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;
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;
182 if (percpu_counter_read_positive(&vm_committed_as) < allowed)
185 vm_unacct_memory(pages);
191 * Requires inode->i_mapping->i_mmap_lock
193 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
194 struct file *file, struct address_space *mapping)
196 if (vma->vm_flags & VM_DENYWRITE)
197 atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
198 if (vma->vm_flags & VM_SHARED)
199 mapping->i_mmap_writable--;
201 flush_dcache_mmap_lock(mapping);
202 if (unlikely(vma->vm_flags & VM_NONLINEAR))
203 list_del_init(&vma->shared.vm_set.list);
205 vma_prio_tree_remove(vma, &mapping->i_mmap);
206 flush_dcache_mmap_unlock(mapping);
210 * Unlink a file-based vm structure from its prio_tree, to hide
211 * vma from rmap and vmtruncate before freeing its page tables.
213 void unlink_file_vma(struct vm_area_struct *vma)
215 struct file *file = vma->vm_file;
218 struct address_space *mapping = file->f_mapping;
219 spin_lock(&mapping->i_mmap_lock);
220 __remove_shared_vm_struct(vma, file, mapping);
221 spin_unlock(&mapping->i_mmap_lock);
226 * Close a vm structure and free it, returning the next.
228 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
230 struct vm_area_struct *next = vma->vm_next;
233 if (vma->vm_ops && vma->vm_ops->close)
234 vma->vm_ops->close(vma);
237 if (vma->vm_flags & VM_EXECUTABLE)
238 removed_exe_file_vma(vma->vm_mm);
240 mpol_put(vma_policy(vma));
241 kmem_cache_free(vm_area_cachep, vma);
245 SYSCALL_DEFINE1(brk, unsigned long, brk)
247 unsigned long rlim, retval;
248 unsigned long newbrk, oldbrk;
249 struct mm_struct *mm = current->mm;
250 unsigned long min_brk;
252 down_write(&mm->mmap_sem);
254 #ifdef CONFIG_COMPAT_BRK
255 min_brk = mm->end_code;
257 min_brk = mm->start_brk;
263 * Check against rlimit here. If this check is done later after the test
264 * of oldbrk with newbrk then it can escape the test and let the data
265 * segment grow beyond its set limit the in case where the limit is
266 * not page aligned -Ram Gupta
268 rlim = rlimit(RLIMIT_DATA);
269 if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
270 (mm->end_data - mm->start_data) > rlim)
273 newbrk = PAGE_ALIGN(brk);
274 oldbrk = PAGE_ALIGN(mm->brk);
275 if (oldbrk == newbrk)
278 /* Always allow shrinking brk. */
279 if (brk <= mm->brk) {
280 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
285 /* Check against existing mmap mappings. */
286 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
289 /* Ok, looks good - let it rip. */
290 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
296 up_write(&mm->mmap_sem);
301 static int browse_rb(struct rb_root *root)
304 struct rb_node *nd, *pn = NULL;
305 unsigned long prev = 0, pend = 0;
307 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
308 struct vm_area_struct *vma;
309 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
310 if (vma->vm_start < prev)
311 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
312 if (vma->vm_start < pend)
313 printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
314 if (vma->vm_start > vma->vm_end)
315 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
318 prev = vma->vm_start;
322 for (nd = pn; nd; nd = rb_prev(nd)) {
326 printk("backwards %d, forwards %d\n", j, i), i = 0;
330 void validate_mm(struct mm_struct *mm)
334 struct vm_area_struct *tmp = mm->mmap;
339 if (i != mm->map_count)
340 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
341 i = browse_rb(&mm->mm_rb);
342 if (i != mm->map_count)
343 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
347 #define validate_mm(mm) do { } while (0)
350 static struct vm_area_struct *
351 find_vma_prepare(struct mm_struct *mm, unsigned long addr,
352 struct vm_area_struct **pprev, struct rb_node ***rb_link,
353 struct rb_node ** rb_parent)
355 struct vm_area_struct * vma;
356 struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
358 __rb_link = &mm->mm_rb.rb_node;
359 rb_prev = __rb_parent = NULL;
363 struct vm_area_struct *vma_tmp;
365 __rb_parent = *__rb_link;
366 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
368 if (vma_tmp->vm_end > addr) {
370 if (vma_tmp->vm_start <= addr)
372 __rb_link = &__rb_parent->rb_left;
374 rb_prev = __rb_parent;
375 __rb_link = &__rb_parent->rb_right;
381 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
382 *rb_link = __rb_link;
383 *rb_parent = __rb_parent;
388 __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma,
389 struct vm_area_struct *prev, struct rb_node *rb_parent)
392 vma->vm_next = prev->vm_next;
397 vma->vm_next = rb_entry(rb_parent,
398 struct vm_area_struct, vm_rb);
404 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
405 struct rb_node **rb_link, struct rb_node *rb_parent)
407 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
408 rb_insert_color(&vma->vm_rb, &mm->mm_rb);
411 static void __vma_link_file(struct vm_area_struct *vma)
417 struct address_space *mapping = file->f_mapping;
419 if (vma->vm_flags & VM_DENYWRITE)
420 atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
421 if (vma->vm_flags & VM_SHARED)
422 mapping->i_mmap_writable++;
424 flush_dcache_mmap_lock(mapping);
425 if (unlikely(vma->vm_flags & VM_NONLINEAR))
426 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
428 vma_prio_tree_insert(vma, &mapping->i_mmap);
429 flush_dcache_mmap_unlock(mapping);
434 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
435 struct vm_area_struct *prev, struct rb_node **rb_link,
436 struct rb_node *rb_parent)
438 __vma_link_list(mm, vma, prev, rb_parent);
439 __vma_link_rb(mm, vma, rb_link, rb_parent);
442 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
443 struct vm_area_struct *prev, struct rb_node **rb_link,
444 struct rb_node *rb_parent)
446 struct address_space *mapping = NULL;
449 mapping = vma->vm_file->f_mapping;
452 spin_lock(&mapping->i_mmap_lock);
453 vma->vm_truncate_count = mapping->truncate_count;
457 __vma_link(mm, vma, prev, rb_link, rb_parent);
458 __vma_link_file(vma);
460 anon_vma_unlock(vma);
462 spin_unlock(&mapping->i_mmap_lock);
469 * Helper for vma_adjust in the split_vma insert case:
470 * insert vm structure into list and rbtree and anon_vma,
471 * but it has already been inserted into prio_tree earlier.
473 static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
475 struct vm_area_struct *__vma, *prev;
476 struct rb_node **rb_link, *rb_parent;
478 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
479 BUG_ON(__vma && __vma->vm_start < vma->vm_end);
480 __vma_link(mm, vma, prev, rb_link, rb_parent);
485 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
486 struct vm_area_struct *prev)
488 prev->vm_next = vma->vm_next;
489 rb_erase(&vma->vm_rb, &mm->mm_rb);
490 if (mm->mmap_cache == vma)
491 mm->mmap_cache = prev;
495 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
496 * is already present in an i_mmap tree without adjusting the tree.
497 * The following helper function should be used when such adjustments
498 * are necessary. The "insert" vma (if any) is to be inserted
499 * before we drop the necessary locks.
501 int vma_adjust(struct vm_area_struct *vma, unsigned long start,
502 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
504 struct mm_struct *mm = vma->vm_mm;
505 struct vm_area_struct *next = vma->vm_next;
506 struct vm_area_struct *importer = NULL;
507 struct address_space *mapping = NULL;
508 struct prio_tree_root *root = NULL;
509 struct file *file = vma->vm_file;
510 struct anon_vma *anon_vma = NULL;
511 long adjust_next = 0;
514 if (next && !insert) {
515 if (end >= next->vm_end) {
517 * vma expands, overlapping all the next, and
518 * perhaps the one after too (mprotect case 6).
520 again: remove_next = 1 + (end > next->vm_end);
522 anon_vma = next->anon_vma;
524 } else if (end > next->vm_start) {
526 * vma expands, overlapping part of the next:
527 * mprotect case 5 shifting the boundary up.
529 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
530 anon_vma = next->anon_vma;
532 } else if (end < vma->vm_end) {
534 * vma shrinks, and !insert tells it's not
535 * split_vma inserting another: so it must be
536 * mprotect case 4 shifting the boundary down.
538 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
539 anon_vma = next->anon_vma;
545 * When changing only vma->vm_end, we don't really need anon_vma lock.
547 if (vma->anon_vma && (insert || importer || start != vma->vm_start))
548 anon_vma = vma->anon_vma;
551 * Easily overlooked: when mprotect shifts the boundary,
552 * make sure the expanding vma has anon_vma set if the
553 * shrinking vma had, to cover any anon pages imported.
555 if (importer && !importer->anon_vma) {
556 /* Block reverse map lookups until things are set up. */
557 if (anon_vma_clone(importer, vma)) {
560 importer->anon_vma = anon_vma;
565 mapping = file->f_mapping;
566 if (!(vma->vm_flags & VM_NONLINEAR))
567 root = &mapping->i_mmap;
568 spin_lock(&mapping->i_mmap_lock);
570 vma->vm_truncate_count != next->vm_truncate_count) {
572 * unmap_mapping_range might be in progress:
573 * ensure that the expanding vma is rescanned.
575 importer->vm_truncate_count = 0;
578 insert->vm_truncate_count = vma->vm_truncate_count;
580 * Put into prio_tree now, so instantiated pages
581 * are visible to arm/parisc __flush_dcache_page
582 * throughout; but we cannot insert into address
583 * space until vma start or end is updated.
585 __vma_link_file(insert);
590 flush_dcache_mmap_lock(mapping);
591 vma_prio_tree_remove(vma, root);
593 vma_prio_tree_remove(next, root);
596 vma->vm_start = start;
598 vma->vm_pgoff = pgoff;
600 next->vm_start += adjust_next << PAGE_SHIFT;
601 next->vm_pgoff += adjust_next;
606 vma_prio_tree_insert(next, root);
607 vma_prio_tree_insert(vma, root);
608 flush_dcache_mmap_unlock(mapping);
613 * vma_merge has merged next into vma, and needs
614 * us to remove next before dropping the locks.
616 __vma_unlink(mm, next, vma);
618 __remove_shared_vm_struct(next, file, mapping);
621 * split_vma has split insert from vma, and needs
622 * us to insert it before dropping the locks
623 * (it may either follow vma or precede it).
625 __insert_vm_struct(mm, insert);
629 spin_unlock(&mapping->i_mmap_lock);
634 if (next->vm_flags & VM_EXECUTABLE)
635 removed_exe_file_vma(mm);
638 anon_vma_merge(vma, next);
640 mpol_put(vma_policy(next));
641 kmem_cache_free(vm_area_cachep, next);
643 * In mprotect's case 6 (see comments on vma_merge),
644 * we must remove another next too. It would clutter
645 * up the code too much to do both in one go.
647 if (remove_next == 2) {
659 * If the vma has a ->close operation then the driver probably needs to release
660 * per-vma resources, so we don't attempt to merge those.
662 static inline int is_mergeable_vma(struct vm_area_struct *vma,
663 struct file *file, unsigned long vm_flags)
665 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
666 if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
668 if (vma->vm_file != file)
670 if (vma->vm_ops && vma->vm_ops->close)
675 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
676 struct anon_vma *anon_vma2)
678 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2);
682 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
683 * in front of (at a lower virtual address and file offset than) the vma.
685 * We cannot merge two vmas if they have differently assigned (non-NULL)
686 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
688 * We don't check here for the merged mmap wrapping around the end of pagecache
689 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
690 * wrap, nor mmaps which cover the final page at index -1UL.
693 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
694 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
696 if (is_mergeable_vma(vma, file, vm_flags) &&
697 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
698 if (vma->vm_pgoff == vm_pgoff)
705 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
706 * beyond (at a higher virtual address and file offset than) the vma.
708 * We cannot merge two vmas if they have differently assigned (non-NULL)
709 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
712 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
713 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
715 if (is_mergeable_vma(vma, file, vm_flags) &&
716 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) {
718 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
719 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
726 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
727 * whether that can be merged with its predecessor or its successor.
728 * Or both (it neatly fills a hole).
730 * In most cases - when called for mmap, brk or mremap - [addr,end) is
731 * certain not to be mapped by the time vma_merge is called; but when
732 * called for mprotect, it is certain to be already mapped (either at
733 * an offset within prev, or at the start of next), and the flags of
734 * this area are about to be changed to vm_flags - and the no-change
735 * case has already been eliminated.
737 * The following mprotect cases have to be considered, where AAAA is
738 * the area passed down from mprotect_fixup, never extending beyond one
739 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
741 * AAAA AAAA AAAA AAAA
742 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
743 * cannot merge might become might become might become
744 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
745 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
746 * mremap move: PPPPNNNNNNNN 8
748 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
749 * might become case 1 below case 2 below case 3 below
751 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
752 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
754 struct vm_area_struct *vma_merge(struct mm_struct *mm,
755 struct vm_area_struct *prev, unsigned long addr,
756 unsigned long end, unsigned long vm_flags,
757 struct anon_vma *anon_vma, struct file *file,
758 pgoff_t pgoff, struct mempolicy *policy)
760 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
761 struct vm_area_struct *area, *next;
765 * We later require that vma->vm_flags == vm_flags,
766 * so this tests vma->vm_flags & VM_SPECIAL, too.
768 if (vm_flags & VM_SPECIAL)
772 next = prev->vm_next;
776 if (next && next->vm_end == end) /* cases 6, 7, 8 */
777 next = next->vm_next;
780 * Can it merge with the predecessor?
782 if (prev && prev->vm_end == addr &&
783 mpol_equal(vma_policy(prev), policy) &&
784 can_vma_merge_after(prev, vm_flags,
785 anon_vma, file, pgoff)) {
787 * OK, it can. Can we now merge in the successor as well?
789 if (next && end == next->vm_start &&
790 mpol_equal(policy, vma_policy(next)) &&
791 can_vma_merge_before(next, vm_flags,
792 anon_vma, file, pgoff+pglen) &&
793 is_mergeable_anon_vma(prev->anon_vma,
796 err = vma_adjust(prev, prev->vm_start,
797 next->vm_end, prev->vm_pgoff, NULL);
798 } else /* cases 2, 5, 7 */
799 err = vma_adjust(prev, prev->vm_start,
800 end, prev->vm_pgoff, NULL);
807 * Can this new request be merged in front of next?
809 if (next && end == next->vm_start &&
810 mpol_equal(policy, vma_policy(next)) &&
811 can_vma_merge_before(next, vm_flags,
812 anon_vma, file, pgoff+pglen)) {
813 if (prev && addr < prev->vm_end) /* case 4 */
814 err = vma_adjust(prev, prev->vm_start,
815 addr, prev->vm_pgoff, NULL);
816 else /* cases 3, 8 */
817 err = vma_adjust(area, addr, next->vm_end,
818 next->vm_pgoff - pglen, NULL);
828 * Rough compatbility check to quickly see if it's even worth looking
829 * at sharing an anon_vma.
831 * They need to have the same vm_file, and the flags can only differ
832 * in things that mprotect may change.
834 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
835 * we can merge the two vma's. For example, we refuse to merge a vma if
836 * there is a vm_ops->close() function, because that indicates that the
837 * driver is doing some kind of reference counting. But that doesn't
838 * really matter for the anon_vma sharing case.
840 static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
842 return a->vm_end == b->vm_start &&
843 mpol_equal(vma_policy(a), vma_policy(b)) &&
844 a->vm_file == b->vm_file &&
845 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
846 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
850 * Do some basic sanity checking to see if we can re-use the anon_vma
851 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
852 * the same as 'old', the other will be the new one that is trying
853 * to share the anon_vma.
855 * NOTE! This runs with mm_sem held for reading, so it is possible that
856 * the anon_vma of 'old' is concurrently in the process of being set up
857 * by another page fault trying to merge _that_. But that's ok: if it
858 * is being set up, that automatically means that it will be a singleton
859 * acceptable for merging, so we can do all of this optimistically. But
860 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
862 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
863 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
864 * is to return an anon_vma that is "complex" due to having gone through
867 * We also make sure that the two vma's are compatible (adjacent,
868 * and with the same memory policies). That's all stable, even with just
869 * a read lock on the mm_sem.
871 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
873 if (anon_vma_compatible(a, b)) {
874 struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
876 if (anon_vma && list_is_singular(&old->anon_vma_chain))
883 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
884 * neighbouring vmas for a suitable anon_vma, before it goes off
885 * to allocate a new anon_vma. It checks because a repetitive
886 * sequence of mprotects and faults may otherwise lead to distinct
887 * anon_vmas being allocated, preventing vma merge in subsequent
890 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
892 struct anon_vma *anon_vma;
893 struct vm_area_struct *near;
899 anon_vma = reusable_anon_vma(near, vma, near);
904 * It is potentially slow to have to call find_vma_prev here.
905 * But it's only on the first write fault on the vma, not
906 * every time, and we could devise a way to avoid it later
907 * (e.g. stash info in next's anon_vma_node when assigning
908 * an anon_vma, or when trying vma_merge). Another time.
910 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma);
914 anon_vma = reusable_anon_vma(near, near, vma);
919 * There's no absolute need to look only at touching neighbours:
920 * we could search further afield for "compatible" anon_vmas.
921 * But it would probably just be a waste of time searching,
922 * or lead to too many vmas hanging off the same anon_vma.
923 * We're trying to allow mprotect remerging later on,
924 * not trying to minimize memory used for anon_vmas.
929 #ifdef CONFIG_PROC_FS
930 void vm_stat_account(struct mm_struct *mm, unsigned long flags,
931 struct file *file, long pages)
933 const unsigned long stack_flags
934 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
937 mm->shared_vm += pages;
938 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
939 mm->exec_vm += pages;
940 } else if (flags & stack_flags)
941 mm->stack_vm += pages;
942 if (flags & (VM_RESERVED|VM_IO))
943 mm->reserved_vm += pages;
945 #endif /* CONFIG_PROC_FS */
948 * The caller must hold down_write(¤t->mm->mmap_sem).
951 unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
952 unsigned long len, unsigned long prot,
953 unsigned long flags, unsigned long pgoff)
955 struct mm_struct * mm = current->mm;
957 unsigned int vm_flags;
959 unsigned long reqprot = prot;
962 * Does the application expect PROT_READ to imply PROT_EXEC?
964 * (the exception is when the underlying filesystem is noexec
965 * mounted, in which case we dont add PROT_EXEC.)
967 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
968 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
974 if (!(flags & MAP_FIXED))
975 addr = round_hint_to_min(addr);
977 /* Careful about overflows.. */
978 len = PAGE_ALIGN(len);
982 /* offset overflow? */
983 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
986 /* Too many mappings? */
987 if (mm->map_count > sysctl_max_map_count)
990 /* Obtain the address to map to. we verify (or select) it and ensure
991 * that it represents a valid section of the address space.
993 addr = get_unmapped_area(file, addr, len, pgoff, flags);
994 if (addr & ~PAGE_MASK)
997 /* Do simple checking here so the lower-level routines won't have
998 * to. we assume access permissions have been handled by the open
999 * of the memory object, so we don't do any here.
1001 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1002 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1004 if (flags & MAP_LOCKED)
1005 if (!can_do_mlock())
1008 /* mlock MCL_FUTURE? */
1009 if (vm_flags & VM_LOCKED) {
1010 unsigned long locked, lock_limit;
1011 locked = len >> PAGE_SHIFT;
1012 locked += mm->locked_vm;
1013 lock_limit = rlimit(RLIMIT_MEMLOCK);
1014 lock_limit >>= PAGE_SHIFT;
1015 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1019 inode = file ? file->f_path.dentry->d_inode : NULL;
1022 switch (flags & MAP_TYPE) {
1024 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1028 * Make sure we don't allow writing to an append-only
1031 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1035 * Make sure there are no mandatory locks on the file.
1037 if (locks_verify_locked(inode))
1040 vm_flags |= VM_SHARED | VM_MAYSHARE;
1041 if (!(file->f_mode & FMODE_WRITE))
1042 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1046 if (!(file->f_mode & FMODE_READ))
1048 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1049 if (vm_flags & VM_EXEC)
1051 vm_flags &= ~VM_MAYEXEC;
1054 if (!file->f_op || !file->f_op->mmap)
1062 switch (flags & MAP_TYPE) {
1068 vm_flags |= VM_SHARED | VM_MAYSHARE;
1072 * Set pgoff according to addr for anon_vma.
1074 pgoff = addr >> PAGE_SHIFT;
1081 error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1085 return mmap_region(file, addr, len, flags, vm_flags, pgoff);
1087 EXPORT_SYMBOL(do_mmap_pgoff);
1089 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1090 unsigned long, prot, unsigned long, flags,
1091 unsigned long, fd, unsigned long, pgoff)
1093 struct file *file = NULL;
1094 unsigned long retval = -EBADF;
1096 if (!(flags & MAP_ANONYMOUS)) {
1097 if (unlikely(flags & MAP_HUGETLB))
1102 } else if (flags & MAP_HUGETLB) {
1103 struct user_struct *user = NULL;
1105 * VM_NORESERVE is used because the reservations will be
1106 * taken when vm_ops->mmap() is called
1107 * A dummy user value is used because we are not locking
1108 * memory so no accounting is necessary
1110 len = ALIGN(len, huge_page_size(&default_hstate));
1111 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1112 &user, HUGETLB_ANONHUGE_INODE);
1114 return PTR_ERR(file);
1117 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1119 down_write(¤t->mm->mmap_sem);
1120 retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1121 up_write(¤t->mm->mmap_sem);
1129 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1130 struct mmap_arg_struct {
1134 unsigned long flags;
1136 unsigned long offset;
1139 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1141 struct mmap_arg_struct a;
1143 if (copy_from_user(&a, arg, sizeof(a)))
1145 if (a.offset & ~PAGE_MASK)
1148 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1149 a.offset >> PAGE_SHIFT);
1151 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1154 * Some shared mappigns will want the pages marked read-only
1155 * to track write events. If so, we'll downgrade vm_page_prot
1156 * to the private version (using protection_map[] without the
1159 int vma_wants_writenotify(struct vm_area_struct *vma)
1161 unsigned int vm_flags = vma->vm_flags;
1163 /* If it was private or non-writable, the write bit is already clear */
1164 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1167 /* The backer wishes to know when pages are first written to? */
1168 if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1171 /* The open routine did something to the protections already? */
1172 if (pgprot_val(vma->vm_page_prot) !=
1173 pgprot_val(vm_get_page_prot(vm_flags)))
1176 /* Specialty mapping? */
1177 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1180 /* Can the mapping track the dirty pages? */
1181 return vma->vm_file && vma->vm_file->f_mapping &&
1182 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1186 * We account for memory if it's a private writeable mapping,
1187 * not hugepages and VM_NORESERVE wasn't set.
1189 static inline int accountable_mapping(struct file *file, unsigned int vm_flags)
1192 * hugetlb has its own accounting separate from the core VM
1193 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1195 if (file && is_file_hugepages(file))
1198 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1201 unsigned long mmap_region(struct file *file, unsigned long addr,
1202 unsigned long len, unsigned long flags,
1203 unsigned int vm_flags, unsigned long pgoff)
1205 struct mm_struct *mm = current->mm;
1206 struct vm_area_struct *vma, *prev;
1207 int correct_wcount = 0;
1209 struct rb_node **rb_link, *rb_parent;
1210 unsigned long charged = 0;
1211 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL;
1213 /* Clear old maps */
1216 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1217 if (vma && vma->vm_start < addr + len) {
1218 if (do_munmap(mm, addr, len))
1223 /* Check against address space limit. */
1224 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
1228 * Set 'VM_NORESERVE' if we should not account for the
1229 * memory use of this mapping.
1231 if ((flags & MAP_NORESERVE)) {
1232 /* We honor MAP_NORESERVE if allowed to overcommit */
1233 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1234 vm_flags |= VM_NORESERVE;
1236 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1237 if (file && is_file_hugepages(file))
1238 vm_flags |= VM_NORESERVE;
1242 * Private writable mapping: check memory availability
1244 if (accountable_mapping(file, vm_flags)) {
1245 charged = len >> PAGE_SHIFT;
1246 if (security_vm_enough_memory(charged))
1248 vm_flags |= VM_ACCOUNT;
1252 * Can we just expand an old mapping?
1254 vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1259 * Determine the object being mapped and call the appropriate
1260 * specific mapper. the address has already been validated, but
1261 * not unmapped, but the maps are removed from the list.
1263 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1270 vma->vm_start = addr;
1271 vma->vm_end = addr + len;
1272 vma->vm_flags = vm_flags;
1273 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1274 vma->vm_pgoff = pgoff;
1275 INIT_LIST_HEAD(&vma->anon_vma_chain);
1279 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1281 if (vm_flags & VM_DENYWRITE) {
1282 error = deny_write_access(file);
1287 vma->vm_file = file;
1289 error = file->f_op->mmap(file, vma);
1291 goto unmap_and_free_vma;
1292 if (vm_flags & VM_EXECUTABLE)
1293 added_exe_file_vma(mm);
1295 /* Can addr have changed??
1297 * Answer: Yes, several device drivers can do it in their
1298 * f_op->mmap method. -DaveM
1300 addr = vma->vm_start;
1301 pgoff = vma->vm_pgoff;
1302 vm_flags = vma->vm_flags;
1303 } else if (vm_flags & VM_SHARED) {
1304 error = shmem_zero_setup(vma);
1309 if (vma_wants_writenotify(vma)) {
1310 pgprot_t pprot = vma->vm_page_prot;
1312 /* Can vma->vm_page_prot have changed??
1314 * Answer: Yes, drivers may have changed it in their
1315 * f_op->mmap method.
1317 * Ensures that vmas marked as uncached stay that way.
1319 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1320 if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1321 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1324 vma_link(mm, vma, prev, rb_link, rb_parent);
1325 file = vma->vm_file;
1327 /* Once vma denies write, undo our temporary denial count */
1329 atomic_inc(&inode->i_writecount);
1331 perf_event_mmap(vma);
1333 mm->total_vm += len >> PAGE_SHIFT;
1334 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1335 if (vm_flags & VM_LOCKED) {
1336 if (!mlock_vma_pages_range(vma, addr, addr + len))
1337 mm->locked_vm += (len >> PAGE_SHIFT);
1338 } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1339 make_pages_present(addr, addr + len);
1344 atomic_inc(&inode->i_writecount);
1345 vma->vm_file = NULL;
1348 /* Undo any partial mapping done by a device driver. */
1349 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1352 kmem_cache_free(vm_area_cachep, vma);
1355 vm_unacct_memory(charged);
1359 /* Get an address range which is currently unmapped.
1360 * For shmat() with addr=0.
1362 * Ugly calling convention alert:
1363 * Return value with the low bits set means error value,
1365 * if (ret & ~PAGE_MASK)
1368 * This function "knows" that -ENOMEM has the bits set.
1370 #ifndef HAVE_ARCH_UNMAPPED_AREA
1372 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1373 unsigned long len, unsigned long pgoff, unsigned long flags)
1375 struct mm_struct *mm = current->mm;
1376 struct vm_area_struct *vma;
1377 unsigned long start_addr;
1379 if (len > TASK_SIZE)
1382 if (flags & MAP_FIXED)
1386 addr = PAGE_ALIGN(addr);
1387 vma = find_vma(mm, addr);
1388 if (TASK_SIZE - len >= addr &&
1389 (!vma || addr + len <= vma->vm_start))
1392 if (len > mm->cached_hole_size) {
1393 start_addr = addr = mm->free_area_cache;
1395 start_addr = addr = TASK_UNMAPPED_BASE;
1396 mm->cached_hole_size = 0;
1400 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1401 /* At this point: (!vma || addr < vma->vm_end). */
1402 if (TASK_SIZE - len < addr) {
1404 * Start a new search - just in case we missed
1407 if (start_addr != TASK_UNMAPPED_BASE) {
1408 addr = TASK_UNMAPPED_BASE;
1410 mm->cached_hole_size = 0;
1415 if (!vma || addr + len <= vma->vm_start) {
1417 * Remember the place where we stopped the search:
1419 mm->free_area_cache = addr + len;
1422 if (addr + mm->cached_hole_size < vma->vm_start)
1423 mm->cached_hole_size = vma->vm_start - addr;
1429 void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
1432 * Is this a new hole at the lowest possible address?
1434 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1435 mm->free_area_cache = addr;
1436 mm->cached_hole_size = ~0UL;
1441 * This mmap-allocator allocates new areas top-down from below the
1442 * stack's low limit (the base):
1444 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1446 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1447 const unsigned long len, const unsigned long pgoff,
1448 const unsigned long flags)
1450 struct vm_area_struct *vma;
1451 struct mm_struct *mm = current->mm;
1452 unsigned long addr = addr0;
1454 /* requested length too big for entire address space */
1455 if (len > TASK_SIZE)
1458 if (flags & MAP_FIXED)
1461 /* requesting a specific address */
1463 addr = PAGE_ALIGN(addr);
1464 vma = find_vma(mm, addr);
1465 if (TASK_SIZE - len >= addr &&
1466 (!vma || addr + len <= vma->vm_start))
1470 /* check if free_area_cache is useful for us */
1471 if (len <= mm->cached_hole_size) {
1472 mm->cached_hole_size = 0;
1473 mm->free_area_cache = mm->mmap_base;
1476 /* either no address requested or can't fit in requested address hole */
1477 addr = mm->free_area_cache;
1479 /* make sure it can fit in the remaining address space */
1481 vma = find_vma(mm, addr-len);
1482 if (!vma || addr <= vma->vm_start)
1483 /* remember the address as a hint for next time */
1484 return (mm->free_area_cache = addr-len);
1487 if (mm->mmap_base < len)
1490 addr = mm->mmap_base-len;
1494 * Lookup failure means no vma is above this address,
1495 * else if new region fits below vma->vm_start,
1496 * return with success:
1498 vma = find_vma(mm, addr);
1499 if (!vma || addr+len <= vma->vm_start)
1500 /* remember the address as a hint for next time */
1501 return (mm->free_area_cache = addr);
1503 /* remember the largest hole we saw so far */
1504 if (addr + mm->cached_hole_size < vma->vm_start)
1505 mm->cached_hole_size = vma->vm_start - addr;
1507 /* try just below the current vma->vm_start */
1508 addr = vma->vm_start-len;
1509 } while (len < vma->vm_start);
1513 * A failed mmap() very likely causes application failure,
1514 * so fall back to the bottom-up function here. This scenario
1515 * can happen with large stack limits and large mmap()
1518 mm->cached_hole_size = ~0UL;
1519 mm->free_area_cache = TASK_UNMAPPED_BASE;
1520 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1522 * Restore the topdown base:
1524 mm->free_area_cache = mm->mmap_base;
1525 mm->cached_hole_size = ~0UL;
1531 void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
1534 * Is this a new hole at the highest possible address?
1536 if (addr > mm->free_area_cache)
1537 mm->free_area_cache = addr;
1539 /* dont allow allocations above current base */
1540 if (mm->free_area_cache > mm->mmap_base)
1541 mm->free_area_cache = mm->mmap_base;
1545 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1546 unsigned long pgoff, unsigned long flags)
1548 unsigned long (*get_area)(struct file *, unsigned long,
1549 unsigned long, unsigned long, unsigned long);
1551 unsigned long error = arch_mmap_check(addr, len, flags);
1555 /* Careful about overflows.. */
1556 if (len > TASK_SIZE)
1559 get_area = current->mm->get_unmapped_area;
1560 if (file && file->f_op && file->f_op->get_unmapped_area)
1561 get_area = file->f_op->get_unmapped_area;
1562 addr = get_area(file, addr, len, pgoff, flags);
1563 if (IS_ERR_VALUE(addr))
1566 if (addr > TASK_SIZE - len)
1568 if (addr & ~PAGE_MASK)
1571 return arch_rebalance_pgtables(addr, len);
1574 EXPORT_SYMBOL(get_unmapped_area);
1576 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1577 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1579 struct vm_area_struct *vma = NULL;
1582 /* Check the cache first. */
1583 /* (Cache hit rate is typically around 35%.) */
1584 vma = mm->mmap_cache;
1585 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1586 struct rb_node * rb_node;
1588 rb_node = mm->mm_rb.rb_node;
1592 struct vm_area_struct * vma_tmp;
1594 vma_tmp = rb_entry(rb_node,
1595 struct vm_area_struct, vm_rb);
1597 if (vma_tmp->vm_end > addr) {
1599 if (vma_tmp->vm_start <= addr)
1601 rb_node = rb_node->rb_left;
1603 rb_node = rb_node->rb_right;
1606 mm->mmap_cache = vma;
1612 EXPORT_SYMBOL(find_vma);
1614 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1615 struct vm_area_struct *
1616 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1617 struct vm_area_struct **pprev)
1619 struct vm_area_struct *vma = NULL, *prev = NULL;
1620 struct rb_node *rb_node;
1624 /* Guard against addr being lower than the first VMA */
1627 /* Go through the RB tree quickly. */
1628 rb_node = mm->mm_rb.rb_node;
1631 struct vm_area_struct *vma_tmp;
1632 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1634 if (addr < vma_tmp->vm_end) {
1635 rb_node = rb_node->rb_left;
1638 if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1640 rb_node = rb_node->rb_right;
1646 return prev ? prev->vm_next : vma;
1650 * Verify that the stack growth is acceptable and
1651 * update accounting. This is shared with both the
1652 * grow-up and grow-down cases.
1654 static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1656 struct mm_struct *mm = vma->vm_mm;
1657 struct rlimit *rlim = current->signal->rlim;
1658 unsigned long new_start;
1660 /* address space limit tests */
1661 if (!may_expand_vm(mm, grow))
1664 /* Stack limit test */
1665 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1668 /* mlock limit tests */
1669 if (vma->vm_flags & VM_LOCKED) {
1670 unsigned long locked;
1671 unsigned long limit;
1672 locked = mm->locked_vm + grow;
1673 limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1674 limit >>= PAGE_SHIFT;
1675 if (locked > limit && !capable(CAP_IPC_LOCK))
1679 /* Check to ensure the stack will not grow into a hugetlb-only region */
1680 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1682 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1686 * Overcommit.. This must be the final test, as it will
1687 * update security statistics.
1689 if (security_vm_enough_memory_mm(mm, grow))
1692 /* Ok, everything looks good - let it rip */
1693 mm->total_vm += grow;
1694 if (vma->vm_flags & VM_LOCKED)
1695 mm->locked_vm += grow;
1696 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1700 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1702 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1703 * vma is the last one with address > vma->vm_end. Have to extend vma.
1708 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1712 if (!(vma->vm_flags & VM_GROWSUP))
1716 * We must make sure the anon_vma is allocated
1717 * so that the anon_vma locking is not a noop.
1719 if (unlikely(anon_vma_prepare(vma)))
1724 * vma->vm_start/vm_end cannot change under us because the caller
1725 * is required to hold the mmap_sem in read mode. We need the
1726 * anon_vma lock to serialize against concurrent expand_stacks.
1727 * Also guard against wrapping around to address 0.
1729 if (address < PAGE_ALIGN(address+4))
1730 address = PAGE_ALIGN(address+4);
1732 anon_vma_unlock(vma);
1737 /* Somebody else might have raced and expanded it already */
1738 if (address > vma->vm_end) {
1739 unsigned long size, grow;
1741 size = address - vma->vm_start;
1742 grow = (address - vma->vm_end) >> PAGE_SHIFT;
1744 error = acct_stack_growth(vma, size, grow);
1746 vma->vm_end = address;
1748 anon_vma_unlock(vma);
1751 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1754 * vma is the first one with address < vma->vm_start. Have to extend vma.
1756 static int expand_downwards(struct vm_area_struct *vma,
1757 unsigned long address)
1762 * We must make sure the anon_vma is allocated
1763 * so that the anon_vma locking is not a noop.
1765 if (unlikely(anon_vma_prepare(vma)))
1768 address &= PAGE_MASK;
1769 error = security_file_mmap(NULL, 0, 0, 0, address, 1);
1776 * vma->vm_start/vm_end cannot change under us because the caller
1777 * is required to hold the mmap_sem in read mode. We need the
1778 * anon_vma lock to serialize against concurrent expand_stacks.
1781 /* Somebody else might have raced and expanded it already */
1782 if (address < vma->vm_start) {
1783 unsigned long size, grow;
1785 size = vma->vm_end - address;
1786 grow = (vma->vm_start - address) >> PAGE_SHIFT;
1788 error = acct_stack_growth(vma, size, grow);
1790 vma->vm_start = address;
1791 vma->vm_pgoff -= grow;
1794 anon_vma_unlock(vma);
1798 int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address)
1800 return expand_downwards(vma, address);
1803 #ifdef CONFIG_STACK_GROWSUP
1804 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1806 return expand_upwards(vma, address);
1809 struct vm_area_struct *
1810 find_extend_vma(struct mm_struct *mm, unsigned long addr)
1812 struct vm_area_struct *vma, *prev;
1815 vma = find_vma_prev(mm, addr, &prev);
1816 if (vma && (vma->vm_start <= addr))
1818 if (!prev || expand_stack(prev, addr))
1820 if (prev->vm_flags & VM_LOCKED) {
1821 mlock_vma_pages_range(prev, addr, prev->vm_end);
1826 int expand_stack(struct vm_area_struct *vma, unsigned long address)
1828 return expand_downwards(vma, address);
1831 struct vm_area_struct *
1832 find_extend_vma(struct mm_struct * mm, unsigned long addr)
1834 struct vm_area_struct * vma;
1835 unsigned long start;
1838 vma = find_vma(mm,addr);
1841 if (vma->vm_start <= addr)
1843 if (!(vma->vm_flags & VM_GROWSDOWN))
1845 start = vma->vm_start;
1846 if (expand_stack(vma, addr))
1848 if (vma->vm_flags & VM_LOCKED) {
1849 mlock_vma_pages_range(vma, addr, start);
1856 * Ok - we have the memory areas we should free on the vma list,
1857 * so release them, and do the vma updates.
1859 * Called with the mm semaphore held.
1861 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1863 /* Update high watermark before we lower total_vm */
1864 update_hiwater_vm(mm);
1866 long nrpages = vma_pages(vma);
1868 mm->total_vm -= nrpages;
1869 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1870 vma = remove_vma(vma);
1876 * Get rid of page table information in the indicated region.
1878 * Called with the mm semaphore held.
1880 static void unmap_region(struct mm_struct *mm,
1881 struct vm_area_struct *vma, struct vm_area_struct *prev,
1882 unsigned long start, unsigned long end)
1884 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1885 struct mmu_gather *tlb;
1886 unsigned long nr_accounted = 0;
1889 tlb = tlb_gather_mmu(mm, 0);
1890 update_hiwater_rss(mm);
1891 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1892 vm_unacct_memory(nr_accounted);
1893 free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS,
1894 next? next->vm_start: 0);
1895 tlb_finish_mmu(tlb, start, end);
1899 * Create a list of vma's touched by the unmap, removing them from the mm's
1900 * vma list as we go..
1903 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1904 struct vm_area_struct *prev, unsigned long end)
1906 struct vm_area_struct **insertion_point;
1907 struct vm_area_struct *tail_vma = NULL;
1910 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1912 rb_erase(&vma->vm_rb, &mm->mm_rb);
1916 } while (vma && vma->vm_start < end);
1917 *insertion_point = vma;
1918 tail_vma->vm_next = NULL;
1919 if (mm->unmap_area == arch_unmap_area)
1920 addr = prev ? prev->vm_end : mm->mmap_base;
1922 addr = vma ? vma->vm_start : mm->mmap_base;
1923 mm->unmap_area(mm, addr);
1924 mm->mmap_cache = NULL; /* Kill the cache. */
1928 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1929 * munmap path where it doesn't make sense to fail.
1931 static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1932 unsigned long addr, int new_below)
1934 struct mempolicy *pol;
1935 struct vm_area_struct *new;
1938 if (is_vm_hugetlb_page(vma) && (addr &
1939 ~(huge_page_mask(hstate_vma(vma)))))
1942 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1946 /* most fields are the same, copy all, and then fixup */
1949 INIT_LIST_HEAD(&new->anon_vma_chain);
1954 new->vm_start = addr;
1955 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1958 pol = mpol_dup(vma_policy(vma));
1963 vma_set_policy(new, pol);
1965 if (anon_vma_clone(new, vma))
1969 get_file(new->vm_file);
1970 if (vma->vm_flags & VM_EXECUTABLE)
1971 added_exe_file_vma(mm);
1974 if (new->vm_ops && new->vm_ops->open)
1975 new->vm_ops->open(new);
1978 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1979 ((addr - new->vm_start) >> PAGE_SHIFT), new);
1981 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1987 /* Clean everything up if vma_adjust failed. */
1988 new->vm_ops->close(new);
1990 if (vma->vm_flags & VM_EXECUTABLE)
1991 removed_exe_file_vma(mm);
1997 kmem_cache_free(vm_area_cachep, new);
2003 * Split a vma into two pieces at address 'addr', a new vma is allocated
2004 * either for the first part or the tail.
2006 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2007 unsigned long addr, int new_below)
2009 if (mm->map_count >= sysctl_max_map_count)
2012 return __split_vma(mm, vma, addr, new_below);
2015 /* Munmap is split into 2 main parts -- this part which finds
2016 * what needs doing, and the areas themselves, which do the
2017 * work. This now handles partial unmappings.
2018 * Jeremy Fitzhardinge <jeremy@goop.org>
2020 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2023 struct vm_area_struct *vma, *prev, *last;
2025 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2028 if ((len = PAGE_ALIGN(len)) == 0)
2031 /* Find the first overlapping VMA */
2032 vma = find_vma_prev(mm, start, &prev);
2035 /* we have start < vma->vm_end */
2037 /* if it doesn't overlap, we have nothing.. */
2039 if (vma->vm_start >= end)
2043 * If we need to split any vma, do it now to save pain later.
2045 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2046 * unmapped vm_area_struct will remain in use: so lower split_vma
2047 * places tmp vma above, and higher split_vma places tmp vma below.
2049 if (start > vma->vm_start) {
2053 * Make sure that map_count on return from munmap() will
2054 * not exceed its limit; but let map_count go just above
2055 * its limit temporarily, to help free resources as expected.
2057 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2060 error = __split_vma(mm, vma, start, 0);
2066 /* Does it split the last one? */
2067 last = find_vma(mm, end);
2068 if (last && end > last->vm_start) {
2069 int error = __split_vma(mm, last, end, 1);
2073 vma = prev? prev->vm_next: mm->mmap;
2076 * unlock any mlock()ed ranges before detaching vmas
2078 if (mm->locked_vm) {
2079 struct vm_area_struct *tmp = vma;
2080 while (tmp && tmp->vm_start < end) {
2081 if (tmp->vm_flags & VM_LOCKED) {
2082 mm->locked_vm -= vma_pages(tmp);
2083 munlock_vma_pages_all(tmp);
2090 * Remove the vma's, and unmap the actual pages
2092 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2093 unmap_region(mm, vma, prev, start, end);
2095 /* Fix up all other VM information */
2096 remove_vma_list(mm, vma);
2101 EXPORT_SYMBOL(do_munmap);
2103 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2106 struct mm_struct *mm = current->mm;
2108 profile_munmap(addr);
2110 down_write(&mm->mmap_sem);
2111 ret = do_munmap(mm, addr, len);
2112 up_write(&mm->mmap_sem);
2116 static inline void verify_mm_writelocked(struct mm_struct *mm)
2118 #ifdef CONFIG_DEBUG_VM
2119 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2121 up_read(&mm->mmap_sem);
2127 * this is really a simplified "do_mmap". it only handles
2128 * anonymous maps. eventually we may be able to do some
2129 * brk-specific accounting here.
2131 unsigned long do_brk(unsigned long addr, unsigned long len)
2133 struct mm_struct * mm = current->mm;
2134 struct vm_area_struct * vma, * prev;
2135 unsigned long flags;
2136 struct rb_node ** rb_link, * rb_parent;
2137 pgoff_t pgoff = addr >> PAGE_SHIFT;
2140 len = PAGE_ALIGN(len);
2144 error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
2148 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2150 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2151 if (error & ~PAGE_MASK)
2157 if (mm->def_flags & VM_LOCKED) {
2158 unsigned long locked, lock_limit;
2159 locked = len >> PAGE_SHIFT;
2160 locked += mm->locked_vm;
2161 lock_limit = rlimit(RLIMIT_MEMLOCK);
2162 lock_limit >>= PAGE_SHIFT;
2163 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2168 * mm->mmap_sem is required to protect against another thread
2169 * changing the mappings in case we sleep.
2171 verify_mm_writelocked(mm);
2174 * Clear old maps. this also does some error checking for us
2177 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2178 if (vma && vma->vm_start < addr + len) {
2179 if (do_munmap(mm, addr, len))
2184 /* Check against address space limits *after* clearing old maps... */
2185 if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2188 if (mm->map_count > sysctl_max_map_count)
2191 if (security_vm_enough_memory(len >> PAGE_SHIFT))
2194 /* Can we just expand an old private anonymous mapping? */
2195 vma = vma_merge(mm, prev, addr, addr + len, flags,
2196 NULL, NULL, pgoff, NULL);
2201 * create a vma struct for an anonymous mapping
2203 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2205 vm_unacct_memory(len >> PAGE_SHIFT);
2209 INIT_LIST_HEAD(&vma->anon_vma_chain);
2211 vma->vm_start = addr;
2212 vma->vm_end = addr + len;
2213 vma->vm_pgoff = pgoff;
2214 vma->vm_flags = flags;
2215 vma->vm_page_prot = vm_get_page_prot(flags);
2216 vma_link(mm, vma, prev, rb_link, rb_parent);
2218 mm->total_vm += len >> PAGE_SHIFT;
2219 if (flags & VM_LOCKED) {
2220 if (!mlock_vma_pages_range(vma, addr, addr + len))
2221 mm->locked_vm += (len >> PAGE_SHIFT);
2226 EXPORT_SYMBOL(do_brk);
2228 /* Release all mmaps. */
2229 void exit_mmap(struct mm_struct *mm)
2231 struct mmu_gather *tlb;
2232 struct vm_area_struct *vma;
2233 unsigned long nr_accounted = 0;
2236 /* mm's last user has gone, and its about to be pulled down */
2237 mmu_notifier_release(mm);
2239 if (mm->locked_vm) {
2242 if (vma->vm_flags & VM_LOCKED)
2243 munlock_vma_pages_all(vma);
2251 if (!vma) /* Can happen if dup_mmap() received an OOM */
2256 tlb = tlb_gather_mmu(mm, 1);
2257 /* update_hiwater_rss(mm) here? but nobody should be looking */
2258 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2259 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2260 vm_unacct_memory(nr_accounted);
2262 free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0);
2263 tlb_finish_mmu(tlb, 0, end);
2266 * Walk the list again, actually closing and freeing it,
2267 * with preemption enabled, without holding any MM locks.
2270 vma = remove_vma(vma);
2272 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2275 /* Insert vm structure into process list sorted by address
2276 * and into the inode's i_mmap tree. If vm_file is non-NULL
2277 * then i_mmap_lock is taken here.
2279 int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2281 struct vm_area_struct * __vma, * prev;
2282 struct rb_node ** rb_link, * rb_parent;
2285 * The vm_pgoff of a purely anonymous vma should be irrelevant
2286 * until its first write fault, when page's anon_vma and index
2287 * are set. But now set the vm_pgoff it will almost certainly
2288 * end up with (unless mremap moves it elsewhere before that
2289 * first wfault), so /proc/pid/maps tells a consistent story.
2291 * By setting it to reflect the virtual start address of the
2292 * vma, merges and splits can happen in a seamless way, just
2293 * using the existing file pgoff checks and manipulations.
2294 * Similarly in do_mmap_pgoff and in do_brk.
2296 if (!vma->vm_file) {
2297 BUG_ON(vma->anon_vma);
2298 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2300 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2301 if (__vma && __vma->vm_start < vma->vm_end)
2303 if ((vma->vm_flags & VM_ACCOUNT) &&
2304 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2306 vma_link(mm, vma, prev, rb_link, rb_parent);
2311 * Copy the vma structure to a new location in the same mm,
2312 * prior to moving page table entries, to effect an mremap move.
2314 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2315 unsigned long addr, unsigned long len, pgoff_t pgoff)
2317 struct vm_area_struct *vma = *vmap;
2318 unsigned long vma_start = vma->vm_start;
2319 struct mm_struct *mm = vma->vm_mm;
2320 struct vm_area_struct *new_vma, *prev;
2321 struct rb_node **rb_link, *rb_parent;
2322 struct mempolicy *pol;
2325 * If anonymous vma has not yet been faulted, update new pgoff
2326 * to match new location, to increase its chance of merging.
2328 if (!vma->vm_file && !vma->anon_vma)
2329 pgoff = addr >> PAGE_SHIFT;
2331 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2332 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2333 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2336 * Source vma may have been merged into new_vma
2338 if (vma_start >= new_vma->vm_start &&
2339 vma_start < new_vma->vm_end)
2342 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2345 pol = mpol_dup(vma_policy(vma));
2348 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2349 if (anon_vma_clone(new_vma, vma))
2350 goto out_free_mempol;
2351 vma_set_policy(new_vma, pol);
2352 new_vma->vm_start = addr;
2353 new_vma->vm_end = addr + len;
2354 new_vma->vm_pgoff = pgoff;
2355 if (new_vma->vm_file) {
2356 get_file(new_vma->vm_file);
2357 if (vma->vm_flags & VM_EXECUTABLE)
2358 added_exe_file_vma(mm);
2360 if (new_vma->vm_ops && new_vma->vm_ops->open)
2361 new_vma->vm_ops->open(new_vma);
2362 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2370 kmem_cache_free(vm_area_cachep, new_vma);
2375 * Return true if the calling process may expand its vm space by the passed
2378 int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2380 unsigned long cur = mm->total_vm; /* pages */
2383 lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
2385 if (cur + npages > lim)
2391 static int special_mapping_fault(struct vm_area_struct *vma,
2392 struct vm_fault *vmf)
2395 struct page **pages;
2398 * special mappings have no vm_file, and in that case, the mm
2399 * uses vm_pgoff internally. So we have to subtract it from here.
2400 * We are allowed to do this because we are the mm; do not copy
2401 * this code into drivers!
2403 pgoff = vmf->pgoff - vma->vm_pgoff;
2405 for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2409 struct page *page = *pages;
2415 return VM_FAULT_SIGBUS;
2419 * Having a close hook prevents vma merging regardless of flags.
2421 static void special_mapping_close(struct vm_area_struct *vma)
2425 static const struct vm_operations_struct special_mapping_vmops = {
2426 .close = special_mapping_close,
2427 .fault = special_mapping_fault,
2431 * Called with mm->mmap_sem held for writing.
2432 * Insert a new vma covering the given region, with the given flags.
2433 * Its pages are supplied by the given array of struct page *.
2434 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2435 * The region past the last page supplied will always produce SIGBUS.
2436 * The array pointer and the pages it points to are assumed to stay alive
2437 * for as long as this mapping might exist.
2439 int install_special_mapping(struct mm_struct *mm,
2440 unsigned long addr, unsigned long len,
2441 unsigned long vm_flags, struct page **pages)
2443 struct vm_area_struct *vma;
2445 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2446 if (unlikely(vma == NULL))
2449 INIT_LIST_HEAD(&vma->anon_vma_chain);
2451 vma->vm_start = addr;
2452 vma->vm_end = addr + len;
2454 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2455 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2457 vma->vm_ops = &special_mapping_vmops;
2458 vma->vm_private_data = pages;
2460 if (unlikely(insert_vm_struct(mm, vma))) {
2461 kmem_cache_free(vm_area_cachep, vma);
2465 mm->total_vm += len >> PAGE_SHIFT;
2467 perf_event_mmap(vma);
2472 static DEFINE_MUTEX(mm_all_locks_mutex);
2474 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2476 if (!test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2478 * The LSB of head.next can't change from under us
2479 * because we hold the mm_all_locks_mutex.
2481 spin_lock_nest_lock(&anon_vma->lock, &mm->mmap_sem);
2483 * We can safely modify head.next after taking the
2484 * anon_vma->lock. If some other vma in this mm shares
2485 * the same anon_vma we won't take it again.
2487 * No need of atomic instructions here, head.next
2488 * can't change from under us thanks to the
2491 if (__test_and_set_bit(0, (unsigned long *)
2492 &anon_vma->head.next))
2497 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2499 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2501 * AS_MM_ALL_LOCKS can't change from under us because
2502 * we hold the mm_all_locks_mutex.
2504 * Operations on ->flags have to be atomic because
2505 * even if AS_MM_ALL_LOCKS is stable thanks to the
2506 * mm_all_locks_mutex, there may be other cpus
2507 * changing other bitflags in parallel to us.
2509 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2511 spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem);
2516 * This operation locks against the VM for all pte/vma/mm related
2517 * operations that could ever happen on a certain mm. This includes
2518 * vmtruncate, try_to_unmap, and all page faults.
2520 * The caller must take the mmap_sem in write mode before calling
2521 * mm_take_all_locks(). The caller isn't allowed to release the
2522 * mmap_sem until mm_drop_all_locks() returns.
2524 * mmap_sem in write mode is required in order to block all operations
2525 * that could modify pagetables and free pages without need of
2526 * altering the vma layout (for example populate_range() with
2527 * nonlinear vmas). It's also needed in write mode to avoid new
2528 * anon_vmas to be associated with existing vmas.
2530 * A single task can't take more than one mm_take_all_locks() in a row
2531 * or it would deadlock.
2533 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2534 * mapping->flags avoid to take the same lock twice, if more than one
2535 * vma in this mm is backed by the same anon_vma or address_space.
2537 * We can take all the locks in random order because the VM code
2538 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2539 * takes more than one of them in a row. Secondly we're protected
2540 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2542 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2543 * that may have to take thousand of locks.
2545 * mm_take_all_locks() can fail if it's interrupted by signals.
2547 int mm_take_all_locks(struct mm_struct *mm)
2549 struct vm_area_struct *vma;
2550 struct anon_vma_chain *avc;
2553 BUG_ON(down_read_trylock(&mm->mmap_sem));
2555 mutex_lock(&mm_all_locks_mutex);
2557 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2558 if (signal_pending(current))
2560 if (vma->vm_file && vma->vm_file->f_mapping)
2561 vm_lock_mapping(mm, vma->vm_file->f_mapping);
2564 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2565 if (signal_pending(current))
2568 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2569 vm_lock_anon_vma(mm, avc->anon_vma);
2576 mm_drop_all_locks(mm);
2581 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2583 if (test_bit(0, (unsigned long *) &anon_vma->head.next)) {
2585 * The LSB of head.next can't change to 0 from under
2586 * us because we hold the mm_all_locks_mutex.
2588 * We must however clear the bitflag before unlocking
2589 * the vma so the users using the anon_vma->head will
2590 * never see our bitflag.
2592 * No need of atomic instructions here, head.next
2593 * can't change from under us until we release the
2596 if (!__test_and_clear_bit(0, (unsigned long *)
2597 &anon_vma->head.next))
2599 spin_unlock(&anon_vma->lock);
2603 static void vm_unlock_mapping(struct address_space *mapping)
2605 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2607 * AS_MM_ALL_LOCKS can't change to 0 from under us
2608 * because we hold the mm_all_locks_mutex.
2610 spin_unlock(&mapping->i_mmap_lock);
2611 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2618 * The mmap_sem cannot be released by the caller until
2619 * mm_drop_all_locks() returns.
2621 void mm_drop_all_locks(struct mm_struct *mm)
2623 struct vm_area_struct *vma;
2624 struct anon_vma_chain *avc;
2626 BUG_ON(down_read_trylock(&mm->mmap_sem));
2627 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2629 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2631 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2632 vm_unlock_anon_vma(avc->anon_vma);
2633 if (vma->vm_file && vma->vm_file->f_mapping)
2634 vm_unlock_mapping(vma->vm_file->f_mapping);
2637 mutex_unlock(&mm_all_locks_mutex);
2641 * initialise the VMA slab
2643 void __init mmap_init(void)
2647 ret = percpu_counter_init(&vm_committed_as, 0);