6 * Address space accounting code <alan@redhat.com>
9 #include <linux/slab.h>
11 #include <linux/shm.h>
12 #include <linux/mman.h>
13 #include <linux/pagemap.h>
14 #include <linux/swap.h>
15 #include <linux/syscalls.h>
16 #include <linux/init.h>
17 #include <linux/file.h>
19 #include <linux/personality.h>
20 #include <linux/security.h>
21 #include <linux/hugetlb.h>
22 #include <linux/profile.h>
23 #include <linux/module.h>
24 #include <linux/mount.h>
25 #include <linux/mempolicy.h>
26 #include <linux/rmap.h>
28 #include <asm/uaccess.h>
29 #include <asm/cacheflush.h>
32 static void unmap_region(struct mm_struct
*mm
,
33 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
34 unsigned long start
, unsigned long end
);
37 * WARNING: the debugging will use recursive algorithms so never enable this
38 * unless you know what you are doing.
42 /* description of effects of mapping type and prot in current implementation.
43 * this is due to the limited x86 page protection hardware. The expected
44 * behavior is in parens:
47 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
48 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
49 * w: (no) no w: (no) no w: (yes) yes w: (no) no
50 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
52 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
53 * w: (no) no w: (no) no w: (copy) copy w: (no) no
54 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
57 pgprot_t protection_map
[16] = {
58 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
59 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
62 int sysctl_overcommit_memory
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
63 int sysctl_overcommit_ratio
= 50; /* default is 50% */
64 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
65 atomic_t vm_committed_space
= ATOMIC_INIT(0);
68 * Check that a process has enough memory to allocate a new virtual
69 * mapping. 0 means there is enough memory for the allocation to
70 * succeed and -ENOMEM implies there is not.
72 * We currently support three overcommit policies, which are set via the
73 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
75 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
76 * Additional code 2002 Jul 20 by Robert Love.
78 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
80 * Note this is a helper function intended to be used by LSMs which
81 * wish to use this logic.
83 int __vm_enough_memory(long pages
, int cap_sys_admin
)
85 unsigned long free
, allowed
;
87 vm_acct_memory(pages
);
90 * Sometimes we want to use more memory than we have
92 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
95 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
98 free
= get_page_cache_size();
99 free
+= nr_swap_pages
;
102 * Any slabs which are created with the
103 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
104 * which are reclaimable, under pressure. The dentry
105 * cache and most inode caches should fall into this
107 free
+= atomic_read(&slab_reclaim_pages
);
110 * Leave the last 3% for root
119 * nr_free_pages() is very expensive on large systems,
120 * only call if we're about to fail.
129 vm_unacct_memory(pages
);
133 allowed
= (totalram_pages
- hugetlb_total_pages())
134 * sysctl_overcommit_ratio
/ 100;
136 * Leave the last 3% for root
139 allowed
-= allowed
/ 32;
140 allowed
+= total_swap_pages
;
142 /* Don't let a single process grow too big:
143 leave 3% of the size of this process for other processes */
144 allowed
-= current
->mm
->total_vm
/ 32;
147 * cast `allowed' as a signed long because vm_committed_space
148 * sometimes has a negative value
150 if (atomic_read(&vm_committed_space
) < (long)allowed
)
153 vm_unacct_memory(pages
);
158 EXPORT_SYMBOL(sysctl_overcommit_memory
);
159 EXPORT_SYMBOL(sysctl_overcommit_ratio
);
160 EXPORT_SYMBOL(sysctl_max_map_count
);
161 EXPORT_SYMBOL(vm_committed_space
);
162 EXPORT_SYMBOL(__vm_enough_memory
);
165 * Requires inode->i_mapping->i_mmap_lock
167 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
168 struct file
*file
, struct address_space
*mapping
)
170 if (vma
->vm_flags
& VM_DENYWRITE
)
171 atomic_inc(&file
->f_dentry
->d_inode
->i_writecount
);
172 if (vma
->vm_flags
& VM_SHARED
)
173 mapping
->i_mmap_writable
--;
175 flush_dcache_mmap_lock(mapping
);
176 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
177 list_del_init(&vma
->shared
.vm_set
.list
);
179 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
180 flush_dcache_mmap_unlock(mapping
);
184 * Unlink a file-based vm structure from its prio_tree, to hide
185 * vma from rmap and vmtruncate before freeing its page tables.
187 void unlink_file_vma(struct vm_area_struct
*vma
)
189 struct file
*file
= vma
->vm_file
;
192 struct address_space
*mapping
= file
->f_mapping
;
193 spin_lock(&mapping
->i_mmap_lock
);
194 __remove_shared_vm_struct(vma
, file
, mapping
);
195 spin_unlock(&mapping
->i_mmap_lock
);
200 * Close a vm structure and free it, returning the next.
202 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
204 struct vm_area_struct
*next
= vma
->vm_next
;
207 * Hide vma from rmap and vmtruncate before freeing page tables:
208 * to be moved into free_pgtables once page_table_lock is lifted
209 * from it, but until then lock ordering forbids that move.
211 anon_vma_unlink(vma
);
212 unlink_file_vma(vma
);
215 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
216 vma
->vm_ops
->close(vma
);
219 mpol_free(vma_policy(vma
));
220 kmem_cache_free(vm_area_cachep
, vma
);
224 asmlinkage
unsigned long sys_brk(unsigned long brk
)
226 unsigned long rlim
, retval
;
227 unsigned long newbrk
, oldbrk
;
228 struct mm_struct
*mm
= current
->mm
;
230 down_write(&mm
->mmap_sem
);
232 if (brk
< mm
->end_code
)
234 newbrk
= PAGE_ALIGN(brk
);
235 oldbrk
= PAGE_ALIGN(mm
->brk
);
236 if (oldbrk
== newbrk
)
239 /* Always allow shrinking brk. */
240 if (brk
<= mm
->brk
) {
241 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
246 /* Check against rlimit.. */
247 rlim
= current
->signal
->rlim
[RLIMIT_DATA
].rlim_cur
;
248 if (rlim
< RLIM_INFINITY
&& brk
- mm
->start_data
> rlim
)
251 /* Check against existing mmap mappings. */
252 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
255 /* Ok, looks good - let it rip. */
256 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
262 up_write(&mm
->mmap_sem
);
267 static int browse_rb(struct rb_root
*root
)
270 struct rb_node
*nd
, *pn
= NULL
;
271 unsigned long prev
= 0, pend
= 0;
273 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
274 struct vm_area_struct
*vma
;
275 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
276 if (vma
->vm_start
< prev
)
277 printk("vm_start %lx prev %lx\n", vma
->vm_start
, prev
), i
= -1;
278 if (vma
->vm_start
< pend
)
279 printk("vm_start %lx pend %lx\n", vma
->vm_start
, pend
);
280 if (vma
->vm_start
> vma
->vm_end
)
281 printk("vm_end %lx < vm_start %lx\n", vma
->vm_end
, vma
->vm_start
);
286 for (nd
= pn
; nd
; nd
= rb_prev(nd
)) {
290 printk("backwards %d, forwards %d\n", j
, i
), i
= 0;
294 void validate_mm(struct mm_struct
*mm
)
298 struct vm_area_struct
*tmp
= mm
->mmap
;
303 if (i
!= mm
->map_count
)
304 printk("map_count %d vm_next %d\n", mm
->map_count
, i
), bug
= 1;
305 i
= browse_rb(&mm
->mm_rb
);
306 if (i
!= mm
->map_count
)
307 printk("map_count %d rb %d\n", mm
->map_count
, i
), bug
= 1;
312 #define validate_mm(mm) do { } while (0)
315 static struct vm_area_struct
*
316 find_vma_prepare(struct mm_struct
*mm
, unsigned long addr
,
317 struct vm_area_struct
**pprev
, struct rb_node
***rb_link
,
318 struct rb_node
** rb_parent
)
320 struct vm_area_struct
* vma
;
321 struct rb_node
** __rb_link
, * __rb_parent
, * rb_prev
;
323 __rb_link
= &mm
->mm_rb
.rb_node
;
324 rb_prev
= __rb_parent
= NULL
;
328 struct vm_area_struct
*vma_tmp
;
330 __rb_parent
= *__rb_link
;
331 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
333 if (vma_tmp
->vm_end
> addr
) {
335 if (vma_tmp
->vm_start
<= addr
)
337 __rb_link
= &__rb_parent
->rb_left
;
339 rb_prev
= __rb_parent
;
340 __rb_link
= &__rb_parent
->rb_right
;
346 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
347 *rb_link
= __rb_link
;
348 *rb_parent
= __rb_parent
;
353 __vma_link_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
354 struct vm_area_struct
*prev
, struct rb_node
*rb_parent
)
357 vma
->vm_next
= prev
->vm_next
;
362 vma
->vm_next
= rb_entry(rb_parent
,
363 struct vm_area_struct
, vm_rb
);
369 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
370 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
372 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
373 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
376 static inline void __vma_link_file(struct vm_area_struct
*vma
)
382 struct address_space
*mapping
= file
->f_mapping
;
384 if (vma
->vm_flags
& VM_DENYWRITE
)
385 atomic_dec(&file
->f_dentry
->d_inode
->i_writecount
);
386 if (vma
->vm_flags
& VM_SHARED
)
387 mapping
->i_mmap_writable
++;
389 flush_dcache_mmap_lock(mapping
);
390 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
391 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
393 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
394 flush_dcache_mmap_unlock(mapping
);
399 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
400 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
401 struct rb_node
*rb_parent
)
403 __vma_link_list(mm
, vma
, prev
, rb_parent
);
404 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
405 __anon_vma_link(vma
);
408 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
409 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
410 struct rb_node
*rb_parent
)
412 struct address_space
*mapping
= NULL
;
415 mapping
= vma
->vm_file
->f_mapping
;
418 spin_lock(&mapping
->i_mmap_lock
);
419 vma
->vm_truncate_count
= mapping
->truncate_count
;
423 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
424 __vma_link_file(vma
);
426 anon_vma_unlock(vma
);
428 spin_unlock(&mapping
->i_mmap_lock
);
435 * Helper for vma_adjust in the split_vma insert case:
436 * insert vm structure into list and rbtree and anon_vma,
437 * but it has already been inserted into prio_tree earlier.
440 __insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
442 struct vm_area_struct
* __vma
, * prev
;
443 struct rb_node
** rb_link
, * rb_parent
;
445 __vma
= find_vma_prepare(mm
, vma
->vm_start
,&prev
, &rb_link
, &rb_parent
);
446 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
448 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
453 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
454 struct vm_area_struct
*prev
)
456 prev
->vm_next
= vma
->vm_next
;
457 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
458 if (mm
->mmap_cache
== vma
)
459 mm
->mmap_cache
= prev
;
463 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
464 * is already present in an i_mmap tree without adjusting the tree.
465 * The following helper function should be used when such adjustments
466 * are necessary. The "insert" vma (if any) is to be inserted
467 * before we drop the necessary locks.
469 void vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
470 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
472 struct mm_struct
*mm
= vma
->vm_mm
;
473 struct vm_area_struct
*next
= vma
->vm_next
;
474 struct vm_area_struct
*importer
= NULL
;
475 struct address_space
*mapping
= NULL
;
476 struct prio_tree_root
*root
= NULL
;
477 struct file
*file
= vma
->vm_file
;
478 struct anon_vma
*anon_vma
= NULL
;
479 long adjust_next
= 0;
482 if (next
&& !insert
) {
483 if (end
>= next
->vm_end
) {
485 * vma expands, overlapping all the next, and
486 * perhaps the one after too (mprotect case 6).
488 again
: remove_next
= 1 + (end
> next
->vm_end
);
490 anon_vma
= next
->anon_vma
;
492 } else if (end
> next
->vm_start
) {
494 * vma expands, overlapping part of the next:
495 * mprotect case 5 shifting the boundary up.
497 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
498 anon_vma
= next
->anon_vma
;
500 } else if (end
< vma
->vm_end
) {
502 * vma shrinks, and !insert tells it's not
503 * split_vma inserting another: so it must be
504 * mprotect case 4 shifting the boundary down.
506 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
507 anon_vma
= next
->anon_vma
;
513 mapping
= file
->f_mapping
;
514 if (!(vma
->vm_flags
& VM_NONLINEAR
))
515 root
= &mapping
->i_mmap
;
516 spin_lock(&mapping
->i_mmap_lock
);
518 vma
->vm_truncate_count
!= next
->vm_truncate_count
) {
520 * unmap_mapping_range might be in progress:
521 * ensure that the expanding vma is rescanned.
523 importer
->vm_truncate_count
= 0;
526 insert
->vm_truncate_count
= vma
->vm_truncate_count
;
528 * Put into prio_tree now, so instantiated pages
529 * are visible to arm/parisc __flush_dcache_page
530 * throughout; but we cannot insert into address
531 * space until vma start or end is updated.
533 __vma_link_file(insert
);
538 * When changing only vma->vm_end, we don't really need
539 * anon_vma lock: but is that case worth optimizing out?
542 anon_vma
= vma
->anon_vma
;
544 spin_lock(&anon_vma
->lock
);
546 * Easily overlooked: when mprotect shifts the boundary,
547 * make sure the expanding vma has anon_vma set if the
548 * shrinking vma had, to cover any anon pages imported.
550 if (importer
&& !importer
->anon_vma
) {
551 importer
->anon_vma
= anon_vma
;
552 __anon_vma_link(importer
);
557 flush_dcache_mmap_lock(mapping
);
558 vma_prio_tree_remove(vma
, root
);
560 vma_prio_tree_remove(next
, root
);
563 vma
->vm_start
= start
;
565 vma
->vm_pgoff
= pgoff
;
567 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
568 next
->vm_pgoff
+= adjust_next
;
573 vma_prio_tree_insert(next
, root
);
574 vma_prio_tree_insert(vma
, root
);
575 flush_dcache_mmap_unlock(mapping
);
580 * vma_merge has merged next into vma, and needs
581 * us to remove next before dropping the locks.
583 __vma_unlink(mm
, next
, vma
);
585 __remove_shared_vm_struct(next
, file
, mapping
);
587 __anon_vma_merge(vma
, next
);
590 * split_vma has split insert from vma, and needs
591 * us to insert it before dropping the locks
592 * (it may either follow vma or precede it).
594 __insert_vm_struct(mm
, insert
);
598 spin_unlock(&anon_vma
->lock
);
600 spin_unlock(&mapping
->i_mmap_lock
);
606 mpol_free(vma_policy(next
));
607 kmem_cache_free(vm_area_cachep
, next
);
609 * In mprotect's case 6 (see comments on vma_merge),
610 * we must remove another next too. It would clutter
611 * up the code too much to do both in one go.
613 if (remove_next
== 2) {
623 * If the vma has a ->close operation then the driver probably needs to release
624 * per-vma resources, so we don't attempt to merge those.
626 #define VM_SPECIAL (VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED)
628 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
629 struct file
*file
, unsigned long vm_flags
)
631 if (vma
->vm_flags
!= vm_flags
)
633 if (vma
->vm_file
!= file
)
635 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
640 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
641 struct anon_vma
*anon_vma2
)
643 return !anon_vma1
|| !anon_vma2
|| (anon_vma1
== anon_vma2
);
647 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
648 * in front of (at a lower virtual address and file offset than) the vma.
650 * We cannot merge two vmas if they have differently assigned (non-NULL)
651 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
653 * We don't check here for the merged mmap wrapping around the end of pagecache
654 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
655 * wrap, nor mmaps which cover the final page at index -1UL.
658 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
659 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
661 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
662 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
663 if (vma
->vm_pgoff
== vm_pgoff
)
670 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
671 * beyond (at a higher virtual address and file offset than) the vma.
673 * We cannot merge two vmas if they have differently assigned (non-NULL)
674 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
677 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
678 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
680 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
681 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
683 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
684 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
691 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
692 * whether that can be merged with its predecessor or its successor.
693 * Or both (it neatly fills a hole).
695 * In most cases - when called for mmap, brk or mremap - [addr,end) is
696 * certain not to be mapped by the time vma_merge is called; but when
697 * called for mprotect, it is certain to be already mapped (either at
698 * an offset within prev, or at the start of next), and the flags of
699 * this area are about to be changed to vm_flags - and the no-change
700 * case has already been eliminated.
702 * The following mprotect cases have to be considered, where AAAA is
703 * the area passed down from mprotect_fixup, never extending beyond one
704 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
706 * AAAA AAAA AAAA AAAA
707 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
708 * cannot merge might become might become might become
709 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
710 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
711 * mremap move: PPPPNNNNNNNN 8
713 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
714 * might become case 1 below case 2 below case 3 below
716 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
717 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
719 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
720 struct vm_area_struct
*prev
, unsigned long addr
,
721 unsigned long end
, unsigned long vm_flags
,
722 struct anon_vma
*anon_vma
, struct file
*file
,
723 pgoff_t pgoff
, struct mempolicy
*policy
)
725 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
726 struct vm_area_struct
*area
, *next
;
729 * We later require that vma->vm_flags == vm_flags,
730 * so this tests vma->vm_flags & VM_SPECIAL, too.
732 if (vm_flags
& VM_SPECIAL
)
736 next
= prev
->vm_next
;
740 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
741 next
= next
->vm_next
;
744 * Can it merge with the predecessor?
746 if (prev
&& prev
->vm_end
== addr
&&
747 mpol_equal(vma_policy(prev
), policy
) &&
748 can_vma_merge_after(prev
, vm_flags
,
749 anon_vma
, file
, pgoff
)) {
751 * OK, it can. Can we now merge in the successor as well?
753 if (next
&& end
== next
->vm_start
&&
754 mpol_equal(policy
, vma_policy(next
)) &&
755 can_vma_merge_before(next
, vm_flags
,
756 anon_vma
, file
, pgoff
+pglen
) &&
757 is_mergeable_anon_vma(prev
->anon_vma
,
760 vma_adjust(prev
, prev
->vm_start
,
761 next
->vm_end
, prev
->vm_pgoff
, NULL
);
762 } else /* cases 2, 5, 7 */
763 vma_adjust(prev
, prev
->vm_start
,
764 end
, prev
->vm_pgoff
, NULL
);
769 * Can this new request be merged in front of next?
771 if (next
&& end
== next
->vm_start
&&
772 mpol_equal(policy
, vma_policy(next
)) &&
773 can_vma_merge_before(next
, vm_flags
,
774 anon_vma
, file
, pgoff
+pglen
)) {
775 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
776 vma_adjust(prev
, prev
->vm_start
,
777 addr
, prev
->vm_pgoff
, NULL
);
778 else /* cases 3, 8 */
779 vma_adjust(area
, addr
, next
->vm_end
,
780 next
->vm_pgoff
- pglen
, NULL
);
788 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
789 * neighbouring vmas for a suitable anon_vma, before it goes off
790 * to allocate a new anon_vma. It checks because a repetitive
791 * sequence of mprotects and faults may otherwise lead to distinct
792 * anon_vmas being allocated, preventing vma merge in subsequent
795 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
797 struct vm_area_struct
*near
;
798 unsigned long vm_flags
;
805 * Since only mprotect tries to remerge vmas, match flags
806 * which might be mprotected into each other later on.
807 * Neither mlock nor madvise tries to remerge at present,
808 * so leave their flags as obstructing a merge.
810 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
811 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
813 if (near
->anon_vma
&& vma
->vm_end
== near
->vm_start
&&
814 mpol_equal(vma_policy(vma
), vma_policy(near
)) &&
815 can_vma_merge_before(near
, vm_flags
,
816 NULL
, vma
->vm_file
, vma
->vm_pgoff
+
817 ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
)))
818 return near
->anon_vma
;
821 * It is potentially slow to have to call find_vma_prev here.
822 * But it's only on the first write fault on the vma, not
823 * every time, and we could devise a way to avoid it later
824 * (e.g. stash info in next's anon_vma_node when assigning
825 * an anon_vma, or when trying vma_merge). Another time.
827 if (find_vma_prev(vma
->vm_mm
, vma
->vm_start
, &near
) != vma
)
832 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
833 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
835 if (near
->anon_vma
&& near
->vm_end
== vma
->vm_start
&&
836 mpol_equal(vma_policy(near
), vma_policy(vma
)) &&
837 can_vma_merge_after(near
, vm_flags
,
838 NULL
, vma
->vm_file
, vma
->vm_pgoff
))
839 return near
->anon_vma
;
842 * There's no absolute need to look only at touching neighbours:
843 * we could search further afield for "compatible" anon_vmas.
844 * But it would probably just be a waste of time searching,
845 * or lead to too many vmas hanging off the same anon_vma.
846 * We're trying to allow mprotect remerging later on,
847 * not trying to minimize memory used for anon_vmas.
852 #ifdef CONFIG_PROC_FS
853 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
854 struct file
*file
, long pages
)
856 const unsigned long stack_flags
857 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
859 #ifdef CONFIG_HUGETLB
860 if (flags
& VM_HUGETLB
) {
861 if (!(flags
& VM_DONTCOPY
))
862 mm
->shared_vm
+= pages
;
865 #endif /* CONFIG_HUGETLB */
868 mm
->shared_vm
+= pages
;
869 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
870 mm
->exec_vm
+= pages
;
871 } else if (flags
& stack_flags
)
872 mm
->stack_vm
+= pages
;
873 if (flags
& (VM_RESERVED
|VM_IO
))
874 mm
->reserved_vm
+= pages
;
876 #endif /* CONFIG_PROC_FS */
879 * The caller must hold down_write(current->mm->mmap_sem).
882 unsigned long do_mmap_pgoff(struct file
* file
, unsigned long addr
,
883 unsigned long len
, unsigned long prot
,
884 unsigned long flags
, unsigned long pgoff
)
886 struct mm_struct
* mm
= current
->mm
;
887 struct vm_area_struct
* vma
, * prev
;
889 unsigned int vm_flags
;
890 int correct_wcount
= 0;
892 struct rb_node
** rb_link
, * rb_parent
;
894 unsigned long charged
= 0, reqprot
= prot
;
897 if (is_file_hugepages(file
))
900 if (!file
->f_op
|| !file
->f_op
->mmap
)
903 if ((prot
& PROT_EXEC
) &&
904 (file
->f_vfsmnt
->mnt_flags
& MNT_NOEXEC
))
908 * Does the application expect PROT_READ to imply PROT_EXEC?
910 * (the exception is when the underlying filesystem is noexec
911 * mounted, in which case we dont add PROT_EXEC.)
913 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
914 if (!(file
&& (file
->f_vfsmnt
->mnt_flags
& MNT_NOEXEC
)))
920 /* Careful about overflows.. */
921 len
= PAGE_ALIGN(len
);
922 if (!len
|| len
> TASK_SIZE
)
925 /* offset overflow? */
926 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
929 /* Too many mappings? */
930 if (mm
->map_count
> sysctl_max_map_count
)
933 /* Obtain the address to map to. we verify (or select) it and ensure
934 * that it represents a valid section of the address space.
936 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
937 if (addr
& ~PAGE_MASK
)
940 /* Do simple checking here so the lower-level routines won't have
941 * to. we assume access permissions have been handled by the open
942 * of the memory object, so we don't do any here.
944 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
945 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
947 if (flags
& MAP_LOCKED
) {
950 vm_flags
|= VM_LOCKED
;
952 /* mlock MCL_FUTURE? */
953 if (vm_flags
& VM_LOCKED
) {
954 unsigned long locked
, lock_limit
;
955 locked
= len
>> PAGE_SHIFT
;
956 locked
+= mm
->locked_vm
;
957 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
958 lock_limit
>>= PAGE_SHIFT
;
959 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
963 inode
= file
? file
->f_dentry
->d_inode
: NULL
;
966 switch (flags
& MAP_TYPE
) {
968 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
972 * Make sure we don't allow writing to an append-only
975 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
979 * Make sure there are no mandatory locks on the file.
981 if (locks_verify_locked(inode
))
984 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
985 if (!(file
->f_mode
& FMODE_WRITE
))
986 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
990 if (!(file
->f_mode
& FMODE_READ
))
998 switch (flags
& MAP_TYPE
) {
1000 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1004 * Set pgoff according to addr for anon_vma.
1006 pgoff
= addr
>> PAGE_SHIFT
;
1013 error
= security_file_mmap(file
, reqprot
, prot
, flags
);
1017 /* Clear old maps */
1020 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1021 if (vma
&& vma
->vm_start
< addr
+ len
) {
1022 if (do_munmap(mm
, addr
, len
))
1027 /* Check against address space limit. */
1028 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1031 if (accountable
&& (!(flags
& MAP_NORESERVE
) ||
1032 sysctl_overcommit_memory
== OVERCOMMIT_NEVER
)) {
1033 if (vm_flags
& VM_SHARED
) {
1034 /* Check memory availability in shmem_file_setup? */
1035 vm_flags
|= VM_ACCOUNT
;
1036 } else if (vm_flags
& VM_WRITE
) {
1038 * Private writable mapping: check memory availability
1040 charged
= len
>> PAGE_SHIFT
;
1041 if (security_vm_enough_memory(charged
))
1043 vm_flags
|= VM_ACCOUNT
;
1048 * Can we just expand an old private anonymous mapping?
1049 * The VM_SHARED test is necessary because shmem_zero_setup
1050 * will create the file object for a shared anonymous map below.
1052 if (!file
&& !(vm_flags
& VM_SHARED
) &&
1053 vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
,
1054 NULL
, NULL
, pgoff
, NULL
))
1058 * Determine the object being mapped and call the appropriate
1059 * specific mapper. the address has already been validated, but
1060 * not unmapped, but the maps are removed from the list.
1062 vma
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
1067 memset(vma
, 0, sizeof(*vma
));
1070 vma
->vm_start
= addr
;
1071 vma
->vm_end
= addr
+ len
;
1072 vma
->vm_flags
= vm_flags
;
1073 vma
->vm_page_prot
= protection_map
[vm_flags
& 0x0f];
1074 vma
->vm_pgoff
= pgoff
;
1078 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1080 if (vm_flags
& VM_DENYWRITE
) {
1081 error
= deny_write_access(file
);
1086 vma
->vm_file
= file
;
1088 error
= file
->f_op
->mmap(file
, vma
);
1090 goto unmap_and_free_vma
;
1091 if ((vma
->vm_flags
& (VM_SHARED
| VM_WRITE
| VM_RESERVED
))
1092 == (VM_WRITE
| VM_RESERVED
)) {
1093 printk(KERN_WARNING
"program %s is using MAP_PRIVATE, "
1094 "PROT_WRITE mmap of VM_RESERVED memory, which "
1095 "is deprecated. Please report this to "
1096 "linux-kernel@vger.kernel.org\n",current
->comm
);
1097 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
1098 vma
->vm_ops
->close(vma
);
1100 goto unmap_and_free_vma
;
1102 } else if (vm_flags
& VM_SHARED
) {
1103 error
= shmem_zero_setup(vma
);
1108 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1109 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1110 * that memory reservation must be checked; but that reservation
1111 * belongs to shared memory object, not to vma: so now clear it.
1113 if ((vm_flags
& (VM_SHARED
|VM_ACCOUNT
)) == (VM_SHARED
|VM_ACCOUNT
))
1114 vma
->vm_flags
&= ~VM_ACCOUNT
;
1116 /* Can addr have changed??
1118 * Answer: Yes, several device drivers can do it in their
1119 * f_op->mmap method. -DaveM
1121 addr
= vma
->vm_start
;
1122 pgoff
= vma
->vm_pgoff
;
1123 vm_flags
= vma
->vm_flags
;
1125 if (!file
|| !vma_merge(mm
, prev
, addr
, vma
->vm_end
,
1126 vma
->vm_flags
, NULL
, file
, pgoff
, vma_policy(vma
))) {
1127 file
= vma
->vm_file
;
1128 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1130 atomic_inc(&inode
->i_writecount
);
1134 atomic_inc(&inode
->i_writecount
);
1137 mpol_free(vma_policy(vma
));
1138 kmem_cache_free(vm_area_cachep
, vma
);
1141 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1142 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1143 if (vm_flags
& VM_LOCKED
) {
1144 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
1145 make_pages_present(addr
, addr
+ len
);
1147 if (flags
& MAP_POPULATE
) {
1148 up_write(&mm
->mmap_sem
);
1149 sys_remap_file_pages(addr
, len
, 0,
1150 pgoff
, flags
& MAP_NONBLOCK
);
1151 down_write(&mm
->mmap_sem
);
1157 atomic_inc(&inode
->i_writecount
);
1158 vma
->vm_file
= NULL
;
1161 /* Undo any partial mapping done by a device driver. */
1162 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1165 kmem_cache_free(vm_area_cachep
, vma
);
1168 vm_unacct_memory(charged
);
1172 EXPORT_SYMBOL(do_mmap_pgoff
);
1174 /* Get an address range which is currently unmapped.
1175 * For shmat() with addr=0.
1177 * Ugly calling convention alert:
1178 * Return value with the low bits set means error value,
1180 * if (ret & ~PAGE_MASK)
1183 * This function "knows" that -ENOMEM has the bits set.
1185 #ifndef HAVE_ARCH_UNMAPPED_AREA
1187 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1188 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1190 struct mm_struct
*mm
= current
->mm
;
1191 struct vm_area_struct
*vma
;
1192 unsigned long start_addr
;
1194 if (len
> TASK_SIZE
)
1198 addr
= PAGE_ALIGN(addr
);
1199 vma
= find_vma(mm
, addr
);
1200 if (TASK_SIZE
- len
>= addr
&&
1201 (!vma
|| addr
+ len
<= vma
->vm_start
))
1204 if (len
> mm
->cached_hole_size
) {
1205 start_addr
= addr
= mm
->free_area_cache
;
1207 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1208 mm
->cached_hole_size
= 0;
1212 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1213 /* At this point: (!vma || addr < vma->vm_end). */
1214 if (TASK_SIZE
- len
< addr
) {
1216 * Start a new search - just in case we missed
1219 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1220 addr
= TASK_UNMAPPED_BASE
;
1222 mm
->cached_hole_size
= 0;
1227 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1229 * Remember the place where we stopped the search:
1231 mm
->free_area_cache
= addr
+ len
;
1234 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1235 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1241 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1244 * Is this a new hole at the lowest possible address?
1246 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
) {
1247 mm
->free_area_cache
= addr
;
1248 mm
->cached_hole_size
= ~0UL;
1253 * This mmap-allocator allocates new areas top-down from below the
1254 * stack's low limit (the base):
1256 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1258 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1259 const unsigned long len
, const unsigned long pgoff
,
1260 const unsigned long flags
)
1262 struct vm_area_struct
*vma
;
1263 struct mm_struct
*mm
= current
->mm
;
1264 unsigned long addr
= addr0
;
1266 /* requested length too big for entire address space */
1267 if (len
> TASK_SIZE
)
1270 /* requesting a specific address */
1272 addr
= PAGE_ALIGN(addr
);
1273 vma
= find_vma(mm
, addr
);
1274 if (TASK_SIZE
- len
>= addr
&&
1275 (!vma
|| addr
+ len
<= vma
->vm_start
))
1279 /* check if free_area_cache is useful for us */
1280 if (len
<= mm
->cached_hole_size
) {
1281 mm
->cached_hole_size
= 0;
1282 mm
->free_area_cache
= mm
->mmap_base
;
1285 /* either no address requested or can't fit in requested address hole */
1286 addr
= mm
->free_area_cache
;
1288 /* make sure it can fit in the remaining address space */
1290 vma
= find_vma(mm
, addr
-len
);
1291 if (!vma
|| addr
<= vma
->vm_start
)
1292 /* remember the address as a hint for next time */
1293 return (mm
->free_area_cache
= addr
-len
);
1296 if (mm
->mmap_base
< len
)
1299 addr
= mm
->mmap_base
-len
;
1303 * Lookup failure means no vma is above this address,
1304 * else if new region fits below vma->vm_start,
1305 * return with success:
1307 vma
= find_vma(mm
, addr
);
1308 if (!vma
|| addr
+len
<= vma
->vm_start
)
1309 /* remember the address as a hint for next time */
1310 return (mm
->free_area_cache
= addr
);
1312 /* remember the largest hole we saw so far */
1313 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1314 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1316 /* try just below the current vma->vm_start */
1317 addr
= vma
->vm_start
-len
;
1318 } while (len
< vma
->vm_start
);
1322 * A failed mmap() very likely causes application failure,
1323 * so fall back to the bottom-up function here. This scenario
1324 * can happen with large stack limits and large mmap()
1327 mm
->cached_hole_size
= ~0UL;
1328 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1329 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1331 * Restore the topdown base:
1333 mm
->free_area_cache
= mm
->mmap_base
;
1334 mm
->cached_hole_size
= ~0UL;
1340 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1343 * Is this a new hole at the highest possible address?
1345 if (addr
> mm
->free_area_cache
)
1346 mm
->free_area_cache
= addr
;
1348 /* dont allow allocations above current base */
1349 if (mm
->free_area_cache
> mm
->mmap_base
)
1350 mm
->free_area_cache
= mm
->mmap_base
;
1354 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1355 unsigned long pgoff
, unsigned long flags
)
1359 if (!(flags
& MAP_FIXED
)) {
1360 unsigned long (*get_area
)(struct file
*, unsigned long, unsigned long, unsigned long, unsigned long);
1362 get_area
= current
->mm
->get_unmapped_area
;
1363 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1364 get_area
= file
->f_op
->get_unmapped_area
;
1365 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1366 if (IS_ERR_VALUE(addr
))
1370 if (addr
> TASK_SIZE
- len
)
1372 if (addr
& ~PAGE_MASK
)
1374 if (file
&& is_file_hugepages(file
)) {
1376 * Check if the given range is hugepage aligned, and
1377 * can be made suitable for hugepages.
1379 ret
= prepare_hugepage_range(addr
, len
);
1382 * Ensure that a normal request is not falling in a
1383 * reserved hugepage range. For some archs like IA-64,
1384 * there is a separate region for hugepages.
1386 ret
= is_hugepage_only_range(current
->mm
, addr
, len
);
1393 EXPORT_SYMBOL(get_unmapped_area
);
1395 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1396 struct vm_area_struct
* find_vma(struct mm_struct
* mm
, unsigned long addr
)
1398 struct vm_area_struct
*vma
= NULL
;
1401 /* Check the cache first. */
1402 /* (Cache hit rate is typically around 35%.) */
1403 vma
= mm
->mmap_cache
;
1404 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1405 struct rb_node
* rb_node
;
1407 rb_node
= mm
->mm_rb
.rb_node
;
1411 struct vm_area_struct
* vma_tmp
;
1413 vma_tmp
= rb_entry(rb_node
,
1414 struct vm_area_struct
, vm_rb
);
1416 if (vma_tmp
->vm_end
> addr
) {
1418 if (vma_tmp
->vm_start
<= addr
)
1420 rb_node
= rb_node
->rb_left
;
1422 rb_node
= rb_node
->rb_right
;
1425 mm
->mmap_cache
= vma
;
1431 EXPORT_SYMBOL(find_vma
);
1433 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1434 struct vm_area_struct
*
1435 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1436 struct vm_area_struct
**pprev
)
1438 struct vm_area_struct
*vma
= NULL
, *prev
= NULL
;
1439 struct rb_node
* rb_node
;
1443 /* Guard against addr being lower than the first VMA */
1446 /* Go through the RB tree quickly. */
1447 rb_node
= mm
->mm_rb
.rb_node
;
1450 struct vm_area_struct
*vma_tmp
;
1451 vma_tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1453 if (addr
< vma_tmp
->vm_end
) {
1454 rb_node
= rb_node
->rb_left
;
1457 if (!prev
->vm_next
|| (addr
< prev
->vm_next
->vm_end
))
1459 rb_node
= rb_node
->rb_right
;
1465 return prev
? prev
->vm_next
: vma
;
1469 * Verify that the stack growth is acceptable and
1470 * update accounting. This is shared with both the
1471 * grow-up and grow-down cases.
1473 static int acct_stack_growth(struct vm_area_struct
* vma
, unsigned long size
, unsigned long grow
)
1475 struct mm_struct
*mm
= vma
->vm_mm
;
1476 struct rlimit
*rlim
= current
->signal
->rlim
;
1478 /* address space limit tests */
1479 if (!may_expand_vm(mm
, grow
))
1482 /* Stack limit test */
1483 if (size
> rlim
[RLIMIT_STACK
].rlim_cur
)
1486 /* mlock limit tests */
1487 if (vma
->vm_flags
& VM_LOCKED
) {
1488 unsigned long locked
;
1489 unsigned long limit
;
1490 locked
= mm
->locked_vm
+ grow
;
1491 limit
= rlim
[RLIMIT_MEMLOCK
].rlim_cur
>> PAGE_SHIFT
;
1492 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1497 * Overcommit.. This must be the final test, as it will
1498 * update security statistics.
1500 if (security_vm_enough_memory(grow
))
1503 /* Ok, everything looks good - let it rip */
1504 mm
->total_vm
+= grow
;
1505 if (vma
->vm_flags
& VM_LOCKED
)
1506 mm
->locked_vm
+= grow
;
1507 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1511 #ifdef CONFIG_STACK_GROWSUP
1513 * vma is the first one with address > vma->vm_end. Have to extend vma.
1515 int expand_stack(struct vm_area_struct
* vma
, unsigned long address
)
1519 if (!(vma
->vm_flags
& VM_GROWSUP
))
1523 * We must make sure the anon_vma is allocated
1524 * so that the anon_vma locking is not a noop.
1526 if (unlikely(anon_vma_prepare(vma
)))
1531 * vma->vm_start/vm_end cannot change under us because the caller
1532 * is required to hold the mmap_sem in read mode. We need the
1533 * anon_vma lock to serialize against concurrent expand_stacks.
1535 address
+= 4 + PAGE_SIZE
- 1;
1536 address
&= PAGE_MASK
;
1539 /* Somebody else might have raced and expanded it already */
1540 if (address
> vma
->vm_end
) {
1541 unsigned long size
, grow
;
1543 size
= address
- vma
->vm_start
;
1544 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1546 error
= acct_stack_growth(vma
, size
, grow
);
1548 vma
->vm_end
= address
;
1550 anon_vma_unlock(vma
);
1554 struct vm_area_struct
*
1555 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1557 struct vm_area_struct
*vma
, *prev
;
1560 vma
= find_vma_prev(mm
, addr
, &prev
);
1561 if (vma
&& (vma
->vm_start
<= addr
))
1563 if (!prev
|| expand_stack(prev
, addr
))
1565 if (prev
->vm_flags
& VM_LOCKED
) {
1566 make_pages_present(addr
, prev
->vm_end
);
1572 * vma is the first one with address < vma->vm_start. Have to extend vma.
1574 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1579 * We must make sure the anon_vma is allocated
1580 * so that the anon_vma locking is not a noop.
1582 if (unlikely(anon_vma_prepare(vma
)))
1587 * vma->vm_start/vm_end cannot change under us because the caller
1588 * is required to hold the mmap_sem in read mode. We need the
1589 * anon_vma lock to serialize against concurrent expand_stacks.
1591 address
&= PAGE_MASK
;
1594 /* Somebody else might have raced and expanded it already */
1595 if (address
< vma
->vm_start
) {
1596 unsigned long size
, grow
;
1598 size
= vma
->vm_end
- address
;
1599 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1601 error
= acct_stack_growth(vma
, size
, grow
);
1603 vma
->vm_start
= address
;
1604 vma
->vm_pgoff
-= grow
;
1607 anon_vma_unlock(vma
);
1611 struct vm_area_struct
*
1612 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1614 struct vm_area_struct
* vma
;
1615 unsigned long start
;
1618 vma
= find_vma(mm
,addr
);
1621 if (vma
->vm_start
<= addr
)
1623 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1625 start
= vma
->vm_start
;
1626 if (expand_stack(vma
, addr
))
1628 if (vma
->vm_flags
& VM_LOCKED
) {
1629 make_pages_present(addr
, start
);
1636 * Ok - we have the memory areas we should free on the vma list,
1637 * so release them, and do the vma updates.
1639 * Called with the mm semaphore held.
1641 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1644 long nrpages
= vma_pages(vma
);
1646 mm
->total_vm
-= nrpages
;
1647 if (vma
->vm_flags
& VM_LOCKED
)
1648 mm
->locked_vm
-= nrpages
;
1649 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1650 vma
= remove_vma(vma
);
1656 * Get rid of page table information in the indicated region.
1658 * Called with the mm semaphore held.
1660 static void unmap_region(struct mm_struct
*mm
,
1661 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1662 unsigned long start
, unsigned long end
)
1664 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1665 struct mmu_gather
*tlb
;
1666 unsigned long nr_accounted
= 0;
1669 spin_lock(&mm
->page_table_lock
);
1670 tlb
= tlb_gather_mmu(mm
, 0);
1671 unmap_vmas(&tlb
, mm
, vma
, start
, end
, &nr_accounted
, NULL
);
1672 vm_unacct_memory(nr_accounted
);
1673 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1674 next
? next
->vm_start
: 0);
1675 tlb_finish_mmu(tlb
, start
, end
);
1676 spin_unlock(&mm
->page_table_lock
);
1680 * Create a list of vma's touched by the unmap, removing them from the mm's
1681 * vma list as we go..
1684 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1685 struct vm_area_struct
*prev
, unsigned long end
)
1687 struct vm_area_struct
**insertion_point
;
1688 struct vm_area_struct
*tail_vma
= NULL
;
1691 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1693 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1697 } while (vma
&& vma
->vm_start
< end
);
1698 *insertion_point
= vma
;
1699 tail_vma
->vm_next
= NULL
;
1700 if (mm
->unmap_area
== arch_unmap_area
)
1701 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1703 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1704 mm
->unmap_area(mm
, addr
);
1705 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1709 * Split a vma into two pieces at address 'addr', a new vma is allocated
1710 * either for the first part or the the tail.
1712 int split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1713 unsigned long addr
, int new_below
)
1715 struct mempolicy
*pol
;
1716 struct vm_area_struct
*new;
1718 if (is_vm_hugetlb_page(vma
) && (addr
& ~HPAGE_MASK
))
1721 if (mm
->map_count
>= sysctl_max_map_count
)
1724 new = kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
1728 /* most fields are the same, copy all, and then fixup */
1734 new->vm_start
= addr
;
1735 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
1738 pol
= mpol_copy(vma_policy(vma
));
1740 kmem_cache_free(vm_area_cachep
, new);
1741 return PTR_ERR(pol
);
1743 vma_set_policy(new, pol
);
1746 get_file(new->vm_file
);
1748 if (new->vm_ops
&& new->vm_ops
->open
)
1749 new->vm_ops
->open(new);
1752 vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
1753 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
1755 vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
1760 /* Munmap is split into 2 main parts -- this part which finds
1761 * what needs doing, and the areas themselves, which do the
1762 * work. This now handles partial unmappings.
1763 * Jeremy Fitzhardinge <jeremy@goop.org>
1765 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1768 struct vm_area_struct
*vma
, *prev
, *last
;
1770 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
1773 if ((len
= PAGE_ALIGN(len
)) == 0)
1776 /* Find the first overlapping VMA */
1777 vma
= find_vma_prev(mm
, start
, &prev
);
1780 /* we have start < vma->vm_end */
1782 /* if it doesn't overlap, we have nothing.. */
1784 if (vma
->vm_start
>= end
)
1788 * If we need to split any vma, do it now to save pain later.
1790 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1791 * unmapped vm_area_struct will remain in use: so lower split_vma
1792 * places tmp vma above, and higher split_vma places tmp vma below.
1794 if (start
> vma
->vm_start
) {
1795 int error
= split_vma(mm
, vma
, start
, 0);
1801 /* Does it split the last one? */
1802 last
= find_vma(mm
, end
);
1803 if (last
&& end
> last
->vm_start
) {
1804 int error
= split_vma(mm
, last
, end
, 1);
1808 vma
= prev
? prev
->vm_next
: mm
->mmap
;
1811 * Remove the vma's, and unmap the actual pages
1813 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
1814 unmap_region(mm
, vma
, prev
, start
, end
);
1816 /* Fix up all other VM information */
1817 remove_vma_list(mm
, vma
);
1822 EXPORT_SYMBOL(do_munmap
);
1824 asmlinkage
long sys_munmap(unsigned long addr
, size_t len
)
1827 struct mm_struct
*mm
= current
->mm
;
1829 profile_munmap(addr
);
1831 down_write(&mm
->mmap_sem
);
1832 ret
= do_munmap(mm
, addr
, len
);
1833 up_write(&mm
->mmap_sem
);
1837 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
1839 #ifdef CONFIG_DEBUG_KERNEL
1840 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
1842 up_read(&mm
->mmap_sem
);
1848 * this is really a simplified "do_mmap". it only handles
1849 * anonymous maps. eventually we may be able to do some
1850 * brk-specific accounting here.
1852 unsigned long do_brk(unsigned long addr
, unsigned long len
)
1854 struct mm_struct
* mm
= current
->mm
;
1855 struct vm_area_struct
* vma
, * prev
;
1856 unsigned long flags
;
1857 struct rb_node
** rb_link
, * rb_parent
;
1858 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
1860 len
= PAGE_ALIGN(len
);
1864 if ((addr
+ len
) > TASK_SIZE
|| (addr
+ len
) < addr
)
1870 if (mm
->def_flags
& VM_LOCKED
) {
1871 unsigned long locked
, lock_limit
;
1872 locked
= len
>> PAGE_SHIFT
;
1873 locked
+= mm
->locked_vm
;
1874 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
1875 lock_limit
>>= PAGE_SHIFT
;
1876 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1881 * mm->mmap_sem is required to protect against another thread
1882 * changing the mappings in case we sleep.
1884 verify_mm_writelocked(mm
);
1887 * Clear old maps. this also does some error checking for us
1890 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1891 if (vma
&& vma
->vm_start
< addr
+ len
) {
1892 if (do_munmap(mm
, addr
, len
))
1897 /* Check against address space limits *after* clearing old maps... */
1898 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1901 if (mm
->map_count
> sysctl_max_map_count
)
1904 if (security_vm_enough_memory(len
>> PAGE_SHIFT
))
1907 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
1909 /* Can we just expand an old private anonymous mapping? */
1910 if (vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
1911 NULL
, NULL
, pgoff
, NULL
))
1915 * create a vma struct for an anonymous mapping
1917 vma
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
1919 vm_unacct_memory(len
>> PAGE_SHIFT
);
1922 memset(vma
, 0, sizeof(*vma
));
1925 vma
->vm_start
= addr
;
1926 vma
->vm_end
= addr
+ len
;
1927 vma
->vm_pgoff
= pgoff
;
1928 vma
->vm_flags
= flags
;
1929 vma
->vm_page_prot
= protection_map
[flags
& 0x0f];
1930 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1932 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1933 if (flags
& VM_LOCKED
) {
1934 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
1935 make_pages_present(addr
, addr
+ len
);
1940 EXPORT_SYMBOL(do_brk
);
1942 /* Release all mmaps. */
1943 void exit_mmap(struct mm_struct
*mm
)
1945 struct mmu_gather
*tlb
;
1946 struct vm_area_struct
*vma
= mm
->mmap
;
1947 unsigned long nr_accounted
= 0;
1952 spin_lock(&mm
->page_table_lock
);
1955 tlb
= tlb_gather_mmu(mm
, 1);
1956 /* Use -1 here to ensure all VMAs in the mm are unmapped */
1957 end
= unmap_vmas(&tlb
, mm
, vma
, 0, -1, &nr_accounted
, NULL
);
1958 vm_unacct_memory(nr_accounted
);
1959 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, 0);
1960 tlb_finish_mmu(tlb
, 0, end
);
1962 spin_unlock(&mm
->page_table_lock
);
1965 * Walk the list again, actually closing and freeing it
1966 * without holding any MM locks.
1969 vma
= remove_vma(vma
);
1971 BUG_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
1974 /* Insert vm structure into process list sorted by address
1975 * and into the inode's i_mmap tree. If vm_file is non-NULL
1976 * then i_mmap_lock is taken here.
1978 int insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
1980 struct vm_area_struct
* __vma
, * prev
;
1981 struct rb_node
** rb_link
, * rb_parent
;
1984 * The vm_pgoff of a purely anonymous vma should be irrelevant
1985 * until its first write fault, when page's anon_vma and index
1986 * are set. But now set the vm_pgoff it will almost certainly
1987 * end up with (unless mremap moves it elsewhere before that
1988 * first wfault), so /proc/pid/maps tells a consistent story.
1990 * By setting it to reflect the virtual start address of the
1991 * vma, merges and splits can happen in a seamless way, just
1992 * using the existing file pgoff checks and manipulations.
1993 * Similarly in do_mmap_pgoff and in do_brk.
1995 if (!vma
->vm_file
) {
1996 BUG_ON(vma
->anon_vma
);
1997 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
1999 __vma
= find_vma_prepare(mm
,vma
->vm_start
,&prev
,&rb_link
,&rb_parent
);
2000 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
2002 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2003 security_vm_enough_memory(vma_pages(vma
)))
2005 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2010 * Copy the vma structure to a new location in the same mm,
2011 * prior to moving page table entries, to effect an mremap move.
2013 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2014 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
2016 struct vm_area_struct
*vma
= *vmap
;
2017 unsigned long vma_start
= vma
->vm_start
;
2018 struct mm_struct
*mm
= vma
->vm_mm
;
2019 struct vm_area_struct
*new_vma
, *prev
;
2020 struct rb_node
**rb_link
, *rb_parent
;
2021 struct mempolicy
*pol
;
2024 * If anonymous vma has not yet been faulted, update new pgoff
2025 * to match new location, to increase its chance of merging.
2027 if (!vma
->vm_file
&& !vma
->anon_vma
)
2028 pgoff
= addr
>> PAGE_SHIFT
;
2030 find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2031 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2032 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2035 * Source vma may have been merged into new_vma
2037 if (vma_start
>= new_vma
->vm_start
&&
2038 vma_start
< new_vma
->vm_end
)
2041 new_vma
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
2044 pol
= mpol_copy(vma_policy(vma
));
2046 kmem_cache_free(vm_area_cachep
, new_vma
);
2049 vma_set_policy(new_vma
, pol
);
2050 new_vma
->vm_start
= addr
;
2051 new_vma
->vm_end
= addr
+ len
;
2052 new_vma
->vm_pgoff
= pgoff
;
2053 if (new_vma
->vm_file
)
2054 get_file(new_vma
->vm_file
);
2055 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2056 new_vma
->vm_ops
->open(new_vma
);
2057 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2064 * Return true if the calling process may expand its vm space by the passed
2067 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2069 unsigned long cur
= mm
->total_vm
; /* pages */
2072 lim
= current
->signal
->rlim
[RLIMIT_AS
].rlim_cur
>> PAGE_SHIFT
;
2074 if (cur
+ npages
> lim
)