6 * Address space accounting code <alan@redhat.com>
9 #include <linux/slab.h>
10 #include <linux/backing-dev.h>
12 #include <linux/shm.h>
13 #include <linux/mman.h>
14 #include <linux/pagemap.h>
15 #include <linux/swap.h>
16 #include <linux/syscalls.h>
17 #include <linux/capability.h>
18 #include <linux/init.h>
19 #include <linux/file.h>
21 #include <linux/personality.h>
22 #include <linux/security.h>
23 #include <linux/ima.h>
24 #include <linux/hugetlb.h>
25 #include <linux/profile.h>
26 #include <linux/module.h>
27 #include <linux/mount.h>
28 #include <linux/mempolicy.h>
29 #include <linux/rmap.h>
30 #include <linux/mmu_notifier.h>
32 #include <asm/uaccess.h>
33 #include <asm/cacheflush.h>
35 #include <asm/mmu_context.h>
39 #ifndef arch_mmap_check
40 #define arch_mmap_check(addr, len, flags) (0)
43 #ifndef arch_rebalance_pgtables
44 #define arch_rebalance_pgtables(addr, len) (addr)
47 static void unmap_region(struct mm_struct *mm,
48 struct vm_area_struct *vma, struct vm_area_struct *prev,
49 unsigned long start, unsigned long end);
52 * WARNING: the debugging will use recursive algorithms so never enable this
53 * unless you know what you are doing.
57 /* description of effects of mapping type and prot in current implementation.
58 * this is due to the limited x86 page protection hardware. The expected
59 * behavior is in parens:
62 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
63 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
64 * w: (no) no w: (no) no w: (yes) yes w: (no) no
65 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
67 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
68 * w: (no) no w: (no) no w: (copy) copy w: (no) no
69 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
72 pgprot_t protection_map[16] = {
73 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
74 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
77 pgprot_t vm_get_page_prot(unsigned long vm_flags)
79 return __pgprot(pgprot_val(protection_map[vm_flags &
80 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
81 pgprot_val(arch_vm_get_page_prot(vm_flags)));
83 EXPORT_SYMBOL(vm_get_page_prot);
85 int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */
86 int sysctl_overcommit_ratio = 50; /* default is 50% */
87 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
88 atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0);
91 * Check that a process has enough memory to allocate a new virtual
92 * mapping. 0 means there is enough memory for the allocation to
93 * succeed and -ENOMEM implies there is not.
95 * We currently support three overcommit policies, which are set via the
96 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
98 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
99 * Additional code 2002 Jul 20 by Robert Love.
101 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
103 * Note this is a helper function intended to be used by LSMs which
104 * wish to use this logic.
106 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
108 unsigned long free, allowed;
110 vm_acct_memory(pages);
113 * Sometimes we want to use more memory than we have
115 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
118 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
121 free = global_page_state(NR_FILE_PAGES);
122 free += nr_swap_pages;
125 * Any slabs which are created with the
126 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
127 * which are reclaimable, under pressure. The dentry
128 * cache and most inode caches should fall into this
130 free += global_page_state(NR_SLAB_RECLAIMABLE);
133 * Leave the last 3% for root
142 * nr_free_pages() is very expensive on large systems,
143 * only call if we're about to fail.
148 * Leave reserved pages. The pages are not for anonymous pages.
150 if (n <= totalreserve_pages)
153 n -= totalreserve_pages;
156 * Leave the last 3% for root
168 allowed = (totalram_pages - hugetlb_total_pages())
169 * sysctl_overcommit_ratio / 100;
171 * Leave the last 3% for root
174 allowed -= allowed / 32;
175 allowed += total_swap_pages;
177 /* Don't let a single process grow too big:
178 leave 3% of the size of this process for other processes */
180 allowed -= mm->total_vm / 32;
183 * cast `allowed' as a signed long because vm_committed_space
184 * sometimes has a negative value
186 if (atomic_long_read(&vm_committed_space) < (long)allowed)
189 vm_unacct_memory(pages);
195 * Requires inode->i_mapping->i_mmap_lock
197 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
198 struct file *file, struct address_space *mapping)
200 if (vma->vm_flags & VM_DENYWRITE)
201 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
202 if (vma->vm_flags & VM_SHARED)
203 mapping->i_mmap_writable--;
205 flush_dcache_mmap_lock(mapping);
206 if (unlikely(vma->vm_flags & VM_NONLINEAR))
207 list_del_init(&vma->shared.vm_set.list);
209 vma_prio_tree_remove(vma, &mapping->i_mmap);
210 flush_dcache_mmap_unlock(mapping);
214 * Unlink a file-based vm structure from its prio_tree, to hide
215 * vma from rmap and vmtruncate before freeing its page tables.
217 void unlink_file_vma(struct vm_area_struct *vma)
219 struct file *file = vma->vm_file;
222 struct address_space *mapping = file->f_mapping;
223 spin_lock(&mapping->i_mmap_lock);
224 __remove_shared_vm_struct(vma, file, mapping);
225 spin_unlock(&mapping->i_mmap_lock);
230 * Close a vm structure and free it, returning the next.
232 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
234 struct vm_area_struct *next = vma->vm_next;
237 if (vma->vm_ops && vma->vm_ops->close)
238 vma->vm_ops->close(vma);
241 if (vma->vm_flags & VM_EXECUTABLE)
242 removed_exe_file_vma(vma->vm_mm);
244 mpol_put(vma_policy(vma));
245 kmem_cache_free(vm_area_cachep, vma);
249 asmlinkage unsigned long sys_brk(unsigned long brk)
251 unsigned long rlim, retval;
252 unsigned long newbrk, oldbrk;
253 struct mm_struct *mm = current->mm;
254 unsigned long min_brk;
256 down_write(&mm->mmap_sem);
258 #ifdef CONFIG_COMPAT_BRK
259 min_brk = mm->end_code;
261 min_brk = mm->start_brk;
267 * Check against rlimit here. If this check is done later after the test
268 * of oldbrk with newbrk then it can escape the test and let the data
269 * segment grow beyond its set limit the in case where the limit is
270 * not page aligned -Ram Gupta
272 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
273 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
274 (mm->end_data - mm->start_data) > rlim)
277 newbrk = PAGE_ALIGN(brk);
278 oldbrk = PAGE_ALIGN(mm->brk);
279 if (oldbrk == newbrk)
282 /* Always allow shrinking brk. */
283 if (brk <= mm->brk) {
284 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
289 /* Check against existing mmap mappings. */
290 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
293 /* Ok, looks good - let it rip. */
294 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
300 up_write(&mm->mmap_sem);
305 static int browse_rb(struct rb_root *root)
308 struct rb_node *nd, *pn = NULL;
309 unsigned long prev = 0, pend = 0;
311 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
312 struct vm_area_struct *vma;
313 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
314 if (vma->vm_start < prev)
315 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
316 if (vma->vm_start < pend)
317 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
318 if (vma->vm_start > vma->vm_end)
319 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
322 prev = vma->vm_start;
326 for (nd = pn; nd; nd = rb_prev(nd)) {
330 printk("backwards %d, forwards %d\n", j, i), i = 0;
334 void validate_mm(struct mm_struct *mm)
338 struct vm_area_struct *tmp = mm->mmap;
343 if (i != mm->map_count)
344 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
345 i = browse_rb(&mm->mm_rb);
346 if (i != mm->map_count)
347 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
351 #define validate_mm(mm) do { } while (0)
354 static struct vm_area_struct *
355 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
356 struct vm_area_struct **pprev, struct rb_node ***rb_link,
357 struct rb_node ** rb_parent)
359 struct vm_area_struct * vma;
360 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
362 __rb_link = &mm->mm_rb.rb_node;
363 rb_prev = __rb_parent = NULL;
367 struct vm_area_struct *vma_tmp;
369 __rb_parent = *__rb_link;
370 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
372 if (vma_tmp->vm_end > addr) {
374 if (vma_tmp->vm_start <= addr)
376 __rb_link = &__rb_parent->rb_left;
378 rb_prev = __rb_parent;
379 __rb_link = &__rb_parent->rb_right;
385 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
386 *rb_link = __rb_link;
387 *rb_parent = __rb_parent;
392 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
393 struct vm_area_struct *prev, struct rb_node *rb_parent)
396 vma->vm_next = prev->vm_next;
401 vma->vm_next = rb_entry(rb_parent,
402 struct vm_area_struct, vm_rb);
408 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
409 struct rb_node **rb_link, struct rb_node *rb_parent)
411 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
412 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
415 static void __vma_link_file(struct vm_area_struct *vma)
421 struct address_space *mapping = file->f_mapping;
423 if (vma->vm_flags & VM_DENYWRITE)
424 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
425 if (vma->vm_flags & VM_SHARED)
426 mapping->i_mmap_writable++;
428 flush_dcache_mmap_lock(mapping);
429 if (unlikely(vma->vm_flags & VM_NONLINEAR))
430 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
432 vma_prio_tree_insert(vma, &mapping->i_mmap);
433 flush_dcache_mmap_unlock(mapping);
438 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
439 struct vm_area_struct *prev, struct rb_node **rb_link,
440 struct rb_node *rb_parent)
442 __vma_link_list(mm, vma, prev, rb_parent);
443 __vma_link_rb(mm, vma, rb_link, rb_parent);
444 __anon_vma_link(vma);
447 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
448 struct vm_area_struct *prev, struct rb_node **rb_link,
449 struct rb_node *rb_parent)
451 struct address_space *mapping = NULL;
454 mapping = vma->vm_file->f_mapping;
457 spin_lock(&mapping->i_mmap_lock);
458 vma->vm_truncate_count = mapping->truncate_count;
462 __vma_link(mm, vma, prev, rb_link, rb_parent);
463 __vma_link_file(vma);
465 anon_vma_unlock(vma);
467 spin_unlock(&mapping->i_mmap_lock);
474 * Helper for vma_adjust in the split_vma insert case:
475 * insert vm structure into list and rbtree and anon_vma,
476 * but it has already been inserted into prio_tree earlier.
479 __insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
481 struct vm_area_struct * __vma, * prev;
482 struct rb_node ** rb_link, * rb_parent;
484 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
485 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
486 __vma_link(mm, vma, prev, rb_link, rb_parent);
491 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
492 struct vm_area_struct *prev)
494 prev->vm_next = vma->vm_next;
495 rb_erase(&vma->vm_rb, &mm->mm_rb);
496 if (mm->mmap_cache == vma)
497 mm->mmap_cache = prev;
501 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
502 * is already present in an i_mmap tree without adjusting the tree.
503 * The following helper function should be used when such adjustments
504 * are necessary. The "insert" vma (if any) is to be inserted
505 * before we drop the necessary locks.
507 void vma_adjust(struct vm_area_struct *vma, unsigned long start,
508 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
510 struct mm_struct *mm = vma->vm_mm;
511 struct vm_area_struct *next = vma->vm_next;
512 struct vm_area_struct *importer = NULL;
513 struct address_space *mapping = NULL;
514 struct prio_tree_root *root = NULL;
515 struct file *file = vma->vm_file;
516 struct anon_vma *anon_vma = NULL;
517 long adjust_next = 0;
520 if (next && !insert) {
521 if (end >= next->vm_end) {
523 * vma expands, overlapping all the next, and
524 * perhaps the one after too (mprotect case 6).
526 again: remove_next = 1 + (end > next->vm_end);
528 anon_vma = next->anon_vma;
530 } else if (end > next->vm_start) {
532 * vma expands, overlapping part of the next:
533 * mprotect case 5 shifting the boundary up.
535 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
536 anon_vma = next->anon_vma;
538 } else if (end < vma->vm_end) {
540 * vma shrinks, and !insert tells it's not
541 * split_vma inserting another: so it must be
542 * mprotect case 4 shifting the boundary down.
544 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
545 anon_vma = next->anon_vma;
551 mapping = file->f_mapping;
552 if (!(vma->vm_flags & VM_NONLINEAR))
553 root = &mapping->i_mmap;
554 spin_lock(&mapping->i_mmap_lock);
556 vma->vm_truncate_count != next->vm_truncate_count) {
558 * unmap_mapping_range might be in progress:
559 * ensure that the expanding vma is rescanned.
561 importer->vm_truncate_count = 0;
564 insert->vm_truncate_count = vma->vm_truncate_count;
566 * Put into prio_tree now, so instantiated pages
567 * are visible to arm/parisc __flush_dcache_page
568 * throughout; but we cannot insert into address
569 * space until vma start or end is updated.
571 __vma_link_file(insert);
576 * When changing only vma->vm_end, we don't really need
577 * anon_vma lock: but is that case worth optimizing out?
580 anon_vma = vma->anon_vma;
582 spin_lock(&anon_vma->lock);
584 * Easily overlooked: when mprotect shifts the boundary,
585 * make sure the expanding vma has anon_vma set if the
586 * shrinking vma had, to cover any anon pages imported.
588 if (importer && !importer->anon_vma) {
589 importer->anon_vma = anon_vma;
590 __anon_vma_link(importer);
595 flush_dcache_mmap_lock(mapping);
596 vma_prio_tree_remove(vma, root);
598 vma_prio_tree_remove(next, root);
601 vma->vm_start = start;
603 vma->vm_pgoff = pgoff;
605 next->vm_start += adjust_next << PAGE_SHIFT;
606 next->vm_pgoff += adjust_next;
611 vma_prio_tree_insert(next, root);
612 vma_prio_tree_insert(vma, root);
613 flush_dcache_mmap_unlock(mapping);
618 * vma_merge has merged next into vma, and needs
619 * us to remove next before dropping the locks.
621 __vma_unlink(mm, next, vma);
623 __remove_shared_vm_struct(next, file, mapping);
625 __anon_vma_merge(vma, next);
628 * split_vma has split insert from vma, and needs
629 * us to insert it before dropping the locks
630 * (it may either follow vma or precede it).
632 __insert_vm_struct(mm, insert);
636 spin_unlock(&anon_vma->lock);
638 spin_unlock(&mapping->i_mmap_lock);
643 if (next->vm_flags & VM_EXECUTABLE)
644 removed_exe_file_vma(mm);
647 mpol_put(vma_policy(next));
648 kmem_cache_free(vm_area_cachep, next);
650 * In mprotect's case 6 (see comments on vma_merge),
651 * we must remove another next too. It would clutter
652 * up the code too much to do both in one go.
654 if (remove_next == 2) {
664 * If the vma has a ->close operation then the driver probably needs to release
665 * per-vma resources, so we don't attempt to merge those.
667 static inline int is_mergeable_vma(struct vm_area_struct *vma,
668 struct file *file, unsigned long vm_flags)
670 if (vma->vm_flags != vm_flags)
672 if (vma->vm_file != file)
674 if (vma->vm_ops && vma->vm_ops->close)
679 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
680 struct anon_vma *anon_vma2)
682 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
686 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
687 * in front of (at a lower virtual address and file offset than) the vma.
689 * We cannot merge two vmas if they have differently assigned (non-NULL)
690 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
692 * We don't check here for the merged mmap wrapping around the end of pagecache
693 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
694 * wrap, nor mmaps which cover the final page at index -1UL.
697 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
698 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
700 if (is_mergeable_vma(vma, file, vm_flags) &&
701 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
702 if (vma->vm_pgoff == vm_pgoff)
709 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
710 * beyond (at a higher virtual address and file offset than) the vma.
712 * We cannot merge two vmas if they have differently assigned (non-NULL)
713 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
716 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
717 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
719 if (is_mergeable_vma(vma, file, vm_flags) &&
720 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
722 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
723 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
730 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
731 * whether that can be merged with its predecessor or its successor.
732 * Or both (it neatly fills a hole).
734 * In most cases - when called for mmap, brk or mremap - [addr,end) is
735 * certain not to be mapped by the time vma_merge is called; but when
736 * called for mprotect, it is certain to be already mapped (either at
737 * an offset within prev, or at the start of next), and the flags of
738 * this area are about to be changed to vm_flags - and the no-change
739 * case has already been eliminated.
741 * The following mprotect cases have to be considered, where AAAA is
742 * the area passed down from mprotect_fixup, never extending beyond one
743 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
745 * AAAA AAAA AAAA AAAA
746 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
747 * cannot merge might become might become might become
748 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
749 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
750 * mremap move: PPPPNNNNNNNN 8
752 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
753 * might become case 1 below case 2 below case 3 below
755 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
756 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
758 struct vm_area_struct *vma_merge(struct mm_struct *mm,
759 struct vm_area_struct *prev, unsigned long addr,
760 unsigned long end, unsigned long vm_flags,
761 struct anon_vma *anon_vma, struct file *file,
762 pgoff_t pgoff, struct mempolicy *policy)
764 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
765 struct vm_area_struct *area, *next;
768 * We later require that vma->vm_flags == vm_flags,
769 * so this tests vma->vm_flags & VM_SPECIAL, too.
771 if (vm_flags & VM_SPECIAL)
775 next = prev->vm_next;
779 if (next && next->vm_end == end) /* cases 6, 7, 8 */
780 next = next->vm_next;
783 * Can it merge with the predecessor?
785 if (prev && prev->vm_end == addr &&
786 mpol_equal(vma_policy(prev), policy) &&
787 can_vma_merge_after(prev, vm_flags,
788 anon_vma, file, pgoff)) {
790 * OK, it can. Can we now merge in the successor as well?
792 if (next && end == next->vm_start &&
793 mpol_equal(policy, vma_policy(next)) &&
794 can_vma_merge_before(next, vm_flags,
795 anon_vma, file, pgoff+pglen) &&
796 is_mergeable_anon_vma(prev->anon_vma,
799 vma_adjust(prev, prev->vm_start,
800 next->vm_end, prev->vm_pgoff, NULL);
801 } else /* cases 2, 5, 7 */
802 vma_adjust(prev, prev->vm_start,
803 end, prev->vm_pgoff, NULL);
808 * Can this new request be merged in front of next?
810 if (next && end == next->vm_start &&
811 mpol_equal(policy, vma_policy(next)) &&
812 can_vma_merge_before(next, vm_flags,
813 anon_vma, file, pgoff+pglen)) {
814 if (prev && addr < prev->vm_end) /* case 4 */
815 vma_adjust(prev, prev->vm_start,
816 addr, prev->vm_pgoff, NULL);
817 else /* cases 3, 8 */
818 vma_adjust(area, addr, next->vm_end,
819 next->vm_pgoff - pglen, NULL);
827 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
828 * neighbouring vmas for a suitable anon_vma, before it goes off
829 * to allocate a new anon_vma. It checks because a repetitive
830 * sequence of mprotects and faults may otherwise lead to distinct
831 * anon_vmas being allocated, preventing vma merge in subsequent
834 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
836 struct vm_area_struct *near;
837 unsigned long vm_flags;
844 * Since only mprotect tries to remerge vmas, match flags
845 * which might be mprotected into each other later on.
846 * Neither mlock nor madvise tries to remerge at present,
847 * so leave their flags as obstructing a merge.
849 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
850 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
852 if (near->anon_vma && vma->vm_end == near->vm_start &&
853 mpol_equal(vma_policy(vma), vma_policy(near)) &&
854 can_vma_merge_before(near, vm_flags,
855 NULL, vma->vm_file, vma->vm_pgoff +
856 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
857 return near->anon_vma;
860 * It is potentially slow to have to call find_vma_prev here.
861 * But it's only on the first write fault on the vma, not
862 * every time, and we could devise a way to avoid it later
863 * (e.g. stash info in next's anon_vma_node when assigning
864 * an anon_vma, or when trying vma_merge). Another time.
866 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
870 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
871 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
873 if (near->anon_vma && near->vm_end == vma->vm_start &&
874 mpol_equal(vma_policy(near), vma_policy(vma)) &&
875 can_vma_merge_after(near, vm_flags,
876 NULL, vma->vm_file, vma->vm_pgoff))
877 return near->anon_vma;
880 * There's no absolute need to look only at touching neighbours:
881 * we could search further afield for "compatible" anon_vmas.
882 * But it would probably just be a waste of time searching,
883 * or lead to too many vmas hanging off the same anon_vma.
884 * We're trying to allow mprotect remerging later on,
885 * not trying to minimize memory used for anon_vmas.
890 #ifdef CONFIG_PROC_FS
891 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
892 struct file *file, long pages)
894 const unsigned long stack_flags
895 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
898 mm->shared_vm += pages;
899 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
900 mm->exec_vm += pages;
901 } else if (flags & stack_flags)
902 mm->stack_vm += pages;
903 if (flags & (VM_RESERVED|VM_IO))
904 mm->reserved_vm += pages;
906 #endif /* CONFIG_PROC_FS */
909 * The caller must hold down_write(current->mm->mmap_sem).
912 unsigned long do_mmap_pgoff(struct file * file, unsigned long addr,
913 unsigned long len, unsigned long prot,
914 unsigned long flags, unsigned long pgoff)
916 struct mm_struct * mm = current->mm;
918 unsigned int vm_flags;
921 unsigned long reqprot = prot;
924 * Does the application expect PROT_READ to imply PROT_EXEC?
926 * (the exception is when the underlying filesystem is noexec
927 * mounted, in which case we dont add PROT_EXEC.)
929 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
930 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
936 if (!(flags & MAP_FIXED))
937 addr = round_hint_to_min(addr);
939 error = arch_mmap_check(addr, len, flags);
943 /* Careful about overflows.. */
944 len = PAGE_ALIGN(len);
945 if (!len || len > TASK_SIZE)
948 /* offset overflow? */
949 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
952 /* Too many mappings? */
953 if (mm->map_count > sysctl_max_map_count)
956 /* Obtain the address to map to. we verify (or select) it and ensure
957 * that it represents a valid section of the address space.
959 addr = get_unmapped_area(file, addr, len, pgoff, flags);
960 if (addr & ~PAGE_MASK)
963 /* Do simple checking here so the lower-level routines won't have
964 * to. we assume access permissions have been handled by the open
965 * of the memory object, so we don't do any here.
967 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
968 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
970 if (flags & MAP_LOCKED) {
973 vm_flags |= VM_LOCKED;
976 /* mlock MCL_FUTURE? */
977 if (vm_flags & VM_LOCKED) {
978 unsigned long locked, lock_limit;
979 locked = len >> PAGE_SHIFT;
980 locked += mm->locked_vm;
981 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
982 lock_limit >>= PAGE_SHIFT;
983 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
987 inode = file ? file->f_path.dentry->d_inode : NULL;
990 switch (flags & MAP_TYPE) {
992 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
996 * Make sure we don't allow writing to an append-only
999 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1003 * Make sure there are no mandatory locks on the file.
1005 if (locks_verify_locked(inode))
1008 vm_flags |= VM_SHARED | VM_MAYSHARE;
1009 if (!(file->f_mode & FMODE_WRITE))
1010 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1014 if (!(file->f_mode & FMODE_READ))
1016 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1017 if (vm_flags & VM_EXEC)
1019 vm_flags &= ~VM_MAYEXEC;
1021 if (is_file_hugepages(file))
1024 if (!file->f_op || !file->f_op->mmap)
1032 switch (flags & MAP_TYPE) {
1038 vm_flags |= VM_SHARED | VM_MAYSHARE;
1042 * Set pgoff according to addr for anon_vma.
1044 pgoff = addr >> PAGE_SHIFT;
1051 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1054 error = ima_file_mmap(file, prot);
1058 return mmap_region(file, addr, len, flags, vm_flags, pgoff,
1061 EXPORT_SYMBOL(do_mmap_pgoff);
1064 * Some shared mappigns will want the pages marked read-only
1065 * to track write events. If so, we'll downgrade vm_page_prot
1066 * to the private version (using protection_map[] without the
1069 int vma_wants_writenotify(struct vm_area_struct *vma)
1071 unsigned int vm_flags = vma->vm_flags;
1073 /* If it was private or non-writable, the write bit is already clear */
1074 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1077 /* The backer wishes to know when pages are first written to? */
1078 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1081 /* The open routine did something to the protections already? */
1082 if (pgprot_val(vma->vm_page_prot) !=
1083 pgprot_val(vm_get_page_prot(vm_flags)))
1086 /* Specialty mapping? */
1087 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1090 /* Can the mapping track the dirty pages? */
1091 return vma->vm_file && vma->vm_file->f_mapping &&
1092 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1095 unsigned long mmap_region(struct file *file, unsigned long addr,
1096 unsigned long len, unsigned long flags,
1097 unsigned int vm_flags, unsigned long pgoff,
1100 struct mm_struct *mm = current->mm;
1101 struct vm_area_struct *vma, *prev;
1102 int correct_wcount = 0;
1104 struct rb_node **rb_link, *rb_parent;
1105 unsigned long charged = 0;
1106 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1108 /* Clear old maps */
1111 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1112 if (vma && vma->vm_start < addr + len) {
1113 if (do_munmap(mm, addr, len))
1118 /* Check against address space limit. */
1119 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1122 if (flags & MAP_NORESERVE)
1123 vm_flags |= VM_NORESERVE;
1125 if (accountable && (!(flags & MAP_NORESERVE) ||
1126 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) {
1127 if (vm_flags & VM_SHARED) {
1128 /* Check memory availability in shmem_file_setup? */
1129 vm_flags |= VM_ACCOUNT;
1130 } else if (vm_flags & VM_WRITE) {
1132 * Private writable mapping: check memory availability
1134 charged = len >> PAGE_SHIFT;
1135 if (security_vm_enough_memory(charged))
1137 vm_flags |= VM_ACCOUNT;
1142 * Can we just expand an old private anonymous mapping?
1143 * The VM_SHARED test is necessary because shmem_zero_setup
1144 * will create the file object for a shared anonymous map below.
1146 if (!file && !(vm_flags & VM_SHARED)) {
1147 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1148 NULL, NULL, pgoff, NULL);
1154 * Determine the object being mapped and call the appropriate
1155 * specific mapper. the address has already been validated, but
1156 * not unmapped, but the maps are removed from the list.
1158 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1165 vma->vm_start = addr;
1166 vma->vm_end = addr + len;
1167 vma->vm_flags = vm_flags;
1168 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1169 vma->vm_pgoff = pgoff;
1173 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1175 if (vm_flags & VM_DENYWRITE) {
1176 error = deny_write_access(file);
1181 vma->vm_file = file;
1183 error = file->f_op->mmap(file, vma);
1185 goto unmap_and_free_vma;
1186 if (vm_flags & VM_EXECUTABLE)
1187 added_exe_file_vma(mm);
1188 } else if (vm_flags & VM_SHARED) {
1189 error = shmem_zero_setup(vma);
1194 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1195 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1196 * that memory reservation must be checked; but that reservation
1197 * belongs to shared memory object, not to vma: so now clear it.
1199 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT))
1200 vma->vm_flags &= ~VM_ACCOUNT;
1202 /* Can addr have changed??
1204 * Answer: Yes, several device drivers can do it in their
1205 * f_op->mmap method. -DaveM
1207 addr = vma->vm_start;
1208 pgoff = vma->vm_pgoff;
1209 vm_flags = vma->vm_flags;
1211 if (vma_wants_writenotify(vma))
1212 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1214 if (file && vma_merge(mm, prev, addr, vma->vm_end,
1215 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) {
1216 mpol_put(vma_policy(vma));
1217 kmem_cache_free(vm_area_cachep, vma);
1219 if (vm_flags & VM_EXECUTABLE)
1220 removed_exe_file_vma(mm);
1222 vma_link(mm, vma, prev, rb_link, rb_parent);
1223 file = vma->vm_file;
1226 /* Once vma denies write, undo our temporary denial count */
1228 atomic_inc(&inode->i_writecount);
1230 mm->total_vm += len >> PAGE_SHIFT;
1231 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1232 if (vm_flags & VM_LOCKED) {
1234 * makes pages present; downgrades, drops, reacquires mmap_sem
1236 long nr_pages = mlock_vma_pages_range(vma, addr, addr + len);
1238 return nr_pages; /* vma gone! */
1239 mm->locked_vm += (len >> PAGE_SHIFT) - nr_pages;
1240 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1241 make_pages_present(addr, addr + len);
1246 atomic_inc(&inode->i_writecount);
1247 vma->vm_file = NULL;
1250 /* Undo any partial mapping done by a device driver. */
1251 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1254 kmem_cache_free(vm_area_cachep, vma);
1257 vm_unacct_memory(charged);
1261 /* Get an address range which is currently unmapped.
1262 * For shmat() with addr=0.
1264 * Ugly calling convention alert:
1265 * Return value with the low bits set means error value,
1267 * if (ret & ~PAGE_MASK)
1270 * This function "knows" that -ENOMEM has the bits set.
1272 #ifndef HAVE_ARCH_UNMAPPED_AREA
1274 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1275 unsigned long len, unsigned long pgoff, unsigned long flags)
1277 struct mm_struct *mm = current->mm;
1278 struct vm_area_struct *vma;
1279 unsigned long start_addr;
1281 if (len > TASK_SIZE)
1284 if (flags & MAP_FIXED)
1288 addr = PAGE_ALIGN(addr);
1289 vma = find_vma(mm, addr);
1290 if (TASK_SIZE - len >= addr &&
1291 (!vma || addr + len <= vma->vm_start))
1294 if (len > mm->cached_hole_size) {
1295 start_addr = addr = mm->free_area_cache;
1297 start_addr = addr = TASK_UNMAPPED_BASE;
1298 mm->cached_hole_size = 0;
1302 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1303 /* At this point: (!vma || addr < vma->vm_end). */
1304 if (TASK_SIZE - len < addr) {
1306 * Start a new search - just in case we missed
1309 if (start_addr != TASK_UNMAPPED_BASE) {
1310 addr = TASK_UNMAPPED_BASE;
1312 mm->cached_hole_size = 0;
1317 if (!vma || addr + len <= vma->vm_start) {
1319 * Remember the place where we stopped the search:
1321 mm->free_area_cache = addr + len;
1324 if (addr + mm->cached_hole_size < vma->vm_start)
1325 mm->cached_hole_size = vma->vm_start - addr;
1331 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1334 * Is this a new hole at the lowest possible address?
1336 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1337 mm->free_area_cache = addr;
1338 mm->cached_hole_size = ~0UL;
1343 * This mmap-allocator allocates new areas top-down from below the
1344 * stack's low limit (the base):
1346 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1348 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1349 const unsigned long len, const unsigned long pgoff,
1350 const unsigned long flags)
1352 struct vm_area_struct *vma;
1353 struct mm_struct *mm = current->mm;
1354 unsigned long addr = addr0;
1356 /* requested length too big for entire address space */
1357 if (len > TASK_SIZE)
1360 if (flags & MAP_FIXED)
1363 /* requesting a specific address */
1365 addr = PAGE_ALIGN(addr);
1366 vma = find_vma(mm, addr);
1367 if (TASK_SIZE - len >= addr &&
1368 (!vma || addr + len <= vma->vm_start))
1372 /* check if free_area_cache is useful for us */
1373 if (len <= mm->cached_hole_size) {
1374 mm->cached_hole_size = 0;
1375 mm->free_area_cache = mm->mmap_base;
1378 /* either no address requested or can't fit in requested address hole */
1379 addr = mm->free_area_cache;
1381 /* make sure it can fit in the remaining address space */
1383 vma = find_vma(mm, addr-len);
1384 if (!vma || addr <= vma->vm_start)
1385 /* remember the address as a hint for next time */
1386 return (mm->free_area_cache = addr-len);
1389 if (mm->mmap_base < len)
1392 addr = mm->mmap_base-len;
1396 * Lookup failure means no vma is above this address,
1397 * else if new region fits below vma->vm_start,
1398 * return with success:
1400 vma = find_vma(mm, addr);
1401 if (!vma || addr+len <= vma->vm_start)
1402 /* remember the address as a hint for next time */
1403 return (mm->free_area_cache = addr);
1405 /* remember the largest hole we saw so far */
1406 if (addr + mm->cached_hole_size < vma->vm_start)
1407 mm->cached_hole_size = vma->vm_start - addr;
1409 /* try just below the current vma->vm_start */
1410 addr = vma->vm_start-len;
1411 } while (len < vma->vm_start);
1415 * A failed mmap() very likely causes application failure,
1416 * so fall back to the bottom-up function here. This scenario
1417 * can happen with large stack limits and large mmap()
1420 mm->cached_hole_size = ~0UL;
1421 mm->free_area_cache = TASK_UNMAPPED_BASE;
1422 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1424 * Restore the topdown base:
1426 mm->free_area_cache = mm->mmap_base;
1427 mm->cached_hole_size = ~0UL;
1433 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1436 * Is this a new hole at the highest possible address?
1438 if (addr > mm->free_area_cache)
1439 mm->free_area_cache = addr;
1441 /* dont allow allocations above current base */
1442 if (mm->free_area_cache > mm->mmap_base)
1443 mm->free_area_cache = mm->mmap_base;
1447 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1448 unsigned long pgoff, unsigned long flags)
1450 unsigned long (*get_area)(struct file *, unsigned long,
1451 unsigned long, unsigned long, unsigned long);
1453 get_area = current->mm->get_unmapped_area;
1454 if (file && file->f_op && file->f_op->get_unmapped_area)
1455 get_area = file->f_op->get_unmapped_area;
1456 addr = get_area(file, addr, len, pgoff, flags);
1457 if (IS_ERR_VALUE(addr))
1460 if (addr > TASK_SIZE - len)
1462 if (addr & ~PAGE_MASK)
1465 return arch_rebalance_pgtables(addr, len);
1468 EXPORT_SYMBOL(get_unmapped_area);
1470 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1471 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
1473 struct vm_area_struct *vma = NULL;
1476 /* Check the cache first. */
1477 /* (Cache hit rate is typically around 35%.) */
1478 vma = mm->mmap_cache;
1479 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1480 struct rb_node * rb_node;
1482 rb_node = mm->mm_rb.rb_node;
1486 struct vm_area_struct * vma_tmp;
1488 vma_tmp = rb_entry(rb_node,
1489 struct vm_area_struct, vm_rb);
1491 if (vma_tmp->vm_end > addr) {
1493 if (vma_tmp->vm_start <= addr)
1495 rb_node = rb_node->rb_left;
1497 rb_node = rb_node->rb_right;
1500 mm->mmap_cache = vma;
1506 EXPORT_SYMBOL(find_vma);
1508 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1509 struct vm_area_struct *
1510 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1511 struct vm_area_struct **pprev)
1513 struct vm_area_struct *vma = NULL, *prev = NULL;
1514 struct rb_node * rb_node;
1518 /* Guard against addr being lower than the first VMA */
1521 /* Go through the RB tree quickly. */
1522 rb_node = mm->mm_rb.rb_node;
1525 struct vm_area_struct *vma_tmp;
1526 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1528 if (addr < vma_tmp->vm_end) {
1529 rb_node = rb_node->rb_left;
1532 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1534 rb_node = rb_node->rb_right;
1540 return prev ? prev->vm_next : vma;
1544 * Verify that the stack growth is acceptable and
1545 * update accounting. This is shared with both the
1546 * grow-up and grow-down cases.
1548 static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow)
1550 struct mm_struct *mm = vma->vm_mm;
1551 struct rlimit *rlim = current->signal->rlim;
1552 unsigned long new_start;
1554 /* address space limit tests */
1555 if (!may_expand_vm(mm, grow))
1558 /* Stack limit test */
1559 if (size > rlim[RLIMIT_STACK].rlim_cur)
1562 /* mlock limit tests */
1563 if (vma->vm_flags & VM_LOCKED) {
1564 unsigned long locked;
1565 unsigned long limit;
1566 locked = mm->locked_vm + grow;
1567 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
1568 if (locked > limit && !capable(CAP_IPC_LOCK))
1572 /* Check to ensure the stack will not grow into a hugetlb-only region */
1573 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1575 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1579 * Overcommit.. This must be the final test, as it will
1580 * update security statistics.
1582 if (security_vm_enough_memory(grow))
1585 /* Ok, everything looks good - let it rip */
1586 mm->total_vm += grow;
1587 if (vma->vm_flags & VM_LOCKED)
1588 mm->locked_vm += grow;
1589 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1593 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1595 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1596 * vma is the last one with address > vma->vm_end. Have to extend vma.
1601 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1605 if (!(vma->vm_flags & VM_GROWSUP))
1609 * We must make sure the anon_vma is allocated
1610 * so that the anon_vma locking is not a noop.
1612 if (unlikely(anon_vma_prepare(vma)))
1617 * vma->vm_start/vm_end cannot change under us because the caller
1618 * is required to hold the mmap_sem in read mode. We need the
1619 * anon_vma lock to serialize against concurrent expand_stacks.
1620 * Also guard against wrapping around to address 0.
1622 if (address < PAGE_ALIGN(address+4))
1623 address = PAGE_ALIGN(address+4);
1625 anon_vma_unlock(vma);
1630 /* Somebody else might have raced and expanded it already */
1631 if (address > vma->vm_end) {
1632 unsigned long size, grow;
1634 size = address - vma->vm_start;
1635 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1637 error = acct_stack_growth(vma, size, grow);
1639 vma->vm_end = address;
1641 anon_vma_unlock(vma);
1644 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1647 * vma is the first one with address < vma->vm_start. Have to extend vma.
1649 static int expand_downwards(struct vm_area_struct *vma,
1650 unsigned long address)
1655 * We must make sure the anon_vma is allocated
1656 * so that the anon_vma locking is not a noop.
1658 if (unlikely(anon_vma_prepare(vma)))
1661 address &= PAGE_MASK;
1662 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
1669 * vma->vm_start/vm_end cannot change under us because the caller
1670 * is required to hold the mmap_sem in read mode. We need the
1671 * anon_vma lock to serialize against concurrent expand_stacks.
1674 /* Somebody else might have raced and expanded it already */
1675 if (address < vma->vm_start) {
1676 unsigned long size, grow;
1678 size = vma->vm_end - address;
1679 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1681 error = acct_stack_growth(vma, size, grow);
1683 vma->vm_start = address;
1684 vma->vm_pgoff -= grow;
1687 anon_vma_unlock(vma);
1691 int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address)
1693 return expand_downwards(vma, address);
1696 #ifdef CONFIG_STACK_GROWSUP
1697 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1699 return expand_upwards(vma, address);
1702 struct vm_area_struct *
1703 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1705 struct vm_area_struct *vma, *prev;
1708 vma = find_vma_prev(mm, addr, &prev);
1709 if (vma && (vma->vm_start <= addr))
1711 if (!prev || expand_stack(prev, addr))
1713 if (prev->vm_flags & VM_LOCKED) {
1714 if (mlock_vma_pages_range(prev, addr, prev->vm_end) < 0)
1715 return NULL; /* vma gone! */
1720 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1722 return expand_downwards(vma, address);
1725 struct vm_area_struct *
1726 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1728 struct vm_area_struct * vma;
1729 unsigned long start;
1732 vma = find_vma(mm,addr);
1735 if (vma->vm_start <= addr)
1737 if (!(vma->vm_flags & VM_GROWSDOWN))
1739 start = vma->vm_start;
1740 if (expand_stack(vma, addr))
1742 if (vma->vm_flags & VM_LOCKED) {
1743 if (mlock_vma_pages_range(vma, addr, start) < 0)
1744 return NULL; /* vma gone! */
1751 * Ok - we have the memory areas we should free on the vma list,
1752 * so release them, and do the vma updates.
1754 * Called with the mm semaphore held.
1756 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1758 /* Update high watermark before we lower total_vm */
1759 update_hiwater_vm(mm);
1761 long nrpages = vma_pages(vma);
1763 mm->total_vm -= nrpages;
1764 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1765 vma = remove_vma(vma);
1771 * Get rid of page table information in the indicated region.
1773 * Called with the mm semaphore held.
1775 static void unmap_region(struct mm_struct *mm,
1776 struct vm_area_struct *vma, struct vm_area_struct *prev,
1777 unsigned long start, unsigned long end)
1779 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1780 struct mmu_gather *tlb;
1781 unsigned long nr_accounted = 0;
1784 tlb = tlb_gather_mmu(mm, 0);
1785 update_hiwater_rss(mm);
1786 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1787 vm_unacct_memory(nr_accounted);
1788 free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
1789 next? next->vm_start: 0);
1790 tlb_finish_mmu(tlb, start, end);
1794 * Create a list of vma's touched by the unmap, removing them from the mm's
1795 * vma list as we go..
1798 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1799 struct vm_area_struct *prev, unsigned long end)
1801 struct vm_area_struct **insertion_point;
1802 struct vm_area_struct *tail_vma = NULL;
1805 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1807 rb_erase(&vma->vm_rb, &mm->mm_rb);
1811 } while (vma && vma->vm_start < end);
1812 *insertion_point = vma;
1813 tail_vma->vm_next = NULL;
1814 if (mm->unmap_area == arch_unmap_area)
1815 addr = prev ? prev->vm_end : mm->mmap_base;
1817 addr = vma ? vma->vm_start : mm->mmap_base;
1818 mm->unmap_area(mm, addr);
1819 mm->mmap_cache = NULL; /* Kill the cache. */
1823 * Split a vma into two pieces at address 'addr', a new vma is allocated
1824 * either for the first part or the tail.
1826 int split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1827 unsigned long addr, int new_below)
1829 struct mempolicy *pol;
1830 struct vm_area_struct *new;
1832 if (is_vm_hugetlb_page(vma) && (addr &
1833 ~(huge_page_mask(hstate_vma(vma)))))
1836 if (mm->map_count >= sysctl_max_map_count)
1839 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1843 /* most fields are the same, copy all, and then fixup */
1849 new->vm_start = addr;
1850 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1853 pol = mpol_dup(vma_policy(vma));
1855 kmem_cache_free(vm_area_cachep, new);
1856 return PTR_ERR(pol);
1858 vma_set_policy(new, pol);
1861 get_file(new->vm_file);
1862 if (vma->vm_flags & VM_EXECUTABLE)
1863 added_exe_file_vma(mm);
1866 if (new->vm_ops && new->vm_ops->open)
1867 new->vm_ops->open(new);
1870 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1871 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1873 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1878 /* Munmap is split into 2 main parts -- this part which finds
1879 * what needs doing, and the areas themselves, which do the
1880 * work. This now handles partial unmappings.
1881 * Jeremy Fitzhardinge <jeremy@goop.org>
1883 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
1886 struct vm_area_struct *vma, *prev, *last;
1888 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
1891 if ((len = PAGE_ALIGN(len)) == 0)
1894 /* Find the first overlapping VMA */
1895 vma = find_vma_prev(mm, start, &prev);
1898 /* we have start < vma->vm_end */
1900 /* if it doesn't overlap, we have nothing.. */
1902 if (vma->vm_start >= end)
1906 * If we need to split any vma, do it now to save pain later.
1908 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1909 * unmapped vm_area_struct will remain in use: so lower split_vma
1910 * places tmp vma above, and higher split_vma places tmp vma below.
1912 if (start > vma->vm_start) {
1913 int error = split_vma(mm, vma, start, 0);
1919 /* Does it split the last one? */
1920 last = find_vma(mm, end);
1921 if (last && end > last->vm_start) {
1922 int error = split_vma(mm, last, end, 1);
1926 vma = prev? prev->vm_next: mm->mmap;
1929 * unlock any mlock()ed ranges before detaching vmas
1931 if (mm->locked_vm) {
1932 struct vm_area_struct *tmp = vma;
1933 while (tmp && tmp->vm_start < end) {
1934 if (tmp->vm_flags & VM_LOCKED) {
1935 mm->locked_vm -= vma_pages(tmp);
1936 munlock_vma_pages_all(tmp);
1943 * Remove the vma's, and unmap the actual pages
1945 detach_vmas_to_be_unmapped(mm, vma, prev, end);
1946 unmap_region(mm, vma, prev, start, end);
1948 /* Fix up all other VM information */
1949 remove_vma_list(mm, vma);
1954 EXPORT_SYMBOL(do_munmap);
1956 asmlinkage long sys_munmap(unsigned long addr, size_t len)
1959 struct mm_struct *mm = current->mm;
1961 profile_munmap(addr);
1963 down_write(&mm->mmap_sem);
1964 ret = do_munmap(mm, addr, len);
1965 up_write(&mm->mmap_sem);
1969 static inline void verify_mm_writelocked(struct mm_struct *mm)
1971 #ifdef CONFIG_DEBUG_VM
1972 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
1974 up_read(&mm->mmap_sem);
1980 * this is really a simplified "do_mmap". it only handles
1981 * anonymous maps. eventually we may be able to do some
1982 * brk-specific accounting here.
1984 unsigned long do_brk(unsigned long addr, unsigned long len)
1986 struct mm_struct * mm = current->mm;
1987 struct vm_area_struct * vma, * prev;
1988 unsigned long flags;
1989 struct rb_node ** rb_link, * rb_parent;
1990 pgoff_t pgoff = addr >> PAGE_SHIFT;
1993 len = PAGE_ALIGN(len);
1997 if ((addr + len) > TASK_SIZE || (addr + len) < addr)
2000 if (is_hugepage_only_range(mm, addr, len))
2003 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
2007 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2009 error = arch_mmap_check(addr, len, flags);
2016 if (mm->def_flags & VM_LOCKED) {
2017 unsigned long locked, lock_limit;
2018 locked = len >> PAGE_SHIFT;
2019 locked += mm->locked_vm;
2020 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
2021 lock_limit >>= PAGE_SHIFT;
2022 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2027 * mm->mmap_sem is required to protect against another thread
2028 * changing the mappings in case we sleep.
2030 verify_mm_writelocked(mm);
2033 * Clear old maps. this also does some error checking for us
2036 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2037 if (vma && vma->vm_start < addr + len) {
2038 if (do_munmap(mm, addr, len))
2043 /* Check against address space limits *after* clearing old maps... */
2044 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2047 if (mm->map_count > sysctl_max_map_count)
2050 if (security_vm_enough_memory(len >> PAGE_SHIFT))
2053 /* Can we just expand an old private anonymous mapping? */
2054 vma = vma_merge(mm, prev, addr, addr + len, flags,
2055 NULL, NULL, pgoff, NULL);
2060 * create a vma struct for an anonymous mapping
2062 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2064 vm_unacct_memory(len >> PAGE_SHIFT);
2069 vma->vm_start = addr;
2070 vma->vm_end = addr + len;
2071 vma->vm_pgoff = pgoff;
2072 vma->vm_flags = flags;
2073 vma->vm_page_prot = vm_get_page_prot(flags);
2074 vma_link(mm, vma, prev, rb_link, rb_parent);
2076 mm->total_vm += len >> PAGE_SHIFT;
2077 if (flags & VM_LOCKED) {
2078 if (!mlock_vma_pages_range(vma, addr, addr + len))
2079 mm->locked_vm += (len >> PAGE_SHIFT);
2084 EXPORT_SYMBOL(do_brk);
2086 /* Release all mmaps. */
2087 void exit_mmap(struct mm_struct *mm)
2089 struct mmu_gather *tlb;
2090 struct vm_area_struct *vma;
2091 unsigned long nr_accounted = 0;
2094 /* mm's last user has gone, and its about to be pulled down */
2096 mmu_notifier_release(mm);
2098 if (mm->locked_vm) {
2101 if (vma->vm_flags & VM_LOCKED)
2102 munlock_vma_pages_all(vma);
2109 tlb = tlb_gather_mmu(mm, 1);
2110 /* Don't update_hiwater_rss(mm) here, do_exit already did */
2111 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2112 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2113 vm_unacct_memory(nr_accounted);
2114 free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0);
2115 tlb_finish_mmu(tlb, 0, end);
2118 * Walk the list again, actually closing and freeing it,
2119 * with preemption enabled, without holding any MM locks.
2122 vma = remove_vma(vma);
2124 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2127 /* Insert vm structure into process list sorted by address
2128 * and into the inode's i_mmap tree. If vm_file is non-NULL
2129 * then i_mmap_lock is taken here.
2131 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2133 struct vm_area_struct * __vma, * prev;
2134 struct rb_node ** rb_link, * rb_parent;
2137 * The vm_pgoff of a purely anonymous vma should be irrelevant
2138 * until its first write fault, when page's anon_vma and index
2139 * are set. But now set the vm_pgoff it will almost certainly
2140 * end up with (unless mremap moves it elsewhere before that
2141 * first wfault), so /proc/pid/maps tells a consistent story.
2143 * By setting it to reflect the virtual start address of the
2144 * vma, merges and splits can happen in a seamless way, just
2145 * using the existing file pgoff checks and manipulations.
2146 * Similarly in do_mmap_pgoff and in do_brk.
2148 if (!vma->vm_file) {
2149 BUG_ON(vma->anon_vma);
2150 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2152 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2153 if (__vma && __vma->vm_start < vma->vm_end)
2155 if ((vma->vm_flags & VM_ACCOUNT) &&
2156 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2158 vma_link(mm, vma, prev, rb_link, rb_parent);
2163 * Copy the vma structure to a new location in the same mm,
2164 * prior to moving page table entries, to effect an mremap move.
2166 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2167 unsigned long addr, unsigned long len, pgoff_t pgoff)
2169 struct vm_area_struct *vma = *vmap;
2170 unsigned long vma_start = vma->vm_start;
2171 struct mm_struct *mm = vma->vm_mm;
2172 struct vm_area_struct *new_vma, *prev;
2173 struct rb_node **rb_link, *rb_parent;
2174 struct mempolicy *pol;
2177 * If anonymous vma has not yet been faulted, update new pgoff
2178 * to match new location, to increase its chance of merging.
2180 if (!vma->vm_file && !vma->anon_vma)
2181 pgoff = addr >> PAGE_SHIFT;
2183 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2184 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2185 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2188 * Source vma may have been merged into new_vma
2190 if (vma_start >= new_vma->vm_start &&
2191 vma_start < new_vma->vm_end)
2194 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2197 pol = mpol_dup(vma_policy(vma));
2199 kmem_cache_free(vm_area_cachep, new_vma);
2202 vma_set_policy(new_vma, pol);
2203 new_vma->vm_start = addr;
2204 new_vma->vm_end = addr + len;
2205 new_vma->vm_pgoff = pgoff;
2206 if (new_vma->vm_file) {
2207 get_file(new_vma->vm_file);
2208 if (vma->vm_flags & VM_EXECUTABLE)
2209 added_exe_file_vma(mm);
2211 if (new_vma->vm_ops && new_vma->vm_ops->open)
2212 new_vma->vm_ops->open(new_vma);
2213 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2220 * Return true if the calling process may expand its vm space by the passed
2223 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2225 unsigned long cur = mm->total_vm; /* pages */
2228 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
2230 if (cur + npages > lim)
2236 static int special_mapping_fault(struct vm_area_struct *vma,
2237 struct vm_fault *vmf)
2240 struct page **pages;
2243 * special mappings have no vm_file, and in that case, the mm
2244 * uses vm_pgoff internally. So we have to subtract it from here.
2245 * We are allowed to do this because we are the mm; do not copy
2246 * this code into drivers!
2248 pgoff = vmf->pgoff - vma->vm_pgoff;
2250 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2254 struct page *page = *pages;
2260 return VM_FAULT_SIGBUS;
2264 * Having a close hook prevents vma merging regardless of flags.
2266 static void special_mapping_close(struct vm_area_struct *vma)
2270 static struct vm_operations_struct special_mapping_vmops = {
2271 .close = special_mapping_close,
2272 .fault = special_mapping_fault,
2276 * Called with mm->mmap_sem held for writing.
2277 * Insert a new vma covering the given region, with the given flags.
2278 * Its pages are supplied by the given array of struct page *.
2279 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2280 * The region past the last page supplied will always produce SIGBUS.
2281 * The array pointer and the pages it points to are assumed to stay alive
2282 * for as long as this mapping might exist.
2284 int install_special_mapping(struct mm_struct *mm,
2285 unsigned long addr, unsigned long len,
2286 unsigned long vm_flags, struct page **pages)
2288 struct vm_area_struct *vma;
2290 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2291 if (unlikely(vma == NULL))
2295 vma->vm_start = addr;
2296 vma->vm_end = addr + len;
2298 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2299 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2301 vma->vm_ops = &special_mapping_vmops;
2302 vma->vm_private_data = pages;
2304 if (unlikely(insert_vm_struct(mm, vma))) {
2305 kmem_cache_free(vm_area_cachep, vma);
2309 mm->total_vm += len >> PAGE_SHIFT;
2314 static DEFINE_MUTEX(mm_all_locks_mutex);
2316 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2318 if (!test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2320 * The LSB of head.next can't change from under us
2321 * because we hold the mm_all_locks_mutex.
2323 spin_lock_nest_lock(&anon_vma->lock, &mm->mmap_sem);
2325 * We can safely modify head.next after taking the
2326 * anon_vma->lock. If some other vma in this mm shares
2327 * the same anon_vma we won't take it again.
2329 * No need of atomic instructions here, head.next
2330 * can't change from under us thanks to the
2333 if (__test_and_set_bit(0, (unsigned long *)
2334 &anon_vma->head.next))
2339 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2341 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2343 * AS_MM_ALL_LOCKS can't change from under us because
2344 * we hold the mm_all_locks_mutex.
2346 * Operations on ->flags have to be atomic because
2347 * even if AS_MM_ALL_LOCKS is stable thanks to the
2348 * mm_all_locks_mutex, there may be other cpus
2349 * changing other bitflags in parallel to us.
2351 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2353 spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem);
2358 * This operation locks against the VM for all pte/vma/mm related
2359 * operations that could ever happen on a certain mm. This includes
2360 * vmtruncate, try_to_unmap, and all page faults.
2362 * The caller must take the mmap_sem in write mode before calling
2363 * mm_take_all_locks(). The caller isn't allowed to release the
2364 * mmap_sem until mm_drop_all_locks() returns.
2366 * mmap_sem in write mode is required in order to block all operations
2367 * that could modify pagetables and free pages without need of
2368 * altering the vma layout (for example populate_range() with
2369 * nonlinear vmas). It's also needed in write mode to avoid new
2370 * anon_vmas to be associated with existing vmas.
2372 * A single task can't take more than one mm_take_all_locks() in a row
2373 * or it would deadlock.
2375 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2376 * mapping->flags avoid to take the same lock twice, if more than one
2377 * vma in this mm is backed by the same anon_vma or address_space.
2379 * We can take all the locks in random order because the VM code
2380 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2381 * takes more than one of them in a row. Secondly we're protected
2382 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2384 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2385 * that may have to take thousand of locks.
2387 * mm_take_all_locks() can fail if it's interrupted by signals.
2389 int mm_take_all_locks(struct mm_struct *mm)
2391 struct vm_area_struct *vma;
2394 BUG_ON(down_read_trylock(&mm->mmap_sem));
2396 mutex_lock(&mm_all_locks_mutex);
2398 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2399 if (signal_pending(current))
2401 if (vma->vm_file && vma->vm_file->f_mapping)
2402 vm_lock_mapping(mm, vma->vm_file->f_mapping);
2405 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2406 if (signal_pending(current))
2409 vm_lock_anon_vma(mm, vma->anon_vma);
2416 mm_drop_all_locks(mm);
2421 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2423 if (test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2425 * The LSB of head.next can't change to 0 from under
2426 * us because we hold the mm_all_locks_mutex.
2428 * We must however clear the bitflag before unlocking
2429 * the vma so the users using the anon_vma->head will
2430 * never see our bitflag.
2432 * No need of atomic instructions here, head.next
2433 * can't change from under us until we release the
2436 if (!__test_and_clear_bit(0, (unsigned long *)
2437 &anon_vma->head.next))
2439 spin_unlock(&anon_vma->lock);
2443 static void vm_unlock_mapping(struct address_space *mapping)
2445 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2447 * AS_MM_ALL_LOCKS can't change to 0 from under us
2448 * because we hold the mm_all_locks_mutex.
2450 spin_unlock(&mapping->i_mmap_lock);
2451 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2458 * The mmap_sem cannot be released by the caller until
2459 * mm_drop_all_locks() returns.
2461 void mm_drop_all_locks(struct mm_struct *mm)
2463 struct vm_area_struct *vma;
2465 BUG_ON(down_read_trylock(&mm->mmap_sem));
2466 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2468 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2470 vm_unlock_anon_vma(vma->anon_vma);
2471 if (vma->vm_file && vma->vm_file->f_mapping)
2472 vm_unlock_mapping(vma->vm_file->f_mapping);
2475 mutex_unlock(&mm_all_locks_mutex);