6 * Address space accounting code <alan@redhat.com>
9 #include <linux/slab.h>
10 #include <linux/backing-dev.h>
12 #include <linux/shm.h>
13 #include <linux/mman.h>
14 #include <linux/pagemap.h>
15 #include <linux/swap.h>
16 #include <linux/syscalls.h>
17 #include <linux/capability.h>
18 #include <linux/init.h>
19 #include <linux/file.h>
21 #include <linux/personality.h>
22 #include <linux/security.h>
23 #include <linux/hugetlb.h>
24 #include <linux/profile.h>
25 #include <linux/module.h>
26 #include <linux/mount.h>
27 #include <linux/mempolicy.h>
28 #include <linux/rmap.h>
29 #include <linux/mmu_notifier.h>
31 #include <asm/uaccess.h>
32 #include <asm/cacheflush.h>
34 #include <asm/mmu_context.h>
38 #ifndef arch_mmap_check
39 #define arch_mmap_check(addr, len, flags) (0)
42 #ifndef arch_rebalance_pgtables
43 #define arch_rebalance_pgtables(addr, len) (addr)
46 static void unmap_region(struct mm_struct
*mm
,
47 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
48 unsigned long start
, unsigned long end
);
51 * WARNING: the debugging will use recursive algorithms so never enable this
52 * unless you know what you are doing.
56 /* description of effects of mapping type and prot in current implementation.
57 * this is due to the limited x86 page protection hardware. The expected
58 * behavior is in parens:
61 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
62 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
63 * w: (no) no w: (no) no w: (yes) yes w: (no) no
64 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
66 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
67 * w: (no) no w: (no) no w: (copy) copy w: (no) no
68 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
71 pgprot_t protection_map
[16] = {
72 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
73 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
76 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
78 return __pgprot(pgprot_val(protection_map
[vm_flags
&
79 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
80 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
82 EXPORT_SYMBOL(vm_get_page_prot
);
84 int sysctl_overcommit_memory
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
85 int sysctl_overcommit_ratio
= 50; /* default is 50% */
86 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
87 atomic_long_t vm_committed_space
= ATOMIC_LONG_INIT(0);
89 /* amount of vm to protect from userspace access */
90 unsigned long mmap_min_addr
= CONFIG_DEFAULT_MMAP_MIN_ADDR
;
93 * Check that a process has enough memory to allocate a new virtual
94 * mapping. 0 means there is enough memory for the allocation to
95 * succeed and -ENOMEM implies there is not.
97 * We currently support three overcommit policies, which are set via the
98 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
100 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
101 * Additional code 2002 Jul 20 by Robert Love.
103 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
105 * Note this is a helper function intended to be used by LSMs which
106 * wish to use this logic.
108 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
110 unsigned long free
, allowed
;
112 vm_acct_memory(pages
);
115 * Sometimes we want to use more memory than we have
117 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
120 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
123 free
= global_page_state(NR_FILE_PAGES
);
124 free
+= nr_swap_pages
;
127 * Any slabs which are created with the
128 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
129 * which are reclaimable, under pressure. The dentry
130 * cache and most inode caches should fall into this
132 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
135 * Leave the last 3% for root
144 * nr_free_pages() is very expensive on large systems,
145 * only call if we're about to fail.
150 * Leave reserved pages. The pages are not for anonymous pages.
152 if (n
<= totalreserve_pages
)
155 n
-= totalreserve_pages
;
158 * Leave the last 3% for root
170 allowed
= (totalram_pages
- hugetlb_total_pages())
171 * sysctl_overcommit_ratio
/ 100;
173 * Leave the last 3% for root
176 allowed
-= allowed
/ 32;
177 allowed
+= total_swap_pages
;
179 /* Don't let a single process grow too big:
180 leave 3% of the size of this process for other processes */
182 allowed
-= mm
->total_vm
/ 32;
185 * cast `allowed' as a signed long because vm_committed_space
186 * sometimes has a negative value
188 if (atomic_long_read(&vm_committed_space
) < (long)allowed
)
191 vm_unacct_memory(pages
);
197 * Requires inode->i_mapping->i_mmap_lock
199 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
200 struct file
*file
, struct address_space
*mapping
)
202 if (vma
->vm_flags
& VM_DENYWRITE
)
203 atomic_inc(&file
->f_path
.dentry
->d_inode
->i_writecount
);
204 if (vma
->vm_flags
& VM_SHARED
)
205 mapping
->i_mmap_writable
--;
207 flush_dcache_mmap_lock(mapping
);
208 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
209 list_del_init(&vma
->shared
.vm_set
.list
);
211 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
212 flush_dcache_mmap_unlock(mapping
);
216 * Unlink a file-based vm structure from its prio_tree, to hide
217 * vma from rmap and vmtruncate before freeing its page tables.
219 void unlink_file_vma(struct vm_area_struct
*vma
)
221 struct file
*file
= vma
->vm_file
;
224 struct address_space
*mapping
= file
->f_mapping
;
225 spin_lock(&mapping
->i_mmap_lock
);
226 __remove_shared_vm_struct(vma
, file
, mapping
);
227 spin_unlock(&mapping
->i_mmap_lock
);
232 * Close a vm structure and free it, returning the next.
234 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
236 struct vm_area_struct
*next
= vma
->vm_next
;
239 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
240 vma
->vm_ops
->close(vma
);
243 if (vma
->vm_flags
& VM_EXECUTABLE
)
244 removed_exe_file_vma(vma
->vm_mm
);
246 mpol_put(vma_policy(vma
));
247 kmem_cache_free(vm_area_cachep
, vma
);
251 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
253 unsigned long rlim
, retval
;
254 unsigned long newbrk
, oldbrk
;
255 struct mm_struct
*mm
= current
->mm
;
256 unsigned long min_brk
;
258 down_write(&mm
->mmap_sem
);
260 #ifdef CONFIG_COMPAT_BRK
261 min_brk
= mm
->end_code
;
263 min_brk
= mm
->start_brk
;
269 * Check against rlimit here. If this check is done later after the test
270 * of oldbrk with newbrk then it can escape the test and let the data
271 * segment grow beyond its set limit the in case where the limit is
272 * not page aligned -Ram Gupta
274 rlim
= current
->signal
->rlim
[RLIMIT_DATA
].rlim_cur
;
275 if (rlim
< RLIM_INFINITY
&& (brk
- mm
->start_brk
) +
276 (mm
->end_data
- mm
->start_data
) > rlim
)
279 newbrk
= PAGE_ALIGN(brk
);
280 oldbrk
= PAGE_ALIGN(mm
->brk
);
281 if (oldbrk
== newbrk
)
284 /* Always allow shrinking brk. */
285 if (brk
<= mm
->brk
) {
286 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
291 /* Check against existing mmap mappings. */
292 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
295 /* Ok, looks good - let it rip. */
296 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
302 up_write(&mm
->mmap_sem
);
307 static int browse_rb(struct rb_root
*root
)
310 struct rb_node
*nd
, *pn
= NULL
;
311 unsigned long prev
= 0, pend
= 0;
313 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
314 struct vm_area_struct
*vma
;
315 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
316 if (vma
->vm_start
< prev
)
317 printk("vm_start %lx prev %lx\n", vma
->vm_start
, prev
), i
= -1;
318 if (vma
->vm_start
< pend
)
319 printk("vm_start %lx pend %lx\n", vma
->vm_start
, pend
);
320 if (vma
->vm_start
> vma
->vm_end
)
321 printk("vm_end %lx < vm_start %lx\n", vma
->vm_end
, vma
->vm_start
);
324 prev
= vma
->vm_start
;
328 for (nd
= pn
; nd
; nd
= rb_prev(nd
)) {
332 printk("backwards %d, forwards %d\n", j
, i
), i
= 0;
336 void validate_mm(struct mm_struct
*mm
)
340 struct vm_area_struct
*tmp
= mm
->mmap
;
345 if (i
!= mm
->map_count
)
346 printk("map_count %d vm_next %d\n", mm
->map_count
, i
), bug
= 1;
347 i
= browse_rb(&mm
->mm_rb
);
348 if (i
!= mm
->map_count
)
349 printk("map_count %d rb %d\n", mm
->map_count
, i
), bug
= 1;
353 #define validate_mm(mm) do { } while (0)
356 static struct vm_area_struct
*
357 find_vma_prepare(struct mm_struct
*mm
, unsigned long addr
,
358 struct vm_area_struct
**pprev
, struct rb_node
***rb_link
,
359 struct rb_node
** rb_parent
)
361 struct vm_area_struct
* vma
;
362 struct rb_node
** __rb_link
, * __rb_parent
, * rb_prev
;
364 __rb_link
= &mm
->mm_rb
.rb_node
;
365 rb_prev
= __rb_parent
= NULL
;
369 struct vm_area_struct
*vma_tmp
;
371 __rb_parent
= *__rb_link
;
372 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
374 if (vma_tmp
->vm_end
> addr
) {
376 if (vma_tmp
->vm_start
<= addr
)
378 __rb_link
= &__rb_parent
->rb_left
;
380 rb_prev
= __rb_parent
;
381 __rb_link
= &__rb_parent
->rb_right
;
387 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
388 *rb_link
= __rb_link
;
389 *rb_parent
= __rb_parent
;
394 __vma_link_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
395 struct vm_area_struct
*prev
, struct rb_node
*rb_parent
)
398 vma
->vm_next
= prev
->vm_next
;
403 vma
->vm_next
= rb_entry(rb_parent
,
404 struct vm_area_struct
, vm_rb
);
410 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
411 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
413 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
414 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
417 static inline void __vma_link_file(struct vm_area_struct
*vma
)
423 struct address_space
*mapping
= file
->f_mapping
;
425 if (vma
->vm_flags
& VM_DENYWRITE
)
426 atomic_dec(&file
->f_path
.dentry
->d_inode
->i_writecount
);
427 if (vma
->vm_flags
& VM_SHARED
)
428 mapping
->i_mmap_writable
++;
430 flush_dcache_mmap_lock(mapping
);
431 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
432 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
434 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
435 flush_dcache_mmap_unlock(mapping
);
440 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
441 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
442 struct rb_node
*rb_parent
)
444 __vma_link_list(mm
, vma
, prev
, rb_parent
);
445 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
446 __anon_vma_link(vma
);
449 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
450 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
451 struct rb_node
*rb_parent
)
453 struct address_space
*mapping
= NULL
;
456 mapping
= vma
->vm_file
->f_mapping
;
459 spin_lock(&mapping
->i_mmap_lock
);
460 vma
->vm_truncate_count
= mapping
->truncate_count
;
464 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
465 __vma_link_file(vma
);
467 anon_vma_unlock(vma
);
469 spin_unlock(&mapping
->i_mmap_lock
);
476 * Helper for vma_adjust in the split_vma insert case:
477 * insert vm structure into list and rbtree and anon_vma,
478 * but it has already been inserted into prio_tree earlier.
481 __insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
483 struct vm_area_struct
* __vma
, * prev
;
484 struct rb_node
** rb_link
, * rb_parent
;
486 __vma
= find_vma_prepare(mm
, vma
->vm_start
,&prev
, &rb_link
, &rb_parent
);
487 BUG_ON(__vma
&& __vma
->vm_start
< vma
->vm_end
);
488 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
493 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
494 struct vm_area_struct
*prev
)
496 prev
->vm_next
= vma
->vm_next
;
497 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
498 if (mm
->mmap_cache
== vma
)
499 mm
->mmap_cache
= prev
;
503 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
504 * is already present in an i_mmap tree without adjusting the tree.
505 * The following helper function should be used when such adjustments
506 * are necessary. The "insert" vma (if any) is to be inserted
507 * before we drop the necessary locks.
509 void vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
510 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
512 struct mm_struct
*mm
= vma
->vm_mm
;
513 struct vm_area_struct
*next
= vma
->vm_next
;
514 struct vm_area_struct
*importer
= NULL
;
515 struct address_space
*mapping
= NULL
;
516 struct prio_tree_root
*root
= NULL
;
517 struct file
*file
= vma
->vm_file
;
518 struct anon_vma
*anon_vma
= NULL
;
519 long adjust_next
= 0;
522 if (next
&& !insert
) {
523 if (end
>= next
->vm_end
) {
525 * vma expands, overlapping all the next, and
526 * perhaps the one after too (mprotect case 6).
528 again
: remove_next
= 1 + (end
> next
->vm_end
);
530 anon_vma
= next
->anon_vma
;
532 } else if (end
> next
->vm_start
) {
534 * vma expands, overlapping part of the next:
535 * mprotect case 5 shifting the boundary up.
537 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
538 anon_vma
= next
->anon_vma
;
540 } else if (end
< vma
->vm_end
) {
542 * vma shrinks, and !insert tells it's not
543 * split_vma inserting another: so it must be
544 * mprotect case 4 shifting the boundary down.
546 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
547 anon_vma
= next
->anon_vma
;
553 mapping
= file
->f_mapping
;
554 if (!(vma
->vm_flags
& VM_NONLINEAR
))
555 root
= &mapping
->i_mmap
;
556 spin_lock(&mapping
->i_mmap_lock
);
558 vma
->vm_truncate_count
!= next
->vm_truncate_count
) {
560 * unmap_mapping_range might be in progress:
561 * ensure that the expanding vma is rescanned.
563 importer
->vm_truncate_count
= 0;
566 insert
->vm_truncate_count
= vma
->vm_truncate_count
;
568 * Put into prio_tree now, so instantiated pages
569 * are visible to arm/parisc __flush_dcache_page
570 * throughout; but we cannot insert into address
571 * space until vma start or end is updated.
573 __vma_link_file(insert
);
578 * When changing only vma->vm_end, we don't really need
581 if (vma
->anon_vma
&& (insert
|| importer
|| start
!= vma
->vm_start
))
582 anon_vma
= vma
->anon_vma
;
584 spin_lock(&anon_vma
->lock
);
586 * Easily overlooked: when mprotect shifts the boundary,
587 * make sure the expanding vma has anon_vma set if the
588 * shrinking vma had, to cover any anon pages imported.
590 if (importer
&& !importer
->anon_vma
) {
591 importer
->anon_vma
= anon_vma
;
592 __anon_vma_link(importer
);
597 flush_dcache_mmap_lock(mapping
);
598 vma_prio_tree_remove(vma
, root
);
600 vma_prio_tree_remove(next
, root
);
603 vma
->vm_start
= start
;
605 vma
->vm_pgoff
= pgoff
;
607 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
608 next
->vm_pgoff
+= adjust_next
;
613 vma_prio_tree_insert(next
, root
);
614 vma_prio_tree_insert(vma
, root
);
615 flush_dcache_mmap_unlock(mapping
);
620 * vma_merge has merged next into vma, and needs
621 * us to remove next before dropping the locks.
623 __vma_unlink(mm
, next
, vma
);
625 __remove_shared_vm_struct(next
, file
, mapping
);
627 __anon_vma_merge(vma
, next
);
630 * split_vma has split insert from vma, and needs
631 * us to insert it before dropping the locks
632 * (it may either follow vma or precede it).
634 __insert_vm_struct(mm
, insert
);
638 spin_unlock(&anon_vma
->lock
);
640 spin_unlock(&mapping
->i_mmap_lock
);
645 if (next
->vm_flags
& VM_EXECUTABLE
)
646 removed_exe_file_vma(mm
);
649 mpol_put(vma_policy(next
));
650 kmem_cache_free(vm_area_cachep
, next
);
652 * In mprotect's case 6 (see comments on vma_merge),
653 * we must remove another next too. It would clutter
654 * up the code too much to do both in one go.
656 if (remove_next
== 2) {
666 * If the vma has a ->close operation then the driver probably needs to release
667 * per-vma resources, so we don't attempt to merge those.
669 #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP)
671 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
672 struct file
*file
, unsigned long vm_flags
)
674 if (vma
->vm_flags
!= vm_flags
)
676 if (vma
->vm_file
!= file
)
678 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
683 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
684 struct anon_vma
*anon_vma2
)
686 return !anon_vma1
|| !anon_vma2
|| (anon_vma1
== anon_vma2
);
690 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
691 * in front of (at a lower virtual address and file offset than) the vma.
693 * We cannot merge two vmas if they have differently assigned (non-NULL)
694 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
696 * We don't check here for the merged mmap wrapping around the end of pagecache
697 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
698 * wrap, nor mmaps which cover the final page at index -1UL.
701 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
702 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
704 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
705 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
706 if (vma
->vm_pgoff
== vm_pgoff
)
713 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
714 * beyond (at a higher virtual address and file offset than) the vma.
716 * We cannot merge two vmas if they have differently assigned (non-NULL)
717 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
720 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
721 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
723 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
724 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
726 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
727 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
734 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
735 * whether that can be merged with its predecessor or its successor.
736 * Or both (it neatly fills a hole).
738 * In most cases - when called for mmap, brk or mremap - [addr,end) is
739 * certain not to be mapped by the time vma_merge is called; but when
740 * called for mprotect, it is certain to be already mapped (either at
741 * an offset within prev, or at the start of next), and the flags of
742 * this area are about to be changed to vm_flags - and the no-change
743 * case has already been eliminated.
745 * The following mprotect cases have to be considered, where AAAA is
746 * the area passed down from mprotect_fixup, never extending beyond one
747 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
749 * AAAA AAAA AAAA AAAA
750 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
751 * cannot merge might become might become might become
752 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
753 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
754 * mremap move: PPPPNNNNNNNN 8
756 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
757 * might become case 1 below case 2 below case 3 below
759 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
760 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
762 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
763 struct vm_area_struct
*prev
, unsigned long addr
,
764 unsigned long end
, unsigned long vm_flags
,
765 struct anon_vma
*anon_vma
, struct file
*file
,
766 pgoff_t pgoff
, struct mempolicy
*policy
)
768 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
769 struct vm_area_struct
*area
, *next
;
772 * We later require that vma->vm_flags == vm_flags,
773 * so this tests vma->vm_flags & VM_SPECIAL, too.
775 if (vm_flags
& VM_SPECIAL
)
779 next
= prev
->vm_next
;
783 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
784 next
= next
->vm_next
;
787 * Can it merge with the predecessor?
789 if (prev
&& prev
->vm_end
== addr
&&
790 mpol_equal(vma_policy(prev
), policy
) &&
791 can_vma_merge_after(prev
, vm_flags
,
792 anon_vma
, file
, pgoff
)) {
794 * OK, it can. Can we now merge in the successor as well?
796 if (next
&& end
== next
->vm_start
&&
797 mpol_equal(policy
, vma_policy(next
)) &&
798 can_vma_merge_before(next
, vm_flags
,
799 anon_vma
, file
, pgoff
+pglen
) &&
800 is_mergeable_anon_vma(prev
->anon_vma
,
803 vma_adjust(prev
, prev
->vm_start
,
804 next
->vm_end
, prev
->vm_pgoff
, NULL
);
805 } else /* cases 2, 5, 7 */
806 vma_adjust(prev
, prev
->vm_start
,
807 end
, prev
->vm_pgoff
, NULL
);
812 * Can this new request be merged in front of next?
814 if (next
&& end
== next
->vm_start
&&
815 mpol_equal(policy
, vma_policy(next
)) &&
816 can_vma_merge_before(next
, vm_flags
,
817 anon_vma
, file
, pgoff
+pglen
)) {
818 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
819 vma_adjust(prev
, prev
->vm_start
,
820 addr
, prev
->vm_pgoff
, NULL
);
821 else /* cases 3, 8 */
822 vma_adjust(area
, addr
, next
->vm_end
,
823 next
->vm_pgoff
- pglen
, NULL
);
831 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
832 * neighbouring vmas for a suitable anon_vma, before it goes off
833 * to allocate a new anon_vma. It checks because a repetitive
834 * sequence of mprotects and faults may otherwise lead to distinct
835 * anon_vmas being allocated, preventing vma merge in subsequent
838 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
840 struct vm_area_struct
*near
;
841 unsigned long vm_flags
;
848 * Since only mprotect tries to remerge vmas, match flags
849 * which might be mprotected into each other later on.
850 * Neither mlock nor madvise tries to remerge at present,
851 * so leave their flags as obstructing a merge.
853 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
854 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
856 if (near
->anon_vma
&& vma
->vm_end
== near
->vm_start
&&
857 mpol_equal(vma_policy(vma
), vma_policy(near
)) &&
858 can_vma_merge_before(near
, vm_flags
,
859 NULL
, vma
->vm_file
, vma
->vm_pgoff
+
860 ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
)))
861 return near
->anon_vma
;
864 * It is potentially slow to have to call find_vma_prev here.
865 * But it's only on the first write fault on the vma, not
866 * every time, and we could devise a way to avoid it later
867 * (e.g. stash info in next's anon_vma_node when assigning
868 * an anon_vma, or when trying vma_merge). Another time.
870 BUG_ON(find_vma_prev(vma
->vm_mm
, vma
->vm_start
, &near
) != vma
);
874 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
875 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
877 if (near
->anon_vma
&& near
->vm_end
== vma
->vm_start
&&
878 mpol_equal(vma_policy(near
), vma_policy(vma
)) &&
879 can_vma_merge_after(near
, vm_flags
,
880 NULL
, vma
->vm_file
, vma
->vm_pgoff
))
881 return near
->anon_vma
;
884 * There's no absolute need to look only at touching neighbours:
885 * we could search further afield for "compatible" anon_vmas.
886 * But it would probably just be a waste of time searching,
887 * or lead to too many vmas hanging off the same anon_vma.
888 * We're trying to allow mprotect remerging later on,
889 * not trying to minimize memory used for anon_vmas.
894 #ifdef CONFIG_PROC_FS
895 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
896 struct file
*file
, long pages
)
898 const unsigned long stack_flags
899 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
902 mm
->shared_vm
+= pages
;
903 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
904 mm
->exec_vm
+= pages
;
905 } else if (flags
& stack_flags
)
906 mm
->stack_vm
+= pages
;
907 if (flags
& (VM_RESERVED
|VM_IO
))
908 mm
->reserved_vm
+= pages
;
910 #endif /* CONFIG_PROC_FS */
913 * The caller must hold down_write(current->mm->mmap_sem).
916 unsigned long do_mmap_pgoff(struct file
* file
, unsigned long addr
,
917 unsigned long len
, unsigned long prot
,
918 unsigned long flags
, unsigned long pgoff
)
920 struct mm_struct
* mm
= current
->mm
;
922 unsigned int vm_flags
;
925 unsigned long reqprot
= prot
;
928 * Does the application expect PROT_READ to imply PROT_EXEC?
930 * (the exception is when the underlying filesystem is noexec
931 * mounted, in which case we dont add PROT_EXEC.)
933 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
934 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
940 if (!(flags
& MAP_FIXED
))
941 addr
= round_hint_to_min(addr
);
943 error
= arch_mmap_check(addr
, len
, flags
);
947 /* Careful about overflows.. */
948 len
= PAGE_ALIGN(len
);
949 if (!len
|| len
> TASK_SIZE
)
952 /* offset overflow? */
953 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
956 /* Too many mappings? */
957 if (mm
->map_count
> sysctl_max_map_count
)
960 /* Obtain the address to map to. we verify (or select) it and ensure
961 * that it represents a valid section of the address space.
963 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
964 if (addr
& ~PAGE_MASK
)
967 /* Do simple checking here so the lower-level routines won't have
968 * to. we assume access permissions have been handled by the open
969 * of the memory object, so we don't do any here.
971 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
972 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
974 if (flags
& MAP_LOCKED
) {
977 vm_flags
|= VM_LOCKED
;
979 /* mlock MCL_FUTURE? */
980 if (vm_flags
& VM_LOCKED
) {
981 unsigned long locked
, lock_limit
;
982 locked
= len
>> PAGE_SHIFT
;
983 locked
+= mm
->locked_vm
;
984 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
985 lock_limit
>>= PAGE_SHIFT
;
986 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
990 inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
993 switch (flags
& MAP_TYPE
) {
995 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
999 * Make sure we don't allow writing to an append-only
1002 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1006 * Make sure there are no mandatory locks on the file.
1008 if (locks_verify_locked(inode
))
1011 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1012 if (!(file
->f_mode
& FMODE_WRITE
))
1013 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1017 if (!(file
->f_mode
& FMODE_READ
))
1019 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1020 if (vm_flags
& VM_EXEC
)
1022 vm_flags
&= ~VM_MAYEXEC
;
1024 if (is_file_hugepages(file
))
1027 if (!file
->f_op
|| !file
->f_op
->mmap
)
1035 switch (flags
& MAP_TYPE
) {
1041 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1045 * Set pgoff according to addr for anon_vma.
1047 pgoff
= addr
>> PAGE_SHIFT
;
1054 error
= security_file_mmap(file
, reqprot
, prot
, flags
, addr
, 0);
1058 return mmap_region(file
, addr
, len
, flags
, vm_flags
, pgoff
,
1061 EXPORT_SYMBOL(do_mmap_pgoff
);
1064 * Some shared mappigns will want the pages marked read-only
1065 * to track write events. If so, we'll downgrade vm_page_prot
1066 * to the private version (using protection_map[] without the
1069 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1071 unsigned int vm_flags
= vma
->vm_flags
;
1073 /* If it was private or non-writable, the write bit is already clear */
1074 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1077 /* The backer wishes to know when pages are first written to? */
1078 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1081 /* The open routine did something to the protections already? */
1082 if (pgprot_val(vma
->vm_page_prot
) !=
1083 pgprot_val(vm_get_page_prot(vm_flags
)))
1086 /* Specialty mapping? */
1087 if (vm_flags
& (VM_PFNMAP
|VM_INSERTPAGE
))
1090 /* Can the mapping track the dirty pages? */
1091 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1092 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1095 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1096 unsigned long len
, unsigned long flags
,
1097 unsigned int vm_flags
, unsigned long pgoff
,
1100 struct mm_struct
*mm
= current
->mm
;
1101 struct vm_area_struct
*vma
, *prev
;
1102 int correct_wcount
= 0;
1104 struct rb_node
**rb_link
, *rb_parent
;
1105 unsigned long charged
= 0;
1106 struct inode
*inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1108 /* Clear old maps */
1111 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1112 if (vma
&& vma
->vm_start
< addr
+ len
) {
1113 if (do_munmap(mm
, addr
, len
))
1118 /* Check against address space limit. */
1119 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1122 if (flags
& MAP_NORESERVE
)
1123 vm_flags
|= VM_NORESERVE
;
1125 if (accountable
&& (!(flags
& MAP_NORESERVE
) ||
1126 sysctl_overcommit_memory
== OVERCOMMIT_NEVER
)) {
1127 if (vm_flags
& VM_SHARED
) {
1128 /* Check memory availability in shmem_file_setup? */
1129 vm_flags
|= VM_ACCOUNT
;
1130 } else if (vm_flags
& VM_WRITE
) {
1132 * Private writable mapping: check memory availability
1134 charged
= len
>> PAGE_SHIFT
;
1135 if (security_vm_enough_memory(charged
))
1137 vm_flags
|= VM_ACCOUNT
;
1142 * Can we just expand an old private anonymous mapping?
1143 * The VM_SHARED test is necessary because shmem_zero_setup
1144 * will create the file object for a shared anonymous map below.
1146 if (!file
&& !(vm_flags
& VM_SHARED
) &&
1147 vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
,
1148 NULL
, NULL
, pgoff
, NULL
))
1152 * Determine the object being mapped and call the appropriate
1153 * specific mapper. the address has already been validated, but
1154 * not unmapped, but the maps are removed from the list.
1156 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1163 vma
->vm_start
= addr
;
1164 vma
->vm_end
= addr
+ len
;
1165 vma
->vm_flags
= vm_flags
;
1166 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1167 vma
->vm_pgoff
= pgoff
;
1171 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1173 if (vm_flags
& VM_DENYWRITE
) {
1174 error
= deny_write_access(file
);
1179 vma
->vm_file
= file
;
1181 error
= file
->f_op
->mmap(file
, vma
);
1183 goto unmap_and_free_vma
;
1184 if (vm_flags
& VM_EXECUTABLE
)
1185 added_exe_file_vma(mm
);
1186 } else if (vm_flags
& VM_SHARED
) {
1187 error
= shmem_zero_setup(vma
);
1192 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1193 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1194 * that memory reservation must be checked; but that reservation
1195 * belongs to shared memory object, not to vma: so now clear it.
1197 if ((vm_flags
& (VM_SHARED
|VM_ACCOUNT
)) == (VM_SHARED
|VM_ACCOUNT
))
1198 vma
->vm_flags
&= ~VM_ACCOUNT
;
1200 /* Can addr have changed??
1202 * Answer: Yes, several device drivers can do it in their
1203 * f_op->mmap method. -DaveM
1205 addr
= vma
->vm_start
;
1206 pgoff
= vma
->vm_pgoff
;
1207 vm_flags
= vma
->vm_flags
;
1209 if (vma_wants_writenotify(vma
))
1210 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
& ~VM_SHARED
);
1212 if (file
&& vma_merge(mm
, prev
, addr
, vma
->vm_end
,
1213 vma
->vm_flags
, NULL
, file
, pgoff
, vma_policy(vma
))) {
1214 mpol_put(vma_policy(vma
));
1215 kmem_cache_free(vm_area_cachep
, vma
);
1217 if (vm_flags
& VM_EXECUTABLE
)
1218 removed_exe_file_vma(mm
);
1220 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1221 file
= vma
->vm_file
;
1224 /* Once vma denies write, undo our temporary denial count */
1226 atomic_inc(&inode
->i_writecount
);
1228 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1229 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1230 if (vm_flags
& VM_LOCKED
) {
1231 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
1232 make_pages_present(addr
, addr
+ len
);
1234 if ((flags
& MAP_POPULATE
) && !(flags
& MAP_NONBLOCK
))
1235 make_pages_present(addr
, addr
+ len
);
1240 atomic_inc(&inode
->i_writecount
);
1241 vma
->vm_file
= NULL
;
1244 /* Undo any partial mapping done by a device driver. */
1245 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1248 kmem_cache_free(vm_area_cachep
, vma
);
1251 vm_unacct_memory(charged
);
1255 /* Get an address range which is currently unmapped.
1256 * For shmat() with addr=0.
1258 * Ugly calling convention alert:
1259 * Return value with the low bits set means error value,
1261 * if (ret & ~PAGE_MASK)
1264 * This function "knows" that -ENOMEM has the bits set.
1266 #ifndef HAVE_ARCH_UNMAPPED_AREA
1268 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1269 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1271 struct mm_struct
*mm
= current
->mm
;
1272 struct vm_area_struct
*vma
;
1273 unsigned long start_addr
;
1275 if (len
> TASK_SIZE
)
1278 if (flags
& MAP_FIXED
)
1282 addr
= PAGE_ALIGN(addr
);
1283 vma
= find_vma(mm
, addr
);
1284 if (TASK_SIZE
- len
>= addr
&&
1285 (!vma
|| addr
+ len
<= vma
->vm_start
))
1288 if (len
> mm
->cached_hole_size
) {
1289 start_addr
= addr
= mm
->free_area_cache
;
1291 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1292 mm
->cached_hole_size
= 0;
1296 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1297 /* At this point: (!vma || addr < vma->vm_end). */
1298 if (TASK_SIZE
- len
< addr
) {
1300 * Start a new search - just in case we missed
1303 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1304 addr
= TASK_UNMAPPED_BASE
;
1306 mm
->cached_hole_size
= 0;
1311 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1313 * Remember the place where we stopped the search:
1315 mm
->free_area_cache
= addr
+ len
;
1318 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1319 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1325 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1328 * Is this a new hole at the lowest possible address?
1330 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
) {
1331 mm
->free_area_cache
= addr
;
1332 mm
->cached_hole_size
= ~0UL;
1337 * This mmap-allocator allocates new areas top-down from below the
1338 * stack's low limit (the base):
1340 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1342 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1343 const unsigned long len
, const unsigned long pgoff
,
1344 const unsigned long flags
)
1346 struct vm_area_struct
*vma
;
1347 struct mm_struct
*mm
= current
->mm
;
1348 unsigned long addr
= addr0
;
1350 /* requested length too big for entire address space */
1351 if (len
> TASK_SIZE
)
1354 if (flags
& MAP_FIXED
)
1357 /* requesting a specific address */
1359 addr
= PAGE_ALIGN(addr
);
1360 vma
= find_vma(mm
, addr
);
1361 if (TASK_SIZE
- len
>= addr
&&
1362 (!vma
|| addr
+ len
<= vma
->vm_start
))
1366 /* check if free_area_cache is useful for us */
1367 if (len
<= mm
->cached_hole_size
) {
1368 mm
->cached_hole_size
= 0;
1369 mm
->free_area_cache
= mm
->mmap_base
;
1372 /* either no address requested or can't fit in requested address hole */
1373 addr
= mm
->free_area_cache
;
1375 /* make sure it can fit in the remaining address space */
1377 vma
= find_vma(mm
, addr
-len
);
1378 if (!vma
|| addr
<= vma
->vm_start
)
1379 /* remember the address as a hint for next time */
1380 return (mm
->free_area_cache
= addr
-len
);
1383 if (mm
->mmap_base
< len
)
1386 addr
= mm
->mmap_base
-len
;
1390 * Lookup failure means no vma is above this address,
1391 * else if new region fits below vma->vm_start,
1392 * return with success:
1394 vma
= find_vma(mm
, addr
);
1395 if (!vma
|| addr
+len
<= vma
->vm_start
)
1396 /* remember the address as a hint for next time */
1397 return (mm
->free_area_cache
= addr
);
1399 /* remember the largest hole we saw so far */
1400 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1401 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1403 /* try just below the current vma->vm_start */
1404 addr
= vma
->vm_start
-len
;
1405 } while (len
< vma
->vm_start
);
1409 * A failed mmap() very likely causes application failure,
1410 * so fall back to the bottom-up function here. This scenario
1411 * can happen with large stack limits and large mmap()
1414 mm
->cached_hole_size
= ~0UL;
1415 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1416 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1418 * Restore the topdown base:
1420 mm
->free_area_cache
= mm
->mmap_base
;
1421 mm
->cached_hole_size
= ~0UL;
1427 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1430 * Is this a new hole at the highest possible address?
1432 if (addr
> mm
->free_area_cache
)
1433 mm
->free_area_cache
= addr
;
1435 /* dont allow allocations above current base */
1436 if (mm
->free_area_cache
> mm
->mmap_base
)
1437 mm
->free_area_cache
= mm
->mmap_base
;
1441 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1442 unsigned long pgoff
, unsigned long flags
)
1444 unsigned long (*get_area
)(struct file
*, unsigned long,
1445 unsigned long, unsigned long, unsigned long);
1447 get_area
= current
->mm
->get_unmapped_area
;
1448 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1449 get_area
= file
->f_op
->get_unmapped_area
;
1450 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1451 if (IS_ERR_VALUE(addr
))
1454 if (addr
> TASK_SIZE
- len
)
1456 if (addr
& ~PAGE_MASK
)
1459 return arch_rebalance_pgtables(addr
, len
);
1462 EXPORT_SYMBOL(get_unmapped_area
);
1464 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1465 struct vm_area_struct
* find_vma(struct mm_struct
* mm
, unsigned long addr
)
1467 struct vm_area_struct
*vma
= NULL
;
1470 /* Check the cache first. */
1471 /* (Cache hit rate is typically around 35%.) */
1472 vma
= mm
->mmap_cache
;
1473 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1474 struct rb_node
* rb_node
;
1476 rb_node
= mm
->mm_rb
.rb_node
;
1480 struct vm_area_struct
* vma_tmp
;
1482 vma_tmp
= rb_entry(rb_node
,
1483 struct vm_area_struct
, vm_rb
);
1485 if (vma_tmp
->vm_end
> addr
) {
1487 if (vma_tmp
->vm_start
<= addr
)
1489 rb_node
= rb_node
->rb_left
;
1491 rb_node
= rb_node
->rb_right
;
1494 mm
->mmap_cache
= vma
;
1500 EXPORT_SYMBOL(find_vma
);
1502 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1503 struct vm_area_struct
*
1504 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1505 struct vm_area_struct
**pprev
)
1507 struct vm_area_struct
*vma
= NULL
, *prev
= NULL
;
1508 struct rb_node
* rb_node
;
1512 /* Guard against addr being lower than the first VMA */
1515 /* Go through the RB tree quickly. */
1516 rb_node
= mm
->mm_rb
.rb_node
;
1519 struct vm_area_struct
*vma_tmp
;
1520 vma_tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1522 if (addr
< vma_tmp
->vm_end
) {
1523 rb_node
= rb_node
->rb_left
;
1526 if (!prev
->vm_next
|| (addr
< prev
->vm_next
->vm_end
))
1528 rb_node
= rb_node
->rb_right
;
1534 return prev
? prev
->vm_next
: vma
;
1538 * Verify that the stack growth is acceptable and
1539 * update accounting. This is shared with both the
1540 * grow-up and grow-down cases.
1542 static int acct_stack_growth(struct vm_area_struct
* vma
, unsigned long size
, unsigned long grow
)
1544 struct mm_struct
*mm
= vma
->vm_mm
;
1545 struct rlimit
*rlim
= current
->signal
->rlim
;
1546 unsigned long new_start
;
1548 /* address space limit tests */
1549 if (!may_expand_vm(mm
, grow
))
1552 /* Stack limit test */
1553 if (size
> rlim
[RLIMIT_STACK
].rlim_cur
)
1556 /* mlock limit tests */
1557 if (vma
->vm_flags
& VM_LOCKED
) {
1558 unsigned long locked
;
1559 unsigned long limit
;
1560 locked
= mm
->locked_vm
+ grow
;
1561 limit
= rlim
[RLIMIT_MEMLOCK
].rlim_cur
>> PAGE_SHIFT
;
1562 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1566 /* Check to ensure the stack will not grow into a hugetlb-only region */
1567 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
1569 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
1573 * Overcommit.. This must be the final test, as it will
1574 * update security statistics.
1576 if (security_vm_enough_memory_mm(mm
, grow
))
1579 /* Ok, everything looks good - let it rip */
1580 mm
->total_vm
+= grow
;
1581 if (vma
->vm_flags
& VM_LOCKED
)
1582 mm
->locked_vm
+= grow
;
1583 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1587 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1589 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1590 * vma is the last one with address > vma->vm_end. Have to extend vma.
1592 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1596 if (!(vma
->vm_flags
& VM_GROWSUP
))
1600 * We must make sure the anon_vma is allocated
1601 * so that the anon_vma locking is not a noop.
1603 if (unlikely(anon_vma_prepare(vma
)))
1608 * vma->vm_start/vm_end cannot change under us because the caller
1609 * is required to hold the mmap_sem in read mode. We need the
1610 * anon_vma lock to serialize against concurrent expand_stacks.
1611 * Also guard against wrapping around to address 0.
1613 if (address
< PAGE_ALIGN(address
+4))
1614 address
= PAGE_ALIGN(address
+4);
1616 anon_vma_unlock(vma
);
1621 /* Somebody else might have raced and expanded it already */
1622 if (address
> vma
->vm_end
) {
1623 unsigned long size
, grow
;
1625 size
= address
- vma
->vm_start
;
1626 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1628 error
= acct_stack_growth(vma
, size
, grow
);
1630 vma
->vm_end
= address
;
1632 anon_vma_unlock(vma
);
1635 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1638 * vma is the first one with address < vma->vm_start. Have to extend vma.
1640 static inline int expand_downwards(struct vm_area_struct
*vma
,
1641 unsigned long address
)
1646 * We must make sure the anon_vma is allocated
1647 * so that the anon_vma locking is not a noop.
1649 if (unlikely(anon_vma_prepare(vma
)))
1652 address
&= PAGE_MASK
;
1653 error
= security_file_mmap(NULL
, 0, 0, 0, address
, 1);
1660 * vma->vm_start/vm_end cannot change under us because the caller
1661 * is required to hold the mmap_sem in read mode. We need the
1662 * anon_vma lock to serialize against concurrent expand_stacks.
1665 /* Somebody else might have raced and expanded it already */
1666 if (address
< vma
->vm_start
) {
1667 unsigned long size
, grow
;
1669 size
= vma
->vm_end
- address
;
1670 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1672 error
= acct_stack_growth(vma
, size
, grow
);
1674 vma
->vm_start
= address
;
1675 vma
->vm_pgoff
-= grow
;
1678 anon_vma_unlock(vma
);
1682 int expand_stack_downwards(struct vm_area_struct
*vma
, unsigned long address
)
1684 return expand_downwards(vma
, address
);
1687 #ifdef CONFIG_STACK_GROWSUP
1688 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1690 return expand_upwards(vma
, address
);
1693 struct vm_area_struct
*
1694 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1696 struct vm_area_struct
*vma
, *prev
;
1699 vma
= find_vma_prev(mm
, addr
, &prev
);
1700 if (vma
&& (vma
->vm_start
<= addr
))
1702 if (!prev
|| expand_stack(prev
, addr
))
1704 if (prev
->vm_flags
& VM_LOCKED
)
1705 make_pages_present(addr
, prev
->vm_end
);
1709 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1711 return expand_downwards(vma
, address
);
1714 struct vm_area_struct
*
1715 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1717 struct vm_area_struct
* vma
;
1718 unsigned long start
;
1721 vma
= find_vma(mm
,addr
);
1724 if (vma
->vm_start
<= addr
)
1726 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1728 start
= vma
->vm_start
;
1729 if (expand_stack(vma
, addr
))
1731 if (vma
->vm_flags
& VM_LOCKED
)
1732 make_pages_present(addr
, start
);
1738 * Ok - we have the memory areas we should free on the vma list,
1739 * so release them, and do the vma updates.
1741 * Called with the mm semaphore held.
1743 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1745 /* Update high watermark before we lower total_vm */
1746 update_hiwater_vm(mm
);
1748 long nrpages
= vma_pages(vma
);
1750 mm
->total_vm
-= nrpages
;
1751 if (vma
->vm_flags
& VM_LOCKED
)
1752 mm
->locked_vm
-= nrpages
;
1753 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1754 vma
= remove_vma(vma
);
1760 * Get rid of page table information in the indicated region.
1762 * Called with the mm semaphore held.
1764 static void unmap_region(struct mm_struct
*mm
,
1765 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1766 unsigned long start
, unsigned long end
)
1768 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1769 struct mmu_gather
*tlb
;
1770 unsigned long nr_accounted
= 0;
1773 tlb
= tlb_gather_mmu(mm
, 0);
1774 update_hiwater_rss(mm
);
1775 unmap_vmas(&tlb
, vma
, start
, end
, &nr_accounted
, NULL
);
1776 vm_unacct_memory(nr_accounted
);
1777 free_pgtables(tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1778 next
? next
->vm_start
: 0);
1779 tlb_finish_mmu(tlb
, start
, end
);
1783 * Create a list of vma's touched by the unmap, removing them from the mm's
1784 * vma list as we go..
1787 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1788 struct vm_area_struct
*prev
, unsigned long end
)
1790 struct vm_area_struct
**insertion_point
;
1791 struct vm_area_struct
*tail_vma
= NULL
;
1794 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1796 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1800 } while (vma
&& vma
->vm_start
< end
);
1801 *insertion_point
= vma
;
1802 tail_vma
->vm_next
= NULL
;
1803 if (mm
->unmap_area
== arch_unmap_area
)
1804 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1806 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1807 mm
->unmap_area(mm
, addr
);
1808 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1812 * Split a vma into two pieces at address 'addr', a new vma is allocated
1813 * either for the first part or the tail.
1815 int split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1816 unsigned long addr
, int new_below
)
1818 struct mempolicy
*pol
;
1819 struct vm_area_struct
*new;
1821 if (is_vm_hugetlb_page(vma
) && (addr
&
1822 ~(huge_page_mask(hstate_vma(vma
)))))
1825 if (mm
->map_count
>= sysctl_max_map_count
)
1828 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1832 /* most fields are the same, copy all, and then fixup */
1838 new->vm_start
= addr
;
1839 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
1842 pol
= mpol_dup(vma_policy(vma
));
1844 kmem_cache_free(vm_area_cachep
, new);
1845 return PTR_ERR(pol
);
1847 vma_set_policy(new, pol
);
1850 get_file(new->vm_file
);
1851 if (vma
->vm_flags
& VM_EXECUTABLE
)
1852 added_exe_file_vma(mm
);
1855 if (new->vm_ops
&& new->vm_ops
->open
)
1856 new->vm_ops
->open(new);
1859 vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
1860 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
1862 vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
1867 /* Munmap is split into 2 main parts -- this part which finds
1868 * what needs doing, and the areas themselves, which do the
1869 * work. This now handles partial unmappings.
1870 * Jeremy Fitzhardinge <jeremy@goop.org>
1872 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1875 struct vm_area_struct
*vma
, *prev
, *last
;
1877 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
1880 if ((len
= PAGE_ALIGN(len
)) == 0)
1883 /* Find the first overlapping VMA */
1884 vma
= find_vma_prev(mm
, start
, &prev
);
1887 /* we have start < vma->vm_end */
1889 /* if it doesn't overlap, we have nothing.. */
1891 if (vma
->vm_start
>= end
)
1895 * If we need to split any vma, do it now to save pain later.
1897 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1898 * unmapped vm_area_struct will remain in use: so lower split_vma
1899 * places tmp vma above, and higher split_vma places tmp vma below.
1901 if (start
> vma
->vm_start
) {
1902 int error
= split_vma(mm
, vma
, start
, 0);
1908 /* Does it split the last one? */
1909 last
= find_vma(mm
, end
);
1910 if (last
&& end
> last
->vm_start
) {
1911 int error
= split_vma(mm
, last
, end
, 1);
1915 vma
= prev
? prev
->vm_next
: mm
->mmap
;
1918 * Remove the vma's, and unmap the actual pages
1920 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
1921 unmap_region(mm
, vma
, prev
, start
, end
);
1923 /* Fix up all other VM information */
1924 remove_vma_list(mm
, vma
);
1929 EXPORT_SYMBOL(do_munmap
);
1931 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1934 struct mm_struct
*mm
= current
->mm
;
1936 profile_munmap(addr
);
1938 down_write(&mm
->mmap_sem
);
1939 ret
= do_munmap(mm
, addr
, len
);
1940 up_write(&mm
->mmap_sem
);
1944 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
1946 #ifdef CONFIG_DEBUG_VM
1947 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
1949 up_read(&mm
->mmap_sem
);
1955 * this is really a simplified "do_mmap". it only handles
1956 * anonymous maps. eventually we may be able to do some
1957 * brk-specific accounting here.
1959 unsigned long do_brk(unsigned long addr
, unsigned long len
)
1961 struct mm_struct
* mm
= current
->mm
;
1962 struct vm_area_struct
* vma
, * prev
;
1963 unsigned long flags
;
1964 struct rb_node
** rb_link
, * rb_parent
;
1965 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
1968 len
= PAGE_ALIGN(len
);
1972 if ((addr
+ len
) > TASK_SIZE
|| (addr
+ len
) < addr
)
1975 if (is_hugepage_only_range(mm
, addr
, len
))
1978 error
= security_file_mmap(NULL
, 0, 0, 0, addr
, 1);
1982 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
1984 error
= arch_mmap_check(addr
, len
, flags
);
1991 if (mm
->def_flags
& VM_LOCKED
) {
1992 unsigned long locked
, lock_limit
;
1993 locked
= len
>> PAGE_SHIFT
;
1994 locked
+= mm
->locked_vm
;
1995 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
1996 lock_limit
>>= PAGE_SHIFT
;
1997 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
2002 * mm->mmap_sem is required to protect against another thread
2003 * changing the mappings in case we sleep.
2005 verify_mm_writelocked(mm
);
2008 * Clear old maps. this also does some error checking for us
2011 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2012 if (vma
&& vma
->vm_start
< addr
+ len
) {
2013 if (do_munmap(mm
, addr
, len
))
2018 /* Check against address space limits *after* clearing old maps... */
2019 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2022 if (mm
->map_count
> sysctl_max_map_count
)
2025 if (security_vm_enough_memory(len
>> PAGE_SHIFT
))
2028 /* Can we just expand an old private anonymous mapping? */
2029 if (vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2030 NULL
, NULL
, pgoff
, NULL
))
2034 * create a vma struct for an anonymous mapping
2036 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2038 vm_unacct_memory(len
>> PAGE_SHIFT
);
2043 vma
->vm_start
= addr
;
2044 vma
->vm_end
= addr
+ len
;
2045 vma
->vm_pgoff
= pgoff
;
2046 vma
->vm_flags
= flags
;
2047 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2048 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2050 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2051 if (flags
& VM_LOCKED
) {
2052 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
2053 make_pages_present(addr
, addr
+ len
);
2058 EXPORT_SYMBOL(do_brk
);
2060 /* Release all mmaps. */
2061 void exit_mmap(struct mm_struct
*mm
)
2063 struct mmu_gather
*tlb
;
2064 struct vm_area_struct
*vma
= mm
->mmap
;
2065 unsigned long nr_accounted
= 0;
2068 /* mm's last user has gone, and its about to be pulled down */
2070 mmu_notifier_release(mm
);
2072 if (!mm
->mmap
) /* Can happen if dup_mmap() received an OOM */
2077 tlb
= tlb_gather_mmu(mm
, 1);
2078 /* Don't update_hiwater_rss(mm) here, do_exit already did */
2079 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2080 end
= unmap_vmas(&tlb
, vma
, 0, -1, &nr_accounted
, NULL
);
2081 vm_unacct_memory(nr_accounted
);
2082 free_pgtables(tlb
, vma
, FIRST_USER_ADDRESS
, 0);
2083 tlb_finish_mmu(tlb
, 0, end
);
2086 * Walk the list again, actually closing and freeing it,
2087 * with preemption enabled, without holding any MM locks.
2090 vma
= remove_vma(vma
);
2092 BUG_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2095 /* Insert vm structure into process list sorted by address
2096 * and into the inode's i_mmap tree. If vm_file is non-NULL
2097 * then i_mmap_lock is taken here.
2099 int insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
2101 struct vm_area_struct
* __vma
, * prev
;
2102 struct rb_node
** rb_link
, * rb_parent
;
2105 * The vm_pgoff of a purely anonymous vma should be irrelevant
2106 * until its first write fault, when page's anon_vma and index
2107 * are set. But now set the vm_pgoff it will almost certainly
2108 * end up with (unless mremap moves it elsewhere before that
2109 * first wfault), so /proc/pid/maps tells a consistent story.
2111 * By setting it to reflect the virtual start address of the
2112 * vma, merges and splits can happen in a seamless way, just
2113 * using the existing file pgoff checks and manipulations.
2114 * Similarly in do_mmap_pgoff and in do_brk.
2116 if (!vma
->vm_file
) {
2117 BUG_ON(vma
->anon_vma
);
2118 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2120 __vma
= find_vma_prepare(mm
,vma
->vm_start
,&prev
,&rb_link
,&rb_parent
);
2121 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
2123 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2124 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2126 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2131 * Copy the vma structure to a new location in the same mm,
2132 * prior to moving page table entries, to effect an mremap move.
2134 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2135 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
2137 struct vm_area_struct
*vma
= *vmap
;
2138 unsigned long vma_start
= vma
->vm_start
;
2139 struct mm_struct
*mm
= vma
->vm_mm
;
2140 struct vm_area_struct
*new_vma
, *prev
;
2141 struct rb_node
**rb_link
, *rb_parent
;
2142 struct mempolicy
*pol
;
2145 * If anonymous vma has not yet been faulted, update new pgoff
2146 * to match new location, to increase its chance of merging.
2148 if (!vma
->vm_file
&& !vma
->anon_vma
)
2149 pgoff
= addr
>> PAGE_SHIFT
;
2151 find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2152 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2153 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2156 * Source vma may have been merged into new_vma
2158 if (vma_start
>= new_vma
->vm_start
&&
2159 vma_start
< new_vma
->vm_end
)
2162 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2165 pol
= mpol_dup(vma_policy(vma
));
2167 kmem_cache_free(vm_area_cachep
, new_vma
);
2170 vma_set_policy(new_vma
, pol
);
2171 new_vma
->vm_start
= addr
;
2172 new_vma
->vm_end
= addr
+ len
;
2173 new_vma
->vm_pgoff
= pgoff
;
2174 if (new_vma
->vm_file
) {
2175 get_file(new_vma
->vm_file
);
2176 if (vma
->vm_flags
& VM_EXECUTABLE
)
2177 added_exe_file_vma(mm
);
2179 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2180 new_vma
->vm_ops
->open(new_vma
);
2181 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2188 * Return true if the calling process may expand its vm space by the passed
2191 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2193 unsigned long cur
= mm
->total_vm
; /* pages */
2196 lim
= current
->signal
->rlim
[RLIMIT_AS
].rlim_cur
>> PAGE_SHIFT
;
2198 if (cur
+ npages
> lim
)
2204 static int special_mapping_fault(struct vm_area_struct
*vma
,
2205 struct vm_fault
*vmf
)
2208 struct page
**pages
;
2211 * special mappings have no vm_file, and in that case, the mm
2212 * uses vm_pgoff internally. So we have to subtract it from here.
2213 * We are allowed to do this because we are the mm; do not copy
2214 * this code into drivers!
2216 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2218 for (pages
= vma
->vm_private_data
; pgoff
&& *pages
; ++pages
)
2222 struct page
*page
= *pages
;
2228 return VM_FAULT_SIGBUS
;
2232 * Having a close hook prevents vma merging regardless of flags.
2234 static void special_mapping_close(struct vm_area_struct
*vma
)
2238 static struct vm_operations_struct special_mapping_vmops
= {
2239 .close
= special_mapping_close
,
2240 .fault
= special_mapping_fault
,
2244 * Called with mm->mmap_sem held for writing.
2245 * Insert a new vma covering the given region, with the given flags.
2246 * Its pages are supplied by the given array of struct page *.
2247 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2248 * The region past the last page supplied will always produce SIGBUS.
2249 * The array pointer and the pages it points to are assumed to stay alive
2250 * for as long as this mapping might exist.
2252 int install_special_mapping(struct mm_struct
*mm
,
2253 unsigned long addr
, unsigned long len
,
2254 unsigned long vm_flags
, struct page
**pages
)
2257 struct vm_area_struct
*vma
;
2259 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2260 if (unlikely(vma
== NULL
))
2264 vma
->vm_start
= addr
;
2265 vma
->vm_end
= addr
+ len
;
2267 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
;
2268 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
2270 vma
->vm_ops
= &special_mapping_vmops
;
2271 vma
->vm_private_data
= pages
;
2273 ret
= security_file_mmap(NULL
, 0, 0, 0, vma
->vm_start
, 1);
2277 ret
= insert_vm_struct(mm
, vma
);
2281 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2286 kmem_cache_free(vm_area_cachep
, vma
);
2290 static DEFINE_MUTEX(mm_all_locks_mutex
);
2292 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
2294 if (!test_bit(0, (unsigned long *) &anon_vma
->head
.next
)) {
2296 * The LSB of head.next can't change from under us
2297 * because we hold the mm_all_locks_mutex.
2299 spin_lock_nest_lock(&anon_vma
->lock
, &mm
->mmap_sem
);
2301 * We can safely modify head.next after taking the
2302 * anon_vma->lock. If some other vma in this mm shares
2303 * the same anon_vma we won't take it again.
2305 * No need of atomic instructions here, head.next
2306 * can't change from under us thanks to the
2309 if (__test_and_set_bit(0, (unsigned long *)
2310 &anon_vma
->head
.next
))
2315 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
2317 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2319 * AS_MM_ALL_LOCKS can't change from under us because
2320 * we hold the mm_all_locks_mutex.
2322 * Operations on ->flags have to be atomic because
2323 * even if AS_MM_ALL_LOCKS is stable thanks to the
2324 * mm_all_locks_mutex, there may be other cpus
2325 * changing other bitflags in parallel to us.
2327 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
2329 spin_lock_nest_lock(&mapping
->i_mmap_lock
, &mm
->mmap_sem
);
2334 * This operation locks against the VM for all pte/vma/mm related
2335 * operations that could ever happen on a certain mm. This includes
2336 * vmtruncate, try_to_unmap, and all page faults.
2338 * The caller must take the mmap_sem in write mode before calling
2339 * mm_take_all_locks(). The caller isn't allowed to release the
2340 * mmap_sem until mm_drop_all_locks() returns.
2342 * mmap_sem in write mode is required in order to block all operations
2343 * that could modify pagetables and free pages without need of
2344 * altering the vma layout (for example populate_range() with
2345 * nonlinear vmas). It's also needed in write mode to avoid new
2346 * anon_vmas to be associated with existing vmas.
2348 * A single task can't take more than one mm_take_all_locks() in a row
2349 * or it would deadlock.
2351 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2352 * mapping->flags avoid to take the same lock twice, if more than one
2353 * vma in this mm is backed by the same anon_vma or address_space.
2355 * We can take all the locks in random order because the VM code
2356 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2357 * takes more than one of them in a row. Secondly we're protected
2358 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2360 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2361 * that may have to take thousand of locks.
2363 * mm_take_all_locks() can fail if it's interrupted by signals.
2365 int mm_take_all_locks(struct mm_struct
*mm
)
2367 struct vm_area_struct
*vma
;
2370 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2372 mutex_lock(&mm_all_locks_mutex
);
2374 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2375 if (signal_pending(current
))
2377 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2378 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
2381 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2382 if (signal_pending(current
))
2385 vm_lock_anon_vma(mm
, vma
->anon_vma
);
2392 mm_drop_all_locks(mm
);
2397 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
2399 if (test_bit(0, (unsigned long *) &anon_vma
->head
.next
)) {
2401 * The LSB of head.next can't change to 0 from under
2402 * us because we hold the mm_all_locks_mutex.
2404 * We must however clear the bitflag before unlocking
2405 * the vma so the users using the anon_vma->head will
2406 * never see our bitflag.
2408 * No need of atomic instructions here, head.next
2409 * can't change from under us until we release the
2412 if (!__test_and_clear_bit(0, (unsigned long *)
2413 &anon_vma
->head
.next
))
2415 spin_unlock(&anon_vma
->lock
);
2419 static void vm_unlock_mapping(struct address_space
*mapping
)
2421 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2423 * AS_MM_ALL_LOCKS can't change to 0 from under us
2424 * because we hold the mm_all_locks_mutex.
2426 spin_unlock(&mapping
->i_mmap_lock
);
2427 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
2434 * The mmap_sem cannot be released by the caller until
2435 * mm_drop_all_locks() returns.
2437 void mm_drop_all_locks(struct mm_struct
*mm
)
2439 struct vm_area_struct
*vma
;
2441 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2442 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
2444 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2446 vm_unlock_anon_vma(vma
->anon_vma
);
2447 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2448 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
2451 mutex_unlock(&mm_all_locks_mutex
);