dst_mm is on swapoff's mmlist. */
450 if (unlikely(list_empty(&dst_mm->mmlist))) {
451 spin_lock(&mmlist_lock);
452 if (list_empty(&dst_mm->mmlist))
453 list_add(&dst_mm->mmlist,
455 spin_unlock(&mmlist_lock);
457 if (is_write_migration_entry(entry) &&
458 is_cow_mapping(vm_flags)) {
460 * COW mappings require pages in both parent
461 * and child to be set to read.
463 make_migration_entry_read(&entry);
464 pte = swp_entry_to_pte(entry);
465 set_pte_at(src_mm, addr, src_pte, pte);
472 * If it's a COW mapping, write protect it both
473 * in the parent and the child
475 if (is_cow_mapping(vm_flags)) {
476 ptep_set_wrprotect(src_mm, addr, src_pte);
477 pte = pte_wrprotect(pte);
481 * If it's a shared mapping, mark it clean in
484 if (vm_flags & VM_SHARED)
485 pte = pte_mkclean(pte);
486 pte = pte_mkold(pte);
488 page = vm_normal_page(vma, addr, pte);
491 page_dup_rmap(page, vma, addr);
492 rss[!!PageAnon(page)]++;
496 set_pte_at(dst_mm, addr, dst_pte, pte);
499 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
500 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
501 unsigned long addr, unsigned long end)
503 pte_t *src_pte, *dst_pte;
504 spinlock_t *src_ptl, *dst_ptl;
510 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
513 src_pte = pte_offset_map_nested(src_pmd, addr);
514 src_ptl = pte_lockptr(src_mm, src_pmd);
515 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
516 arch_enter_lazy_mmu_mode();
520 * We are holding two locks at this point - either of them
521 * could generate latencies in another task on another CPU.
523 if (progress >= 32) {
525 if (need_resched() ||
526 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
529 if (pte_none(*src_pte)) {
533 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
535 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
537 arch_leave_lazy_mmu_mode();
538 spin_unlock(src_ptl);
539 pte_unmap_nested(src_pte - 1);
540 add_mm_rss(dst_mm, rss[0], rss[1]);
541 pte_unmap_unlock(dst_pte - 1, dst_ptl);
548 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
549 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
550 unsigned long addr, unsigned long end)
552 pmd_t *src_pmd, *dst_pmd;
555 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
558 src_pmd = pmd_offset(src_pud, addr);
560 next = pmd_addr_end(addr, end);
561 if (pmd_none_or_clear_bad(src_pmd))
563 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
566 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
570 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
571 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
572 unsigned long addr, unsigned long end)
574 pud_t *src_pud, *dst_pud;
577 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
580 src_pud = pud_offset(src_pgd, addr);
582 next = pud_addr_end(addr, end);
583 if (pud_none_or_clear_bad(src_pud))
585 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
588 } while (dst_pud++, src_pud++, addr = next, addr != end);
592 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
593 struct vm_area_struct *vma)
595 pgd_t *src_pgd, *dst_pgd;
597 unsigned long addr = vma->vm_start;
598 unsigned long end = vma->vm_end;
601 * Don't copy ptes where a page fault will fill them correctly.
602 * Fork becomes much lighter when there are big shared or private
603 * readonly mappings. The tradeoff is that copy_page_range is more
604 * efficient than faulting.
606 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
611 if (is_vm_hugetlb_page(vma))
612 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
614 dst_pgd = pgd_offset(dst_mm, addr);
615 src_pgd = pgd_offset(src_mm, addr);
617 next = pgd_addr_end(addr, end);
618 if (pgd_none_or_clear_bad(src_pgd))
620 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
623 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
627 static unsigned long zap_pte_range(struct mmu_gather *tlb,
628 struct vm_area_struct *vma, pmd_t *pmd,
629 unsigned long addr, unsigned long end,
630 long *zap_work, struct zap_details *details)
632 struct mm_struct *mm = tlb->mm;
638 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
639 arch_enter_lazy_mmu_mode();
642 if (pte_none(ptent)) {
647 (*zap_work) -= PAGE_SIZE;
649 if (pte_present(ptent)) {
652 page = vm_normal_page(vma, addr, ptent);
653 if (unlikely(details) && page) {
655 * unmap_shared_mapping_pages() wants to
656 * invalidate cache without truncating:
657 * unmap shared but keep private pages.
659 if (details->check_mapping &&
660 details->check_mapping != page->mapping)
663 * Each page->index must be checked when
664 * invalidating or truncating nonlinear.
666 if (details->nonlinear_vma &&
667 (page->index < details->first_index ||
668 page->index > details->last_index))
671 ptent = ptep_get_and_clear_full(mm, addr, pte,
673 tlb_remove_tlb_entry(tlb, pte, addr);
676 if (unlikely(details) && details->nonlinear_vma
677 && linear_page_index(details->nonlinear_vma,
678 addr) != page->index)
679 set_pte_at(mm, addr, pte,
680 pgoff_to_pte(page->index));
684 if (pte_dirty(ptent))
685 set_page_dirty(page);
686 if (pte_young(ptent))
687 SetPageReferenced(page);
690 page_remove_rmap(page, vma);
691 tlb_remove_page(tlb, page);
695 * If details->check_mapping, we leave swap entries;
696 * if details->nonlinear_vma, we leave file entries.
698 if (unlikely(details))
700 if (!pte_file(ptent))
701 free_swap_and_cache(pte_to_swp_entry(ptent));
702 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
703 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
705 add_mm_rss(mm, file_rss, anon_rss);
706 arch_leave_lazy_mmu_mode();
707 pte_unmap_unlock(pte - 1, ptl);
712 static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
713 struct vm_area_struct *vma, pud_t *pud,
714 unsigned long addr, unsigned long end,
715 long *zap_work, struct zap_details *details)
720 pmd = pmd_offset(pud, addr);
722 next = pmd_addr_end(addr, end);
723 if (pmd_none_or_clear_bad(pmd)) {
727 next = zap_pte_range(tlb, vma, pmd, addr, next,
729 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
734 static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
735 struct vm_area_struct *vma, pgd_t *pgd,
736 unsigned long addr, unsigned long end,
737 long *zap_work, struct zap_details *details)
742 pud = pud_offset(pgd, addr);
744 next = pud_addr_end(addr, end);
745 if (pud_none_or_clear_bad(pud)) {
749 next = zap_pmd_range(tlb, vma, pud, addr, next,
751 } while (pud++, addr = next, (addr != end && *zap_work > 0));
756 static unsigned long unmap_page_range(struct mmu_gather *tlb,
757 struct vm_area_struct *vma,
758 unsigned long addr, unsigned long end,
759 long *zap_work, struct zap_details *details)
764 if (details && !details->check_mapping && !details->nonlinear_vma)
768 tlb_start_vma(tlb, vma);
769 pgd = pgd_offset(vma->vm_mm, addr);
771 next = pgd_addr_end(addr, end);
772 if (pgd_none_or_clear_bad(pgd)) {
776 next = zap_pud_range(tlb, vma, pgd, addr, next,
778 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
779 tlb_end_vma(tlb, vma);
784 #ifdef CONFIG_PREEMPT
785 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
787 /* No preempt: go for improved straight-line efficiency */
788 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
792 * unmap_vmas - unmap a range of memory covered by a list of vma's
793 * @tlbp: address of the caller's struct mmu_gather
794 * @vma: the starting vma
795 * @start_addr: virtual address at which to start unmapping
796 * @end_addr: virtual address at which to end unmapping
797 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
798 * @details: details of nonlinear truncation or shared cache invalidation
800 * Returns the end address of the unmapping (restart addr if interrupted).
802 * Unmap all pages in the vma list.
804 * We aim to not hold locks for too long (for scheduling latency reasons).
805 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
806 * return the ending mmu_gather to the caller.
808 * Only addresses between `start' and `end' will be unmapped.
810 * The VMA list must be sorted in ascending virtual address order.
812 * unmap_vmas() assumes that the caller will flush the whole unmapped address
813 * range after unmap_vmas() returns. So the only responsibility here is to
814 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
815 * drops the lock and schedules.
817 unsigned long unmap_vmas(struct mmu_gather **tlbp,
818 struct vm_area_struct *vma, unsigned long start_addr,
819 unsigned long end_addr, unsigned long *nr_accounted,
820 struct zap_details *details)
822 long zap_work = ZAP_BLOCK_SIZE;
823 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
824 int tlb_start_valid = 0;
825 unsigned long start = start_addr;
826 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
827 int fullmm = (*tlbp)->fullmm;
829 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
832 start = max(vma->vm_start, start_addr);
833 if (start >= vma->vm_end)
835 end = min(vma->vm_end, end_addr);
836 if (end <= vma->vm_start)
839 if (vma->vm_flags & VM_ACCOUNT)
840 *nr_accounted += (end - start) >> PAGE_SHIFT;
842 while (start != end) {
843 if (!tlb_start_valid) {
848 if (unlikely(is_vm_hugetlb_page(vma))) {
849 unmap_hugepage_range(vma, start, end);
850 zap_work -= (end - start) /
851 (HPAGE_SIZE / PAGE_SIZE);
854 start = unmap_page_range(*tlbp, vma,
855 start, end, &zap_work, details);
858 BUG_ON(start != end);
862 tlb_finish_mmu(*tlbp, tlb_start, start);
864 if (need_resched() ||
865 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
873 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
875 zap_work = ZAP_BLOCK_SIZE;
879 return start; /* which is now the end (or restart) address */
883 * zap_page_range - remove user pages in a given range
884 * @vma: vm_area_struct holding the applicable pages
885 * @address: starting address of pages to zap
886 * @size: number of bytes to zap
887 * @details: details of nonlinear truncation or shared cache invalidation
889 unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
890 unsigned long size, struct zap_details *details)
892 struct mm_struct *mm = vma->vm_mm;
893 struct mmu_gather *tlb;
894 unsigned long end = address + size;
895 unsigned long nr_accounted = 0;
898 tlb = tlb_gather_mmu(mm, 0);
899 update_hiwater_rss(mm);
900 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
902 tlb_finish_mmu(tlb, address, end);
907 * Do a quick page-table lookup for a single page.
909 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
918 struct mm_struct *mm = vma->vm_mm;
920 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
922 BUG_ON(flags & FOLL_GET);
927 pgd = pgd_offset(mm, address);
928 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
931 pud = pud_offset(pgd, address);
932 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
935 pmd = pmd_offset(pud, address);
936 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
939 if (pmd_huge(*pmd)) {
940 BUG_ON(flags & FOLL_GET);
941 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
945 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
950 if (!pte_present(pte))
952 if ((flags & FOLL_WRITE) && !pte_write(pte))
954 page = vm_normal_page(vma, address, pte);
958 if (flags & FOLL_GET)
960 if (flags & FOLL_TOUCH) {
961 if ((flags & FOLL_WRITE) &&
962 !pte_dirty(pte) && !PageDirty(page))
963 set_page_dirty(page);
964 mark_page_accessed(page);
967 pte_unmap_unlock(ptep, ptl);
973 * When core dumping an enormous anonymous area that nobody
974 * has touched so far, we don't want to allocate page tables.
976 if (flags & FOLL_ANON) {
978 if (flags & FOLL_GET)
980 BUG_ON(flags & FOLL_WRITE);
985 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
986 unsigned long start, int len, int write, int force,
987 struct page **pages, struct vm_area_struct **vmas)
990 unsigned int vm_flags;
995 * Require read or write permissions.
996 * If 'force' is set, we only require the "MAY" flags.
998 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
999 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
1003 struct vm_area_struct *vma;
1004 unsigned int foll_flags;
1006 vma = find_extend_vma(mm, start);
1007 if (!vma && in_gate_area(tsk, start)) {
1008 unsigned long pg = start & PAGE_MASK;
1009 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1014 if (write) /* user gate pages are read-only */
1015 return i ? : -EFAULT;
1017 pgd = pgd_offset_k(pg);
1019 pgd = pgd_offset_gate(mm, pg);
1020 BUG_ON(pgd_none(*pgd));
1021 pud = pud_offset(pgd, pg);
1022 BUG_ON(pud_none(*pud));
1023 pmd = pmd_offset(pud, pg);
1025 return i ? : -EFAULT;
1026 pte = pte_offset_map(pmd, pg);
1027 if (pte_none(*pte)) {
1029 return i ? : -EFAULT;
1032 struct page *page = vm_normal_page(gate_vma, start, *pte);
1046 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
1047 || !(vm_flags & vma->vm_flags))
1048 return i ? : -EFAULT;
1050 if (is_vm_hugetlb_page(vma)) {
1051 i = follow_hugetlb_page(mm, vma, pages, vmas,
1052 &start, &len, i, write);
1056 foll_flags = FOLL_TOUCH;
1058 foll_flags |= FOLL_GET;
1059 if (!write && !(vma->vm_flags & VM_LOCKED) &&
1060 (!vma->vm_ops || !vma->vm_ops->fault))
1061 foll_flags |= FOLL_ANON;
1067 * If tsk is ooming, cut off its access to large memory
1068 * allocations. It has a pending SIGKILL, but it can't
1069 * be processed until returning to user space.
1071 if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE)))
1075 foll_flags |= FOLL_WRITE;
1078 while (!(page = follow_page(vma, start, foll_flags))) {
1080 ret = handle_mm_fault(mm, vma, start,
1081 foll_flags & FOLL_WRITE);
1082 if (ret & VM_FAULT_ERROR) {
1083 if (ret & VM_FAULT_OOM)
1084 return i ? i : -ENOMEM;
1085 else if (ret & VM_FAULT_SIGBUS)
1086 return i ? i : -EFAULT;
1089 if (ret & VM_FAULT_MAJOR)
1095 * The VM_FAULT_WRITE bit tells us that
1096 * do_wp_page has broken COW when necessary,
1097 * even if maybe_mkwrite decided not to set
1098 * pte_write. We can thus safely do subsequent
1099 * page lookups as if they were reads.
1101 if (ret & VM_FAULT_WRITE)
1102 foll_flags &= ~FOLL_WRITE;
1109 flush_anon_page(vma, page, start);
1110 flush_dcache_page(page);
1117 } while (len && start < vma->vm_end);
1121 EXPORT_SYMBOL(get_user_pages);
1123 pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1126 pgd_t * pgd = pgd_offset(mm, addr);
1127 pud_t * pud = pud_alloc(mm, pgd, addr);
1129 pmd_t * pmd = pmd_alloc(mm, pud, addr);
1131 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1137 * This is the old fallback for page remapping.
1139 * For historical reasons, it only allows reserved pages. Only
1140 * old drivers should use this, and they needed to mark their
1141 * pages reserved for the old functions anyway.
1143 static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1149 retval = mem_cgroup_charge(page, mm, GFP_KERNEL);
1157 flush_dcache_page(page);
1158 pte = get_locked_pte(mm, addr, &ptl);
1162 if (!pte_none(*pte))
1165 /* Ok, finally just insert the thing.. */
1167 inc_mm_counter(mm, file_rss);
1168 page_add_file_rmap(page);
1169 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1172 pte_unmap_unlock(pte, ptl);
1175 pte_unmap_unlock(pte, ptl);
1177 mem_cgroup_uncharge_page(page);
1183 * vm_insert_page - insert single page into user vma
1184 * @vma: user vma to map to
1185 * @addr: target user address of this page
1186 * @page: source kernel page
1188 * This allows drivers to insert individual pages they've allocated
1191 * The page has to be a nice clean _individual_ kernel allocation.
1192 * If you allocate a compound page, you need to have marked it as
1193 * such (__GFP_COMP), or manually just split the page up yourself
1194 * (see split_page()).
1196 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1197 * took an arbitrary page protection parameter. This doesn't allow
1198 * that. Your vma protection will have to be set up correctly, which
1199 * means that if you want a shared writable mapping, you'd better
1200 * ask for a shared writable mapping!
1202 * The page does not need to be reserved.
1204 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1206 if (addr < vma->vm_start || addr >= vma->vm_end)
1208 if (!page_count(page))
1210 vma->vm_flags |= VM_INSERTPAGE;
1211 return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1213 EXPORT_SYMBOL(vm_insert_page);
1216 * vm_insert_pfn - insert single pfn into user vma
1217 * @vma: user vma to map to
1218 * @addr: target user address of this page
1219 * @pfn: source kernel pfn
1221 * Similar to vm_inert_page, this allows drivers to insert individual pages
1222 * they've allocated into a user vma. Same comments apply.
1224 * This function should only be called from a vm_ops->fault handler, and
1225 * in that case the handler should return NULL.
1227 int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1230 struct mm_struct *mm = vma->vm_mm;
1235 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
1236 BUG_ON(is_cow_mapping(vma->vm_flags));
1239 pte = get_locked_pte(mm, addr, &ptl);
1243 if (!pte_none(*pte))
1246 /* Ok, finally just insert the thing.. */
1247 entry = pfn_pte(pfn, vma->vm_page_prot);
1248 set_pte_at(mm, addr, pte, entry);
1249 update_mmu_cache(vma, addr, entry);
1253 pte_unmap_unlock(pte, ptl);
1258 EXPORT_SYMBOL(vm_insert_pfn);
1261 * maps a range of physical memory into the requested pages. the old
1262 * mappings are removed. any references to nonexistent pages results
1263 * in null mappings (currently treated as "copy-on-access")
1265 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1266 unsigned long addr, unsigned long end,
1267 unsigned long pfn, pgprot_t prot)
1272 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1275 arch_enter_lazy_mmu_mode();
1277 BUG_ON(!pte_none(*pte));
1278 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
1280 } while (pte++, addr += PAGE_SIZE, addr != end);
1281 arch_leave_lazy_mmu_mode();
1282 pte_unmap_unlock(pte - 1, ptl);
1286 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1287 unsigned long addr, unsigned long end,
1288 unsigned long pfn, pgprot_t prot)
1293 pfn -= addr >> PAGE_SHIFT;
1294 pmd = pmd_alloc(mm, pud, addr);
1298 next = pmd_addr_end(addr, end);
1299 if (remap_pte_range(mm, pmd, addr, next,
1300 pfn + (addr >> PAGE_SHIFT), prot))
1302 } while (pmd++, addr = next, addr != end);
1306 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1307 unsigned long addr, unsigned long end,
1308 unsigned long pfn, pgprot_t prot)
1313 pfn -= addr >> PAGE_SHIFT;
1314 pud = pud_alloc(mm, pgd, addr);
1318 next = pud_addr_end(addr, end);
1319 if (remap_pmd_range(mm, pud, addr, next,
1320 pfn + (addr >> PAGE_SHIFT), prot))
1322 } while (pud++, addr = next, addr != end);
1327 * remap_pfn_range - remap kernel memory to userspace
1328 * @vma: user vma to map to
1329 * @addr: target user address to start at
1330 * @pfn: physical address of kernel memory
1331 * @size: size of map area
1332 * @prot: page protection flags for this mapping
1334 * Note: this is only safe if the mm semaphore is held when called.
1336 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1337 unsigned long pfn, unsigned long size, pgprot_t prot)
1341 unsigned long end = addr + PAGE_ALIGN(size);
1342 struct mm_struct *mm = vma->vm_mm;
1346 * Physically remapped pages are special. Tell the
1347 * rest of the world about it:
1348 * VM_IO tells people not to look at these pages
1349 * (accesses can have side effects).
1350 * VM_RESERVED is specified all over the place, because
1351 * in 2.4 it kept swapout's vma scan off this vma; but
1352 * in 2.6 the LRU scan won't even find its pages, so this
1353 * flag means no more than count its pages in reserved_vm,
1354 * and omit it from core dump, even when VM_IO turned off.
1355 * VM_PFNMAP tells the core MM that the base pages are just
1356 * raw PFN mappings, and do not have a "struct page" associated
1359 * There's a horrible special case to handle copy-on-write
1360 * behaviour that some programs depend on. We mark the "original"
1361 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1363 if (is_cow_mapping(vma->vm_flags)) {
1364 if (addr != vma->vm_start || end != vma->vm_end)
1366 vma->vm_pgoff = pfn;
1369 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
1371 BUG_ON(addr >= end);
1372 pfn -= addr >> PAGE_SHIFT;
1373 pgd = pgd_offset(mm, addr);
1374 flush_cache_range(vma, addr, end);
1376 next = pgd_addr_end(addr, end);
1377 err = remap_pud_range(mm, pgd, addr, next,
1378 pfn + (addr >> PAGE_SHIFT), prot);
1381 } while (pgd++, addr = next, addr != end);
1384 EXPORT_SYMBOL(remap_pfn_range);
1386 static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1387 unsigned long addr, unsigned long end,
1388 pte_fn_t fn, void *data)
1393 spinlock_t *uninitialized_var(ptl);
1395 pte = (mm == &init_mm) ?
1396 pte_alloc_kernel(pmd, addr) :
1397 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1401 BUG_ON(pmd_huge(*pmd));
1403 token = pmd_pgtable(*pmd);
1406 err = fn(pte, token, addr, data);
1409 } while (pte++, addr += PAGE_SIZE, addr != end);
1412 pte_unmap_unlock(pte-1, ptl);
1416 static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1417 unsigned long addr, unsigned long end,
1418 pte_fn_t fn, void *data)
1424 pmd = pmd_alloc(mm, pud, addr);
1428 next = pmd_addr_end(addr, end);
1429 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1432 } while (pmd++, addr = next, addr != end);
1436 static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1437 unsigned long addr, unsigned long end,
1438 pte_fn_t fn, void *data)
1444 pud = pud_alloc(mm, pgd, addr);
1448 next = pud_addr_end(addr, end);
1449 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1452 } while (pud++, addr = next, addr != end);
1457 * Scan a region of virtual memory, filling in page tables as necessary
1458 * and calling a provided function on each leaf page table.
1460 int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1461 unsigned long size, pte_fn_t fn, void *data)
1465 unsigned long end = addr + size;
1468 BUG_ON(addr >= end);
1469 pgd = pgd_offset(mm, addr);
1471 next = pgd_addr_end(addr, end);
1472 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1475 } while (pgd++, addr = next, addr != end);
1478 EXPORT_SYMBOL_GPL(apply_to_page_range);
1481 * handle_pte_fault chooses page fault handler according to an entry
1482 * which was read non-atomically. Before making any commitment, on
1483 * those architectures or configurations (e.g. i386 with PAE) which
1484 * might give a mix of unmatched parts, do_swap_page and do_file_page
1485 * must check under lock before unmapping the pte and proceeding
1486 * (but do_wp_page is only called after already making such a check;
1487 * and do_anonymous_page and do_no_page can safely check later on).
1489 static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
1490 pte_t *page_table, pte_t orig_pte)
1493 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1494 if (sizeof(pte_t) > sizeof(unsigned long)) {
1495 spinlock_t *ptl = pte_lockptr(mm, pmd);
1497 same = pte_same(*page_table, orig_pte);
1501 pte_unmap(page_table);
1506 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1507 * servicing faults for write access. In the normal case, do always want
1508 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1509 * that do not have writing enabled, when used by access_process_vm.
1511 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1513 if (likely(vma->vm_flags & VM_WRITE))
1514 pte = pte_mkwrite(pte);
1518 static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
1521 * If the source page was a PFN mapping, we don't have
1522 * a "struct page" for it. We do a best-effort copy by
1523 * just copying from the original user address. If that
1524 * fails, we just zero-fill it. Live with it.
1526 if (unlikely(!src)) {
1527 void *kaddr = kmap_atomic(dst, KM_USER0);
1528 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1531 * This really shouldn't fail, because the page is there
1532 * in the page tables. But it might just be unreadable,
1533 * in which case we just give up and fill the result with
1536 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
1537 memset(kaddr, 0, PAGE_SIZE);
1538 kunmap_atomic(kaddr, KM_USER0);
1539 flush_dcache_page(dst);
1541 copy_user_highpage(dst, src, va, vma);
1545 * This routine handles present pages, when users try to write
1546 * to a shared page. It is done by copying the page to a new address
1547 * and decrementing the shared-page counter for the old page.
1549 * Note that this routine assumes that the protection checks have been
1550 * done by the caller (the low-level page fault routine in most cases).
1551 * Thus we can safely just mark it writable once we've done any necessary
1554 * We also mark the page dirty at this point even though the page will
1555 * change only once the write actually happens. This avoids a few races,
1556 * and potentially makes it more efficient.
1558 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1559 * but allow concurrent faults), with pte both mapped and locked.
1560 * We return with mmap_sem still held, but pte unmapped and unlocked.
1562 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1563 unsigned long address, pte_t *page_table, pmd_t *pmd,
1564 spinlock_t *ptl, pte_t orig_pte)
1566 struct page *old_page, *new_page;
1568 int reuse = 0, ret = 0;
1569 int page_mkwrite = 0;
1570 struct page *dirty_page = NULL;
1572 old_page = vm_normal_page(vma, address, orig_pte);
1577 * Take out anonymous pages first, anonymous shared vmas are
1578 * not dirty accountable.
1580 if (PageAnon(old_page)) {
1581 if (!TestSetPageLocked(old_page)) {
1582 reuse = can_share_swap_page(old_page);
1583 unlock_page(old_page);
1585 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
1586 (VM_WRITE|VM_SHARED))) {
1588 * Only catch write-faults on shared writable pages,
1589 * read-only shared pages can get COWed by
1590 * get_user_pages(.write=1, .force=1).
1592 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1594 * Notify the address space that the page is about to
1595 * become writable so that it can prohibit this or wait
1596 * for the page to get into an appropriate state.
1598 * We do this without the lock held, so that it can
1599 * sleep if it needs to.
1601 page_cache_get(old_page);
1602 pte_unmap_unlock(page_table, ptl);
1604 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1605 goto unwritable_page;
1608 * Since we dropped the lock we need to revalidate
1609 * the PTE as someone else may have changed it. If
1610 * they did, we just return, as we can count on the
1611 * MMU to tell us if they didn't also make it writable.
1613 page_table = pte_offset_map_lock(mm, pmd, address,
1615 page_cache_release(old_page);
1616 if (!pte_same(*page_table, orig_pte))
1621 dirty_page = old_page;
1622 get_page(dirty_page);
1627 flush_cache_page(vma, address, pte_pfn(orig_pte));
1628 entry = pte_mkyoung(orig_pte);
1629 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1630 if (ptep_set_access_flags(vma, address, page_table, entry,1))
1631 update_mmu_cache(vma, address, entry);
1632 ret |= VM_FAULT_WRITE;
1637 * Ok, we need to copy. Oh, well..
1639 page_cache_get(old_page);
1641 pte_unmap_unlock(page_table, ptl);
1643 if (unlikely(anon_vma_prepare(vma)))
1645 VM_BUG_ON(old_page == ZERO_PAGE(0));
1646 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1649 cow_user_page(new_page, old_page, address, vma);
1650 __SetPageUptodate(new_page);
1652 if (mem_cgroup_charge(new_page, mm, GFP_KERNEL))
1656 * Re-check the pte - we dropped the lock
1658 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1659 if (likely(pte_same(*page_table, orig_pte))) {
1661 page_remove_rmap(old_page, vma);
1662 if (!PageAnon(old_page)) {
1663 dec_mm_counter(mm, file_rss);
1664 inc_mm_counter(mm, anon_rss);
1667 inc_mm_counter(mm, anon_rss);
1668 flush_cache_page(vma, address, pte_pfn(orig_pte));
1669 entry = mk_pte(new_page, vma->vm_page_prot);
1670 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1672 * Clear the pte entry and flush it first, before updating the
1673 * pte with the new entry. This will avoid a race condition
1674 * seen in the presence of one thread doing SMC and another
1677 ptep_clear_flush(vma, address, page_table);
1678 set_pte_at(mm, address, page_table, entry);
1679 update_mmu_cache(vma, address, entry);
1680 lru_cache_add_active(new_page);
1681 page_add_new_anon_rmap(new_page, vma, address);
1683 /* Free the old page.. */
1684 new_page = old_page;
1685 ret |= VM_FAULT_WRITE;
1687 mem_cgroup_uncharge_page(new_page);
1690 page_cache_release(new_page);
1692 page_cache_release(old_page);
1694 pte_unmap_unlock(page_table, ptl);
1697 file_update_time(vma->vm_file);
1700 * Yes, Virginia, this is actually required to prevent a race
1701 * with clear_page_dirty_for_io() from clearing the page dirty
1702 * bit after it clear all dirty ptes, but before a racing
1703 * do_wp_page installs a dirty pte.
1705 * do_no_page is protected similarly.
1707 wait_on_page_locked(dirty_page);
1708 set_page_dirty_balance(dirty_page, page_mkwrite);
1709 put_page(dirty_page);
1713 page_cache_release(new_page);
1716 page_cache_release(old_page);
1717 return VM_FAULT_OOM;
1720 page_cache_release(old_page);
1721 return VM_FAULT_SIGBUS;
1725 * Helper functions for unmap_mapping_range().
1727 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1729 * We have to restart searching the prio_tree whenever we drop the lock,
1730 * since the iterator is only valid while the lock is held, and anyway
1731 * a later vma might be split and reinserted earlier while lock dropped.
1733 * The list of nonlinear vmas could be handled more efficiently, using
1734 * a placeholder, but handle it in the same way until a need is shown.
1735 * It is important to search the prio_tree before nonlinear list: a vma
1736 * may become nonlinear and be shifted from prio_tree to nonlinear list
1737 * while the lock is dropped; but never shifted from list to prio_tree.
1739 * In order to make forward progress despite restarting the search,
1740 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1741 * quickly skip it next time around. Since the prio_tree search only
1742 * shows us those vmas affected by unmapping the range in question, we
1743 * can't efficiently keep all vmas in step with mapping->truncate_count:
1744 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1745 * mapping->truncate_count and vma->vm_truncate_count are protected by
1748 * In order to make forward progress despite repeatedly restarting some
1749 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1750 * and restart from that address when we reach that vma again. It might
1751 * have been split or merged, shrunk or extended, but never shifted: so
1752 * restart_addr remains valid so long as it remains in the vma's range.
1753 * unmap_mapping_range forces truncate_count to leap over page-aligned
1754 * values so we can save vma's restart_addr in its truncate_count field.
1756 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1758 static void reset_vma_truncate_counts(struct address_space *mapping)
1760 struct vm_area_struct *vma;
1761 struct prio_tree_iter iter;
1763 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1764 vma->vm_truncate_count = 0;
1765 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1766 vma->vm_truncate_count = 0;
1769 static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1770 unsigned long start_addr, unsigned long end_addr,
1771 struct zap_details *details)
1773 unsigned long restart_addr;
1777 * files that support invalidating or truncating portions of the
1778 * file from under mmaped areas must have their ->fault function
1779 * return a locked page (and set VM_FAULT_LOCKED in the return).
1780 * This provides synchronisation against concurrent unmapping here.
1784 restart_addr = vma->vm_truncate_count;
1785 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1786 start_addr = restart_addr;
1787 if (start_addr >= end_addr) {
1788 /* Top of vma has been split off since last time */
1789 vma->vm_truncate_count = details->truncate_count;
1794 restart_addr = zap_page_range(vma, start_addr,
1795 end_addr - start_addr, details);
1796 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
1798 if (restart_addr >= end_addr) {
1799 /* We have now completed this vma: mark it so */
1800 vma->vm_truncate_count = details->truncate_count;
1804 /* Note restart_addr in vma's truncate_count field */
1805 vma->vm_truncate_count = restart_addr;
1810 spin_unlock(details->i_mmap_lock);
1812 spin_lock(details->i_mmap_lock);
1816 static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1817 struct zap_details *details)
1819 struct vm_area_struct *vma;
1820 struct prio_tree_iter iter;
1821 pgoff_t vba, vea, zba, zea;
1824 vma_prio_tree_foreach(vma, &iter, root,
1825 details->first_index, details->last_index) {
1826 /* Skip quickly over those we have already dealt with */
1827 if (vma->vm_truncate_count == details->truncate_count)
1830 vba = vma->vm_pgoff;
1831 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1832 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1833 zba = details->first_index;
1836 zea = details->last_index;
1840 if (unmap_mapping_range_vma(vma,
1841 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1842 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1848 static inline void unmap_mapping_range_list(struct list_head *head,
1849 struct zap_details *details)
1851 struct vm_area_struct *vma;
1854 * In nonlinear VMAs there is no correspondence between virtual address
1855 * offset and file offset. So we must perform an exhaustive search
1856 * across *all* the pages in each nonlinear VMA, not just the pages
1857 * whose virtual address lies outside the file truncation point.
1860 list_for_each_entry(vma, head, shared.vm_set.list) {
1861 /* Skip quickly over those we have already dealt with */
1862 if (vma->vm_truncate_count == details->truncate_count)
1864 details->nonlinear_vma = vma;
1865 if (unmap_mapping_range_vma(vma, vma->vm_start,
1866 vma->vm_end, details) < 0)
1872 * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
1873 * @mapping: the address space containing mmaps to be unmapped.
1874 * @holebegin: byte in first page to unmap, relative to the start of
1875 * the underlying file. This will be rounded down to a PAGE_SIZE
1876 * boundary. Note that this is different from vmtruncate(), which
1877 * must keep the partial page. In contrast, we must get rid of
1879 * @holelen: size of prospective hole in bytes. This will be rounded
1880 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1882 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1883 * but 0 when invalidating pagecache, don't throw away private data.
1885 void unmap_mapping_range(struct address_space *mapping,
1886 loff_t const holebegin, loff_t const holelen, int even_cows)
1888 struct zap_details details;
1889 pgoff_t hba = holebegin >> PAGE_SHIFT;
1890 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1892 /* Check for overflow. */
1893 if (sizeof(holelen) > sizeof(hlen)) {
1895 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1896 if (holeend & ~(long long)ULONG_MAX)
1897 hlen = ULONG_MAX - hba + 1;
1900 details.check_mapping = even_cows? NULL: mapping;
1901 details.nonlinear_vma = NULL;
1902 details.first_index = hba;
1903 details.last_index = hba + hlen - 1;
1904 if (details.last_index < details.first_index)
1905 details.last_index = ULONG_MAX;
1906 details.i_mmap_lock = &mapping->i_mmap_lock;
1908 spin_lock(&mapping->i_mmap_lock);
1910 /* Protect against endless unmapping loops */
1911 mapping->truncate_count++;
1912 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1913 if (mapping->truncate_count == 0)
1914 reset_vma_truncate_counts(mapping);
1915 mapping->truncate_count++;
1917 details.truncate_count = mapping->truncate_count;
1919 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1920 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1921 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1922 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1923 spin_unlock(&mapping->i_mmap_lock);
1925 EXPORT_SYMBOL(unmap_mapping_range);
1928 * vmtruncate - unmap mappings "freed" by truncate() syscall
1929 * @inode: inode of the file used
1930 * @offset: file offset to start truncating
1932 * NOTE! We have to be ready to update the memory sharing
1933 * between the file and the memory map for a potential last
1934 * incomplete page. Ugly, but necessary.
1936 int vmtruncate(struct inode * inode, loff_t offset)
1938 if (inode->i_size < offset) {
1939 unsigned long limit;
1941 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1942 if (limit != RLIM_INFINITY && offset > limit)
1944 if (offset > inode->i_sb->s_maxbytes)
1946 i_size_write(inode, offset);
1948 struct address_space *mapping = inode->i_mapping;
1951 * truncation of in-use swapfiles is disallowed - it would
1952 * cause subsequent swapout to scribble on the now-freed
1955 if (IS_SWAPFILE(inode))
1957 i_size_write(inode, offset);
1960 * unmap_mapping_range is called twice, first simply for
1961 * efficiency so that truncate_inode_pages does fewer
1962 * single-page unmaps. However after this first call, and
1963 * before truncate_inode_pages finishes, it is possible for
1964 * private pages to be COWed, which remain after
1965 * truncate_inode_pages finishes, hence the second
1966 * unmap_mapping_range call must be made for correctness.
1968 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1969 truncate_inode_pages(mapping, offset);
1970 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1973 if (inode->i_op && inode->i_op->truncate)
1974 inode->i_op->truncate(inode);
1978 send_sig(SIGXFSZ, current, 0);
1982 EXPORT_SYMBOL(vmtruncate);
1984 int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
1986 struct address_space *mapping = inode->i_mapping;
1989 * If the underlying filesystem is not going to provide
1990 * a way to truncate a range of blocks (punch a hole) -
1991 * we should return failure right now.
1993 if (!inode->i_op || !inode->i_op->truncate_range)
1996 mutex_lock(&inode->i_mutex);
1997 down_write(&inode->i_alloc_sem);
1998 unmap_mapping_range(mapping, offset, (end - offset), 1);
1999 truncate_inode_pages_range(mapping, offset, end);
2000 unmap_mapping_range(mapping, offset, (end - offset), 1);
2001 inode->i_op->truncate_range(inode, offset, end);
2002 up_write(&inode->i_alloc_sem);
2003 mutex_unlock(&inode->i_mutex);
2009 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2010 * but allow concurrent faults), and pte mapped but not yet locked.
2011 * We return with mmap_sem still held, but pte unmapped and unlocked.
2013 static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2014 unsigned long address, pte_t *page_table, pmd_t *pmd,
2015 int write_access, pte_t orig_pte)
2023 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2026 entry = pte_to_swp_entry(orig_pte);
2027 if (is_migration_entry(entry)) {
2028 migration_entry_wait(mm, pmd, address);
2031 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
2032 page = lookup_swap_cache(entry);
2034 grab_swap_token(); /* Contend for token _before_ read-in */
2035 page = swapin_readahead(entry,
2036 GFP_HIGHUSER_MOVABLE, vma, address);
2039 * Back out if somebody else faulted in this pte
2040 * while we released the pte lock.
2042 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2043 if (likely(pte_same(*page_table, orig_pte)))
2045 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2049 /* Had to read the page from swap area: Major fault */
2050 ret = VM_FAULT_MAJOR;
2051 count_vm_event(PGMAJFAULT);
2054 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
2055 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2060 mark_page_accessed(page);
2062 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2065 * Back out if somebody else already faulted in this pte.
2067 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2068 if (unlikely(!pte_same(*page_table, orig_pte)))
2071 if (unlikely(!PageUptodate(page))) {
2072 ret = VM_FAULT_SIGBUS;
2076 /* The page isn't present yet, go ahead with the fault. */
2078 inc_mm_counter(mm, anon_rss);
2079 pte = mk_pte(page, vma->vm_page_prot);
2080 if (write_access && can_share_swap_page(page)) {
2081 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2085 flush_icache_page(vma, page);
2086 set_pte_at(mm, address, page_table, pte);
2087 page_add_anon_rmap(page, vma, address);
2091 remove_exclusive_swap_page(page);
2095 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2096 if (ret & VM_FAULT_ERROR)
2097 ret &= VM_FAULT_ERROR;
2101 /* No need to invalidate - it was non-present before */
2102 update_mmu_cache(vma, address, pte);
2104 pte_unmap_unlock(page_table, ptl);
2108 mem_cgroup_uncharge_page(page);
2109 pte_unmap_unlock(page_table, ptl);
2111 page_cache_release(page);
2116 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2117 * but allow concurrent faults), and pte mapped but not yet locked.
2118 * We return with mmap_sem still held, but pte unmapped and unlocked.
2120 static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2121 unsigned long address, pte_t *page_table, pmd_t *pmd,
2128 /* Allocate our own private page. */
2129 pte_unmap(page_table);
2131 if (unlikely(anon_vma_prepare(vma)))
2133 page = alloc_zeroed_user_highpage_movable(vma, address);
2136 __SetPageUptodate(page);
2138 if (mem_cgroup_charge(page, mm, GFP_KERNEL))
2141 entry = mk_pte(page, vma->vm_page_prot);
2142 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2144 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2145 if (!pte_none(*page_table))
2147 inc_mm_counter(mm, anon_rss);
2148 lru_cache_add_active(page);
2149 page_add_new_anon_rmap(page, vma, address);
2150 set_pte_at(mm, address, page_table, entry);
2152 /* No need to invalidate - it was non-present before */
2153 update_mmu_cache(vma, address, entry);
2155 pte_unmap_unlock(page_table, ptl);
2158 mem_cgroup_uncharge_page(page);
2159 page_cache_release(page);
2162 page_cache_release(page);
2164 return VM_FAULT_OOM;
2168 * __do_fault() tries to create a new page mapping. It aggressively
2169 * tries to share with existing pages, but makes a separate copy if
2170 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2171 * the next page fault.
2173 * As this is called only for pages that do not currently exist, we
2174 * do not need to flush old virtual caches or the TLB.
2176 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2177 * but allow concurrent faults), and pte neither mapped nor locked.
2178 * We return with mmap_sem still held, but pte unmapped and unlocked.
2180 static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2181 unsigned long address, pmd_t *pmd,
2182 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
2189 struct page *dirty_page = NULL;
2190 struct vm_fault vmf;
2192 int page_mkwrite = 0;
2194 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2199 BUG_ON(vma->vm_flags & VM_PFNMAP);
2201 ret = vma->vm_ops->fault(vma, &vmf);
2202 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2206 * For consistency in subsequent calls, make the faulted page always
2209 if (unlikely(!(ret & VM_FAULT_LOCKED)))
2210 lock_page(vmf.page);
2212 VM_BUG_ON(!PageLocked(vmf.page));
2215 * Should we do an early C-O-W break?
2218 if (flags & FAULT_FLAG_WRITE) {
2219 if (!(vma->vm_flags & VM_SHARED)) {
2221 if (unlikely(anon_vma_prepare(vma))) {
2225 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2231 copy_user_highpage(page, vmf.page, address, vma);
2232 __SetPageUptodate(page);
2235 * If the page will be shareable, see if the backing
2236 * address space wants to know that the page is about
2237 * to become writable
2239 if (vma->vm_ops->page_mkwrite) {
2241 if (vma->vm_ops->page_mkwrite(vma, page) < 0) {
2242 ret = VM_FAULT_SIGBUS;
2243 anon = 1; /* no anon but release vmf.page */
2248 * XXX: this is not quite right (racy vs
2249 * invalidate) to unlock and relock the page
2250 * like this, however a better fix requires
2251 * reworking page_mkwrite locking API, which
2252 * is better done later.
2254 if (!page->mapping) {
2256 anon = 1; /* no anon but release vmf.page */
2265 if (mem_cgroup_charge(page, mm, GFP_KERNEL)) {
2270 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2273 * This silly early PAGE_DIRTY setting removes a race
2274 * due to the bad i386 page protection. But it's valid
2275 * for other architectures too.
2277 * Note that if write_access is true, we either now have
2278 * an exclusive copy of the page, or this is a shared mapping,
2279 * so we can make it writable and dirty to avoid having to
2280 * handle that later.
2282 /* Only go through if we didn't race with anybody else... */
2283 if (likely(pte_same(*page_table, orig_pte))) {
2284 flush_icache_page(vma, page);
2285 entry = mk_pte(page, vma->vm_page_prot);
2286 if (flags & FAULT_FLAG_WRITE)
2287 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2288 set_pte_at(mm, address, page_table, entry);
2290 inc_mm_counter(mm, anon_rss);
2291 lru_cache_add_active(page);
2292 page_add_new_anon_rmap(page, vma, address);
2294 inc_mm_counter(mm, file_rss);
2295 page_add_file_rmap(page);
2296 if (flags & FAULT_FLAG_WRITE) {
2298 get_page(dirty_page);
2302 /* no need to invalidate: a not-present page won't be cached */
2303 update_mmu_cache(vma, address, entry);
2305 mem_cgroup_uncharge_page(page);
2307 page_cache_release(page);
2309 anon = 1; /* no anon but release faulted_page */
2312 pte_unmap_unlock(page_table, ptl);
2315 unlock_page(vmf.page);
2318 page_cache_release(vmf.page);
2319 else if (dirty_page) {
2321 file_update_time(vma->vm_file);
2323 set_page_dirty_balance(dirty_page, page_mkwrite);
2324 put_page(dirty_page);
2330 static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2331 unsigned long address, pte_t *page_table, pmd_t *pmd,
2332 int write_access, pte_t orig_pte)
2334 pgoff_t pgoff = (((address & PAGE_MASK)
2335 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2336 unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0);
2338 pte_unmap(page_table);
2339 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
2344 * do_no_pfn() tries to create a new page mapping for a page without
2345 * a struct_page backing it
2347 * As this is called only for pages that do not currently exist, we
2348 * do not need to flush old virtual caches or the TLB.
2350 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2351 * but allow concurrent faults), and pte mapped but not yet locked.
2352 * We return with mmap_sem still held, but pte unmapped and unlocked.
2354 * It is expected that the ->nopfn handler always returns the same pfn
2355 * for a given virtual mapping.
2357 * Mark this `noinline' to prevent it from bloating the main pagefault code.
2359 static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2360 unsigned long address, pte_t *page_table, pmd_t *pmd,
2367 pte_unmap(page_table);
2368 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
2369 BUG_ON(is_cow_mapping(vma->vm_flags));
2371 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
2372 if (unlikely(pfn == NOPFN_OOM))
2373 return VM_FAULT_OOM;
2374 else if (unlikely(pfn == NOPFN_SIGBUS))
2375 return VM_FAULT_SIGBUS;
2376 else if (unlikely(pfn == NOPFN_REFAULT))
2379 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2381 /* Only go through if we didn't race with anybody else... */
2382 if (pte_none(*page_table)) {
2383 entry = pfn_pte(pfn, vma->vm_page_prot);
2385 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2386 set_pte_at(mm, address, page_table, entry);
2388 pte_unmap_unlock(page_table, ptl);
2393 * Fault of a previously existing named mapping. Repopulate the pte
2394 * from the encoded file_pte if possible. This enables swappable
2397 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2398 * but allow concurrent faults), and pte mapped but not yet locked.
2399 * We return with mmap_sem still held, but pte unmapped and unlocked.
2401 static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2402 unsigned long address, pte_t *page_table, pmd_t *pmd,
2403 int write_access, pte_t orig_pte)
2405 unsigned int flags = FAULT_FLAG_NONLINEAR |
2406 (write_access ? FAULT_FLAG_WRITE : 0);
2409 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2412 if (unlikely(!(vma->vm_flags & VM_NONLINEAR) ||
2413 !(vma->vm_flags & VM_CAN_NONLINEAR))) {
2415 * Page table corrupted: show pte and kill process.
2417 print_bad_pte(vma, orig_pte, address);
2418 return VM_FAULT_OOM;
2421 pgoff = pte_to_pgoff(orig_pte);
2422 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
2426 * These routines also need to handle stuff like marking pages dirty
2427 * and/or accessed for architectures that don't do it in hardware (most
2428 * RISC architectures). The early dirtying is also good on the i386.
2430 * There is also a hook called "update_mmu_cache()" that architectures
2431 * with external mmu caches can use to update those (ie the Sparc or
2432 * PowerPC hashed page tables that act as extended TLBs).
2434 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2435 * but allow concurrent faults), and pte mapped but not yet locked.
2436 * We return with mmap_sem still held, but pte unmapped and unlocked.
2438 static inline int handle_pte_fault(struct mm_struct *mm,
2439 struct vm_area_struct *vma, unsigned long address,
2440 pte_t *pte, pmd_t *pmd, int write_access)
2446 if (!pte_present(entry)) {
2447 if (pte_none(entry)) {
2449 if (likely(vma->vm_ops->fault))
2450 return do_linear_fault(mm, vma, address,
2451 pte, pmd, write_access, entry);
2452 if (unlikely(vma->vm_ops->nopfn))
2453 return do_no_pfn(mm, vma, address, pte,
2456 return do_anonymous_page(mm, vma, address,
2457 pte, pmd, write_access);
2459 if (pte_file(entry))
2460 return do_nonlinear_fault(mm, vma, address,
2461 pte, pmd, write_access, entry);
2462 return do_swap_page(mm, vma, address,
2463 pte, pmd, write_access, entry);
2466 ptl = pte_lockptr(mm, pmd);
2468 if (unlikely(!pte_same(*pte, entry)))
2471 if (!pte_write(entry))
2472 return do_wp_page(mm, vma, address,
2473 pte, pmd, ptl, entry);
2474 entry = pte_mkdirty(entry);
2476 entry = pte_mkyoung(entry);
2477 if (ptep_set_access_flags(vma, address, pte, entry, write_access)) {
2478 update_mmu_cache(vma, address, entry);
2481 * This is needed only for protection faults but the arch code
2482 * is not yet telling us if this is a protection fault or not.
2483 * This still avoids useless tlb flushes for .text page faults
2487 flush_tlb_page(vma, address);
2490 pte_unmap_unlock(pte, ptl);
2495 * By the time we get here, we already hold the mm semaphore
2497 int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2498 unsigned long address, int write_access)
2505 __set_current_state(TASK_RUNNING);
2507 count_vm_event(PGFAULT);
2509 if (unlikely(is_vm_hugetlb_page(vma)))
2510 return hugetlb_fault(mm, vma, address, write_access);
2512 pgd = pgd_offset(mm, address);
2513 pud = pud_alloc(mm, pgd, address);
2515 return VM_FAULT_OOM;
2516 pmd = pmd_alloc(mm, pud, address);
2518 return VM_FAULT_OOM;
2519 pte = pte_alloc_map(mm, pmd, address);
2521 return VM_FAULT_OOM;
2523 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
2526 #ifndef __PAGETABLE_PUD_FOLDED
2528 * Allocate page upper directory.
2529 * We've already handled the fast-path in-line.
2531 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2533 pud_t *new = pud_alloc_one(mm, address);
2537 spin_lock(&mm->page_table_lock);
2538 if (pgd_present(*pgd)) /* Another has populated it */
2541 pgd_populate(mm, pgd, new);
2542 spin_unlock(&mm->page_table_lock);
2545 #endif /* __PAGETABLE_PUD_FOLDED */
2547 #ifndef __PAGETABLE_PMD_FOLDED
2549 * Allocate page middle directory.
2550 * We've already handled the fast-path in-line.
2552 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2554 pmd_t *new = pmd_alloc_one(mm, address);
2558 spin_lock(&mm->page_table_lock);
2559 #ifndef __ARCH_HAS_4LEVEL_HACK
2560 if (pud_present(*pud)) /* Another has populated it */
2563 pud_populate(mm, pud, new);
2565 if (pgd_present(*pud)) /* Another has populated it */
2568 pgd_populate(mm, pud, new);
2569 #endif /* __ARCH_HAS_4LEVEL_HACK */
2570 spin_unlock(&mm->page_table_lock);
2573 #endif /* __PAGETABLE_PMD_FOLDED */
2575 int make_pages_present(unsigned long addr, unsigned long end)
2577 int ret, len, write;
2578 struct vm_area_struct * vma;
2580 vma = find_vma(current->mm, addr);
2583 write = (vma->vm_flags & VM_WRITE) != 0;
2584 BUG_ON(addr >= end);
2585 BUG_ON(end > vma->vm_end);
2586 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
2587 ret = get_user_pages(current, current->mm, addr,
2588 len, write, 0, NULL, NULL);
2591 return ret == len ? 0 : -1;
2594 #if !defined(__HAVE_ARCH_GATE_AREA)
2596 #if defined(AT_SYSINFO_EHDR)
2597 static struct vm_area_struct gate_vma;
2599 static int __init gate_vma_init(void)
2601 gate_vma.vm_mm = NULL;
2602 gate_vma.vm_start = FIXADDR_USER_START;
2603 gate_vma.vm_end = FIXADDR_USER_END;
2604 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2605 gate_vma.vm_page_prot = __P101;
2607 * Make sure the vDSO gets into every core dump.
2608 * Dumping its contents makes post-mortem fully interpretable later
2609 * without matching up the same kernel and hardware config to see
2610 * what PC values meant.
2612 gate_vma.vm_flags |= VM_ALWAYSDUMP;
2615 __initcall(gate_vma_init);
2618 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2620 #ifdef AT_SYSINFO_EHDR
2627 int in_gate_area_no_task(unsigned long addr)
2629 #ifdef AT_SYSINFO_EHDR
2630 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2636 #endif /* __HAVE_ARCH_GATE_AREA */
2639 * Access another process' address space.
2640 * Source/target buffer must be kernel space,
2641 * Do not walk the page table directly, use get_user_pages
2643 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2645 struct mm_struct *mm;
2646 struct vm_area_struct *vma;
2648 void *old_buf = buf;
2650 mm = get_task_mm(tsk);
2654 down_read(&mm->mmap_sem);
2655 /* ignore errors, just check how much was successfully transferred */
2657 int bytes, ret, offset;
2660 ret = get_user_pages(tsk, mm, addr, 1,
2661 write, 1, &page, &vma);
2666 offset = addr & (PAGE_SIZE-1);
2667 if (bytes > PAGE_SIZE-offset)
2668 bytes = PAGE_SIZE-offset;
2672 copy_to_user_page(vma, page, addr,
2673 maddr + offset, buf, bytes);
2674 set_page_dirty_lock(page);
2676 copy_from_user_page(vma, page, addr,
2677 buf, maddr + offset, bytes);
2680 page_cache_release(page);
2685 up_read(&mm->mmap_sem);
2688 return buf - old_buf;
2692 * Print the name of a VMA.
2694 void print_vma_addr(char *prefix, unsigned long ip)
2696 struct mm_struct *mm = current->mm;
2697 struct vm_area_struct *vma;
2700 * Do not print if we are in atomic
2701 * contexts (in exception stacks, etc.):
2703 if (preempt_count())
2706 down_read(&mm->mmap_sem);
2707 vma = find_vma(mm, ip);
2708 if (vma && vma->vm_file) {
2709 struct file *f = vma->vm_file;
2710 char *buf = (char *)__get_free_page(GFP_KERNEL);
2714 p = d_path(&f->f_path, buf, PAGE_SIZE);
2717 s = strrchr(p, '/');
2720 printk("%s%s[%lx+%lx]", prefix, p,
2722 vma->vm_end - vma->vm_start);
2723 free_page((unsigned long)buf);
2726 up_read(¤t->mm->mmap_sem);