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
)
225 * Check against rlimit here. If this check is done later after the test
226 * of oldbrk with newbrk then it can escape the test and let the data
227 * segment grow beyond its set limit the in case where the limit is
228 * not page aligned -Ram Gupta
230 rlim
= current
->signal
->rlim
[RLIMIT_DATA
].rlim_cur
;
231 if (rlim
< RLIM_INFINITY
&& brk
- mm
->start_data
> rlim
)
234 newbrk
= PAGE_ALIGN(brk
);
235 oldbrk
= PAGE_ALIGN(mm
->brk
);
236 if (oldbrk
== newbrk
)
239 /* Always allow shrinking brk. */
240 if (brk
<= mm
->brk
) {
241 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
246 /* Check against existing mmap mappings. */
247 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
250 /* Ok, looks good - let it rip. */
251 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
257 up_write(&mm
->mmap_sem
);
262 static int browse_rb(struct rb_root
*root
)
265 struct rb_node
*nd
, *pn
= NULL
;
266 unsigned long prev
= 0, pend
= 0;
268 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
269 struct vm_area_struct
*vma
;
270 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
271 if (vma
->vm_start
< prev
)
272 printk("vm_start %lx prev %lx\n", vma
->vm_start
, prev
), i
= -1;
273 if (vma
->vm_start
< pend
)
274 printk("vm_start %lx pend %lx\n", vma
->vm_start
, pend
);
275 if (vma
->vm_start
> vma
->vm_end
)
276 printk("vm_end %lx < vm_start %lx\n", vma
->vm_end
, vma
->vm_start
);
281 for (nd
= pn
; nd
; nd
= rb_prev(nd
)) {
285 printk("backwards %d, forwards %d\n", j
, i
), i
= 0;
289 void validate_mm(struct mm_struct
*mm
)
293 struct vm_area_struct
*tmp
= mm
->mmap
;
298 if (i
!= mm
->map_count
)
299 printk("map_count %d vm_next %d\n", mm
->map_count
, i
), bug
= 1;
300 i
= browse_rb(&mm
->mm_rb
);
301 if (i
!= mm
->map_count
)
302 printk("map_count %d rb %d\n", mm
->map_count
, i
), bug
= 1;
306 #define validate_mm(mm) do { } while (0)
309 static struct vm_area_struct
*
310 find_vma_prepare(struct mm_struct
*mm
, unsigned long addr
,
311 struct vm_area_struct
**pprev
, struct rb_node
***rb_link
,
312 struct rb_node
** rb_parent
)
314 struct vm_area_struct
* vma
;
315 struct rb_node
** __rb_link
, * __rb_parent
, * rb_prev
;
317 __rb_link
= &mm
->mm_rb
.rb_node
;
318 rb_prev
= __rb_parent
= NULL
;
322 struct vm_area_struct
*vma_tmp
;
324 __rb_parent
= *__rb_link
;
325 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
327 if (vma_tmp
->vm_end
> addr
) {
329 if (vma_tmp
->vm_start
<= addr
)
331 __rb_link
= &__rb_parent
->rb_left
;
333 rb_prev
= __rb_parent
;
334 __rb_link
= &__rb_parent
->rb_right
;
340 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
341 *rb_link
= __rb_link
;
342 *rb_parent
= __rb_parent
;
347 __vma_link_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
348 struct vm_area_struct
*prev
, struct rb_node
*rb_parent
)
351 vma
->vm_next
= prev
->vm_next
;
356 vma
->vm_next
= rb_entry(rb_parent
,
357 struct vm_area_struct
, vm_rb
);
363 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
364 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
366 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
367 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
370 static inline void __vma_link_file(struct vm_area_struct
*vma
)
376 struct address_space
*mapping
= file
->f_mapping
;
378 if (vma
->vm_flags
& VM_DENYWRITE
)
379 atomic_dec(&file
->f_dentry
->d_inode
->i_writecount
);
380 if (vma
->vm_flags
& VM_SHARED
)
381 mapping
->i_mmap_writable
++;
383 flush_dcache_mmap_lock(mapping
);
384 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
385 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
387 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
388 flush_dcache_mmap_unlock(mapping
);
393 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
394 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
395 struct rb_node
*rb_parent
)
397 __vma_link_list(mm
, vma
, prev
, rb_parent
);
398 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
399 __anon_vma_link(vma
);
402 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
403 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
404 struct rb_node
*rb_parent
)
406 struct address_space
*mapping
= NULL
;
409 mapping
= vma
->vm_file
->f_mapping
;
412 spin_lock(&mapping
->i_mmap_lock
);
413 vma
->vm_truncate_count
= mapping
->truncate_count
;
417 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
418 __vma_link_file(vma
);
420 anon_vma_unlock(vma
);
422 spin_unlock(&mapping
->i_mmap_lock
);
429 * Helper for vma_adjust in the split_vma insert case:
430 * insert vm structure into list and rbtree and anon_vma,
431 * but it has already been inserted into prio_tree earlier.
434 __insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
436 struct vm_area_struct
* __vma
, * prev
;
437 struct rb_node
** rb_link
, * rb_parent
;
439 __vma
= find_vma_prepare(mm
, vma
->vm_start
,&prev
, &rb_link
, &rb_parent
);
440 BUG_ON(__vma
&& __vma
->vm_start
< vma
->vm_end
);
441 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
446 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
447 struct vm_area_struct
*prev
)
449 prev
->vm_next
= vma
->vm_next
;
450 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
451 if (mm
->mmap_cache
== vma
)
452 mm
->mmap_cache
= prev
;
456 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
457 * is already present in an i_mmap tree without adjusting the tree.
458 * The following helper function should be used when such adjustments
459 * are necessary. The "insert" vma (if any) is to be inserted
460 * before we drop the necessary locks.
462 void vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
463 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
465 struct mm_struct
*mm
= vma
->vm_mm
;
466 struct vm_area_struct
*next
= vma
->vm_next
;
467 struct vm_area_struct
*importer
= NULL
;
468 struct address_space
*mapping
= NULL
;
469 struct prio_tree_root
*root
= NULL
;
470 struct file
*file
= vma
->vm_file
;
471 struct anon_vma
*anon_vma
= NULL
;
472 long adjust_next
= 0;
475 if (next
&& !insert
) {
476 if (end
>= next
->vm_end
) {
478 * vma expands, overlapping all the next, and
479 * perhaps the one after too (mprotect case 6).
481 again
: remove_next
= 1 + (end
> next
->vm_end
);
483 anon_vma
= next
->anon_vma
;
485 } else if (end
> next
->vm_start
) {
487 * vma expands, overlapping part of the next:
488 * mprotect case 5 shifting the boundary up.
490 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
491 anon_vma
= next
->anon_vma
;
493 } else if (end
< vma
->vm_end
) {
495 * vma shrinks, and !insert tells it's not
496 * split_vma inserting another: so it must be
497 * mprotect case 4 shifting the boundary down.
499 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
500 anon_vma
= next
->anon_vma
;
506 mapping
= file
->f_mapping
;
507 if (!(vma
->vm_flags
& VM_NONLINEAR
))
508 root
= &mapping
->i_mmap
;
509 spin_lock(&mapping
->i_mmap_lock
);
511 vma
->vm_truncate_count
!= next
->vm_truncate_count
) {
513 * unmap_mapping_range might be in progress:
514 * ensure that the expanding vma is rescanned.
516 importer
->vm_truncate_count
= 0;
519 insert
->vm_truncate_count
= vma
->vm_truncate_count
;
521 * Put into prio_tree now, so instantiated pages
522 * are visible to arm/parisc __flush_dcache_page
523 * throughout; but we cannot insert into address
524 * space until vma start or end is updated.
526 __vma_link_file(insert
);
531 * When changing only vma->vm_end, we don't really need
532 * anon_vma lock: but is that case worth optimizing out?
535 anon_vma
= vma
->anon_vma
;
537 spin_lock(&anon_vma
->lock
);
539 * Easily overlooked: when mprotect shifts the boundary,
540 * make sure the expanding vma has anon_vma set if the
541 * shrinking vma had, to cover any anon pages imported.
543 if (importer
&& !importer
->anon_vma
) {
544 importer
->anon_vma
= anon_vma
;
545 __anon_vma_link(importer
);
550 flush_dcache_mmap_lock(mapping
);
551 vma_prio_tree_remove(vma
, root
);
553 vma_prio_tree_remove(next
, root
);
556 vma
->vm_start
= start
;
558 vma
->vm_pgoff
= pgoff
;
560 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
561 next
->vm_pgoff
+= adjust_next
;
566 vma_prio_tree_insert(next
, root
);
567 vma_prio_tree_insert(vma
, root
);
568 flush_dcache_mmap_unlock(mapping
);
573 * vma_merge has merged next into vma, and needs
574 * us to remove next before dropping the locks.
576 __vma_unlink(mm
, next
, vma
);
578 __remove_shared_vm_struct(next
, file
, mapping
);
580 __anon_vma_merge(vma
, next
);
583 * split_vma has split insert from vma, and needs
584 * us to insert it before dropping the locks
585 * (it may either follow vma or precede it).
587 __insert_vm_struct(mm
, insert
);
591 spin_unlock(&anon_vma
->lock
);
593 spin_unlock(&mapping
->i_mmap_lock
);
599 mpol_free(vma_policy(next
));
600 kmem_cache_free(vm_area_cachep
, next
);
602 * In mprotect's case 6 (see comments on vma_merge),
603 * we must remove another next too. It would clutter
604 * up the code too much to do both in one go.
606 if (remove_next
== 2) {
616 * If the vma has a ->close operation then the driver probably needs to release
617 * per-vma resources, so we don't attempt to merge those.
619 #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP)
621 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
622 struct file
*file
, unsigned long vm_flags
)
624 if (vma
->vm_flags
!= vm_flags
)
626 if (vma
->vm_file
!= file
)
628 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
633 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
634 struct anon_vma
*anon_vma2
)
636 return !anon_vma1
|| !anon_vma2
|| (anon_vma1
== anon_vma2
);
640 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
641 * in front of (at a lower virtual address and file offset than) the vma.
643 * We cannot merge two vmas if they have differently assigned (non-NULL)
644 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
646 * We don't check here for the merged mmap wrapping around the end of pagecache
647 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
648 * wrap, nor mmaps which cover the final page at index -1UL.
651 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
652 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
654 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
655 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
656 if (vma
->vm_pgoff
== vm_pgoff
)
663 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
664 * beyond (at a higher 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.
670 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
671 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
673 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
674 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
676 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
677 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
684 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
685 * whether that can be merged with its predecessor or its successor.
686 * Or both (it neatly fills a hole).
688 * In most cases - when called for mmap, brk or mremap - [addr,end) is
689 * certain not to be mapped by the time vma_merge is called; but when
690 * called for mprotect, it is certain to be already mapped (either at
691 * an offset within prev, or at the start of next), and the flags of
692 * this area are about to be changed to vm_flags - and the no-change
693 * case has already been eliminated.
695 * The following mprotect cases have to be considered, where AAAA is
696 * the area passed down from mprotect_fixup, never extending beyond one
697 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
699 * AAAA AAAA AAAA AAAA
700 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
701 * cannot merge might become might become might become
702 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
703 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
704 * mremap move: PPPPNNNNNNNN 8
706 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
707 * might become case 1 below case 2 below case 3 below
709 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
710 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
712 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
713 struct vm_area_struct
*prev
, unsigned long addr
,
714 unsigned long end
, unsigned long vm_flags
,
715 struct anon_vma
*anon_vma
, struct file
*file
,
716 pgoff_t pgoff
, struct mempolicy
*policy
)
718 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
719 struct vm_area_struct
*area
, *next
;
722 * We later require that vma->vm_flags == vm_flags,
723 * so this tests vma->vm_flags & VM_SPECIAL, too.
725 if (vm_flags
& VM_SPECIAL
)
729 next
= prev
->vm_next
;
733 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
734 next
= next
->vm_next
;
737 * Can it merge with the predecessor?
739 if (prev
&& prev
->vm_end
== addr
&&
740 mpol_equal(vma_policy(prev
), policy
) &&
741 can_vma_merge_after(prev
, vm_flags
,
742 anon_vma
, file
, pgoff
)) {
744 * OK, it can. Can we now merge in the successor as well?
746 if (next
&& end
== next
->vm_start
&&
747 mpol_equal(policy
, vma_policy(next
)) &&
748 can_vma_merge_before(next
, vm_flags
,
749 anon_vma
, file
, pgoff
+pglen
) &&
750 is_mergeable_anon_vma(prev
->anon_vma
,
753 vma_adjust(prev
, prev
->vm_start
,
754 next
->vm_end
, prev
->vm_pgoff
, NULL
);
755 } else /* cases 2, 5, 7 */
756 vma_adjust(prev
, prev
->vm_start
,
757 end
, prev
->vm_pgoff
, NULL
);
762 * Can this new request be merged in front of next?
764 if (next
&& end
== next
->vm_start
&&
765 mpol_equal(policy
, vma_policy(next
)) &&
766 can_vma_merge_before(next
, vm_flags
,
767 anon_vma
, file
, pgoff
+pglen
)) {
768 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
769 vma_adjust(prev
, prev
->vm_start
,
770 addr
, prev
->vm_pgoff
, NULL
);
771 else /* cases 3, 8 */
772 vma_adjust(area
, addr
, next
->vm_end
,
773 next
->vm_pgoff
- pglen
, NULL
);
781 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
782 * neighbouring vmas for a suitable anon_vma, before it goes off
783 * to allocate a new anon_vma. It checks because a repetitive
784 * sequence of mprotects and faults may otherwise lead to distinct
785 * anon_vmas being allocated, preventing vma merge in subsequent
788 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
790 struct vm_area_struct
*near
;
791 unsigned long vm_flags
;
798 * Since only mprotect tries to remerge vmas, match flags
799 * which might be mprotected into each other later on.
800 * Neither mlock nor madvise tries to remerge at present,
801 * so leave their flags as obstructing a merge.
803 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
804 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
806 if (near
->anon_vma
&& vma
->vm_end
== near
->vm_start
&&
807 mpol_equal(vma_policy(vma
), vma_policy(near
)) &&
808 can_vma_merge_before(near
, vm_flags
,
809 NULL
, vma
->vm_file
, vma
->vm_pgoff
+
810 ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
)))
811 return near
->anon_vma
;
814 * It is potentially slow to have to call find_vma_prev here.
815 * But it's only on the first write fault on the vma, not
816 * every time, and we could devise a way to avoid it later
817 * (e.g. stash info in next's anon_vma_node when assigning
818 * an anon_vma, or when trying vma_merge). Another time.
820 BUG_ON(find_vma_prev(vma
->vm_mm
, vma
->vm_start
, &near
) != vma
);
824 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
825 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
827 if (near
->anon_vma
&& near
->vm_end
== vma
->vm_start
&&
828 mpol_equal(vma_policy(near
), vma_policy(vma
)) &&
829 can_vma_merge_after(near
, vm_flags
,
830 NULL
, vma
->vm_file
, vma
->vm_pgoff
))
831 return near
->anon_vma
;
834 * There's no absolute need to look only at touching neighbours:
835 * we could search further afield for "compatible" anon_vmas.
836 * But it would probably just be a waste of time searching,
837 * or lead to too many vmas hanging off the same anon_vma.
838 * We're trying to allow mprotect remerging later on,
839 * not trying to minimize memory used for anon_vmas.
844 #ifdef CONFIG_PROC_FS
845 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
846 struct file
*file
, long pages
)
848 const unsigned long stack_flags
849 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
852 mm
->shared_vm
+= pages
;
853 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
854 mm
->exec_vm
+= pages
;
855 } else if (flags
& stack_flags
)
856 mm
->stack_vm
+= pages
;
857 if (flags
& (VM_RESERVED
|VM_IO
))
858 mm
->reserved_vm
+= pages
;
860 #endif /* CONFIG_PROC_FS */
863 * The caller must hold down_write(current->mm->mmap_sem).
866 unsigned long do_mmap_pgoff(struct file
* file
, unsigned long addr
,
867 unsigned long len
, unsigned long prot
,
868 unsigned long flags
, unsigned long pgoff
)
870 struct mm_struct
* mm
= current
->mm
;
871 struct vm_area_struct
* vma
, * prev
;
873 unsigned int vm_flags
;
874 int correct_wcount
= 0;
876 struct rb_node
** rb_link
, * rb_parent
;
878 unsigned long charged
= 0, reqprot
= prot
;
881 if (is_file_hugepages(file
))
884 if (!file
->f_op
|| !file
->f_op
->mmap
)
887 if ((prot
& PROT_EXEC
) &&
888 (file
->f_vfsmnt
->mnt_flags
& MNT_NOEXEC
))
892 * Does the application expect PROT_READ to imply PROT_EXEC?
894 * (the exception is when the underlying filesystem is noexec
895 * mounted, in which case we dont add PROT_EXEC.)
897 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
898 if (!(file
&& (file
->f_vfsmnt
->mnt_flags
& MNT_NOEXEC
)))
904 /* Careful about overflows.. */
905 len
= PAGE_ALIGN(len
);
906 if (!len
|| len
> TASK_SIZE
)
909 /* offset overflow? */
910 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
913 /* Too many mappings? */
914 if (mm
->map_count
> sysctl_max_map_count
)
917 /* Obtain the address to map to. we verify (or select) it and ensure
918 * that it represents a valid section of the address space.
920 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
921 if (addr
& ~PAGE_MASK
)
924 /* Do simple checking here so the lower-level routines won't have
925 * to. we assume access permissions have been handled by the open
926 * of the memory object, so we don't do any here.
928 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
929 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
931 if (flags
& MAP_LOCKED
) {
934 vm_flags
|= VM_LOCKED
;
936 /* mlock MCL_FUTURE? */
937 if (vm_flags
& VM_LOCKED
) {
938 unsigned long locked
, lock_limit
;
939 locked
= len
>> PAGE_SHIFT
;
940 locked
+= mm
->locked_vm
;
941 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
942 lock_limit
>>= PAGE_SHIFT
;
943 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
947 inode
= file
? file
->f_dentry
->d_inode
: NULL
;
950 switch (flags
& MAP_TYPE
) {
952 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
956 * Make sure we don't allow writing to an append-only
959 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
963 * Make sure there are no mandatory locks on the file.
965 if (locks_verify_locked(inode
))
968 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
969 if (!(file
->f_mode
& FMODE_WRITE
))
970 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
974 if (!(file
->f_mode
& FMODE_READ
))
982 switch (flags
& MAP_TYPE
) {
984 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
988 * Set pgoff according to addr for anon_vma.
990 pgoff
= addr
>> PAGE_SHIFT
;
997 error
= security_file_mmap(file
, reqprot
, prot
, flags
);
1001 /* Clear old maps */
1004 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1005 if (vma
&& vma
->vm_start
< addr
+ len
) {
1006 if (do_munmap(mm
, addr
, len
))
1011 /* Check against address space limit. */
1012 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1015 if (accountable
&& (!(flags
& MAP_NORESERVE
) ||
1016 sysctl_overcommit_memory
== OVERCOMMIT_NEVER
)) {
1017 if (vm_flags
& VM_SHARED
) {
1018 /* Check memory availability in shmem_file_setup? */
1019 vm_flags
|= VM_ACCOUNT
;
1020 } else if (vm_flags
& VM_WRITE
) {
1022 * Private writable mapping: check memory availability
1024 charged
= len
>> PAGE_SHIFT
;
1025 if (security_vm_enough_memory(charged
))
1027 vm_flags
|= VM_ACCOUNT
;
1032 * Can we just expand an old private anonymous mapping?
1033 * The VM_SHARED test is necessary because shmem_zero_setup
1034 * will create the file object for a shared anonymous map below.
1036 if (!file
&& !(vm_flags
& VM_SHARED
) &&
1037 vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
,
1038 NULL
, NULL
, pgoff
, NULL
))
1042 * Determine the object being mapped and call the appropriate
1043 * specific mapper. the address has already been validated, but
1044 * not unmapped, but the maps are removed from the list.
1046 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1053 vma
->vm_start
= addr
;
1054 vma
->vm_end
= addr
+ len
;
1055 vma
->vm_flags
= vm_flags
;
1056 vma
->vm_page_prot
= protection_map
[vm_flags
& 0x0f];
1057 vma
->vm_pgoff
= pgoff
;
1061 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1063 if (vm_flags
& VM_DENYWRITE
) {
1064 error
= deny_write_access(file
);
1069 vma
->vm_file
= file
;
1071 error
= file
->f_op
->mmap(file
, vma
);
1073 goto unmap_and_free_vma
;
1074 } else if (vm_flags
& VM_SHARED
) {
1075 error
= shmem_zero_setup(vma
);
1080 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1081 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1082 * that memory reservation must be checked; but that reservation
1083 * belongs to shared memory object, not to vma: so now clear it.
1085 if ((vm_flags
& (VM_SHARED
|VM_ACCOUNT
)) == (VM_SHARED
|VM_ACCOUNT
))
1086 vma
->vm_flags
&= ~VM_ACCOUNT
;
1088 /* Can addr have changed??
1090 * Answer: Yes, several device drivers can do it in their
1091 * f_op->mmap method. -DaveM
1093 addr
= vma
->vm_start
;
1094 pgoff
= vma
->vm_pgoff
;
1095 vm_flags
= vma
->vm_flags
;
1097 if (!file
|| !vma_merge(mm
, prev
, addr
, vma
->vm_end
,
1098 vma
->vm_flags
, NULL
, file
, pgoff
, vma_policy(vma
))) {
1099 file
= vma
->vm_file
;
1100 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1102 atomic_inc(&inode
->i_writecount
);
1106 atomic_inc(&inode
->i_writecount
);
1109 mpol_free(vma_policy(vma
));
1110 kmem_cache_free(vm_area_cachep
, vma
);
1113 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1114 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1115 if (vm_flags
& VM_LOCKED
) {
1116 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
1117 make_pages_present(addr
, addr
+ len
);
1119 if (flags
& MAP_POPULATE
) {
1120 up_write(&mm
->mmap_sem
);
1121 sys_remap_file_pages(addr
, len
, 0,
1122 pgoff
, flags
& MAP_NONBLOCK
);
1123 down_write(&mm
->mmap_sem
);
1129 atomic_inc(&inode
->i_writecount
);
1130 vma
->vm_file
= NULL
;
1133 /* Undo any partial mapping done by a device driver. */
1134 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1137 kmem_cache_free(vm_area_cachep
, vma
);
1140 vm_unacct_memory(charged
);
1144 EXPORT_SYMBOL(do_mmap_pgoff
);
1146 /* Get an address range which is currently unmapped.
1147 * For shmat() with addr=0.
1149 * Ugly calling convention alert:
1150 * Return value with the low bits set means error value,
1152 * if (ret & ~PAGE_MASK)
1155 * This function "knows" that -ENOMEM has the bits set.
1157 #ifndef HAVE_ARCH_UNMAPPED_AREA
1159 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1160 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1162 struct mm_struct
*mm
= current
->mm
;
1163 struct vm_area_struct
*vma
;
1164 unsigned long start_addr
;
1166 if (len
> TASK_SIZE
)
1170 addr
= PAGE_ALIGN(addr
);
1171 vma
= find_vma(mm
, addr
);
1172 if (TASK_SIZE
- len
>= addr
&&
1173 (!vma
|| addr
+ len
<= vma
->vm_start
))
1176 if (len
> mm
->cached_hole_size
) {
1177 start_addr
= addr
= mm
->free_area_cache
;
1179 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1180 mm
->cached_hole_size
= 0;
1184 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1185 /* At this point: (!vma || addr < vma->vm_end). */
1186 if (TASK_SIZE
- len
< addr
) {
1188 * Start a new search - just in case we missed
1191 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1192 addr
= TASK_UNMAPPED_BASE
;
1194 mm
->cached_hole_size
= 0;
1199 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1201 * Remember the place where we stopped the search:
1203 mm
->free_area_cache
= addr
+ len
;
1206 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1207 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1213 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1216 * Is this a new hole at the lowest possible address?
1218 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
) {
1219 mm
->free_area_cache
= addr
;
1220 mm
->cached_hole_size
= ~0UL;
1225 * This mmap-allocator allocates new areas top-down from below the
1226 * stack's low limit (the base):
1228 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1230 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1231 const unsigned long len
, const unsigned long pgoff
,
1232 const unsigned long flags
)
1234 struct vm_area_struct
*vma
;
1235 struct mm_struct
*mm
= current
->mm
;
1236 unsigned long addr
= addr0
;
1238 /* requested length too big for entire address space */
1239 if (len
> TASK_SIZE
)
1242 /* requesting a specific address */
1244 addr
= PAGE_ALIGN(addr
);
1245 vma
= find_vma(mm
, addr
);
1246 if (TASK_SIZE
- len
>= addr
&&
1247 (!vma
|| addr
+ len
<= vma
->vm_start
))
1251 /* check if free_area_cache is useful for us */
1252 if (len
<= mm
->cached_hole_size
) {
1253 mm
->cached_hole_size
= 0;
1254 mm
->free_area_cache
= mm
->mmap_base
;
1257 /* either no address requested or can't fit in requested address hole */
1258 addr
= mm
->free_area_cache
;
1260 /* make sure it can fit in the remaining address space */
1262 vma
= find_vma(mm
, addr
-len
);
1263 if (!vma
|| addr
<= vma
->vm_start
)
1264 /* remember the address as a hint for next time */
1265 return (mm
->free_area_cache
= addr
-len
);
1268 if (mm
->mmap_base
< len
)
1271 addr
= mm
->mmap_base
-len
;
1275 * Lookup failure means no vma is above this address,
1276 * else if new region fits below vma->vm_start,
1277 * return with success:
1279 vma
= find_vma(mm
, addr
);
1280 if (!vma
|| addr
+len
<= vma
->vm_start
)
1281 /* remember the address as a hint for next time */
1282 return (mm
->free_area_cache
= addr
);
1284 /* remember the largest hole we saw so far */
1285 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1286 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1288 /* try just below the current vma->vm_start */
1289 addr
= vma
->vm_start
-len
;
1290 } while (len
< vma
->vm_start
);
1294 * A failed mmap() very likely causes application failure,
1295 * so fall back to the bottom-up function here. This scenario
1296 * can happen with large stack limits and large mmap()
1299 mm
->cached_hole_size
= ~0UL;
1300 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1301 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1303 * Restore the topdown base:
1305 mm
->free_area_cache
= mm
->mmap_base
;
1306 mm
->cached_hole_size
= ~0UL;
1312 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1315 * Is this a new hole at the highest possible address?
1317 if (addr
> mm
->free_area_cache
)
1318 mm
->free_area_cache
= addr
;
1320 /* dont allow allocations above current base */
1321 if (mm
->free_area_cache
> mm
->mmap_base
)
1322 mm
->free_area_cache
= mm
->mmap_base
;
1326 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1327 unsigned long pgoff
, unsigned long flags
)
1331 if (!(flags
& MAP_FIXED
)) {
1332 unsigned long (*get_area
)(struct file
*, unsigned long, unsigned long, unsigned long, unsigned long);
1334 get_area
= current
->mm
->get_unmapped_area
;
1335 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1336 get_area
= file
->f_op
->get_unmapped_area
;
1337 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1338 if (IS_ERR_VALUE(addr
))
1342 if (addr
> TASK_SIZE
- len
)
1344 if (addr
& ~PAGE_MASK
)
1346 if (file
&& is_file_hugepages(file
)) {
1348 * Check if the given range is hugepage aligned, and
1349 * can be made suitable for hugepages.
1351 ret
= prepare_hugepage_range(addr
, len
);
1354 * Ensure that a normal request is not falling in a
1355 * reserved hugepage range. For some archs like IA-64,
1356 * there is a separate region for hugepages.
1358 ret
= is_hugepage_only_range(current
->mm
, addr
, len
);
1365 EXPORT_SYMBOL(get_unmapped_area
);
1367 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1368 struct vm_area_struct
* find_vma(struct mm_struct
* mm
, unsigned long addr
)
1370 struct vm_area_struct
*vma
= NULL
;
1373 /* Check the cache first. */
1374 /* (Cache hit rate is typically around 35%.) */
1375 vma
= mm
->mmap_cache
;
1376 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1377 struct rb_node
* rb_node
;
1379 rb_node
= mm
->mm_rb
.rb_node
;
1383 struct vm_area_struct
* vma_tmp
;
1385 vma_tmp
= rb_entry(rb_node
,
1386 struct vm_area_struct
, vm_rb
);
1388 if (vma_tmp
->vm_end
> addr
) {
1390 if (vma_tmp
->vm_start
<= addr
)
1392 rb_node
= rb_node
->rb_left
;
1394 rb_node
= rb_node
->rb_right
;
1397 mm
->mmap_cache
= vma
;
1403 EXPORT_SYMBOL(find_vma
);
1405 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1406 struct vm_area_struct
*
1407 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1408 struct vm_area_struct
**pprev
)
1410 struct vm_area_struct
*vma
= NULL
, *prev
= NULL
;
1411 struct rb_node
* rb_node
;
1415 /* Guard against addr being lower than the first VMA */
1418 /* Go through the RB tree quickly. */
1419 rb_node
= mm
->mm_rb
.rb_node
;
1422 struct vm_area_struct
*vma_tmp
;
1423 vma_tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1425 if (addr
< vma_tmp
->vm_end
) {
1426 rb_node
= rb_node
->rb_left
;
1429 if (!prev
->vm_next
|| (addr
< prev
->vm_next
->vm_end
))
1431 rb_node
= rb_node
->rb_right
;
1437 return prev
? prev
->vm_next
: vma
;
1441 * Verify that the stack growth is acceptable and
1442 * update accounting. This is shared with both the
1443 * grow-up and grow-down cases.
1445 static int acct_stack_growth(struct vm_area_struct
* vma
, unsigned long size
, unsigned long grow
)
1447 struct mm_struct
*mm
= vma
->vm_mm
;
1448 struct rlimit
*rlim
= current
->signal
->rlim
;
1450 /* address space limit tests */
1451 if (!may_expand_vm(mm
, grow
))
1454 /* Stack limit test */
1455 if (size
> rlim
[RLIMIT_STACK
].rlim_cur
)
1458 /* mlock limit tests */
1459 if (vma
->vm_flags
& VM_LOCKED
) {
1460 unsigned long locked
;
1461 unsigned long limit
;
1462 locked
= mm
->locked_vm
+ grow
;
1463 limit
= rlim
[RLIMIT_MEMLOCK
].rlim_cur
>> PAGE_SHIFT
;
1464 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1469 * Overcommit.. This must be the final test, as it will
1470 * update security statistics.
1472 if (security_vm_enough_memory(grow
))
1475 /* Ok, everything looks good - let it rip */
1476 mm
->total_vm
+= grow
;
1477 if (vma
->vm_flags
& VM_LOCKED
)
1478 mm
->locked_vm
+= grow
;
1479 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1483 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1485 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1486 * vma is the last one with address > vma->vm_end. Have to extend vma.
1491 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1495 if (!(vma
->vm_flags
& VM_GROWSUP
))
1499 * We must make sure the anon_vma is allocated
1500 * so that the anon_vma locking is not a noop.
1502 if (unlikely(anon_vma_prepare(vma
)))
1507 * vma->vm_start/vm_end cannot change under us because the caller
1508 * is required to hold the mmap_sem in read mode. We need the
1509 * anon_vma lock to serialize against concurrent expand_stacks.
1511 address
+= 4 + PAGE_SIZE
- 1;
1512 address
&= PAGE_MASK
;
1515 /* Somebody else might have raced and expanded it already */
1516 if (address
> vma
->vm_end
) {
1517 unsigned long size
, grow
;
1519 size
= address
- vma
->vm_start
;
1520 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1522 error
= acct_stack_growth(vma
, size
, grow
);
1524 vma
->vm_end
= address
;
1526 anon_vma_unlock(vma
);
1529 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1531 #ifdef CONFIG_STACK_GROWSUP
1532 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1534 return expand_upwards(vma
, address
);
1537 struct vm_area_struct
*
1538 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1540 struct vm_area_struct
*vma
, *prev
;
1543 vma
= find_vma_prev(mm
, addr
, &prev
);
1544 if (vma
&& (vma
->vm_start
<= addr
))
1546 if (!prev
|| expand_stack(prev
, addr
))
1548 if (prev
->vm_flags
& VM_LOCKED
) {
1549 make_pages_present(addr
, prev
->vm_end
);
1555 * vma is the first one with address < vma->vm_start. Have to extend vma.
1557 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1562 * We must make sure the anon_vma is allocated
1563 * so that the anon_vma locking is not a noop.
1565 if (unlikely(anon_vma_prepare(vma
)))
1570 * vma->vm_start/vm_end cannot change under us because the caller
1571 * is required to hold the mmap_sem in read mode. We need the
1572 * anon_vma lock to serialize against concurrent expand_stacks.
1574 address
&= PAGE_MASK
;
1577 /* Somebody else might have raced and expanded it already */
1578 if (address
< vma
->vm_start
) {
1579 unsigned long size
, grow
;
1581 size
= vma
->vm_end
- address
;
1582 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1584 error
= acct_stack_growth(vma
, size
, grow
);
1586 vma
->vm_start
= address
;
1587 vma
->vm_pgoff
-= grow
;
1590 anon_vma_unlock(vma
);
1594 struct vm_area_struct
*
1595 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1597 struct vm_area_struct
* vma
;
1598 unsigned long start
;
1601 vma
= find_vma(mm
,addr
);
1604 if (vma
->vm_start
<= addr
)
1606 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1608 start
= vma
->vm_start
;
1609 if (expand_stack(vma
, addr
))
1611 if (vma
->vm_flags
& VM_LOCKED
) {
1612 make_pages_present(addr
, start
);
1619 * Ok - we have the memory areas we should free on the vma list,
1620 * so release them, and do the vma updates.
1622 * Called with the mm semaphore held.
1624 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1626 /* Update high watermark before we lower total_vm */
1627 update_hiwater_vm(mm
);
1629 long nrpages
= vma_pages(vma
);
1631 mm
->total_vm
-= nrpages
;
1632 if (vma
->vm_flags
& VM_LOCKED
)
1633 mm
->locked_vm
-= nrpages
;
1634 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1635 vma
= remove_vma(vma
);
1641 * Get rid of page table information in the indicated region.
1643 * Called with the mm semaphore held.
1645 static void unmap_region(struct mm_struct
*mm
,
1646 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1647 unsigned long start
, unsigned long end
)
1649 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1650 struct mmu_gather
*tlb
;
1651 unsigned long nr_accounted
= 0;
1654 tlb
= tlb_gather_mmu(mm
, 0);
1655 update_hiwater_rss(mm
);
1656 unmap_vmas(&tlb
, vma
, start
, end
, &nr_accounted
, NULL
);
1657 vm_unacct_memory(nr_accounted
);
1658 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1659 next
? next
->vm_start
: 0);
1660 tlb_finish_mmu(tlb
, start
, end
);
1664 * Create a list of vma's touched by the unmap, removing them from the mm's
1665 * vma list as we go..
1668 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1669 struct vm_area_struct
*prev
, unsigned long end
)
1671 struct vm_area_struct
**insertion_point
;
1672 struct vm_area_struct
*tail_vma
= NULL
;
1675 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1677 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1681 } while (vma
&& vma
->vm_start
< end
);
1682 *insertion_point
= vma
;
1683 tail_vma
->vm_next
= NULL
;
1684 if (mm
->unmap_area
== arch_unmap_area
)
1685 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1687 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1688 mm
->unmap_area(mm
, addr
);
1689 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1693 * Split a vma into two pieces at address 'addr', a new vma is allocated
1694 * either for the first part or the the tail.
1696 int split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1697 unsigned long addr
, int new_below
)
1699 struct mempolicy
*pol
;
1700 struct vm_area_struct
*new;
1702 if (is_vm_hugetlb_page(vma
) && (addr
& ~HPAGE_MASK
))
1705 if (mm
->map_count
>= sysctl_max_map_count
)
1708 new = kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
1712 /* most fields are the same, copy all, and then fixup */
1718 new->vm_start
= addr
;
1719 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
1722 pol
= mpol_copy(vma_policy(vma
));
1724 kmem_cache_free(vm_area_cachep
, new);
1725 return PTR_ERR(pol
);
1727 vma_set_policy(new, pol
);
1730 get_file(new->vm_file
);
1732 if (new->vm_ops
&& new->vm_ops
->open
)
1733 new->vm_ops
->open(new);
1736 vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
1737 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
1739 vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
1744 /* Munmap is split into 2 main parts -- this part which finds
1745 * what needs doing, and the areas themselves, which do the
1746 * work. This now handles partial unmappings.
1747 * Jeremy Fitzhardinge <jeremy@goop.org>
1749 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1752 struct vm_area_struct
*vma
, *prev
, *last
;
1754 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
1757 if ((len
= PAGE_ALIGN(len
)) == 0)
1760 /* Find the first overlapping VMA */
1761 vma
= find_vma_prev(mm
, start
, &prev
);
1764 /* we have start < vma->vm_end */
1766 /* if it doesn't overlap, we have nothing.. */
1768 if (vma
->vm_start
>= end
)
1772 * If we need to split any vma, do it now to save pain later.
1774 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1775 * unmapped vm_area_struct will remain in use: so lower split_vma
1776 * places tmp vma above, and higher split_vma places tmp vma below.
1778 if (start
> vma
->vm_start
) {
1779 int error
= split_vma(mm
, vma
, start
, 0);
1785 /* Does it split the last one? */
1786 last
= find_vma(mm
, end
);
1787 if (last
&& end
> last
->vm_start
) {
1788 int error
= split_vma(mm
, last
, end
, 1);
1792 vma
= prev
? prev
->vm_next
: mm
->mmap
;
1795 * Remove the vma's, and unmap the actual pages
1797 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
1798 unmap_region(mm
, vma
, prev
, start
, end
);
1800 /* Fix up all other VM information */
1801 remove_vma_list(mm
, vma
);
1806 EXPORT_SYMBOL(do_munmap
);
1808 asmlinkage
long sys_munmap(unsigned long addr
, size_t len
)
1811 struct mm_struct
*mm
= current
->mm
;
1813 profile_munmap(addr
);
1815 down_write(&mm
->mmap_sem
);
1816 ret
= do_munmap(mm
, addr
, len
);
1817 up_write(&mm
->mmap_sem
);
1821 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
1823 #ifdef CONFIG_DEBUG_VM
1824 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
1826 up_read(&mm
->mmap_sem
);
1832 * this is really a simplified "do_mmap". it only handles
1833 * anonymous maps. eventually we may be able to do some
1834 * brk-specific accounting here.
1836 unsigned long do_brk(unsigned long addr
, unsigned long len
)
1838 struct mm_struct
* mm
= current
->mm
;
1839 struct vm_area_struct
* vma
, * prev
;
1840 unsigned long flags
;
1841 struct rb_node
** rb_link
, * rb_parent
;
1842 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
1844 len
= PAGE_ALIGN(len
);
1848 if ((addr
+ len
) > TASK_SIZE
|| (addr
+ len
) < addr
)
1854 if (mm
->def_flags
& VM_LOCKED
) {
1855 unsigned long locked
, lock_limit
;
1856 locked
= len
>> PAGE_SHIFT
;
1857 locked
+= mm
->locked_vm
;
1858 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
1859 lock_limit
>>= PAGE_SHIFT
;
1860 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1865 * mm->mmap_sem is required to protect against another thread
1866 * changing the mappings in case we sleep.
1868 verify_mm_writelocked(mm
);
1871 * Clear old maps. this also does some error checking for us
1874 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1875 if (vma
&& vma
->vm_start
< addr
+ len
) {
1876 if (do_munmap(mm
, addr
, len
))
1881 /* Check against address space limits *after* clearing old maps... */
1882 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1885 if (mm
->map_count
> sysctl_max_map_count
)
1888 if (security_vm_enough_memory(len
>> PAGE_SHIFT
))
1891 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
1893 /* Can we just expand an old private anonymous mapping? */
1894 if (vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
1895 NULL
, NULL
, pgoff
, NULL
))
1899 * create a vma struct for an anonymous mapping
1901 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1903 vm_unacct_memory(len
>> PAGE_SHIFT
);
1908 vma
->vm_start
= addr
;
1909 vma
->vm_end
= addr
+ len
;
1910 vma
->vm_pgoff
= pgoff
;
1911 vma
->vm_flags
= flags
;
1912 vma
->vm_page_prot
= protection_map
[flags
& 0x0f];
1913 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1915 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1916 if (flags
& VM_LOCKED
) {
1917 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
1918 make_pages_present(addr
, addr
+ len
);
1923 EXPORT_SYMBOL(do_brk
);
1925 /* Release all mmaps. */
1926 void exit_mmap(struct mm_struct
*mm
)
1928 struct mmu_gather
*tlb
;
1929 struct vm_area_struct
*vma
= mm
->mmap
;
1930 unsigned long nr_accounted
= 0;
1935 tlb
= tlb_gather_mmu(mm
, 1);
1936 /* Don't update_hiwater_rss(mm) here, do_exit already did */
1937 /* Use -1 here to ensure all VMAs in the mm are unmapped */
1938 end
= unmap_vmas(&tlb
, vma
, 0, -1, &nr_accounted
, NULL
);
1939 vm_unacct_memory(nr_accounted
);
1940 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, 0);
1941 tlb_finish_mmu(tlb
, 0, end
);
1944 * Walk the list again, actually closing and freeing it,
1945 * with preemption enabled, without holding any MM locks.
1948 vma
= remove_vma(vma
);
1950 BUG_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
1953 /* Insert vm structure into process list sorted by address
1954 * and into the inode's i_mmap tree. If vm_file is non-NULL
1955 * then i_mmap_lock is taken here.
1957 int insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
1959 struct vm_area_struct
* __vma
, * prev
;
1960 struct rb_node
** rb_link
, * rb_parent
;
1963 * The vm_pgoff of a purely anonymous vma should be irrelevant
1964 * until its first write fault, when page's anon_vma and index
1965 * are set. But now set the vm_pgoff it will almost certainly
1966 * end up with (unless mremap moves it elsewhere before that
1967 * first wfault), so /proc/pid/maps tells a consistent story.
1969 * By setting it to reflect the virtual start address of the
1970 * vma, merges and splits can happen in a seamless way, just
1971 * using the existing file pgoff checks and manipulations.
1972 * Similarly in do_mmap_pgoff and in do_brk.
1974 if (!vma
->vm_file
) {
1975 BUG_ON(vma
->anon_vma
);
1976 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
1978 __vma
= find_vma_prepare(mm
,vma
->vm_start
,&prev
,&rb_link
,&rb_parent
);
1979 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
1981 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
1982 security_vm_enough_memory(vma_pages(vma
)))
1984 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1989 * Copy the vma structure to a new location in the same mm,
1990 * prior to moving page table entries, to effect an mremap move.
1992 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
1993 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
1995 struct vm_area_struct
*vma
= *vmap
;
1996 unsigned long vma_start
= vma
->vm_start
;
1997 struct mm_struct
*mm
= vma
->vm_mm
;
1998 struct vm_area_struct
*new_vma
, *prev
;
1999 struct rb_node
**rb_link
, *rb_parent
;
2000 struct mempolicy
*pol
;
2003 * If anonymous vma has not yet been faulted, update new pgoff
2004 * to match new location, to increase its chance of merging.
2006 if (!vma
->vm_file
&& !vma
->anon_vma
)
2007 pgoff
= addr
>> PAGE_SHIFT
;
2009 find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2010 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2011 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2014 * Source vma may have been merged into new_vma
2016 if (vma_start
>= new_vma
->vm_start
&&
2017 vma_start
< new_vma
->vm_end
)
2020 new_vma
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
2023 pol
= mpol_copy(vma_policy(vma
));
2025 kmem_cache_free(vm_area_cachep
, new_vma
);
2028 vma_set_policy(new_vma
, pol
);
2029 new_vma
->vm_start
= addr
;
2030 new_vma
->vm_end
= addr
+ len
;
2031 new_vma
->vm_pgoff
= pgoff
;
2032 if (new_vma
->vm_file
)
2033 get_file(new_vma
->vm_file
);
2034 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2035 new_vma
->vm_ops
->open(new_vma
);
2036 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2043 * Return true if the calling process may expand its vm space by the passed
2046 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2048 unsigned long cur
= mm
->total_vm
; /* pages */
2051 lim
= current
->signal
->rlim
[RLIMIT_AS
].rlim_cur
>> PAGE_SHIFT
;
2053 if (cur
+ npages
> lim
)