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/capability.h>
17 #include <linux/init.h>
18 #include <linux/file.h>
20 #include <linux/personality.h>
21 #include <linux/security.h>
22 #include <linux/hugetlb.h>
23 #include <linux/profile.h>
24 #include <linux/module.h>
25 #include <linux/mount.h>
26 #include <linux/mempolicy.h>
27 #include <linux/rmap.h>
29 #include <asm/uaccess.h>
30 #include <asm/cacheflush.h>
33 static void unmap_region(struct mm_struct
*mm
,
34 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
35 unsigned long start
, unsigned long end
);
38 * WARNING: the debugging will use recursive algorithms so never enable this
39 * unless you know what you are doing.
43 /* description of effects of mapping type and prot in current implementation.
44 * this is due to the limited x86 page protection hardware. The expected
45 * behavior is in parens:
48 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
49 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
50 * w: (no) no w: (no) no w: (yes) yes w: (no) no
51 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
53 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
54 * w: (no) no w: (no) no w: (copy) copy w: (no) no
55 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
58 pgprot_t protection_map
[16] = {
59 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
60 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
63 int sysctl_overcommit_memory
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
64 int sysctl_overcommit_ratio
= 50; /* default is 50% */
65 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
66 atomic_t vm_committed_space
= ATOMIC_INIT(0);
69 * Check that a process has enough memory to allocate a new virtual
70 * mapping. 0 means there is enough memory for the allocation to
71 * succeed and -ENOMEM implies there is not.
73 * We currently support three overcommit policies, which are set via the
74 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
76 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
77 * Additional code 2002 Jul 20 by Robert Love.
79 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
81 * Note this is a helper function intended to be used by LSMs which
82 * wish to use this logic.
84 int __vm_enough_memory(long pages
, int cap_sys_admin
)
86 unsigned long free
, allowed
;
88 vm_acct_memory(pages
);
91 * Sometimes we want to use more memory than we have
93 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
96 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
99 free
= get_page_cache_size();
100 free
+= nr_swap_pages
;
103 * Any slabs which are created with the
104 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
105 * which are reclaimable, under pressure. The dentry
106 * cache and most inode caches should fall into this
108 free
+= atomic_read(&slab_reclaim_pages
);
111 * Leave the last 3% for root
120 * nr_free_pages() is very expensive on large systems,
121 * only call if we're about to fail.
130 vm_unacct_memory(pages
);
134 allowed
= (totalram_pages
- hugetlb_total_pages())
135 * sysctl_overcommit_ratio
/ 100;
137 * Leave the last 3% for root
140 allowed
-= allowed
/ 32;
141 allowed
+= total_swap_pages
;
143 /* Don't let a single process grow too big:
144 leave 3% of the size of this process for other processes */
145 allowed
-= current
->mm
->total_vm
/ 32;
148 * cast `allowed' as a signed long because vm_committed_space
149 * sometimes has a negative value
151 if (atomic_read(&vm_committed_space
) < (long)allowed
)
154 vm_unacct_memory(pages
);
159 EXPORT_SYMBOL(__vm_enough_memory
);
162 * Requires inode->i_mapping->i_mmap_lock
164 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
165 struct file
*file
, struct address_space
*mapping
)
167 if (vma
->vm_flags
& VM_DENYWRITE
)
168 atomic_inc(&file
->f_dentry
->d_inode
->i_writecount
);
169 if (vma
->vm_flags
& VM_SHARED
)
170 mapping
->i_mmap_writable
--;
172 flush_dcache_mmap_lock(mapping
);
173 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
174 list_del_init(&vma
->shared
.vm_set
.list
);
176 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
177 flush_dcache_mmap_unlock(mapping
);
181 * Unlink a file-based vm structure from its prio_tree, to hide
182 * vma from rmap and vmtruncate before freeing its page tables.
184 void unlink_file_vma(struct vm_area_struct
*vma
)
186 struct file
*file
= vma
->vm_file
;
189 struct address_space
*mapping
= file
->f_mapping
;
190 spin_lock(&mapping
->i_mmap_lock
);
191 __remove_shared_vm_struct(vma
, file
, mapping
);
192 spin_unlock(&mapping
->i_mmap_lock
);
197 * Close a vm structure and free it, returning the next.
199 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
201 struct vm_area_struct
*next
= vma
->vm_next
;
204 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
205 vma
->vm_ops
->close(vma
);
208 mpol_free(vma_policy(vma
));
209 kmem_cache_free(vm_area_cachep
, vma
);
213 asmlinkage
unsigned long sys_brk(unsigned long brk
)
215 unsigned long rlim
, retval
;
216 unsigned long newbrk
, oldbrk
;
217 struct mm_struct
*mm
= current
->mm
;
219 down_write(&mm
->mmap_sem
);
221 if (brk
< mm
->end_code
)
223 newbrk
= PAGE_ALIGN(brk
);
224 oldbrk
= PAGE_ALIGN(mm
->brk
);
225 if (oldbrk
== newbrk
)
228 /* Always allow shrinking brk. */
229 if (brk
<= mm
->brk
) {
230 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
235 /* Check against rlimit.. */
236 rlim
= current
->signal
->rlim
[RLIMIT_DATA
].rlim_cur
;
237 if (rlim
< RLIM_INFINITY
&& brk
- mm
->start_data
> rlim
)
240 /* Check against existing mmap mappings. */
241 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
244 /* Ok, looks good - let it rip. */
245 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
251 up_write(&mm
->mmap_sem
);
256 static int browse_rb(struct rb_root
*root
)
259 struct rb_node
*nd
, *pn
= NULL
;
260 unsigned long prev
= 0, pend
= 0;
262 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
263 struct vm_area_struct
*vma
;
264 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
265 if (vma
->vm_start
< prev
)
266 printk("vm_start %lx prev %lx\n", vma
->vm_start
, prev
), i
= -1;
267 if (vma
->vm_start
< pend
)
268 printk("vm_start %lx pend %lx\n", vma
->vm_start
, pend
);
269 if (vma
->vm_start
> vma
->vm_end
)
270 printk("vm_end %lx < vm_start %lx\n", vma
->vm_end
, vma
->vm_start
);
275 for (nd
= pn
; nd
; nd
= rb_prev(nd
)) {
279 printk("backwards %d, forwards %d\n", j
, i
), i
= 0;
283 void validate_mm(struct mm_struct
*mm
)
287 struct vm_area_struct
*tmp
= mm
->mmap
;
292 if (i
!= mm
->map_count
)
293 printk("map_count %d vm_next %d\n", mm
->map_count
, i
), bug
= 1;
294 i
= browse_rb(&mm
->mm_rb
);
295 if (i
!= mm
->map_count
)
296 printk("map_count %d rb %d\n", mm
->map_count
, i
), bug
= 1;
300 #define validate_mm(mm) do { } while (0)
303 static struct vm_area_struct
*
304 find_vma_prepare(struct mm_struct
*mm
, unsigned long addr
,
305 struct vm_area_struct
**pprev
, struct rb_node
***rb_link
,
306 struct rb_node
** rb_parent
)
308 struct vm_area_struct
* vma
;
309 struct rb_node
** __rb_link
, * __rb_parent
, * rb_prev
;
311 __rb_link
= &mm
->mm_rb
.rb_node
;
312 rb_prev
= __rb_parent
= NULL
;
316 struct vm_area_struct
*vma_tmp
;
318 __rb_parent
= *__rb_link
;
319 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
321 if (vma_tmp
->vm_end
> addr
) {
323 if (vma_tmp
->vm_start
<= addr
)
325 __rb_link
= &__rb_parent
->rb_left
;
327 rb_prev
= __rb_parent
;
328 __rb_link
= &__rb_parent
->rb_right
;
334 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
335 *rb_link
= __rb_link
;
336 *rb_parent
= __rb_parent
;
341 __vma_link_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
342 struct vm_area_struct
*prev
, struct rb_node
*rb_parent
)
345 vma
->vm_next
= prev
->vm_next
;
350 vma
->vm_next
= rb_entry(rb_parent
,
351 struct vm_area_struct
, vm_rb
);
357 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
358 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
360 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
361 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
364 static inline void __vma_link_file(struct vm_area_struct
*vma
)
370 struct address_space
*mapping
= file
->f_mapping
;
372 if (vma
->vm_flags
& VM_DENYWRITE
)
373 atomic_dec(&file
->f_dentry
->d_inode
->i_writecount
);
374 if (vma
->vm_flags
& VM_SHARED
)
375 mapping
->i_mmap_writable
++;
377 flush_dcache_mmap_lock(mapping
);
378 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
379 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
381 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
382 flush_dcache_mmap_unlock(mapping
);
387 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
388 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
389 struct rb_node
*rb_parent
)
391 __vma_link_list(mm
, vma
, prev
, rb_parent
);
392 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
393 __anon_vma_link(vma
);
396 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
397 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
398 struct rb_node
*rb_parent
)
400 struct address_space
*mapping
= NULL
;
403 mapping
= vma
->vm_file
->f_mapping
;
406 spin_lock(&mapping
->i_mmap_lock
);
407 vma
->vm_truncate_count
= mapping
->truncate_count
;
411 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
412 __vma_link_file(vma
);
414 anon_vma_unlock(vma
);
416 spin_unlock(&mapping
->i_mmap_lock
);
423 * Helper for vma_adjust in the split_vma insert case:
424 * insert vm structure into list and rbtree and anon_vma,
425 * but it has already been inserted into prio_tree earlier.
428 __insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
430 struct vm_area_struct
* __vma
, * prev
;
431 struct rb_node
** rb_link
, * rb_parent
;
433 __vma
= find_vma_prepare(mm
, vma
->vm_start
,&prev
, &rb_link
, &rb_parent
);
434 BUG_ON(__vma
&& __vma
->vm_start
< vma
->vm_end
);
435 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
440 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
441 struct vm_area_struct
*prev
)
443 prev
->vm_next
= vma
->vm_next
;
444 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
445 if (mm
->mmap_cache
== vma
)
446 mm
->mmap_cache
= prev
;
450 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
451 * is already present in an i_mmap tree without adjusting the tree.
452 * The following helper function should be used when such adjustments
453 * are necessary. The "insert" vma (if any) is to be inserted
454 * before we drop the necessary locks.
456 void vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
457 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
459 struct mm_struct
*mm
= vma
->vm_mm
;
460 struct vm_area_struct
*next
= vma
->vm_next
;
461 struct vm_area_struct
*importer
= NULL
;
462 struct address_space
*mapping
= NULL
;
463 struct prio_tree_root
*root
= NULL
;
464 struct file
*file
= vma
->vm_file
;
465 struct anon_vma
*anon_vma
= NULL
;
466 long adjust_next
= 0;
469 if (next
&& !insert
) {
470 if (end
>= next
->vm_end
) {
472 * vma expands, overlapping all the next, and
473 * perhaps the one after too (mprotect case 6).
475 again
: remove_next
= 1 + (end
> next
->vm_end
);
477 anon_vma
= next
->anon_vma
;
479 } else if (end
> next
->vm_start
) {
481 * vma expands, overlapping part of the next:
482 * mprotect case 5 shifting the boundary up.
484 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
485 anon_vma
= next
->anon_vma
;
487 } else if (end
< vma
->vm_end
) {
489 * vma shrinks, and !insert tells it's not
490 * split_vma inserting another: so it must be
491 * mprotect case 4 shifting the boundary down.
493 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
494 anon_vma
= next
->anon_vma
;
500 mapping
= file
->f_mapping
;
501 if (!(vma
->vm_flags
& VM_NONLINEAR
))
502 root
= &mapping
->i_mmap
;
503 spin_lock(&mapping
->i_mmap_lock
);
505 vma
->vm_truncate_count
!= next
->vm_truncate_count
) {
507 * unmap_mapping_range might be in progress:
508 * ensure that the expanding vma is rescanned.
510 importer
->vm_truncate_count
= 0;
513 insert
->vm_truncate_count
= vma
->vm_truncate_count
;
515 * Put into prio_tree now, so instantiated pages
516 * are visible to arm/parisc __flush_dcache_page
517 * throughout; but we cannot insert into address
518 * space until vma start or end is updated.
520 __vma_link_file(insert
);
525 * When changing only vma->vm_end, we don't really need
526 * anon_vma lock: but is that case worth optimizing out?
529 anon_vma
= vma
->anon_vma
;
531 spin_lock(&anon_vma
->lock
);
533 * Easily overlooked: when mprotect shifts the boundary,
534 * make sure the expanding vma has anon_vma set if the
535 * shrinking vma had, to cover any anon pages imported.
537 if (importer
&& !importer
->anon_vma
) {
538 importer
->anon_vma
= anon_vma
;
539 __anon_vma_link(importer
);
544 flush_dcache_mmap_lock(mapping
);
545 vma_prio_tree_remove(vma
, root
);
547 vma_prio_tree_remove(next
, root
);
550 vma
->vm_start
= start
;
552 vma
->vm_pgoff
= pgoff
;
554 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
555 next
->vm_pgoff
+= adjust_next
;
560 vma_prio_tree_insert(next
, root
);
561 vma_prio_tree_insert(vma
, root
);
562 flush_dcache_mmap_unlock(mapping
);
567 * vma_merge has merged next into vma, and needs
568 * us to remove next before dropping the locks.
570 __vma_unlink(mm
, next
, vma
);
572 __remove_shared_vm_struct(next
, file
, mapping
);
574 __anon_vma_merge(vma
, next
);
577 * split_vma has split insert from vma, and needs
578 * us to insert it before dropping the locks
579 * (it may either follow vma or precede it).
581 __insert_vm_struct(mm
, insert
);
585 spin_unlock(&anon_vma
->lock
);
587 spin_unlock(&mapping
->i_mmap_lock
);
593 mpol_free(vma_policy(next
));
594 kmem_cache_free(vm_area_cachep
, next
);
596 * In mprotect's case 6 (see comments on vma_merge),
597 * we must remove another next too. It would clutter
598 * up the code too much to do both in one go.
600 if (remove_next
== 2) {
610 * If the vma has a ->close operation then the driver probably needs to release
611 * per-vma resources, so we don't attempt to merge those.
613 #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP)
615 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
616 struct file
*file
, unsigned long vm_flags
)
618 if (vma
->vm_flags
!= vm_flags
)
620 if (vma
->vm_file
!= file
)
622 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
627 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
628 struct anon_vma
*anon_vma2
)
630 return !anon_vma1
|| !anon_vma2
|| (anon_vma1
== anon_vma2
);
634 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
635 * in front of (at a lower virtual address and file offset than) the vma.
637 * We cannot merge two vmas if they have differently assigned (non-NULL)
638 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
640 * We don't check here for the merged mmap wrapping around the end of pagecache
641 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
642 * wrap, nor mmaps which cover the final page at index -1UL.
645 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
646 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
648 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
649 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
650 if (vma
->vm_pgoff
== vm_pgoff
)
657 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
658 * beyond (at a higher virtual address and file offset than) the vma.
660 * We cannot merge two vmas if they have differently assigned (non-NULL)
661 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
664 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
665 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
667 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
668 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
670 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
671 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
678 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
679 * whether that can be merged with its predecessor or its successor.
680 * Or both (it neatly fills a hole).
682 * In most cases - when called for mmap, brk or mremap - [addr,end) is
683 * certain not to be mapped by the time vma_merge is called; but when
684 * called for mprotect, it is certain to be already mapped (either at
685 * an offset within prev, or at the start of next), and the flags of
686 * this area are about to be changed to vm_flags - and the no-change
687 * case has already been eliminated.
689 * The following mprotect cases have to be considered, where AAAA is
690 * the area passed down from mprotect_fixup, never extending beyond one
691 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
693 * AAAA AAAA AAAA AAAA
694 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
695 * cannot merge might become might become might become
696 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
697 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
698 * mremap move: PPPPNNNNNNNN 8
700 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
701 * might become case 1 below case 2 below case 3 below
703 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
704 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
706 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
707 struct vm_area_struct
*prev
, unsigned long addr
,
708 unsigned long end
, unsigned long vm_flags
,
709 struct anon_vma
*anon_vma
, struct file
*file
,
710 pgoff_t pgoff
, struct mempolicy
*policy
)
712 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
713 struct vm_area_struct
*area
, *next
;
716 * We later require that vma->vm_flags == vm_flags,
717 * so this tests vma->vm_flags & VM_SPECIAL, too.
719 if (vm_flags
& VM_SPECIAL
)
723 next
= prev
->vm_next
;
727 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
728 next
= next
->vm_next
;
731 * Can it merge with the predecessor?
733 if (prev
&& prev
->vm_end
== addr
&&
734 mpol_equal(vma_policy(prev
), policy
) &&
735 can_vma_merge_after(prev
, vm_flags
,
736 anon_vma
, file
, pgoff
)) {
738 * OK, it can. Can we now merge in the successor as well?
740 if (next
&& end
== next
->vm_start
&&
741 mpol_equal(policy
, vma_policy(next
)) &&
742 can_vma_merge_before(next
, vm_flags
,
743 anon_vma
, file
, pgoff
+pglen
) &&
744 is_mergeable_anon_vma(prev
->anon_vma
,
747 vma_adjust(prev
, prev
->vm_start
,
748 next
->vm_end
, prev
->vm_pgoff
, NULL
);
749 } else /* cases 2, 5, 7 */
750 vma_adjust(prev
, prev
->vm_start
,
751 end
, prev
->vm_pgoff
, NULL
);
756 * Can this new request be merged in front of next?
758 if (next
&& end
== next
->vm_start
&&
759 mpol_equal(policy
, vma_policy(next
)) &&
760 can_vma_merge_before(next
, vm_flags
,
761 anon_vma
, file
, pgoff
+pglen
)) {
762 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
763 vma_adjust(prev
, prev
->vm_start
,
764 addr
, prev
->vm_pgoff
, NULL
);
765 else /* cases 3, 8 */
766 vma_adjust(area
, addr
, next
->vm_end
,
767 next
->vm_pgoff
- pglen
, NULL
);
775 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
776 * neighbouring vmas for a suitable anon_vma, before it goes off
777 * to allocate a new anon_vma. It checks because a repetitive
778 * sequence of mprotects and faults may otherwise lead to distinct
779 * anon_vmas being allocated, preventing vma merge in subsequent
782 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
784 struct vm_area_struct
*near
;
785 unsigned long vm_flags
;
792 * Since only mprotect tries to remerge vmas, match flags
793 * which might be mprotected into each other later on.
794 * Neither mlock nor madvise tries to remerge at present,
795 * so leave their flags as obstructing a merge.
797 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
798 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
800 if (near
->anon_vma
&& vma
->vm_end
== near
->vm_start
&&
801 mpol_equal(vma_policy(vma
), vma_policy(near
)) &&
802 can_vma_merge_before(near
, vm_flags
,
803 NULL
, vma
->vm_file
, vma
->vm_pgoff
+
804 ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
)))
805 return near
->anon_vma
;
808 * It is potentially slow to have to call find_vma_prev here.
809 * But it's only on the first write fault on the vma, not
810 * every time, and we could devise a way to avoid it later
811 * (e.g. stash info in next's anon_vma_node when assigning
812 * an anon_vma, or when trying vma_merge). Another time.
814 BUG_ON(find_vma_prev(vma
->vm_mm
, vma
->vm_start
, &near
) != vma
);
818 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
819 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
821 if (near
->anon_vma
&& near
->vm_end
== vma
->vm_start
&&
822 mpol_equal(vma_policy(near
), vma_policy(vma
)) &&
823 can_vma_merge_after(near
, vm_flags
,
824 NULL
, vma
->vm_file
, vma
->vm_pgoff
))
825 return near
->anon_vma
;
828 * There's no absolute need to look only at touching neighbours:
829 * we could search further afield for "compatible" anon_vmas.
830 * But it would probably just be a waste of time searching,
831 * or lead to too many vmas hanging off the same anon_vma.
832 * We're trying to allow mprotect remerging later on,
833 * not trying to minimize memory used for anon_vmas.
838 #ifdef CONFIG_PROC_FS
839 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
840 struct file
*file
, long pages
)
842 const unsigned long stack_flags
843 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
846 mm
->shared_vm
+= pages
;
847 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
848 mm
->exec_vm
+= pages
;
849 } else if (flags
& stack_flags
)
850 mm
->stack_vm
+= pages
;
851 if (flags
& (VM_RESERVED
|VM_IO
))
852 mm
->reserved_vm
+= pages
;
854 #endif /* CONFIG_PROC_FS */
857 * The caller must hold down_write(current->mm->mmap_sem).
860 unsigned long do_mmap_pgoff(struct file
* file
, unsigned long addr
,
861 unsigned long len
, unsigned long prot
,
862 unsigned long flags
, unsigned long pgoff
)
864 struct mm_struct
* mm
= current
->mm
;
865 struct vm_area_struct
* vma
, * prev
;
867 unsigned int vm_flags
;
868 int correct_wcount
= 0;
870 struct rb_node
** rb_link
, * rb_parent
;
872 unsigned long charged
= 0, reqprot
= prot
;
875 if (is_file_hugepages(file
))
878 if (!file
->f_op
|| !file
->f_op
->mmap
)
881 if ((prot
& PROT_EXEC
) &&
882 (file
->f_vfsmnt
->mnt_flags
& MNT_NOEXEC
))
886 * Does the application expect PROT_READ to imply PROT_EXEC?
888 * (the exception is when the underlying filesystem is noexec
889 * mounted, in which case we dont add PROT_EXEC.)
891 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
892 if (!(file
&& (file
->f_vfsmnt
->mnt_flags
& MNT_NOEXEC
)))
898 /* Careful about overflows.. */
899 len
= PAGE_ALIGN(len
);
900 if (!len
|| len
> TASK_SIZE
)
903 /* offset overflow? */
904 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
907 /* Too many mappings? */
908 if (mm
->map_count
> sysctl_max_map_count
)
911 /* Obtain the address to map to. we verify (or select) it and ensure
912 * that it represents a valid section of the address space.
914 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
915 if (addr
& ~PAGE_MASK
)
918 /* Do simple checking here so the lower-level routines won't have
919 * to. we assume access permissions have been handled by the open
920 * of the memory object, so we don't do any here.
922 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
923 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
925 if (flags
& MAP_LOCKED
) {
928 vm_flags
|= VM_LOCKED
;
930 /* mlock MCL_FUTURE? */
931 if (vm_flags
& VM_LOCKED
) {
932 unsigned long locked
, lock_limit
;
933 locked
= len
>> PAGE_SHIFT
;
934 locked
+= mm
->locked_vm
;
935 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
936 lock_limit
>>= PAGE_SHIFT
;
937 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
941 inode
= file
? file
->f_dentry
->d_inode
: NULL
;
944 switch (flags
& MAP_TYPE
) {
946 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
950 * Make sure we don't allow writing to an append-only
953 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
957 * Make sure there are no mandatory locks on the file.
959 if (locks_verify_locked(inode
))
962 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
963 if (!(file
->f_mode
& FMODE_WRITE
))
964 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
968 if (!(file
->f_mode
& FMODE_READ
))
976 switch (flags
& MAP_TYPE
) {
978 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
982 * Set pgoff according to addr for anon_vma.
984 pgoff
= addr
>> PAGE_SHIFT
;
991 error
= security_file_mmap(file
, reqprot
, prot
, flags
);
998 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
999 if (vma
&& vma
->vm_start
< addr
+ len
) {
1000 if (do_munmap(mm
, addr
, len
))
1005 /* Check against address space limit. */
1006 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1009 if (accountable
&& (!(flags
& MAP_NORESERVE
) ||
1010 sysctl_overcommit_memory
== OVERCOMMIT_NEVER
)) {
1011 if (vm_flags
& VM_SHARED
) {
1012 /* Check memory availability in shmem_file_setup? */
1013 vm_flags
|= VM_ACCOUNT
;
1014 } else if (vm_flags
& VM_WRITE
) {
1016 * Private writable mapping: check memory availability
1018 charged
= len
>> PAGE_SHIFT
;
1019 if (security_vm_enough_memory(charged
))
1021 vm_flags
|= VM_ACCOUNT
;
1026 * Can we just expand an old private anonymous mapping?
1027 * The VM_SHARED test is necessary because shmem_zero_setup
1028 * will create the file object for a shared anonymous map below.
1030 if (!file
&& !(vm_flags
& VM_SHARED
) &&
1031 vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
,
1032 NULL
, NULL
, pgoff
, NULL
))
1036 * Determine the object being mapped and call the appropriate
1037 * specific mapper. the address has already been validated, but
1038 * not unmapped, but the maps are removed from the list.
1040 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1047 vma
->vm_start
= addr
;
1048 vma
->vm_end
= addr
+ len
;
1049 vma
->vm_flags
= vm_flags
;
1050 vma
->vm_page_prot
= protection_map
[vm_flags
& 0x0f];
1051 vma
->vm_pgoff
= pgoff
;
1055 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1057 if (vm_flags
& VM_DENYWRITE
) {
1058 error
= deny_write_access(file
);
1063 vma
->vm_file
= file
;
1065 error
= file
->f_op
->mmap(file
, vma
);
1067 goto unmap_and_free_vma
;
1068 } else if (vm_flags
& VM_SHARED
) {
1069 error
= shmem_zero_setup(vma
);
1074 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1075 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1076 * that memory reservation must be checked; but that reservation
1077 * belongs to shared memory object, not to vma: so now clear it.
1079 if ((vm_flags
& (VM_SHARED
|VM_ACCOUNT
)) == (VM_SHARED
|VM_ACCOUNT
))
1080 vma
->vm_flags
&= ~VM_ACCOUNT
;
1082 /* Can addr have changed??
1084 * Answer: Yes, several device drivers can do it in their
1085 * f_op->mmap method. -DaveM
1087 addr
= vma
->vm_start
;
1088 pgoff
= vma
->vm_pgoff
;
1089 vm_flags
= vma
->vm_flags
;
1091 if (!file
|| !vma_merge(mm
, prev
, addr
, vma
->vm_end
,
1092 vma
->vm_flags
, NULL
, file
, pgoff
, vma_policy(vma
))) {
1093 file
= vma
->vm_file
;
1094 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1096 atomic_inc(&inode
->i_writecount
);
1100 atomic_inc(&inode
->i_writecount
);
1103 mpol_free(vma_policy(vma
));
1104 kmem_cache_free(vm_area_cachep
, vma
);
1107 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1108 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1109 if (vm_flags
& VM_LOCKED
) {
1110 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
1111 make_pages_present(addr
, addr
+ len
);
1113 if (flags
& MAP_POPULATE
) {
1114 up_write(&mm
->mmap_sem
);
1115 sys_remap_file_pages(addr
, len
, 0,
1116 pgoff
, flags
& MAP_NONBLOCK
);
1117 down_write(&mm
->mmap_sem
);
1123 atomic_inc(&inode
->i_writecount
);
1124 vma
->vm_file
= NULL
;
1127 /* Undo any partial mapping done by a device driver. */
1128 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1131 kmem_cache_free(vm_area_cachep
, vma
);
1134 vm_unacct_memory(charged
);
1138 EXPORT_SYMBOL(do_mmap_pgoff
);
1140 /* Get an address range which is currently unmapped.
1141 * For shmat() with addr=0.
1143 * Ugly calling convention alert:
1144 * Return value with the low bits set means error value,
1146 * if (ret & ~PAGE_MASK)
1149 * This function "knows" that -ENOMEM has the bits set.
1151 #ifndef HAVE_ARCH_UNMAPPED_AREA
1153 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1154 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1156 struct mm_struct
*mm
= current
->mm
;
1157 struct vm_area_struct
*vma
;
1158 unsigned long start_addr
;
1160 if (len
> TASK_SIZE
)
1164 addr
= PAGE_ALIGN(addr
);
1165 vma
= find_vma(mm
, addr
);
1166 if (TASK_SIZE
- len
>= addr
&&
1167 (!vma
|| addr
+ len
<= vma
->vm_start
))
1170 if (len
> mm
->cached_hole_size
) {
1171 start_addr
= addr
= mm
->free_area_cache
;
1173 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1174 mm
->cached_hole_size
= 0;
1178 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1179 /* At this point: (!vma || addr < vma->vm_end). */
1180 if (TASK_SIZE
- len
< addr
) {
1182 * Start a new search - just in case we missed
1185 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1186 addr
= TASK_UNMAPPED_BASE
;
1188 mm
->cached_hole_size
= 0;
1193 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1195 * Remember the place where we stopped the search:
1197 mm
->free_area_cache
= addr
+ len
;
1200 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1201 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1207 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1210 * Is this a new hole at the lowest possible address?
1212 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
) {
1213 mm
->free_area_cache
= addr
;
1214 mm
->cached_hole_size
= ~0UL;
1219 * This mmap-allocator allocates new areas top-down from below the
1220 * stack's low limit (the base):
1222 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1224 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1225 const unsigned long len
, const unsigned long pgoff
,
1226 const unsigned long flags
)
1228 struct vm_area_struct
*vma
;
1229 struct mm_struct
*mm
= current
->mm
;
1230 unsigned long addr
= addr0
;
1232 /* requested length too big for entire address space */
1233 if (len
> TASK_SIZE
)
1236 /* requesting a specific address */
1238 addr
= PAGE_ALIGN(addr
);
1239 vma
= find_vma(mm
, addr
);
1240 if (TASK_SIZE
- len
>= addr
&&
1241 (!vma
|| addr
+ len
<= vma
->vm_start
))
1245 /* check if free_area_cache is useful for us */
1246 if (len
<= mm
->cached_hole_size
) {
1247 mm
->cached_hole_size
= 0;
1248 mm
->free_area_cache
= mm
->mmap_base
;
1251 /* either no address requested or can't fit in requested address hole */
1252 addr
= mm
->free_area_cache
;
1254 /* make sure it can fit in the remaining address space */
1256 vma
= find_vma(mm
, addr
-len
);
1257 if (!vma
|| addr
<= vma
->vm_start
)
1258 /* remember the address as a hint for next time */
1259 return (mm
->free_area_cache
= addr
-len
);
1262 if (mm
->mmap_base
< len
)
1265 addr
= mm
->mmap_base
-len
;
1269 * Lookup failure means no vma is above this address,
1270 * else if new region fits below vma->vm_start,
1271 * return with success:
1273 vma
= find_vma(mm
, addr
);
1274 if (!vma
|| addr
+len
<= vma
->vm_start
)
1275 /* remember the address as a hint for next time */
1276 return (mm
->free_area_cache
= addr
);
1278 /* remember the largest hole we saw so far */
1279 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1280 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1282 /* try just below the current vma->vm_start */
1283 addr
= vma
->vm_start
-len
;
1284 } while (len
< vma
->vm_start
);
1288 * A failed mmap() very likely causes application failure,
1289 * so fall back to the bottom-up function here. This scenario
1290 * can happen with large stack limits and large mmap()
1293 mm
->cached_hole_size
= ~0UL;
1294 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1295 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1297 * Restore the topdown base:
1299 mm
->free_area_cache
= mm
->mmap_base
;
1300 mm
->cached_hole_size
= ~0UL;
1306 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1309 * Is this a new hole at the highest possible address?
1311 if (addr
> mm
->free_area_cache
)
1312 mm
->free_area_cache
= addr
;
1314 /* dont allow allocations above current base */
1315 if (mm
->free_area_cache
> mm
->mmap_base
)
1316 mm
->free_area_cache
= mm
->mmap_base
;
1320 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1321 unsigned long pgoff
, unsigned long flags
)
1325 if (!(flags
& MAP_FIXED
)) {
1326 unsigned long (*get_area
)(struct file
*, unsigned long, unsigned long, unsigned long, unsigned long);
1328 get_area
= current
->mm
->get_unmapped_area
;
1329 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1330 get_area
= file
->f_op
->get_unmapped_area
;
1331 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1332 if (IS_ERR_VALUE(addr
))
1336 if (addr
> TASK_SIZE
- len
)
1338 if (addr
& ~PAGE_MASK
)
1340 if (file
&& is_file_hugepages(file
)) {
1342 * Check if the given range is hugepage aligned, and
1343 * can be made suitable for hugepages.
1345 ret
= prepare_hugepage_range(addr
, len
);
1348 * Ensure that a normal request is not falling in a
1349 * reserved hugepage range. For some archs like IA-64,
1350 * there is a separate region for hugepages.
1352 ret
= is_hugepage_only_range(current
->mm
, addr
, len
);
1359 EXPORT_SYMBOL(get_unmapped_area
);
1361 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1362 struct vm_area_struct
* find_vma(struct mm_struct
* mm
, unsigned long addr
)
1364 struct vm_area_struct
*vma
= NULL
;
1367 /* Check the cache first. */
1368 /* (Cache hit rate is typically around 35%.) */
1369 vma
= mm
->mmap_cache
;
1370 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1371 struct rb_node
* rb_node
;
1373 rb_node
= mm
->mm_rb
.rb_node
;
1377 struct vm_area_struct
* vma_tmp
;
1379 vma_tmp
= rb_entry(rb_node
,
1380 struct vm_area_struct
, vm_rb
);
1382 if (vma_tmp
->vm_end
> addr
) {
1384 if (vma_tmp
->vm_start
<= addr
)
1386 rb_node
= rb_node
->rb_left
;
1388 rb_node
= rb_node
->rb_right
;
1391 mm
->mmap_cache
= vma
;
1397 EXPORT_SYMBOL(find_vma
);
1399 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1400 struct vm_area_struct
*
1401 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1402 struct vm_area_struct
**pprev
)
1404 struct vm_area_struct
*vma
= NULL
, *prev
= NULL
;
1405 struct rb_node
* rb_node
;
1409 /* Guard against addr being lower than the first VMA */
1412 /* Go through the RB tree quickly. */
1413 rb_node
= mm
->mm_rb
.rb_node
;
1416 struct vm_area_struct
*vma_tmp
;
1417 vma_tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1419 if (addr
< vma_tmp
->vm_end
) {
1420 rb_node
= rb_node
->rb_left
;
1423 if (!prev
->vm_next
|| (addr
< prev
->vm_next
->vm_end
))
1425 rb_node
= rb_node
->rb_right
;
1431 return prev
? prev
->vm_next
: vma
;
1435 * Verify that the stack growth is acceptable and
1436 * update accounting. This is shared with both the
1437 * grow-up and grow-down cases.
1439 static int acct_stack_growth(struct vm_area_struct
* vma
, unsigned long size
, unsigned long grow
)
1441 struct mm_struct
*mm
= vma
->vm_mm
;
1442 struct rlimit
*rlim
= current
->signal
->rlim
;
1444 /* address space limit tests */
1445 if (!may_expand_vm(mm
, grow
))
1448 /* Stack limit test */
1449 if (size
> rlim
[RLIMIT_STACK
].rlim_cur
)
1452 /* mlock limit tests */
1453 if (vma
->vm_flags
& VM_LOCKED
) {
1454 unsigned long locked
;
1455 unsigned long limit
;
1456 locked
= mm
->locked_vm
+ grow
;
1457 limit
= rlim
[RLIMIT_MEMLOCK
].rlim_cur
>> PAGE_SHIFT
;
1458 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1463 * Overcommit.. This must be the final test, as it will
1464 * update security statistics.
1466 if (security_vm_enough_memory(grow
))
1469 /* Ok, everything looks good - let it rip */
1470 mm
->total_vm
+= grow
;
1471 if (vma
->vm_flags
& VM_LOCKED
)
1472 mm
->locked_vm
+= grow
;
1473 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1477 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1479 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1480 * vma is the last one with address > vma->vm_end. Have to extend vma.
1485 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1489 if (!(vma
->vm_flags
& VM_GROWSUP
))
1493 * We must make sure the anon_vma is allocated
1494 * so that the anon_vma locking is not a noop.
1496 if (unlikely(anon_vma_prepare(vma
)))
1501 * vma->vm_start/vm_end cannot change under us because the caller
1502 * is required to hold the mmap_sem in read mode. We need the
1503 * anon_vma lock to serialize against concurrent expand_stacks.
1505 address
+= 4 + PAGE_SIZE
- 1;
1506 address
&= PAGE_MASK
;
1509 /* Somebody else might have raced and expanded it already */
1510 if (address
> vma
->vm_end
) {
1511 unsigned long size
, grow
;
1513 size
= address
- vma
->vm_start
;
1514 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1516 error
= acct_stack_growth(vma
, size
, grow
);
1518 vma
->vm_end
= address
;
1520 anon_vma_unlock(vma
);
1523 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1525 #ifdef CONFIG_STACK_GROWSUP
1526 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1528 return expand_upwards(vma
, address
);
1531 struct vm_area_struct
*
1532 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1534 struct vm_area_struct
*vma
, *prev
;
1537 vma
= find_vma_prev(mm
, addr
, &prev
);
1538 if (vma
&& (vma
->vm_start
<= addr
))
1540 if (!prev
|| expand_stack(prev
, addr
))
1542 if (prev
->vm_flags
& VM_LOCKED
) {
1543 make_pages_present(addr
, prev
->vm_end
);
1549 * vma is the first one with address < vma->vm_start. Have to extend vma.
1551 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1556 * We must make sure the anon_vma is allocated
1557 * so that the anon_vma locking is not a noop.
1559 if (unlikely(anon_vma_prepare(vma
)))
1564 * vma->vm_start/vm_end cannot change under us because the caller
1565 * is required to hold the mmap_sem in read mode. We need the
1566 * anon_vma lock to serialize against concurrent expand_stacks.
1568 address
&= PAGE_MASK
;
1571 /* Somebody else might have raced and expanded it already */
1572 if (address
< vma
->vm_start
) {
1573 unsigned long size
, grow
;
1575 size
= vma
->vm_end
- address
;
1576 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1578 error
= acct_stack_growth(vma
, size
, grow
);
1580 vma
->vm_start
= address
;
1581 vma
->vm_pgoff
-= grow
;
1584 anon_vma_unlock(vma
);
1588 struct vm_area_struct
*
1589 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1591 struct vm_area_struct
* vma
;
1592 unsigned long start
;
1595 vma
= find_vma(mm
,addr
);
1598 if (vma
->vm_start
<= addr
)
1600 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1602 start
= vma
->vm_start
;
1603 if (expand_stack(vma
, addr
))
1605 if (vma
->vm_flags
& VM_LOCKED
) {
1606 make_pages_present(addr
, start
);
1613 * Ok - we have the memory areas we should free on the vma list,
1614 * so release them, and do the vma updates.
1616 * Called with the mm semaphore held.
1618 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1620 /* Update high watermark before we lower total_vm */
1621 update_hiwater_vm(mm
);
1623 long nrpages
= vma_pages(vma
);
1625 mm
->total_vm
-= nrpages
;
1626 if (vma
->vm_flags
& VM_LOCKED
)
1627 mm
->locked_vm
-= nrpages
;
1628 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1629 vma
= remove_vma(vma
);
1635 * Get rid of page table information in the indicated region.
1637 * Called with the mm semaphore held.
1639 static void unmap_region(struct mm_struct
*mm
,
1640 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1641 unsigned long start
, unsigned long end
)
1643 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1644 struct mmu_gather
*tlb
;
1645 unsigned long nr_accounted
= 0;
1648 tlb
= tlb_gather_mmu(mm
, 0);
1649 update_hiwater_rss(mm
);
1650 unmap_vmas(&tlb
, vma
, start
, end
, &nr_accounted
, NULL
);
1651 vm_unacct_memory(nr_accounted
);
1652 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1653 next
? next
->vm_start
: 0);
1654 tlb_finish_mmu(tlb
, start
, end
);
1658 * Create a list of vma's touched by the unmap, removing them from the mm's
1659 * vma list as we go..
1662 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1663 struct vm_area_struct
*prev
, unsigned long end
)
1665 struct vm_area_struct
**insertion_point
;
1666 struct vm_area_struct
*tail_vma
= NULL
;
1669 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1671 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1675 } while (vma
&& vma
->vm_start
< end
);
1676 *insertion_point
= vma
;
1677 tail_vma
->vm_next
= NULL
;
1678 if (mm
->unmap_area
== arch_unmap_area
)
1679 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1681 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1682 mm
->unmap_area(mm
, addr
);
1683 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1687 * Split a vma into two pieces at address 'addr', a new vma is allocated
1688 * either for the first part or the the tail.
1690 int split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1691 unsigned long addr
, int new_below
)
1693 struct mempolicy
*pol
;
1694 struct vm_area_struct
*new;
1696 if (is_vm_hugetlb_page(vma
) && (addr
& ~HPAGE_MASK
))
1699 if (mm
->map_count
>= sysctl_max_map_count
)
1702 new = kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
1706 /* most fields are the same, copy all, and then fixup */
1712 new->vm_start
= addr
;
1713 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
1716 pol
= mpol_copy(vma_policy(vma
));
1718 kmem_cache_free(vm_area_cachep
, new);
1719 return PTR_ERR(pol
);
1721 vma_set_policy(new, pol
);
1724 get_file(new->vm_file
);
1726 if (new->vm_ops
&& new->vm_ops
->open
)
1727 new->vm_ops
->open(new);
1730 vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
1731 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
1733 vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
1738 /* Munmap is split into 2 main parts -- this part which finds
1739 * what needs doing, and the areas themselves, which do the
1740 * work. This now handles partial unmappings.
1741 * Jeremy Fitzhardinge <jeremy@goop.org>
1743 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1746 struct vm_area_struct
*vma
, *prev
, *last
;
1748 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
1751 if ((len
= PAGE_ALIGN(len
)) == 0)
1754 /* Find the first overlapping VMA */
1755 vma
= find_vma_prev(mm
, start
, &prev
);
1758 /* we have start < vma->vm_end */
1760 /* if it doesn't overlap, we have nothing.. */
1762 if (vma
->vm_start
>= end
)
1766 * If we need to split any vma, do it now to save pain later.
1768 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1769 * unmapped vm_area_struct will remain in use: so lower split_vma
1770 * places tmp vma above, and higher split_vma places tmp vma below.
1772 if (start
> vma
->vm_start
) {
1773 int error
= split_vma(mm
, vma
, start
, 0);
1779 /* Does it split the last one? */
1780 last
= find_vma(mm
, end
);
1781 if (last
&& end
> last
->vm_start
) {
1782 int error
= split_vma(mm
, last
, end
, 1);
1786 vma
= prev
? prev
->vm_next
: mm
->mmap
;
1789 * Remove the vma's, and unmap the actual pages
1791 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
1792 unmap_region(mm
, vma
, prev
, start
, end
);
1794 /* Fix up all other VM information */
1795 remove_vma_list(mm
, vma
);
1800 EXPORT_SYMBOL(do_munmap
);
1802 asmlinkage
long sys_munmap(unsigned long addr
, size_t len
)
1805 struct mm_struct
*mm
= current
->mm
;
1807 profile_munmap(addr
);
1809 down_write(&mm
->mmap_sem
);
1810 ret
= do_munmap(mm
, addr
, len
);
1811 up_write(&mm
->mmap_sem
);
1815 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
1817 #ifdef CONFIG_DEBUG_VM
1818 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
1820 up_read(&mm
->mmap_sem
);
1826 * this is really a simplified "do_mmap". it only handles
1827 * anonymous maps. eventually we may be able to do some
1828 * brk-specific accounting here.
1830 unsigned long do_brk(unsigned long addr
, unsigned long len
)
1832 struct mm_struct
* mm
= current
->mm
;
1833 struct vm_area_struct
* vma
, * prev
;
1834 unsigned long flags
;
1835 struct rb_node
** rb_link
, * rb_parent
;
1836 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
1838 len
= PAGE_ALIGN(len
);
1842 if ((addr
+ len
) > TASK_SIZE
|| (addr
+ len
) < addr
)
1848 if (mm
->def_flags
& VM_LOCKED
) {
1849 unsigned long locked
, lock_limit
;
1850 locked
= len
>> PAGE_SHIFT
;
1851 locked
+= mm
->locked_vm
;
1852 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
1853 lock_limit
>>= PAGE_SHIFT
;
1854 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1859 * mm->mmap_sem is required to protect against another thread
1860 * changing the mappings in case we sleep.
1862 verify_mm_writelocked(mm
);
1865 * Clear old maps. this also does some error checking for us
1868 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1869 if (vma
&& vma
->vm_start
< addr
+ len
) {
1870 if (do_munmap(mm
, addr
, len
))
1875 /* Check against address space limits *after* clearing old maps... */
1876 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1879 if (mm
->map_count
> sysctl_max_map_count
)
1882 if (security_vm_enough_memory(len
>> PAGE_SHIFT
))
1885 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
1887 /* Can we just expand an old private anonymous mapping? */
1888 if (vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
1889 NULL
, NULL
, pgoff
, NULL
))
1893 * create a vma struct for an anonymous mapping
1895 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1897 vm_unacct_memory(len
>> PAGE_SHIFT
);
1902 vma
->vm_start
= addr
;
1903 vma
->vm_end
= addr
+ len
;
1904 vma
->vm_pgoff
= pgoff
;
1905 vma
->vm_flags
= flags
;
1906 vma
->vm_page_prot
= protection_map
[flags
& 0x0f];
1907 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1909 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1910 if (flags
& VM_LOCKED
) {
1911 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
1912 make_pages_present(addr
, addr
+ len
);
1917 EXPORT_SYMBOL(do_brk
);
1919 /* Release all mmaps. */
1920 void exit_mmap(struct mm_struct
*mm
)
1922 struct mmu_gather
*tlb
;
1923 struct vm_area_struct
*vma
= mm
->mmap
;
1924 unsigned long nr_accounted
= 0;
1929 tlb
= tlb_gather_mmu(mm
, 1);
1930 /* Don't update_hiwater_rss(mm) here, do_exit already did */
1931 /* Use -1 here to ensure all VMAs in the mm are unmapped */
1932 end
= unmap_vmas(&tlb
, vma
, 0, -1, &nr_accounted
, NULL
);
1933 vm_unacct_memory(nr_accounted
);
1934 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, 0);
1935 tlb_finish_mmu(tlb
, 0, end
);
1938 * Walk the list again, actually closing and freeing it,
1939 * with preemption enabled, without holding any MM locks.
1942 vma
= remove_vma(vma
);
1944 BUG_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
1947 /* Insert vm structure into process list sorted by address
1948 * and into the inode's i_mmap tree. If vm_file is non-NULL
1949 * then i_mmap_lock is taken here.
1951 int insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
1953 struct vm_area_struct
* __vma
, * prev
;
1954 struct rb_node
** rb_link
, * rb_parent
;
1957 * The vm_pgoff of a purely anonymous vma should be irrelevant
1958 * until its first write fault, when page's anon_vma and index
1959 * are set. But now set the vm_pgoff it will almost certainly
1960 * end up with (unless mremap moves it elsewhere before that
1961 * first wfault), so /proc/pid/maps tells a consistent story.
1963 * By setting it to reflect the virtual start address of the
1964 * vma, merges and splits can happen in a seamless way, just
1965 * using the existing file pgoff checks and manipulations.
1966 * Similarly in do_mmap_pgoff and in do_brk.
1968 if (!vma
->vm_file
) {
1969 BUG_ON(vma
->anon_vma
);
1970 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
1972 __vma
= find_vma_prepare(mm
,vma
->vm_start
,&prev
,&rb_link
,&rb_parent
);
1973 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
1975 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
1976 security_vm_enough_memory(vma_pages(vma
)))
1978 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1983 * Copy the vma structure to a new location in the same mm,
1984 * prior to moving page table entries, to effect an mremap move.
1986 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
1987 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
1989 struct vm_area_struct
*vma
= *vmap
;
1990 unsigned long vma_start
= vma
->vm_start
;
1991 struct mm_struct
*mm
= vma
->vm_mm
;
1992 struct vm_area_struct
*new_vma
, *prev
;
1993 struct rb_node
**rb_link
, *rb_parent
;
1994 struct mempolicy
*pol
;
1997 * If anonymous vma has not yet been faulted, update new pgoff
1998 * to match new location, to increase its chance of merging.
2000 if (!vma
->vm_file
&& !vma
->anon_vma
)
2001 pgoff
= addr
>> PAGE_SHIFT
;
2003 find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2004 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2005 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2008 * Source vma may have been merged into new_vma
2010 if (vma_start
>= new_vma
->vm_start
&&
2011 vma_start
< new_vma
->vm_end
)
2014 new_vma
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
2017 pol
= mpol_copy(vma_policy(vma
));
2019 kmem_cache_free(vm_area_cachep
, new_vma
);
2022 vma_set_policy(new_vma
, pol
);
2023 new_vma
->vm_start
= addr
;
2024 new_vma
->vm_end
= addr
+ len
;
2025 new_vma
->vm_pgoff
= pgoff
;
2026 if (new_vma
->vm_file
)
2027 get_file(new_vma
->vm_file
);
2028 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2029 new_vma
->vm_ops
->open(new_vma
);
2030 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2037 * Return true if the calling process may expand its vm space by the passed
2040 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2042 unsigned long cur
= mm
->total_vm
; /* pages */
2045 lim
= current
->signal
->rlim
[RLIMIT_AS
].rlim_cur
>> PAGE_SHIFT
;
2047 if (cur
+ npages
> lim
)