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
= rlimit(RLIMIT_DATA
);
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
);
442 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
443 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
444 struct rb_node
*rb_parent
)
446 struct address_space
*mapping
= NULL
;
449 mapping
= vma
->vm_file
->f_mapping
;
452 spin_lock(&mapping
->i_mmap_lock
);
453 vma
->vm_truncate_count
= mapping
->truncate_count
;
457 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
458 __vma_link_file(vma
);
460 anon_vma_unlock(vma
);
462 spin_unlock(&mapping
->i_mmap_lock
);
469 * Helper for vma_adjust in the split_vma insert case:
470 * insert vm structure into list and rbtree and anon_vma,
471 * but it has already been inserted into prio_tree earlier.
473 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
475 struct vm_area_struct
*__vma
, *prev
;
476 struct rb_node
**rb_link
, *rb_parent
;
478 __vma
= find_vma_prepare(mm
, vma
->vm_start
,&prev
, &rb_link
, &rb_parent
);
479 BUG_ON(__vma
&& __vma
->vm_start
< vma
->vm_end
);
480 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
485 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
486 struct vm_area_struct
*prev
)
488 prev
->vm_next
= vma
->vm_next
;
489 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
490 if (mm
->mmap_cache
== vma
)
491 mm
->mmap_cache
= prev
;
495 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
496 * is already present in an i_mmap tree without adjusting the tree.
497 * The following helper function should be used when such adjustments
498 * are necessary. The "insert" vma (if any) is to be inserted
499 * before we drop the necessary locks.
501 int vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
502 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
504 struct mm_struct
*mm
= vma
->vm_mm
;
505 struct vm_area_struct
*next
= vma
->vm_next
;
506 struct vm_area_struct
*importer
= NULL
;
507 struct address_space
*mapping
= NULL
;
508 struct prio_tree_root
*root
= NULL
;
509 struct file
*file
= vma
->vm_file
;
510 struct anon_vma
*anon_vma
= NULL
;
511 long adjust_next
= 0;
514 if (next
&& !insert
) {
515 if (end
>= next
->vm_end
) {
517 * vma expands, overlapping all the next, and
518 * perhaps the one after too (mprotect case 6).
520 again
: remove_next
= 1 + (end
> next
->vm_end
);
522 anon_vma
= next
->anon_vma
;
524 } else if (end
> next
->vm_start
) {
526 * vma expands, overlapping part of the next:
527 * mprotect case 5 shifting the boundary up.
529 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
530 anon_vma
= next
->anon_vma
;
532 } else if (end
< vma
->vm_end
) {
534 * vma shrinks, and !insert tells it's not
535 * split_vma inserting another: so it must be
536 * mprotect case 4 shifting the boundary down.
538 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
539 anon_vma
= next
->anon_vma
;
545 * When changing only vma->vm_end, we don't really need anon_vma lock.
547 if (vma
->anon_vma
&& (insert
|| importer
|| start
!= vma
->vm_start
))
548 anon_vma
= vma
->anon_vma
;
551 * Easily overlooked: when mprotect shifts the boundary,
552 * make sure the expanding vma has anon_vma set if the
553 * shrinking vma had, to cover any anon pages imported.
555 if (importer
&& !importer
->anon_vma
) {
556 /* Block reverse map lookups until things are set up. */
557 if (anon_vma_clone(importer
, vma
)) {
560 importer
->anon_vma
= anon_vma
;
565 mapping
= file
->f_mapping
;
566 if (!(vma
->vm_flags
& VM_NONLINEAR
))
567 root
= &mapping
->i_mmap
;
568 spin_lock(&mapping
->i_mmap_lock
);
570 vma
->vm_truncate_count
!= next
->vm_truncate_count
) {
572 * unmap_mapping_range might be in progress:
573 * ensure that the expanding vma is rescanned.
575 importer
->vm_truncate_count
= 0;
578 insert
->vm_truncate_count
= vma
->vm_truncate_count
;
580 * Put into prio_tree now, so instantiated pages
581 * are visible to arm/parisc __flush_dcache_page
582 * throughout; but we cannot insert into address
583 * space until vma start or end is updated.
585 __vma_link_file(insert
);
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
);
621 * split_vma has split insert from vma, and needs
622 * us to insert it before dropping the locks
623 * (it may either follow vma or precede it).
625 __insert_vm_struct(mm
, insert
);
629 spin_unlock(&mapping
->i_mmap_lock
);
634 if (next
->vm_flags
& VM_EXECUTABLE
)
635 removed_exe_file_vma(mm
);
638 anon_vma_merge(vma
, next
);
640 mpol_put(vma_policy(next
));
641 kmem_cache_free(vm_area_cachep
, next
);
643 * In mprotect's case 6 (see comments on vma_merge),
644 * we must remove another next too. It would clutter
645 * up the code too much to do both in one go.
647 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
;
765 * We later require that vma->vm_flags == vm_flags,
766 * so this tests vma->vm_flags & VM_SPECIAL, too.
768 if (vm_flags
& VM_SPECIAL
)
772 next
= prev
->vm_next
;
776 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
777 next
= next
->vm_next
;
780 * Can it merge with the predecessor?
782 if (prev
&& prev
->vm_end
== addr
&&
783 mpol_equal(vma_policy(prev
), policy
) &&
784 can_vma_merge_after(prev
, vm_flags
,
785 anon_vma
, file
, pgoff
)) {
787 * OK, it can. Can we now merge in the successor as well?
789 if (next
&& end
== next
->vm_start
&&
790 mpol_equal(policy
, vma_policy(next
)) &&
791 can_vma_merge_before(next
, vm_flags
,
792 anon_vma
, file
, pgoff
+pglen
) &&
793 is_mergeable_anon_vma(prev
->anon_vma
,
796 err
= vma_adjust(prev
, prev
->vm_start
,
797 next
->vm_end
, prev
->vm_pgoff
, NULL
);
798 } else /* cases 2, 5, 7 */
799 err
= vma_adjust(prev
, prev
->vm_start
,
800 end
, prev
->vm_pgoff
, NULL
);
807 * Can this new request be merged in front of next?
809 if (next
&& end
== next
->vm_start
&&
810 mpol_equal(policy
, vma_policy(next
)) &&
811 can_vma_merge_before(next
, vm_flags
,
812 anon_vma
, file
, pgoff
+pglen
)) {
813 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
814 err
= vma_adjust(prev
, prev
->vm_start
,
815 addr
, prev
->vm_pgoff
, NULL
);
816 else /* cases 3, 8 */
817 err
= vma_adjust(area
, addr
, next
->vm_end
,
818 next
->vm_pgoff
- pglen
, NULL
);
828 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
829 * neighbouring vmas for a suitable anon_vma, before it goes off
830 * to allocate a new anon_vma. It checks because a repetitive
831 * sequence of mprotects and faults may otherwise lead to distinct
832 * anon_vmas being allocated, preventing vma merge in subsequent
835 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
837 struct vm_area_struct
*near
;
838 unsigned long vm_flags
;
845 * Since only mprotect tries to remerge vmas, match flags
846 * which might be mprotected into each other later on.
847 * Neither mlock nor madvise tries to remerge at present,
848 * so leave their flags as obstructing a merge.
850 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
851 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
853 if (near
->anon_vma
&& vma
->vm_end
== near
->vm_start
&&
854 mpol_equal(vma_policy(vma
), vma_policy(near
)) &&
855 can_vma_merge_before(near
, vm_flags
,
856 NULL
, vma
->vm_file
, vma
->vm_pgoff
+
857 ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
)))
858 return near
->anon_vma
;
861 * It is potentially slow to have to call find_vma_prev here.
862 * But it's only on the first write fault on the vma, not
863 * every time, and we could devise a way to avoid it later
864 * (e.g. stash info in next's anon_vma_node when assigning
865 * an anon_vma, or when trying vma_merge). Another time.
867 BUG_ON(find_vma_prev(vma
->vm_mm
, vma
->vm_start
, &near
) != vma
);
871 vm_flags
= vma
->vm_flags
& ~(VM_READ
|VM_WRITE
|VM_EXEC
);
872 vm_flags
|= near
->vm_flags
& (VM_READ
|VM_WRITE
|VM_EXEC
);
874 if (near
->anon_vma
&& near
->vm_end
== vma
->vm_start
&&
875 mpol_equal(vma_policy(near
), vma_policy(vma
)) &&
876 can_vma_merge_after(near
, vm_flags
,
877 NULL
, vma
->vm_file
, vma
->vm_pgoff
))
878 return near
->anon_vma
;
881 * There's no absolute need to look only at touching neighbours:
882 * we could search further afield for "compatible" anon_vmas.
883 * But it would probably just be a waste of time searching,
884 * or lead to too many vmas hanging off the same anon_vma.
885 * We're trying to allow mprotect remerging later on,
886 * not trying to minimize memory used for anon_vmas.
891 #ifdef CONFIG_PROC_FS
892 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
893 struct file
*file
, long pages
)
895 const unsigned long stack_flags
896 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
899 mm
->shared_vm
+= pages
;
900 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
901 mm
->exec_vm
+= pages
;
902 } else if (flags
& stack_flags
)
903 mm
->stack_vm
+= pages
;
904 if (flags
& (VM_RESERVED
|VM_IO
))
905 mm
->reserved_vm
+= pages
;
907 #endif /* CONFIG_PROC_FS */
910 * The caller must hold down_write(¤t->mm->mmap_sem).
913 unsigned long do_mmap_pgoff(struct file
*file
, unsigned long addr
,
914 unsigned long len
, unsigned long prot
,
915 unsigned long flags
, unsigned long pgoff
)
917 struct mm_struct
* mm
= current
->mm
;
919 unsigned int vm_flags
;
921 unsigned long reqprot
= prot
;
924 * Does the application expect PROT_READ to imply PROT_EXEC?
926 * (the exception is when the underlying filesystem is noexec
927 * mounted, in which case we dont add PROT_EXEC.)
929 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
930 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
936 if (!(flags
& MAP_FIXED
))
937 addr
= round_hint_to_min(addr
);
939 /* Careful about overflows.. */
940 len
= PAGE_ALIGN(len
);
944 /* offset overflow? */
945 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
948 /* Too many mappings? */
949 if (mm
->map_count
> sysctl_max_map_count
)
952 /* Obtain the address to map to. we verify (or select) it and ensure
953 * that it represents a valid section of the address space.
955 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
956 if (addr
& ~PAGE_MASK
)
959 /* Do simple checking here so the lower-level routines won't have
960 * to. we assume access permissions have been handled by the open
961 * of the memory object, so we don't do any here.
963 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
964 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
966 if (flags
& MAP_LOCKED
)
970 /* mlock MCL_FUTURE? */
971 if (vm_flags
& VM_LOCKED
) {
972 unsigned long locked
, lock_limit
;
973 locked
= len
>> PAGE_SHIFT
;
974 locked
+= mm
->locked_vm
;
975 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
976 lock_limit
>>= PAGE_SHIFT
;
977 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
981 inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
984 switch (flags
& MAP_TYPE
) {
986 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
990 * Make sure we don't allow writing to an append-only
993 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
997 * Make sure there are no mandatory locks on the file.
999 if (locks_verify_locked(inode
))
1002 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1003 if (!(file
->f_mode
& FMODE_WRITE
))
1004 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1008 if (!(file
->f_mode
& FMODE_READ
))
1010 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1011 if (vm_flags
& VM_EXEC
)
1013 vm_flags
&= ~VM_MAYEXEC
;
1016 if (!file
->f_op
|| !file
->f_op
->mmap
)
1024 switch (flags
& MAP_TYPE
) {
1030 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1034 * Set pgoff according to addr for anon_vma.
1036 pgoff
= addr
>> PAGE_SHIFT
;
1043 error
= security_file_mmap(file
, reqprot
, prot
, flags
, addr
, 0);
1047 return mmap_region(file
, addr
, len
, flags
, vm_flags
, pgoff
);
1049 EXPORT_SYMBOL(do_mmap_pgoff
);
1051 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1052 unsigned long, prot
, unsigned long, flags
,
1053 unsigned long, fd
, unsigned long, pgoff
)
1055 struct file
*file
= NULL
;
1056 unsigned long retval
= -EBADF
;
1058 if (!(flags
& MAP_ANONYMOUS
)) {
1059 if (unlikely(flags
& MAP_HUGETLB
))
1064 } else if (flags
& MAP_HUGETLB
) {
1065 struct user_struct
*user
= NULL
;
1067 * VM_NORESERVE is used because the reservations will be
1068 * taken when vm_ops->mmap() is called
1069 * A dummy user value is used because we are not locking
1070 * memory so no accounting is necessary
1072 len
= ALIGN(len
, huge_page_size(&default_hstate
));
1073 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, len
, VM_NORESERVE
,
1074 &user
, HUGETLB_ANONHUGE_INODE
);
1076 return PTR_ERR(file
);
1079 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1081 down_write(¤t
->mm
->mmap_sem
);
1082 retval
= do_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1083 up_write(¤t
->mm
->mmap_sem
);
1092 * Some shared mappigns will want the pages marked read-only
1093 * to track write events. If so, we'll downgrade vm_page_prot
1094 * to the private version (using protection_map[] without the
1097 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1099 unsigned int vm_flags
= vma
->vm_flags
;
1101 /* If it was private or non-writable, the write bit is already clear */
1102 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1105 /* The backer wishes to know when pages are first written to? */
1106 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1109 /* The open routine did something to the protections already? */
1110 if (pgprot_val(vma
->vm_page_prot
) !=
1111 pgprot_val(vm_get_page_prot(vm_flags
)))
1114 /* Specialty mapping? */
1115 if (vm_flags
& (VM_PFNMAP
|VM_INSERTPAGE
))
1118 /* Can the mapping track the dirty pages? */
1119 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1120 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1124 * We account for memory if it's a private writeable mapping,
1125 * not hugepages and VM_NORESERVE wasn't set.
1127 static inline int accountable_mapping(struct file
*file
, unsigned int vm_flags
)
1130 * hugetlb has its own accounting separate from the core VM
1131 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1133 if (file
&& is_file_hugepages(file
))
1136 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1139 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1140 unsigned long len
, unsigned long flags
,
1141 unsigned int vm_flags
, unsigned long pgoff
)
1143 struct mm_struct
*mm
= current
->mm
;
1144 struct vm_area_struct
*vma
, *prev
;
1145 int correct_wcount
= 0;
1147 struct rb_node
**rb_link
, *rb_parent
;
1148 unsigned long charged
= 0;
1149 struct inode
*inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1151 /* Clear old maps */
1154 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1155 if (vma
&& vma
->vm_start
< addr
+ len
) {
1156 if (do_munmap(mm
, addr
, len
))
1161 /* Check against address space limit. */
1162 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1166 * Set 'VM_NORESERVE' if we should not account for the
1167 * memory use of this mapping.
1169 if ((flags
& MAP_NORESERVE
)) {
1170 /* We honor MAP_NORESERVE if allowed to overcommit */
1171 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1172 vm_flags
|= VM_NORESERVE
;
1174 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1175 if (file
&& is_file_hugepages(file
))
1176 vm_flags
|= VM_NORESERVE
;
1180 * Private writable mapping: check memory availability
1182 if (accountable_mapping(file
, vm_flags
)) {
1183 charged
= len
>> PAGE_SHIFT
;
1184 if (security_vm_enough_memory(charged
))
1186 vm_flags
|= VM_ACCOUNT
;
1190 * Can we just expand an old mapping?
1192 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
, NULL
, file
, pgoff
, NULL
);
1197 * Determine the object being mapped and call the appropriate
1198 * specific mapper. the address has already been validated, but
1199 * not unmapped, but the maps are removed from the list.
1201 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1208 vma
->vm_start
= addr
;
1209 vma
->vm_end
= addr
+ len
;
1210 vma
->vm_flags
= vm_flags
;
1211 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1212 vma
->vm_pgoff
= pgoff
;
1213 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1217 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1219 if (vm_flags
& VM_DENYWRITE
) {
1220 error
= deny_write_access(file
);
1225 vma
->vm_file
= file
;
1227 error
= file
->f_op
->mmap(file
, vma
);
1229 goto unmap_and_free_vma
;
1230 if (vm_flags
& VM_EXECUTABLE
)
1231 added_exe_file_vma(mm
);
1233 /* Can addr have changed??
1235 * Answer: Yes, several device drivers can do it in their
1236 * f_op->mmap method. -DaveM
1238 addr
= vma
->vm_start
;
1239 pgoff
= vma
->vm_pgoff
;
1240 vm_flags
= vma
->vm_flags
;
1241 } else if (vm_flags
& VM_SHARED
) {
1242 error
= shmem_zero_setup(vma
);
1247 if (vma_wants_writenotify(vma
)) {
1248 pgprot_t pprot
= vma
->vm_page_prot
;
1250 /* Can vma->vm_page_prot have changed??
1252 * Answer: Yes, drivers may have changed it in their
1253 * f_op->mmap method.
1255 * Ensures that vmas marked as uncached stay that way.
1257 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
& ~VM_SHARED
);
1258 if (pgprot_val(pprot
) == pgprot_val(pgprot_noncached(pprot
)))
1259 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
1262 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1263 file
= vma
->vm_file
;
1265 /* Once vma denies write, undo our temporary denial count */
1267 atomic_inc(&inode
->i_writecount
);
1269 perf_event_mmap(vma
);
1271 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1272 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1273 if (vm_flags
& VM_LOCKED
) {
1274 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
1275 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1276 } else if ((flags
& MAP_POPULATE
) && !(flags
& MAP_NONBLOCK
))
1277 make_pages_present(addr
, addr
+ len
);
1282 atomic_inc(&inode
->i_writecount
);
1283 vma
->vm_file
= NULL
;
1286 /* Undo any partial mapping done by a device driver. */
1287 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1290 kmem_cache_free(vm_area_cachep
, vma
);
1293 vm_unacct_memory(charged
);
1297 /* Get an address range which is currently unmapped.
1298 * For shmat() with addr=0.
1300 * Ugly calling convention alert:
1301 * Return value with the low bits set means error value,
1303 * if (ret & ~PAGE_MASK)
1306 * This function "knows" that -ENOMEM has the bits set.
1308 #ifndef HAVE_ARCH_UNMAPPED_AREA
1310 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1311 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1313 struct mm_struct
*mm
= current
->mm
;
1314 struct vm_area_struct
*vma
;
1315 unsigned long start_addr
;
1317 if (len
> TASK_SIZE
)
1320 if (flags
& MAP_FIXED
)
1324 addr
= PAGE_ALIGN(addr
);
1325 vma
= find_vma(mm
, addr
);
1326 if (TASK_SIZE
- len
>= addr
&&
1327 (!vma
|| addr
+ len
<= vma
->vm_start
))
1330 if (len
> mm
->cached_hole_size
) {
1331 start_addr
= addr
= mm
->free_area_cache
;
1333 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1334 mm
->cached_hole_size
= 0;
1338 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1339 /* At this point: (!vma || addr < vma->vm_end). */
1340 if (TASK_SIZE
- len
< addr
) {
1342 * Start a new search - just in case we missed
1345 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1346 addr
= TASK_UNMAPPED_BASE
;
1348 mm
->cached_hole_size
= 0;
1353 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1355 * Remember the place where we stopped the search:
1357 mm
->free_area_cache
= addr
+ len
;
1360 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1361 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1367 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1370 * Is this a new hole at the lowest possible address?
1372 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
) {
1373 mm
->free_area_cache
= addr
;
1374 mm
->cached_hole_size
= ~0UL;
1379 * This mmap-allocator allocates new areas top-down from below the
1380 * stack's low limit (the base):
1382 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1384 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1385 const unsigned long len
, const unsigned long pgoff
,
1386 const unsigned long flags
)
1388 struct vm_area_struct
*vma
;
1389 struct mm_struct
*mm
= current
->mm
;
1390 unsigned long addr
= addr0
;
1392 /* requested length too big for entire address space */
1393 if (len
> TASK_SIZE
)
1396 if (flags
& MAP_FIXED
)
1399 /* requesting a specific address */
1401 addr
= PAGE_ALIGN(addr
);
1402 vma
= find_vma(mm
, addr
);
1403 if (TASK_SIZE
- len
>= addr
&&
1404 (!vma
|| addr
+ len
<= vma
->vm_start
))
1408 /* check if free_area_cache is useful for us */
1409 if (len
<= mm
->cached_hole_size
) {
1410 mm
->cached_hole_size
= 0;
1411 mm
->free_area_cache
= mm
->mmap_base
;
1414 /* either no address requested or can't fit in requested address hole */
1415 addr
= mm
->free_area_cache
;
1417 /* make sure it can fit in the remaining address space */
1419 vma
= find_vma(mm
, addr
-len
);
1420 if (!vma
|| addr
<= vma
->vm_start
)
1421 /* remember the address as a hint for next time */
1422 return (mm
->free_area_cache
= addr
-len
);
1425 if (mm
->mmap_base
< len
)
1428 addr
= mm
->mmap_base
-len
;
1432 * Lookup failure means no vma is above this address,
1433 * else if new region fits below vma->vm_start,
1434 * return with success:
1436 vma
= find_vma(mm
, addr
);
1437 if (!vma
|| addr
+len
<= vma
->vm_start
)
1438 /* remember the address as a hint for next time */
1439 return (mm
->free_area_cache
= addr
);
1441 /* remember the largest hole we saw so far */
1442 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1443 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1445 /* try just below the current vma->vm_start */
1446 addr
= vma
->vm_start
-len
;
1447 } while (len
< vma
->vm_start
);
1451 * A failed mmap() very likely causes application failure,
1452 * so fall back to the bottom-up function here. This scenario
1453 * can happen with large stack limits and large mmap()
1456 mm
->cached_hole_size
= ~0UL;
1457 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1458 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1460 * Restore the topdown base:
1462 mm
->free_area_cache
= mm
->mmap_base
;
1463 mm
->cached_hole_size
= ~0UL;
1469 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1472 * Is this a new hole at the highest possible address?
1474 if (addr
> mm
->free_area_cache
)
1475 mm
->free_area_cache
= addr
;
1477 /* dont allow allocations above current base */
1478 if (mm
->free_area_cache
> mm
->mmap_base
)
1479 mm
->free_area_cache
= mm
->mmap_base
;
1483 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1484 unsigned long pgoff
, unsigned long flags
)
1486 unsigned long (*get_area
)(struct file
*, unsigned long,
1487 unsigned long, unsigned long, unsigned long);
1489 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
1493 /* Careful about overflows.. */
1494 if (len
> TASK_SIZE
)
1497 get_area
= current
->mm
->get_unmapped_area
;
1498 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1499 get_area
= file
->f_op
->get_unmapped_area
;
1500 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1501 if (IS_ERR_VALUE(addr
))
1504 if (addr
> TASK_SIZE
- len
)
1506 if (addr
& ~PAGE_MASK
)
1509 return arch_rebalance_pgtables(addr
, len
);
1512 EXPORT_SYMBOL(get_unmapped_area
);
1514 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1515 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
1517 struct vm_area_struct
*vma
= NULL
;
1520 /* Check the cache first. */
1521 /* (Cache hit rate is typically around 35%.) */
1522 vma
= mm
->mmap_cache
;
1523 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1524 struct rb_node
* rb_node
;
1526 rb_node
= mm
->mm_rb
.rb_node
;
1530 struct vm_area_struct
* vma_tmp
;
1532 vma_tmp
= rb_entry(rb_node
,
1533 struct vm_area_struct
, vm_rb
);
1535 if (vma_tmp
->vm_end
> addr
) {
1537 if (vma_tmp
->vm_start
<= addr
)
1539 rb_node
= rb_node
->rb_left
;
1541 rb_node
= rb_node
->rb_right
;
1544 mm
->mmap_cache
= vma
;
1550 EXPORT_SYMBOL(find_vma
);
1552 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1553 struct vm_area_struct
*
1554 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1555 struct vm_area_struct
**pprev
)
1557 struct vm_area_struct
*vma
= NULL
, *prev
= NULL
;
1558 struct rb_node
*rb_node
;
1562 /* Guard against addr being lower than the first VMA */
1565 /* Go through the RB tree quickly. */
1566 rb_node
= mm
->mm_rb
.rb_node
;
1569 struct vm_area_struct
*vma_tmp
;
1570 vma_tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1572 if (addr
< vma_tmp
->vm_end
) {
1573 rb_node
= rb_node
->rb_left
;
1576 if (!prev
->vm_next
|| (addr
< prev
->vm_next
->vm_end
))
1578 rb_node
= rb_node
->rb_right
;
1584 return prev
? prev
->vm_next
: vma
;
1588 * Verify that the stack growth is acceptable and
1589 * update accounting. This is shared with both the
1590 * grow-up and grow-down cases.
1592 static int acct_stack_growth(struct vm_area_struct
*vma
, unsigned long size
, unsigned long grow
)
1594 struct mm_struct
*mm
= vma
->vm_mm
;
1595 struct rlimit
*rlim
= current
->signal
->rlim
;
1596 unsigned long new_start
;
1598 /* address space limit tests */
1599 if (!may_expand_vm(mm
, grow
))
1602 /* Stack limit test */
1603 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
))
1606 /* mlock limit tests */
1607 if (vma
->vm_flags
& VM_LOCKED
) {
1608 unsigned long locked
;
1609 unsigned long limit
;
1610 locked
= mm
->locked_vm
+ grow
;
1611 limit
= ACCESS_ONCE(rlim
[RLIMIT_MEMLOCK
].rlim_cur
);
1612 limit
>>= PAGE_SHIFT
;
1613 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1617 /* Check to ensure the stack will not grow into a hugetlb-only region */
1618 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
1620 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
1624 * Overcommit.. This must be the final test, as it will
1625 * update security statistics.
1627 if (security_vm_enough_memory_mm(mm
, grow
))
1630 /* Ok, everything looks good - let it rip */
1631 mm
->total_vm
+= grow
;
1632 if (vma
->vm_flags
& VM_LOCKED
)
1633 mm
->locked_vm
+= grow
;
1634 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1638 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1640 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1641 * vma is the last one with address > vma->vm_end. Have to extend vma.
1646 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1650 if (!(vma
->vm_flags
& VM_GROWSUP
))
1654 * We must make sure the anon_vma is allocated
1655 * so that the anon_vma locking is not a noop.
1657 if (unlikely(anon_vma_prepare(vma
)))
1662 * vma->vm_start/vm_end cannot change under us because the caller
1663 * is required to hold the mmap_sem in read mode. We need the
1664 * anon_vma lock to serialize against concurrent expand_stacks.
1665 * Also guard against wrapping around to address 0.
1667 if (address
< PAGE_ALIGN(address
+4))
1668 address
= PAGE_ALIGN(address
+4);
1670 anon_vma_unlock(vma
);
1675 /* Somebody else might have raced and expanded it already */
1676 if (address
> vma
->vm_end
) {
1677 unsigned long size
, grow
;
1679 size
= address
- vma
->vm_start
;
1680 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1682 error
= acct_stack_growth(vma
, size
, grow
);
1684 vma
->vm_end
= address
;
1686 anon_vma_unlock(vma
);
1689 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1692 * vma is the first one with address < vma->vm_start. Have to extend vma.
1694 static int expand_downwards(struct vm_area_struct
*vma
,
1695 unsigned long address
)
1700 * We must make sure the anon_vma is allocated
1701 * so that the anon_vma locking is not a noop.
1703 if (unlikely(anon_vma_prepare(vma
)))
1706 address
&= PAGE_MASK
;
1707 error
= security_file_mmap(NULL
, 0, 0, 0, address
, 1);
1714 * vma->vm_start/vm_end cannot change under us because the caller
1715 * is required to hold the mmap_sem in read mode. We need the
1716 * anon_vma lock to serialize against concurrent expand_stacks.
1719 /* Somebody else might have raced and expanded it already */
1720 if (address
< vma
->vm_start
) {
1721 unsigned long size
, grow
;
1723 size
= vma
->vm_end
- address
;
1724 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1726 error
= acct_stack_growth(vma
, size
, grow
);
1728 vma
->vm_start
= address
;
1729 vma
->vm_pgoff
-= grow
;
1732 anon_vma_unlock(vma
);
1736 int expand_stack_downwards(struct vm_area_struct
*vma
, unsigned long address
)
1738 return expand_downwards(vma
, address
);
1741 #ifdef CONFIG_STACK_GROWSUP
1742 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1744 return expand_upwards(vma
, address
);
1747 struct vm_area_struct
*
1748 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1750 struct vm_area_struct
*vma
, *prev
;
1753 vma
= find_vma_prev(mm
, addr
, &prev
);
1754 if (vma
&& (vma
->vm_start
<= addr
))
1756 if (!prev
|| expand_stack(prev
, addr
))
1758 if (prev
->vm_flags
& VM_LOCKED
) {
1759 mlock_vma_pages_range(prev
, addr
, prev
->vm_end
);
1764 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1766 return expand_downwards(vma
, address
);
1769 struct vm_area_struct
*
1770 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1772 struct vm_area_struct
* vma
;
1773 unsigned long start
;
1776 vma
= find_vma(mm
,addr
);
1779 if (vma
->vm_start
<= addr
)
1781 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1783 start
= vma
->vm_start
;
1784 if (expand_stack(vma
, addr
))
1786 if (vma
->vm_flags
& VM_LOCKED
) {
1787 mlock_vma_pages_range(vma
, addr
, start
);
1794 * Ok - we have the memory areas we should free on the vma list,
1795 * so release them, and do the vma updates.
1797 * Called with the mm semaphore held.
1799 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1801 /* Update high watermark before we lower total_vm */
1802 update_hiwater_vm(mm
);
1804 long nrpages
= vma_pages(vma
);
1806 mm
->total_vm
-= nrpages
;
1807 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1808 vma
= remove_vma(vma
);
1814 * Get rid of page table information in the indicated region.
1816 * Called with the mm semaphore held.
1818 static void unmap_region(struct mm_struct
*mm
,
1819 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1820 unsigned long start
, unsigned long end
)
1822 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1823 struct mmu_gather
*tlb
;
1824 unsigned long nr_accounted
= 0;
1827 tlb
= tlb_gather_mmu(mm
, 0);
1828 update_hiwater_rss(mm
);
1829 unmap_vmas(&tlb
, vma
, start
, end
, &nr_accounted
, NULL
);
1830 vm_unacct_memory(nr_accounted
);
1831 free_pgtables(tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1832 next
? next
->vm_start
: 0);
1833 tlb_finish_mmu(tlb
, start
, end
);
1837 * Create a list of vma's touched by the unmap, removing them from the mm's
1838 * vma list as we go..
1841 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1842 struct vm_area_struct
*prev
, unsigned long end
)
1844 struct vm_area_struct
**insertion_point
;
1845 struct vm_area_struct
*tail_vma
= NULL
;
1848 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1850 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1854 } while (vma
&& vma
->vm_start
< end
);
1855 *insertion_point
= vma
;
1856 tail_vma
->vm_next
= NULL
;
1857 if (mm
->unmap_area
== arch_unmap_area
)
1858 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1860 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1861 mm
->unmap_area(mm
, addr
);
1862 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1866 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1867 * munmap path where it doesn't make sense to fail.
1869 static int __split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1870 unsigned long addr
, int new_below
)
1872 struct mempolicy
*pol
;
1873 struct vm_area_struct
*new;
1876 if (is_vm_hugetlb_page(vma
) && (addr
&
1877 ~(huge_page_mask(hstate_vma(vma
)))))
1880 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1884 /* most fields are the same, copy all, and then fixup */
1887 INIT_LIST_HEAD(&new->anon_vma_chain
);
1892 new->vm_start
= addr
;
1893 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
1896 pol
= mpol_dup(vma_policy(vma
));
1901 vma_set_policy(new, pol
);
1903 if (anon_vma_clone(new, vma
))
1907 get_file(new->vm_file
);
1908 if (vma
->vm_flags
& VM_EXECUTABLE
)
1909 added_exe_file_vma(mm
);
1912 if (new->vm_ops
&& new->vm_ops
->open
)
1913 new->vm_ops
->open(new);
1916 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
1917 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
1919 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
1925 /* Clean everything up if vma_adjust failed. */
1926 new->vm_ops
->close(new);
1928 if (vma
->vm_flags
& VM_EXECUTABLE
)
1929 removed_exe_file_vma(mm
);
1935 kmem_cache_free(vm_area_cachep
, new);
1941 * Split a vma into two pieces at address 'addr', a new vma is allocated
1942 * either for the first part or the tail.
1944 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1945 unsigned long addr
, int new_below
)
1947 if (mm
->map_count
>= sysctl_max_map_count
)
1950 return __split_vma(mm
, vma
, addr
, new_below
);
1953 /* Munmap is split into 2 main parts -- this part which finds
1954 * what needs doing, and the areas themselves, which do the
1955 * work. This now handles partial unmappings.
1956 * Jeremy Fitzhardinge <jeremy@goop.org>
1958 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
1961 struct vm_area_struct
*vma
, *prev
, *last
;
1963 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
1966 if ((len
= PAGE_ALIGN(len
)) == 0)
1969 /* Find the first overlapping VMA */
1970 vma
= find_vma_prev(mm
, start
, &prev
);
1973 /* we have start < vma->vm_end */
1975 /* if it doesn't overlap, we have nothing.. */
1977 if (vma
->vm_start
>= end
)
1981 * If we need to split any vma, do it now to save pain later.
1983 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
1984 * unmapped vm_area_struct will remain in use: so lower split_vma
1985 * places tmp vma above, and higher split_vma places tmp vma below.
1987 if (start
> vma
->vm_start
) {
1991 * Make sure that map_count on return from munmap() will
1992 * not exceed its limit; but let map_count go just above
1993 * its limit temporarily, to help free resources as expected.
1995 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
1998 error
= __split_vma(mm
, vma
, start
, 0);
2004 /* Does it split the last one? */
2005 last
= find_vma(mm
, end
);
2006 if (last
&& end
> last
->vm_start
) {
2007 int error
= __split_vma(mm
, last
, end
, 1);
2011 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2014 * unlock any mlock()ed ranges before detaching vmas
2016 if (mm
->locked_vm
) {
2017 struct vm_area_struct
*tmp
= vma
;
2018 while (tmp
&& tmp
->vm_start
< end
) {
2019 if (tmp
->vm_flags
& VM_LOCKED
) {
2020 mm
->locked_vm
-= vma_pages(tmp
);
2021 munlock_vma_pages_all(tmp
);
2028 * Remove the vma's, and unmap the actual pages
2030 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2031 unmap_region(mm
, vma
, prev
, start
, end
);
2033 /* Fix up all other VM information */
2034 remove_vma_list(mm
, vma
);
2039 EXPORT_SYMBOL(do_munmap
);
2041 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2044 struct mm_struct
*mm
= current
->mm
;
2046 profile_munmap(addr
);
2048 down_write(&mm
->mmap_sem
);
2049 ret
= do_munmap(mm
, addr
, len
);
2050 up_write(&mm
->mmap_sem
);
2054 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
2056 #ifdef CONFIG_DEBUG_VM
2057 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
2059 up_read(&mm
->mmap_sem
);
2065 * this is really a simplified "do_mmap". it only handles
2066 * anonymous maps. eventually we may be able to do some
2067 * brk-specific accounting here.
2069 unsigned long do_brk(unsigned long addr
, unsigned long len
)
2071 struct mm_struct
* mm
= current
->mm
;
2072 struct vm_area_struct
* vma
, * prev
;
2073 unsigned long flags
;
2074 struct rb_node
** rb_link
, * rb_parent
;
2075 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2078 len
= PAGE_ALIGN(len
);
2082 error
= security_file_mmap(NULL
, 0, 0, 0, addr
, 1);
2086 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2088 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2089 if (error
& ~PAGE_MASK
)
2095 if (mm
->def_flags
& VM_LOCKED
) {
2096 unsigned long locked
, lock_limit
;
2097 locked
= len
>> PAGE_SHIFT
;
2098 locked
+= mm
->locked_vm
;
2099 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
2100 lock_limit
>>= PAGE_SHIFT
;
2101 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
2106 * mm->mmap_sem is required to protect against another thread
2107 * changing the mappings in case we sleep.
2109 verify_mm_writelocked(mm
);
2112 * Clear old maps. this also does some error checking for us
2115 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2116 if (vma
&& vma
->vm_start
< addr
+ len
) {
2117 if (do_munmap(mm
, addr
, len
))
2122 /* Check against address space limits *after* clearing old maps... */
2123 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2126 if (mm
->map_count
> sysctl_max_map_count
)
2129 if (security_vm_enough_memory(len
>> PAGE_SHIFT
))
2132 /* Can we just expand an old private anonymous mapping? */
2133 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2134 NULL
, NULL
, pgoff
, NULL
);
2139 * create a vma struct for an anonymous mapping
2141 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2143 vm_unacct_memory(len
>> PAGE_SHIFT
);
2147 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2149 vma
->vm_start
= addr
;
2150 vma
->vm_end
= addr
+ len
;
2151 vma
->vm_pgoff
= pgoff
;
2152 vma
->vm_flags
= flags
;
2153 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2154 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2156 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2157 if (flags
& VM_LOCKED
) {
2158 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
2159 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2164 EXPORT_SYMBOL(do_brk
);
2166 /* Release all mmaps. */
2167 void exit_mmap(struct mm_struct
*mm
)
2169 struct mmu_gather
*tlb
;
2170 struct vm_area_struct
*vma
;
2171 unsigned long nr_accounted
= 0;
2174 /* mm's last user has gone, and its about to be pulled down */
2175 mmu_notifier_release(mm
);
2177 if (mm
->locked_vm
) {
2180 if (vma
->vm_flags
& VM_LOCKED
)
2181 munlock_vma_pages_all(vma
);
2189 if (!vma
) /* Can happen if dup_mmap() received an OOM */
2194 tlb
= tlb_gather_mmu(mm
, 1);
2195 /* update_hiwater_rss(mm) here? but nobody should be looking */
2196 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2197 end
= unmap_vmas(&tlb
, vma
, 0, -1, &nr_accounted
, NULL
);
2198 vm_unacct_memory(nr_accounted
);
2200 free_pgtables(tlb
, vma
, FIRST_USER_ADDRESS
, 0);
2201 tlb_finish_mmu(tlb
, 0, end
);
2204 * Walk the list again, actually closing and freeing it,
2205 * with preemption enabled, without holding any MM locks.
2208 vma
= remove_vma(vma
);
2210 BUG_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2213 /* Insert vm structure into process list sorted by address
2214 * and into the inode's i_mmap tree. If vm_file is non-NULL
2215 * then i_mmap_lock is taken here.
2217 int insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
2219 struct vm_area_struct
* __vma
, * prev
;
2220 struct rb_node
** rb_link
, * rb_parent
;
2223 * The vm_pgoff of a purely anonymous vma should be irrelevant
2224 * until its first write fault, when page's anon_vma and index
2225 * are set. But now set the vm_pgoff it will almost certainly
2226 * end up with (unless mremap moves it elsewhere before that
2227 * first wfault), so /proc/pid/maps tells a consistent story.
2229 * By setting it to reflect the virtual start address of the
2230 * vma, merges and splits can happen in a seamless way, just
2231 * using the existing file pgoff checks and manipulations.
2232 * Similarly in do_mmap_pgoff and in do_brk.
2234 if (!vma
->vm_file
) {
2235 BUG_ON(vma
->anon_vma
);
2236 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2238 __vma
= find_vma_prepare(mm
,vma
->vm_start
,&prev
,&rb_link
,&rb_parent
);
2239 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
2241 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2242 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2244 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2249 * Copy the vma structure to a new location in the same mm,
2250 * prior to moving page table entries, to effect an mremap move.
2252 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2253 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
2255 struct vm_area_struct
*vma
= *vmap
;
2256 unsigned long vma_start
= vma
->vm_start
;
2257 struct mm_struct
*mm
= vma
->vm_mm
;
2258 struct vm_area_struct
*new_vma
, *prev
;
2259 struct rb_node
**rb_link
, *rb_parent
;
2260 struct mempolicy
*pol
;
2263 * If anonymous vma has not yet been faulted, update new pgoff
2264 * to match new location, to increase its chance of merging.
2266 if (!vma
->vm_file
&& !vma
->anon_vma
)
2267 pgoff
= addr
>> PAGE_SHIFT
;
2269 find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2270 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2271 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2274 * Source vma may have been merged into new_vma
2276 if (vma_start
>= new_vma
->vm_start
&&
2277 vma_start
< new_vma
->vm_end
)
2280 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2283 pol
= mpol_dup(vma_policy(vma
));
2286 INIT_LIST_HEAD(&new_vma
->anon_vma_chain
);
2287 if (anon_vma_clone(new_vma
, vma
))
2288 goto out_free_mempol
;
2289 vma_set_policy(new_vma
, pol
);
2290 new_vma
->vm_start
= addr
;
2291 new_vma
->vm_end
= addr
+ len
;
2292 new_vma
->vm_pgoff
= pgoff
;
2293 if (new_vma
->vm_file
) {
2294 get_file(new_vma
->vm_file
);
2295 if (vma
->vm_flags
& VM_EXECUTABLE
)
2296 added_exe_file_vma(mm
);
2298 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2299 new_vma
->vm_ops
->open(new_vma
);
2300 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2308 kmem_cache_free(vm_area_cachep
, new_vma
);
2313 * Return true if the calling process may expand its vm space by the passed
2316 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2318 unsigned long cur
= mm
->total_vm
; /* pages */
2321 lim
= rlimit(RLIMIT_AS
) >> PAGE_SHIFT
;
2323 if (cur
+ npages
> lim
)
2329 static int special_mapping_fault(struct vm_area_struct
*vma
,
2330 struct vm_fault
*vmf
)
2333 struct page
**pages
;
2336 * special mappings have no vm_file, and in that case, the mm
2337 * uses vm_pgoff internally. So we have to subtract it from here.
2338 * We are allowed to do this because we are the mm; do not copy
2339 * this code into drivers!
2341 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2343 for (pages
= vma
->vm_private_data
; pgoff
&& *pages
; ++pages
)
2347 struct page
*page
= *pages
;
2353 return VM_FAULT_SIGBUS
;
2357 * Having a close hook prevents vma merging regardless of flags.
2359 static void special_mapping_close(struct vm_area_struct
*vma
)
2363 static const struct vm_operations_struct special_mapping_vmops
= {
2364 .close
= special_mapping_close
,
2365 .fault
= special_mapping_fault
,
2369 * Called with mm->mmap_sem held for writing.
2370 * Insert a new vma covering the given region, with the given flags.
2371 * Its pages are supplied by the given array of struct page *.
2372 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2373 * The region past the last page supplied will always produce SIGBUS.
2374 * The array pointer and the pages it points to are assumed to stay alive
2375 * for as long as this mapping might exist.
2377 int install_special_mapping(struct mm_struct
*mm
,
2378 unsigned long addr
, unsigned long len
,
2379 unsigned long vm_flags
, struct page
**pages
)
2381 struct vm_area_struct
*vma
;
2383 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2384 if (unlikely(vma
== NULL
))
2387 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2389 vma
->vm_start
= addr
;
2390 vma
->vm_end
= addr
+ len
;
2392 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
;
2393 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
2395 vma
->vm_ops
= &special_mapping_vmops
;
2396 vma
->vm_private_data
= pages
;
2398 if (unlikely(insert_vm_struct(mm
, vma
))) {
2399 kmem_cache_free(vm_area_cachep
, vma
);
2403 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2405 perf_event_mmap(vma
);
2410 static DEFINE_MUTEX(mm_all_locks_mutex
);
2412 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
2414 if (!test_bit(0, (unsigned long *) &anon_vma
->head
.next
)) {
2416 * The LSB of head.next can't change from under us
2417 * because we hold the mm_all_locks_mutex.
2419 spin_lock_nest_lock(&anon_vma
->lock
, &mm
->mmap_sem
);
2421 * We can safely modify head.next after taking the
2422 * anon_vma->lock. If some other vma in this mm shares
2423 * the same anon_vma we won't take it again.
2425 * No need of atomic instructions here, head.next
2426 * can't change from under us thanks to the
2429 if (__test_and_set_bit(0, (unsigned long *)
2430 &anon_vma
->head
.next
))
2435 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
2437 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2439 * AS_MM_ALL_LOCKS can't change from under us because
2440 * we hold the mm_all_locks_mutex.
2442 * Operations on ->flags have to be atomic because
2443 * even if AS_MM_ALL_LOCKS is stable thanks to the
2444 * mm_all_locks_mutex, there may be other cpus
2445 * changing other bitflags in parallel to us.
2447 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
2449 spin_lock_nest_lock(&mapping
->i_mmap_lock
, &mm
->mmap_sem
);
2454 * This operation locks against the VM for all pte/vma/mm related
2455 * operations that could ever happen on a certain mm. This includes
2456 * vmtruncate, try_to_unmap, and all page faults.
2458 * The caller must take the mmap_sem in write mode before calling
2459 * mm_take_all_locks(). The caller isn't allowed to release the
2460 * mmap_sem until mm_drop_all_locks() returns.
2462 * mmap_sem in write mode is required in order to block all operations
2463 * that could modify pagetables and free pages without need of
2464 * altering the vma layout (for example populate_range() with
2465 * nonlinear vmas). It's also needed in write mode to avoid new
2466 * anon_vmas to be associated with existing vmas.
2468 * A single task can't take more than one mm_take_all_locks() in a row
2469 * or it would deadlock.
2471 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2472 * mapping->flags avoid to take the same lock twice, if more than one
2473 * vma in this mm is backed by the same anon_vma or address_space.
2475 * We can take all the locks in random order because the VM code
2476 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2477 * takes more than one of them in a row. Secondly we're protected
2478 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2480 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2481 * that may have to take thousand of locks.
2483 * mm_take_all_locks() can fail if it's interrupted by signals.
2485 int mm_take_all_locks(struct mm_struct
*mm
)
2487 struct vm_area_struct
*vma
;
2488 struct anon_vma_chain
*avc
;
2491 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2493 mutex_lock(&mm_all_locks_mutex
);
2495 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2496 if (signal_pending(current
))
2498 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2499 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
2502 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2503 if (signal_pending(current
))
2506 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2507 vm_lock_anon_vma(mm
, avc
->anon_vma
);
2514 mm_drop_all_locks(mm
);
2519 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
2521 if (test_bit(0, (unsigned long *) &anon_vma
->head
.next
)) {
2523 * The LSB of head.next can't change to 0 from under
2524 * us because we hold the mm_all_locks_mutex.
2526 * We must however clear the bitflag before unlocking
2527 * the vma so the users using the anon_vma->head will
2528 * never see our bitflag.
2530 * No need of atomic instructions here, head.next
2531 * can't change from under us until we release the
2534 if (!__test_and_clear_bit(0, (unsigned long *)
2535 &anon_vma
->head
.next
))
2537 spin_unlock(&anon_vma
->lock
);
2541 static void vm_unlock_mapping(struct address_space
*mapping
)
2543 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2545 * AS_MM_ALL_LOCKS can't change to 0 from under us
2546 * because we hold the mm_all_locks_mutex.
2548 spin_unlock(&mapping
->i_mmap_lock
);
2549 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
2556 * The mmap_sem cannot be released by the caller until
2557 * mm_drop_all_locks() returns.
2559 void mm_drop_all_locks(struct mm_struct
*mm
)
2561 struct vm_area_struct
*vma
;
2562 struct anon_vma_chain
*avc
;
2564 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2565 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
2567 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2569 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2570 vm_unlock_anon_vma(avc
->anon_vma
);
2571 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2572 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
2575 mutex_unlock(&mm_all_locks_mutex
);
2579 * initialise the VMA slab
2581 void __init
mmap_init(void)
2585 ret
= percpu_counter_init(&vm_committed_as
, 0);