6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
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/hugetlb.h>
24 #include <linux/profile.h>
25 #include <linux/module.h>
26 #include <linux/mount.h>
27 #include <linux/mempolicy.h>
28 #include <linux/rmap.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/perf_event.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 struct percpu_counter vm_committed_as;
89 int heap_stack_gap __read_mostly = 1;
92 * Check that a process has enough memory to allocate a new virtual
93 * mapping. 0 means there is enough memory for the allocation to
94 * succeed and -ENOMEM implies there is not.
96 * We currently support three overcommit policies, which are set via the
97 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
99 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
100 * Additional code 2002 Jul 20 by Robert Love.
102 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
104 * Note this is a helper function intended to be used by LSMs which
105 * wish to use this logic.
107 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
109 unsigned long free, allowed;
111 vm_acct_memory(pages);
114 * Sometimes we want to use more memory than we have
116 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
119 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
122 free = global_page_state(NR_FILE_PAGES);
123 free += nr_swap_pages;
126 * Any slabs which are created with the
127 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
128 * which are reclaimable, under pressure. The dentry
129 * cache and most inode caches should fall into this
131 free += global_page_state(NR_SLAB_RECLAIMABLE);
134 * Leave the last 3% for root
143 * nr_free_pages() is very expensive on large systems,
144 * only call if we're about to fail.
149 * Leave reserved pages. The pages are not for anonymous pages.
151 if (n <= totalreserve_pages)
154 n -= totalreserve_pages;
157 * Leave the last 3% for root
169 allowed = (totalram_pages - hugetlb_total_pages())
170 * sysctl_overcommit_ratio / 100;
172 * Leave the last 3% for root
175 allowed -= allowed / 32;
176 allowed += total_swap_pages;
178 /* Don't let a single process grow too big:
179 leave 3% of the size of this process for other processes */
181 allowed -= mm->total_vm / 32;
183 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
186 vm_unacct_memory(pages);
192 * Requires inode->i_mapping->i_mmap_lock
194 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
195 struct file *file, struct address_space *mapping)
197 if (vma->vm_flags & VM_DENYWRITE)
198 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
199 if (vma->vm_flags & VM_SHARED)
200 mapping->i_mmap_writable--;
202 flush_dcache_mmap_lock(mapping);
203 if (unlikely(vma->vm_flags & VM_NONLINEAR))
204 list_del_init(&vma->shared.vm_set.list);
206 vma_prio_tree_remove(vma, &mapping->i_mmap);
207 flush_dcache_mmap_unlock(mapping);
211 * Unlink a file-based vm structure from its prio_tree, to hide
212 * vma from rmap and vmtruncate before freeing its page tables.
214 void unlink_file_vma(struct vm_area_struct *vma)
216 struct file *file = vma->vm_file;
219 struct address_space *mapping = file->f_mapping;
220 spin_lock(&mapping->i_mmap_lock);
221 __remove_shared_vm_struct(vma, file, mapping);
222 spin_unlock(&mapping->i_mmap_lock);
227 * Close a vm structure and free it, returning the next.
229 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
231 struct vm_area_struct *next = vma->vm_next;
234 if (vma->vm_ops && vma->vm_ops->close)
235 vma->vm_ops->close(vma);
238 if (vma->vm_flags & VM_EXECUTABLE)
239 removed_exe_file_vma(vma->vm_mm);
241 mpol_put(vma_policy(vma));
242 kmem_cache_free(vm_area_cachep, vma);
246 SYSCALL_DEFINE1(brk, unsigned long, brk)
248 unsigned long rlim, retval;
249 unsigned long newbrk, oldbrk;
250 struct mm_struct *mm = current->mm;
251 unsigned long min_brk;
253 down_write(&mm->mmap_sem);
255 #ifdef CONFIG_COMPAT_BRK
256 min_brk = mm->end_code;
258 min_brk = mm->start_brk;
264 * Check against rlimit here. If this check is done later after the test
265 * of oldbrk with newbrk then it can escape the test and let the data
266 * segment grow beyond its set limit the in case where the limit is
267 * not page aligned -Ram Gupta
269 rlim = rlimit(RLIMIT_DATA);
270 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
271 (mm->end_data - mm->start_data) > rlim)
274 newbrk = PAGE_ALIGN(brk);
275 oldbrk = PAGE_ALIGN(mm->brk);
276 if (oldbrk == newbrk)
279 /* Always allow shrinking brk. */
280 if (brk <= mm->brk) {
281 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
286 /* Check against existing mmap mappings. */
287 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
290 /* Ok, looks good - let it rip. */
291 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
297 up_write(&mm->mmap_sem);
302 static int browse_rb(struct rb_root *root)
305 struct rb_node *nd, *pn = NULL;
306 unsigned long prev = 0, pend = 0;
308 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
309 struct vm_area_struct *vma;
310 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
311 if (vma->vm_start < prev)
312 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
313 if (vma->vm_start < pend)
314 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
315 if (vma->vm_start > vma->vm_end)
316 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
319 prev = vma->vm_start;
323 for (nd = pn; nd; nd = rb_prev(nd)) {
327 printk("backwards %d, forwards %d\n", j, i), i = 0;
331 void validate_mm(struct mm_struct *mm)
335 struct vm_area_struct *tmp = mm->mmap;
340 if (i != mm->map_count)
341 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
342 i = browse_rb(&mm->mm_rb);
343 if (i != mm->map_count)
344 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
348 #define validate_mm(mm) do { } while (0)
351 static struct vm_area_struct *
352 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
353 struct vm_area_struct **pprev, struct rb_node ***rb_link,
354 struct rb_node ** rb_parent)
356 struct vm_area_struct * vma;
357 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
359 __rb_link = &mm->mm_rb.rb_node;
360 rb_prev = __rb_parent = NULL;
364 struct vm_area_struct *vma_tmp;
366 __rb_parent = *__rb_link;
367 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
369 if (vma_tmp->vm_end > addr) {
371 if (vma_tmp->vm_start <= addr)
373 __rb_link = &__rb_parent->rb_left;
375 rb_prev = __rb_parent;
376 __rb_link = &__rb_parent->rb_right;
382 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
383 *rb_link = __rb_link;
384 *rb_parent = __rb_parent;
389 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
390 struct vm_area_struct *prev, struct rb_node *rb_parent)
393 vma->vm_next = prev->vm_next;
398 vma->vm_next = rb_entry(rb_parent,
399 struct vm_area_struct, vm_rb);
405 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
406 struct rb_node **rb_link, struct rb_node *rb_parent)
408 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
409 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
412 static void __vma_link_file(struct vm_area_struct *vma)
418 struct address_space *mapping = file->f_mapping;
420 if (vma->vm_flags & VM_DENYWRITE)
421 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
422 if (vma->vm_flags & VM_SHARED)
423 mapping->i_mmap_writable++;
425 flush_dcache_mmap_lock(mapping);
426 if (unlikely(vma->vm_flags & VM_NONLINEAR))
427 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
429 vma_prio_tree_insert(vma, &mapping->i_mmap);
430 flush_dcache_mmap_unlock(mapping);
435 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
436 struct vm_area_struct *prev, struct rb_node **rb_link,
437 struct rb_node *rb_parent)
439 __vma_link_list(mm, vma, prev, rb_parent);
440 __vma_link_rb(mm, vma, rb_link, rb_parent);
443 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
444 struct vm_area_struct *prev, struct rb_node **rb_link,
445 struct rb_node *rb_parent)
447 struct address_space *mapping = NULL;
450 mapping = vma->vm_file->f_mapping;
453 spin_lock(&mapping->i_mmap_lock);
454 vma->vm_truncate_count = mapping->truncate_count;
458 __vma_link(mm, vma, prev, rb_link, rb_parent);
459 __vma_link_file(vma);
461 anon_vma_unlock(vma);
463 spin_unlock(&mapping->i_mmap_lock);
470 * Helper for vma_adjust in the split_vma insert case:
471 * insert vm structure into list and rbtree and anon_vma,
472 * but it has already been inserted into prio_tree earlier.
474 static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
476 struct vm_area_struct *__vma, *prev;
477 struct rb_node **rb_link, *rb_parent;
479 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
480 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
481 __vma_link(mm, vma, prev, rb_link, rb_parent);
486 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
487 struct vm_area_struct *prev)
489 prev->vm_next = vma->vm_next;
490 rb_erase(&vma->vm_rb, &mm->mm_rb);
491 if (mm->mmap_cache == vma)
492 mm->mmap_cache = prev;
496 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
497 * is already present in an i_mmap tree without adjusting the tree.
498 * The following helper function should be used when such adjustments
499 * are necessary. The "insert" vma (if any) is to be inserted
500 * before we drop the necessary locks.
502 int vma_adjust(struct vm_area_struct *vma, unsigned long start,
503 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
505 struct mm_struct *mm = vma->vm_mm;
506 struct vm_area_struct *next = vma->vm_next;
507 struct vm_area_struct *importer = NULL;
508 struct address_space *mapping = NULL;
509 struct prio_tree_root *root = NULL;
510 struct file *file = vma->vm_file;
511 struct anon_vma *anon_vma = NULL;
512 long adjust_next = 0;
515 if (next && !insert) {
516 if (end >= next->vm_end) {
518 * vma expands, overlapping all the next, and
519 * perhaps the one after too (mprotect case 6).
521 again: remove_next = 1 + (end > next->vm_end);
523 anon_vma = next->anon_vma;
525 } else if (end > next->vm_start) {
527 * vma expands, overlapping part of the next:
528 * mprotect case 5 shifting the boundary up.
530 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
531 anon_vma = next->anon_vma;
533 } else if (end < vma->vm_end) {
535 * vma shrinks, and !insert tells it's not
536 * split_vma inserting another: so it must be
537 * mprotect case 4 shifting the boundary down.
539 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
540 anon_vma = next->anon_vma;
546 * When changing only vma->vm_end, we don't really need anon_vma lock.
548 if (vma->anon_vma && (insert || importer || start != vma->vm_start))
549 anon_vma = vma->anon_vma;
552 * Easily overlooked: when mprotect shifts the boundary,
553 * make sure the expanding vma has anon_vma set if the
554 * shrinking vma had, to cover any anon pages imported.
556 if (importer && !importer->anon_vma) {
557 /* Block reverse map lookups until things are set up. */
558 if (anon_vma_clone(importer, vma)) {
561 importer->anon_vma = anon_vma;
566 mapping = file->f_mapping;
567 if (!(vma->vm_flags & VM_NONLINEAR))
568 root = &mapping->i_mmap;
569 spin_lock(&mapping->i_mmap_lock);
571 vma->vm_truncate_count != next->vm_truncate_count) {
573 * unmap_mapping_range might be in progress:
574 * ensure that the expanding vma is rescanned.
576 importer->vm_truncate_count = 0;
579 insert->vm_truncate_count = vma->vm_truncate_count;
581 * Put into prio_tree now, so instantiated pages
582 * are visible to arm/parisc __flush_dcache_page
583 * throughout; but we cannot insert into address
584 * space until vma start or end is updated.
586 __vma_link_file(insert);
591 flush_dcache_mmap_lock(mapping);
592 vma_prio_tree_remove(vma, root);
594 vma_prio_tree_remove(next, root);
597 vma->vm_start = start;
599 vma->vm_pgoff = pgoff;
601 next->vm_start += adjust_next << PAGE_SHIFT;
602 next->vm_pgoff += adjust_next;
607 vma_prio_tree_insert(next, root);
608 vma_prio_tree_insert(vma, root);
609 flush_dcache_mmap_unlock(mapping);
614 * vma_merge has merged next into vma, and needs
615 * us to remove next before dropping the locks.
617 __vma_unlink(mm, next, vma);
619 __remove_shared_vm_struct(next, file, mapping);
622 * split_vma has split insert from vma, and needs
623 * us to insert it before dropping the locks
624 * (it may either follow vma or precede it).
626 __insert_vm_struct(mm, insert);
630 spin_unlock(&mapping->i_mmap_lock);
635 if (next->vm_flags & VM_EXECUTABLE)
636 removed_exe_file_vma(mm);
639 anon_vma_merge(vma, next);
641 mpol_put(vma_policy(next));
642 kmem_cache_free(vm_area_cachep, next);
644 * In mprotect's case 6 (see comments on vma_merge),
645 * we must remove another next too. It would clutter
646 * up the code too much to do both in one go.
648 if (remove_next == 2) {
660 * If the vma has a ->close operation then the driver probably needs to release
661 * per-vma resources, so we don't attempt to merge those.
663 static inline int is_mergeable_vma(struct vm_area_struct *vma,
664 struct file *file, unsigned long vm_flags)
666 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
667 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
669 if (vma->vm_file != file)
671 if (vma->vm_ops && vma->vm_ops->close)
676 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
677 struct anon_vma *anon_vma2)
679 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
683 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
684 * in front of (at a lower virtual address and file offset than) the vma.
686 * We cannot merge two vmas if they have differently assigned (non-NULL)
687 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
689 * We don't check here for the merged mmap wrapping around the end of pagecache
690 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
691 * wrap, nor mmaps which cover the final page at index -1UL.
694 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
695 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
697 if (is_mergeable_vma(vma, file, vm_flags) &&
698 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
699 if (vma->vm_pgoff == vm_pgoff)
706 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
707 * beyond (at a higher virtual address and file offset than) the vma.
709 * We cannot merge two vmas if they have differently assigned (non-NULL)
710 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
713 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
714 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
716 if (is_mergeable_vma(vma, file, vm_flags) &&
717 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
719 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
720 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
727 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
728 * whether that can be merged with its predecessor or its successor.
729 * Or both (it neatly fills a hole).
731 * In most cases - when called for mmap, brk or mremap - [addr,end) is
732 * certain not to be mapped by the time vma_merge is called; but when
733 * called for mprotect, it is certain to be already mapped (either at
734 * an offset within prev, or at the start of next), and the flags of
735 * this area are about to be changed to vm_flags - and the no-change
736 * case has already been eliminated.
738 * The following mprotect cases have to be considered, where AAAA is
739 * the area passed down from mprotect_fixup, never extending beyond one
740 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
742 * AAAA AAAA AAAA AAAA
743 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
744 * cannot merge might become might become might become
745 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
746 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
747 * mremap move: PPPPNNNNNNNN 8
749 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
750 * might become case 1 below case 2 below case 3 below
752 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
753 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
755 struct vm_area_struct *vma_merge(struct mm_struct *mm,
756 struct vm_area_struct *prev, unsigned long addr,
757 unsigned long end, unsigned long vm_flags,
758 struct anon_vma *anon_vma, struct file *file,
759 pgoff_t pgoff, struct mempolicy *policy)
761 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
762 struct vm_area_struct *area, *next;
766 * We later require that vma->vm_flags == vm_flags,
767 * so this tests vma->vm_flags & VM_SPECIAL, too.
769 if (vm_flags & VM_SPECIAL)
773 next = prev->vm_next;
777 if (next && next->vm_end == end) /* cases 6, 7, 8 */
778 next = next->vm_next;
781 * Can it merge with the predecessor?
783 if (prev && prev->vm_end == addr &&
784 mpol_equal(vma_policy(prev), policy) &&
785 can_vma_merge_after(prev, vm_flags,
786 anon_vma, file, pgoff)) {
788 * OK, it can. Can we now merge in the successor as well?
790 if (next && end == next->vm_start &&
791 mpol_equal(policy, vma_policy(next)) &&
792 can_vma_merge_before(next, vm_flags,
793 anon_vma, file, pgoff+pglen) &&
794 is_mergeable_anon_vma(prev->anon_vma,
797 err = vma_adjust(prev, prev->vm_start,
798 next->vm_end, prev->vm_pgoff, NULL);
799 } else /* cases 2, 5, 7 */
800 err = vma_adjust(prev, prev->vm_start,
801 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 err = vma_adjust(prev, prev->vm_start,
816 addr, prev->vm_pgoff, NULL);
817 else /* cases 3, 8 */
818 err = vma_adjust(area, addr, next->vm_end,
819 next->vm_pgoff - pglen, NULL);
829 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
830 * neighbouring vmas for a suitable anon_vma, before it goes off
831 * to allocate a new anon_vma. It checks because a repetitive
832 * sequence of mprotects and faults may otherwise lead to distinct
833 * anon_vmas being allocated, preventing vma merge in subsequent
836 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
838 struct vm_area_struct *near;
839 unsigned long vm_flags;
846 * Since only mprotect tries to remerge vmas, match flags
847 * which might be mprotected into each other later on.
848 * Neither mlock nor madvise tries to remerge at present,
849 * so leave their flags as obstructing a merge.
851 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
852 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
854 if (near->anon_vma && vma->vm_end == near->vm_start &&
855 mpol_equal(vma_policy(vma), vma_policy(near)) &&
856 can_vma_merge_before(near, vm_flags,
857 NULL, vma->vm_file, vma->vm_pgoff +
858 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT)))
859 return near->anon_vma;
862 * It is potentially slow to have to call find_vma_prev here.
863 * But it's only on the first write fault on the vma, not
864 * every time, and we could devise a way to avoid it later
865 * (e.g. stash info in next's anon_vma_node when assigning
866 * an anon_vma, or when trying vma_merge). Another time.
868 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
872 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC);
873 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC);
875 if (near->anon_vma && near->vm_end == vma->vm_start &&
876 mpol_equal(vma_policy(near), vma_policy(vma)) &&
877 can_vma_merge_after(near, vm_flags,
878 NULL, vma->vm_file, vma->vm_pgoff))
879 return near->anon_vma;
882 * There's no absolute need to look only at touching neighbours:
883 * we could search further afield for "compatible" anon_vmas.
884 * But it would probably just be a waste of time searching,
885 * or lead to too many vmas hanging off the same anon_vma.
886 * We're trying to allow mprotect remerging later on,
887 * not trying to minimize memory used for anon_vmas.
892 #ifdef CONFIG_PROC_FS
893 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
894 struct file *file, long pages)
896 const unsigned long stack_flags
897 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
900 mm->shared_vm += pages;
901 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
902 mm->exec_vm += pages;
903 } else if (flags & stack_flags)
904 mm->stack_vm += pages;
905 if (flags & (VM_RESERVED|VM_IO))
906 mm->reserved_vm += pages;
908 #endif /* CONFIG_PROC_FS */
911 * The caller must hold down_write(¤t->mm->mmap_sem).
914 unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
915 unsigned long len, unsigned long prot,
916 unsigned long flags, unsigned long pgoff)
918 struct mm_struct * mm = current->mm;
920 unsigned int vm_flags;
922 unsigned long reqprot = prot;
925 * Does the application expect PROT_READ to imply PROT_EXEC?
927 * (the exception is when the underlying filesystem is noexec
928 * mounted, in which case we dont add PROT_EXEC.)
930 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
931 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
937 if (!(flags & MAP_FIXED))
938 addr = round_hint_to_min(addr);
940 /* Careful about overflows.. */
941 len = PAGE_ALIGN(len);
945 /* offset overflow? */
946 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
949 /* Too many mappings? */
950 if (mm->map_count > sysctl_max_map_count)
953 /* Obtain the address to map to. we verify (or select) it and ensure
954 * that it represents a valid section of the address space.
956 addr = get_unmapped_area(file, addr, len, pgoff, flags);
957 if (addr & ~PAGE_MASK)
960 /* Do simple checking here so the lower-level routines won't have
961 * to. we assume access permissions have been handled by the open
962 * of the memory object, so we don't do any here.
964 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
965 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
967 if (flags & MAP_LOCKED)
971 /* mlock MCL_FUTURE? */
972 if (vm_flags & VM_LOCKED) {
973 unsigned long locked, lock_limit;
974 locked = len >> PAGE_SHIFT;
975 locked += mm->locked_vm;
976 lock_limit = rlimit(RLIMIT_MEMLOCK);
977 lock_limit >>= PAGE_SHIFT;
978 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
982 inode = file ? file->f_path.dentry->d_inode : NULL;
985 switch (flags & MAP_TYPE) {
987 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
991 * Make sure we don't allow writing to an append-only
994 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
998 * Make sure there are no mandatory locks on the file.
1000 if (locks_verify_locked(inode))
1003 vm_flags |= VM_SHARED | VM_MAYSHARE;
1004 if (!(file->f_mode & FMODE_WRITE))
1005 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1009 if (!(file->f_mode & FMODE_READ))
1011 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1012 if (vm_flags & VM_EXEC)
1014 vm_flags &= ~VM_MAYEXEC;
1017 if (!file->f_op || !file->f_op->mmap)
1025 switch (flags & MAP_TYPE) {
1031 vm_flags |= VM_SHARED | VM_MAYSHARE;
1035 * Set pgoff according to addr for anon_vma.
1037 pgoff = addr >> PAGE_SHIFT;
1044 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1048 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
1050 EXPORT_SYMBOL(do_mmap_pgoff);
1052 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1053 unsigned long, prot, unsigned long, flags,
1054 unsigned long, fd, unsigned long, pgoff)
1056 struct file *file = NULL;
1057 unsigned long retval = -EBADF;
1059 if (!(flags & MAP_ANONYMOUS)) {
1060 if (unlikely(flags & MAP_HUGETLB))
1065 } else if (flags & MAP_HUGETLB) {
1066 struct user_struct *user = NULL;
1068 * VM_NORESERVE is used because the reservations will be
1069 * taken when vm_ops->mmap() is called
1070 * A dummy user value is used because we are not locking
1071 * memory so no accounting is necessary
1073 len = ALIGN(len, huge_page_size(&default_hstate));
1074 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1075 &user, HUGETLB_ANONHUGE_INODE);
1077 return PTR_ERR(file);
1080 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1082 down_write(¤t->mm->mmap_sem);
1083 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1084 up_write(¤t->mm->mmap_sem);
1093 * Some shared mappigns will want the pages marked read-only
1094 * to track write events. If so, we'll downgrade vm_page_prot
1095 * to the private version (using protection_map[] without the
1098 int vma_wants_writenotify(struct vm_area_struct *vma)
1100 unsigned int vm_flags = vma->vm_flags;
1102 /* If it was private or non-writable, the write bit is already clear */
1103 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1106 /* The backer wishes to know when pages are first written to? */
1107 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1110 /* The open routine did something to the protections already? */
1111 if (pgprot_val(vma->vm_page_prot) !=
1112 pgprot_val(vm_get_page_prot(vm_flags)))
1115 /* Specialty mapping? */
1116 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1119 /* Can the mapping track the dirty pages? */
1120 return vma->vm_file && vma->vm_file->f_mapping &&
1121 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1125 * We account for memory if it's a private writeable mapping,
1126 * not hugepages and VM_NORESERVE wasn't set.
1128 static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
1131 * hugetlb has its own accounting separate from the core VM
1132 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1134 if (file && is_file_hugepages(file))
1137 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1140 unsigned long mmap_region(struct file *file, unsigned long addr,
1141 unsigned long len, unsigned long flags,
1142 unsigned int vm_flags, unsigned long pgoff)
1144 struct mm_struct *mm = current->mm;
1145 struct vm_area_struct *vma, *prev;
1146 int correct_wcount = 0;
1148 struct rb_node **rb_link, *rb_parent;
1149 unsigned long charged = 0;
1150 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1152 /* Clear old maps */
1155 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1156 if (vma && vma->vm_start < addr + len) {
1157 if (do_munmap(mm, addr, len))
1162 /* Check against address space limit. */
1163 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1167 * Set 'VM_NORESERVE' if we should not account for the
1168 * memory use of this mapping.
1170 if ((flags & MAP_NORESERVE)) {
1171 /* We honor MAP_NORESERVE if allowed to overcommit */
1172 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1173 vm_flags |= VM_NORESERVE;
1175 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1176 if (file && is_file_hugepages(file))
1177 vm_flags |= VM_NORESERVE;
1181 * Private writable mapping: check memory availability
1183 if (accountable_mapping(file, vm_flags)) {
1184 charged = len >> PAGE_SHIFT;
1185 if (security_vm_enough_memory(charged))
1187 vm_flags |= VM_ACCOUNT;
1191 * Can we just expand an old mapping?
1193 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1198 * Determine the object being mapped and call the appropriate
1199 * specific mapper. the address has already been validated, but
1200 * not unmapped, but the maps are removed from the list.
1202 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1209 vma->vm_start = addr;
1210 vma->vm_end = addr + len;
1211 vma->vm_flags = vm_flags;
1212 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1213 vma->vm_pgoff = pgoff;
1214 INIT_LIST_HEAD(&vma->anon_vma_chain);
1218 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1220 if (vm_flags & VM_DENYWRITE) {
1221 error = deny_write_access(file);
1226 vma->vm_file = file;
1228 error = file->f_op->mmap(file, vma);
1230 goto unmap_and_free_vma;
1231 if (vm_flags & VM_EXECUTABLE)
1232 added_exe_file_vma(mm);
1234 /* Can addr have changed??
1236 * Answer: Yes, several device drivers can do it in their
1237 * f_op->mmap method. -DaveM
1239 addr = vma->vm_start;
1240 pgoff = vma->vm_pgoff;
1241 vm_flags = vma->vm_flags;
1242 } else if (vm_flags & VM_SHARED) {
1243 error = shmem_zero_setup(vma);
1248 if (vma_wants_writenotify(vma)) {
1249 pgprot_t pprot = vma->vm_page_prot;
1251 /* Can vma->vm_page_prot have changed??
1253 * Answer: Yes, drivers may have changed it in their
1254 * f_op->mmap method.
1256 * Ensures that vmas marked as uncached stay that way.
1258 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1259 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1260 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1263 vma_link(mm, vma, prev, rb_link, rb_parent);
1264 file = vma->vm_file;
1266 /* Once vma denies write, undo our temporary denial count */
1268 atomic_inc(&inode->i_writecount);
1270 perf_event_mmap(vma);
1272 mm->total_vm += len >> PAGE_SHIFT;
1273 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1274 if (vm_flags & VM_LOCKED) {
1275 if (!mlock_vma_pages_range(vma, addr, addr + len))
1276 mm->locked_vm += (len >> PAGE_SHIFT);
1277 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1278 make_pages_present(addr, addr + len);
1283 atomic_inc(&inode->i_writecount);
1284 vma->vm_file = NULL;
1287 /* Undo any partial mapping done by a device driver. */
1288 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1291 kmem_cache_free(vm_area_cachep, vma);
1294 vm_unacct_memory(charged);
1298 /* Get an address range which is currently unmapped.
1299 * For shmat() with addr=0.
1301 * Ugly calling convention alert:
1302 * Return value with the low bits set means error value,
1304 * if (ret & ~PAGE_MASK)
1307 * This function "knows" that -ENOMEM has the bits set.
1309 #ifndef HAVE_ARCH_UNMAPPED_AREA
1311 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1312 unsigned long len, unsigned long pgoff, unsigned long flags)
1314 struct mm_struct *mm = current->mm;
1315 struct vm_area_struct *vma;
1316 unsigned long start_addr;
1318 if (len > TASK_SIZE)
1321 if (flags & MAP_FIXED)
1325 addr = PAGE_ALIGN(addr);
1326 vma = find_vma(mm, addr);
1327 if (TASK_SIZE - len >= addr &&
1328 (!vma || addr + len <= vma->vm_start))
1331 if (len > mm->cached_hole_size) {
1332 start_addr = addr = mm->free_area_cache;
1334 start_addr = addr = TASK_UNMAPPED_BASE;
1335 mm->cached_hole_size = 0;
1339 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1340 unsigned long guard;
1342 /* At this point: (!vma || addr < vma->vm_end). */
1343 if (TASK_SIZE - len < addr) {
1345 * Start a new search - just in case we missed
1348 if (start_addr != TASK_UNMAPPED_BASE) {
1349 addr = TASK_UNMAPPED_BASE;
1351 mm->cached_hole_size = 0;
1359 if (vma->vm_flags & VM_GROWSDOWN)
1360 guard = min(TASK_SIZE - (addr + len),
1361 (unsigned long)heap_stack_gap << PAGE_SHIFT);
1362 if (addr + len + guard <= vma->vm_start) {
1365 * Remember the place where we stopped the search:
1367 mm->free_area_cache = addr + len;
1370 if (addr + guard + mm->cached_hole_size < vma->vm_start)
1371 mm->cached_hole_size = vma->vm_start - (addr + guard);
1378 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1381 * Is this a new hole at the lowest possible address?
1383 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1384 mm->free_area_cache = addr;
1385 mm->cached_hole_size = ~0UL;
1390 * This mmap-allocator allocates new areas top-down from below the
1391 * stack's low limit (the base):
1393 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1395 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1396 const unsigned long len, const unsigned long pgoff,
1397 const unsigned long flags)
1399 struct vm_area_struct *vma;
1400 struct mm_struct *mm = current->mm;
1401 unsigned long addr = addr0;
1403 /* requested length too big for entire address space */
1404 if (len > TASK_SIZE)
1407 if (flags & MAP_FIXED)
1410 /* requesting a specific address */
1412 addr = PAGE_ALIGN(addr);
1413 vma = find_vma(mm, addr);
1414 if (TASK_SIZE - len >= addr &&
1415 (!vma || addr + len <= vma->vm_start))
1419 /* check if free_area_cache is useful for us */
1420 if (len <= mm->cached_hole_size) {
1421 mm->cached_hole_size = 0;
1422 mm->free_area_cache = mm->mmap_base;
1425 /* either no address requested or can't fit in requested address hole */
1426 addr = mm->free_area_cache;
1428 /* make sure it can fit in the remaining address space */
1430 unsigned long guard;
1433 vma = find_vma(mm, addr);
1437 if (vma->vm_flags & VM_GROWSDOWN)
1438 guard = min(TASK_SIZE - (addr + len),
1439 (unsigned long)heap_stack_gap << PAGE_SHIFT);
1440 if (addr + len + guard <= vma->vm_start)
1444 if (mm->mmap_base < len)
1447 addr = mm->mmap_base-len;
1449 unsigned long guard;
1451 * Lookup failure means no vma is above this address,
1452 * else if new region fits below vma->vm_start,
1453 * return with success:
1455 vma = find_vma(mm, addr);
1459 if (vma->vm_flags & VM_GROWSDOWN)
1460 guard = min(TASK_SIZE - (addr + len),
1461 (unsigned long)heap_stack_gap << PAGE_SHIFT);
1462 if (addr + len + guard <= vma->vm_start) {
1464 /* remember the address as a hint for next time */
1465 mm->free_area_cache = addr;
1469 /* remember the largest hole we saw so far */
1470 if (addr + guard + mm->cached_hole_size < vma->vm_start)
1471 mm->cached_hole_size = vma->vm_start - (addr + guard);
1473 /* try just below the current vma->vm_start */
1474 addr = vma->vm_start - (len + guard);
1475 } while (len < vma->vm_start);
1479 * A failed mmap() very likely causes application failure,
1480 * so fall back to the bottom-up function here. This scenario
1481 * can happen with large stack limits and large mmap()
1484 mm->cached_hole_size = ~0UL;
1485 mm->free_area_cache = TASK_UNMAPPED_BASE;
1486 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1488 * Restore the topdown base:
1490 mm->free_area_cache = mm->mmap_base;
1491 mm->cached_hole_size = ~0UL;
1497 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1500 * Is this a new hole at the highest possible address?
1502 if (addr > mm->free_area_cache)
1503 mm->free_area_cache = addr;
1505 /* dont allow allocations above current base */
1506 if (mm->free_area_cache > mm->mmap_base)
1507 mm->free_area_cache = mm->mmap_base;
1511 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1512 unsigned long pgoff, unsigned long flags)
1514 unsigned long (*get_area)(struct file *, unsigned long,
1515 unsigned long, unsigned long, unsigned long);
1517 unsigned long error = arch_mmap_check(addr, len, flags);
1521 /* Careful about overflows.. */
1522 if (len > TASK_SIZE)
1525 get_area = current->mm->get_unmapped_area;
1526 if (file && file->f_op && file->f_op->get_unmapped_area)
1527 get_area = file->f_op->get_unmapped_area;
1528 addr = get_area(file, addr, len, pgoff, flags);
1529 if (IS_ERR_VALUE(addr))
1532 if (addr > TASK_SIZE - len)
1534 if (addr & ~PAGE_MASK)
1537 return arch_rebalance_pgtables(addr, len);
1540 EXPORT_SYMBOL(get_unmapped_area);
1542 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1543 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1545 struct vm_area_struct *vma = NULL;
1548 /* Check the cache first. */
1549 /* (Cache hit rate is typically around 35%.) */
1550 vma = mm->mmap_cache;
1551 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1552 struct rb_node * rb_node;
1554 rb_node = mm->mm_rb.rb_node;
1558 struct vm_area_struct * vma_tmp;
1560 vma_tmp = rb_entry(rb_node,
1561 struct vm_area_struct, vm_rb);
1563 if (vma_tmp->vm_end > addr) {
1565 if (vma_tmp->vm_start <= addr)
1567 rb_node = rb_node->rb_left;
1569 rb_node = rb_node->rb_right;
1572 mm->mmap_cache = vma;
1578 EXPORT_SYMBOL(find_vma);
1580 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1581 struct vm_area_struct *
1582 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1583 struct vm_area_struct **pprev)
1585 struct vm_area_struct *vma = NULL, *prev = NULL;
1586 struct rb_node *rb_node;
1590 /* Guard against addr being lower than the first VMA */
1593 /* Go through the RB tree quickly. */
1594 rb_node = mm->mm_rb.rb_node;
1597 struct vm_area_struct *vma_tmp;
1598 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1600 if (addr < vma_tmp->vm_end) {
1601 rb_node = rb_node->rb_left;
1604 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1606 rb_node = rb_node->rb_right;
1612 return prev ? prev->vm_next : vma;
1616 * Verify that the stack growth is acceptable and
1617 * update accounting. This is shared with both the
1618 * grow-up and grow-down cases.
1620 static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1622 struct mm_struct *mm = vma->vm_mm;
1623 struct rlimit *rlim = current->signal->rlim;
1624 unsigned long new_start;
1626 /* address space limit tests */
1627 if (!may_expand_vm(mm, grow))
1630 /* Stack limit test */
1631 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1634 /* mlock limit tests */
1635 if (vma->vm_flags & VM_LOCKED) {
1636 unsigned long locked;
1637 unsigned long limit;
1638 locked = mm->locked_vm + grow;
1639 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1640 limit >>= PAGE_SHIFT;
1641 if (locked > limit && !capable(CAP_IPC_LOCK))
1645 /* Check to ensure the stack will not grow into a hugetlb-only region */
1646 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1648 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1652 * Overcommit.. This must be the final test, as it will
1653 * update security statistics.
1655 if (security_vm_enough_memory_mm(mm, grow))
1658 /* Ok, everything looks good - let it rip */
1659 mm->total_vm += grow;
1660 if (vma->vm_flags & VM_LOCKED)
1661 mm->locked_vm += grow;
1662 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1666 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1668 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1669 * vma is the last one with address > vma->vm_end. Have to extend vma.
1674 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1678 if (!(vma->vm_flags & VM_GROWSUP))
1682 * We must make sure the anon_vma is allocated
1683 * so that the anon_vma locking is not a noop.
1685 if (unlikely(anon_vma_prepare(vma)))
1690 * vma->vm_start/vm_end cannot change under us because the caller
1691 * is required to hold the mmap_sem in read mode. We need the
1692 * anon_vma lock to serialize against concurrent expand_stacks.
1693 * Also guard against wrapping around to address 0.
1695 if (address < PAGE_ALIGN(address+4))
1696 address = PAGE_ALIGN(address+4);
1698 anon_vma_unlock(vma);
1703 /* Somebody else might have raced and expanded it already */
1704 if (address > vma->vm_end) {
1705 unsigned long size, grow;
1706 #ifdef CONFIG_STACK_GROWSUP
1707 unsigned long guard;
1708 struct vm_area_struct *vm_next;
1711 guard = min(TASK_SIZE - address,
1712 (unsigned long)heap_stack_gap << PAGE_SHIFT);
1713 vm_next = find_vma(vma->vm_mm, address + guard);
1714 if (unlikely(vm_next && vm_next != vma)) {
1715 /* stack collision with another vma */
1720 size = address - vma->vm_start;
1721 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1723 error = acct_stack_growth(vma, size, grow);
1725 vma->vm_end = address;
1727 out_unlock: __maybe_unused
1728 anon_vma_unlock(vma);
1731 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1734 * vma is the first one with address < vma->vm_start. Have to extend vma.
1736 static int expand_downwards(struct vm_area_struct *vma,
1737 unsigned long address)
1742 * We must make sure the anon_vma is allocated
1743 * so that the anon_vma locking is not a noop.
1745 if (unlikely(anon_vma_prepare(vma)))
1748 address &= PAGE_MASK;
1749 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
1756 * vma->vm_start/vm_end cannot change under us because the caller
1757 * is required to hold the mmap_sem in read mode. We need the
1758 * anon_vma lock to serialize against concurrent expand_stacks.
1761 /* Somebody else might have raced and expanded it already */
1762 if (address < vma->vm_start) {
1763 unsigned long size, grow;
1764 struct vm_area_struct *prev_vma;
1766 find_vma_prev(vma->vm_mm, address, &prev_vma);
1770 unsigned long guard;
1772 guard = min(TASK_SIZE - prev_vma->vm_end,
1773 (unsigned long)heap_stack_gap << PAGE_SHIFT);
1774 if (unlikely(prev_vma->vm_end + guard > address)) {
1775 /* stack collision with another vma */
1779 size = vma->vm_end - address;
1780 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1782 error = acct_stack_growth(vma, size, grow);
1784 vma->vm_start = address;
1785 vma->vm_pgoff -= grow;
1789 anon_vma_unlock(vma);
1793 int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address)
1795 return expand_downwards(vma, address);
1798 #ifdef CONFIG_STACK_GROWSUP
1799 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1801 return expand_upwards(vma, address);
1804 struct vm_area_struct *
1805 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1807 struct vm_area_struct *vma, *prev;
1810 vma = find_vma_prev(mm, addr, &prev);
1811 if (vma && (vma->vm_start <= addr))
1813 if (!prev || expand_stack(prev, addr))
1815 if (prev->vm_flags & VM_LOCKED) {
1816 mlock_vma_pages_range(prev, addr, prev->vm_end);
1821 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1823 return expand_downwards(vma, address);
1826 struct vm_area_struct *
1827 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1829 struct vm_area_struct * vma;
1830 unsigned long start;
1833 vma = find_vma(mm,addr);
1836 if (vma->vm_start <= addr)
1838 if (!(vma->vm_flags & VM_GROWSDOWN))
1840 start = vma->vm_start;
1841 if (expand_stack(vma, addr))
1843 if (vma->vm_flags & VM_LOCKED) {
1844 mlock_vma_pages_range(vma, addr, start);
1851 * Ok - we have the memory areas we should free on the vma list,
1852 * so release them, and do the vma updates.
1854 * Called with the mm semaphore held.
1856 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1858 /* Update high watermark before we lower total_vm */
1859 update_hiwater_vm(mm);
1861 long nrpages = vma_pages(vma);
1863 mm->total_vm -= nrpages;
1864 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1865 vma = remove_vma(vma);
1871 * Get rid of page table information in the indicated region.
1873 * Called with the mm semaphore held.
1875 static void unmap_region(struct mm_struct *mm,
1876 struct vm_area_struct *vma, struct vm_area_struct *prev,
1877 unsigned long start, unsigned long end)
1879 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1880 struct mmu_gather *tlb;
1881 unsigned long nr_accounted = 0;
1884 tlb = tlb_gather_mmu(mm, 0);
1885 update_hiwater_rss(mm);
1886 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1887 vm_unacct_memory(nr_accounted);
1888 free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
1889 next? next->vm_start: 0);
1890 tlb_finish_mmu(tlb, start, end);
1894 * Create a list of vma's touched by the unmap, removing them from the mm's
1895 * vma list as we go..
1898 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1899 struct vm_area_struct *prev, unsigned long end)
1901 struct vm_area_struct **insertion_point;
1902 struct vm_area_struct *tail_vma = NULL;
1905 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1907 rb_erase(&vma->vm_rb, &mm->mm_rb);
1911 } while (vma && vma->vm_start < end);
1912 *insertion_point = vma;
1913 tail_vma->vm_next = NULL;
1914 if (mm->unmap_area == arch_unmap_area)
1915 addr = prev ? prev->vm_end : mm->mmap_base;
1917 addr = vma ? vma->vm_start : mm->mmap_base;
1918 mm->unmap_area(mm, addr);
1919 mm->mmap_cache = NULL; /* Kill the cache. */
1923 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1924 * munmap path where it doesn't make sense to fail.
1926 static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1927 unsigned long addr, int new_below)
1929 struct mempolicy *pol;
1930 struct vm_area_struct *new;
1933 if (is_vm_hugetlb_page(vma) && (addr &
1934 ~(huge_page_mask(hstate_vma(vma)))))
1937 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1941 /* most fields are the same, copy all, and then fixup */
1944 INIT_LIST_HEAD(&new->anon_vma_chain);
1949 new->vm_start = addr;
1950 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1953 pol = mpol_dup(vma_policy(vma));
1958 vma_set_policy(new, pol);
1960 if (anon_vma_clone(new, vma))
1964 get_file(new->vm_file);
1965 if (vma->vm_flags & VM_EXECUTABLE)
1966 added_exe_file_vma(mm);
1969 if (new->vm_ops && new->vm_ops->open)
1970 new->vm_ops->open(new);
1973 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1974 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1976 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1982 /* Clean everything up if vma_adjust failed. */
1983 new->vm_ops->close(new);
1985 if (vma->vm_flags & VM_EXECUTABLE)
1986 removed_exe_file_vma(mm);
1992 kmem_cache_free(vm_area_cachep, new);
1998 * Split a vma into two pieces at address 'addr', a new vma is allocated
1999 * either for the first part or the tail.
2001 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2002 unsigned long addr, int new_below)
2004 if (mm->map_count >= sysctl_max_map_count)
2007 return __split_vma(mm, vma, addr, new_below);
2010 /* Munmap is split into 2 main parts -- this part which finds
2011 * what needs doing, and the areas themselves, which do the
2012 * work. This now handles partial unmappings.
2013 * Jeremy Fitzhardinge <jeremy@goop.org>
2015 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2018 struct vm_area_struct *vma, *prev, *last;
2020 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2023 if ((len = PAGE_ALIGN(len)) == 0)
2026 /* Find the first overlapping VMA */
2027 vma = find_vma_prev(mm, start, &prev);
2030 /* we have start < vma->vm_end */
2032 /* if it doesn't overlap, we have nothing.. */
2034 if (vma->vm_start >= end)
2038 * If we need to split any vma, do it now to save pain later.
2040 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2041 * unmapped vm_area_struct will remain in use: so lower split_vma
2042 * places tmp vma above, and higher split_vma places tmp vma below.
2044 if (start > vma->vm_start) {
2048 * Make sure that map_count on return from munmap() will
2049 * not exceed its limit; but let map_count go just above
2050 * its limit temporarily, to help free resources as expected.
2052 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2055 error = __split_vma(mm, vma, start, 0);
2061 /* Does it split the last one? */
2062 last = find_vma(mm, end);
2063 if (last && end > last->vm_start) {
2064 int error = __split_vma(mm, last, end, 1);
2068 vma = prev? prev->vm_next: mm->mmap;
2071 * unlock any mlock()ed ranges before detaching vmas
2073 if (mm->locked_vm) {
2074 struct vm_area_struct *tmp = vma;
2075 while (tmp && tmp->vm_start < end) {
2076 if (tmp->vm_flags & VM_LOCKED) {
2077 mm->locked_vm -= vma_pages(tmp);
2078 munlock_vma_pages_all(tmp);
2085 * Remove the vma's, and unmap the actual pages
2087 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2088 unmap_region(mm, vma, prev, start, end);
2090 /* Fix up all other VM information */
2091 remove_vma_list(mm, vma);
2096 EXPORT_SYMBOL(do_munmap);
2098 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2101 struct mm_struct *mm = current->mm;
2103 profile_munmap(addr);
2105 down_write(&mm->mmap_sem);
2106 ret = do_munmap(mm, addr, len);
2107 up_write(&mm->mmap_sem);
2111 static inline void verify_mm_writelocked(struct mm_struct *mm)
2113 #ifdef CONFIG_DEBUG_VM
2114 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2116 up_read(&mm->mmap_sem);
2122 * this is really a simplified "do_mmap". it only handles
2123 * anonymous maps. eventually we may be able to do some
2124 * brk-specific accounting here.
2126 unsigned long do_brk(unsigned long addr, unsigned long len)
2128 struct mm_struct * mm = current->mm;
2129 struct vm_area_struct * vma, * prev;
2130 unsigned long flags;
2131 struct rb_node ** rb_link, * rb_parent;
2132 pgoff_t pgoff = addr >> PAGE_SHIFT;
2135 len = PAGE_ALIGN(len);
2139 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
2143 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2145 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2146 if (error & ~PAGE_MASK)
2152 if (mm->def_flags & VM_LOCKED) {
2153 unsigned long locked, lock_limit;
2154 locked = len >> PAGE_SHIFT;
2155 locked += mm->locked_vm;
2156 lock_limit = rlimit(RLIMIT_MEMLOCK);
2157 lock_limit >>= PAGE_SHIFT;
2158 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2163 * mm->mmap_sem is required to protect against another thread
2164 * changing the mappings in case we sleep.
2166 verify_mm_writelocked(mm);
2169 * Clear old maps. this also does some error checking for us
2172 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2173 if (vma && vma->vm_start < addr + len) {
2174 if (do_munmap(mm, addr, len))
2179 /* Check against address space limits *after* clearing old maps... */
2180 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2183 if (mm->map_count > sysctl_max_map_count)
2186 if (security_vm_enough_memory(len >> PAGE_SHIFT))
2189 /* Can we just expand an old private anonymous mapping? */
2190 vma = vma_merge(mm, prev, addr, addr + len, flags,
2191 NULL, NULL, pgoff, NULL);
2196 * create a vma struct for an anonymous mapping
2198 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2200 vm_unacct_memory(len >> PAGE_SHIFT);
2204 INIT_LIST_HEAD(&vma->anon_vma_chain);
2206 vma->vm_start = addr;
2207 vma->vm_end = addr + len;
2208 vma->vm_pgoff = pgoff;
2209 vma->vm_flags = flags;
2210 vma->vm_page_prot = vm_get_page_prot(flags);
2211 vma_link(mm, vma, prev, rb_link, rb_parent);
2213 mm->total_vm += len >> PAGE_SHIFT;
2214 if (flags & VM_LOCKED) {
2215 if (!mlock_vma_pages_range(vma, addr, addr + len))
2216 mm->locked_vm += (len >> PAGE_SHIFT);
2221 EXPORT_SYMBOL(do_brk);
2223 /* Release all mmaps. */
2224 void exit_mmap(struct mm_struct *mm)
2226 struct mmu_gather *tlb;
2227 struct vm_area_struct *vma;
2228 unsigned long nr_accounted = 0;
2231 /* mm's last user has gone, and its about to be pulled down */
2232 mmu_notifier_release(mm);
2234 if (mm->locked_vm) {
2237 if (vma->vm_flags & VM_LOCKED)
2238 munlock_vma_pages_all(vma);
2246 if (!vma) /* Can happen if dup_mmap() received an OOM */
2251 tlb = tlb_gather_mmu(mm, 1);
2252 /* update_hiwater_rss(mm) here? but nobody should be looking */
2253 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2254 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2255 vm_unacct_memory(nr_accounted);
2257 free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0);
2258 tlb_finish_mmu(tlb, 0, end);
2261 * Walk the list again, actually closing and freeing it,
2262 * with preemption enabled, without holding any MM locks.
2265 vma = remove_vma(vma);
2267 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2270 /* Insert vm structure into process list sorted by address
2271 * and into the inode's i_mmap tree. If vm_file is non-NULL
2272 * then i_mmap_lock is taken here.
2274 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2276 struct vm_area_struct * __vma, * prev;
2277 struct rb_node ** rb_link, * rb_parent;
2280 * The vm_pgoff of a purely anonymous vma should be irrelevant
2281 * until its first write fault, when page's anon_vma and index
2282 * are set. But now set the vm_pgoff it will almost certainly
2283 * end up with (unless mremap moves it elsewhere before that
2284 * first wfault), so /proc/pid/maps tells a consistent story.
2286 * By setting it to reflect the virtual start address of the
2287 * vma, merges and splits can happen in a seamless way, just
2288 * using the existing file pgoff checks and manipulations.
2289 * Similarly in do_mmap_pgoff and in do_brk.
2291 if (!vma->vm_file) {
2292 BUG_ON(vma->anon_vma);
2293 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2295 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2296 if (__vma && __vma->vm_start < vma->vm_end)
2298 if ((vma->vm_flags & VM_ACCOUNT) &&
2299 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2301 vma_link(mm, vma, prev, rb_link, rb_parent);
2306 * Copy the vma structure to a new location in the same mm,
2307 * prior to moving page table entries, to effect an mremap move.
2309 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2310 unsigned long addr, unsigned long len, pgoff_t pgoff)
2312 struct vm_area_struct *vma = *vmap;
2313 unsigned long vma_start = vma->vm_start;
2314 struct mm_struct *mm = vma->vm_mm;
2315 struct vm_area_struct *new_vma, *prev;
2316 struct rb_node **rb_link, *rb_parent;
2317 struct mempolicy *pol;
2320 * If anonymous vma has not yet been faulted, update new pgoff
2321 * to match new location, to increase its chance of merging.
2323 if (!vma->vm_file && !vma->anon_vma)
2324 pgoff = addr >> PAGE_SHIFT;
2326 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2327 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2328 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2331 * Source vma may have been merged into new_vma
2333 if (vma_start >= new_vma->vm_start &&
2334 vma_start < new_vma->vm_end)
2337 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2340 pol = mpol_dup(vma_policy(vma));
2343 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2344 if (anon_vma_clone(new_vma, vma))
2345 goto out_free_mempol;
2346 vma_set_policy(new_vma, pol);
2347 new_vma->vm_start = addr;
2348 new_vma->vm_end = addr + len;
2349 new_vma->vm_pgoff = pgoff;
2350 if (new_vma->vm_file) {
2351 get_file(new_vma->vm_file);
2352 if (vma->vm_flags & VM_EXECUTABLE)
2353 added_exe_file_vma(mm);
2355 if (new_vma->vm_ops && new_vma->vm_ops->open)
2356 new_vma->vm_ops->open(new_vma);
2357 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2365 kmem_cache_free(vm_area_cachep, new_vma);
2370 * Return true if the calling process may expand its vm space by the passed
2373 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2375 unsigned long cur = mm->total_vm; /* pages */
2378 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2380 if (cur + npages > lim)
2386 static int special_mapping_fault(struct vm_area_struct *vma,
2387 struct vm_fault *vmf)
2390 struct page **pages;
2393 * special mappings have no vm_file, and in that case, the mm
2394 * uses vm_pgoff internally. So we have to subtract it from here.
2395 * We are allowed to do this because we are the mm; do not copy
2396 * this code into drivers!
2398 pgoff = vmf->pgoff - vma->vm_pgoff;
2400 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2404 struct page *page = *pages;
2410 return VM_FAULT_SIGBUS;
2414 * Having a close hook prevents vma merging regardless of flags.
2416 static void special_mapping_close(struct vm_area_struct *vma)
2420 static const struct vm_operations_struct special_mapping_vmops = {
2421 .close = special_mapping_close,
2422 .fault = special_mapping_fault,
2426 * Called with mm->mmap_sem held for writing.
2427 * Insert a new vma covering the given region, with the given flags.
2428 * Its pages are supplied by the given array of struct page *.
2429 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2430 * The region past the last page supplied will always produce SIGBUS.
2431 * The array pointer and the pages it points to are assumed to stay alive
2432 * for as long as this mapping might exist.
2434 int install_special_mapping(struct mm_struct *mm,
2435 unsigned long addr, unsigned long len,
2436 unsigned long vm_flags, struct page **pages)
2438 struct vm_area_struct *vma;
2440 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2441 if (unlikely(vma == NULL))
2444 INIT_LIST_HEAD(&vma->anon_vma_chain);
2446 vma->vm_start = addr;
2447 vma->vm_end = addr + len;
2449 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2450 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2452 vma->vm_ops = &special_mapping_vmops;
2453 vma->vm_private_data = pages;
2455 if (unlikely(insert_vm_struct(mm, vma))) {
2456 kmem_cache_free(vm_area_cachep, vma);
2460 mm->total_vm += len >> PAGE_SHIFT;
2462 perf_event_mmap(vma);
2467 static DEFINE_MUTEX(mm_all_locks_mutex);
2469 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2471 if (!test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2473 * The LSB of head.next can't change from under us
2474 * because we hold the mm_all_locks_mutex.
2476 spin_lock_nest_lock(&anon_vma->lock, &mm->mmap_sem);
2478 * We can safely modify head.next after taking the
2479 * anon_vma->lock. If some other vma in this mm shares
2480 * the same anon_vma we won't take it again.
2482 * No need of atomic instructions here, head.next
2483 * can't change from under us thanks to the
2486 if (__test_and_set_bit(0, (unsigned long *)
2487 &anon_vma->head.next))
2492 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2494 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2496 * AS_MM_ALL_LOCKS can't change from under us because
2497 * we hold the mm_all_locks_mutex.
2499 * Operations on ->flags have to be atomic because
2500 * even if AS_MM_ALL_LOCKS is stable thanks to the
2501 * mm_all_locks_mutex, there may be other cpus
2502 * changing other bitflags in parallel to us.
2504 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2506 spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem);
2511 * This operation locks against the VM for all pte/vma/mm related
2512 * operations that could ever happen on a certain mm. This includes
2513 * vmtruncate, try_to_unmap, and all page faults.
2515 * The caller must take the mmap_sem in write mode before calling
2516 * mm_take_all_locks(). The caller isn't allowed to release the
2517 * mmap_sem until mm_drop_all_locks() returns.
2519 * mmap_sem in write mode is required in order to block all operations
2520 * that could modify pagetables and free pages without need of
2521 * altering the vma layout (for example populate_range() with
2522 * nonlinear vmas). It's also needed in write mode to avoid new
2523 * anon_vmas to be associated with existing vmas.
2525 * A single task can't take more than one mm_take_all_locks() in a row
2526 * or it would deadlock.
2528 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2529 * mapping->flags avoid to take the same lock twice, if more than one
2530 * vma in this mm is backed by the same anon_vma or address_space.
2532 * We can take all the locks in random order because the VM code
2533 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2534 * takes more than one of them in a row. Secondly we're protected
2535 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2537 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2538 * that may have to take thousand of locks.
2540 * mm_take_all_locks() can fail if it's interrupted by signals.
2542 int mm_take_all_locks(struct mm_struct *mm)
2544 struct vm_area_struct *vma;
2545 struct anon_vma_chain *avc;
2548 BUG_ON(down_read_trylock(&mm->mmap_sem));
2550 mutex_lock(&mm_all_locks_mutex);
2552 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2553 if (signal_pending(current))
2555 if (vma->vm_file && vma->vm_file->f_mapping)
2556 vm_lock_mapping(mm, vma->vm_file->f_mapping);
2559 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2560 if (signal_pending(current))
2563 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2564 vm_lock_anon_vma(mm, avc->anon_vma);
2571 mm_drop_all_locks(mm);
2576 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2578 if (test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2580 * The LSB of head.next can't change to 0 from under
2581 * us because we hold the mm_all_locks_mutex.
2583 * We must however clear the bitflag before unlocking
2584 * the vma so the users using the anon_vma->head will
2585 * never see our bitflag.
2587 * No need of atomic instructions here, head.next
2588 * can't change from under us until we release the
2591 if (!__test_and_clear_bit(0, (unsigned long *)
2592 &anon_vma->head.next))
2594 spin_unlock(&anon_vma->lock);
2598 static void vm_unlock_mapping(struct address_space *mapping)
2600 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2602 * AS_MM_ALL_LOCKS can't change to 0 from under us
2603 * because we hold the mm_all_locks_mutex.
2605 spin_unlock(&mapping->i_mmap_lock);
2606 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2613 * The mmap_sem cannot be released by the caller until
2614 * mm_drop_all_locks() returns.
2616 void mm_drop_all_locks(struct mm_struct *mm)
2618 struct vm_area_struct *vma;
2619 struct anon_vma_chain *avc;
2621 BUG_ON(down_read_trylock(&mm->mmap_sem));
2622 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2624 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2626 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2627 vm_unlock_anon_vma(avc->anon_vma);
2628 if (vma->vm_file && vma->vm_file->f_mapping)
2629 vm_unlock_mapping(vma->vm_file->f_mapping);
2632 mutex_unlock(&mm_all_locks_mutex);
2636 * initialise the VMA slab
2638 void __init mmap_init(void)
2642 ret = percpu_counter_init(&vm_committed_as, 0);