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/ima.h>
24 #include <linux/hugetlb.h>
25 #include <linux/profile.h>
26 #include <linux/module.h>
27 #include <linux/mount.h>
28 #include <linux/mempolicy.h>
29 #include <linux/rmap.h>
30 #include <linux/ksm.h>
31 #include <linux/mmu_notifier.h>
32 #include <linux/perf_event.h>
34 #include <asm/uaccess.h>
35 #include <asm/cacheflush.h>
37 #include <asm/mmu_context.h>
41 #ifndef arch_mmap_check
42 #define arch_mmap_check(addr, len, flags) (0)
45 #ifndef arch_rebalance_pgtables
46 #define arch_rebalance_pgtables(addr, len) (addr)
49 static void unmap_region(struct mm_struct
*mm
,
50 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
51 unsigned long start
, unsigned long end
);
54 * WARNING: the debugging will use recursive algorithms so never enable this
55 * unless you know what you are doing.
59 /* description of effects of mapping type and prot in current implementation.
60 * this is due to the limited x86 page protection hardware. The expected
61 * behavior is in parens:
64 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
65 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
66 * w: (no) no w: (no) no w: (yes) yes w: (no) no
67 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
69 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
70 * w: (no) no w: (no) no w: (copy) copy w: (no) no
71 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
74 pgprot_t protection_map
[16] = {
75 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
76 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
79 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
81 return __pgprot(pgprot_val(protection_map
[vm_flags
&
82 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
83 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
85 EXPORT_SYMBOL(vm_get_page_prot
);
87 int sysctl_overcommit_memory
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
88 int sysctl_overcommit_ratio
= 50; /* default is 50% */
89 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
90 struct percpu_counter vm_committed_as
;
93 * Check that a process has enough memory to allocate a new virtual
94 * mapping. 0 means there is enough memory for the allocation to
95 * succeed and -ENOMEM implies there is not.
97 * We currently support three overcommit policies, which are set via the
98 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
100 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
101 * Additional code 2002 Jul 20 by Robert Love.
103 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
105 * Note this is a helper function intended to be used by LSMs which
106 * wish to use this logic.
108 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
110 unsigned long free
, allowed
;
112 vm_acct_memory(pages
);
115 * Sometimes we want to use more memory than we have
117 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
120 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
123 free
= global_page_state(NR_FILE_PAGES
);
124 free
+= nr_swap_pages
;
127 * Any slabs which are created with the
128 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
129 * which are reclaimable, under pressure. The dentry
130 * cache and most inode caches should fall into this
132 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
135 * Leave the last 3% for root
144 * nr_free_pages() is very expensive on large systems,
145 * only call if we're about to fail.
150 * Leave reserved pages. The pages are not for anonymous pages.
152 if (n
<= totalreserve_pages
)
155 n
-= totalreserve_pages
;
158 * Leave the last 3% for root
170 allowed
= (totalram_pages
- hugetlb_total_pages())
171 * sysctl_overcommit_ratio
/ 100;
173 * Leave the last 3% for root
176 allowed
-= allowed
/ 32;
177 allowed
+= total_swap_pages
;
179 /* Don't let a single process grow too big:
180 leave 3% of the size of this process for other processes */
182 allowed
-= mm
->total_vm
/ 32;
184 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
187 vm_unacct_memory(pages
);
193 * Requires inode->i_mapping->i_mmap_lock
195 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
196 struct file
*file
, struct address_space
*mapping
)
198 if (vma
->vm_flags
& VM_DENYWRITE
)
199 atomic_inc(&file
->f_path
.dentry
->d_inode
->i_writecount
);
200 if (vma
->vm_flags
& VM_SHARED
)
201 mapping
->i_mmap_writable
--;
203 flush_dcache_mmap_lock(mapping
);
204 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
205 list_del_init(&vma
->shared
.vm_set
.list
);
207 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
208 flush_dcache_mmap_unlock(mapping
);
212 * Unlink a file-based vm structure from its prio_tree, to hide
213 * vma from rmap and vmtruncate before freeing its page tables.
215 void unlink_file_vma(struct vm_area_struct
*vma
)
217 struct file
*file
= vma
->vm_file
;
220 struct address_space
*mapping
= file
->f_mapping
;
221 spin_lock(&mapping
->i_mmap_lock
);
222 __remove_shared_vm_struct(vma
, file
, mapping
);
223 spin_unlock(&mapping
->i_mmap_lock
);
228 * Close a vm structure and free it, returning the next.
230 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
232 struct vm_area_struct
*next
= vma
->vm_next
;
235 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
236 vma
->vm_ops
->close(vma
);
239 if (vma
->vm_flags
& VM_EXECUTABLE
)
240 removed_exe_file_vma(vma
->vm_mm
);
242 mpol_put(vma_policy(vma
));
243 kmem_cache_free(vm_area_cachep
, vma
);
247 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
249 unsigned long rlim
, retval
;
250 unsigned long newbrk
, oldbrk
;
251 struct mm_struct
*mm
= current
->mm
;
252 unsigned long min_brk
;
254 down_write(&mm
->mmap_sem
);
256 #ifdef CONFIG_COMPAT_BRK
257 min_brk
= mm
->end_code
;
259 min_brk
= mm
->start_brk
;
265 * Check against rlimit here. If this check is done later after the test
266 * of oldbrk with newbrk then it can escape the test and let the data
267 * segment grow beyond its set limit the in case where the limit is
268 * not page aligned -Ram Gupta
270 rlim
= current
->signal
->rlim
[RLIMIT_DATA
].rlim_cur
;
271 if (rlim
< RLIM_INFINITY
&& (brk
- mm
->start_brk
) +
272 (mm
->end_data
- mm
->start_data
) > rlim
)
275 newbrk
= PAGE_ALIGN(brk
);
276 oldbrk
= PAGE_ALIGN(mm
->brk
);
277 if (oldbrk
== newbrk
)
280 /* Always allow shrinking brk. */
281 if (brk
<= mm
->brk
) {
282 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
287 /* Check against existing mmap mappings. */
288 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
291 /* Ok, looks good - let it rip. */
292 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
298 up_write(&mm
->mmap_sem
);
303 static int browse_rb(struct rb_root
*root
)
306 struct rb_node
*nd
, *pn
= NULL
;
307 unsigned long prev
= 0, pend
= 0;
309 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
310 struct vm_area_struct
*vma
;
311 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
312 if (vma
->vm_start
< prev
)
313 printk("vm_start %lx prev %lx\n", vma
->vm_start
, prev
), i
= -1;
314 if (vma
->vm_start
< pend
)
315 printk("vm_start %lx pend %lx\n", vma
->vm_start
, pend
);
316 if (vma
->vm_start
> vma
->vm_end
)
317 printk("vm_end %lx < vm_start %lx\n", vma
->vm_end
, vma
->vm_start
);
320 prev
= vma
->vm_start
;
324 for (nd
= pn
; nd
; nd
= rb_prev(nd
)) {
328 printk("backwards %d, forwards %d\n", j
, i
), i
= 0;
332 void validate_mm(struct mm_struct
*mm
)
336 struct vm_area_struct
*tmp
= mm
->mmap
;
341 if (i
!= mm
->map_count
)
342 printk("map_count %d vm_next %d\n", mm
->map_count
, i
), bug
= 1;
343 i
= browse_rb(&mm
->mm_rb
);
344 if (i
!= mm
->map_count
)
345 printk("map_count %d rb %d\n", mm
->map_count
, i
), bug
= 1;
349 #define validate_mm(mm) do { } while (0)
352 static struct vm_area_struct
*
353 find_vma_prepare(struct mm_struct
*mm
, unsigned long addr
,
354 struct vm_area_struct
**pprev
, struct rb_node
***rb_link
,
355 struct rb_node
** rb_parent
)
357 struct vm_area_struct
* vma
;
358 struct rb_node
** __rb_link
, * __rb_parent
, * rb_prev
;
360 __rb_link
= &mm
->mm_rb
.rb_node
;
361 rb_prev
= __rb_parent
= NULL
;
365 struct vm_area_struct
*vma_tmp
;
367 __rb_parent
= *__rb_link
;
368 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
370 if (vma_tmp
->vm_end
> addr
) {
372 if (vma_tmp
->vm_start
<= addr
)
374 __rb_link
= &__rb_parent
->rb_left
;
376 rb_prev
= __rb_parent
;
377 __rb_link
= &__rb_parent
->rb_right
;
383 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
384 *rb_link
= __rb_link
;
385 *rb_parent
= __rb_parent
;
390 __vma_link_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
391 struct vm_area_struct
*prev
, struct rb_node
*rb_parent
)
394 vma
->vm_next
= prev
->vm_next
;
399 vma
->vm_next
= rb_entry(rb_parent
,
400 struct vm_area_struct
, vm_rb
);
406 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
407 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
409 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
410 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
413 static void __vma_link_file(struct vm_area_struct
*vma
)
419 struct address_space
*mapping
= file
->f_mapping
;
421 if (vma
->vm_flags
& VM_DENYWRITE
)
422 atomic_dec(&file
->f_path
.dentry
->d_inode
->i_writecount
);
423 if (vma
->vm_flags
& VM_SHARED
)
424 mapping
->i_mmap_writable
++;
426 flush_dcache_mmap_lock(mapping
);
427 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
428 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
430 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
431 flush_dcache_mmap_unlock(mapping
);
436 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
437 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
438 struct rb_node
*rb_parent
)
440 __vma_link_list(mm
, vma
, prev
, rb_parent
);
441 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
442 __anon_vma_link(vma
);
445 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
446 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
447 struct rb_node
*rb_parent
)
449 struct address_space
*mapping
= NULL
;
452 mapping
= vma
->vm_file
->f_mapping
;
455 spin_lock(&mapping
->i_mmap_lock
);
456 vma
->vm_truncate_count
= mapping
->truncate_count
;
460 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
461 __vma_link_file(vma
);
463 anon_vma_unlock(vma
);
465 spin_unlock(&mapping
->i_mmap_lock
);
472 * Helper for vma_adjust in the split_vma insert case:
473 * insert vm structure into list and rbtree and anon_vma,
474 * but it has already been inserted into prio_tree earlier.
476 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
478 struct vm_area_struct
*__vma
, *prev
;
479 struct rb_node
**rb_link
, *rb_parent
;
481 __vma
= find_vma_prepare(mm
, vma
->vm_start
,&prev
, &rb_link
, &rb_parent
);
482 BUG_ON(__vma
&& __vma
->vm_start
< vma
->vm_end
);
483 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
488 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
489 struct vm_area_struct
*prev
)
491 prev
->vm_next
= vma
->vm_next
;
492 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
493 if (mm
->mmap_cache
== vma
)
494 mm
->mmap_cache
= prev
;
498 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
499 * is already present in an i_mmap tree without adjusting the tree.
500 * The following helper function should be used when such adjustments
501 * are necessary. The "insert" vma (if any) is to be inserted
502 * before we drop the necessary locks.
504 void vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
505 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
507 struct mm_struct
*mm
= vma
->vm_mm
;
508 struct vm_area_struct
*next
= vma
->vm_next
;
509 struct vm_area_struct
*importer
= NULL
;
510 struct address_space
*mapping
= NULL
;
511 struct prio_tree_root
*root
= NULL
;
512 struct file
*file
= vma
->vm_file
;
513 struct anon_vma
*anon_vma
= NULL
;
514 long adjust_next
= 0;
517 if (next
&& !insert
) {
518 if (end
>= next
->vm_end
) {
520 * vma expands, overlapping all the next, and
521 * perhaps the one after too (mprotect case 6).
523 again
: remove_next
= 1 + (end
> next
->vm_end
);
525 anon_vma
= next
->anon_vma
;
527 } else if (end
> next
->vm_start
) {
529 * vma expands, overlapping part of the next:
530 * mprotect case 5 shifting the boundary up.
532 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
533 anon_vma
= next
->anon_vma
;
535 } else if (end
< vma
->vm_end
) {
537 * vma shrinks, and !insert tells it's not
538 * split_vma inserting another: so it must be
539 * mprotect case 4 shifting the boundary down.
541 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
542 anon_vma
= next
->anon_vma
;
548 mapping
= file
->f_mapping
;
549 if (!(vma
->vm_flags
& VM_NONLINEAR
))
550 root
= &mapping
->i_mmap
;
551 spin_lock(&mapping
->i_mmap_lock
);
553 vma
->vm_truncate_count
!= next
->vm_truncate_count
) {
555 * unmap_mapping_range might be in progress:
556 * ensure that the expanding vma is rescanned.
558 importer
->vm_truncate_count
= 0;
561 insert
->vm_truncate_count
= vma
->vm_truncate_count
;
563 * Put into prio_tree now, so instantiated pages
564 * are visible to arm/parisc __flush_dcache_page
565 * throughout; but we cannot insert into address
566 * space until vma start or end is updated.
568 __vma_link_file(insert
);
573 * When changing only vma->vm_end, we don't really need
574 * anon_vma lock: but is that case worth optimizing out?
577 anon_vma
= vma
->anon_vma
;
579 spin_lock(&anon_vma
->lock
);
581 * Easily overlooked: when mprotect shifts the boundary,
582 * make sure the expanding vma has anon_vma set if the
583 * shrinking vma had, to cover any anon pages imported.
585 if (importer
&& !importer
->anon_vma
) {
586 importer
->anon_vma
= anon_vma
;
587 __anon_vma_link(importer
);
592 flush_dcache_mmap_lock(mapping
);
593 vma_prio_tree_remove(vma
, root
);
595 vma_prio_tree_remove(next
, root
);
598 vma
->vm_start
= start
;
600 vma
->vm_pgoff
= pgoff
;
602 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
603 next
->vm_pgoff
+= adjust_next
;
608 vma_prio_tree_insert(next
, root
);
609 vma_prio_tree_insert(vma
, root
);
610 flush_dcache_mmap_unlock(mapping
);
615 * vma_merge has merged next into vma, and needs
616 * us to remove next before dropping the locks.
618 __vma_unlink(mm
, next
, vma
);
620 __remove_shared_vm_struct(next
, file
, mapping
);
622 __anon_vma_merge(vma
, next
);
625 * split_vma has split insert from vma, and needs
626 * us to insert it before dropping the locks
627 * (it may either follow vma or precede it).
629 __insert_vm_struct(mm
, insert
);
633 spin_unlock(&anon_vma
->lock
);
635 spin_unlock(&mapping
->i_mmap_lock
);
640 if (next
->vm_flags
& VM_EXECUTABLE
)
641 removed_exe_file_vma(mm
);
644 mpol_put(vma_policy(next
));
645 kmem_cache_free(vm_area_cachep
, next
);
647 * In mprotect's case 6 (see comments on vma_merge),
648 * we must remove another next too. It would clutter
649 * up the code too much to do both in one go.
651 if (remove_next
== 2) {
661 * If the vma has a ->close operation then the driver probably needs to release
662 * per-vma resources, so we don't attempt to merge those.
664 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
665 struct file
*file
, unsigned long vm_flags
)
667 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
668 if ((vma
->vm_flags
^ vm_flags
) & ~VM_CAN_NONLINEAR
)
670 if (vma
->vm_file
!= file
)
672 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
677 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
678 struct anon_vma
*anon_vma2
)
680 return !anon_vma1
|| !anon_vma2
|| (anon_vma1
== anon_vma2
);
684 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
685 * in front of (at a lower virtual address and file offset than) the vma.
687 * We cannot merge two vmas if they have differently assigned (non-NULL)
688 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
690 * We don't check here for the merged mmap wrapping around the end of pagecache
691 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
692 * wrap, nor mmaps which cover the final page at index -1UL.
695 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
696 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
698 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
699 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
700 if (vma
->vm_pgoff
== vm_pgoff
)
707 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
708 * beyond (at a higher virtual address and file offset than) the vma.
710 * We cannot merge two vmas if they have differently assigned (non-NULL)
711 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
714 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
715 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
717 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
718 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
720 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
721 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
728 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
729 * whether that can be merged with its predecessor or its successor.
730 * Or both (it neatly fills a hole).
732 * In most cases - when called for mmap, brk or mremap - [addr,end) is
733 * certain not to be mapped by the time vma_merge is called; but when
734 * called for mprotect, it is certain to be already mapped (either at
735 * an offset within prev, or at the start of next), and the flags of
736 * this area are about to be changed to vm_flags - and the no-change
737 * case has already been eliminated.
739 * The following mprotect cases have to be considered, where AAAA is
740 * the area passed down from mprotect_fixup, never extending beyond one
741 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
743 * AAAA AAAA AAAA AAAA
744 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
745 * cannot merge might become might become might become
746 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
747 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
748 * mremap move: PPPPNNNNNNNN 8
750 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
751 * might become case 1 below case 2 below case 3 below
753 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
754 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
756 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
757 struct vm_area_struct
*prev
, unsigned long addr
,
758 unsigned long end
, unsigned long vm_flags
,
759 struct anon_vma
*anon_vma
, struct file
*file
,
760 pgoff_t pgoff
, struct mempolicy
*policy
)
762 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
763 struct vm_area_struct
*area
, *next
;
766 * We later require that vma->vm_flags == vm_flags,
767 * so this tests vma->vm_flags & VM_SPECIAL, too.
769 if (vm_flags
& VM_SPECIAL
)
773 next
= prev
->vm_next
;
777 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
778 next
= next
->vm_next
;
781 * Can it merge with the predecessor?
783 if (prev
&& prev
->vm_end
== addr
&&
784 mpol_equal(vma_policy(prev
), policy
) &&
785 can_vma_merge_after(prev
, vm_flags
,
786 anon_vma
, file
, pgoff
)) {
788 * OK, it can. Can we now merge in the successor as well?
790 if (next
&& end
== next
->vm_start
&&
791 mpol_equal(policy
, vma_policy(next
)) &&
792 can_vma_merge_before(next
, vm_flags
,
793 anon_vma
, file
, pgoff
+pglen
) &&
794 is_mergeable_anon_vma(prev
->anon_vma
,
797 vma_adjust(prev
, prev
->vm_start
,
798 next
->vm_end
, prev
->vm_pgoff
, NULL
);
799 } else /* cases 2, 5, 7 */
800 vma_adjust(prev
, prev
->vm_start
,
801 end
, prev
->vm_pgoff
, NULL
);
806 * Can this new request be merged in front of next?
808 if (next
&& end
== next
->vm_start
&&
809 mpol_equal(policy
, vma_policy(next
)) &&
810 can_vma_merge_before(next
, vm_flags
,
811 anon_vma
, file
, pgoff
+pglen
)) {
812 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
813 vma_adjust(prev
, prev
->vm_start
,
814 addr
, prev
->vm_pgoff
, NULL
);
815 else /* cases 3, 8 */
816 vma_adjust(area
, addr
, next
->vm_end
,
817 next
->vm_pgoff
- pglen
, NULL
);
825 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
826 * neighbouring vmas for a suitable anon_vma, before it goes off
827 * to allocate a new anon_vma. It checks because a repetitive
828 * sequence of mprotects and faults may otherwise lead to distinct
829 * anon_vmas being allocated, preventing vma merge in subsequent
832 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
834 struct vm_area_struct
*near
;
835 unsigned long vm_flags
;
842 * Since only mprotect tries to remerge vmas, match flags
843 * which might be mprotected into each other later on.
844 * Neither mlock nor madvise tries to remerge at present,
845 * so leave their flags as obstructing a merge.
847 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
848 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
850 if (near
->anon_vma
&& vma
->vm_end
== near
->vm_start
&&
851 mpol_equal(vma_policy(vma
), vma_policy(near
)) &&
852 can_vma_merge_before(near
, vm_flags
,
853 NULL
, vma
->vm_file
, vma
->vm_pgoff
+
854 ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
)))
855 return near
->anon_vma
;
858 * It is potentially slow to have to call find_vma_prev here.
859 * But it's only on the first write fault on the vma, not
860 * every time, and we could devise a way to avoid it later
861 * (e.g. stash info in next's anon_vma_node when assigning
862 * an anon_vma, or when trying vma_merge). Another time.
864 BUG_ON(find_vma_prev(vma
->vm_mm
, vma
->vm_start
, &near
) != vma
);
868 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
869 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
871 if (near
->anon_vma
&& near
->vm_end
== vma
->vm_start
&&
872 mpol_equal(vma_policy(near
), vma_policy(vma
)) &&
873 can_vma_merge_after(near
, vm_flags
,
874 NULL
, vma
->vm_file
, vma
->vm_pgoff
))
875 return near
->anon_vma
;
878 * There's no absolute need to look only at touching neighbours:
879 * we could search further afield for "compatible" anon_vmas.
880 * But it would probably just be a waste of time searching,
881 * or lead to too many vmas hanging off the same anon_vma.
882 * We're trying to allow mprotect remerging later on,
883 * not trying to minimize memory used for anon_vmas.
888 #ifdef CONFIG_PROC_FS
889 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
890 struct file
*file
, long pages
)
892 const unsigned long stack_flags
893 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
896 mm
->shared_vm
+= pages
;
897 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
898 mm
->exec_vm
+= pages
;
899 } else if (flags
& stack_flags
)
900 mm
->stack_vm
+= pages
;
901 if (flags
& (VM_RESERVED
|VM_IO
))
902 mm
->reserved_vm
+= pages
;
904 #endif /* CONFIG_PROC_FS */
907 * The caller must hold down_write(¤t->mm->mmap_sem).
910 unsigned long do_mmap_pgoff(struct file
*file
, unsigned long addr
,
911 unsigned long len
, unsigned long prot
,
912 unsigned long flags
, unsigned long pgoff
)
914 struct mm_struct
* mm
= current
->mm
;
916 unsigned int vm_flags
;
918 unsigned long reqprot
= prot
;
921 * Does the application expect PROT_READ to imply PROT_EXEC?
923 * (the exception is when the underlying filesystem is noexec
924 * mounted, in which case we dont add PROT_EXEC.)
926 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
927 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
933 if (!(flags
& MAP_FIXED
))
934 addr
= round_hint_to_min(addr
);
936 error
= arch_mmap_check(addr
, len
, flags
);
940 /* Careful about overflows.. */
941 len
= PAGE_ALIGN(len
);
942 if (!len
|| len
> TASK_SIZE
)
945 /* offset overflow? */
946 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
949 /* Too many mappings? */
950 if (mm
->map_count
> sysctl_max_map_count
)
953 /* Obtain the address to map to. we verify (or select) it and ensure
954 * that it represents a valid section of the address space.
956 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
957 if (addr
& ~PAGE_MASK
)
960 /* Do simple checking here so the lower-level routines won't have
961 * to. we assume access permissions have been handled by the open
962 * of the memory object, so we don't do any here.
964 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
965 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
967 if (flags
& MAP_LOCKED
) {
970 vm_flags
|= VM_LOCKED
;
973 /* mlock MCL_FUTURE? */
974 if (vm_flags
& VM_LOCKED
) {
975 unsigned long locked
, lock_limit
;
976 locked
= len
>> PAGE_SHIFT
;
977 locked
+= mm
->locked_vm
;
978 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
979 lock_limit
>>= PAGE_SHIFT
;
980 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
984 inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
987 switch (flags
& MAP_TYPE
) {
989 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
993 * Make sure we don't allow writing to an append-only
996 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1000 * Make sure there are no mandatory locks on the file.
1002 if (locks_verify_locked(inode
))
1005 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1006 if (!(file
->f_mode
& FMODE_WRITE
))
1007 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1011 if (!(file
->f_mode
& FMODE_READ
))
1013 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1014 if (vm_flags
& VM_EXEC
)
1016 vm_flags
&= ~VM_MAYEXEC
;
1019 if (!file
->f_op
|| !file
->f_op
->mmap
)
1027 switch (flags
& MAP_TYPE
) {
1033 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1037 * Set pgoff according to addr for anon_vma.
1039 pgoff
= addr
>> PAGE_SHIFT
;
1046 error
= security_file_mmap(file
, reqprot
, prot
, flags
, addr
, 0);
1049 error
= ima_file_mmap(file
, prot
);
1053 return mmap_region(file
, addr
, len
, flags
, vm_flags
, pgoff
);
1055 EXPORT_SYMBOL(do_mmap_pgoff
);
1058 * Some shared mappigns will want the pages marked read-only
1059 * to track write events. If so, we'll downgrade vm_page_prot
1060 * to the private version (using protection_map[] without the
1063 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1065 unsigned int vm_flags
= vma
->vm_flags
;
1067 /* If it was private or non-writable, the write bit is already clear */
1068 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1071 /* The backer wishes to know when pages are first written to? */
1072 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1075 /* The open routine did something to the protections already? */
1076 if (pgprot_val(vma
->vm_page_prot
) !=
1077 pgprot_val(vm_get_page_prot(vm_flags
)))
1080 /* Specialty mapping? */
1081 if (vm_flags
& (VM_PFNMAP
|VM_INSERTPAGE
))
1084 /* Can the mapping track the dirty pages? */
1085 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1086 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1090 * We account for memory if it's a private writeable mapping,
1091 * not hugepages and VM_NORESERVE wasn't set.
1093 static inline int accountable_mapping(struct file
*file
, unsigned int vm_flags
)
1096 * hugetlb has its own accounting separate from the core VM
1097 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1099 if (file
&& is_file_hugepages(file
))
1102 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1105 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1106 unsigned long len
, unsigned long flags
,
1107 unsigned int vm_flags
, unsigned long pgoff
)
1109 struct mm_struct
*mm
= current
->mm
;
1110 struct vm_area_struct
*vma
, *prev
;
1111 int correct_wcount
= 0;
1113 struct rb_node
**rb_link
, *rb_parent
;
1114 unsigned long charged
= 0;
1115 struct inode
*inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1117 /* Clear old maps */
1120 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1121 if (vma
&& vma
->vm_start
< addr
+ len
) {
1122 if (do_munmap(mm
, addr
, len
))
1127 /* Check against address space limit. */
1128 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1132 * Set 'VM_NORESERVE' if we should not account for the
1133 * memory use of this mapping.
1135 if ((flags
& MAP_NORESERVE
)) {
1136 /* We honor MAP_NORESERVE if allowed to overcommit */
1137 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1138 vm_flags
|= VM_NORESERVE
;
1140 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1141 if (file
&& is_file_hugepages(file
))
1142 vm_flags
|= VM_NORESERVE
;
1146 * Private writable mapping: check memory availability
1148 if (accountable_mapping(file
, vm_flags
)) {
1149 charged
= len
>> PAGE_SHIFT
;
1150 if (security_vm_enough_memory(charged
))
1152 vm_flags
|= VM_ACCOUNT
;
1156 * Can we just expand an old mapping?
1158 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
, NULL
, file
, pgoff
, NULL
);
1163 * Determine the object being mapped and call the appropriate
1164 * specific mapper. the address has already been validated, but
1165 * not unmapped, but the maps are removed from the list.
1167 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1174 vma
->vm_start
= addr
;
1175 vma
->vm_end
= addr
+ len
;
1176 vma
->vm_flags
= vm_flags
;
1177 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1178 vma
->vm_pgoff
= pgoff
;
1182 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1184 if (vm_flags
& VM_DENYWRITE
) {
1185 error
= deny_write_access(file
);
1190 vma
->vm_file
= file
;
1192 error
= file
->f_op
->mmap(file
, vma
);
1194 goto unmap_and_free_vma
;
1195 if (vm_flags
& VM_EXECUTABLE
)
1196 added_exe_file_vma(mm
);
1197 } else if (vm_flags
& VM_SHARED
) {
1198 error
= shmem_zero_setup(vma
);
1203 /* Can addr have changed??
1205 * Answer: Yes, several device drivers can do it in their
1206 * f_op->mmap method. -DaveM
1208 addr
= vma
->vm_start
;
1209 pgoff
= vma
->vm_pgoff
;
1210 vm_flags
= vma
->vm_flags
;
1212 if (vma_wants_writenotify(vma
))
1213 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
& ~VM_SHARED
);
1215 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1216 file
= vma
->vm_file
;
1218 /* Once vma denies write, undo our temporary denial count */
1220 atomic_inc(&inode
->i_writecount
);
1222 perf_event_mmap(vma
);
1224 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1225 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1226 if (vm_flags
& VM_LOCKED
) {
1228 * makes pages present; downgrades, drops, reacquires mmap_sem
1230 long nr_pages
= mlock_vma_pages_range(vma
, addr
, addr
+ len
);
1232 return nr_pages
; /* vma gone! */
1233 mm
->locked_vm
+= (len
>> PAGE_SHIFT
) - nr_pages
;
1234 } else if ((flags
& MAP_POPULATE
) && !(flags
& MAP_NONBLOCK
))
1235 make_pages_present(addr
, addr
+ len
);
1240 atomic_inc(&inode
->i_writecount
);
1241 vma
->vm_file
= NULL
;
1244 /* Undo any partial mapping done by a device driver. */
1245 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1248 kmem_cache_free(vm_area_cachep
, vma
);
1251 vm_unacct_memory(charged
);
1255 /* Get an address range which is currently unmapped.
1256 * For shmat() with addr=0.
1258 * Ugly calling convention alert:
1259 * Return value with the low bits set means error value,
1261 * if (ret & ~PAGE_MASK)
1264 * This function "knows" that -ENOMEM has the bits set.
1266 #ifndef HAVE_ARCH_UNMAPPED_AREA
1268 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1269 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1271 struct mm_struct
*mm
= current
->mm
;
1272 struct vm_area_struct
*vma
;
1273 unsigned long start_addr
;
1275 if (len
> TASK_SIZE
)
1278 if (flags
& MAP_FIXED
)
1282 addr
= PAGE_ALIGN(addr
);
1283 vma
= find_vma(mm
, addr
);
1284 if (TASK_SIZE
- len
>= addr
&&
1285 (!vma
|| addr
+ len
<= vma
->vm_start
))
1288 if (len
> mm
->cached_hole_size
) {
1289 start_addr
= addr
= mm
->free_area_cache
;
1291 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1292 mm
->cached_hole_size
= 0;
1296 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1297 /* At this point: (!vma || addr < vma->vm_end). */
1298 if (TASK_SIZE
- len
< addr
) {
1300 * Start a new search - just in case we missed
1303 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1304 addr
= TASK_UNMAPPED_BASE
;
1306 mm
->cached_hole_size
= 0;
1311 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1313 * Remember the place where we stopped the search:
1315 mm
->free_area_cache
= addr
+ len
;
1318 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1319 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1325 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1328 * Is this a new hole at the lowest possible address?
1330 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
) {
1331 mm
->free_area_cache
= addr
;
1332 mm
->cached_hole_size
= ~0UL;
1337 * This mmap-allocator allocates new areas top-down from below the
1338 * stack's low limit (the base):
1340 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1342 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1343 const unsigned long len
, const unsigned long pgoff
,
1344 const unsigned long flags
)
1346 struct vm_area_struct
*vma
;
1347 struct mm_struct
*mm
= current
->mm
;
1348 unsigned long addr
= addr0
;
1350 /* requested length too big for entire address space */
1351 if (len
> TASK_SIZE
)
1354 if (flags
& MAP_FIXED
)
1357 /* requesting a specific address */
1359 addr
= PAGE_ALIGN(addr
);
1360 vma
= find_vma(mm
, addr
);
1361 if (TASK_SIZE
- len
>= addr
&&
1362 (!vma
|| addr
+ len
<= vma
->vm_start
))
1366 /* check if free_area_cache is useful for us */
1367 if (len
<= mm
->cached_hole_size
) {
1368 mm
->cached_hole_size
= 0;
1369 mm
->free_area_cache
= mm
->mmap_base
;
1372 /* either no address requested or can't fit in requested address hole */
1373 addr
= mm
->free_area_cache
;
1375 /* make sure it can fit in the remaining address space */
1377 vma
= find_vma(mm
, addr
-len
);
1378 if (!vma
|| addr
<= vma
->vm_start
)
1379 /* remember the address as a hint for next time */
1380 return (mm
->free_area_cache
= addr
-len
);
1383 if (mm
->mmap_base
< len
)
1386 addr
= mm
->mmap_base
-len
;
1390 * Lookup failure means no vma is above this address,
1391 * else if new region fits below vma->vm_start,
1392 * return with success:
1394 vma
= find_vma(mm
, addr
);
1395 if (!vma
|| addr
+len
<= vma
->vm_start
)
1396 /* remember the address as a hint for next time */
1397 return (mm
->free_area_cache
= addr
);
1399 /* remember the largest hole we saw so far */
1400 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1401 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1403 /* try just below the current vma->vm_start */
1404 addr
= vma
->vm_start
-len
;
1405 } while (len
< vma
->vm_start
);
1409 * A failed mmap() very likely causes application failure,
1410 * so fall back to the bottom-up function here. This scenario
1411 * can happen with large stack limits and large mmap()
1414 mm
->cached_hole_size
= ~0UL;
1415 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1416 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1418 * Restore the topdown base:
1420 mm
->free_area_cache
= mm
->mmap_base
;
1421 mm
->cached_hole_size
= ~0UL;
1427 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1430 * Is this a new hole at the highest possible address?
1432 if (addr
> mm
->free_area_cache
)
1433 mm
->free_area_cache
= addr
;
1435 /* dont allow allocations above current base */
1436 if (mm
->free_area_cache
> mm
->mmap_base
)
1437 mm
->free_area_cache
= mm
->mmap_base
;
1441 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1442 unsigned long pgoff
, unsigned long flags
)
1444 unsigned long (*get_area
)(struct file
*, unsigned long,
1445 unsigned long, unsigned long, unsigned long);
1447 get_area
= current
->mm
->get_unmapped_area
;
1448 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1449 get_area
= file
->f_op
->get_unmapped_area
;
1450 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1451 if (IS_ERR_VALUE(addr
))
1454 if (addr
> TASK_SIZE
- len
)
1456 if (addr
& ~PAGE_MASK
)
1459 return arch_rebalance_pgtables(addr
, len
);
1462 EXPORT_SYMBOL(get_unmapped_area
);
1464 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1465 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
1467 struct vm_area_struct
*vma
= NULL
;
1470 /* Check the cache first. */
1471 /* (Cache hit rate is typically around 35%.) */
1472 vma
= mm
->mmap_cache
;
1473 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1474 struct rb_node
* rb_node
;
1476 rb_node
= mm
->mm_rb
.rb_node
;
1480 struct vm_area_struct
* vma_tmp
;
1482 vma_tmp
= rb_entry(rb_node
,
1483 struct vm_area_struct
, vm_rb
);
1485 if (vma_tmp
->vm_end
> addr
) {
1487 if (vma_tmp
->vm_start
<= addr
)
1489 rb_node
= rb_node
->rb_left
;
1491 rb_node
= rb_node
->rb_right
;
1494 mm
->mmap_cache
= vma
;
1500 EXPORT_SYMBOL(find_vma
);
1502 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1503 struct vm_area_struct
*
1504 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1505 struct vm_area_struct
**pprev
)
1507 struct vm_area_struct
*vma
= NULL
, *prev
= NULL
;
1508 struct rb_node
*rb_node
;
1512 /* Guard against addr being lower than the first VMA */
1515 /* Go through the RB tree quickly. */
1516 rb_node
= mm
->mm_rb
.rb_node
;
1519 struct vm_area_struct
*vma_tmp
;
1520 vma_tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1522 if (addr
< vma_tmp
->vm_end
) {
1523 rb_node
= rb_node
->rb_left
;
1526 if (!prev
->vm_next
|| (addr
< prev
->vm_next
->vm_end
))
1528 rb_node
= rb_node
->rb_right
;
1534 return prev
? prev
->vm_next
: vma
;
1538 * Verify that the stack growth is acceptable and
1539 * update accounting. This is shared with both the
1540 * grow-up and grow-down cases.
1542 static int acct_stack_growth(struct vm_area_struct
*vma
, unsigned long size
, unsigned long grow
)
1544 struct mm_struct
*mm
= vma
->vm_mm
;
1545 struct rlimit
*rlim
= current
->signal
->rlim
;
1546 unsigned long new_start
;
1548 /* address space limit tests */
1549 if (!may_expand_vm(mm
, grow
))
1552 /* Stack limit test */
1553 if (size
> rlim
[RLIMIT_STACK
].rlim_cur
)
1556 /* mlock limit tests */
1557 if (vma
->vm_flags
& VM_LOCKED
) {
1558 unsigned long locked
;
1559 unsigned long limit
;
1560 locked
= mm
->locked_vm
+ grow
;
1561 limit
= rlim
[RLIMIT_MEMLOCK
].rlim_cur
>> PAGE_SHIFT
;
1562 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1566 /* Check to ensure the stack will not grow into a hugetlb-only region */
1567 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
1569 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
1573 * Overcommit.. This must be the final test, as it will
1574 * update security statistics.
1576 if (security_vm_enough_memory_mm(mm
, grow
))
1579 /* Ok, everything looks good - let it rip */
1580 mm
->total_vm
+= grow
;
1581 if (vma
->vm_flags
& VM_LOCKED
)
1582 mm
->locked_vm
+= grow
;
1583 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1587 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1589 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1590 * vma is the last one with address > vma->vm_end. Have to extend vma.
1595 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1599 if (!(vma
->vm_flags
& VM_GROWSUP
))
1603 * We must make sure the anon_vma is allocated
1604 * so that the anon_vma locking is not a noop.
1606 if (unlikely(anon_vma_prepare(vma
)))
1611 * vma->vm_start/vm_end cannot change under us because the caller
1612 * is required to hold the mmap_sem in read mode. We need the
1613 * anon_vma lock to serialize against concurrent expand_stacks.
1614 * Also guard against wrapping around to address 0.
1616 if (address
< PAGE_ALIGN(address
+4))
1617 address
= PAGE_ALIGN(address
+4);
1619 anon_vma_unlock(vma
);
1624 /* Somebody else might have raced and expanded it already */
1625 if (address
> vma
->vm_end
) {
1626 unsigned long size
, grow
;
1628 size
= address
- vma
->vm_start
;
1629 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1631 error
= acct_stack_growth(vma
, size
, grow
);
1633 vma
->vm_end
= address
;
1635 anon_vma_unlock(vma
);
1638 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1641 * vma is the first one with address < vma->vm_start. Have to extend vma.
1643 static int expand_downwards(struct vm_area_struct
*vma
,
1644 unsigned long address
)
1649 * We must make sure the anon_vma is allocated
1650 * so that the anon_vma locking is not a noop.
1652 if (unlikely(anon_vma_prepare(vma
)))
1655 address
&= PAGE_MASK
;
1656 error
= security_file_mmap(NULL
, 0, 0, 0, address
, 1);
1663 * vma->vm_start/vm_end cannot change under us because the caller
1664 * is required to hold the mmap_sem in read mode. We need the
1665 * anon_vma lock to serialize against concurrent expand_stacks.
1668 /* Somebody else might have raced and expanded it already */
1669 if (address
< vma
->vm_start
) {
1670 unsigned long size
, grow
;
1672 size
= vma
->vm_end
- address
;
1673 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1675 error
= acct_stack_growth(vma
, size
, grow
);
1677 vma
->vm_start
= address
;
1678 vma
->vm_pgoff
-= grow
;
1681 anon_vma_unlock(vma
);
1685 int expand_stack_downwards(struct vm_area_struct
*vma
, unsigned long address
)
1687 return expand_downwards(vma
, address
);
1690 #ifdef CONFIG_STACK_GROWSUP
1691 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1693 return expand_upwards(vma
, address
);
1696 struct vm_area_struct
*
1697 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1699 struct vm_area_struct
*vma
, *prev
;
1702 vma
= find_vma_prev(mm
, addr
, &prev
);
1703 if (vma
&& (vma
->vm_start
<= addr
))
1705 if (!prev
|| expand_stack(prev
, addr
))
1707 if (prev
->vm_flags
& VM_LOCKED
) {
1708 if (mlock_vma_pages_range(prev
, addr
, prev
->vm_end
) < 0)
1709 return NULL
; /* vma gone! */
1714 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1716 return expand_downwards(vma
, address
);
1719 struct vm_area_struct
*
1720 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1722 struct vm_area_struct
* vma
;
1723 unsigned long start
;
1726 vma
= find_vma(mm
,addr
);
1729 if (vma
->vm_start
<= addr
)
1731 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1733 start
= vma
->vm_start
;
1734 if (expand_stack(vma
, addr
))
1736 if (vma
->vm_flags
& VM_LOCKED
) {
1737 if (mlock_vma_pages_range(vma
, addr
, start
) < 0)
1738 return NULL
; /* vma gone! */
1745 * Ok - we have the memory areas we should free on the vma list,
1746 * so release them, and do the vma updates.
1748 * Called with the mm semaphore held.
1750 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1752 /* Update high watermark before we lower total_vm */
1753 update_hiwater_vm(mm
);
1755 long nrpages
= vma_pages(vma
);
1757 mm
->total_vm
-= nrpages
;
1758 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1759 vma
= remove_vma(vma
);
1765 * Get rid of page table information in the indicated region.
1767 * Called with the mm semaphore held.
1769 static void unmap_region(struct mm_struct
*mm
,
1770 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1771 unsigned long start
, unsigned long end
)
1773 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1774 struct mmu_gather
*tlb
;
1775 unsigned long nr_accounted
= 0;
1778 tlb
= tlb_gather_mmu(mm
, 0);
1779 update_hiwater_rss(mm
);
1780 unmap_vmas(&tlb
, vma
, start
, end
, &nr_accounted
, NULL
);
1781 vm_unacct_memory(nr_accounted
);
1782 free_pgtables(tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1783 next
? next
->vm_start
: 0);
1784 tlb_finish_mmu(tlb
, start
, end
);
1788 * Create a list of vma's touched by the unmap, removing them from the mm's
1789 * vma list as we go..
1792 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1793 struct vm_area_struct
*prev
, unsigned long end
)
1795 struct vm_area_struct
**insertion_point
;
1796 struct vm_area_struct
*tail_vma
= NULL
;
1799 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1801 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1805 } while (vma
&& vma
->vm_start
< end
);
1806 *insertion_point
= vma
;
1807 tail_vma
->vm_next
= NULL
;
1808 if (mm
->unmap_area
== arch_unmap_area
)
1809 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1811 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1812 mm
->unmap_area(mm
, addr
);
1813 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1817 * Split a vma into two pieces at address 'addr', a new vma is allocated
1818 * either for the first part or the tail.
1820 int split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1821 unsigned long addr
, int new_below
)
1823 struct mempolicy
*pol
;
1824 struct vm_area_struct
*new;
1826 if (is_vm_hugetlb_page(vma
) && (addr
&
1827 ~(huge_page_mask(hstate_vma(vma
)))))
1830 if (mm
->map_count
>= sysctl_max_map_count
)
1833 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1837 /* most fields are the same, copy all, and then fixup */
1843 new->vm_start
= addr
;
1844 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
1847 pol
= mpol_dup(vma_policy(vma
));
1849 kmem_cache_free(vm_area_cachep
, new);
1850 return PTR_ERR(pol
);
1852 vma_set_policy(new, pol
);
1855 get_file(new->vm_file
);
1856 if (vma
->vm_flags
& VM_EXECUTABLE
)
1857 added_exe_file_vma(mm
);
1860 if (new->vm_ops
&& new->vm_ops
->open
)
1861 new->vm_ops
->open(new);
1864 vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
1865 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
1867 vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
1872 /* Munmap is split into 2 main parts -- this part which finds
1873 * what needs doing, and the areas themselves, which do the
1874 * work. This now handles partial unmappings.
1875 * Jeremy Fitzhardinge <jeremy@goop.org>
1877 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1880 struct vm_area_struct
*vma
, *prev
, *last
;
1882 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
1885 if ((len
= PAGE_ALIGN(len
)) == 0)
1888 /* Find the first overlapping VMA */
1889 vma
= find_vma_prev(mm
, start
, &prev
);
1892 /* we have start < vma->vm_end */
1894 /* if it doesn't overlap, we have nothing.. */
1896 if (vma
->vm_start
>= end
)
1900 * If we need to split any vma, do it now to save pain later.
1902 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1903 * unmapped vm_area_struct will remain in use: so lower split_vma
1904 * places tmp vma above, and higher split_vma places tmp vma below.
1906 if (start
> vma
->vm_start
) {
1907 int error
= split_vma(mm
, vma
, start
, 0);
1913 /* Does it split the last one? */
1914 last
= find_vma(mm
, end
);
1915 if (last
&& end
> last
->vm_start
) {
1916 int error
= split_vma(mm
, last
, end
, 1);
1920 vma
= prev
? prev
->vm_next
: mm
->mmap
;
1923 * unlock any mlock()ed ranges before detaching vmas
1925 if (mm
->locked_vm
) {
1926 struct vm_area_struct
*tmp
= vma
;
1927 while (tmp
&& tmp
->vm_start
< end
) {
1928 if (tmp
->vm_flags
& VM_LOCKED
) {
1929 mm
->locked_vm
-= vma_pages(tmp
);
1930 munlock_vma_pages_all(tmp
);
1937 * Remove the vma's, and unmap the actual pages
1939 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
1940 unmap_region(mm
, vma
, prev
, start
, end
);
1942 /* Fix up all other VM information */
1943 remove_vma_list(mm
, vma
);
1948 EXPORT_SYMBOL(do_munmap
);
1950 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
1953 struct mm_struct
*mm
= current
->mm
;
1955 profile_munmap(addr
);
1957 down_write(&mm
->mmap_sem
);
1958 ret
= do_munmap(mm
, addr
, len
);
1959 up_write(&mm
->mmap_sem
);
1963 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
1965 #ifdef CONFIG_DEBUG_VM
1966 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
1968 up_read(&mm
->mmap_sem
);
1974 * this is really a simplified "do_mmap". it only handles
1975 * anonymous maps. eventually we may be able to do some
1976 * brk-specific accounting here.
1978 unsigned long do_brk(unsigned long addr
, unsigned long len
)
1980 struct mm_struct
* mm
= current
->mm
;
1981 struct vm_area_struct
* vma
, * prev
;
1982 unsigned long flags
;
1983 struct rb_node
** rb_link
, * rb_parent
;
1984 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
1987 len
= PAGE_ALIGN(len
);
1991 if ((addr
+ len
) > TASK_SIZE
|| (addr
+ len
) < addr
)
1994 if (is_hugepage_only_range(mm
, addr
, len
))
1997 error
= security_file_mmap(NULL
, 0, 0, 0, addr
, 1);
2001 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2003 error
= arch_mmap_check(addr
, len
, flags
);
2010 if (mm
->def_flags
& VM_LOCKED
) {
2011 unsigned long locked
, lock_limit
;
2012 locked
= len
>> PAGE_SHIFT
;
2013 locked
+= mm
->locked_vm
;
2014 lock_limit
= current
->signal
->rlim
[RLIMIT_MEMLOCK
].rlim_cur
;
2015 lock_limit
>>= PAGE_SHIFT
;
2016 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
2021 * mm->mmap_sem is required to protect against another thread
2022 * changing the mappings in case we sleep.
2024 verify_mm_writelocked(mm
);
2027 * Clear old maps. this also does some error checking for us
2030 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2031 if (vma
&& vma
->vm_start
< addr
+ len
) {
2032 if (do_munmap(mm
, addr
, len
))
2037 /* Check against address space limits *after* clearing old maps... */
2038 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2041 if (mm
->map_count
> sysctl_max_map_count
)
2044 if (security_vm_enough_memory(len
>> PAGE_SHIFT
))
2047 /* Can we just expand an old private anonymous mapping? */
2048 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2049 NULL
, NULL
, pgoff
, NULL
);
2054 * create a vma struct for an anonymous mapping
2056 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2058 vm_unacct_memory(len
>> PAGE_SHIFT
);
2063 vma
->vm_start
= addr
;
2064 vma
->vm_end
= addr
+ len
;
2065 vma
->vm_pgoff
= pgoff
;
2066 vma
->vm_flags
= flags
;
2067 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2068 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2070 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2071 if (flags
& VM_LOCKED
) {
2072 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
2073 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2078 EXPORT_SYMBOL(do_brk
);
2080 /* Release all mmaps. */
2081 void exit_mmap(struct mm_struct
*mm
)
2083 struct mmu_gather
*tlb
;
2084 struct vm_area_struct
*vma
;
2085 unsigned long nr_accounted
= 0;
2088 /* mm's last user has gone, and its about to be pulled down */
2089 mmu_notifier_release(mm
);
2091 if (mm
->locked_vm
) {
2094 if (vma
->vm_flags
& VM_LOCKED
)
2095 munlock_vma_pages_all(vma
);
2103 if (!vma
) /* Can happen if dup_mmap() received an OOM */
2108 tlb
= tlb_gather_mmu(mm
, 1);
2109 /* update_hiwater_rss(mm) here? but nobody should be looking */
2110 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2111 end
= unmap_vmas(&tlb
, vma
, 0, -1, &nr_accounted
, NULL
);
2112 vm_unacct_memory(nr_accounted
);
2114 free_pgtables(tlb
, vma
, FIRST_USER_ADDRESS
, 0);
2115 tlb_finish_mmu(tlb
, 0, end
);
2118 * Walk the list again, actually closing and freeing it,
2119 * with preemption enabled, without holding any MM locks.
2122 vma
= remove_vma(vma
);
2124 BUG_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2127 /* Insert vm structure into process list sorted by address
2128 * and into the inode's i_mmap tree. If vm_file is non-NULL
2129 * then i_mmap_lock is taken here.
2131 int insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
2133 struct vm_area_struct
* __vma
, * prev
;
2134 struct rb_node
** rb_link
, * rb_parent
;
2137 * The vm_pgoff of a purely anonymous vma should be irrelevant
2138 * until its first write fault, when page's anon_vma and index
2139 * are set. But now set the vm_pgoff it will almost certainly
2140 * end up with (unless mremap moves it elsewhere before that
2141 * first wfault), so /proc/pid/maps tells a consistent story.
2143 * By setting it to reflect the virtual start address of the
2144 * vma, merges and splits can happen in a seamless way, just
2145 * using the existing file pgoff checks and manipulations.
2146 * Similarly in do_mmap_pgoff and in do_brk.
2148 if (!vma
->vm_file
) {
2149 BUG_ON(vma
->anon_vma
);
2150 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2152 __vma
= find_vma_prepare(mm
,vma
->vm_start
,&prev
,&rb_link
,&rb_parent
);
2153 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
2155 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2156 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2158 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2163 * Copy the vma structure to a new location in the same mm,
2164 * prior to moving page table entries, to effect an mremap move.
2166 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2167 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
2169 struct vm_area_struct
*vma
= *vmap
;
2170 unsigned long vma_start
= vma
->vm_start
;
2171 struct mm_struct
*mm
= vma
->vm_mm
;
2172 struct vm_area_struct
*new_vma
, *prev
;
2173 struct rb_node
**rb_link
, *rb_parent
;
2174 struct mempolicy
*pol
;
2177 * If anonymous vma has not yet been faulted, update new pgoff
2178 * to match new location, to increase its chance of merging.
2180 if (!vma
->vm_file
&& !vma
->anon_vma
)
2181 pgoff
= addr
>> PAGE_SHIFT
;
2183 find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2184 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2185 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2188 * Source vma may have been merged into new_vma
2190 if (vma_start
>= new_vma
->vm_start
&&
2191 vma_start
< new_vma
->vm_end
)
2194 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2197 pol
= mpol_dup(vma_policy(vma
));
2199 kmem_cache_free(vm_area_cachep
, new_vma
);
2202 vma_set_policy(new_vma
, pol
);
2203 new_vma
->vm_start
= addr
;
2204 new_vma
->vm_end
= addr
+ len
;
2205 new_vma
->vm_pgoff
= pgoff
;
2206 if (new_vma
->vm_file
) {
2207 get_file(new_vma
->vm_file
);
2208 if (vma
->vm_flags
& VM_EXECUTABLE
)
2209 added_exe_file_vma(mm
);
2211 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2212 new_vma
->vm_ops
->open(new_vma
);
2213 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2220 * Return true if the calling process may expand its vm space by the passed
2223 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2225 unsigned long cur
= mm
->total_vm
; /* pages */
2228 lim
= current
->signal
->rlim
[RLIMIT_AS
].rlim_cur
>> PAGE_SHIFT
;
2230 if (cur
+ npages
> lim
)
2236 static int special_mapping_fault(struct vm_area_struct
*vma
,
2237 struct vm_fault
*vmf
)
2240 struct page
**pages
;
2243 * special mappings have no vm_file, and in that case, the mm
2244 * uses vm_pgoff internally. So we have to subtract it from here.
2245 * We are allowed to do this because we are the mm; do not copy
2246 * this code into drivers!
2248 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2250 for (pages
= vma
->vm_private_data
; pgoff
&& *pages
; ++pages
)
2254 struct page
*page
= *pages
;
2260 return VM_FAULT_SIGBUS
;
2264 * Having a close hook prevents vma merging regardless of flags.
2266 static void special_mapping_close(struct vm_area_struct
*vma
)
2270 static struct vm_operations_struct special_mapping_vmops
= {
2271 .close
= special_mapping_close
,
2272 .fault
= special_mapping_fault
,
2276 * Called with mm->mmap_sem held for writing.
2277 * Insert a new vma covering the given region, with the given flags.
2278 * Its pages are supplied by the given array of struct page *.
2279 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2280 * The region past the last page supplied will always produce SIGBUS.
2281 * The array pointer and the pages it points to are assumed to stay alive
2282 * for as long as this mapping might exist.
2284 int install_special_mapping(struct mm_struct
*mm
,
2285 unsigned long addr
, unsigned long len
,
2286 unsigned long vm_flags
, struct page
**pages
)
2288 struct vm_area_struct
*vma
;
2290 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2291 if (unlikely(vma
== NULL
))
2295 vma
->vm_start
= addr
;
2296 vma
->vm_end
= addr
+ len
;
2298 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
;
2299 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
2301 vma
->vm_ops
= &special_mapping_vmops
;
2302 vma
->vm_private_data
= pages
;
2304 if (unlikely(insert_vm_struct(mm
, vma
))) {
2305 kmem_cache_free(vm_area_cachep
, vma
);
2309 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2311 perf_event_mmap(vma
);
2316 static DEFINE_MUTEX(mm_all_locks_mutex
);
2318 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
2320 if (!test_bit(0, (unsigned long *) &anon_vma
->head
.next
)) {
2322 * The LSB of head.next can't change from under us
2323 * because we hold the mm_all_locks_mutex.
2325 spin_lock_nest_lock(&anon_vma
->lock
, &mm
->mmap_sem
);
2327 * We can safely modify head.next after taking the
2328 * anon_vma->lock. If some other vma in this mm shares
2329 * the same anon_vma we won't take it again.
2331 * No need of atomic instructions here, head.next
2332 * can't change from under us thanks to the
2335 if (__test_and_set_bit(0, (unsigned long *)
2336 &anon_vma
->head
.next
))
2341 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
2343 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2345 * AS_MM_ALL_LOCKS can't change from under us because
2346 * we hold the mm_all_locks_mutex.
2348 * Operations on ->flags have to be atomic because
2349 * even if AS_MM_ALL_LOCKS is stable thanks to the
2350 * mm_all_locks_mutex, there may be other cpus
2351 * changing other bitflags in parallel to us.
2353 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
2355 spin_lock_nest_lock(&mapping
->i_mmap_lock
, &mm
->mmap_sem
);
2360 * This operation locks against the VM for all pte/vma/mm related
2361 * operations that could ever happen on a certain mm. This includes
2362 * vmtruncate, try_to_unmap, and all page faults.
2364 * The caller must take the mmap_sem in write mode before calling
2365 * mm_take_all_locks(). The caller isn't allowed to release the
2366 * mmap_sem until mm_drop_all_locks() returns.
2368 * mmap_sem in write mode is required in order to block all operations
2369 * that could modify pagetables and free pages without need of
2370 * altering the vma layout (for example populate_range() with
2371 * nonlinear vmas). It's also needed in write mode to avoid new
2372 * anon_vmas to be associated with existing vmas.
2374 * A single task can't take more than one mm_take_all_locks() in a row
2375 * or it would deadlock.
2377 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2378 * mapping->flags avoid to take the same lock twice, if more than one
2379 * vma in this mm is backed by the same anon_vma or address_space.
2381 * We can take all the locks in random order because the VM code
2382 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2383 * takes more than one of them in a row. Secondly we're protected
2384 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2386 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2387 * that may have to take thousand of locks.
2389 * mm_take_all_locks() can fail if it's interrupted by signals.
2391 int mm_take_all_locks(struct mm_struct
*mm
)
2393 struct vm_area_struct
*vma
;
2396 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2398 mutex_lock(&mm_all_locks_mutex
);
2400 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2401 if (signal_pending(current
))
2403 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2404 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
2407 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2408 if (signal_pending(current
))
2411 vm_lock_anon_vma(mm
, vma
->anon_vma
);
2418 mm_drop_all_locks(mm
);
2423 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
2425 if (test_bit(0, (unsigned long *) &anon_vma
->head
.next
)) {
2427 * The LSB of head.next can't change to 0 from under
2428 * us because we hold the mm_all_locks_mutex.
2430 * We must however clear the bitflag before unlocking
2431 * the vma so the users using the anon_vma->head will
2432 * never see our bitflag.
2434 * No need of atomic instructions here, head.next
2435 * can't change from under us until we release the
2438 if (!__test_and_clear_bit(0, (unsigned long *)
2439 &anon_vma
->head
.next
))
2441 spin_unlock(&anon_vma
->lock
);
2445 static void vm_unlock_mapping(struct address_space
*mapping
)
2447 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2449 * AS_MM_ALL_LOCKS can't change to 0 from under us
2450 * because we hold the mm_all_locks_mutex.
2452 spin_unlock(&mapping
->i_mmap_lock
);
2453 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
2460 * The mmap_sem cannot be released by the caller until
2461 * mm_drop_all_locks() returns.
2463 void mm_drop_all_locks(struct mm_struct
*mm
)
2465 struct vm_area_struct
*vma
;
2467 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2468 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
2470 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2472 vm_unlock_anon_vma(vma
->anon_vma
);
2473 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2474 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
2477 mutex_unlock(&mm_all_locks_mutex
);
2481 * initialise the VMA slab
2483 void __init
mmap_init(void)
2487 ret
= percpu_counter_init(&vm_committed_as
, 0);