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 */
181 allowed
-= mm
->total_vm
/ 32;
184 * cast `allowed' as a signed long because vm_committed_space
185 * sometimes has a negative value
187 if (atomic_long_read(&vm_committed_space
) < (long)allowed
)
190 vm_unacct_memory(pages
);
196 * Requires inode->i_mapping->i_mmap_lock
198 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
199 struct file
*file
, struct address_space
*mapping
)
201 if (vma
->vm_flags
& VM_DENYWRITE
)
202 atomic_inc(&file
->f_path
.dentry
->d_inode
->i_writecount
);
203 if (vma
->vm_flags
& VM_SHARED
)
204 mapping
->i_mmap_writable
--;
206 flush_dcache_mmap_lock(mapping
);
207 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
208 list_del_init(&vma
->shared
.vm_set
.list
);
210 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
211 flush_dcache_mmap_unlock(mapping
);
215 * Unlink a file-based vm structure from its prio_tree, to hide
216 * vma from rmap and vmtruncate before freeing its page tables.
218 void unlink_file_vma(struct vm_area_struct
*vma
)
220 struct file
*file
= vma
->vm_file
;
223 struct address_space
*mapping
= file
->f_mapping
;
224 spin_lock(&mapping
->i_mmap_lock
);
225 __remove_shared_vm_struct(vma
, file
, mapping
);
226 spin_unlock(&mapping
->i_mmap_lock
);
231 * Close a vm structure and free it, returning the next.
233 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
235 struct vm_area_struct
*next
= vma
->vm_next
;
238 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
239 vma
->vm_ops
->close(vma
);
242 if (vma
->vm_flags
& VM_EXECUTABLE
)
243 removed_exe_file_vma(vma
->vm_mm
);
245 mpol_put(vma_policy(vma
));
246 kmem_cache_free(vm_area_cachep
, vma
);
250 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
252 unsigned long rlim
, retval
;
253 unsigned long newbrk
, oldbrk
;
254 struct mm_struct
*mm
= current
->mm
;
255 unsigned long min_brk
;
257 down_write(&mm
->mmap_sem
);
259 #ifdef CONFIG_COMPAT_BRK
260 min_brk
= mm
->end_code
;
262 min_brk
= mm
->start_brk
;
268 * Check against rlimit here. If this check is done later after the test
269 * of oldbrk with newbrk then it can escape the test and let the data
270 * segment grow beyond its set limit the in case where the limit is
271 * not page aligned -Ram Gupta
273 rlim
= current
->signal
->rlim
[RLIMIT_DATA
].rlim_cur
;
274 if (rlim
< RLIM_INFINITY
&& (brk
- mm
->start_brk
) +
275 (mm
->end_data
- mm
->start_data
) > rlim
)
278 newbrk
= PAGE_ALIGN(brk
);
279 oldbrk
= PAGE_ALIGN(mm
->brk
);
280 if (oldbrk
== newbrk
)
283 /* Always allow shrinking brk. */
284 if (brk
<= mm
->brk
) {
285 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
290 /* Check against existing mmap mappings. */
291 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
294 /* Ok, looks good - let it rip. */
295 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
301 up_write(&mm
->mmap_sem
);
306 static int browse_rb(struct rb_root
*root
)
309 struct rb_node
*nd
, *pn
= NULL
;
310 unsigned long prev
= 0, pend
= 0;
312 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
313 struct vm_area_struct
*vma
;
314 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
315 if (vma
->vm_start
< prev
)
316 printk("vm_start %lx prev %lx\n", vma
->vm_start
, prev
), i
= -1;
317 if (vma
->vm_start
< pend
)
318 printk("vm_start %lx pend %lx\n", vma
->vm_start
, pend
);
319 if (vma
->vm_start
> vma
->vm_end
)
320 printk("vm_end %lx < vm_start %lx\n", vma
->vm_end
, vma
->vm_start
);
323 prev
= vma
->vm_start
;
327 for (nd
= pn
; nd
; nd
= rb_prev(nd
)) {
331 printk("backwards %d, forwards %d\n", j
, i
), i
= 0;
335 void validate_mm(struct mm_struct
*mm
)
339 struct vm_area_struct
*tmp
= mm
->mmap
;
344 if (i
!= mm
->map_count
)
345 printk("map_count %d vm_next %d\n", mm
->map_count
, i
), bug
= 1;
346 i
= browse_rb(&mm
->mm_rb
);
347 if (i
!= mm
->map_count
)
348 printk("map_count %d rb %d\n", mm
->map_count
, i
), bug
= 1;
352 #define validate_mm(mm) do { } while (0)
355 static struct vm_area_struct
*
356 find_vma_prepare(struct mm_struct
*mm
, unsigned long addr
,
357 struct vm_area_struct
**pprev
, struct rb_node
***rb_link
,
358 struct rb_node
** rb_parent
)
360 struct vm_area_struct
* vma
;
361 struct rb_node
** __rb_link
, * __rb_parent
, * rb_prev
;
363 __rb_link
= &mm
->mm_rb
.rb_node
;
364 rb_prev
= __rb_parent
= NULL
;
368 struct vm_area_struct
*vma_tmp
;
370 __rb_parent
= *__rb_link
;
371 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
373 if (vma_tmp
->vm_end
> addr
) {
375 if (vma_tmp
->vm_start
<= addr
)
377 __rb_link
= &__rb_parent
->rb_left
;
379 rb_prev
= __rb_parent
;
380 __rb_link
= &__rb_parent
->rb_right
;
386 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
387 *rb_link
= __rb_link
;
388 *rb_parent
= __rb_parent
;
393 __vma_link_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
394 struct vm_area_struct
*prev
, struct rb_node
*rb_parent
)
397 vma
->vm_next
= prev
->vm_next
;
402 vma
->vm_next
= rb_entry(rb_parent
,
403 struct vm_area_struct
, vm_rb
);
409 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
410 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
412 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
413 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
416 static inline void __vma_link_file(struct vm_area_struct
*vma
)
422 struct address_space
*mapping
= file
->f_mapping
;
424 if (vma
->vm_flags
& VM_DENYWRITE
)
425 atomic_dec(&file
->f_path
.dentry
->d_inode
->i_writecount
);
426 if (vma
->vm_flags
& VM_SHARED
)
427 mapping
->i_mmap_writable
++;
429 flush_dcache_mmap_lock(mapping
);
430 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
431 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
433 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
434 flush_dcache_mmap_unlock(mapping
);
439 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
440 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
441 struct rb_node
*rb_parent
)
443 __vma_link_list(mm
, vma
, prev
, rb_parent
);
444 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
445 __anon_vma_link(vma
);
448 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
449 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
450 struct rb_node
*rb_parent
)
452 struct address_space
*mapping
= NULL
;
455 mapping
= vma
->vm_file
->f_mapping
;
458 spin_lock(&mapping
->i_mmap_lock
);
459 vma
->vm_truncate_count
= mapping
->truncate_count
;
463 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
464 __vma_link_file(vma
);
466 anon_vma_unlock(vma
);
468 spin_unlock(&mapping
->i_mmap_lock
);
475 * Helper for vma_adjust in the split_vma insert case:
476 * insert vm structure into list and rbtree and anon_vma,
477 * but it has already been inserted into prio_tree earlier.
480 __insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
482 struct vm_area_struct
* __vma
, * prev
;
483 struct rb_node
** rb_link
, * rb_parent
;
485 __vma
= find_vma_prepare(mm
, vma
->vm_start
,&prev
, &rb_link
, &rb_parent
);
486 BUG_ON(__vma
&& __vma
->vm_start
< vma
->vm_end
);
487 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
492 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
493 struct vm_area_struct
*prev
)
495 prev
->vm_next
= vma
->vm_next
;
496 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
497 if (mm
->mmap_cache
== vma
)
498 mm
->mmap_cache
= prev
;
502 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
503 * is already present in an i_mmap tree without adjusting the tree.
504 * The following helper function should be used when such adjustments
505 * are necessary. The "insert" vma (if any) is to be inserted
506 * before we drop the necessary locks.
508 void vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
509 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
511 struct mm_struct
*mm
= vma
->vm_mm
;
512 struct vm_area_struct
*next
= vma
->vm_next
;
513 struct vm_area_struct
*importer
= NULL
;
514 struct address_space
*mapping
= NULL
;
515 struct prio_tree_root
*root
= NULL
;
516 struct file
*file
= vma
->vm_file
;
517 struct anon_vma
*anon_vma
= NULL
;
518 long adjust_next
= 0;
521 if (next
&& !insert
) {
522 if (end
>= next
->vm_end
) {
524 * vma expands, overlapping all the next, and
525 * perhaps the one after too (mprotect case 6).
527 again
: remove_next
= 1 + (end
> next
->vm_end
);
529 anon_vma
= next
->anon_vma
;
531 } else if (end
> next
->vm_start
) {
533 * vma expands, overlapping part of the next:
534 * mprotect case 5 shifting the boundary up.
536 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
537 anon_vma
= next
->anon_vma
;
539 } else if (end
< vma
->vm_end
) {
541 * vma shrinks, and !insert tells it's not
542 * split_vma inserting another: so it must be
543 * mprotect case 4 shifting the boundary down.
545 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
546 anon_vma
= next
->anon_vma
;
552 mapping
= file
->f_mapping
;
553 if (!(vma
->vm_flags
& VM_NONLINEAR
))
554 root
= &mapping
->i_mmap
;
555 spin_lock(&mapping
->i_mmap_lock
);
557 vma
->vm_truncate_count
!= next
->vm_truncate_count
) {
559 * unmap_mapping_range might be in progress:
560 * ensure that the expanding vma is rescanned.
562 importer
->vm_truncate_count
= 0;
565 insert
->vm_truncate_count
= vma
->vm_truncate_count
;
567 * Put into prio_tree now, so instantiated pages
568 * are visible to arm/parisc __flush_dcache_page
569 * throughout; but we cannot insert into address
570 * space until vma start or end is updated.
572 __vma_link_file(insert
);
577 * When changing only vma->vm_end, we don't really need
578 * anon_vma lock: but is that case worth optimizing out?
581 anon_vma
= vma
->anon_vma
;
583 spin_lock(&anon_vma
->lock
);
585 * Easily overlooked: when mprotect shifts the boundary,
586 * make sure the expanding vma has anon_vma set if the
587 * shrinking vma had, to cover any anon pages imported.
589 if (importer
&& !importer
->anon_vma
) {
590 importer
->anon_vma
= anon_vma
;
591 __anon_vma_link(importer
);
596 flush_dcache_mmap_lock(mapping
);
597 vma_prio_tree_remove(vma
, root
);
599 vma_prio_tree_remove(next
, root
);
602 vma
->vm_start
= start
;
604 vma
->vm_pgoff
= pgoff
;
606 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
607 next
->vm_pgoff
+= adjust_next
;
612 vma_prio_tree_insert(next
, root
);
613 vma_prio_tree_insert(vma
, root
);
614 flush_dcache_mmap_unlock(mapping
);
619 * vma_merge has merged next into vma, and needs
620 * us to remove next before dropping the locks.
622 __vma_unlink(mm
, next
, vma
);
624 __remove_shared_vm_struct(next
, file
, mapping
);
626 __anon_vma_merge(vma
, next
);
629 * split_vma has split insert from vma, and needs
630 * us to insert it before dropping the locks
631 * (it may either follow vma or precede it).
633 __insert_vm_struct(mm
, insert
);
637 spin_unlock(&anon_vma
->lock
);
639 spin_unlock(&mapping
->i_mmap_lock
);
644 if (next
->vm_flags
& VM_EXECUTABLE
)
645 removed_exe_file_vma(mm
);
648 mpol_put(vma_policy(next
));
649 kmem_cache_free(vm_area_cachep
, next
);
651 * In mprotect's case 6 (see comments on vma_merge),
652 * we must remove another next too. It would clutter
653 * up the code too much to do both in one go.
655 if (remove_next
== 2) {
665 * If the vma has a ->close operation then the driver probably needs to release
666 * per-vma resources, so we don't attempt to merge those.
668 #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP)
670 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
671 struct file
*file
, unsigned long vm_flags
)
673 if (vma
->vm_flags
!= vm_flags
)
675 if (vma
->vm_file
!= file
)
677 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
682 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
683 struct anon_vma
*anon_vma2
)
685 return !anon_vma1
|| !anon_vma2
|| (anon_vma1
== anon_vma2
);
689 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
690 * in front of (at a lower virtual address and file offset than) the vma.
692 * We cannot merge two vmas if they have differently assigned (non-NULL)
693 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
695 * We don't check here for the merged mmap wrapping around the end of pagecache
696 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
697 * wrap, nor mmaps which cover the final page at index -1UL.
700 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
701 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
703 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
704 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
705 if (vma
->vm_pgoff
== vm_pgoff
)
712 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
713 * beyond (at a higher virtual address and file offset than) the vma.
715 * We cannot merge two vmas if they have differently assigned (non-NULL)
716 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
719 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
720 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
722 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
723 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
725 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
726 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
733 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
734 * whether that can be merged with its predecessor or its successor.
735 * Or both (it neatly fills a hole).
737 * In most cases - when called for mmap, brk or mremap - [addr,end) is
738 * certain not to be mapped by the time vma_merge is called; but when
739 * called for mprotect, it is certain to be already mapped (either at
740 * an offset within prev, or at the start of next), and the flags of
741 * this area are about to be changed to vm_flags - and the no-change
742 * case has already been eliminated.
744 * The following mprotect cases have to be considered, where AAAA is
745 * the area passed down from mprotect_fixup, never extending beyond one
746 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
748 * AAAA AAAA AAAA AAAA
749 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
750 * cannot merge might become might become might become
751 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
752 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
753 * mremap move: PPPPNNNNNNNN 8
755 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
756 * might become case 1 below case 2 below case 3 below
758 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
759 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
761 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
762 struct vm_area_struct
*prev
, unsigned long addr
,
763 unsigned long end
, unsigned long vm_flags
,
764 struct anon_vma
*anon_vma
, struct file
*file
,
765 pgoff_t pgoff
, struct mempolicy
*policy
)
767 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
768 struct vm_area_struct
*area
, *next
;
771 * We later require that vma->vm_flags == vm_flags,
772 * so this tests vma->vm_flags & VM_SPECIAL, too.
774 if (vm_flags
& VM_SPECIAL
)
778 next
= prev
->vm_next
;
782 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
783 next
= next
->vm_next
;
786 * Can it merge with the predecessor?
788 if (prev
&& prev
->vm_end
== addr
&&
789 mpol_equal(vma_policy(prev
), policy
) &&
790 can_vma_merge_after(prev
, vm_flags
,
791 anon_vma
, file
, pgoff
)) {
793 * OK, it can. Can we now merge in the successor as well?
795 if (next
&& end
== next
->vm_start
&&
796 mpol_equal(policy
, vma_policy(next
)) &&
797 can_vma_merge_before(next
, vm_flags
,
798 anon_vma
, file
, pgoff
+pglen
) &&
799 is_mergeable_anon_vma(prev
->anon_vma
,
802 vma_adjust(prev
, prev
->vm_start
,
803 next
->vm_end
, prev
->vm_pgoff
, NULL
);
804 } else /* cases 2, 5, 7 */
805 vma_adjust(prev
, prev
->vm_start
,
806 end
, prev
->vm_pgoff
, NULL
);
811 * Can this new request be merged in front of next?
813 if (next
&& end
== next
->vm_start
&&
814 mpol_equal(policy
, vma_policy(next
)) &&
815 can_vma_merge_before(next
, vm_flags
,
816 anon_vma
, file
, pgoff
+pglen
)) {
817 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
818 vma_adjust(prev
, prev
->vm_start
,
819 addr
, prev
->vm_pgoff
, NULL
);
820 else /* cases 3, 8 */
821 vma_adjust(area
, addr
, next
->vm_end
,
822 next
->vm_pgoff
- pglen
, NULL
);
830 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
831 * neighbouring vmas for a suitable anon_vma, before it goes off
832 * to allocate a new anon_vma. It checks because a repetitive
833 * sequence of mprotects and faults may otherwise lead to distinct
834 * anon_vmas being allocated, preventing vma merge in subsequent
837 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
839 struct vm_area_struct
*near
;
840 unsigned long vm_flags
;
847 * Since only mprotect tries to remerge vmas, match flags
848 * which might be mprotected into each other later on.
849 * Neither mlock nor madvise tries to remerge at present,
850 * so leave their flags as obstructing a merge.
852 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
853 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
855 if (near
->anon_vma
&& vma
->vm_end
== near
->vm_start
&&
856 mpol_equal(vma_policy(vma
), vma_policy(near
)) &&
857 can_vma_merge_before(near
, vm_flags
,
858 NULL
, vma
->vm_file
, vma
->vm_pgoff
+
859 ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
)))
860 return near
->anon_vma
;
863 * It is potentially slow to have to call find_vma_prev here.
864 * But it's only on the first write fault on the vma, not
865 * every time, and we could devise a way to avoid it later
866 * (e.g. stash info in next's anon_vma_node when assigning
867 * an anon_vma, or when trying vma_merge). Another time.
869 BUG_ON(find_vma_prev(vma
->vm_mm
, vma
->vm_start
, &near
) != vma
);
873 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
874 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
876 if (near
->anon_vma
&& near
->vm_end
== vma
->vm_start
&&
877 mpol_equal(vma_policy(near
), vma_policy(vma
)) &&
878 can_vma_merge_after(near
, vm_flags
,
879 NULL
, vma
->vm_file
, vma
->vm_pgoff
))
880 return near
->anon_vma
;
883 * There's no absolute need to look only at touching neighbours:
884 * we could search further afield for "compatible" anon_vmas.
885 * But it would probably just be a waste of time searching,
886 * or lead to too many vmas hanging off the same anon_vma.
887 * We're trying to allow mprotect remerging later on,
888 * not trying to minimize memory used for anon_vmas.
893 #ifdef CONFIG_PROC_FS
894 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
895 struct file
*file
, long pages
)
897 const unsigned long stack_flags
898 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
901 mm
->shared_vm
+= pages
;
902 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
903 mm
->exec_vm
+= pages
;
904 } else if (flags
& stack_flags
)
905 mm
->stack_vm
+= pages
;
906 if (flags
& (VM_RESERVED
|VM_IO
))
907 mm
->reserved_vm
+= pages
;
909 #endif /* CONFIG_PROC_FS */
912 * The caller must hold down_write(current->mm->mmap_sem).
915 unsigned long do_mmap_pgoff(struct file
* file
, unsigned long addr
,
916 unsigned long len
, unsigned long prot
,
917 unsigned long flags
, unsigned long pgoff
)
919 struct mm_struct
* mm
= current
->mm
;
921 unsigned int vm_flags
;
924 unsigned long reqprot
= prot
;
927 * Does the application expect PROT_READ to imply PROT_EXEC?
929 * (the exception is when the underlying filesystem is noexec
930 * mounted, in which case we dont add PROT_EXEC.)
932 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
933 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
939 if (!(flags
& MAP_FIXED
))
940 addr
= round_hint_to_min(addr
);
942 error
= arch_mmap_check(addr
, len
, flags
);
946 /* Careful about overflows.. */
947 len
= PAGE_ALIGN(len
);
948 if (!len
|| len
> TASK_SIZE
)
951 /* offset overflow? */
952 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
955 /* Too many mappings? */
956 if (mm
->map_count
> sysctl_max_map_count
)
959 /* Obtain the address to map to. we verify (or select) it and ensure
960 * that it represents a valid section of the address space.
962 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
963 if (addr
& ~PAGE_MASK
)
966 /* Do simple checking here so the lower-level routines won't have
967 * to. we assume access permissions have been handled by the open
968 * of the memory object, so we don't do any here.
970 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
971 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
973 if (flags
& MAP_LOCKED
) {
976 vm_flags
|= VM_LOCKED
;
978 /* mlock MCL_FUTURE? */
979 if (vm_flags
& VM_LOCKED
) {
980 unsigned long locked
, lock_limit
;
981 locked
= len
>> PAGE_SHIFT
;
982 locked
+= mm
->locked_vm
;
983 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
984 lock_limit
>>= PAGE_SHIFT
;
985 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
989 inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
992 switch (flags
& MAP_TYPE
) {
994 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
998 * Make sure we don't allow writing to an append-only
1001 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1005 * Make sure there are no mandatory locks on the file.
1007 if (locks_verify_locked(inode
))
1010 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1011 if (!(file
->f_mode
& FMODE_WRITE
))
1012 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1016 if (!(file
->f_mode
& FMODE_READ
))
1018 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1019 if (vm_flags
& VM_EXEC
)
1021 vm_flags
&= ~VM_MAYEXEC
;
1023 if (is_file_hugepages(file
))
1026 if (!file
->f_op
|| !file
->f_op
->mmap
)
1034 switch (flags
& MAP_TYPE
) {
1040 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1044 * Set pgoff according to addr for anon_vma.
1046 pgoff
= addr
>> PAGE_SHIFT
;
1053 error
= security_file_mmap(file
, reqprot
, prot
, flags
, addr
, 0);
1057 return mmap_region(file
, addr
, len
, flags
, vm_flags
, pgoff
,
1060 EXPORT_SYMBOL(do_mmap_pgoff
);
1063 * Some shared mappigns will want the pages marked read-only
1064 * to track write events. If so, we'll downgrade vm_page_prot
1065 * to the private version (using protection_map[] without the
1068 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1070 unsigned int vm_flags
= vma
->vm_flags
;
1072 /* If it was private or non-writable, the write bit is already clear */
1073 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1076 /* The backer wishes to know when pages are first written to? */
1077 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1080 /* The open routine did something to the protections already? */
1081 if (pgprot_val(vma
->vm_page_prot
) !=
1082 pgprot_val(vm_get_page_prot(vm_flags
)))
1085 /* Specialty mapping? */
1086 if (vm_flags
& (VM_PFNMAP
|VM_INSERTPAGE
))
1089 /* Can the mapping track the dirty pages? */
1090 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1091 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1094 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1095 unsigned long len
, unsigned long flags
,
1096 unsigned int vm_flags
, unsigned long pgoff
,
1099 struct mm_struct
*mm
= current
->mm
;
1100 struct vm_area_struct
*vma
, *prev
;
1101 int correct_wcount
= 0;
1103 struct rb_node
**rb_link
, *rb_parent
;
1104 unsigned long charged
= 0;
1105 struct inode
*inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1107 /* Clear old maps */
1110 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1111 if (vma
&& vma
->vm_start
< addr
+ len
) {
1112 if (do_munmap(mm
, addr
, len
))
1117 /* Check against address space limit. */
1118 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1121 if (flags
& MAP_NORESERVE
)
1122 vm_flags
|= VM_NORESERVE
;
1124 if (accountable
&& (!(flags
& MAP_NORESERVE
) ||
1125 sysctl_overcommit_memory
== OVERCOMMIT_NEVER
)) {
1126 if (vm_flags
& VM_SHARED
) {
1127 /* Check memory availability in shmem_file_setup? */
1128 vm_flags
|= VM_ACCOUNT
;
1129 } else if (vm_flags
& VM_WRITE
) {
1131 * Private writable mapping: check memory availability
1133 charged
= len
>> PAGE_SHIFT
;
1134 if (security_vm_enough_memory(charged
))
1136 vm_flags
|= VM_ACCOUNT
;
1141 * Can we just expand an old private anonymous mapping?
1142 * The VM_SHARED test is necessary because shmem_zero_setup
1143 * will create the file object for a shared anonymous map below.
1145 if (!file
&& !(vm_flags
& VM_SHARED
) &&
1146 vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
,
1147 NULL
, NULL
, pgoff
, NULL
))
1151 * Determine the object being mapped and call the appropriate
1152 * specific mapper. the address has already been validated, but
1153 * not unmapped, but the maps are removed from the list.
1155 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1162 vma
->vm_start
= addr
;
1163 vma
->vm_end
= addr
+ len
;
1164 vma
->vm_flags
= vm_flags
;
1165 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1166 vma
->vm_pgoff
= pgoff
;
1170 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1172 if (vm_flags
& VM_DENYWRITE
) {
1173 error
= deny_write_access(file
);
1178 vma
->vm_file
= file
;
1180 error
= file
->f_op
->mmap(file
, vma
);
1182 goto unmap_and_free_vma
;
1183 if (vm_flags
& VM_EXECUTABLE
)
1184 added_exe_file_vma(mm
);
1185 } else if (vm_flags
& VM_SHARED
) {
1186 error
= shmem_zero_setup(vma
);
1191 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform
1192 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap)
1193 * that memory reservation must be checked; but that reservation
1194 * belongs to shared memory object, not to vma: so now clear it.
1196 if ((vm_flags
& (VM_SHARED
|VM_ACCOUNT
)) == (VM_SHARED
|VM_ACCOUNT
))
1197 vma
->vm_flags
&= ~VM_ACCOUNT
;
1199 /* Can addr have changed??
1201 * Answer: Yes, several device drivers can do it in their
1202 * f_op->mmap method. -DaveM
1204 addr
= vma
->vm_start
;
1205 pgoff
= vma
->vm_pgoff
;
1206 vm_flags
= vma
->vm_flags
;
1208 if (vma_wants_writenotify(vma
))
1209 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
& ~VM_SHARED
);
1211 if (file
&& vma_merge(mm
, prev
, addr
, vma
->vm_end
,
1212 vma
->vm_flags
, NULL
, file
, pgoff
, vma_policy(vma
))) {
1213 mpol_put(vma_policy(vma
));
1214 kmem_cache_free(vm_area_cachep
, vma
);
1216 if (vm_flags
& VM_EXECUTABLE
)
1217 removed_exe_file_vma(mm
);
1219 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1220 file
= vma
->vm_file
;
1223 /* Once vma denies write, undo our temporary denial count */
1225 atomic_inc(&inode
->i_writecount
);
1227 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1228 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1229 if (vm_flags
& VM_LOCKED
) {
1230 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
1231 make_pages_present(addr
, addr
+ len
);
1233 if ((flags
& MAP_POPULATE
) && !(flags
& MAP_NONBLOCK
))
1234 make_pages_present(addr
, addr
+ len
);
1239 atomic_inc(&inode
->i_writecount
);
1240 vma
->vm_file
= NULL
;
1243 /* Undo any partial mapping done by a device driver. */
1244 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1247 kmem_cache_free(vm_area_cachep
, vma
);
1250 vm_unacct_memory(charged
);
1254 /* Get an address range which is currently unmapped.
1255 * For shmat() with addr=0.
1257 * Ugly calling convention alert:
1258 * Return value with the low bits set means error value,
1260 * if (ret & ~PAGE_MASK)
1263 * This function "knows" that -ENOMEM has the bits set.
1265 #ifndef HAVE_ARCH_UNMAPPED_AREA
1267 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1268 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1270 struct mm_struct
*mm
= current
->mm
;
1271 struct vm_area_struct
*vma
;
1272 unsigned long start_addr
;
1274 if (len
> TASK_SIZE
)
1277 if (flags
& MAP_FIXED
)
1281 addr
= PAGE_ALIGN(addr
);
1282 vma
= find_vma(mm
, addr
);
1283 if (TASK_SIZE
- len
>= addr
&&
1284 (!vma
|| addr
+ len
<= vma
->vm_start
))
1287 if (len
> mm
->cached_hole_size
) {
1288 start_addr
= addr
= mm
->free_area_cache
;
1290 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1291 mm
->cached_hole_size
= 0;
1295 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1296 /* At this point: (!vma || addr < vma->vm_end). */
1297 if (TASK_SIZE
- len
< addr
) {
1299 * Start a new search - just in case we missed
1302 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1303 addr
= TASK_UNMAPPED_BASE
;
1305 mm
->cached_hole_size
= 0;
1310 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1312 * Remember the place where we stopped the search:
1314 mm
->free_area_cache
= addr
+ len
;
1317 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1318 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1324 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1327 * Is this a new hole at the lowest possible address?
1329 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
) {
1330 mm
->free_area_cache
= addr
;
1331 mm
->cached_hole_size
= ~0UL;
1336 * This mmap-allocator allocates new areas top-down from below the
1337 * stack's low limit (the base):
1339 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1341 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1342 const unsigned long len
, const unsigned long pgoff
,
1343 const unsigned long flags
)
1345 struct vm_area_struct
*vma
;
1346 struct mm_struct
*mm
= current
->mm
;
1347 unsigned long addr
= addr0
;
1349 /* requested length too big for entire address space */
1350 if (len
> TASK_SIZE
)
1353 if (flags
& MAP_FIXED
)
1356 /* requesting a specific address */
1358 addr
= PAGE_ALIGN(addr
);
1359 vma
= find_vma(mm
, addr
);
1360 if (TASK_SIZE
- len
>= addr
&&
1361 (!vma
|| addr
+ len
<= vma
->vm_start
))
1365 /* check if free_area_cache is useful for us */
1366 if (len
<= mm
->cached_hole_size
) {
1367 mm
->cached_hole_size
= 0;
1368 mm
->free_area_cache
= mm
->mmap_base
;
1371 /* either no address requested or can't fit in requested address hole */
1372 addr
= mm
->free_area_cache
;
1374 /* make sure it can fit in the remaining address space */
1376 vma
= find_vma(mm
, addr
-len
);
1377 if (!vma
|| addr
<= vma
->vm_start
)
1378 /* remember the address as a hint for next time */
1379 return (mm
->free_area_cache
= addr
-len
);
1382 if (mm
->mmap_base
< len
)
1385 addr
= mm
->mmap_base
-len
;
1389 * Lookup failure means no vma is above this address,
1390 * else if new region fits below vma->vm_start,
1391 * return with success:
1393 vma
= find_vma(mm
, addr
);
1394 if (!vma
|| addr
+len
<= vma
->vm_start
)
1395 /* remember the address as a hint for next time */
1396 return (mm
->free_area_cache
= addr
);
1398 /* remember the largest hole we saw so far */
1399 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1400 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1402 /* try just below the current vma->vm_start */
1403 addr
= vma
->vm_start
-len
;
1404 } while (len
< vma
->vm_start
);
1408 * A failed mmap() very likely causes application failure,
1409 * so fall back to the bottom-up function here. This scenario
1410 * can happen with large stack limits and large mmap()
1413 mm
->cached_hole_size
= ~0UL;
1414 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1415 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1417 * Restore the topdown base:
1419 mm
->free_area_cache
= mm
->mmap_base
;
1420 mm
->cached_hole_size
= ~0UL;
1426 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1429 * Is this a new hole at the highest possible address?
1431 if (addr
> mm
->free_area_cache
)
1432 mm
->free_area_cache
= addr
;
1434 /* dont allow allocations above current base */
1435 if (mm
->free_area_cache
> mm
->mmap_base
)
1436 mm
->free_area_cache
= mm
->mmap_base
;
1440 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1441 unsigned long pgoff
, unsigned long flags
)
1443 unsigned long (*get_area
)(struct file
*, unsigned long,
1444 unsigned long, unsigned long, unsigned long);
1446 get_area
= current
->mm
->get_unmapped_area
;
1447 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1448 get_area
= file
->f_op
->get_unmapped_area
;
1449 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1450 if (IS_ERR_VALUE(addr
))
1453 if (addr
> TASK_SIZE
- len
)
1455 if (addr
& ~PAGE_MASK
)
1458 return arch_rebalance_pgtables(addr
, len
);
1461 EXPORT_SYMBOL(get_unmapped_area
);
1463 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1464 struct vm_area_struct
* find_vma(struct mm_struct
* mm
, unsigned long addr
)
1466 struct vm_area_struct
*vma
= NULL
;
1469 /* Check the cache first. */
1470 /* (Cache hit rate is typically around 35%.) */
1471 vma
= mm
->mmap_cache
;
1472 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1473 struct rb_node
* rb_node
;
1475 rb_node
= mm
->mm_rb
.rb_node
;
1479 struct vm_area_struct
* vma_tmp
;
1481 vma_tmp
= rb_entry(rb_node
,
1482 struct vm_area_struct
, vm_rb
);
1484 if (vma_tmp
->vm_end
> addr
) {
1486 if (vma_tmp
->vm_start
<= addr
)
1488 rb_node
= rb_node
->rb_left
;
1490 rb_node
= rb_node
->rb_right
;
1493 mm
->mmap_cache
= vma
;
1499 EXPORT_SYMBOL(find_vma
);
1501 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1502 struct vm_area_struct
*
1503 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1504 struct vm_area_struct
**pprev
)
1506 struct vm_area_struct
*vma
= NULL
, *prev
= NULL
;
1507 struct rb_node
* rb_node
;
1511 /* Guard against addr being lower than the first VMA */
1514 /* Go through the RB tree quickly. */
1515 rb_node
= mm
->mm_rb
.rb_node
;
1518 struct vm_area_struct
*vma_tmp
;
1519 vma_tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1521 if (addr
< vma_tmp
->vm_end
) {
1522 rb_node
= rb_node
->rb_left
;
1525 if (!prev
->vm_next
|| (addr
< prev
->vm_next
->vm_end
))
1527 rb_node
= rb_node
->rb_right
;
1533 return prev
? prev
->vm_next
: vma
;
1537 * Verify that the stack growth is acceptable and
1538 * update accounting. This is shared with both the
1539 * grow-up and grow-down cases.
1541 static int acct_stack_growth(struct vm_area_struct
* vma
, unsigned long size
, unsigned long grow
)
1543 struct mm_struct
*mm
= vma
->vm_mm
;
1544 struct rlimit
*rlim
= current
->signal
->rlim
;
1545 unsigned long new_start
;
1547 /* address space limit tests */
1548 if (!may_expand_vm(mm
, grow
))
1551 /* Stack limit test */
1552 if (size
> rlim
[RLIMIT_STACK
].rlim_cur
)
1555 /* mlock limit tests */
1556 if (vma
->vm_flags
& VM_LOCKED
) {
1557 unsigned long locked
;
1558 unsigned long limit
;
1559 locked
= mm
->locked_vm
+ grow
;
1560 limit
= rlim
[RLIMIT_MEMLOCK
].rlim_cur
>> PAGE_SHIFT
;
1561 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1565 /* Check to ensure the stack will not grow into a hugetlb-only region */
1566 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
1568 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
1572 * Overcommit.. This must be the final test, as it will
1573 * update security statistics.
1575 if (security_vm_enough_memory(grow
))
1578 /* Ok, everything looks good - let it rip */
1579 mm
->total_vm
+= grow
;
1580 if (vma
->vm_flags
& VM_LOCKED
)
1581 mm
->locked_vm
+= grow
;
1582 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1586 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1588 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1589 * vma is the last one with address > vma->vm_end. Have to extend vma.
1594 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1598 if (!(vma
->vm_flags
& VM_GROWSUP
))
1602 * We must make sure the anon_vma is allocated
1603 * so that the anon_vma locking is not a noop.
1605 if (unlikely(anon_vma_prepare(vma
)))
1610 * vma->vm_start/vm_end cannot change under us because the caller
1611 * is required to hold the mmap_sem in read mode. We need the
1612 * anon_vma lock to serialize against concurrent expand_stacks.
1613 * Also guard against wrapping around to address 0.
1615 if (address
< PAGE_ALIGN(address
+4))
1616 address
= PAGE_ALIGN(address
+4);
1618 anon_vma_unlock(vma
);
1623 /* Somebody else might have raced and expanded it already */
1624 if (address
> vma
->vm_end
) {
1625 unsigned long size
, grow
;
1627 size
= address
- vma
->vm_start
;
1628 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1630 error
= acct_stack_growth(vma
, size
, grow
);
1632 vma
->vm_end
= address
;
1634 anon_vma_unlock(vma
);
1637 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1640 * vma is the first one with address < vma->vm_start. Have to extend vma.
1642 static inline int expand_downwards(struct vm_area_struct
*vma
,
1643 unsigned long address
)
1648 * We must make sure the anon_vma is allocated
1649 * so that the anon_vma locking is not a noop.
1651 if (unlikely(anon_vma_prepare(vma
)))
1654 address
&= PAGE_MASK
;
1655 error
= security_file_mmap(NULL
, 0, 0, 0, address
, 1);
1662 * vma->vm_start/vm_end cannot change under us because the caller
1663 * is required to hold the mmap_sem in read mode. We need the
1664 * anon_vma lock to serialize against concurrent expand_stacks.
1667 /* Somebody else might have raced and expanded it already */
1668 if (address
< vma
->vm_start
) {
1669 unsigned long size
, grow
;
1671 size
= vma
->vm_end
- address
;
1672 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1674 error
= acct_stack_growth(vma
, size
, grow
);
1676 vma
->vm_start
= address
;
1677 vma
->vm_pgoff
-= grow
;
1680 anon_vma_unlock(vma
);
1684 int expand_stack_downwards(struct vm_area_struct
*vma
, unsigned long address
)
1686 return expand_downwards(vma
, address
);
1689 #ifdef CONFIG_STACK_GROWSUP
1690 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1692 return expand_upwards(vma
, address
);
1695 struct vm_area_struct
*
1696 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1698 struct vm_area_struct
*vma
, *prev
;
1701 vma
= find_vma_prev(mm
, addr
, &prev
);
1702 if (vma
&& (vma
->vm_start
<= addr
))
1704 if (!prev
|| expand_stack(prev
, addr
))
1706 if (prev
->vm_flags
& VM_LOCKED
)
1707 make_pages_present(addr
, prev
->vm_end
);
1711 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1713 return expand_downwards(vma
, address
);
1716 struct vm_area_struct
*
1717 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1719 struct vm_area_struct
* vma
;
1720 unsigned long start
;
1723 vma
= find_vma(mm
,addr
);
1726 if (vma
->vm_start
<= addr
)
1728 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1730 start
= vma
->vm_start
;
1731 if (expand_stack(vma
, addr
))
1733 if (vma
->vm_flags
& VM_LOCKED
)
1734 make_pages_present(addr
, start
);
1740 * Ok - we have the memory areas we should free on the vma list,
1741 * so release them, and do the vma updates.
1743 * Called with the mm semaphore held.
1745 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1747 /* Update high watermark before we lower total_vm */
1748 update_hiwater_vm(mm
);
1750 long nrpages
= vma_pages(vma
);
1752 mm
->total_vm
-= nrpages
;
1753 if (vma
->vm_flags
& VM_LOCKED
)
1754 mm
->locked_vm
-= nrpages
;
1755 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1756 vma
= remove_vma(vma
);
1762 * Get rid of page table information in the indicated region.
1764 * Called with the mm semaphore held.
1766 static void unmap_region(struct mm_struct
*mm
,
1767 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1768 unsigned long start
, unsigned long end
)
1770 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1771 struct mmu_gather
*tlb
;
1772 unsigned long nr_accounted
= 0;
1775 tlb
= tlb_gather_mmu(mm
, 0);
1776 update_hiwater_rss(mm
);
1777 unmap_vmas(&tlb
, vma
, start
, end
, &nr_accounted
, NULL
);
1778 vm_unacct_memory(nr_accounted
);
1779 free_pgtables(tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1780 next
? next
->vm_start
: 0);
1781 tlb_finish_mmu(tlb
, start
, end
);
1785 * Create a list of vma's touched by the unmap, removing them from the mm's
1786 * vma list as we go..
1789 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1790 struct vm_area_struct
*prev
, unsigned long end
)
1792 struct vm_area_struct
**insertion_point
;
1793 struct vm_area_struct
*tail_vma
= NULL
;
1796 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1798 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1802 } while (vma
&& vma
->vm_start
< end
);
1803 *insertion_point
= vma
;
1804 tail_vma
->vm_next
= NULL
;
1805 if (mm
->unmap_area
== arch_unmap_area
)
1806 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1808 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1809 mm
->unmap_area(mm
, addr
);
1810 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1814 * Split a vma into two pieces at address 'addr', a new vma is allocated
1815 * either for the first part or the tail.
1817 int split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1818 unsigned long addr
, int new_below
)
1820 struct mempolicy
*pol
;
1821 struct vm_area_struct
*new;
1823 if (is_vm_hugetlb_page(vma
) && (addr
&
1824 ~(huge_page_mask(hstate_vma(vma
)))))
1827 if (mm
->map_count
>= sysctl_max_map_count
)
1830 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1834 /* most fields are the same, copy all, and then fixup */
1840 new->vm_start
= addr
;
1841 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
1844 pol
= mpol_dup(vma_policy(vma
));
1846 kmem_cache_free(vm_area_cachep
, new);
1847 return PTR_ERR(pol
);
1849 vma_set_policy(new, pol
);
1852 get_file(new->vm_file
);
1853 if (vma
->vm_flags
& VM_EXECUTABLE
)
1854 added_exe_file_vma(mm
);
1857 if (new->vm_ops
&& new->vm_ops
->open
)
1858 new->vm_ops
->open(new);
1861 vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
1862 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
1864 vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
1869 /* Munmap is split into 2 main parts -- this part which finds
1870 * what needs doing, and the areas themselves, which do the
1871 * work. This now handles partial unmappings.
1872 * Jeremy Fitzhardinge <jeremy@goop.org>
1874 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1877 struct vm_area_struct
*vma
, *prev
, *last
;
1879 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
1882 if ((len
= PAGE_ALIGN(len
)) == 0)
1885 /* Find the first overlapping VMA */
1886 vma
= find_vma_prev(mm
, start
, &prev
);
1889 /* we have start < vma->vm_end */
1891 /* if it doesn't overlap, we have nothing.. */
1893 if (vma
->vm_start
>= end
)
1897 * If we need to split any vma, do it now to save pain later.
1899 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1900 * unmapped vm_area_struct will remain in use: so lower split_vma
1901 * places tmp vma above, and higher split_vma places tmp vma below.
1903 if (start
> vma
->vm_start
) {
1904 int error
= split_vma(mm
, vma
, start
, 0);
1910 /* Does it split the last one? */
1911 last
= find_vma(mm
, end
);
1912 if (last
&& end
> last
->vm_start
) {
1913 int error
= split_vma(mm
, last
, end
, 1);
1917 vma
= prev
? prev
->vm_next
: mm
->mmap
;
1920 * Remove the vma's, and unmap the actual pages
1922 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
1923 unmap_region(mm
, vma
, prev
, start
, end
);
1925 /* Fix up all other VM information */
1926 remove_vma_list(mm
, vma
);
1931 EXPORT_SYMBOL(do_munmap
);
1933 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1936 struct mm_struct
*mm
= current
->mm
;
1938 profile_munmap(addr
);
1940 down_write(&mm
->mmap_sem
);
1941 ret
= do_munmap(mm
, addr
, len
);
1942 up_write(&mm
->mmap_sem
);
1946 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
1948 #ifdef CONFIG_DEBUG_VM
1949 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
1951 up_read(&mm
->mmap_sem
);
1957 * this is really a simplified "do_mmap". it only handles
1958 * anonymous maps. eventually we may be able to do some
1959 * brk-specific accounting here.
1961 unsigned long do_brk(unsigned long addr
, unsigned long len
)
1963 struct mm_struct
* mm
= current
->mm
;
1964 struct vm_area_struct
* vma
, * prev
;
1965 unsigned long flags
;
1966 struct rb_node
** rb_link
, * rb_parent
;
1967 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
1970 len
= PAGE_ALIGN(len
);
1974 if ((addr
+ len
) > TASK_SIZE
|| (addr
+ len
) < addr
)
1977 if (is_hugepage_only_range(mm
, addr
, len
))
1980 error
= security_file_mmap(NULL
, 0, 0, 0, addr
, 1);
1984 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
1986 error
= arch_mmap_check(addr
, len
, flags
);
1993 if (mm
->def_flags
& VM_LOCKED
) {
1994 unsigned long locked
, lock_limit
;
1995 locked
= len
>> PAGE_SHIFT
;
1996 locked
+= mm
->locked_vm
;
1997 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
1998 lock_limit
>>= PAGE_SHIFT
;
1999 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
2004 * mm->mmap_sem is required to protect against another thread
2005 * changing the mappings in case we sleep.
2007 verify_mm_writelocked(mm
);
2010 * Clear old maps. this also does some error checking for us
2013 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2014 if (vma
&& vma
->vm_start
< addr
+ len
) {
2015 if (do_munmap(mm
, addr
, len
))
2020 /* Check against address space limits *after* clearing old maps... */
2021 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2024 if (mm
->map_count
> sysctl_max_map_count
)
2027 if (security_vm_enough_memory(len
>> PAGE_SHIFT
))
2030 /* Can we just expand an old private anonymous mapping? */
2031 if (vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2032 NULL
, NULL
, pgoff
, NULL
))
2036 * create a vma struct for an anonymous mapping
2038 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2040 vm_unacct_memory(len
>> PAGE_SHIFT
);
2045 vma
->vm_start
= addr
;
2046 vma
->vm_end
= addr
+ len
;
2047 vma
->vm_pgoff
= pgoff
;
2048 vma
->vm_flags
= flags
;
2049 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2050 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2052 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2053 if (flags
& VM_LOCKED
) {
2054 mm
->locked_vm
+= len
>> PAGE_SHIFT
;
2055 make_pages_present(addr
, addr
+ len
);
2060 EXPORT_SYMBOL(do_brk
);
2062 /* Release all mmaps. */
2063 void exit_mmap(struct mm_struct
*mm
)
2065 struct mmu_gather
*tlb
;
2066 struct vm_area_struct
*vma
= mm
->mmap
;
2067 unsigned long nr_accounted
= 0;
2070 /* mm's last user has gone, and its about to be pulled down */
2072 mmu_notifier_release(mm
);
2074 if (!mm
->mmap
) /* Can happen if dup_mmap() received an OOM */
2079 tlb
= tlb_gather_mmu(mm
, 1);
2080 /* Don't update_hiwater_rss(mm) here, do_exit already did */
2081 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2082 end
= unmap_vmas(&tlb
, vma
, 0, -1, &nr_accounted
, NULL
);
2083 vm_unacct_memory(nr_accounted
);
2084 free_pgtables(tlb
, vma
, FIRST_USER_ADDRESS
, 0);
2085 tlb_finish_mmu(tlb
, 0, end
);
2088 * Walk the list again, actually closing and freeing it,
2089 * with preemption enabled, without holding any MM locks.
2092 vma
= remove_vma(vma
);
2094 BUG_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2097 /* Insert vm structure into process list sorted by address
2098 * and into the inode's i_mmap tree. If vm_file is non-NULL
2099 * then i_mmap_lock is taken here.
2101 int insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
2103 struct vm_area_struct
* __vma
, * prev
;
2104 struct rb_node
** rb_link
, * rb_parent
;
2107 * The vm_pgoff of a purely anonymous vma should be irrelevant
2108 * until its first write fault, when page's anon_vma and index
2109 * are set. But now set the vm_pgoff it will almost certainly
2110 * end up with (unless mremap moves it elsewhere before that
2111 * first wfault), so /proc/pid/maps tells a consistent story.
2113 * By setting it to reflect the virtual start address of the
2114 * vma, merges and splits can happen in a seamless way, just
2115 * using the existing file pgoff checks and manipulations.
2116 * Similarly in do_mmap_pgoff and in do_brk.
2118 if (!vma
->vm_file
) {
2119 BUG_ON(vma
->anon_vma
);
2120 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2122 __vma
= find_vma_prepare(mm
,vma
->vm_start
,&prev
,&rb_link
,&rb_parent
);
2123 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
2125 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2126 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2128 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2133 * Copy the vma structure to a new location in the same mm,
2134 * prior to moving page table entries, to effect an mremap move.
2136 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2137 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
2139 struct vm_area_struct
*vma
= *vmap
;
2140 unsigned long vma_start
= vma
->vm_start
;
2141 struct mm_struct
*mm
= vma
->vm_mm
;
2142 struct vm_area_struct
*new_vma
, *prev
;
2143 struct rb_node
**rb_link
, *rb_parent
;
2144 struct mempolicy
*pol
;
2147 * If anonymous vma has not yet been faulted, update new pgoff
2148 * to match new location, to increase its chance of merging.
2150 if (!vma
->vm_file
&& !vma
->anon_vma
)
2151 pgoff
= addr
>> PAGE_SHIFT
;
2153 find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2154 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2155 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2158 * Source vma may have been merged into new_vma
2160 if (vma_start
>= new_vma
->vm_start
&&
2161 vma_start
< new_vma
->vm_end
)
2164 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2167 pol
= mpol_dup(vma_policy(vma
));
2169 kmem_cache_free(vm_area_cachep
, new_vma
);
2172 vma_set_policy(new_vma
, pol
);
2173 new_vma
->vm_start
= addr
;
2174 new_vma
->vm_end
= addr
+ len
;
2175 new_vma
->vm_pgoff
= pgoff
;
2176 if (new_vma
->vm_file
) {
2177 get_file(new_vma
->vm_file
);
2178 if (vma
->vm_flags
& VM_EXECUTABLE
)
2179 added_exe_file_vma(mm
);
2181 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2182 new_vma
->vm_ops
->open(new_vma
);
2183 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2190 * Return true if the calling process may expand its vm space by the passed
2193 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2195 unsigned long cur
= mm
->total_vm
; /* pages */
2198 lim
= current
->signal
->rlim
[RLIMIT_AS
].rlim_cur
>> PAGE_SHIFT
;
2200 if (cur
+ npages
> lim
)
2206 static int special_mapping_fault(struct vm_area_struct
*vma
,
2207 struct vm_fault
*vmf
)
2210 struct page
**pages
;
2213 * special mappings have no vm_file, and in that case, the mm
2214 * uses vm_pgoff internally. So we have to subtract it from here.
2215 * We are allowed to do this because we are the mm; do not copy
2216 * this code into drivers!
2218 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2220 for (pages
= vma
->vm_private_data
; pgoff
&& *pages
; ++pages
)
2224 struct page
*page
= *pages
;
2230 return VM_FAULT_SIGBUS
;
2234 * Having a close hook prevents vma merging regardless of flags.
2236 static void special_mapping_close(struct vm_area_struct
*vma
)
2240 static struct vm_operations_struct special_mapping_vmops
= {
2241 .close
= special_mapping_close
,
2242 .fault
= special_mapping_fault
,
2246 * Called with mm->mmap_sem held for writing.
2247 * Insert a new vma covering the given region, with the given flags.
2248 * Its pages are supplied by the given array of struct page *.
2249 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2250 * The region past the last page supplied will always produce SIGBUS.
2251 * The array pointer and the pages it points to are assumed to stay alive
2252 * for as long as this mapping might exist.
2254 int install_special_mapping(struct mm_struct
*mm
,
2255 unsigned long addr
, unsigned long len
,
2256 unsigned long vm_flags
, struct page
**pages
)
2258 struct vm_area_struct
*vma
;
2260 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2261 if (unlikely(vma
== NULL
))
2265 vma
->vm_start
= addr
;
2266 vma
->vm_end
= addr
+ len
;
2268 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
;
2269 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
2271 vma
->vm_ops
= &special_mapping_vmops
;
2272 vma
->vm_private_data
= pages
;
2274 if (unlikely(insert_vm_struct(mm
, vma
))) {
2275 kmem_cache_free(vm_area_cachep
, vma
);
2279 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2284 static DEFINE_MUTEX(mm_all_locks_mutex
);
2286 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
2288 if (!test_bit(0, (unsigned long *) &anon_vma
->head
.next
)) {
2290 * The LSB of head.next can't change from under us
2291 * because we hold the mm_all_locks_mutex.
2293 spin_lock_nest_lock(&anon_vma
->lock
, &mm
->mmap_sem
);
2295 * We can safely modify head.next after taking the
2296 * anon_vma->lock. If some other vma in this mm shares
2297 * the same anon_vma we won't take it again.
2299 * No need of atomic instructions here, head.next
2300 * can't change from under us thanks to the
2303 if (__test_and_set_bit(0, (unsigned long *)
2304 &anon_vma
->head
.next
))
2309 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
2311 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2313 * AS_MM_ALL_LOCKS can't change from under us because
2314 * we hold the mm_all_locks_mutex.
2316 * Operations on ->flags have to be atomic because
2317 * even if AS_MM_ALL_LOCKS is stable thanks to the
2318 * mm_all_locks_mutex, there may be other cpus
2319 * changing other bitflags in parallel to us.
2321 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
2323 spin_lock_nest_lock(&mapping
->i_mmap_lock
, &mm
->mmap_sem
);
2328 * This operation locks against the VM for all pte/vma/mm related
2329 * operations that could ever happen on a certain mm. This includes
2330 * vmtruncate, try_to_unmap, and all page faults.
2332 * The caller must take the mmap_sem in write mode before calling
2333 * mm_take_all_locks(). The caller isn't allowed to release the
2334 * mmap_sem until mm_drop_all_locks() returns.
2336 * mmap_sem in write mode is required in order to block all operations
2337 * that could modify pagetables and free pages without need of
2338 * altering the vma layout (for example populate_range() with
2339 * nonlinear vmas). It's also needed in write mode to avoid new
2340 * anon_vmas to be associated with existing vmas.
2342 * A single task can't take more than one mm_take_all_locks() in a row
2343 * or it would deadlock.
2345 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2346 * mapping->flags avoid to take the same lock twice, if more than one
2347 * vma in this mm is backed by the same anon_vma or address_space.
2349 * We can take all the locks in random order because the VM code
2350 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2351 * takes more than one of them in a row. Secondly we're protected
2352 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2354 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2355 * that may have to take thousand of locks.
2357 * mm_take_all_locks() can fail if it's interrupted by signals.
2359 int mm_take_all_locks(struct mm_struct
*mm
)
2361 struct vm_area_struct
*vma
;
2364 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2366 mutex_lock(&mm_all_locks_mutex
);
2368 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2369 if (signal_pending(current
))
2371 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2372 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
2375 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2376 if (signal_pending(current
))
2379 vm_lock_anon_vma(mm
, vma
->anon_vma
);
2386 mm_drop_all_locks(mm
);
2391 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
2393 if (test_bit(0, (unsigned long *) &anon_vma
->head
.next
)) {
2395 * The LSB of head.next can't change to 0 from under
2396 * us because we hold the mm_all_locks_mutex.
2398 * We must however clear the bitflag before unlocking
2399 * the vma so the users using the anon_vma->head will
2400 * never see our bitflag.
2402 * No need of atomic instructions here, head.next
2403 * can't change from under us until we release the
2406 if (!__test_and_clear_bit(0, (unsigned long *)
2407 &anon_vma
->head
.next
))
2409 spin_unlock(&anon_vma
->lock
);
2413 static void vm_unlock_mapping(struct address_space
*mapping
)
2415 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2417 * AS_MM_ALL_LOCKS can't change to 0 from under us
2418 * because we hold the mm_all_locks_mutex.
2420 spin_unlock(&mapping
->i_mmap_lock
);
2421 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
2428 * The mmap_sem cannot be released by the caller until
2429 * mm_drop_all_locks() returns.
2431 void mm_drop_all_locks(struct mm_struct
*mm
)
2433 struct vm_area_struct
*vma
;
2435 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2436 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
2438 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2440 vm_unlock_anon_vma(vma
->anon_vma
);
2441 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2442 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
2445 mutex_unlock(&mm_all_locks_mutex
);