6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
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>
30 #include <linux/perf_event.h>
32 #include <asm/uaccess.h>
33 #include <asm/cacheflush.h>
35 #include <asm/mmu_context.h>
39 #ifndef arch_mmap_check
40 #define arch_mmap_check(addr, len, flags) (0)
43 #ifndef arch_rebalance_pgtables
44 #define arch_rebalance_pgtables(addr, len) (addr)
47 static void unmap_region(struct mm_struct
*mm
,
48 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
49 unsigned long start
, unsigned long end
);
52 * WARNING: the debugging will use recursive algorithms so never enable this
53 * unless you know what you are doing.
57 /* description of effects of mapping type and prot in current implementation.
58 * this is due to the limited x86 page protection hardware. The expected
59 * behavior is in parens:
62 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
63 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
64 * w: (no) no w: (no) no w: (yes) yes w: (no) no
65 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
67 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
68 * w: (no) no w: (no) no w: (copy) copy w: (no) no
69 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
72 pgprot_t protection_map
[16] = {
73 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
74 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
77 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
79 return __pgprot(pgprot_val(protection_map
[vm_flags
&
80 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
81 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
83 EXPORT_SYMBOL(vm_get_page_prot
);
85 int sysctl_overcommit_memory
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
86 int sysctl_overcommit_ratio
= 50; /* default is 50% */
87 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
88 struct percpu_counter vm_committed_as
;
91 * Check that a process has enough memory to allocate a new virtual
92 * mapping. 0 means there is enough memory for the allocation to
93 * succeed and -ENOMEM implies there is not.
95 * We currently support three overcommit policies, which are set via the
96 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
98 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
99 * Additional code 2002 Jul 20 by Robert Love.
101 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
103 * Note this is a helper function intended to be used by LSMs which
104 * wish to use this logic.
106 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
108 unsigned long free
, allowed
;
110 vm_acct_memory(pages
);
113 * Sometimes we want to use more memory than we have
115 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
118 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
121 free
= global_page_state(NR_FILE_PAGES
);
122 free
+= nr_swap_pages
;
125 * Any slabs which are created with the
126 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
127 * which are reclaimable, under pressure. The dentry
128 * cache and most inode caches should fall into this
130 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
133 * Leave the last 3% for root
142 * nr_free_pages() is very expensive on large systems,
143 * only call if we're about to fail.
148 * Leave reserved pages. The pages are not for anonymous pages.
150 if (n
<= totalreserve_pages
)
153 n
-= totalreserve_pages
;
156 * Leave the last 3% for root
168 allowed
= (totalram_pages
- hugetlb_total_pages())
169 * sysctl_overcommit_ratio
/ 100;
171 * Leave the last 3% for root
174 allowed
-= allowed
/ 32;
175 allowed
+= total_swap_pages
;
177 /* Don't let a single process grow too big:
178 leave 3% of the size of this process for other processes */
180 allowed
-= mm
->total_vm
/ 32;
182 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
185 vm_unacct_memory(pages
);
191 * Requires inode->i_mapping->i_mmap_lock
193 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
194 struct file
*file
, struct address_space
*mapping
)
196 if (vma
->vm_flags
& VM_DENYWRITE
)
197 atomic_inc(&file
->f_path
.dentry
->d_inode
->i_writecount
);
198 if (vma
->vm_flags
& VM_SHARED
)
199 mapping
->i_mmap_writable
--;
201 flush_dcache_mmap_lock(mapping
);
202 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
203 list_del_init(&vma
->shared
.vm_set
.list
);
205 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
206 flush_dcache_mmap_unlock(mapping
);
210 * Unlink a file-based vm structure from its prio_tree, to hide
211 * vma from rmap and vmtruncate before freeing its page tables.
213 void unlink_file_vma(struct vm_area_struct
*vma
)
215 struct file
*file
= vma
->vm_file
;
218 struct address_space
*mapping
= file
->f_mapping
;
219 spin_lock(&mapping
->i_mmap_lock
);
220 __remove_shared_vm_struct(vma
, file
, mapping
);
221 spin_unlock(&mapping
->i_mmap_lock
);
226 * Close a vm structure and free it, returning the next.
228 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
230 struct vm_area_struct
*next
= vma
->vm_next
;
233 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
234 vma
->vm_ops
->close(vma
);
237 if (vma
->vm_flags
& VM_EXECUTABLE
)
238 removed_exe_file_vma(vma
->vm_mm
);
240 mpol_put(vma_policy(vma
));
241 kmem_cache_free(vm_area_cachep
, vma
);
245 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
247 unsigned long rlim
, retval
;
248 unsigned long newbrk
, oldbrk
;
249 struct mm_struct
*mm
= current
->mm
;
250 unsigned long min_brk
;
252 down_write(&mm
->mmap_sem
);
254 #ifdef CONFIG_COMPAT_BRK
255 min_brk
= mm
->end_code
;
257 min_brk
= mm
->start_brk
;
263 * Check against rlimit here. If this check is done later after the test
264 * of oldbrk with newbrk then it can escape the test and let the data
265 * segment grow beyond its set limit the in case where the limit is
266 * not page aligned -Ram Gupta
268 rlim
= current
->signal
->rlim
[RLIMIT_DATA
].rlim_cur
;
269 if (rlim
< RLIM_INFINITY
&& (brk
- mm
->start_brk
) +
270 (mm
->end_data
- mm
->start_data
) > rlim
)
273 newbrk
= PAGE_ALIGN(brk
);
274 oldbrk
= PAGE_ALIGN(mm
->brk
);
275 if (oldbrk
== newbrk
)
278 /* Always allow shrinking brk. */
279 if (brk
<= mm
->brk
) {
280 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
285 /* Check against existing mmap mappings. */
286 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
289 /* Ok, looks good - let it rip. */
290 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
296 up_write(&mm
->mmap_sem
);
301 static int browse_rb(struct rb_root
*root
)
304 struct rb_node
*nd
, *pn
= NULL
;
305 unsigned long prev
= 0, pend
= 0;
307 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
308 struct vm_area_struct
*vma
;
309 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
310 if (vma
->vm_start
< prev
)
311 printk("vm_start %lx prev %lx\n", vma
->vm_start
, prev
), i
= -1;
312 if (vma
->vm_start
< pend
)
313 printk("vm_start %lx pend %lx\n", vma
->vm_start
, pend
);
314 if (vma
->vm_start
> vma
->vm_end
)
315 printk("vm_end %lx < vm_start %lx\n", vma
->vm_end
, vma
->vm_start
);
318 prev
= vma
->vm_start
;
322 for (nd
= pn
; nd
; nd
= rb_prev(nd
)) {
326 printk("backwards %d, forwards %d\n", j
, i
), i
= 0;
330 void validate_mm(struct mm_struct
*mm
)
334 struct vm_area_struct
*tmp
= mm
->mmap
;
339 if (i
!= mm
->map_count
)
340 printk("map_count %d vm_next %d\n", mm
->map_count
, i
), bug
= 1;
341 i
= browse_rb(&mm
->mm_rb
);
342 if (i
!= mm
->map_count
)
343 printk("map_count %d rb %d\n", mm
->map_count
, i
), bug
= 1;
347 #define validate_mm(mm) do { } while (0)
350 static struct vm_area_struct
*
351 find_vma_prepare(struct mm_struct
*mm
, unsigned long addr
,
352 struct vm_area_struct
**pprev
, struct rb_node
***rb_link
,
353 struct rb_node
** rb_parent
)
355 struct vm_area_struct
* vma
;
356 struct rb_node
** __rb_link
, * __rb_parent
, * rb_prev
;
358 __rb_link
= &mm
->mm_rb
.rb_node
;
359 rb_prev
= __rb_parent
= NULL
;
363 struct vm_area_struct
*vma_tmp
;
365 __rb_parent
= *__rb_link
;
366 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
368 if (vma_tmp
->vm_end
> addr
) {
370 if (vma_tmp
->vm_start
<= addr
)
372 __rb_link
= &__rb_parent
->rb_left
;
374 rb_prev
= __rb_parent
;
375 __rb_link
= &__rb_parent
->rb_right
;
381 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
382 *rb_link
= __rb_link
;
383 *rb_parent
= __rb_parent
;
388 __vma_link_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
389 struct vm_area_struct
*prev
, struct rb_node
*rb_parent
)
392 vma
->vm_next
= prev
->vm_next
;
397 vma
->vm_next
= rb_entry(rb_parent
,
398 struct vm_area_struct
, vm_rb
);
404 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
405 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
407 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
408 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
411 static void __vma_link_file(struct vm_area_struct
*vma
)
417 struct address_space
*mapping
= file
->f_mapping
;
419 if (vma
->vm_flags
& VM_DENYWRITE
)
420 atomic_dec(&file
->f_path
.dentry
->d_inode
->i_writecount
);
421 if (vma
->vm_flags
& VM_SHARED
)
422 mapping
->i_mmap_writable
++;
424 flush_dcache_mmap_lock(mapping
);
425 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
426 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
428 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
429 flush_dcache_mmap_unlock(mapping
);
434 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
435 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
436 struct rb_node
*rb_parent
)
438 __vma_link_list(mm
, vma
, prev
, rb_parent
);
439 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
440 __anon_vma_link(vma
);
443 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
444 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
445 struct rb_node
*rb_parent
)
447 struct address_space
*mapping
= NULL
;
450 mapping
= vma
->vm_file
->f_mapping
;
453 spin_lock(&mapping
->i_mmap_lock
);
454 vma
->vm_truncate_count
= mapping
->truncate_count
;
458 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
459 __vma_link_file(vma
);
461 anon_vma_unlock(vma
);
463 spin_unlock(&mapping
->i_mmap_lock
);
470 * Helper for vma_adjust in the split_vma insert case:
471 * insert vm structure into list and rbtree and anon_vma,
472 * but it has already been inserted into prio_tree earlier.
474 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
476 struct vm_area_struct
*__vma
, *prev
;
477 struct rb_node
**rb_link
, *rb_parent
;
479 __vma
= find_vma_prepare(mm
, vma
->vm_start
,&prev
, &rb_link
, &rb_parent
);
480 BUG_ON(__vma
&& __vma
->vm_start
< vma
->vm_end
);
481 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
486 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
487 struct vm_area_struct
*prev
)
489 prev
->vm_next
= vma
->vm_next
;
490 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
491 if (mm
->mmap_cache
== vma
)
492 mm
->mmap_cache
= prev
;
496 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
497 * is already present in an i_mmap tree without adjusting the tree.
498 * The following helper function should be used when such adjustments
499 * are necessary. The "insert" vma (if any) is to be inserted
500 * before we drop the necessary locks.
502 void vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
503 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
505 struct mm_struct
*mm
= vma
->vm_mm
;
506 struct vm_area_struct
*next
= vma
->vm_next
;
507 struct vm_area_struct
*importer
= NULL
;
508 struct address_space
*mapping
= NULL
;
509 struct prio_tree_root
*root
= NULL
;
510 struct file
*file
= vma
->vm_file
;
511 struct anon_vma
*anon_vma
= NULL
;
512 long adjust_next
= 0;
515 if (next
&& !insert
) {
516 if (end
>= next
->vm_end
) {
518 * vma expands, overlapping all the next, and
519 * perhaps the one after too (mprotect case 6).
521 again
: remove_next
= 1 + (end
> next
->vm_end
);
523 anon_vma
= next
->anon_vma
;
525 } else if (end
> next
->vm_start
) {
527 * vma expands, overlapping part of the next:
528 * mprotect case 5 shifting the boundary up.
530 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
531 anon_vma
= next
->anon_vma
;
533 } else if (end
< vma
->vm_end
) {
535 * vma shrinks, and !insert tells it's not
536 * split_vma inserting another: so it must be
537 * mprotect case 4 shifting the boundary down.
539 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
540 anon_vma
= next
->anon_vma
;
546 mapping
= file
->f_mapping
;
547 if (!(vma
->vm_flags
& VM_NONLINEAR
))
548 root
= &mapping
->i_mmap
;
549 spin_lock(&mapping
->i_mmap_lock
);
551 vma
->vm_truncate_count
!= next
->vm_truncate_count
) {
553 * unmap_mapping_range might be in progress:
554 * ensure that the expanding vma is rescanned.
556 importer
->vm_truncate_count
= 0;
559 insert
->vm_truncate_count
= vma
->vm_truncate_count
;
561 * Put into prio_tree now, so instantiated pages
562 * are visible to arm/parisc __flush_dcache_page
563 * throughout; but we cannot insert into address
564 * space until vma start or end is updated.
566 __vma_link_file(insert
);
571 * When changing only vma->vm_end, we don't really need
574 if (vma
->anon_vma
&& (insert
|| importer
|| start
!= vma
->vm_start
))
575 anon_vma
= vma
->anon_vma
;
577 spin_lock(&anon_vma
->lock
);
579 * Easily overlooked: when mprotect shifts the boundary,
580 * make sure the expanding vma has anon_vma set if the
581 * shrinking vma had, to cover any anon pages imported.
583 if (importer
&& !importer
->anon_vma
) {
584 importer
->anon_vma
= anon_vma
;
585 __anon_vma_link(importer
);
590 flush_dcache_mmap_lock(mapping
);
591 vma_prio_tree_remove(vma
, root
);
593 vma_prio_tree_remove(next
, root
);
596 vma
->vm_start
= start
;
598 vma
->vm_pgoff
= pgoff
;
600 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
601 next
->vm_pgoff
+= adjust_next
;
606 vma_prio_tree_insert(next
, root
);
607 vma_prio_tree_insert(vma
, root
);
608 flush_dcache_mmap_unlock(mapping
);
613 * vma_merge has merged next into vma, and needs
614 * us to remove next before dropping the locks.
616 __vma_unlink(mm
, next
, vma
);
618 __remove_shared_vm_struct(next
, file
, mapping
);
620 __anon_vma_merge(vma
, next
);
623 * split_vma has split insert from vma, and needs
624 * us to insert it before dropping the locks
625 * (it may either follow vma or precede it).
627 __insert_vm_struct(mm
, insert
);
631 spin_unlock(&anon_vma
->lock
);
633 spin_unlock(&mapping
->i_mmap_lock
);
638 if (next
->vm_flags
& VM_EXECUTABLE
)
639 removed_exe_file_vma(mm
);
642 mpol_put(vma_policy(next
));
643 kmem_cache_free(vm_area_cachep
, next
);
645 * In mprotect's case 6 (see comments on vma_merge),
646 * we must remove another next too. It would clutter
647 * up the code too much to do both in one go.
649 if (remove_next
== 2) {
659 * If the vma has a ->close operation then the driver probably needs to release
660 * per-vma resources, so we don't attempt to merge those.
662 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
663 struct file
*file
, unsigned long vm_flags
)
665 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
666 if ((vma
->vm_flags
^ vm_flags
) & ~VM_CAN_NONLINEAR
)
668 if (vma
->vm_file
!= file
)
670 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
675 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
676 struct anon_vma
*anon_vma2
)
678 return !anon_vma1
|| !anon_vma2
|| (anon_vma1
== anon_vma2
);
682 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
683 * in front of (at a lower virtual address and file offset than) the vma.
685 * We cannot merge two vmas if they have differently assigned (non-NULL)
686 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
688 * We don't check here for the merged mmap wrapping around the end of pagecache
689 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
690 * wrap, nor mmaps which cover the final page at index -1UL.
693 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
694 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
696 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
697 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
698 if (vma
->vm_pgoff
== vm_pgoff
)
705 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
706 * beyond (at a higher virtual address and file offset than) the vma.
708 * We cannot merge two vmas if they have differently assigned (non-NULL)
709 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
712 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
713 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
715 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
716 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
718 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
719 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
726 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
727 * whether that can be merged with its predecessor or its successor.
728 * Or both (it neatly fills a hole).
730 * In most cases - when called for mmap, brk or mremap - [addr,end) is
731 * certain not to be mapped by the time vma_merge is called; but when
732 * called for mprotect, it is certain to be already mapped (either at
733 * an offset within prev, or at the start of next), and the flags of
734 * this area are about to be changed to vm_flags - and the no-change
735 * case has already been eliminated.
737 * The following mprotect cases have to be considered, where AAAA is
738 * the area passed down from mprotect_fixup, never extending beyond one
739 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
741 * AAAA AAAA AAAA AAAA
742 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
743 * cannot merge might become might become might become
744 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
745 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
746 * mremap move: PPPPNNNNNNNN 8
748 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
749 * might become case 1 below case 2 below case 3 below
751 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
752 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
754 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
755 struct vm_area_struct
*prev
, unsigned long addr
,
756 unsigned long end
, unsigned long vm_flags
,
757 struct anon_vma
*anon_vma
, struct file
*file
,
758 pgoff_t pgoff
, struct mempolicy
*policy
)
760 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
761 struct vm_area_struct
*area
, *next
;
764 * We later require that vma->vm_flags == vm_flags,
765 * so this tests vma->vm_flags & VM_SPECIAL, too.
767 if (vm_flags
& VM_SPECIAL
)
771 next
= prev
->vm_next
;
775 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
776 next
= next
->vm_next
;
779 * Can it merge with the predecessor?
781 if (prev
&& prev
->vm_end
== addr
&&
782 mpol_equal(vma_policy(prev
), policy
) &&
783 can_vma_merge_after(prev
, vm_flags
,
784 anon_vma
, file
, pgoff
)) {
786 * OK, it can. Can we now merge in the successor as well?
788 if (next
&& end
== next
->vm_start
&&
789 mpol_equal(policy
, vma_policy(next
)) &&
790 can_vma_merge_before(next
, vm_flags
,
791 anon_vma
, file
, pgoff
+pglen
) &&
792 is_mergeable_anon_vma(prev
->anon_vma
,
795 vma_adjust(prev
, prev
->vm_start
,
796 next
->vm_end
, prev
->vm_pgoff
, NULL
);
797 } else /* cases 2, 5, 7 */
798 vma_adjust(prev
, prev
->vm_start
,
799 end
, prev
->vm_pgoff
, NULL
);
804 * Can this new request be merged in front of next?
806 if (next
&& end
== next
->vm_start
&&
807 mpol_equal(policy
, vma_policy(next
)) &&
808 can_vma_merge_before(next
, vm_flags
,
809 anon_vma
, file
, pgoff
+pglen
)) {
810 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
811 vma_adjust(prev
, prev
->vm_start
,
812 addr
, prev
->vm_pgoff
, NULL
);
813 else /* cases 3, 8 */
814 vma_adjust(area
, addr
, next
->vm_end
,
815 next
->vm_pgoff
- pglen
, NULL
);
823 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
824 * neighbouring vmas for a suitable anon_vma, before it goes off
825 * to allocate a new anon_vma. It checks because a repetitive
826 * sequence of mprotects and faults may otherwise lead to distinct
827 * anon_vmas being allocated, preventing vma merge in subsequent
830 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
832 struct vm_area_struct
*near
;
833 unsigned long vm_flags
;
840 * Since only mprotect tries to remerge vmas, match flags
841 * which might be mprotected into each other later on.
842 * Neither mlock nor madvise tries to remerge at present,
843 * so leave their flags as obstructing a merge.
845 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
846 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
848 if (near
->anon_vma
&& vma
->vm_end
== near
->vm_start
&&
849 mpol_equal(vma_policy(vma
), vma_policy(near
)) &&
850 can_vma_merge_before(near
, vm_flags
,
851 NULL
, vma
->vm_file
, vma
->vm_pgoff
+
852 ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
)))
853 return near
->anon_vma
;
856 * It is potentially slow to have to call find_vma_prev here.
857 * But it's only on the first write fault on the vma, not
858 * every time, and we could devise a way to avoid it later
859 * (e.g. stash info in next's anon_vma_node when assigning
860 * an anon_vma, or when trying vma_merge). Another time.
862 BUG_ON(find_vma_prev(vma
->vm_mm
, vma
->vm_start
, &near
) != vma
);
866 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
867 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
869 if (near
->anon_vma
&& near
->vm_end
== vma
->vm_start
&&
870 mpol_equal(vma_policy(near
), vma_policy(vma
)) &&
871 can_vma_merge_after(near
, vm_flags
,
872 NULL
, vma
->vm_file
, vma
->vm_pgoff
))
873 return near
->anon_vma
;
876 * There's no absolute need to look only at touching neighbours:
877 * we could search further afield for "compatible" anon_vmas.
878 * But it would probably just be a waste of time searching,
879 * or lead to too many vmas hanging off the same anon_vma.
880 * We're trying to allow mprotect remerging later on,
881 * not trying to minimize memory used for anon_vmas.
886 #ifdef CONFIG_PROC_FS
887 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
888 struct file
*file
, long pages
)
890 const unsigned long stack_flags
891 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
894 mm
->shared_vm
+= pages
;
895 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
896 mm
->exec_vm
+= pages
;
897 } else if (flags
& stack_flags
)
898 mm
->stack_vm
+= pages
;
899 if (flags
& (VM_RESERVED
|VM_IO
))
900 mm
->reserved_vm
+= pages
;
902 #endif /* CONFIG_PROC_FS */
905 * The caller must hold down_write(¤t->mm->mmap_sem).
908 unsigned long do_mmap_pgoff(struct file
*file
, unsigned long addr
,
909 unsigned long len
, unsigned long prot
,
910 unsigned long flags
, unsigned long pgoff
)
912 struct mm_struct
* mm
= current
->mm
;
914 unsigned int vm_flags
;
916 unsigned long reqprot
= prot
;
919 * Does the application expect PROT_READ to imply PROT_EXEC?
921 * (the exception is when the underlying filesystem is noexec
922 * mounted, in which case we dont add PROT_EXEC.)
924 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
925 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
931 if (!(flags
& MAP_FIXED
))
932 addr
= round_hint_to_min(addr
);
934 /* Careful about overflows.. */
935 len
= PAGE_ALIGN(len
);
939 /* offset overflow? */
940 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
943 /* Too many mappings? */
944 if (mm
->map_count
> sysctl_max_map_count
)
947 /* Obtain the address to map to. we verify (or select) it and ensure
948 * that it represents a valid section of the address space.
950 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
951 if (addr
& ~PAGE_MASK
)
954 /* Do simple checking here so the lower-level routines won't have
955 * to. we assume access permissions have been handled by the open
956 * of the memory object, so we don't do any here.
958 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
959 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
961 if (flags
& MAP_LOCKED
)
965 /* mlock MCL_FUTURE? */
966 if (vm_flags
& VM_LOCKED
) {
967 unsigned long locked
, lock_limit
;
968 locked
= len
>> PAGE_SHIFT
;
969 locked
+= mm
->locked_vm
;
970 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
971 lock_limit
>>= PAGE_SHIFT
;
972 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
976 inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
979 switch (flags
& MAP_TYPE
) {
981 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
985 * Make sure we don't allow writing to an append-only
988 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
992 * Make sure there are no mandatory locks on the file.
994 if (locks_verify_locked(inode
))
997 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
998 if (!(file
->f_mode
& FMODE_WRITE
))
999 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1003 if (!(file
->f_mode
& FMODE_READ
))
1005 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1006 if (vm_flags
& VM_EXEC
)
1008 vm_flags
&= ~VM_MAYEXEC
;
1011 if (!file
->f_op
|| !file
->f_op
->mmap
)
1019 switch (flags
& MAP_TYPE
) {
1025 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1029 * Set pgoff according to addr for anon_vma.
1031 pgoff
= addr
>> PAGE_SHIFT
;
1038 error
= security_file_mmap(file
, reqprot
, prot
, flags
, addr
, 0);
1042 return mmap_region(file
, addr
, len
, flags
, vm_flags
, pgoff
);
1044 EXPORT_SYMBOL(do_mmap_pgoff
);
1047 * Some shared mappigns will want the pages marked read-only
1048 * to track write events. If so, we'll downgrade vm_page_prot
1049 * to the private version (using protection_map[] without the
1052 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1054 unsigned int vm_flags
= vma
->vm_flags
;
1056 /* If it was private or non-writable, the write bit is already clear */
1057 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1060 /* The backer wishes to know when pages are first written to? */
1061 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1064 /* The open routine did something to the protections already? */
1065 if (pgprot_val(vma
->vm_page_prot
) !=
1066 pgprot_val(vm_get_page_prot(vm_flags
)))
1069 /* Specialty mapping? */
1070 if (vm_flags
& (VM_PFNMAP
|VM_INSERTPAGE
))
1073 /* Can the mapping track the dirty pages? */
1074 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1075 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1079 * We account for memory if it's a private writeable mapping,
1080 * not hugepages and VM_NORESERVE wasn't set.
1082 static inline int accountable_mapping(struct file
*file
, unsigned int vm_flags
)
1085 * hugetlb has its own accounting separate from the core VM
1086 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1088 if (file
&& is_file_hugepages(file
))
1091 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
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
)
1098 struct mm_struct
*mm
= current
->mm
;
1099 struct vm_area_struct
*vma
, *prev
;
1100 int correct_wcount
= 0;
1102 struct rb_node
**rb_link
, *rb_parent
;
1103 unsigned long charged
= 0;
1104 struct inode
*inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1106 /* Clear old maps */
1109 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1110 if (vma
&& vma
->vm_start
< addr
+ len
) {
1111 if (do_munmap(mm
, addr
, len
))
1116 /* Check against address space limit. */
1117 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1121 * Set 'VM_NORESERVE' if we should not account for the
1122 * memory use of this mapping.
1124 if ((flags
& MAP_NORESERVE
)) {
1125 /* We honor MAP_NORESERVE if allowed to overcommit */
1126 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1127 vm_flags
|= VM_NORESERVE
;
1129 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1130 if (file
&& is_file_hugepages(file
))
1131 vm_flags
|= VM_NORESERVE
;
1135 * Private writable mapping: check memory availability
1137 if (accountable_mapping(file
, vm_flags
)) {
1138 charged
= len
>> PAGE_SHIFT
;
1139 if (security_vm_enough_memory(charged
))
1141 vm_flags
|= VM_ACCOUNT
;
1145 * Can we just expand an old mapping?
1147 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
, NULL
, file
, 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
);
1187 /* Can addr have changed??
1189 * Answer: Yes, several device drivers can do it in their
1190 * f_op->mmap method. -DaveM
1192 addr
= vma
->vm_start
;
1193 pgoff
= vma
->vm_pgoff
;
1194 vm_flags
= vma
->vm_flags
;
1195 } else if (vm_flags
& VM_SHARED
) {
1196 error
= shmem_zero_setup(vma
);
1201 if (vma_wants_writenotify(vma
)) {
1202 pgprot_t pprot
= vma
->vm_page_prot
;
1204 /* Can vma->vm_page_prot have changed??
1206 * Answer: Yes, drivers may have changed it in their
1207 * f_op->mmap method.
1209 * Ensures that vmas marked as uncached stay that way.
1211 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
& ~VM_SHARED
);
1212 if (pgprot_val(pprot
) == pgprot_val(pgprot_noncached(pprot
)))
1213 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
1216 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1217 file
= vma
->vm_file
;
1219 /* Once vma denies write, undo our temporary denial count */
1221 atomic_inc(&inode
->i_writecount
);
1223 perf_event_mmap(vma
);
1225 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1226 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1227 if (vm_flags
& VM_LOCKED
) {
1229 * makes pages present; downgrades, drops, reacquires mmap_sem
1231 long nr_pages
= mlock_vma_pages_range(vma
, addr
, addr
+ len
);
1233 return nr_pages
; /* vma gone! */
1234 mm
->locked_vm
+= (len
>> PAGE_SHIFT
) - nr_pages
;
1235 } else if ((flags
& MAP_POPULATE
) && !(flags
& MAP_NONBLOCK
))
1236 make_pages_present(addr
, addr
+ len
);
1241 atomic_inc(&inode
->i_writecount
);
1242 vma
->vm_file
= NULL
;
1245 /* Undo any partial mapping done by a device driver. */
1246 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1249 kmem_cache_free(vm_area_cachep
, vma
);
1252 vm_unacct_memory(charged
);
1256 /* Get an address range which is currently unmapped.
1257 * For shmat() with addr=0.
1259 * Ugly calling convention alert:
1260 * Return value with the low bits set means error value,
1262 * if (ret & ~PAGE_MASK)
1265 * This function "knows" that -ENOMEM has the bits set.
1267 #ifndef HAVE_ARCH_UNMAPPED_AREA
1269 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1270 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1272 struct mm_struct
*mm
= current
->mm
;
1273 struct vm_area_struct
*vma
;
1274 unsigned long start_addr
;
1276 if (len
> TASK_SIZE
)
1279 if (flags
& MAP_FIXED
)
1283 addr
= PAGE_ALIGN(addr
);
1284 vma
= find_vma(mm
, addr
);
1285 if (TASK_SIZE
- len
>= addr
&&
1286 (!vma
|| addr
+ len
<= vma
->vm_start
))
1289 if (len
> mm
->cached_hole_size
) {
1290 start_addr
= addr
= mm
->free_area_cache
;
1292 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1293 mm
->cached_hole_size
= 0;
1297 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1298 /* At this point: (!vma || addr < vma->vm_end). */
1299 if (TASK_SIZE
- len
< addr
) {
1301 * Start a new search - just in case we missed
1304 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1305 addr
= TASK_UNMAPPED_BASE
;
1307 mm
->cached_hole_size
= 0;
1312 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1314 * Remember the place where we stopped the search:
1316 mm
->free_area_cache
= addr
+ len
;
1319 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1320 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1326 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1329 * Is this a new hole at the lowest possible address?
1331 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
) {
1332 mm
->free_area_cache
= addr
;
1333 mm
->cached_hole_size
= ~0UL;
1338 * This mmap-allocator allocates new areas top-down from below the
1339 * stack's low limit (the base):
1341 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1343 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1344 const unsigned long len
, const unsigned long pgoff
,
1345 const unsigned long flags
)
1347 struct vm_area_struct
*vma
;
1348 struct mm_struct
*mm
= current
->mm
;
1349 unsigned long addr
= addr0
;
1351 /* requested length too big for entire address space */
1352 if (len
> TASK_SIZE
)
1355 if (flags
& MAP_FIXED
)
1358 /* requesting a specific address */
1360 addr
= PAGE_ALIGN(addr
);
1361 vma
= find_vma(mm
, addr
);
1362 if (TASK_SIZE
- len
>= addr
&&
1363 (!vma
|| addr
+ len
<= vma
->vm_start
))
1367 /* check if free_area_cache is useful for us */
1368 if (len
<= mm
->cached_hole_size
) {
1369 mm
->cached_hole_size
= 0;
1370 mm
->free_area_cache
= mm
->mmap_base
;
1373 /* either no address requested or can't fit in requested address hole */
1374 addr
= mm
->free_area_cache
;
1376 /* make sure it can fit in the remaining address space */
1378 vma
= find_vma(mm
, addr
-len
);
1379 if (!vma
|| addr
<= vma
->vm_start
)
1380 /* remember the address as a hint for next time */
1381 return (mm
->free_area_cache
= addr
-len
);
1384 if (mm
->mmap_base
< len
)
1387 addr
= mm
->mmap_base
-len
;
1391 * Lookup failure means no vma is above this address,
1392 * else if new region fits below vma->vm_start,
1393 * return with success:
1395 vma
= find_vma(mm
, addr
);
1396 if (!vma
|| addr
+len
<= vma
->vm_start
)
1397 /* remember the address as a hint for next time */
1398 return (mm
->free_area_cache
= addr
);
1400 /* remember the largest hole we saw so far */
1401 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1402 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1404 /* try just below the current vma->vm_start */
1405 addr
= vma
->vm_start
-len
;
1406 } while (len
< vma
->vm_start
);
1410 * A failed mmap() very likely causes application failure,
1411 * so fall back to the bottom-up function here. This scenario
1412 * can happen with large stack limits and large mmap()
1415 mm
->cached_hole_size
= ~0UL;
1416 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1417 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1419 * Restore the topdown base:
1421 mm
->free_area_cache
= mm
->mmap_base
;
1422 mm
->cached_hole_size
= ~0UL;
1428 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1431 * Is this a new hole at the highest possible address?
1433 if (addr
> mm
->free_area_cache
)
1434 mm
->free_area_cache
= addr
;
1436 /* dont allow allocations above current base */
1437 if (mm
->free_area_cache
> mm
->mmap_base
)
1438 mm
->free_area_cache
= mm
->mmap_base
;
1442 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1443 unsigned long pgoff
, unsigned long flags
)
1445 unsigned long (*get_area
)(struct file
*, unsigned long,
1446 unsigned long, unsigned long, unsigned long);
1448 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
1452 /* Careful about overflows.. */
1453 if (len
> TASK_SIZE
)
1456 get_area
= current
->mm
->get_unmapped_area
;
1457 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1458 get_area
= file
->f_op
->get_unmapped_area
;
1459 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1460 if (IS_ERR_VALUE(addr
))
1463 if (addr
> TASK_SIZE
- len
)
1465 if (addr
& ~PAGE_MASK
)
1468 return arch_rebalance_pgtables(addr
, len
);
1471 EXPORT_SYMBOL(get_unmapped_area
);
1473 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1474 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
1476 struct vm_area_struct
*vma
= NULL
;
1479 /* Check the cache first. */
1480 /* (Cache hit rate is typically around 35%.) */
1481 vma
= mm
->mmap_cache
;
1482 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1483 struct rb_node
* rb_node
;
1485 rb_node
= mm
->mm_rb
.rb_node
;
1489 struct vm_area_struct
* vma_tmp
;
1491 vma_tmp
= rb_entry(rb_node
,
1492 struct vm_area_struct
, vm_rb
);
1494 if (vma_tmp
->vm_end
> addr
) {
1496 if (vma_tmp
->vm_start
<= addr
)
1498 rb_node
= rb_node
->rb_left
;
1500 rb_node
= rb_node
->rb_right
;
1503 mm
->mmap_cache
= vma
;
1509 EXPORT_SYMBOL(find_vma
);
1511 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1512 struct vm_area_struct
*
1513 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1514 struct vm_area_struct
**pprev
)
1516 struct vm_area_struct
*vma
= NULL
, *prev
= NULL
;
1517 struct rb_node
*rb_node
;
1521 /* Guard against addr being lower than the first VMA */
1524 /* Go through the RB tree quickly. */
1525 rb_node
= mm
->mm_rb
.rb_node
;
1528 struct vm_area_struct
*vma_tmp
;
1529 vma_tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1531 if (addr
< vma_tmp
->vm_end
) {
1532 rb_node
= rb_node
->rb_left
;
1535 if (!prev
->vm_next
|| (addr
< prev
->vm_next
->vm_end
))
1537 rb_node
= rb_node
->rb_right
;
1543 return prev
? prev
->vm_next
: vma
;
1547 * Verify that the stack growth is acceptable and
1548 * update accounting. This is shared with both the
1549 * grow-up and grow-down cases.
1551 static int acct_stack_growth(struct vm_area_struct
*vma
, unsigned long size
, unsigned long grow
)
1553 struct mm_struct
*mm
= vma
->vm_mm
;
1554 struct rlimit
*rlim
= current
->signal
->rlim
;
1555 unsigned long new_start
;
1557 /* address space limit tests */
1558 if (!may_expand_vm(mm
, grow
))
1561 /* Stack limit test */
1562 if (size
> rlim
[RLIMIT_STACK
].rlim_cur
)
1565 /* mlock limit tests */
1566 if (vma
->vm_flags
& VM_LOCKED
) {
1567 unsigned long locked
;
1568 unsigned long limit
;
1569 locked
= mm
->locked_vm
+ grow
;
1570 limit
= rlim
[RLIMIT_MEMLOCK
].rlim_cur
>> PAGE_SHIFT
;
1571 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1575 /* Check to ensure the stack will not grow into a hugetlb-only region */
1576 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
1578 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
1582 * Overcommit.. This must be the final test, as it will
1583 * update security statistics.
1585 if (security_vm_enough_memory_mm(mm
, grow
))
1588 /* Ok, everything looks good - let it rip */
1589 mm
->total_vm
+= grow
;
1590 if (vma
->vm_flags
& VM_LOCKED
)
1591 mm
->locked_vm
+= grow
;
1592 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1596 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1598 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1599 * vma is the last one with address > vma->vm_end. Have to extend vma.
1604 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1608 if (!(vma
->vm_flags
& VM_GROWSUP
))
1612 * We must make sure the anon_vma is allocated
1613 * so that the anon_vma locking is not a noop.
1615 if (unlikely(anon_vma_prepare(vma
)))
1620 * vma->vm_start/vm_end cannot change under us because the caller
1621 * is required to hold the mmap_sem in read mode. We need the
1622 * anon_vma lock to serialize against concurrent expand_stacks.
1623 * Also guard against wrapping around to address 0.
1625 if (address
< PAGE_ALIGN(address
+4))
1626 address
= PAGE_ALIGN(address
+4);
1628 anon_vma_unlock(vma
);
1633 /* Somebody else might have raced and expanded it already */
1634 if (address
> vma
->vm_end
) {
1635 unsigned long size
, grow
;
1637 size
= address
- vma
->vm_start
;
1638 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1640 error
= acct_stack_growth(vma
, size
, grow
);
1642 vma
->vm_end
= address
;
1644 anon_vma_unlock(vma
);
1647 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1650 * vma is the first one with address < vma->vm_start. Have to extend vma.
1652 static int expand_downwards(struct vm_area_struct
*vma
,
1653 unsigned long address
)
1658 * We must make sure the anon_vma is allocated
1659 * so that the anon_vma locking is not a noop.
1661 if (unlikely(anon_vma_prepare(vma
)))
1664 address
&= PAGE_MASK
;
1665 error
= security_file_mmap(NULL
, 0, 0, 0, address
, 1);
1672 * vma->vm_start/vm_end cannot change under us because the caller
1673 * is required to hold the mmap_sem in read mode. We need the
1674 * anon_vma lock to serialize against concurrent expand_stacks.
1677 /* Somebody else might have raced and expanded it already */
1678 if (address
< vma
->vm_start
) {
1679 unsigned long size
, grow
;
1681 size
= vma
->vm_end
- address
;
1682 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1684 error
= acct_stack_growth(vma
, size
, grow
);
1686 vma
->vm_start
= address
;
1687 vma
->vm_pgoff
-= grow
;
1690 anon_vma_unlock(vma
);
1694 int expand_stack_downwards(struct vm_area_struct
*vma
, unsigned long address
)
1696 return expand_downwards(vma
, address
);
1699 #ifdef CONFIG_STACK_GROWSUP
1700 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1702 return expand_upwards(vma
, address
);
1705 struct vm_area_struct
*
1706 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1708 struct vm_area_struct
*vma
, *prev
;
1711 vma
= find_vma_prev(mm
, addr
, &prev
);
1712 if (vma
&& (vma
->vm_start
<= addr
))
1714 if (!prev
|| expand_stack(prev
, addr
))
1716 if (prev
->vm_flags
& VM_LOCKED
) {
1717 if (mlock_vma_pages_range(prev
, addr
, prev
->vm_end
) < 0)
1718 return NULL
; /* vma gone! */
1723 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1725 return expand_downwards(vma
, address
);
1728 struct vm_area_struct
*
1729 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1731 struct vm_area_struct
* vma
;
1732 unsigned long start
;
1735 vma
= find_vma(mm
,addr
);
1738 if (vma
->vm_start
<= addr
)
1740 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1742 start
= vma
->vm_start
;
1743 if (expand_stack(vma
, addr
))
1745 if (vma
->vm_flags
& VM_LOCKED
) {
1746 if (mlock_vma_pages_range(vma
, addr
, start
) < 0)
1747 return NULL
; /* vma gone! */
1754 * Ok - we have the memory areas we should free on the vma list,
1755 * so release them, and do the vma updates.
1757 * Called with the mm semaphore held.
1759 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1761 /* Update high watermark before we lower total_vm */
1762 update_hiwater_vm(mm
);
1764 long nrpages
= vma_pages(vma
);
1766 mm
->total_vm
-= nrpages
;
1767 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1768 vma
= remove_vma(vma
);
1774 * Get rid of page table information in the indicated region.
1776 * Called with the mm semaphore held.
1778 static void unmap_region(struct mm_struct
*mm
,
1779 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1780 unsigned long start
, unsigned long end
)
1782 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1783 struct mmu_gather
*tlb
;
1784 unsigned long nr_accounted
= 0;
1787 tlb
= tlb_gather_mmu(mm
, 0);
1788 update_hiwater_rss(mm
);
1789 unmap_vmas(&tlb
, vma
, start
, end
, &nr_accounted
, NULL
);
1790 vm_unacct_memory(nr_accounted
);
1791 free_pgtables(tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1792 next
? next
->vm_start
: 0);
1793 tlb_finish_mmu(tlb
, start
, end
);
1797 * Create a list of vma's touched by the unmap, removing them from the mm's
1798 * vma list as we go..
1801 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1802 struct vm_area_struct
*prev
, unsigned long end
)
1804 struct vm_area_struct
**insertion_point
;
1805 struct vm_area_struct
*tail_vma
= NULL
;
1808 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1810 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1814 } while (vma
&& vma
->vm_start
< end
);
1815 *insertion_point
= vma
;
1816 tail_vma
->vm_next
= NULL
;
1817 if (mm
->unmap_area
== arch_unmap_area
)
1818 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1820 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1821 mm
->unmap_area(mm
, addr
);
1822 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1826 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1827 * munmap path where it doesn't make sense to fail.
1829 static int __split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1830 unsigned long addr
, int new_below
)
1832 struct mempolicy
*pol
;
1833 struct vm_area_struct
*new;
1835 if (is_vm_hugetlb_page(vma
) && (addr
&
1836 ~(huge_page_mask(hstate_vma(vma
)))))
1839 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1843 /* most fields are the same, copy all, and then fixup */
1849 new->vm_start
= addr
;
1850 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
1853 pol
= mpol_dup(vma_policy(vma
));
1855 kmem_cache_free(vm_area_cachep
, new);
1856 return PTR_ERR(pol
);
1858 vma_set_policy(new, pol
);
1861 get_file(new->vm_file
);
1862 if (vma
->vm_flags
& VM_EXECUTABLE
)
1863 added_exe_file_vma(mm
);
1866 if (new->vm_ops
&& new->vm_ops
->open
)
1867 new->vm_ops
->open(new);
1870 vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
1871 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
1873 vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
1879 * Split a vma into two pieces at address 'addr', a new vma is allocated
1880 * either for the first part or the tail.
1882 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1883 unsigned long addr
, int new_below
)
1885 if (mm
->map_count
>= sysctl_max_map_count
)
1888 return __split_vma(mm
, vma
, addr
, new_below
);
1891 /* Munmap is split into 2 main parts -- this part which finds
1892 * what needs doing, and the areas themselves, which do the
1893 * work. This now handles partial unmappings.
1894 * Jeremy Fitzhardinge <jeremy@goop.org>
1896 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1899 struct vm_area_struct
*vma
, *prev
, *last
;
1901 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
1904 if ((len
= PAGE_ALIGN(len
)) == 0)
1907 /* Find the first overlapping VMA */
1908 vma
= find_vma_prev(mm
, start
, &prev
);
1911 /* we have start < vma->vm_end */
1913 /* if it doesn't overlap, we have nothing.. */
1915 if (vma
->vm_start
>= end
)
1919 * If we need to split any vma, do it now to save pain later.
1921 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1922 * unmapped vm_area_struct will remain in use: so lower split_vma
1923 * places tmp vma above, and higher split_vma places tmp vma below.
1925 if (start
> vma
->vm_start
) {
1929 * Make sure that map_count on return from munmap() will
1930 * not exceed its limit; but let map_count go just above
1931 * its limit temporarily, to help free resources as expected.
1933 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
1936 error
= __split_vma(mm
, vma
, start
, 0);
1942 /* Does it split the last one? */
1943 last
= find_vma(mm
, end
);
1944 if (last
&& end
> last
->vm_start
) {
1945 int error
= __split_vma(mm
, last
, end
, 1);
1949 vma
= prev
? prev
->vm_next
: mm
->mmap
;
1952 * unlock any mlock()ed ranges before detaching vmas
1954 if (mm
->locked_vm
) {
1955 struct vm_area_struct
*tmp
= vma
;
1956 while (tmp
&& tmp
->vm_start
< end
) {
1957 if (tmp
->vm_flags
& VM_LOCKED
) {
1958 mm
->locked_vm
-= vma_pages(tmp
);
1959 munlock_vma_pages_all(tmp
);
1966 * Remove the vma's, and unmap the actual pages
1968 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
1969 unmap_region(mm
, vma
, prev
, start
, end
);
1971 /* Fix up all other VM information */
1972 remove_vma_list(mm
, vma
);
1977 EXPORT_SYMBOL(do_munmap
);
1979 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1982 struct mm_struct
*mm
= current
->mm
;
1984 profile_munmap(addr
);
1986 down_write(&mm
->mmap_sem
);
1987 ret
= do_munmap(mm
, addr
, len
);
1988 up_write(&mm
->mmap_sem
);
1992 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
1994 #ifdef CONFIG_DEBUG_VM
1995 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
1997 up_read(&mm
->mmap_sem
);
2003 * this is really a simplified "do_mmap". it only handles
2004 * anonymous maps. eventually we may be able to do some
2005 * brk-specific accounting here.
2007 unsigned long do_brk(unsigned long addr
, unsigned long len
)
2009 struct mm_struct
* mm
= current
->mm
;
2010 struct vm_area_struct
* vma
, * prev
;
2011 unsigned long flags
;
2012 struct rb_node
** rb_link
, * rb_parent
;
2013 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2016 len
= PAGE_ALIGN(len
);
2020 error
= security_file_mmap(NULL
, 0, 0, 0, addr
, 1);
2024 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2026 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2027 if (error
& ~PAGE_MASK
)
2033 if (mm
->def_flags
& VM_LOCKED
) {
2034 unsigned long locked
, lock_limit
;
2035 locked
= len
>> PAGE_SHIFT
;
2036 locked
+= mm
->locked_vm
;
2037 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
2038 lock_limit
>>= PAGE_SHIFT
;
2039 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
2044 * mm->mmap_sem is required to protect against another thread
2045 * changing the mappings in case we sleep.
2047 verify_mm_writelocked(mm
);
2050 * Clear old maps. this also does some error checking for us
2053 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2054 if (vma
&& vma
->vm_start
< addr
+ len
) {
2055 if (do_munmap(mm
, addr
, len
))
2060 /* Check against address space limits *after* clearing old maps... */
2061 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2064 if (mm
->map_count
> sysctl_max_map_count
)
2067 if (security_vm_enough_memory(len
>> PAGE_SHIFT
))
2070 /* Can we just expand an old private anonymous mapping? */
2071 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2072 NULL
, NULL
, pgoff
, NULL
);
2077 * create a vma struct for an anonymous mapping
2079 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2081 vm_unacct_memory(len
>> PAGE_SHIFT
);
2086 vma
->vm_start
= addr
;
2087 vma
->vm_end
= addr
+ len
;
2088 vma
->vm_pgoff
= pgoff
;
2089 vma
->vm_flags
= flags
;
2090 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2091 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2093 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2094 if (flags
& VM_LOCKED
) {
2095 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
2096 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2101 EXPORT_SYMBOL(do_brk
);
2103 /* Release all mmaps. */
2104 void exit_mmap(struct mm_struct
*mm
)
2106 struct mmu_gather
*tlb
;
2107 struct vm_area_struct
*vma
;
2108 unsigned long nr_accounted
= 0;
2111 /* mm's last user has gone, and its about to be pulled down */
2112 mmu_notifier_release(mm
);
2114 if (mm
->locked_vm
) {
2117 if (vma
->vm_flags
& VM_LOCKED
)
2118 munlock_vma_pages_all(vma
);
2126 if (!vma
) /* Can happen if dup_mmap() received an OOM */
2131 tlb
= tlb_gather_mmu(mm
, 1);
2132 /* update_hiwater_rss(mm) here? but nobody should be looking */
2133 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2134 end
= unmap_vmas(&tlb
, vma
, 0, -1, &nr_accounted
, NULL
);
2135 vm_unacct_memory(nr_accounted
);
2137 free_pgtables(tlb
, vma
, FIRST_USER_ADDRESS
, 0);
2138 tlb_finish_mmu(tlb
, 0, end
);
2141 * Walk the list again, actually closing and freeing it,
2142 * with preemption enabled, without holding any MM locks.
2145 vma
= remove_vma(vma
);
2147 BUG_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2150 /* Insert vm structure into process list sorted by address
2151 * and into the inode's i_mmap tree. If vm_file is non-NULL
2152 * then i_mmap_lock is taken here.
2154 int insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
2156 struct vm_area_struct
* __vma
, * prev
;
2157 struct rb_node
** rb_link
, * rb_parent
;
2160 * The vm_pgoff of a purely anonymous vma should be irrelevant
2161 * until its first write fault, when page's anon_vma and index
2162 * are set. But now set the vm_pgoff it will almost certainly
2163 * end up with (unless mremap moves it elsewhere before that
2164 * first wfault), so /proc/pid/maps tells a consistent story.
2166 * By setting it to reflect the virtual start address of the
2167 * vma, merges and splits can happen in a seamless way, just
2168 * using the existing file pgoff checks and manipulations.
2169 * Similarly in do_mmap_pgoff and in do_brk.
2171 if (!vma
->vm_file
) {
2172 BUG_ON(vma
->anon_vma
);
2173 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2175 __vma
= find_vma_prepare(mm
,vma
->vm_start
,&prev
,&rb_link
,&rb_parent
);
2176 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
2178 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2179 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2181 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2186 * Copy the vma structure to a new location in the same mm,
2187 * prior to moving page table entries, to effect an mremap move.
2189 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2190 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
2192 struct vm_area_struct
*vma
= *vmap
;
2193 unsigned long vma_start
= vma
->vm_start
;
2194 struct mm_struct
*mm
= vma
->vm_mm
;
2195 struct vm_area_struct
*new_vma
, *prev
;
2196 struct rb_node
**rb_link
, *rb_parent
;
2197 struct mempolicy
*pol
;
2200 * If anonymous vma has not yet been faulted, update new pgoff
2201 * to match new location, to increase its chance of merging.
2203 if (!vma
->vm_file
&& !vma
->anon_vma
)
2204 pgoff
= addr
>> PAGE_SHIFT
;
2206 find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2207 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2208 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2211 * Source vma may have been merged into new_vma
2213 if (vma_start
>= new_vma
->vm_start
&&
2214 vma_start
< new_vma
->vm_end
)
2217 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2220 pol
= mpol_dup(vma_policy(vma
));
2222 kmem_cache_free(vm_area_cachep
, new_vma
);
2225 vma_set_policy(new_vma
, pol
);
2226 new_vma
->vm_start
= addr
;
2227 new_vma
->vm_end
= addr
+ len
;
2228 new_vma
->vm_pgoff
= pgoff
;
2229 if (new_vma
->vm_file
) {
2230 get_file(new_vma
->vm_file
);
2231 if (vma
->vm_flags
& VM_EXECUTABLE
)
2232 added_exe_file_vma(mm
);
2234 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2235 new_vma
->vm_ops
->open(new_vma
);
2236 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2243 * Return true if the calling process may expand its vm space by the passed
2246 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2248 unsigned long cur
= mm
->total_vm
; /* pages */
2251 lim
= current
->signal
->rlim
[RLIMIT_AS
].rlim_cur
>> PAGE_SHIFT
;
2253 if (cur
+ npages
> lim
)
2259 static int special_mapping_fault(struct vm_area_struct
*vma
,
2260 struct vm_fault
*vmf
)
2263 struct page
**pages
;
2266 * special mappings have no vm_file, and in that case, the mm
2267 * uses vm_pgoff internally. So we have to subtract it from here.
2268 * We are allowed to do this because we are the mm; do not copy
2269 * this code into drivers!
2271 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2273 for (pages
= vma
->vm_private_data
; pgoff
&& *pages
; ++pages
)
2277 struct page
*page
= *pages
;
2283 return VM_FAULT_SIGBUS
;
2287 * Having a close hook prevents vma merging regardless of flags.
2289 static void special_mapping_close(struct vm_area_struct
*vma
)
2293 static const struct vm_operations_struct special_mapping_vmops
= {
2294 .close
= special_mapping_close
,
2295 .fault
= special_mapping_fault
,
2299 * Called with mm->mmap_sem held for writing.
2300 * Insert a new vma covering the given region, with the given flags.
2301 * Its pages are supplied by the given array of struct page *.
2302 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2303 * The region past the last page supplied will always produce SIGBUS.
2304 * The array pointer and the pages it points to are assumed to stay alive
2305 * for as long as this mapping might exist.
2307 int install_special_mapping(struct mm_struct
*mm
,
2308 unsigned long addr
, unsigned long len
,
2309 unsigned long vm_flags
, struct page
**pages
)
2311 struct vm_area_struct
*vma
;
2313 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2314 if (unlikely(vma
== NULL
))
2318 vma
->vm_start
= addr
;
2319 vma
->vm_end
= addr
+ len
;
2321 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
;
2322 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
2324 vma
->vm_ops
= &special_mapping_vmops
;
2325 vma
->vm_private_data
= pages
;
2327 if (unlikely(insert_vm_struct(mm
, vma
))) {
2328 kmem_cache_free(vm_area_cachep
, vma
);
2332 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2334 perf_event_mmap(vma
);
2339 static DEFINE_MUTEX(mm_all_locks_mutex
);
2341 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
2343 if (!test_bit(0, (unsigned long *) &anon_vma
->head
.next
)) {
2345 * The LSB of head.next can't change from under us
2346 * because we hold the mm_all_locks_mutex.
2348 spin_lock_nest_lock(&anon_vma
->lock
, &mm
->mmap_sem
);
2350 * We can safely modify head.next after taking the
2351 * anon_vma->lock. If some other vma in this mm shares
2352 * the same anon_vma we won't take it again.
2354 * No need of atomic instructions here, head.next
2355 * can't change from under us thanks to the
2358 if (__test_and_set_bit(0, (unsigned long *)
2359 &anon_vma
->head
.next
))
2364 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
2366 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2368 * AS_MM_ALL_LOCKS can't change from under us because
2369 * we hold the mm_all_locks_mutex.
2371 * Operations on ->flags have to be atomic because
2372 * even if AS_MM_ALL_LOCKS is stable thanks to the
2373 * mm_all_locks_mutex, there may be other cpus
2374 * changing other bitflags in parallel to us.
2376 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
2378 spin_lock_nest_lock(&mapping
->i_mmap_lock
, &mm
->mmap_sem
);
2383 * This operation locks against the VM for all pte/vma/mm related
2384 * operations that could ever happen on a certain mm. This includes
2385 * vmtruncate, try_to_unmap, and all page faults.
2387 * The caller must take the mmap_sem in write mode before calling
2388 * mm_take_all_locks(). The caller isn't allowed to release the
2389 * mmap_sem until mm_drop_all_locks() returns.
2391 * mmap_sem in write mode is required in order to block all operations
2392 * that could modify pagetables and free pages without need of
2393 * altering the vma layout (for example populate_range() with
2394 * nonlinear vmas). It's also needed in write mode to avoid new
2395 * anon_vmas to be associated with existing vmas.
2397 * A single task can't take more than one mm_take_all_locks() in a row
2398 * or it would deadlock.
2400 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2401 * mapping->flags avoid to take the same lock twice, if more than one
2402 * vma in this mm is backed by the same anon_vma or address_space.
2404 * We can take all the locks in random order because the VM code
2405 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2406 * takes more than one of them in a row. Secondly we're protected
2407 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2409 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2410 * that may have to take thousand of locks.
2412 * mm_take_all_locks() can fail if it's interrupted by signals.
2414 int mm_take_all_locks(struct mm_struct
*mm
)
2416 struct vm_area_struct
*vma
;
2419 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2421 mutex_lock(&mm_all_locks_mutex
);
2423 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2424 if (signal_pending(current
))
2426 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2427 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
2430 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2431 if (signal_pending(current
))
2434 vm_lock_anon_vma(mm
, vma
->anon_vma
);
2441 mm_drop_all_locks(mm
);
2446 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
2448 if (test_bit(0, (unsigned long *) &anon_vma
->head
.next
)) {
2450 * The LSB of head.next can't change to 0 from under
2451 * us because we hold the mm_all_locks_mutex.
2453 * We must however clear the bitflag before unlocking
2454 * the vma so the users using the anon_vma->head will
2455 * never see our bitflag.
2457 * No need of atomic instructions here, head.next
2458 * can't change from under us until we release the
2461 if (!__test_and_clear_bit(0, (unsigned long *)
2462 &anon_vma
->head
.next
))
2464 spin_unlock(&anon_vma
->lock
);
2468 static void vm_unlock_mapping(struct address_space
*mapping
)
2470 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2472 * AS_MM_ALL_LOCKS can't change to 0 from under us
2473 * because we hold the mm_all_locks_mutex.
2475 spin_unlock(&mapping
->i_mmap_lock
);
2476 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
2483 * The mmap_sem cannot be released by the caller until
2484 * mm_drop_all_locks() returns.
2486 void mm_drop_all_locks(struct mm_struct
*mm
)
2488 struct vm_area_struct
*vma
;
2490 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2491 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
2493 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2495 vm_unlock_anon_vma(vma
->anon_vma
);
2496 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2497 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
2500 mutex_unlock(&mm_all_locks_mutex
);
2504 * initialise the VMA slab
2506 void __init
mmap_init(void)
2510 ret
= percpu_counter_init(&vm_committed_as
, 0);