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 #ifndef arch_mmap_check
34 #define arch_mmap_check(addr, len, flags) (0)
37 static void unmap_region(struct mm_struct
*mm
,
38 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
39 unsigned long start
, unsigned long end
);
42 * WARNING: the debugging will use recursive algorithms so never enable this
43 * unless you know what you are doing.
47 /* description of effects of mapping type and prot in current implementation.
48 * this is due to the limited x86 page protection hardware. The expected
49 * behavior is in parens:
52 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
53 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
54 * w: (no) no w: (no) no w: (yes) yes w: (no) no
55 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
57 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
58 * w: (no) no w: (no) no w: (copy) copy w: (no) no
59 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
62 pgprot_t protection_map
[16] = {
63 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
64 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
67 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
69 return protection_map
[vm_flags
&
70 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)];
72 EXPORT_SYMBOL(vm_get_page_prot
);
74 int sysctl_overcommit_memory
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
75 int sysctl_overcommit_ratio
= 50; /* default is 50% */
76 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
77 atomic_t vm_committed_space
= ATOMIC_INIT(0);
80 * Check that a process has enough memory to allocate a new virtual
81 * mapping. 0 means there is enough memory for the allocation to
82 * succeed and -ENOMEM implies there is not.
84 * We currently support three overcommit policies, which are set via the
85 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
87 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
88 * Additional code 2002 Jul 20 by Robert Love.
90 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
92 * Note this is a helper function intended to be used by LSMs which
93 * wish to use this logic.
95 int __vm_enough_memory(long pages
, int cap_sys_admin
)
97 unsigned long free
, allowed
;
99 vm_acct_memory(pages
);
102 * Sometimes we want to use more memory than we have
104 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
107 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
110 free
= global_page_state(NR_FILE_PAGES
);
111 free
+= nr_swap_pages
;
114 * Any slabs which are created with the
115 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
116 * which are reclaimable, under pressure. The dentry
117 * cache and most inode caches should fall into this
119 free
+= atomic_read(&slab_reclaim_pages
);
122 * Leave the last 3% for root
131 * nr_free_pages() is very expensive on large systems,
132 * only call if we're about to fail.
137 * Leave reserved pages. The pages are not for anonymous pages.
139 if (n
<= totalreserve_pages
)
142 n
-= totalreserve_pages
;
145 * Leave the last 3% for root
157 allowed
= (totalram_pages
- hugetlb_total_pages())
158 * sysctl_overcommit_ratio
/ 100;
160 * Leave the last 3% for root
163 allowed
-= allowed
/ 32;
164 allowed
+= total_swap_pages
;
166 /* Don't let a single process grow too big:
167 leave 3% of the size of this process for other processes */
168 allowed
-= current
->mm
->total_vm
/ 32;
171 * cast `allowed' as a signed long because vm_committed_space
172 * sometimes has a negative value
174 if (atomic_read(&vm_committed_space
) < (long)allowed
)
177 vm_unacct_memory(pages
);
182 EXPORT_SYMBOL(__vm_enough_memory
);
185 * Requires inode->i_mapping->i_mmap_lock
187 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
188 struct file
*file
, struct address_space
*mapping
)
190 if (vma
->vm_flags
& VM_DENYWRITE
)
191 atomic_inc(&file
->f_dentry
->d_inode
->i_writecount
);
192 if (vma
->vm_flags
& VM_SHARED
)
193 mapping
->i_mmap_writable
--;
195 flush_dcache_mmap_lock(mapping
);
196 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
197 list_del_init(&vma
->shared
.vm_set
.list
);
199 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
200 flush_dcache_mmap_unlock(mapping
);
204 * Unlink a file-based vm structure from its prio_tree, to hide
205 * vma from rmap and vmtruncate before freeing its page tables.
207 void unlink_file_vma(struct vm_area_struct
*vma
)
209 struct file
*file
= vma
->vm_file
;
212 struct address_space
*mapping
= file
->f_mapping
;
213 spin_lock(&mapping
->i_mmap_lock
);
214 __remove_shared_vm_struct(vma
, file
, mapping
);
215 spin_unlock(&mapping
->i_mmap_lock
);
220 * Close a vm structure and free it, returning the next.
222 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
224 struct vm_area_struct
*next
= vma
->vm_next
;
227 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
228 vma
->vm_ops
->close(vma
);
231 mpol_free(vma_policy(vma
));
232 kmem_cache_free(vm_area_cachep
, vma
);
236 asmlinkage
unsigned long sys_brk(unsigned long brk
)
238 unsigned long rlim
, retval
;
239 unsigned long newbrk
, oldbrk
;
240 struct mm_struct
*mm
= current
->mm
;
242 down_write(&mm
->mmap_sem
);
244 if (brk
< mm
->end_code
)
248 * Check against rlimit here. If this check is done later after the test
249 * of oldbrk with newbrk then it can escape the test and let the data
250 * segment grow beyond its set limit the in case where the limit is
251 * not page aligned -Ram Gupta
253 rlim
= current
->signal
->rlim
[RLIMIT_DATA
].rlim_cur
;
254 if (rlim
< RLIM_INFINITY
&& brk
- mm
->start_data
> rlim
)
257 newbrk
= PAGE_ALIGN(brk
);
258 oldbrk
= PAGE_ALIGN(mm
->brk
);
259 if (oldbrk
== newbrk
)
262 /* Always allow shrinking brk. */
263 if (brk
<= mm
->brk
) {
264 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
269 /* Check against existing mmap mappings. */
270 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
273 /* Ok, looks good - let it rip. */
274 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
280 up_write(&mm
->mmap_sem
);
285 static int browse_rb(struct rb_root
*root
)
288 struct rb_node
*nd
, *pn
= NULL
;
289 unsigned long prev
= 0, pend
= 0;
291 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
292 struct vm_area_struct
*vma
;
293 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
294 if (vma
->vm_start
< prev
)
295 printk("vm_start %lx prev %lx\n", vma
->vm_start
, prev
), i
= -1;
296 if (vma
->vm_start
< pend
)
297 printk("vm_start %lx pend %lx\n", vma
->vm_start
, pend
);
298 if (vma
->vm_start
> vma
->vm_end
)
299 printk("vm_end %lx < vm_start %lx\n", vma
->vm_end
, vma
->vm_start
);
304 for (nd
= pn
; nd
; nd
= rb_prev(nd
)) {
308 printk("backwards %d, forwards %d\n", j
, i
), i
= 0;
312 void validate_mm(struct mm_struct
*mm
)
316 struct vm_area_struct
*tmp
= mm
->mmap
;
321 if (i
!= mm
->map_count
)
322 printk("map_count %d vm_next %d\n", mm
->map_count
, i
), bug
= 1;
323 i
= browse_rb(&mm
->mm_rb
);
324 if (i
!= mm
->map_count
)
325 printk("map_count %d rb %d\n", mm
->map_count
, i
), bug
= 1;
329 #define validate_mm(mm) do { } while (0)
332 static struct vm_area_struct
*
333 find_vma_prepare(struct mm_struct
*mm
, unsigned long addr
,
334 struct vm_area_struct
**pprev
, struct rb_node
***rb_link
,
335 struct rb_node
** rb_parent
)
337 struct vm_area_struct
* vma
;
338 struct rb_node
** __rb_link
, * __rb_parent
, * rb_prev
;
340 __rb_link
= &mm
->mm_rb
.rb_node
;
341 rb_prev
= __rb_parent
= NULL
;
345 struct vm_area_struct
*vma_tmp
;
347 __rb_parent
= *__rb_link
;
348 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
350 if (vma_tmp
->vm_end
> addr
) {
352 if (vma_tmp
->vm_start
<= addr
)
354 __rb_link
= &__rb_parent
->rb_left
;
356 rb_prev
= __rb_parent
;
357 __rb_link
= &__rb_parent
->rb_right
;
363 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
364 *rb_link
= __rb_link
;
365 *rb_parent
= __rb_parent
;
370 __vma_link_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
371 struct vm_area_struct
*prev
, struct rb_node
*rb_parent
)
374 vma
->vm_next
= prev
->vm_next
;
379 vma
->vm_next
= rb_entry(rb_parent
,
380 struct vm_area_struct
, vm_rb
);
386 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
387 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
389 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
390 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
393 static inline void __vma_link_file(struct vm_area_struct
*vma
)
399 struct address_space
*mapping
= file
->f_mapping
;
401 if (vma
->vm_flags
& VM_DENYWRITE
)
402 atomic_dec(&file
->f_dentry
->d_inode
->i_writecount
);
403 if (vma
->vm_flags
& VM_SHARED
)
404 mapping
->i_mmap_writable
++;
406 flush_dcache_mmap_lock(mapping
);
407 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
408 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
410 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
411 flush_dcache_mmap_unlock(mapping
);
416 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
417 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
418 struct rb_node
*rb_parent
)
420 __vma_link_list(mm
, vma
, prev
, rb_parent
);
421 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
422 __anon_vma_link(vma
);
425 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
426 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
427 struct rb_node
*rb_parent
)
429 struct address_space
*mapping
= NULL
;
432 mapping
= vma
->vm_file
->f_mapping
;
435 spin_lock(&mapping
->i_mmap_lock
);
436 vma
->vm_truncate_count
= mapping
->truncate_count
;
440 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
441 __vma_link_file(vma
);
443 anon_vma_unlock(vma
);
445 spin_unlock(&mapping
->i_mmap_lock
);
452 * Helper for vma_adjust in the split_vma insert case:
453 * insert vm structure into list and rbtree and anon_vma,
454 * but it has already been inserted into prio_tree earlier.
457 __insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
459 struct vm_area_struct
* __vma
, * prev
;
460 struct rb_node
** rb_link
, * rb_parent
;
462 __vma
= find_vma_prepare(mm
, vma
->vm_start
,&prev
, &rb_link
, &rb_parent
);
463 BUG_ON(__vma
&& __vma
->vm_start
< vma
->vm_end
);
464 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
469 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
470 struct vm_area_struct
*prev
)
472 prev
->vm_next
= vma
->vm_next
;
473 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
474 if (mm
->mmap_cache
== vma
)
475 mm
->mmap_cache
= prev
;
479 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
480 * is already present in an i_mmap tree without adjusting the tree.
481 * The following helper function should be used when such adjustments
482 * are necessary. The "insert" vma (if any) is to be inserted
483 * before we drop the necessary locks.
485 void vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
486 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
488 struct mm_struct
*mm
= vma
->vm_mm
;
489 struct vm_area_struct
*next
= vma
->vm_next
;
490 struct vm_area_struct
*importer
= NULL
;
491 struct address_space
*mapping
= NULL
;
492 struct prio_tree_root
*root
= NULL
;
493 struct file
*file
= vma
->vm_file
;
494 struct anon_vma
*anon_vma
= NULL
;
495 long adjust_next
= 0;
498 if (next
&& !insert
) {
499 if (end
>= next
->vm_end
) {
501 * vma expands, overlapping all the next, and
502 * perhaps the one after too (mprotect case 6).
504 again
: remove_next
= 1 + (end
> next
->vm_end
);
506 anon_vma
= next
->anon_vma
;
508 } else if (end
> next
->vm_start
) {
510 * vma expands, overlapping part of the next:
511 * mprotect case 5 shifting the boundary up.
513 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
514 anon_vma
= next
->anon_vma
;
516 } else if (end
< vma
->vm_end
) {
518 * vma shrinks, and !insert tells it's not
519 * split_vma inserting another: so it must be
520 * mprotect case 4 shifting the boundary down.
522 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
523 anon_vma
= next
->anon_vma
;
529 mapping
= file
->f_mapping
;
530 if (!(vma
->vm_flags
& VM_NONLINEAR
))
531 root
= &mapping
->i_mmap
;
532 spin_lock(&mapping
->i_mmap_lock
);
534 vma
->vm_truncate_count
!= next
->vm_truncate_count
) {
536 * unmap_mapping_range might be in progress:
537 * ensure that the expanding vma is rescanned.
539 importer
->vm_truncate_count
= 0;
542 insert
->vm_truncate_count
= vma
->vm_truncate_count
;
544 * Put into prio_tree now, so instantiated pages
545 * are visible to arm/parisc __flush_dcache_page
546 * throughout; but we cannot insert into address
547 * space until vma start or end is updated.
549 __vma_link_file(insert
);
554 * When changing only vma->vm_end, we don't really need
555 * anon_vma lock: but is that case worth optimizing out?
558 anon_vma
= vma
->anon_vma
;
560 spin_lock(&anon_vma
->lock
);
562 * Easily overlooked: when mprotect shifts the boundary,
563 * make sure the expanding vma has anon_vma set if the
564 * shrinking vma had, to cover any anon pages imported.
566 if (importer
&& !importer
->anon_vma
) {
567 importer
->anon_vma
= anon_vma
;
568 __anon_vma_link(importer
);
573 flush_dcache_mmap_lock(mapping
);
574 vma_prio_tree_remove(vma
, root
);
576 vma_prio_tree_remove(next
, root
);
579 vma
->vm_start
= start
;
581 vma
->vm_pgoff
= pgoff
;
583 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
584 next
->vm_pgoff
+= adjust_next
;
589 vma_prio_tree_insert(next
, root
);
590 vma_prio_tree_insert(vma
, root
);
591 flush_dcache_mmap_unlock(mapping
);
596 * vma_merge has merged next into vma, and needs
597 * us to remove next before dropping the locks.
599 __vma_unlink(mm
, next
, vma
);
601 __remove_shared_vm_struct(next
, file
, mapping
);
603 __anon_vma_merge(vma
, next
);
606 * split_vma has split insert from vma, and needs
607 * us to insert it before dropping the locks
608 * (it may either follow vma or precede it).
610 __insert_vm_struct(mm
, insert
);
614 spin_unlock(&anon_vma
->lock
);
616 spin_unlock(&mapping
->i_mmap_lock
);
622 mpol_free(vma_policy(next
));
623 kmem_cache_free(vm_area_cachep
, next
);
625 * In mprotect's case 6 (see comments on vma_merge),
626 * we must remove another next too. It would clutter
627 * up the code too much to do both in one go.
629 if (remove_next
== 2) {
639 * If the vma has a ->close operation then the driver probably needs to release
640 * per-vma resources, so we don't attempt to merge those.
642 #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP)
644 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
645 struct file
*file
, unsigned long vm_flags
)
647 if (vma
->vm_flags
!= vm_flags
)
649 if (vma
->vm_file
!= file
)
651 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
656 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
657 struct anon_vma
*anon_vma2
)
659 return !anon_vma1
|| !anon_vma2
|| (anon_vma1
== anon_vma2
);
663 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
664 * in front of (at a lower virtual address and file offset than) the vma.
666 * We cannot merge two vmas if they have differently assigned (non-NULL)
667 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
669 * We don't check here for the merged mmap wrapping around the end of pagecache
670 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
671 * wrap, nor mmaps which cover the final page at index -1UL.
674 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
675 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
677 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
678 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
679 if (vma
->vm_pgoff
== vm_pgoff
)
686 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
687 * beyond (at a higher virtual address and file offset than) the vma.
689 * We cannot merge two vmas if they have differently assigned (non-NULL)
690 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
693 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
694 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
696 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
697 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
699 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
700 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
707 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
708 * whether that can be merged with its predecessor or its successor.
709 * Or both (it neatly fills a hole).
711 * In most cases - when called for mmap, brk or mremap - [addr,end) is
712 * certain not to be mapped by the time vma_merge is called; but when
713 * called for mprotect, it is certain to be already mapped (either at
714 * an offset within prev, or at the start of next), and the flags of
715 * this area are about to be changed to vm_flags - and the no-change
716 * case has already been eliminated.
718 * The following mprotect cases have to be considered, where AAAA is
719 * the area passed down from mprotect_fixup, never extending beyond one
720 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
722 * AAAA AAAA AAAA AAAA
723 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
724 * cannot merge might become might become might become
725 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
726 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
727 * mremap move: PPPPNNNNNNNN 8
729 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
730 * might become case 1 below case 2 below case 3 below
732 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
733 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
735 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
736 struct vm_area_struct
*prev
, unsigned long addr
,
737 unsigned long end
, unsigned long vm_flags
,
738 struct anon_vma
*anon_vma
, struct file
*file
,
739 pgoff_t pgoff
, struct mempolicy
*policy
)
741 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
742 struct vm_area_struct
*area
, *next
;
745 * We later require that vma->vm_flags == vm_flags,
746 * so this tests vma->vm_flags & VM_SPECIAL, too.
748 if (vm_flags
& VM_SPECIAL
)
752 next
= prev
->vm_next
;
756 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
757 next
= next
->vm_next
;
760 * Can it merge with the predecessor?
762 if (prev
&& prev
->vm_end
== addr
&&
763 mpol_equal(vma_policy(prev
), policy
) &&
764 can_vma_merge_after(prev
, vm_flags
,
765 anon_vma
, file
, pgoff
)) {
767 * OK, it can. Can we now merge in the successor as well?
769 if (next
&& end
== next
->vm_start
&&
770 mpol_equal(policy
, vma_policy(next
)) &&
771 can_vma_merge_before(next
, vm_flags
,
772 anon_vma
, file
, pgoff
+pglen
) &&
773 is_mergeable_anon_vma(prev
->anon_vma
,
776 vma_adjust(prev
, prev
->vm_start
,
777 next
->vm_end
, prev
->vm_pgoff
, NULL
);
778 } else /* cases 2, 5, 7 */
779 vma_adjust(prev
, prev
->vm_start
,
780 end
, prev
->vm_pgoff
, NULL
);
785 * Can this new request be merged in front of next?
787 if (next
&& end
== next
->vm_start
&&
788 mpol_equal(policy
, vma_policy(next
)) &&
789 can_vma_merge_before(next
, vm_flags
,
790 anon_vma
, file
, pgoff
+pglen
)) {
791 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
792 vma_adjust(prev
, prev
->vm_start
,
793 addr
, prev
->vm_pgoff
, NULL
);
794 else /* cases 3, 8 */
795 vma_adjust(area
, addr
, next
->vm_end
,
796 next
->vm_pgoff
- pglen
, NULL
);
804 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
805 * neighbouring vmas for a suitable anon_vma, before it goes off
806 * to allocate a new anon_vma. It checks because a repetitive
807 * sequence of mprotects and faults may otherwise lead to distinct
808 * anon_vmas being allocated, preventing vma merge in subsequent
811 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
813 struct vm_area_struct
*near
;
814 unsigned long vm_flags
;
821 * Since only mprotect tries to remerge vmas, match flags
822 * which might be mprotected into each other later on.
823 * Neither mlock nor madvise tries to remerge at present,
824 * so leave their flags as obstructing a merge.
826 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
827 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
829 if (near
->anon_vma
&& vma
->vm_end
== near
->vm_start
&&
830 mpol_equal(vma_policy(vma
), vma_policy(near
)) &&
831 can_vma_merge_before(near
, vm_flags
,
832 NULL
, vma
->vm_file
, vma
->vm_pgoff
+
833 ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
)))
834 return near
->anon_vma
;
837 * It is potentially slow to have to call find_vma_prev here.
838 * But it's only on the first write fault on the vma, not
839 * every time, and we could devise a way to avoid it later
840 * (e.g. stash info in next's anon_vma_node when assigning
841 * an anon_vma, or when trying vma_merge). Another time.
843 BUG_ON(find_vma_prev(vma
->vm_mm
, vma
->vm_start
, &near
) != vma
);
847 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
848 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
850 if (near
->anon_vma
&& near
->vm_end
== vma
->vm_start
&&
851 mpol_equal(vma_policy(near
), vma_policy(vma
)) &&
852 can_vma_merge_after(near
, vm_flags
,
853 NULL
, vma
->vm_file
, vma
->vm_pgoff
))
854 return near
->anon_vma
;
857 * There's no absolute need to look only at touching neighbours:
858 * we could search further afield for "compatible" anon_vmas.
859 * But it would probably just be a waste of time searching,
860 * or lead to too many vmas hanging off the same anon_vma.
861 * We're trying to allow mprotect remerging later on,
862 * not trying to minimize memory used for anon_vmas.
867 #ifdef CONFIG_PROC_FS
868 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
869 struct file
*file
, long pages
)
871 const unsigned long stack_flags
872 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
875 mm
->shared_vm
+= pages
;
876 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
877 mm
->exec_vm
+= pages
;
878 } else if (flags
& stack_flags
)
879 mm
->stack_vm
+= pages
;
880 if (flags
& (VM_RESERVED
|VM_IO
))
881 mm
->reserved_vm
+= pages
;
883 #endif /* CONFIG_PROC_FS */
886 * The caller must hold down_write(current->mm->mmap_sem).
889 unsigned long do_mmap_pgoff(struct file
* file
, unsigned long addr
,
890 unsigned long len
, unsigned long prot
,
891 unsigned long flags
, unsigned long pgoff
)
893 struct mm_struct
* mm
= current
->mm
;
894 struct vm_area_struct
* vma
, * prev
;
896 unsigned int vm_flags
;
897 int correct_wcount
= 0;
899 struct rb_node
** rb_link
, * rb_parent
;
901 unsigned long charged
= 0, reqprot
= prot
;
904 if (is_file_hugepages(file
))
907 if (!file
->f_op
|| !file
->f_op
->mmap
)
910 if ((prot
& PROT_EXEC
) &&
911 (file
->f_vfsmnt
->mnt_flags
& MNT_NOEXEC
))
915 * Does the application expect PROT_READ to imply PROT_EXEC?
917 * (the exception is when the underlying filesystem is noexec
918 * mounted, in which case we dont add PROT_EXEC.)
920 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
921 if (!(file
&& (file
->f_vfsmnt
->mnt_flags
& MNT_NOEXEC
)))
927 error
= arch_mmap_check(addr
, len
, flags
);
931 /* Careful about overflows.. */
932 len
= PAGE_ALIGN(len
);
933 if (!len
|| len
> TASK_SIZE
)
936 /* offset overflow? */
937 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
940 /* Too many mappings? */
941 if (mm
->map_count
> sysctl_max_map_count
)
944 /* Obtain the address to map to. we verify (or select) it and ensure
945 * that it represents a valid section of the address space.
947 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
948 if (addr
& ~PAGE_MASK
)
951 /* Do simple checking here so the lower-level routines won't have
952 * to. we assume access permissions have been handled by the open
953 * of the memory object, so we don't do any here.
955 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
956 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
958 if (flags
& MAP_LOCKED
) {
961 vm_flags
|= VM_LOCKED
;
963 /* mlock MCL_FUTURE? */
964 if (vm_flags
& VM_LOCKED
) {
965 unsigned long locked
, lock_limit
;
966 locked
= len
>> PAGE_SHIFT
;
967 locked
+= mm
->locked_vm
;
968 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
969 lock_limit
>>= PAGE_SHIFT
;
970 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
974 inode
= file
? file
->f_dentry
->d_inode
: NULL
;
977 switch (flags
& MAP_TYPE
) {
979 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
983 * Make sure we don't allow writing to an append-only
986 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
990 * Make sure there are no mandatory locks on the file.
992 if (locks_verify_locked(inode
))
995 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
996 if (!(file
->f_mode
& FMODE_WRITE
))
997 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1001 if (!(file
->f_mode
& FMODE_READ
))
1009 switch (flags
& MAP_TYPE
) {
1011 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1015 * Set pgoff according to addr for anon_vma.
1017 pgoff
= addr
>> PAGE_SHIFT
;
1024 error
= security_file_mmap(file
, reqprot
, prot
, flags
);
1028 /* Clear old maps */
1031 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1032 if (vma
&& vma
->vm_start
< addr
+ len
) {
1033 if (do_munmap(mm
, addr
, len
))
1038 /* Check against address space limit. */
1039 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1042 if (accountable
&& (!(flags
& MAP_NORESERVE
) ||
1043 sysctl_overcommit_memory
== OVERCOMMIT_NEVER
)) {
1044 if (vm_flags
& VM_SHARED
) {
1045 /* Check memory availability in shmem_file_setup? */
1046 vm_flags
|= VM_ACCOUNT
;
1047 } else if (vm_flags
& VM_WRITE
) {
1049 * Private writable mapping: check memory availability
1051 charged
= len
>> PAGE_SHIFT
;
1052 if (security_vm_enough_memory(charged
))
1054 vm_flags
|= VM_ACCOUNT
;
1059 * Can we just expand an old private anonymous mapping?
1060 * The VM_SHARED test is necessary because shmem_zero_setup
1061 * will create the file object for a shared anonymous map below.
1063 if (!file
&& !(vm_flags
& VM_SHARED
) &&
1064 vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
,
1065 NULL
, NULL
, pgoff
, NULL
))
1069 * Determine the object being mapped and call the appropriate
1070 * specific mapper. the address has already been validated, but
1071 * not unmapped, but the maps are removed from the list.
1073 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1080 vma
->vm_start
= addr
;
1081 vma
->vm_end
= addr
+ len
;
1082 vma
->vm_flags
= vm_flags
;
1083 vma
->vm_page_prot
= protection_map
[vm_flags
&
1084 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)];
1085 vma
->vm_pgoff
= pgoff
;
1089 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1091 if (vm_flags
& VM_DENYWRITE
) {
1092 error
= deny_write_access(file
);
1097 vma
->vm_file
= file
;
1099 error
= file
->f_op
->mmap(file
, vma
);
1101 goto unmap_and_free_vma
;
1102 } else if (vm_flags
& VM_SHARED
) {
1103 error
= shmem_zero_setup(vma
);
1108 /* Don't make the VMA automatically writable if it's shared, but the
1109 * backer wishes to know when pages are first written to */
1110 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1112 protection_map
[vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
)];
1114 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1115 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1116 * that memory reservation must be checked; but that reservation
1117 * belongs to shared memory object, not to vma: so now clear it.
1119 if ((vm_flags
& (VM_SHARED
|VM_ACCOUNT
)) == (VM_SHARED
|VM_ACCOUNT
))
1120 vma
->vm_flags
&= ~VM_ACCOUNT
;
1122 /* Can addr have changed??
1124 * Answer: Yes, several device drivers can do it in their
1125 * f_op->mmap method. -DaveM
1127 addr
= vma
->vm_start
;
1128 pgoff
= vma
->vm_pgoff
;
1129 vm_flags
= vma
->vm_flags
;
1131 if (!file
|| !vma_merge(mm
, prev
, addr
, vma
->vm_end
,
1132 vma
->vm_flags
, NULL
, file
, pgoff
, vma_policy(vma
))) {
1133 file
= vma
->vm_file
;
1134 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1136 atomic_inc(&inode
->i_writecount
);
1140 atomic_inc(&inode
->i_writecount
);
1143 mpol_free(vma_policy(vma
));
1144 kmem_cache_free(vm_area_cachep
, vma
);
1147 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1148 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1149 if (vm_flags
& VM_LOCKED
) {
1150 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
1151 make_pages_present(addr
, addr
+ len
);
1153 if (flags
& MAP_POPULATE
) {
1154 up_write(&mm
->mmap_sem
);
1155 sys_remap_file_pages(addr
, len
, 0,
1156 pgoff
, flags
& MAP_NONBLOCK
);
1157 down_write(&mm
->mmap_sem
);
1163 atomic_inc(&inode
->i_writecount
);
1164 vma
->vm_file
= NULL
;
1167 /* Undo any partial mapping done by a device driver. */
1168 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1171 kmem_cache_free(vm_area_cachep
, vma
);
1174 vm_unacct_memory(charged
);
1178 EXPORT_SYMBOL(do_mmap_pgoff
);
1180 /* Get an address range which is currently unmapped.
1181 * For shmat() with addr=0.
1183 * Ugly calling convention alert:
1184 * Return value with the low bits set means error value,
1186 * if (ret & ~PAGE_MASK)
1189 * This function "knows" that -ENOMEM has the bits set.
1191 #ifndef HAVE_ARCH_UNMAPPED_AREA
1193 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1194 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1196 struct mm_struct
*mm
= current
->mm
;
1197 struct vm_area_struct
*vma
;
1198 unsigned long start_addr
;
1200 if (len
> TASK_SIZE
)
1204 addr
= PAGE_ALIGN(addr
);
1205 vma
= find_vma(mm
, addr
);
1206 if (TASK_SIZE
- len
>= addr
&&
1207 (!vma
|| addr
+ len
<= vma
->vm_start
))
1210 if (len
> mm
->cached_hole_size
) {
1211 start_addr
= addr
= mm
->free_area_cache
;
1213 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1214 mm
->cached_hole_size
= 0;
1218 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1219 /* At this point: (!vma || addr < vma->vm_end). */
1220 if (TASK_SIZE
- len
< addr
) {
1222 * Start a new search - just in case we missed
1225 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1226 addr
= TASK_UNMAPPED_BASE
;
1228 mm
->cached_hole_size
= 0;
1233 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1235 * Remember the place where we stopped the search:
1237 mm
->free_area_cache
= addr
+ len
;
1240 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1241 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1247 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1250 * Is this a new hole at the lowest possible address?
1252 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
) {
1253 mm
->free_area_cache
= addr
;
1254 mm
->cached_hole_size
= ~0UL;
1259 * This mmap-allocator allocates new areas top-down from below the
1260 * stack's low limit (the base):
1262 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1264 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1265 const unsigned long len
, const unsigned long pgoff
,
1266 const unsigned long flags
)
1268 struct vm_area_struct
*vma
;
1269 struct mm_struct
*mm
= current
->mm
;
1270 unsigned long addr
= addr0
;
1272 /* requested length too big for entire address space */
1273 if (len
> TASK_SIZE
)
1276 /* requesting a specific address */
1278 addr
= PAGE_ALIGN(addr
);
1279 vma
= find_vma(mm
, addr
);
1280 if (TASK_SIZE
- len
>= addr
&&
1281 (!vma
|| addr
+ len
<= vma
->vm_start
))
1285 /* check if free_area_cache is useful for us */
1286 if (len
<= mm
->cached_hole_size
) {
1287 mm
->cached_hole_size
= 0;
1288 mm
->free_area_cache
= mm
->mmap_base
;
1291 /* either no address requested or can't fit in requested address hole */
1292 addr
= mm
->free_area_cache
;
1294 /* make sure it can fit in the remaining address space */
1296 vma
= find_vma(mm
, addr
-len
);
1297 if (!vma
|| addr
<= vma
->vm_start
)
1298 /* remember the address as a hint for next time */
1299 return (mm
->free_area_cache
= addr
-len
);
1302 if (mm
->mmap_base
< len
)
1305 addr
= mm
->mmap_base
-len
;
1309 * Lookup failure means no vma is above this address,
1310 * else if new region fits below vma->vm_start,
1311 * return with success:
1313 vma
= find_vma(mm
, addr
);
1314 if (!vma
|| addr
+len
<= vma
->vm_start
)
1315 /* remember the address as a hint for next time */
1316 return (mm
->free_area_cache
= addr
);
1318 /* remember the largest hole we saw so far */
1319 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1320 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1322 /* try just below the current vma->vm_start */
1323 addr
= vma
->vm_start
-len
;
1324 } while (len
< vma
->vm_start
);
1328 * A failed mmap() very likely causes application failure,
1329 * so fall back to the bottom-up function here. This scenario
1330 * can happen with large stack limits and large mmap()
1333 mm
->cached_hole_size
= ~0UL;
1334 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1335 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1337 * Restore the topdown base:
1339 mm
->free_area_cache
= mm
->mmap_base
;
1340 mm
->cached_hole_size
= ~0UL;
1346 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1349 * Is this a new hole at the highest possible address?
1351 if (addr
> mm
->free_area_cache
)
1352 mm
->free_area_cache
= addr
;
1354 /* dont allow allocations above current base */
1355 if (mm
->free_area_cache
> mm
->mmap_base
)
1356 mm
->free_area_cache
= mm
->mmap_base
;
1360 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1361 unsigned long pgoff
, unsigned long flags
)
1365 if (!(flags
& MAP_FIXED
)) {
1366 unsigned long (*get_area
)(struct file
*, unsigned long, unsigned long, unsigned long, unsigned long);
1368 get_area
= current
->mm
->get_unmapped_area
;
1369 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1370 get_area
= file
->f_op
->get_unmapped_area
;
1371 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1372 if (IS_ERR_VALUE(addr
))
1376 if (addr
> TASK_SIZE
- len
)
1378 if (addr
& ~PAGE_MASK
)
1380 if (file
&& is_file_hugepages(file
)) {
1382 * Check if the given range is hugepage aligned, and
1383 * can be made suitable for hugepages.
1385 ret
= prepare_hugepage_range(addr
, len
);
1388 * Ensure that a normal request is not falling in a
1389 * reserved hugepage range. For some archs like IA-64,
1390 * there is a separate region for hugepages.
1392 ret
= is_hugepage_only_range(current
->mm
, addr
, len
);
1399 EXPORT_SYMBOL(get_unmapped_area
);
1401 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1402 struct vm_area_struct
* find_vma(struct mm_struct
* mm
, unsigned long addr
)
1404 struct vm_area_struct
*vma
= NULL
;
1407 /* Check the cache first. */
1408 /* (Cache hit rate is typically around 35%.) */
1409 vma
= mm
->mmap_cache
;
1410 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1411 struct rb_node
* rb_node
;
1413 rb_node
= mm
->mm_rb
.rb_node
;
1417 struct vm_area_struct
* vma_tmp
;
1419 vma_tmp
= rb_entry(rb_node
,
1420 struct vm_area_struct
, vm_rb
);
1422 if (vma_tmp
->vm_end
> addr
) {
1424 if (vma_tmp
->vm_start
<= addr
)
1426 rb_node
= rb_node
->rb_left
;
1428 rb_node
= rb_node
->rb_right
;
1431 mm
->mmap_cache
= vma
;
1437 EXPORT_SYMBOL(find_vma
);
1439 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1440 struct vm_area_struct
*
1441 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1442 struct vm_area_struct
**pprev
)
1444 struct vm_area_struct
*vma
= NULL
, *prev
= NULL
;
1445 struct rb_node
* rb_node
;
1449 /* Guard against addr being lower than the first VMA */
1452 /* Go through the RB tree quickly. */
1453 rb_node
= mm
->mm_rb
.rb_node
;
1456 struct vm_area_struct
*vma_tmp
;
1457 vma_tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1459 if (addr
< vma_tmp
->vm_end
) {
1460 rb_node
= rb_node
->rb_left
;
1463 if (!prev
->vm_next
|| (addr
< prev
->vm_next
->vm_end
))
1465 rb_node
= rb_node
->rb_right
;
1471 return prev
? prev
->vm_next
: vma
;
1475 * Verify that the stack growth is acceptable and
1476 * update accounting. This is shared with both the
1477 * grow-up and grow-down cases.
1479 static int acct_stack_growth(struct vm_area_struct
* vma
, unsigned long size
, unsigned long grow
)
1481 struct mm_struct
*mm
= vma
->vm_mm
;
1482 struct rlimit
*rlim
= current
->signal
->rlim
;
1484 /* address space limit tests */
1485 if (!may_expand_vm(mm
, grow
))
1488 /* Stack limit test */
1489 if (size
> rlim
[RLIMIT_STACK
].rlim_cur
)
1492 /* mlock limit tests */
1493 if (vma
->vm_flags
& VM_LOCKED
) {
1494 unsigned long locked
;
1495 unsigned long limit
;
1496 locked
= mm
->locked_vm
+ grow
;
1497 limit
= rlim
[RLIMIT_MEMLOCK
].rlim_cur
>> PAGE_SHIFT
;
1498 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1503 * Overcommit.. This must be the final test, as it will
1504 * update security statistics.
1506 if (security_vm_enough_memory(grow
))
1509 /* Ok, everything looks good - let it rip */
1510 mm
->total_vm
+= grow
;
1511 if (vma
->vm_flags
& VM_LOCKED
)
1512 mm
->locked_vm
+= grow
;
1513 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1517 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1519 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1520 * vma is the last one with address > vma->vm_end. Have to extend vma.
1525 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1529 if (!(vma
->vm_flags
& VM_GROWSUP
))
1533 * We must make sure the anon_vma is allocated
1534 * so that the anon_vma locking is not a noop.
1536 if (unlikely(anon_vma_prepare(vma
)))
1541 * vma->vm_start/vm_end cannot change under us because the caller
1542 * is required to hold the mmap_sem in read mode. We need the
1543 * anon_vma lock to serialize against concurrent expand_stacks.
1545 address
+= 4 + PAGE_SIZE
- 1;
1546 address
&= PAGE_MASK
;
1549 /* Somebody else might have raced and expanded it already */
1550 if (address
> vma
->vm_end
) {
1551 unsigned long size
, grow
;
1553 size
= address
- vma
->vm_start
;
1554 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1556 error
= acct_stack_growth(vma
, size
, grow
);
1558 vma
->vm_end
= address
;
1560 anon_vma_unlock(vma
);
1563 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1565 #ifdef CONFIG_STACK_GROWSUP
1566 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1568 return expand_upwards(vma
, address
);
1571 struct vm_area_struct
*
1572 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1574 struct vm_area_struct
*vma
, *prev
;
1577 vma
= find_vma_prev(mm
, addr
, &prev
);
1578 if (vma
&& (vma
->vm_start
<= addr
))
1580 if (!prev
|| expand_stack(prev
, addr
))
1582 if (prev
->vm_flags
& VM_LOCKED
) {
1583 make_pages_present(addr
, prev
->vm_end
);
1589 * vma is the first one with address < vma->vm_start. Have to extend vma.
1591 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1596 * We must make sure the anon_vma is allocated
1597 * so that the anon_vma locking is not a noop.
1599 if (unlikely(anon_vma_prepare(vma
)))
1604 * vma->vm_start/vm_end cannot change under us because the caller
1605 * is required to hold the mmap_sem in read mode. We need the
1606 * anon_vma lock to serialize against concurrent expand_stacks.
1608 address
&= PAGE_MASK
;
1611 /* Somebody else might have raced and expanded it already */
1612 if (address
< vma
->vm_start
) {
1613 unsigned long size
, grow
;
1615 size
= vma
->vm_end
- address
;
1616 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1618 error
= acct_stack_growth(vma
, size
, grow
);
1620 vma
->vm_start
= address
;
1621 vma
->vm_pgoff
-= grow
;
1624 anon_vma_unlock(vma
);
1628 struct vm_area_struct
*
1629 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1631 struct vm_area_struct
* vma
;
1632 unsigned long start
;
1635 vma
= find_vma(mm
,addr
);
1638 if (vma
->vm_start
<= addr
)
1640 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1642 start
= vma
->vm_start
;
1643 if (expand_stack(vma
, addr
))
1645 if (vma
->vm_flags
& VM_LOCKED
) {
1646 make_pages_present(addr
, start
);
1653 * Ok - we have the memory areas we should free on the vma list,
1654 * so release them, and do the vma updates.
1656 * Called with the mm semaphore held.
1658 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1660 /* Update high watermark before we lower total_vm */
1661 update_hiwater_vm(mm
);
1663 long nrpages
= vma_pages(vma
);
1665 mm
->total_vm
-= nrpages
;
1666 if (vma
->vm_flags
& VM_LOCKED
)
1667 mm
->locked_vm
-= nrpages
;
1668 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1669 vma
= remove_vma(vma
);
1675 * Get rid of page table information in the indicated region.
1677 * Called with the mm semaphore held.
1679 static void unmap_region(struct mm_struct
*mm
,
1680 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1681 unsigned long start
, unsigned long end
)
1683 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1684 struct mmu_gather
*tlb
;
1685 unsigned long nr_accounted
= 0;
1688 tlb
= tlb_gather_mmu(mm
, 0);
1689 update_hiwater_rss(mm
);
1690 unmap_vmas(&tlb
, vma
, start
, end
, &nr_accounted
, NULL
);
1691 vm_unacct_memory(nr_accounted
);
1692 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1693 next
? next
->vm_start
: 0);
1694 tlb_finish_mmu(tlb
, start
, end
);
1698 * Create a list of vma's touched by the unmap, removing them from the mm's
1699 * vma list as we go..
1702 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1703 struct vm_area_struct
*prev
, unsigned long end
)
1705 struct vm_area_struct
**insertion_point
;
1706 struct vm_area_struct
*tail_vma
= NULL
;
1709 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1711 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1715 } while (vma
&& vma
->vm_start
< end
);
1716 *insertion_point
= vma
;
1717 tail_vma
->vm_next
= NULL
;
1718 if (mm
->unmap_area
== arch_unmap_area
)
1719 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1721 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1722 mm
->unmap_area(mm
, addr
);
1723 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1727 * Split a vma into two pieces at address 'addr', a new vma is allocated
1728 * either for the first part or the the tail.
1730 int split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1731 unsigned long addr
, int new_below
)
1733 struct mempolicy
*pol
;
1734 struct vm_area_struct
*new;
1736 if (is_vm_hugetlb_page(vma
) && (addr
& ~HPAGE_MASK
))
1739 if (mm
->map_count
>= sysctl_max_map_count
)
1742 new = kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
1746 /* most fields are the same, copy all, and then fixup */
1752 new->vm_start
= addr
;
1753 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
1756 pol
= mpol_copy(vma_policy(vma
));
1758 kmem_cache_free(vm_area_cachep
, new);
1759 return PTR_ERR(pol
);
1761 vma_set_policy(new, pol
);
1764 get_file(new->vm_file
);
1766 if (new->vm_ops
&& new->vm_ops
->open
)
1767 new->vm_ops
->open(new);
1770 vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
1771 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
1773 vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
1778 /* Munmap is split into 2 main parts -- this part which finds
1779 * what needs doing, and the areas themselves, which do the
1780 * work. This now handles partial unmappings.
1781 * Jeremy Fitzhardinge <jeremy@goop.org>
1783 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1786 struct vm_area_struct
*vma
, *prev
, *last
;
1788 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
1791 if ((len
= PAGE_ALIGN(len
)) == 0)
1794 /* Find the first overlapping VMA */
1795 vma
= find_vma_prev(mm
, start
, &prev
);
1798 /* we have start < vma->vm_end */
1800 /* if it doesn't overlap, we have nothing.. */
1802 if (vma
->vm_start
>= end
)
1806 * If we need to split any vma, do it now to save pain later.
1808 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1809 * unmapped vm_area_struct will remain in use: so lower split_vma
1810 * places tmp vma above, and higher split_vma places tmp vma below.
1812 if (start
> vma
->vm_start
) {
1813 int error
= split_vma(mm
, vma
, start
, 0);
1819 /* Does it split the last one? */
1820 last
= find_vma(mm
, end
);
1821 if (last
&& end
> last
->vm_start
) {
1822 int error
= split_vma(mm
, last
, end
, 1);
1826 vma
= prev
? prev
->vm_next
: mm
->mmap
;
1829 * Remove the vma's, and unmap the actual pages
1831 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
1832 unmap_region(mm
, vma
, prev
, start
, end
);
1834 /* Fix up all other VM information */
1835 remove_vma_list(mm
, vma
);
1840 EXPORT_SYMBOL(do_munmap
);
1842 asmlinkage
long sys_munmap(unsigned long addr
, size_t len
)
1845 struct mm_struct
*mm
= current
->mm
;
1847 profile_munmap(addr
);
1849 down_write(&mm
->mmap_sem
);
1850 ret
= do_munmap(mm
, addr
, len
);
1851 up_write(&mm
->mmap_sem
);
1855 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
1857 #ifdef CONFIG_DEBUG_VM
1858 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
1860 up_read(&mm
->mmap_sem
);
1866 * this is really a simplified "do_mmap". it only handles
1867 * anonymous maps. eventually we may be able to do some
1868 * brk-specific accounting here.
1870 unsigned long do_brk(unsigned long addr
, unsigned long len
)
1872 struct mm_struct
* mm
= current
->mm
;
1873 struct vm_area_struct
* vma
, * prev
;
1874 unsigned long flags
;
1875 struct rb_node
** rb_link
, * rb_parent
;
1876 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
1879 len
= PAGE_ALIGN(len
);
1883 if ((addr
+ len
) > TASK_SIZE
|| (addr
+ len
) < addr
)
1886 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
1888 error
= arch_mmap_check(addr
, len
, flags
);
1895 if (mm
->def_flags
& VM_LOCKED
) {
1896 unsigned long locked
, lock_limit
;
1897 locked
= len
>> PAGE_SHIFT
;
1898 locked
+= mm
->locked_vm
;
1899 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
1900 lock_limit
>>= PAGE_SHIFT
;
1901 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1906 * mm->mmap_sem is required to protect against another thread
1907 * changing the mappings in case we sleep.
1909 verify_mm_writelocked(mm
);
1912 * Clear old maps. this also does some error checking for us
1915 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1916 if (vma
&& vma
->vm_start
< addr
+ len
) {
1917 if (do_munmap(mm
, addr
, len
))
1922 /* Check against address space limits *after* clearing old maps... */
1923 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1926 if (mm
->map_count
> sysctl_max_map_count
)
1929 if (security_vm_enough_memory(len
>> PAGE_SHIFT
))
1932 /* Can we just expand an old private anonymous mapping? */
1933 if (vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
1934 NULL
, NULL
, pgoff
, NULL
))
1938 * create a vma struct for an anonymous mapping
1940 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1942 vm_unacct_memory(len
>> PAGE_SHIFT
);
1947 vma
->vm_start
= addr
;
1948 vma
->vm_end
= addr
+ len
;
1949 vma
->vm_pgoff
= pgoff
;
1950 vma
->vm_flags
= flags
;
1951 vma
->vm_page_prot
= protection_map
[flags
&
1952 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)];
1953 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1955 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1956 if (flags
& VM_LOCKED
) {
1957 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
1958 make_pages_present(addr
, addr
+ len
);
1963 EXPORT_SYMBOL(do_brk
);
1965 /* Release all mmaps. */
1966 void exit_mmap(struct mm_struct
*mm
)
1968 struct mmu_gather
*tlb
;
1969 struct vm_area_struct
*vma
= mm
->mmap
;
1970 unsigned long nr_accounted
= 0;
1975 tlb
= tlb_gather_mmu(mm
, 1);
1976 /* Don't update_hiwater_rss(mm) here, do_exit already did */
1977 /* Use -1 here to ensure all VMAs in the mm are unmapped */
1978 end
= unmap_vmas(&tlb
, vma
, 0, -1, &nr_accounted
, NULL
);
1979 vm_unacct_memory(nr_accounted
);
1980 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, 0);
1981 tlb_finish_mmu(tlb
, 0, end
);
1984 * Walk the list again, actually closing and freeing it,
1985 * with preemption enabled, without holding any MM locks.
1988 vma
= remove_vma(vma
);
1990 BUG_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
1993 /* Insert vm structure into process list sorted by address
1994 * and into the inode's i_mmap tree. If vm_file is non-NULL
1995 * then i_mmap_lock is taken here.
1997 int insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
1999 struct vm_area_struct
* __vma
, * prev
;
2000 struct rb_node
** rb_link
, * rb_parent
;
2003 * The vm_pgoff of a purely anonymous vma should be irrelevant
2004 * until its first write fault, when page's anon_vma and index
2005 * are set. But now set the vm_pgoff it will almost certainly
2006 * end up with (unless mremap moves it elsewhere before that
2007 * first wfault), so /proc/pid/maps tells a consistent story.
2009 * By setting it to reflect the virtual start address of the
2010 * vma, merges and splits can happen in a seamless way, just
2011 * using the existing file pgoff checks and manipulations.
2012 * Similarly in do_mmap_pgoff and in do_brk.
2014 if (!vma
->vm_file
) {
2015 BUG_ON(vma
->anon_vma
);
2016 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2018 __vma
= find_vma_prepare(mm
,vma
->vm_start
,&prev
,&rb_link
,&rb_parent
);
2019 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
2021 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2022 security_vm_enough_memory(vma_pages(vma
)))
2024 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2029 * Copy the vma structure to a new location in the same mm,
2030 * prior to moving page table entries, to effect an mremap move.
2032 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2033 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
2035 struct vm_area_struct
*vma
= *vmap
;
2036 unsigned long vma_start
= vma
->vm_start
;
2037 struct mm_struct
*mm
= vma
->vm_mm
;
2038 struct vm_area_struct
*new_vma
, *prev
;
2039 struct rb_node
**rb_link
, *rb_parent
;
2040 struct mempolicy
*pol
;
2043 * If anonymous vma has not yet been faulted, update new pgoff
2044 * to match new location, to increase its chance of merging.
2046 if (!vma
->vm_file
&& !vma
->anon_vma
)
2047 pgoff
= addr
>> PAGE_SHIFT
;
2049 find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2050 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2051 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2054 * Source vma may have been merged into new_vma
2056 if (vma_start
>= new_vma
->vm_start
&&
2057 vma_start
< new_vma
->vm_end
)
2060 new_vma
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
2063 pol
= mpol_copy(vma_policy(vma
));
2065 kmem_cache_free(vm_area_cachep
, new_vma
);
2068 vma_set_policy(new_vma
, pol
);
2069 new_vma
->vm_start
= addr
;
2070 new_vma
->vm_end
= addr
+ len
;
2071 new_vma
->vm_pgoff
= pgoff
;
2072 if (new_vma
->vm_file
)
2073 get_file(new_vma
->vm_file
);
2074 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2075 new_vma
->vm_ops
->open(new_vma
);
2076 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2083 * Return true if the calling process may expand its vm space by the passed
2086 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2088 unsigned long cur
= mm
->total_vm
; /* pages */
2091 lim
= current
->signal
->rlim
[RLIMIT_AS
].rlim_cur
>> PAGE_SHIFT
;
2093 if (cur
+ npages
> lim
)