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>
59 #include <linux/gfp.h>
62 #include <asm/pgalloc.h>
63 #include <asm/uaccess.h>
65 #include <asm/tlbflush.h>
66 #include <asm/pgtable.h>
70 #ifndef CONFIG_NEED_MULTIPLE_NODES
71 /* use the per-pgdat data instead for discontigmem - mbligh */
72 unsigned long max_mapnr;
75 EXPORT_SYMBOL(max_mapnr);
76 EXPORT_SYMBOL(mem_map);
79 unsigned long num_physpages;
81 * A number of key systems in x86 including ioremap() rely on the assumption
82 * that high_memory defines the upper bound on direct map memory, then end
83 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
84 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
89 EXPORT_SYMBOL(num_physpages);
90 EXPORT_SYMBOL(high_memory);
93 * Randomize the address space (stacks, mmaps, brk, etc.).
95 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
96 * as ancient (libc5 based) binaries can segfault. )
98 int randomize_va_space __read_mostly =
99 #ifdef CONFIG_COMPAT_BRK
105 static int __init disable_randmaps(char *s)
107 randomize_va_space = 0;
110 __setup("norandmaps", disable_randmaps);
112 unsigned long zero_pfn __read_mostly;
113 unsigned long highest_memmap_pfn __read_mostly;
116 * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
118 static int __init init_zero_pfn(void)
120 zero_pfn = page_to_pfn(ZERO_PAGE(0));
123 core_initcall(init_zero_pfn);
126 #if defined(SPLIT_RSS_COUNTING)
128 static void __sync_task_rss_stat(struct task_struct *task, struct mm_struct *mm)
132 for (i = 0; i < NR_MM_COUNTERS; i++) {
133 if (task->rss_stat.count[i]) {
134 add_mm_counter(mm, i, task->rss_stat.count[i]);
135 task->rss_stat.count[i] = 0;
138 task->rss_stat.events = 0;
141 static void add_mm_counter_fast(struct mm_struct *mm, int member, int val)
143 struct task_struct *task = current;
145 if (likely(task->mm == mm))
146 task->rss_stat.count[member] += val;
148 add_mm_counter(mm, member, val);
150 #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
151 #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
153 /* sync counter once per 64 page faults */
154 #define TASK_RSS_EVENTS_THRESH (64)
155 static void check_sync_rss_stat(struct task_struct *task)
157 if (unlikely(task != current))
159 if (unlikely(task->rss_stat.events++ > TASK_RSS_EVENTS_THRESH))
160 __sync_task_rss_stat(task, task->mm);
163 unsigned long get_mm_counter(struct mm_struct *mm, int member)
168 * Don't use task->mm here...for avoiding to use task_get_mm()..
169 * The caller must guarantee task->mm is not invalid.
171 val = atomic_long_read(&mm->rss_stat.count[member]);
173 * counter is updated in asynchronous manner and may go to minus.
174 * But it's never be expected number for users.
178 return (unsigned long)val;
181 void sync_mm_rss(struct task_struct *task, struct mm_struct *mm)
183 __sync_task_rss_stat(task, mm);
187 #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
188 #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
190 static void check_sync_rss_stat(struct task_struct *task)
197 * If a p?d_bad entry is found while walking page tables, report
198 * the error, before resetting entry to p?d_none. Usually (but
199 * very seldom) called out from the p?d_none_or_clear_bad macros.
202 void pgd_clear_bad(pgd_t *pgd)
208 void pud_clear_bad(pud_t *pud)
214 void pmd_clear_bad(pmd_t *pmd)
221 * Note: this doesn't free the actual pages themselves. That
222 * has been handled earlier when unmapping all the memory regions.
224 static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
227 pgtable_t token = pmd_pgtable(*pmd);
229 pte_free_tlb(tlb, token, addr);
233 static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
234 unsigned long addr, unsigned long end,
235 unsigned long floor, unsigned long ceiling)
242 pmd = pmd_offset(pud, addr);
244 next = pmd_addr_end(addr, end);
245 if (pmd_none_or_clear_bad(pmd))
247 free_pte_range(tlb, pmd, addr);
248 } while (pmd++, addr = next, addr != end);
258 if (end - 1 > ceiling - 1)
261 pmd = pmd_offset(pud, start);
263 pmd_free_tlb(tlb, pmd, start);
266 static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
267 unsigned long addr, unsigned long end,
268 unsigned long floor, unsigned long ceiling)
275 pud = pud_offset(pgd, addr);
277 next = pud_addr_end(addr, end);
278 if (pud_none_or_clear_bad(pud))
280 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
281 } while (pud++, addr = next, addr != end);
287 ceiling &= PGDIR_MASK;
291 if (end - 1 > ceiling - 1)
294 pud = pud_offset(pgd, start);
296 pud_free_tlb(tlb, pud, start);
300 * This function frees user-level page tables of a process.
302 * Must be called with pagetable lock held.
304 void free_pgd_range(struct mmu_gather *tlb,
305 unsigned long addr, unsigned long end,
306 unsigned long floor, unsigned long ceiling)
313 * The next few lines have given us lots of grief...
315 * Why are we testing PMD* at this top level? Because often
316 * there will be no work to do at all, and we'd prefer not to
317 * go all the way down to the bottom just to discover that.
319 * Why all these "- 1"s? Because 0 represents both the bottom
320 * of the address space and the top of it (using -1 for the
321 * top wouldn't help much: the masks would do the wrong thing).
322 * The rule is that addr 0 and floor 0 refer to the bottom of
323 * the address space, but end 0 and ceiling 0 refer to the top
324 * Comparisons need to use "end - 1" and "ceiling - 1" (though
325 * that end 0 case should be mythical).
327 * Wherever addr is brought up or ceiling brought down, we must
328 * be careful to reject "the opposite 0" before it confuses the
329 * subsequent tests. But what about where end is brought down
330 * by PMD_SIZE below? no, end can't go down to 0 there.
332 * Whereas we round start (addr) and ceiling down, by different
333 * masks at different levels, in order to test whether a table
334 * now has no other vmas using it, so can be freed, we don't
335 * bother to round floor or end up - the tests don't need that.
349 if (end - 1 > ceiling - 1)
355 pgd = pgd_offset(tlb->mm, addr);
357 next = pgd_addr_end(addr, end);
358 if (pgd_none_or_clear_bad(pgd))
360 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
361 } while (pgd++, addr = next, addr != end);
364 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
365 unsigned long floor, unsigned long ceiling)
368 struct vm_area_struct *next = vma->vm_next;
369 unsigned long addr = vma->vm_start;
372 * Hide vma from rmap and truncate_pagecache before freeing
375 unlink_anon_vmas(vma);
376 unlink_file_vma(vma);
378 if (is_vm_hugetlb_page(vma)) {
379 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
380 floor, next? next->vm_start: ceiling);
383 * Optimization: gather nearby vmas into one call down
385 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
386 && !is_vm_hugetlb_page(next)) {
389 unlink_anon_vmas(vma);
390 unlink_file_vma(vma);
392 free_pgd_range(tlb, addr, vma->vm_end,
393 floor, next? next->vm_start: ceiling);
399 int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
401 pgtable_t new = pte_alloc_one(mm, address);
406 * Ensure all pte setup (eg. pte page lock and page clearing) are
407 * visible before the pte is made visible to other CPUs by being
408 * put into page tables.
410 * The other side of the story is the pointer chasing in the page
411 * table walking code (when walking the page table without locking;
412 * ie. most of the time). Fortunately, these data accesses consist
413 * of a chain of data-dependent loads, meaning most CPUs (alpha
414 * being the notable exception) will already guarantee loads are
415 * seen in-order. See the alpha page table accessors for the
416 * smp_read_barrier_depends() barriers in page table walking code.
418 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
420 spin_lock(&mm->page_table_lock);
421 if (!pmd_present(*pmd)) { /* Has another populated it ? */
423 pmd_populate(mm, pmd, new);
426 spin_unlock(&mm->page_table_lock);
432 int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
434 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
438 smp_wmb(); /* See comment in __pte_alloc */
440 spin_lock(&init_mm.page_table_lock);
441 if (!pmd_present(*pmd)) { /* Has another populated it ? */
442 pmd_populate_kernel(&init_mm, pmd, new);
445 spin_unlock(&init_mm.page_table_lock);
447 pte_free_kernel(&init_mm, new);
451 static inline void init_rss_vec(int *rss)
453 memset(rss, 0, sizeof(int) * NR_MM_COUNTERS);
456 static inline void add_mm_rss_vec(struct mm_struct *mm, int *rss)
460 if (current->mm == mm)
461 sync_mm_rss(current, mm);
462 for (i = 0; i < NR_MM_COUNTERS; i++)
464 add_mm_counter(mm, i, rss[i]);
468 * This function is called to print an error when a bad pte
469 * is found. For example, we might have a PFN-mapped pte in
470 * a region that doesn't allow it.
472 * The calling function must still handle the error.
474 static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
475 pte_t pte, struct page *page)
477 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
478 pud_t *pud = pud_offset(pgd, addr);
479 pmd_t *pmd = pmd_offset(pud, addr);
480 struct address_space *mapping;
482 static unsigned long resume;
483 static unsigned long nr_shown;
484 static unsigned long nr_unshown;
487 * Allow a burst of 60 reports, then keep quiet for that minute;
488 * or allow a steady drip of one report per second.
490 if (nr_shown == 60) {
491 if (time_before(jiffies, resume)) {
497 "BUG: Bad page map: %lu messages suppressed\n",
504 resume = jiffies + 60 * HZ;
506 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
507 index = linear_page_index(vma, addr);
510 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
512 (long long)pte_val(pte), (long long)pmd_val(*pmd));
516 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
517 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
519 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
522 print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n",
523 (unsigned long)vma->vm_ops->fault);
524 if (vma->vm_file && vma->vm_file->f_op)
525 print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n",
526 (unsigned long)vma->vm_file->f_op->mmap);
528 add_taint(TAINT_BAD_PAGE);
531 static inline int is_cow_mapping(unsigned int flags)
533 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
537 static inline int is_zero_pfn(unsigned long pfn)
539 return pfn == zero_pfn;
544 static inline unsigned long my_zero_pfn(unsigned long addr)
551 * vm_normal_page -- This function gets the "struct page" associated with a pte.
553 * "Special" mappings do not wish to be associated with a "struct page" (either
554 * it doesn't exist, or it exists but they don't want to touch it). In this
555 * case, NULL is returned here. "Normal" mappings do have a struct page.
557 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
558 * pte bit, in which case this function is trivial. Secondly, an architecture
559 * may not have a spare pte bit, which requires a more complicated scheme,
562 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
563 * special mapping (even if there are underlying and valid "struct pages").
564 * COWed pages of a VM_PFNMAP are always normal.
566 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
567 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
568 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
569 * mapping will always honor the rule
571 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
573 * And for normal mappings this is false.
575 * This restricts such mappings to be a linear translation from virtual address
576 * to pfn. To get around this restriction, we allow arbitrary mappings so long
577 * as the vma is not a COW mapping; in that case, we know that all ptes are
578 * special (because none can have been COWed).
581 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
583 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
584 * page" backing, however the difference is that _all_ pages with a struct
585 * page (that is, those where pfn_valid is true) are refcounted and considered
586 * normal pages by the VM. The disadvantage is that pages are refcounted
587 * (which can be slower and simply not an option for some PFNMAP users). The
588 * advantage is that we don't have to follow the strict linearity rule of
589 * PFNMAP mappings in order to support COWable mappings.
592 #ifdef __HAVE_ARCH_PTE_SPECIAL
593 # define HAVE_PTE_SPECIAL 1
595 # define HAVE_PTE_SPECIAL 0
597 struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
600 unsigned long pfn = pte_pfn(pte);
602 if (HAVE_PTE_SPECIAL) {
603 if (likely(!pte_special(pte)))
605 if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
607 if (!is_zero_pfn(pfn))
608 print_bad_pte(vma, addr, pte, NULL);
612 /* !HAVE_PTE_SPECIAL case follows: */
614 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
615 if (vma->vm_flags & VM_MIXEDMAP) {
621 off = (addr - vma->vm_start) >> PAGE_SHIFT;
622 if (pfn == vma->vm_pgoff + off)
624 if (!is_cow_mapping(vma->vm_flags))
629 if (is_zero_pfn(pfn))
632 if (unlikely(pfn > highest_memmap_pfn)) {
633 print_bad_pte(vma, addr, pte, NULL);
638 * NOTE! We still have PageReserved() pages in the page tables.
639 * eg. VDSO mappings can cause them to exist.
642 return pfn_to_page(pfn);
646 * copy one vm_area from one task to the other. Assumes the page tables
647 * already present in the new task to be cleared in the whole range
648 * covered by this vma.
651 static inline unsigned long
652 copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
653 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
654 unsigned long addr, int *rss)
656 unsigned long vm_flags = vma->vm_flags;
657 pte_t pte = *src_pte;
660 /* pte contains position in swap or file, so copy. */
661 if (unlikely(!pte_present(pte))) {
662 if (!pte_file(pte)) {
663 swp_entry_t entry = pte_to_swp_entry(pte);
665 if (swap_duplicate(entry) < 0)
668 /* make sure dst_mm is on swapoff's mmlist. */
669 if (unlikely(list_empty(&dst_mm->mmlist))) {
670 spin_lock(&mmlist_lock);
671 if (list_empty(&dst_mm->mmlist))
672 list_add(&dst_mm->mmlist,
674 spin_unlock(&mmlist_lock);
676 if (likely(!non_swap_entry(entry)))
678 else if (is_write_migration_entry(entry) &&
679 is_cow_mapping(vm_flags)) {
681 * COW mappings require pages in both parent
682 * and child to be set to read.
684 make_migration_entry_read(&entry);
685 pte = swp_entry_to_pte(entry);
686 set_pte_at(src_mm, addr, src_pte, pte);
693 * If it's a COW mapping, write protect it both
694 * in the parent and the child
696 if (is_cow_mapping(vm_flags)) {
697 ptep_set_wrprotect(src_mm, addr, src_pte);
698 pte = pte_wrprotect(pte);
702 * If it's a shared mapping, mark it clean in
705 if (vm_flags & VM_SHARED)
706 pte = pte_mkclean(pte);
707 pte = pte_mkold(pte);
709 page = vm_normal_page(vma, addr, pte);
720 set_pte_at(dst_mm, addr, dst_pte, pte);
724 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
725 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
726 unsigned long addr, unsigned long end)
728 pte_t *orig_src_pte, *orig_dst_pte;
729 pte_t *src_pte, *dst_pte;
730 spinlock_t *src_ptl, *dst_ptl;
732 int rss[NR_MM_COUNTERS];
733 swp_entry_t entry = (swp_entry_t){0};
738 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
741 src_pte = pte_offset_map_nested(src_pmd, addr);
742 src_ptl = pte_lockptr(src_mm, src_pmd);
743 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
744 orig_src_pte = src_pte;
745 orig_dst_pte = dst_pte;
746 arch_enter_lazy_mmu_mode();
750 * We are holding two locks at this point - either of them
751 * could generate latencies in another task on another CPU.
753 if (progress >= 32) {
755 if (need_resched() ||
756 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
759 if (pte_none(*src_pte)) {
763 entry.val = copy_one_pte(dst_mm, src_mm, dst_pte, src_pte,
768 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
770 arch_leave_lazy_mmu_mode();
771 spin_unlock(src_ptl);
772 pte_unmap_nested(orig_src_pte);
773 add_mm_rss_vec(dst_mm, rss);
774 pte_unmap_unlock(orig_dst_pte, dst_ptl);
778 if (add_swap_count_continuation(entry, GFP_KERNEL) < 0)
787 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
788 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
789 unsigned long addr, unsigned long end)
791 pmd_t *src_pmd, *dst_pmd;
794 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
797 src_pmd = pmd_offset(src_pud, addr);
799 next = pmd_addr_end(addr, end);
800 if (pmd_none_or_clear_bad(src_pmd))
802 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
805 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
809 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
810 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
811 unsigned long addr, unsigned long end)
813 pud_t *src_pud, *dst_pud;
816 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
819 src_pud = pud_offset(src_pgd, addr);
821 next = pud_addr_end(addr, end);
822 if (pud_none_or_clear_bad(src_pud))
824 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
827 } while (dst_pud++, src_pud++, addr = next, addr != end);
831 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
832 struct vm_area_struct *vma)
834 pgd_t *src_pgd, *dst_pgd;
836 unsigned long addr = vma->vm_start;
837 unsigned long end = vma->vm_end;
841 * Don't copy ptes where a page fault will fill them correctly.
842 * Fork becomes much lighter when there are big shared or private
843 * readonly mappings. The tradeoff is that copy_page_range is more
844 * efficient than faulting.
846 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
851 if (is_vm_hugetlb_page(vma))
852 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
854 if (unlikely(is_pfn_mapping(vma))) {
856 * We do not free on error cases below as remove_vma
857 * gets called on error from higher level routine
859 ret = track_pfn_vma_copy(vma);
865 * We need to invalidate the secondary MMU mappings only when
866 * there could be a permission downgrade on the ptes of the
867 * parent mm. And a permission downgrade will only happen if
868 * is_cow_mapping() returns true.
870 if (is_cow_mapping(vma->vm_flags))
871 mmu_notifier_invalidate_range_start(src_mm, addr, end);
874 dst_pgd = pgd_offset(dst_mm, addr);
875 src_pgd = pgd_offset(src_mm, addr);
877 next = pgd_addr_end(addr, end);
878 if (pgd_none_or_clear_bad(src_pgd))
880 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
885 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
887 if (is_cow_mapping(vma->vm_flags))
888 mmu_notifier_invalidate_range_end(src_mm,
893 static unsigned long zap_pte_range(struct mmu_gather *tlb,
894 struct vm_area_struct *vma, pmd_t *pmd,
895 unsigned long addr, unsigned long end,
896 long *zap_work, struct zap_details *details)
898 struct mm_struct *mm = tlb->mm;
901 int rss[NR_MM_COUNTERS];
905 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
906 arch_enter_lazy_mmu_mode();
909 if (pte_none(ptent)) {
914 (*zap_work) -= PAGE_SIZE;
916 if (pte_present(ptent)) {
919 page = vm_normal_page(vma, addr, ptent);
920 if (unlikely(details) && page) {
922 * unmap_shared_mapping_pages() wants to
923 * invalidate cache without truncating:
924 * unmap shared but keep private pages.
926 if (details->check_mapping &&
927 details->check_mapping != page->mapping)
930 * Each page->index must be checked when
931 * invalidating or truncating nonlinear.
933 if (details->nonlinear_vma &&
934 (page->index < details->first_index ||
935 page->index > details->last_index))
938 ptent = ptep_get_and_clear_full(mm, addr, pte,
940 tlb_remove_tlb_entry(tlb, pte, addr);
943 if (unlikely(details) && details->nonlinear_vma
944 && linear_page_index(details->nonlinear_vma,
945 addr) != page->index)
946 set_pte_at(mm, addr, pte,
947 pgoff_to_pte(page->index));
951 if (pte_dirty(ptent))
952 set_page_dirty(page);
953 if (pte_young(ptent) &&
954 likely(!VM_SequentialReadHint(vma)))
955 mark_page_accessed(page);
958 page_remove_rmap(page);
959 if (unlikely(page_mapcount(page) < 0))
960 print_bad_pte(vma, addr, ptent, page);
961 tlb_remove_page(tlb, page);
965 * If details->check_mapping, we leave swap entries;
966 * if details->nonlinear_vma, we leave file entries.
968 if (unlikely(details))
970 if (pte_file(ptent)) {
971 if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))
972 print_bad_pte(vma, addr, ptent, NULL);
974 swp_entry_t entry = pte_to_swp_entry(ptent);
976 if (!non_swap_entry(entry))
978 if (unlikely(!free_swap_and_cache(entry)))
979 print_bad_pte(vma, addr, ptent, NULL);
981 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
982 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
984 add_mm_rss_vec(mm, rss);
985 arch_leave_lazy_mmu_mode();
986 pte_unmap_unlock(pte - 1, ptl);
991 static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
992 struct vm_area_struct *vma, pud_t *pud,
993 unsigned long addr, unsigned long end,
994 long *zap_work, struct zap_details *details)
999 pmd = pmd_offset(pud, addr);
1001 next = pmd_addr_end(addr, end);
1002 if (pmd_none_or_clear_bad(pmd)) {
1006 next = zap_pte_range(tlb, vma, pmd, addr, next,
1008 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
1013 static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
1014 struct vm_area_struct *vma, pgd_t *pgd,
1015 unsigned long addr, unsigned long end,
1016 long *zap_work, struct zap_details *details)
1021 pud = pud_offset(pgd, addr);
1023 next = pud_addr_end(addr, end);
1024 if (pud_none_or_clear_bad(pud)) {
1028 next = zap_pmd_range(tlb, vma, pud, addr, next,
1030 } while (pud++, addr = next, (addr != end && *zap_work > 0));
1035 static unsigned long unmap_page_range(struct mmu_gather *tlb,
1036 struct vm_area_struct *vma,
1037 unsigned long addr, unsigned long end,
1038 long *zap_work, struct zap_details *details)
1043 if (details && !details->check_mapping && !details->nonlinear_vma)
1046 BUG_ON(addr >= end);
1047 mem_cgroup_uncharge_start();
1048 tlb_start_vma(tlb, vma);
1049 pgd = pgd_offset(vma->vm_mm, addr);
1051 next = pgd_addr_end(addr, end);
1052 if (pgd_none_or_clear_bad(pgd)) {
1056 next = zap_pud_range(tlb, vma, pgd, addr, next,
1058 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
1059 tlb_end_vma(tlb, vma);
1060 mem_cgroup_uncharge_end();
1066 /* zap one pte page at a time */
1067 #define ZAP_BLOCK_SIZE (FREE_PTE_NR * PAGE_SIZE)
1069 #define ZAP_BLOCK_SIZE (253 * PAGE_SIZE)
1073 * unmap_vmas - unmap a range of memory covered by a list of vma's
1074 * @tlbp: address of the caller's struct mmu_gather
1075 * @vma: the starting vma
1076 * @start_addr: virtual address at which to start unmapping
1077 * @end_addr: virtual address at which to end unmapping
1078 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
1079 * @details: details of nonlinear truncation or shared cache invalidation
1081 * Returns the end address of the unmapping (restart addr if interrupted).
1083 * Unmap all pages in the vma list.
1085 * We aim to not hold locks for too long (for scheduling latency reasons).
1086 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
1087 * return the ending mmu_gather to the caller.
1089 * Only addresses between `start' and `end' will be unmapped.
1091 * The VMA list must be sorted in ascending virtual address order.
1093 * unmap_vmas() assumes that the caller will flush the whole unmapped address
1094 * range after unmap_vmas() returns. So the only responsibility here is to
1095 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1096 * drops the lock and schedules.
1098 unsigned long unmap_vmas(struct mmu_gather **tlbp,
1099 struct vm_area_struct *vma, unsigned long start_addr,
1100 unsigned long end_addr, unsigned long *nr_accounted,
1101 struct zap_details *details)
1103 long zap_work = ZAP_BLOCK_SIZE;
1104 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
1105 int tlb_start_valid = 0;
1106 unsigned long start = start_addr;
1107 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
1108 int fullmm = (*tlbp)->fullmm;
1109 struct mm_struct *mm = vma->vm_mm;
1111 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
1112 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
1115 start = max(vma->vm_start, start_addr);
1116 if (start >= vma->vm_end)
1118 end = min(vma->vm_end, end_addr);
1119 if (end <= vma->vm_start)
1122 if (vma->vm_flags & VM_ACCOUNT)
1123 *nr_accounted += (end - start) >> PAGE_SHIFT;
1125 if (unlikely(is_pfn_mapping(vma)))
1126 untrack_pfn_vma(vma, 0, 0);
1128 while (start != end) {
1129 if (!tlb_start_valid) {
1131 tlb_start_valid = 1;
1134 if (unlikely(is_vm_hugetlb_page(vma))) {
1136 * It is undesirable to test vma->vm_file as it
1137 * should be non-null for valid hugetlb area.
1138 * However, vm_file will be NULL in the error
1139 * cleanup path of do_mmap_pgoff. When
1140 * hugetlbfs ->mmap method fails,
1141 * do_mmap_pgoff() nullifies vma->vm_file
1142 * before calling this function to clean up.
1143 * Since no pte has actually been setup, it is
1144 * safe to do nothing in this case.
1147 unmap_hugepage_range(vma, start, end, NULL);
1148 zap_work -= (end - start) /
1149 pages_per_huge_page(hstate_vma(vma));
1154 start = unmap_page_range(*tlbp, vma,
1155 start, end, &zap_work, details);
1158 BUG_ON(start != end);
1162 tlb_finish_mmu(*tlbp, tlb_start, start);
1164 if (need_resched() ||
1165 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
1173 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
1174 tlb_start_valid = 0;
1175 zap_work = ZAP_BLOCK_SIZE;
1179 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
1180 return start; /* which is now the end (or restart) address */
1184 * zap_page_range - remove user pages in a given range
1185 * @vma: vm_area_struct holding the applicable pages
1186 * @address: starting address of pages to zap
1187 * @size: number of bytes to zap
1188 * @details: details of nonlinear truncation or shared cache invalidation
1190 unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
1191 unsigned long size, struct zap_details *details)
1193 struct mm_struct *mm = vma->vm_mm;
1194 struct mmu_gather *tlb;
1195 unsigned long end = address + size;
1196 unsigned long nr_accounted = 0;
1199 tlb = tlb_gather_mmu(mm, 0);
1200 update_hiwater_rss(mm);
1201 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
1203 tlb_finish_mmu(tlb, address, end);
1208 * zap_vma_ptes - remove ptes mapping the vma
1209 * @vma: vm_area_struct holding ptes to be zapped
1210 * @address: starting address of pages to zap
1211 * @size: number of bytes to zap
1213 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1215 * The entire address range must be fully contained within the vma.
1217 * Returns 0 if successful.
1219 int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1222 if (address < vma->vm_start || address + size > vma->vm_end ||
1223 !(vma->vm_flags & VM_PFNMAP))
1225 zap_page_range(vma, address, size, NULL);
1228 EXPORT_SYMBOL_GPL(zap_vma_ptes);
1231 * follow_page - look up a page descriptor from a user-virtual address
1232 * @vma: vm_area_struct mapping @address
1233 * @address: virtual address to look up
1234 * @flags: flags modifying lookup behaviour
1236 * @flags can have FOLL_ flags set, defined in <linux/mm.h>
1238 * Returns the mapped (struct page *), %NULL if no mapping exists, or
1239 * an error pointer if there is a mapping to something not represented
1240 * by a page descriptor (see also vm_normal_page()).
1242 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
1251 struct mm_struct *mm = vma->vm_mm;
1253 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
1254 if (!IS_ERR(page)) {
1255 BUG_ON(flags & FOLL_GET);
1260 pgd = pgd_offset(mm, address);
1261 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
1264 pud = pud_offset(pgd, address);
1267 if (pud_huge(*pud)) {
1268 BUG_ON(flags & FOLL_GET);
1269 page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
1272 if (unlikely(pud_bad(*pud)))
1275 pmd = pmd_offset(pud, address);
1278 if (pmd_huge(*pmd)) {
1279 BUG_ON(flags & FOLL_GET);
1280 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
1283 if (unlikely(pmd_bad(*pmd)))
1286 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
1289 if (!pte_present(pte))
1291 if ((flags & FOLL_WRITE) && !pte_write(pte))
1294 page = vm_normal_page(vma, address, pte);
1295 if (unlikely(!page)) {
1296 if ((flags & FOLL_DUMP) ||
1297 !is_zero_pfn(pte_pfn(pte)))
1299 page = pte_page(pte);
1302 if (flags & FOLL_GET)
1304 if (flags & FOLL_TOUCH) {
1305 if ((flags & FOLL_WRITE) &&
1306 !pte_dirty(pte) && !PageDirty(page))
1307 set_page_dirty(page);
1309 * pte_mkyoung() would be more correct here, but atomic care
1310 * is needed to avoid losing the dirty bit: it is easier to use
1311 * mark_page_accessed().
1313 mark_page_accessed(page);
1316 pte_unmap_unlock(ptep, ptl);
1321 pte_unmap_unlock(ptep, ptl);
1322 return ERR_PTR(-EFAULT);
1325 pte_unmap_unlock(ptep, ptl);
1331 * When core dumping an enormous anonymous area that nobody
1332 * has touched so far, we don't want to allocate unnecessary pages or
1333 * page tables. Return error instead of NULL to skip handle_mm_fault,
1334 * then get_dump_page() will return NULL to leave a hole in the dump.
1335 * But we can only make this optimization where a hole would surely
1336 * be zero-filled if handle_mm_fault() actually did handle it.
1338 if ((flags & FOLL_DUMP) &&
1339 (!vma->vm_ops || !vma->vm_ops->fault))
1340 return ERR_PTR(-EFAULT);
1344 int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1345 unsigned long start, int nr_pages, unsigned int gup_flags,
1346 struct page **pages, struct vm_area_struct **vmas)
1349 unsigned long vm_flags;
1354 VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET));
1357 * Require read or write permissions.
1358 * If FOLL_FORCE is set, we only require the "MAY" flags.
1360 vm_flags = (gup_flags & FOLL_WRITE) ?
1361 (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1362 vm_flags &= (gup_flags & FOLL_FORCE) ?
1363 (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
1367 struct vm_area_struct *vma;
1369 vma = find_extend_vma(mm, start);
1370 if (!vma && in_gate_area(tsk, start)) {
1371 unsigned long pg = start & PAGE_MASK;
1372 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1378 /* user gate pages are read-only */
1379 if (gup_flags & FOLL_WRITE)
1380 return i ? : -EFAULT;
1382 pgd = pgd_offset_k(pg);
1384 pgd = pgd_offset_gate(mm, pg);
1385 BUG_ON(pgd_none(*pgd));
1386 pud = pud_offset(pgd, pg);
1387 BUG_ON(pud_none(*pud));
1388 pmd = pmd_offset(pud, pg);
1390 return i ? : -EFAULT;
1391 pte = pte_offset_map(pmd, pg);
1392 if (pte_none(*pte)) {
1394 return i ? : -EFAULT;
1397 struct page *page = vm_normal_page(gate_vma, start, *pte);
1412 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
1413 !(vm_flags & vma->vm_flags))
1414 return i ? : -EFAULT;
1416 if (is_vm_hugetlb_page(vma)) {
1417 i = follow_hugetlb_page(mm, vma, pages, vmas,
1418 &start, &nr_pages, i, gup_flags);
1424 unsigned int foll_flags = gup_flags;
1427 * If we have a pending SIGKILL, don't keep faulting
1428 * pages and potentially allocating memory.
1430 if (unlikely(fatal_signal_pending(current)))
1431 return i ? i : -ERESTARTSYS;
1434 while (!(page = follow_page(vma, start, foll_flags))) {
1437 ret = handle_mm_fault(mm, vma, start,
1438 (foll_flags & FOLL_WRITE) ?
1439 FAULT_FLAG_WRITE : 0);
1441 if (ret & VM_FAULT_ERROR) {
1442 if (ret & VM_FAULT_OOM)
1443 return i ? i : -ENOMEM;
1445 (VM_FAULT_HWPOISON|VM_FAULT_SIGBUS))
1446 return i ? i : -EFAULT;
1449 if (ret & VM_FAULT_MAJOR)
1455 * The VM_FAULT_WRITE bit tells us that
1456 * do_wp_page has broken COW when necessary,
1457 * even if maybe_mkwrite decided not to set
1458 * pte_write. We can thus safely do subsequent
1459 * page lookups as if they were reads. But only
1460 * do so when looping for pte_write is futile:
1461 * in some cases userspace may also be wanting
1462 * to write to the gotten user page, which a
1463 * read fault here might prevent (a readonly
1464 * page might get reCOWed by userspace write).
1466 if ((ret & VM_FAULT_WRITE) &&
1467 !(vma->vm_flags & VM_WRITE))
1468 foll_flags &= ~FOLL_WRITE;
1473 return i ? i : PTR_ERR(page);
1477 flush_anon_page(vma, page, start);
1478 flush_dcache_page(page);
1485 } while (nr_pages && start < vma->vm_end);
1491 * get_user_pages() - pin user pages in memory
1492 * @tsk: task_struct of target task
1493 * @mm: mm_struct of target mm
1494 * @start: starting user address
1495 * @nr_pages: number of pages from start to pin
1496 * @write: whether pages will be written to by the caller
1497 * @force: whether to force write access even if user mapping is
1498 * readonly. This will result in the page being COWed even
1499 * in MAP_SHARED mappings. You do not want this.
1500 * @pages: array that receives pointers to the pages pinned.
1501 * Should be at least nr_pages long. Or NULL, if caller
1502 * only intends to ensure the pages are faulted in.
1503 * @vmas: array of pointers to vmas corresponding to each page.
1504 * Or NULL if the caller does not require them.
1506 * Returns number of pages pinned. This may be fewer than the number
1507 * requested. If nr_pages is 0 or negative, returns 0. If no pages
1508 * were pinned, returns -errno. Each page returned must be released
1509 * with a put_page() call when it is finished with. vmas will only
1510 * remain valid while mmap_sem is held.
1512 * Must be called with mmap_sem held for read or write.
1514 * get_user_pages walks a process's page tables and takes a reference to
1515 * each struct page that each user address corresponds to at a given
1516 * instant. That is, it takes the page that would be accessed if a user
1517 * thread accesses the given user virtual address at that instant.
1519 * This does not guarantee that the page exists in the user mappings when
1520 * get_user_pages returns, and there may even be a completely different
1521 * page there in some cases (eg. if mmapped pagecache has been invalidated
1522 * and subsequently re faulted). However it does guarantee that the page
1523 * won't be freed completely. And mostly callers simply care that the page
1524 * contains data that was valid *at some point in time*. Typically, an IO
1525 * or similar operation cannot guarantee anything stronger anyway because
1526 * locks can't be held over the syscall boundary.
1528 * If write=0, the page must not be written to. If the page is written to,
1529 * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
1530 * after the page is finished with, and before put_page is called.
1532 * get_user_pages is typically used for fewer-copy IO operations, to get a
1533 * handle on the memory by some means other than accesses via the user virtual
1534 * addresses. The pages may be submitted for DMA to devices or accessed via
1535 * their kernel linear mapping (via the kmap APIs). Care should be taken to
1536 * use the correct cache flushing APIs.
1538 * See also get_user_pages_fast, for performance critical applications.
1540 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
1541 unsigned long start, int nr_pages, int write, int force,
1542 struct page **pages, struct vm_area_struct **vmas)
1544 int flags = FOLL_TOUCH;
1549 flags |= FOLL_WRITE;
1551 flags |= FOLL_FORCE;
1553 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
1555 EXPORT_SYMBOL(get_user_pages);
1558 * get_dump_page() - pin user page in memory while writing it to core dump
1559 * @addr: user address
1561 * Returns struct page pointer of user page pinned for dump,
1562 * to be freed afterwards by page_cache_release() or put_page().
1564 * Returns NULL on any kind of failure - a hole must then be inserted into
1565 * the corefile, to preserve alignment with its headers; and also returns
1566 * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
1567 * allowing a hole to be left in the corefile to save diskspace.
1569 * Called without mmap_sem, but after all other threads have been killed.
1571 #ifdef CONFIG_ELF_CORE
1572 struct page *get_dump_page(unsigned long addr)
1574 struct vm_area_struct *vma;
1577 if (__get_user_pages(current, current->mm, addr, 1,
1578 FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma) < 1)
1580 flush_cache_page(vma, addr, page_to_pfn(page));
1583 #endif /* CONFIG_ELF_CORE */
1585 static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1586 unsigned long addr, unsigned long end, pgprot_t prot)
1592 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1595 arch_enter_lazy_mmu_mode();
1599 if (unlikely(!pte_none(*pte))) {
1604 zero_pte = pte_mkspecial(pfn_pte(my_zero_pfn(addr), prot));
1605 zero_pte = pte_wrprotect(zero_pte);
1606 set_pte_at(mm, addr, pte, zero_pte);
1607 } while (pte++, addr += PAGE_SIZE, addr != end);
1608 arch_leave_lazy_mmu_mode();
1609 pte_unmap_unlock(pte - 1, ptl);
1613 static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1614 unsigned long addr, unsigned long end, pgprot_t prot)
1620 pmd = pmd_alloc(mm, pud, addr);
1624 next = pmd_addr_end(addr, end);
1625 err = zeromap_pte_range(mm, pmd, addr, next, prot);
1628 } while (pmd++, addr = next, addr != end);
1632 static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1633 unsigned long addr, unsigned long end, pgprot_t prot)
1639 pud = pud_alloc(mm, pgd, addr);
1643 next = pud_addr_end(addr, end);
1644 err = zeromap_pmd_range(mm, pud, addr, next, prot);
1647 } while (pud++, addr = next, addr != end);
1651 int zeromap_page_range(struct vm_area_struct *vma,
1652 unsigned long addr, unsigned long size, pgprot_t prot)
1656 unsigned long end = addr + size;
1657 struct mm_struct *mm = vma->vm_mm;
1660 BUG_ON(addr >= end);
1661 pgd = pgd_offset(mm, addr);
1662 flush_cache_range(vma, addr, end);
1664 next = pgd_addr_end(addr, end);
1665 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1668 } while (pgd++, addr = next, addr != end);
1672 pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1675 pgd_t * pgd = pgd_offset(mm, addr);
1676 pud_t * pud = pud_alloc(mm, pgd, addr);
1678 pmd_t * pmd = pmd_alloc(mm, pud, addr);
1680 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1686 * This is the old fallback for page remapping.
1688 * For historical reasons, it only allows reserved pages. Only
1689 * old drivers should use this, and they needed to mark their
1690 * pages reserved for the old functions anyway.
1692 static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1693 struct page *page, pgprot_t prot)
1695 struct mm_struct *mm = vma->vm_mm;
1704 flush_dcache_page(page);
1705 pte = get_locked_pte(mm, addr, &ptl);
1709 if (!pte_none(*pte))
1712 /* Ok, finally just insert the thing.. */
1714 inc_mm_counter_fast(mm, MM_FILEPAGES);
1715 page_add_file_rmap(page);
1716 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1719 pte_unmap_unlock(pte, ptl);
1722 pte_unmap_unlock(pte, ptl);
1728 * vm_insert_page - insert single page into user vma
1729 * @vma: user vma to map to
1730 * @addr: target user address of this page
1731 * @page: source kernel page
1733 * This allows drivers to insert individual pages they've allocated
1736 * The page has to be a nice clean _individual_ kernel allocation.
1737 * If you allocate a compound page, you need to have marked it as
1738 * such (__GFP_COMP), or manually just split the page up yourself
1739 * (see split_page()).
1741 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1742 * took an arbitrary page protection parameter. This doesn't allow
1743 * that. Your vma protection will have to be set up correctly, which
1744 * means that if you want a shared writable mapping, you'd better
1745 * ask for a shared writable mapping!
1747 * The page does not need to be reserved.
1749 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1752 if (addr < vma->vm_start || addr >= vma->vm_end)
1754 if (!page_count(page))
1756 vma->vm_flags |= VM_INSERTPAGE;
1757 return insert_page(vma, addr, page, vma->vm_page_prot);
1759 EXPORT_SYMBOL(vm_insert_page);
1761 static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1762 unsigned long pfn, pgprot_t prot)
1764 struct mm_struct *mm = vma->vm_mm;
1770 pte = get_locked_pte(mm, addr, &ptl);
1774 if (!pte_none(*pte))
1777 /* Ok, finally just insert the thing.. */
1778 entry = pte_mkspecial(pfn_pte(pfn, prot));
1779 set_pte_at(mm, addr, pte, entry);
1780 update_mmu_cache(vma, addr, pte); /* XXX: why not for insert_page? */
1784 pte_unmap_unlock(pte, ptl);
1790 * vm_insert_pfn - insert single pfn into user vma
1791 * @vma: user vma to map to
1792 * @addr: target user address of this page
1793 * @pfn: source kernel pfn
1795 * Similar to vm_inert_page, this allows drivers to insert individual pages
1796 * they've allocated into a user vma. Same comments apply.
1798 * This function should only be called from a vm_ops->fault handler, and
1799 * in that case the handler should return NULL.
1801 * vma cannot be a COW mapping.
1803 * As this is called only for pages that do not currently exist, we
1804 * do not need to flush old virtual caches or the TLB.
1806 int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1810 pgprot_t pgprot = vma->vm_page_prot;
1812 * Technically, architectures with pte_special can avoid all these
1813 * restrictions (same for remap_pfn_range). However we would like
1814 * consistency in testing and feature parity among all, so we should
1815 * try to keep these invariants in place for everybody.
1817 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1818 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1819 (VM_PFNMAP|VM_MIXEDMAP));
1820 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1821 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
1823 if (addr < vma->vm_start || addr >= vma->vm_end)
1825 if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE))
1828 ret = insert_pfn(vma, addr, pfn, pgprot);
1831 untrack_pfn_vma(vma, pfn, PAGE_SIZE);
1835 EXPORT_SYMBOL(vm_insert_pfn);
1837 int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1840 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
1842 if (addr < vma->vm_start || addr >= vma->vm_end)
1846 * If we don't have pte special, then we have to use the pfn_valid()
1847 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1848 * refcount the page if pfn_valid is true (hence insert_page rather
1849 * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
1850 * without pte special, it would there be refcounted as a normal page.
1852 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1855 page = pfn_to_page(pfn);
1856 return insert_page(vma, addr, page, vma->vm_page_prot);
1858 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
1860 EXPORT_SYMBOL(vm_insert_mixed);
1863 * maps a range of physical memory into the requested pages. the old
1864 * mappings are removed. any references to nonexistent pages results
1865 * in null mappings (currently treated as "copy-on-access")
1867 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1868 unsigned long addr, unsigned long end,
1869 unsigned long pfn, pgprot_t prot)
1874 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1877 arch_enter_lazy_mmu_mode();
1879 BUG_ON(!pte_none(*pte));
1880 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
1882 } while (pte++, addr += PAGE_SIZE, addr != end);
1883 arch_leave_lazy_mmu_mode();
1884 pte_unmap_unlock(pte - 1, ptl);
1888 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1889 unsigned long addr, unsigned long end,
1890 unsigned long pfn, pgprot_t prot)
1895 pfn -= addr >> PAGE_SHIFT;
1896 pmd = pmd_alloc(mm, pud, addr);
1900 next = pmd_addr_end(addr, end);
1901 if (remap_pte_range(mm, pmd, addr, next,
1902 pfn + (addr >> PAGE_SHIFT), prot))
1904 } while (pmd++, addr = next, addr != end);
1908 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1909 unsigned long addr, unsigned long end,
1910 unsigned long pfn, pgprot_t prot)
1915 pfn -= addr >> PAGE_SHIFT;
1916 pud = pud_alloc(mm, pgd, addr);
1920 next = pud_addr_end(addr, end);
1921 if (remap_pmd_range(mm, pud, addr, next,
1922 pfn + (addr >> PAGE_SHIFT), prot))
1924 } while (pud++, addr = next, addr != end);
1929 * remap_pfn_range - remap kernel memory to userspace
1930 * @vma: user vma to map to
1931 * @addr: target user address to start at
1932 * @pfn: physical address of kernel memory
1933 * @size: size of map area
1934 * @prot: page protection flags for this mapping
1936 * Note: this is only safe if the mm semaphore is held when called.
1938 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1939 unsigned long pfn, unsigned long size, pgprot_t prot)
1943 unsigned long end = addr + PAGE_ALIGN(size);
1944 struct mm_struct *mm = vma->vm_mm;
1948 * Physically remapped pages are special. Tell the
1949 * rest of the world about it:
1950 * VM_IO tells people not to look at these pages
1951 * (accesses can have side effects).
1952 * VM_RESERVED is specified all over the place, because
1953 * in 2.4 it kept swapout's vma scan off this vma; but
1954 * in 2.6 the LRU scan won't even find its pages, so this
1955 * flag means no more than count its pages in reserved_vm,
1956 * and omit it from core dump, even when VM_IO turned off.
1957 * VM_PFNMAP tells the core MM that the base pages are just
1958 * raw PFN mappings, and do not have a "struct page" associated
1961 * There's a horrible special case to handle copy-on-write
1962 * behaviour that some programs depend on. We mark the "original"
1963 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1965 if (addr == vma->vm_start && end == vma->vm_end) {
1966 vma->vm_pgoff = pfn;
1967 vma->vm_flags |= VM_PFN_AT_MMAP;
1968 } else if (is_cow_mapping(vma->vm_flags))
1971 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
1973 err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size));
1976 * To indicate that track_pfn related cleanup is not
1977 * needed from higher level routine calling unmap_vmas
1979 vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP);
1980 vma->vm_flags &= ~VM_PFN_AT_MMAP;
1984 BUG_ON(addr >= end);
1985 pfn -= addr >> PAGE_SHIFT;
1986 pgd = pgd_offset(mm, addr);
1987 flush_cache_range(vma, addr, end);
1989 next = pgd_addr_end(addr, end);
1990 err = remap_pud_range(mm, pgd, addr, next,
1991 pfn + (addr >> PAGE_SHIFT), prot);
1994 } while (pgd++, addr = next, addr != end);
1997 untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size));
2001 EXPORT_SYMBOL(remap_pfn_range);
2003 static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
2004 unsigned long addr, unsigned long end,
2005 pte_fn_t fn, void *data)
2010 spinlock_t *uninitialized_var(ptl);
2012 pte = (mm == &init_mm) ?
2013 pte_alloc_kernel(pmd, addr) :
2014 pte_alloc_map_lock(mm, pmd, addr, &ptl);
2018 BUG_ON(pmd_huge(*pmd));
2020 arch_enter_lazy_mmu_mode();
2022 token = pmd_pgtable(*pmd);
2025 err = fn(pte++, token, addr, data);
2028 } while (addr += PAGE_SIZE, addr != end);
2030 arch_leave_lazy_mmu_mode();
2033 pte_unmap_unlock(pte-1, ptl);
2037 static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
2038 unsigned long addr, unsigned long end,
2039 pte_fn_t fn, void *data)
2045 BUG_ON(pud_huge(*pud));
2047 pmd = pmd_alloc(mm, pud, addr);
2051 next = pmd_addr_end(addr, end);
2052 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
2055 } while (pmd++, addr = next, addr != end);
2059 static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
2060 unsigned long addr, unsigned long end,
2061 pte_fn_t fn, void *data)
2067 pud = pud_alloc(mm, pgd, addr);
2071 next = pud_addr_end(addr, end);
2072 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
2075 } while (pud++, addr = next, addr != end);
2080 * Scan a region of virtual memory, filling in page tables as necessary
2081 * and calling a provided function on each leaf page table.
2083 int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
2084 unsigned long size, pte_fn_t fn, void *data)
2088 unsigned long start = addr, end = addr + size;
2091 BUG_ON(addr >= end);
2092 mmu_notifier_invalidate_range_start(mm, start, end);
2093 pgd = pgd_offset(mm, addr);
2095 next = pgd_addr_end(addr, end);
2096 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
2099 } while (pgd++, addr = next, addr != end);
2100 mmu_notifier_invalidate_range_end(mm, start, end);
2103 EXPORT_SYMBOL_GPL(apply_to_page_range);
2106 * handle_pte_fault chooses page fault handler according to an entry
2107 * which was read non-atomically. Before making any commitment, on
2108 * those architectures or configurations (e.g. i386 with PAE) which
2109 * might give a mix of unmatched parts, do_swap_page and do_file_page
2110 * must check under lock before unmapping the pte and proceeding
2111 * (but do_wp_page is only called after already making such a check;
2112 * and do_anonymous_page and do_no_page can safely check later on).
2114 static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
2115 pte_t *page_table, pte_t orig_pte)
2118 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
2119 if (sizeof(pte_t) > sizeof(unsigned long)) {
2120 spinlock_t *ptl = pte_lockptr(mm, pmd);
2122 same = pte_same(*page_table, orig_pte);
2126 pte_unmap(page_table);
2131 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
2132 * servicing faults for write access. In the normal case, do always want
2133 * pte_mkwrite. But get_user_pages can cause write faults for mappings
2134 * that do not have writing enabled, when used by access_process_vm.
2136 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
2138 if (likely(vma->vm_flags & VM_WRITE))
2139 pte = pte_mkwrite(pte);
2143 static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
2146 * If the source page was a PFN mapping, we don't have
2147 * a "struct page" for it. We do a best-effort copy by
2148 * just copying from the original user address. If that
2149 * fails, we just zero-fill it. Live with it.
2151 if (unlikely(!src)) {
2152 void *kaddr = kmap_atomic(dst, KM_USER0);
2153 void __user *uaddr = (void __user *)(va & PAGE_MASK);
2156 * This really shouldn't fail, because the page is there
2157 * in the page tables. But it might just be unreadable,
2158 * in which case we just give up and fill the result with
2161 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
2162 memset(kaddr, 0, PAGE_SIZE);
2163 kunmap_atomic(kaddr, KM_USER0);
2164 flush_dcache_page(dst);
2166 copy_user_highpage(dst, src, va, vma);
2170 * This routine handles present pages, when users try to write
2171 * to a shared page. It is done by copying the page to a new address
2172 * and decrementing the shared-page counter for the old page.
2174 * Note that this routine assumes that the protection checks have been
2175 * done by the caller (the low-level page fault routine in most cases).
2176 * Thus we can safely just mark it writable once we've done any necessary
2179 * We also mark the page dirty at this point even though the page will
2180 * change only once the write actually happens. This avoids a few races,
2181 * and potentially makes it more efficient.
2183 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2184 * but allow concurrent faults), with pte both mapped and locked.
2185 * We return with mmap_sem still held, but pte unmapped and unlocked.
2187 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
2188 unsigned long address, pte_t *page_table, pmd_t *pmd,
2189 spinlock_t *ptl, pte_t orig_pte)
2191 struct page *old_page, *new_page;
2193 int reuse = 0, ret = 0;
2194 int page_mkwrite = 0;
2195 struct page *dirty_page = NULL;
2197 old_page = vm_normal_page(vma, address, orig_pte);
2200 * VM_MIXEDMAP !pfn_valid() case
2202 * We should not cow pages in a shared writeable mapping.
2203 * Just mark the pages writable as we can't do any dirty
2204 * accounting on raw pfn maps.
2206 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2207 (VM_WRITE|VM_SHARED))
2213 * Take out anonymous pages first, anonymous shared vmas are
2214 * not dirty accountable.
2216 if (PageAnon(old_page) && !PageKsm(old_page)) {
2217 if (!trylock_page(old_page)) {
2218 page_cache_get(old_page);
2219 pte_unmap_unlock(page_table, ptl);
2220 lock_page(old_page);
2221 page_table = pte_offset_map_lock(mm, pmd, address,
2223 if (!pte_same(*page_table, orig_pte)) {
2224 unlock_page(old_page);
2225 page_cache_release(old_page);
2228 page_cache_release(old_page);
2230 reuse = reuse_swap_page(old_page);
2233 * The page is all ours. Move it to our anon_vma so
2234 * the rmap code will not search our parent or siblings.
2235 * Protected against the rmap code by the page lock.
2237 page_move_anon_rmap(old_page, vma, address);
2238 unlock_page(old_page);
2239 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
2240 (VM_WRITE|VM_SHARED))) {
2242 * Only catch write-faults on shared writable pages,
2243 * read-only shared pages can get COWed by
2244 * get_user_pages(.write=1, .force=1).
2246 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
2247 struct vm_fault vmf;
2250 vmf.virtual_address = (void __user *)(address &
2252 vmf.pgoff = old_page->index;
2253 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2254 vmf.page = old_page;
2257 * Notify the address space that the page is about to
2258 * become writable so that it can prohibit this or wait
2259 * for the page to get into an appropriate state.
2261 * We do this without the lock held, so that it can
2262 * sleep if it needs to.
2264 page_cache_get(old_page);
2265 pte_unmap_unlock(page_table, ptl);
2267 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2269 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2271 goto unwritable_page;
2273 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2274 lock_page(old_page);
2275 if (!old_page->mapping) {
2276 ret = 0; /* retry the fault */
2277 unlock_page(old_page);
2278 goto unwritable_page;
2281 VM_BUG_ON(!PageLocked(old_page));
2284 * Since we dropped the lock we need to revalidate
2285 * the PTE as someone else may have changed it. If
2286 * they did, we just return, as we can count on the
2287 * MMU to tell us if they didn't also make it writable.
2289 page_table = pte_offset_map_lock(mm, pmd, address,
2291 if (!pte_same(*page_table, orig_pte)) {
2292 unlock_page(old_page);
2293 page_cache_release(old_page);
2299 dirty_page = old_page;
2300 get_page(dirty_page);
2306 flush_cache_page(vma, address, pte_pfn(orig_pte));
2307 entry = pte_mkyoung(orig_pte);
2308 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2309 if (ptep_set_access_flags(vma, address, page_table, entry,1))
2310 update_mmu_cache(vma, address, page_table);
2311 ret |= VM_FAULT_WRITE;
2316 * Ok, we need to copy. Oh, well..
2318 page_cache_get(old_page);
2320 pte_unmap_unlock(page_table, ptl);
2322 if (unlikely(anon_vma_prepare(vma)))
2325 if (is_zero_pfn(pte_pfn(orig_pte))) {
2326 new_page = alloc_zeroed_user_highpage_movable(vma, address);
2330 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2333 cow_user_page(new_page, old_page, address, vma);
2335 __SetPageUptodate(new_page);
2338 * Don't let another task, with possibly unlocked vma,
2339 * keep the mlocked page.
2341 if ((vma->vm_flags & VM_LOCKED) && old_page) {
2342 lock_page(old_page); /* for LRU manipulation */
2343 clear_page_mlock(old_page);
2344 unlock_page(old_page);
2347 if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
2351 * Re-check the pte - we dropped the lock
2353 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2354 if (likely(pte_same(*page_table, orig_pte))) {
2356 if (!PageAnon(old_page)) {
2357 dec_mm_counter_fast(mm, MM_FILEPAGES);
2358 inc_mm_counter_fast(mm, MM_ANONPAGES);
2361 inc_mm_counter_fast(mm, MM_ANONPAGES);
2362 flush_cache_page(vma, address, pte_pfn(orig_pte));
2363 entry = mk_pte(new_page, vma->vm_page_prot);
2364 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2366 * Clear the pte entry and flush it first, before updating the
2367 * pte with the new entry. This will avoid a race condition
2368 * seen in the presence of one thread doing SMC and another
2371 ptep_clear_flush(vma, address, page_table);
2372 page_add_new_anon_rmap(new_page, vma, address);
2374 * We call the notify macro here because, when using secondary
2375 * mmu page tables (such as kvm shadow page tables), we want the
2376 * new page to be mapped directly into the secondary page table.
2378 set_pte_at_notify(mm, address, page_table, entry);
2379 update_mmu_cache(vma, address, page_table);
2382 * Only after switching the pte to the new page may
2383 * we remove the mapcount here. Otherwise another
2384 * process may come and find the rmap count decremented
2385 * before the pte is switched to the new page, and
2386 * "reuse" the old page writing into it while our pte
2387 * here still points into it and can be read by other
2390 * The critical issue is to order this
2391 * page_remove_rmap with the ptp_clear_flush above.
2392 * Those stores are ordered by (if nothing else,)
2393 * the barrier present in the atomic_add_negative
2394 * in page_remove_rmap.
2396 * Then the TLB flush in ptep_clear_flush ensures that
2397 * no process can access the old page before the
2398 * decremented mapcount is visible. And the old page
2399 * cannot be reused until after the decremented
2400 * mapcount is visible. So transitively, TLBs to
2401 * old page will be flushed before it can be reused.
2403 page_remove_rmap(old_page);
2406 /* Free the old page.. */
2407 new_page = old_page;
2408 ret |= VM_FAULT_WRITE;
2410 mem_cgroup_uncharge_page(new_page);
2413 page_cache_release(new_page);
2415 page_cache_release(old_page);
2417 pte_unmap_unlock(page_table, ptl);
2420 * Yes, Virginia, this is actually required to prevent a race
2421 * with clear_page_dirty_for_io() from clearing the page dirty
2422 * bit after it clear all dirty ptes, but before a racing
2423 * do_wp_page installs a dirty pte.
2425 * do_no_page is protected similarly.
2427 if (!page_mkwrite) {
2428 wait_on_page_locked(dirty_page);
2429 set_page_dirty_balance(dirty_page, page_mkwrite);
2431 put_page(dirty_page);
2433 struct address_space *mapping = dirty_page->mapping;
2435 set_page_dirty(dirty_page);
2436 unlock_page(dirty_page);
2437 page_cache_release(dirty_page);
2440 * Some device drivers do not set page.mapping
2441 * but still dirty their pages
2443 balance_dirty_pages_ratelimited(mapping);
2447 /* file_update_time outside page_lock */
2449 file_update_time(vma->vm_file);
2453 page_cache_release(new_page);
2457 unlock_page(old_page);
2458 page_cache_release(old_page);
2460 page_cache_release(old_page);
2462 return VM_FAULT_OOM;
2465 page_cache_release(old_page);
2470 * Helper functions for unmap_mapping_range().
2472 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
2474 * We have to restart searching the prio_tree whenever we drop the lock,
2475 * since the iterator is only valid while the lock is held, and anyway
2476 * a later vma might be split and reinserted earlier while lock dropped.
2478 * The list of nonlinear vmas could be handled more efficiently, using
2479 * a placeholder, but handle it in the same way until a need is shown.
2480 * It is important to search the prio_tree before nonlinear list: a vma
2481 * may become nonlinear and be shifted from prio_tree to nonlinear list
2482 * while the lock is dropped; but never shifted from list to prio_tree.
2484 * In order to make forward progress despite restarting the search,
2485 * vm_truncate_count is used to mark a vma as now dealt with, so we can
2486 * quickly skip it next time around. Since the prio_tree search only
2487 * shows us those vmas affected by unmapping the range in question, we
2488 * can't efficiently keep all vmas in step with mapping->truncate_count:
2489 * so instead reset them all whenever it wraps back to 0 (then go to 1).
2490 * mapping->truncate_count and vma->vm_truncate_count are protected by
2493 * In order to make forward progress despite repeatedly restarting some
2494 * large vma, note the restart_addr from unmap_vmas when it breaks out:
2495 * and restart from that address when we reach that vma again. It might
2496 * have been split or merged, shrunk or extended, but never shifted: so
2497 * restart_addr remains valid so long as it remains in the vma's range.
2498 * unmap_mapping_range forces truncate_count to leap over page-aligned
2499 * values so we can save vma's restart_addr in its truncate_count field.
2501 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
2503 static void reset_vma_truncate_counts(struct address_space *mapping)
2505 struct vm_area_struct *vma;
2506 struct prio_tree_iter iter;
2508 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
2509 vma->vm_truncate_count = 0;
2510 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
2511 vma->vm_truncate_count = 0;
2514 static int unmap_mapping_range_vma(struct vm_area_struct *vma,
2515 unsigned long start_addr, unsigned long end_addr,
2516 struct zap_details *details)
2518 unsigned long restart_addr;
2522 * files that support invalidating or truncating portions of the
2523 * file from under mmaped areas must have their ->fault function
2524 * return a locked page (and set VM_FAULT_LOCKED in the return).
2525 * This provides synchronisation against concurrent unmapping here.
2529 restart_addr = vma->vm_truncate_count;
2530 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
2531 start_addr = restart_addr;
2532 if (start_addr >= end_addr) {
2533 /* Top of vma has been split off since last time */
2534 vma->vm_truncate_count = details->truncate_count;
2539 restart_addr = zap_page_range(vma, start_addr,
2540 end_addr - start_addr, details);
2541 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
2543 if (restart_addr >= end_addr) {
2544 /* We have now completed this vma: mark it so */
2545 vma->vm_truncate_count = details->truncate_count;
2549 /* Note restart_addr in vma's truncate_count field */
2550 vma->vm_truncate_count = restart_addr;
2555 spin_unlock(details->i_mmap_lock);
2557 spin_lock(details->i_mmap_lock);
2561 static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
2562 struct zap_details *details)
2564 struct vm_area_struct *vma;
2565 struct prio_tree_iter iter;
2566 pgoff_t vba, vea, zba, zea;
2569 vma_prio_tree_foreach(vma, &iter, root,
2570 details->first_index, details->last_index) {
2571 /* Skip quickly over those we have already dealt with */
2572 if (vma->vm_truncate_count == details->truncate_count)
2575 vba = vma->vm_pgoff;
2576 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
2577 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2578 zba = details->first_index;
2581 zea = details->last_index;
2585 if (unmap_mapping_range_vma(vma,
2586 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2587 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2593 static inline void unmap_mapping_range_list(struct list_head *head,
2594 struct zap_details *details)
2596 struct vm_area_struct *vma;
2599 * In nonlinear VMAs there is no correspondence between virtual address
2600 * offset and file offset. So we must perform an exhaustive search
2601 * across *all* the pages in each nonlinear VMA, not just the pages
2602 * whose virtual address lies outside the file truncation point.
2605 list_for_each_entry(vma, head, shared.vm_set.list) {
2606 /* Skip quickly over those we have already dealt with */
2607 if (vma->vm_truncate_count == details->truncate_count)
2609 details->nonlinear_vma = vma;
2610 if (unmap_mapping_range_vma(vma, vma->vm_start,
2611 vma->vm_end, details) < 0)
2617 * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
2618 * @mapping: the address space containing mmaps to be unmapped.
2619 * @holebegin: byte in first page to unmap, relative to the start of
2620 * the underlying file. This will be rounded down to a PAGE_SIZE
2621 * boundary. Note that this is different from truncate_pagecache(), which
2622 * must keep the partial page. In contrast, we must get rid of
2624 * @holelen: size of prospective hole in bytes. This will be rounded
2625 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2627 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2628 * but 0 when invalidating pagecache, don't throw away private data.
2630 void unmap_mapping_range(struct address_space *mapping,
2631 loff_t const holebegin, loff_t const holelen, int even_cows)
2633 struct zap_details details;
2634 pgoff_t hba = holebegin >> PAGE_SHIFT;
2635 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2637 /* Check for overflow. */
2638 if (sizeof(holelen) > sizeof(hlen)) {
2640 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2641 if (holeend & ~(long long)ULONG_MAX)
2642 hlen = ULONG_MAX - hba + 1;
2645 details.check_mapping = even_cows? NULL: mapping;
2646 details.nonlinear_vma = NULL;
2647 details.first_index = hba;
2648 details.last_index = hba + hlen - 1;
2649 if (details.last_index < details.first_index)
2650 details.last_index = ULONG_MAX;
2651 details.i_mmap_lock = &mapping->i_mmap_lock;
2653 spin_lock(&mapping->i_mmap_lock);
2655 /* Protect against endless unmapping loops */
2656 mapping->truncate_count++;
2657 if (unlikely(is_restart_addr(mapping->truncate_count))) {
2658 if (mapping->truncate_count == 0)
2659 reset_vma_truncate_counts(mapping);
2660 mapping->truncate_count++;
2662 details.truncate_count = mapping->truncate_count;
2664 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2665 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2666 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2667 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2668 spin_unlock(&mapping->i_mmap_lock);
2670 EXPORT_SYMBOL(unmap_mapping_range);
2672 int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2674 struct address_space *mapping = inode->i_mapping;
2677 * If the underlying filesystem is not going to provide
2678 * a way to truncate a range of blocks (punch a hole) -
2679 * we should return failure right now.
2681 if (!inode->i_op->truncate_range)
2684 mutex_lock(&inode->i_mutex);
2685 down_write(&inode->i_alloc_sem);
2686 unmap_mapping_range(mapping, offset, (end - offset), 1);
2687 truncate_inode_pages_range(mapping, offset, end);
2688 unmap_mapping_range(mapping, offset, (end - offset), 1);
2689 inode->i_op->truncate_range(inode, offset, end);
2690 up_write(&inode->i_alloc_sem);
2691 mutex_unlock(&inode->i_mutex);
2697 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2698 * but allow concurrent faults), and pte mapped but not yet locked.
2699 * We return with mmap_sem still held, but pte unmapped and unlocked.
2701 static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2702 unsigned long address, pte_t *page_table, pmd_t *pmd,
2703 unsigned int flags, pte_t orig_pte)
2709 struct mem_cgroup *ptr = NULL;
2712 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2715 entry = pte_to_swp_entry(orig_pte);
2716 if (unlikely(non_swap_entry(entry))) {
2717 if (is_migration_entry(entry)) {
2718 migration_entry_wait(mm, pmd, address);
2719 } else if (is_hwpoison_entry(entry)) {
2720 ret = VM_FAULT_HWPOISON;
2722 print_bad_pte(vma, address, orig_pte, NULL);
2723 ret = VM_FAULT_SIGBUS;
2727 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
2728 page = lookup_swap_cache(entry);
2730 grab_swap_token(mm); /* Contend for token _before_ read-in */
2731 page = swapin_readahead(entry,
2732 GFP_HIGHUSER_MOVABLE, vma, address);
2735 * Back out if somebody else faulted in this pte
2736 * while we released the pte lock.
2738 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2739 if (likely(pte_same(*page_table, orig_pte)))
2741 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2745 /* Had to read the page from swap area: Major fault */
2746 ret = VM_FAULT_MAJOR;
2747 count_vm_event(PGMAJFAULT);
2748 } else if (PageHWPoison(page)) {
2750 * hwpoisoned dirty swapcache pages are kept for killing
2751 * owner processes (which may be unknown at hwpoison time)
2753 ret = VM_FAULT_HWPOISON;
2754 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2759 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2761 page = ksm_might_need_to_copy(page, vma, address);
2767 if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
2773 * Back out if somebody else already faulted in this pte.
2775 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2776 if (unlikely(!pte_same(*page_table, orig_pte)))
2779 if (unlikely(!PageUptodate(page))) {
2780 ret = VM_FAULT_SIGBUS;
2785 * The page isn't present yet, go ahead with the fault.
2787 * Be careful about the sequence of operations here.
2788 * To get its accounting right, reuse_swap_page() must be called
2789 * while the page is counted on swap but not yet in mapcount i.e.
2790 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2791 * must be called after the swap_free(), or it will never succeed.
2792 * Because delete_from_swap_page() may be called by reuse_swap_page(),
2793 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
2794 * in page->private. In this case, a record in swap_cgroup is silently
2795 * discarded at swap_free().
2798 inc_mm_counter_fast(mm, MM_ANONPAGES);
2799 dec_mm_counter_fast(mm, MM_SWAPENTS);
2800 pte = mk_pte(page, vma->vm_page_prot);
2801 if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
2802 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2803 flags &= ~FAULT_FLAG_WRITE;
2805 flush_icache_page(vma, page);
2806 set_pte_at(mm, address, page_table, pte);
2807 page_add_anon_rmap(page, vma, address);
2808 /* It's better to call commit-charge after rmap is established */
2809 mem_cgroup_commit_charge_swapin(page, ptr);
2812 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
2813 try_to_free_swap(page);
2816 if (flags & FAULT_FLAG_WRITE) {
2817 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2818 if (ret & VM_FAULT_ERROR)
2819 ret &= VM_FAULT_ERROR;
2823 /* No need to invalidate - it was non-present before */
2824 update_mmu_cache(vma, address, page_table);
2826 pte_unmap_unlock(page_table, ptl);
2830 mem_cgroup_cancel_charge_swapin(ptr);
2831 pte_unmap_unlock(page_table, ptl);
2835 page_cache_release(page);
2840 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2841 * but allow concurrent faults), and pte mapped but not yet locked.
2842 * We return with mmap_sem still held, but pte unmapped and unlocked.
2844 static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2845 unsigned long address, pte_t *page_table, pmd_t *pmd,
2852 if (!(flags & FAULT_FLAG_WRITE)) {
2853 entry = pte_mkspecial(pfn_pte(my_zero_pfn(address),
2854 vma->vm_page_prot));
2855 ptl = pte_lockptr(mm, pmd);
2857 if (!pte_none(*page_table))
2862 /* Allocate our own private page. */
2863 pte_unmap(page_table);
2865 if (unlikely(anon_vma_prepare(vma)))
2867 page = alloc_zeroed_user_highpage_movable(vma, address);
2870 __SetPageUptodate(page);
2872 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))
2875 entry = mk_pte(page, vma->vm_page_prot);
2876 if (vma->vm_flags & VM_WRITE)
2877 entry = pte_mkwrite(pte_mkdirty(entry));
2879 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2880 if (!pte_none(*page_table))
2883 inc_mm_counter_fast(mm, MM_ANONPAGES);
2884 page_add_new_anon_rmap(page, vma, address);
2886 set_pte_at(mm, address, page_table, entry);
2888 /* No need to invalidate - it was non-present before */
2889 update_mmu_cache(vma, address, page_table);
2891 pte_unmap_unlock(page_table, ptl);
2894 mem_cgroup_uncharge_page(page);
2895 page_cache_release(page);
2898 page_cache_release(page);
2900 return VM_FAULT_OOM;
2904 * __do_fault() tries to create a new page mapping. It aggressively
2905 * tries to share with existing pages, but makes a separate copy if
2906 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2907 * the next page fault.
2909 * As this is called only for pages that do not currently exist, we
2910 * do not need to flush old virtual caches or the TLB.
2912 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2913 * but allow concurrent faults), and pte neither mapped nor locked.
2914 * We return with mmap_sem still held, but pte unmapped and unlocked.
2916 static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2917 unsigned long address, pmd_t *pmd,
2918 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
2926 struct page *dirty_page = NULL;
2927 struct vm_fault vmf;
2929 int page_mkwrite = 0;
2931 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2936 ret = vma->vm_ops->fault(vma, &vmf);
2937 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2940 if (unlikely(PageHWPoison(vmf.page))) {
2941 if (ret & VM_FAULT_LOCKED)
2942 unlock_page(vmf.page);
2943 return VM_FAULT_HWPOISON;
2947 * For consistency in subsequent calls, make the faulted page always
2950 if (unlikely(!(ret & VM_FAULT_LOCKED)))
2951 lock_page(vmf.page);
2953 VM_BUG_ON(!PageLocked(vmf.page));
2956 * Should we do an early C-O-W break?
2959 if (flags & FAULT_FLAG_WRITE) {
2960 if (!(vma->vm_flags & VM_SHARED)) {
2962 if (unlikely(anon_vma_prepare(vma))) {
2966 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2972 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) {
2974 page_cache_release(page);
2979 * Don't let another task, with possibly unlocked vma,
2980 * keep the mlocked page.
2982 if (vma->vm_flags & VM_LOCKED)
2983 clear_page_mlock(vmf.page);
2984 copy_user_highpage(page, vmf.page, address, vma);
2985 __SetPageUptodate(page);
2988 * If the page will be shareable, see if the backing
2989 * address space wants to know that the page is about
2990 * to become writable
2992 if (vma->vm_ops->page_mkwrite) {
2996 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2997 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2999 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
3001 goto unwritable_page;
3003 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
3005 if (!page->mapping) {
3006 ret = 0; /* retry the fault */
3008 goto unwritable_page;
3011 VM_BUG_ON(!PageLocked(page));
3018 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
3021 * This silly early PAGE_DIRTY setting removes a race
3022 * due to the bad i386 page protection. But it's valid
3023 * for other architectures too.
3025 * Note that if FAULT_FLAG_WRITE is set, we either now have
3026 * an exclusive copy of the page, or this is a shared mapping,
3027 * so we can make it writable and dirty to avoid having to
3028 * handle that later.
3030 /* Only go through if we didn't race with anybody else... */
3031 if (likely(pte_same(*page_table, orig_pte))) {
3032 flush_icache_page(vma, page);
3033 entry = mk_pte(page, vma->vm_page_prot);
3034 if (flags & FAULT_FLAG_WRITE)
3035 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
3037 inc_mm_counter_fast(mm, MM_ANONPAGES);
3038 page_add_new_anon_rmap(page, vma, address);
3040 inc_mm_counter_fast(mm, MM_FILEPAGES);
3041 page_add_file_rmap(page);
3042 if (flags & FAULT_FLAG_WRITE) {
3044 get_page(dirty_page);
3047 set_pte_at(mm, address, page_table, entry);
3049 /* no need to invalidate: a not-present page won't be cached */
3050 update_mmu_cache(vma, address, page_table);
3053 mem_cgroup_uncharge_page(page);
3055 page_cache_release(page);
3057 anon = 1; /* no anon but release faulted_page */
3060 pte_unmap_unlock(page_table, ptl);
3064 struct address_space *mapping = page->mapping;
3066 if (set_page_dirty(dirty_page))
3068 unlock_page(dirty_page);
3069 put_page(dirty_page);
3070 if (page_mkwrite && mapping) {
3072 * Some device drivers do not set page.mapping but still
3075 balance_dirty_pages_ratelimited(mapping);
3078 /* file_update_time outside page_lock */
3080 file_update_time(vma->vm_file);
3082 unlock_page(vmf.page);
3084 page_cache_release(vmf.page);
3090 page_cache_release(page);
3094 static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3095 unsigned long address, pte_t *page_table, pmd_t *pmd,
3096 unsigned int flags, pte_t orig_pte)
3098 pgoff_t pgoff = (((address & PAGE_MASK)
3099 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
3101 pte_unmap(page_table);
3102 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
3106 * Fault of a previously existing named mapping. Repopulate the pte
3107 * from the encoded file_pte if possible. This enables swappable
3110 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3111 * but allow concurrent faults), and pte mapped but not yet locked.
3112 * We return with mmap_sem still held, but pte unmapped and unlocked.
3114 static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3115 unsigned long address, pte_t *page_table, pmd_t *pmd,
3116 unsigned int flags, pte_t orig_pte)
3120 flags |= FAULT_FLAG_NONLINEAR;
3122 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
3125 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
3127 * Page table corrupted: show pte and kill process.
3129 print_bad_pte(vma, address, orig_pte, NULL);
3130 return VM_FAULT_SIGBUS;
3133 pgoff = pte_to_pgoff(orig_pte);
3134 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
3138 * These routines also need to handle stuff like marking pages dirty
3139 * and/or accessed for architectures that don't do it in hardware (most
3140 * RISC architectures). The early dirtying is also good on the i386.
3142 * There is also a hook called "update_mmu_cache()" that architectures
3143 * with external mmu caches can use to update those (ie the Sparc or
3144 * PowerPC hashed page tables that act as extended TLBs).
3146 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3147 * but allow concurrent faults), and pte mapped but not yet locked.
3148 * We return with mmap_sem still held, but pte unmapped and unlocked.
3150 static inline int handle_pte_fault(struct mm_struct *mm,
3151 struct vm_area_struct *vma, unsigned long address,
3152 pte_t *pte, pmd_t *pmd, unsigned int flags)
3158 if (!pte_present(entry)) {
3159 if (pte_none(entry)) {
3161 if (likely(vma->vm_ops->fault))
3162 return do_linear_fault(mm, vma, address,
3163 pte, pmd, flags, entry);
3165 return do_anonymous_page(mm, vma, address,
3168 if (pte_file(entry))
3169 return do_nonlinear_fault(mm, vma, address,
3170 pte, pmd, flags, entry);
3171 return do_swap_page(mm, vma, address,
3172 pte, pmd, flags, entry);
3175 ptl = pte_lockptr(mm, pmd);
3177 if (unlikely(!pte_same(*pte, entry)))
3179 if (flags & FAULT_FLAG_WRITE) {
3180 if (!pte_write(entry))
3181 return do_wp_page(mm, vma, address,
3182 pte, pmd, ptl, entry);
3183 entry = pte_mkdirty(entry);
3185 entry = pte_mkyoung(entry);
3186 if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
3187 update_mmu_cache(vma, address, pte);
3190 * This is needed only for protection faults but the arch code
3191 * is not yet telling us if this is a protection fault or not.
3192 * This still avoids useless tlb flushes for .text page faults
3195 if (flags & FAULT_FLAG_WRITE)
3196 flush_tlb_page(vma, address);
3199 pte_unmap_unlock(pte, ptl);
3204 * By the time we get here, we already hold the mm semaphore
3206 int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
3207 unsigned long address, unsigned int flags)
3214 __set_current_state(TASK_RUNNING);
3216 count_vm_event(PGFAULT);
3218 /* do counter updates before entering really critical section. */
3219 check_sync_rss_stat(current);
3221 if (unlikely(is_vm_hugetlb_page(vma)))
3222 return hugetlb_fault(mm, vma, address, flags);
3224 pgd = pgd_offset(mm, address);
3225 pud = pud_alloc(mm, pgd, address);
3227 return VM_FAULT_OOM;
3228 pmd = pmd_alloc(mm, pud, address);
3230 return VM_FAULT_OOM;
3231 pte = pte_alloc_map(mm, pmd, address);
3233 return VM_FAULT_OOM;
3235 return handle_pte_fault(mm, vma, address, pte, pmd, flags);
3238 #ifndef __PAGETABLE_PUD_FOLDED
3240 * Allocate page upper directory.
3241 * We've already handled the fast-path in-line.
3243 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
3245 pud_t *new = pud_alloc_one(mm, address);
3249 smp_wmb(); /* See comment in __pte_alloc */
3251 spin_lock(&mm->page_table_lock);
3252 if (pgd_present(*pgd)) /* Another has populated it */
3255 pgd_populate(mm, pgd, new);
3256 spin_unlock(&mm->page_table_lock);
3259 #endif /* __PAGETABLE_PUD_FOLDED */
3261 #ifndef __PAGETABLE_PMD_FOLDED
3263 * Allocate page middle directory.
3264 * We've already handled the fast-path in-line.
3266 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
3268 pmd_t *new = pmd_alloc_one(mm, address);
3272 smp_wmb(); /* See comment in __pte_alloc */
3274 spin_lock(&mm->page_table_lock);
3275 #ifndef __ARCH_HAS_4LEVEL_HACK
3276 if (pud_present(*pud)) /* Another has populated it */
3279 pud_populate(mm, pud, new);
3281 if (pgd_present(*pud)) /* Another has populated it */
3284 pgd_populate(mm, pud, new);
3285 #endif /* __ARCH_HAS_4LEVEL_HACK */
3286 spin_unlock(&mm->page_table_lock);
3289 #endif /* __PAGETABLE_PMD_FOLDED */
3291 int make_pages_present(unsigned long addr, unsigned long end)
3293 int ret, len, write;
3294 struct vm_area_struct * vma;
3296 vma = find_vma(current->mm, addr);
3299 write = (vma->vm_flags & VM_WRITE) != 0;
3300 BUG_ON(addr >= end);
3301 BUG_ON(end > vma->vm_end);
3302 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
3303 ret = get_user_pages(current, current->mm, addr,
3304 len, write, 0, NULL, NULL);
3307 return ret == len ? 0 : -EFAULT;
3310 #if !defined(__HAVE_ARCH_GATE_AREA)
3312 #if defined(AT_SYSINFO_EHDR)
3313 static struct vm_area_struct gate_vma;
3315 static int __init gate_vma_init(void)
3317 gate_vma.vm_mm = NULL;
3318 gate_vma.vm_start = FIXADDR_USER_START;
3319 gate_vma.vm_end = FIXADDR_USER_END;
3320 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
3321 gate_vma.vm_page_prot = __P101;
3323 * Make sure the vDSO gets into every core dump.
3324 * Dumping its contents makes post-mortem fully interpretable later
3325 * without matching up the same kernel and hardware config to see
3326 * what PC values meant.
3328 gate_vma.vm_flags |= VM_ALWAYSDUMP;
3331 __initcall(gate_vma_init);
3334 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
3336 #ifdef AT_SYSINFO_EHDR
3343 int in_gate_area_no_task(unsigned long addr)
3345 #ifdef AT_SYSINFO_EHDR
3346 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
3352 #endif /* __HAVE_ARCH_GATE_AREA */
3354 static int follow_pte(struct mm_struct *mm, unsigned long address,
3355 pte_t **ptepp, spinlock_t **ptlp)
3362 pgd = pgd_offset(mm, address);
3363 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3366 pud = pud_offset(pgd, address);
3367 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3370 pmd = pmd_offset(pud, address);
3371 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3374 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3378 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3381 if (!pte_present(*ptep))
3386 pte_unmap_unlock(ptep, *ptlp);
3392 * follow_pfn - look up PFN at a user virtual address
3393 * @vma: memory mapping
3394 * @address: user virtual address
3395 * @pfn: location to store found PFN
3397 * Only IO mappings and raw PFN mappings are allowed.
3399 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3401 int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3408 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3411 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3414 *pfn = pte_pfn(*ptep);
3415 pte_unmap_unlock(ptep, ptl);
3418 EXPORT_SYMBOL(follow_pfn);
3420 #ifdef CONFIG_HAVE_IOREMAP_PROT
3421 int follow_phys(struct vm_area_struct *vma,
3422 unsigned long address, unsigned int flags,
3423 unsigned long *prot, resource_size_t *phys)
3429 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3432 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
3436 if ((flags & FOLL_WRITE) && !pte_write(pte))
3439 *prot = pgprot_val(pte_pgprot(pte));
3440 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
3444 pte_unmap_unlock(ptep, ptl);
3449 int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3450 void *buf, int len, int write)
3452 resource_size_t phys_addr;
3453 unsigned long prot = 0;
3454 void __iomem *maddr;
3455 int offset = addr & (PAGE_SIZE-1);
3457 if (follow_phys(vma, addr, write, &prot, &phys_addr))
3460 maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
3462 memcpy_toio(maddr + offset, buf, len);
3464 memcpy_fromio(buf, maddr + offset, len);
3472 * Access another process' address space.
3473 * Source/target buffer must be kernel space,
3474 * Do not walk the page table directly, use get_user_pages
3476 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
3478 struct mm_struct *mm;
3479 struct vm_area_struct *vma;
3480 void *old_buf = buf;
3482 mm = get_task_mm(tsk);
3486 down_read(&mm->mmap_sem);
3487 /* ignore errors, just check how much was successfully transferred */
3489 int bytes, ret, offset;
3491 struct page *page = NULL;
3493 ret = get_user_pages(tsk, mm, addr, 1,
3494 write, 1, &page, &vma);
3497 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3498 * we can access using slightly different code.
3500 #ifdef CONFIG_HAVE_IOREMAP_PROT
3501 vma = find_vma(mm, addr);
3504 if (vma->vm_ops && vma->vm_ops->access)
3505 ret = vma->vm_ops->access(vma, addr, buf,
3513 offset = addr & (PAGE_SIZE-1);
3514 if (bytes > PAGE_SIZE-offset)
3515 bytes = PAGE_SIZE-offset;
3519 copy_to_user_page(vma, page, addr,
3520 maddr + offset, buf, bytes);
3521 set_page_dirty_lock(page);
3523 copy_from_user_page(vma, page, addr,
3524 buf, maddr + offset, bytes);
3527 page_cache_release(page);
3533 up_read(&mm->mmap_sem);
3536 return buf - old_buf;
3540 * Print the name of a VMA.
3542 void print_vma_addr(char *prefix, unsigned long ip)
3544 struct mm_struct *mm = current->mm;
3545 struct vm_area_struct *vma;
3548 * Do not print if we are in atomic
3549 * contexts (in exception stacks, etc.):
3551 if (preempt_count())
3554 down_read(&mm->mmap_sem);
3555 vma = find_vma(mm, ip);
3556 if (vma && vma->vm_file) {
3557 struct file *f = vma->vm_file;
3558 char *buf = (char *)__get_free_page(GFP_KERNEL);
3562 p = d_path(&f->f_path, buf, PAGE_SIZE);
3565 s = strrchr(p, '/');
3568 printk("%s%s[%lx+%lx]", prefix, p,
3570 vma->vm_end - vma->vm_start);
3571 free_page((unsigned long)buf);
3574 up_read(¤t->mm->mmap_sem);
3577 #ifdef CONFIG_PROVE_LOCKING
3578 void might_fault(void)
3581 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3582 * holding the mmap_sem, this is safe because kernel memory doesn't
3583 * get paged out, therefore we'll never actually fault, and the
3584 * below annotations will generate false positives.
3586 if (segment_eq(get_fs(), KERNEL_DS))
3591 * it would be nicer only to annotate paths which are not under
3592 * pagefault_disable, however that requires a larger audit and
3593 * providing helpers like get_user_atomic.
3595 if (!in_atomic() && current->mm)
3596 might_lock_read(¤t->mm->mmap_sem);
3598 EXPORT_SYMBOL(might_fault);