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
)
391 struct vm_area_struct
*next
;
395 next
= prev
->vm_next
;
400 next
= rb_entry(rb_parent
,
401 struct vm_area_struct
, vm_rb
);
410 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
411 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
413 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
414 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
417 static void __vma_link_file(struct vm_area_struct
*vma
)
423 struct address_space
*mapping
= file
->f_mapping
;
425 if (vma
->vm_flags
& VM_DENYWRITE
)
426 atomic_dec(&file
->f_path
.dentry
->d_inode
->i_writecount
);
427 if (vma
->vm_flags
& VM_SHARED
)
428 mapping
->i_mmap_writable
++;
430 flush_dcache_mmap_lock(mapping
);
431 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
432 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
434 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
435 flush_dcache_mmap_unlock(mapping
);
440 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
441 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
442 struct rb_node
*rb_parent
)
444 __vma_link_list(mm
, vma
, prev
, rb_parent
);
445 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
448 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
449 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
450 struct rb_node
*rb_parent
)
452 struct address_space
*mapping
= NULL
;
455 mapping
= vma
->vm_file
->f_mapping
;
458 spin_lock(&mapping
->i_mmap_lock
);
459 vma
->vm_truncate_count
= mapping
->truncate_count
;
463 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
464 __vma_link_file(vma
);
466 anon_vma_unlock(vma
);
468 spin_unlock(&mapping
->i_mmap_lock
);
475 * Helper for vma_adjust in the split_vma insert case:
476 * insert vm structure into list and rbtree and anon_vma,
477 * but it has already been inserted into prio_tree earlier.
479 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
481 struct vm_area_struct
*__vma
, *prev
;
482 struct rb_node
**rb_link
, *rb_parent
;
484 __vma
= find_vma_prepare(mm
, vma
->vm_start
,&prev
, &rb_link
, &rb_parent
);
485 BUG_ON(__vma
&& __vma
->vm_start
< vma
->vm_end
);
486 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
491 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
492 struct vm_area_struct
*prev
)
494 struct vm_area_struct
*next
= vma
->vm_next
;
496 prev
->vm_next
= next
;
498 next
->vm_prev
= prev
;
499 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
500 if (mm
->mmap_cache
== vma
)
501 mm
->mmap_cache
= prev
;
505 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
506 * is already present in an i_mmap tree without adjusting the tree.
507 * The following helper function should be used when such adjustments
508 * are necessary. The "insert" vma (if any) is to be inserted
509 * before we drop the necessary locks.
511 int vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
512 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
514 struct mm_struct
*mm
= vma
->vm_mm
;
515 struct vm_area_struct
*next
= vma
->vm_next
;
516 struct vm_area_struct
*importer
= NULL
;
517 struct address_space
*mapping
= NULL
;
518 struct prio_tree_root
*root
= NULL
;
519 struct file
*file
= vma
->vm_file
;
520 long adjust_next
= 0;
523 if (next
&& !insert
) {
524 struct vm_area_struct
*exporter
= NULL
;
526 if (end
>= next
->vm_end
) {
528 * vma expands, overlapping all the next, and
529 * perhaps the one after too (mprotect case 6).
531 again
: remove_next
= 1 + (end
> next
->vm_end
);
535 } else if (end
> next
->vm_start
) {
537 * vma expands, overlapping part of the next:
538 * mprotect case 5 shifting the boundary up.
540 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
543 } else if (end
< vma
->vm_end
) {
545 * vma shrinks, and !insert tells it's not
546 * split_vma inserting another: so it must be
547 * mprotect case 4 shifting the boundary down.
549 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
555 * Easily overlooked: when mprotect shifts the boundary,
556 * make sure the expanding vma has anon_vma set if the
557 * shrinking vma had, to cover any anon pages imported.
559 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
560 if (anon_vma_clone(importer
, exporter
))
562 importer
->anon_vma
= exporter
->anon_vma
;
567 mapping
= file
->f_mapping
;
568 if (!(vma
->vm_flags
& VM_NONLINEAR
))
569 root
= &mapping
->i_mmap
;
570 spin_lock(&mapping
->i_mmap_lock
);
572 vma
->vm_truncate_count
!= next
->vm_truncate_count
) {
574 * unmap_mapping_range might be in progress:
575 * ensure that the expanding vma is rescanned.
577 importer
->vm_truncate_count
= 0;
580 insert
->vm_truncate_count
= vma
->vm_truncate_count
;
582 * Put into prio_tree now, so instantiated pages
583 * are visible to arm/parisc __flush_dcache_page
584 * throughout; but we cannot insert into address
585 * space until vma start or end is updated.
587 __vma_link_file(insert
);
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
);
623 * split_vma has split insert from vma, and needs
624 * us to insert it before dropping the locks
625 * (it may either follow vma or precede it).
627 __insert_vm_struct(mm
, insert
);
631 spin_unlock(&mapping
->i_mmap_lock
);
636 if (next
->vm_flags
& VM_EXECUTABLE
)
637 removed_exe_file_vma(mm
);
640 anon_vma_merge(vma
, next
);
642 mpol_put(vma_policy(next
));
643 kmem_cache_free(vm_area_cachep
, next
);
645 * In mprotect's case 6 (see comments on vma_merge),
646 * we must remove another next too. It would clutter
647 * up the code too much to do both in one go.
649 if (remove_next
== 2) {
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
;
767 * We later require that vma->vm_flags == vm_flags,
768 * so this tests vma->vm_flags & VM_SPECIAL, too.
770 if (vm_flags
& VM_SPECIAL
)
774 next
= prev
->vm_next
;
778 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
779 next
= next
->vm_next
;
782 * Can it merge with the predecessor?
784 if (prev
&& prev
->vm_end
== addr
&&
785 mpol_equal(vma_policy(prev
), policy
) &&
786 can_vma_merge_after(prev
, vm_flags
,
787 anon_vma
, file
, pgoff
)) {
789 * OK, it can. Can we now merge in the successor as well?
791 if (next
&& end
== next
->vm_start
&&
792 mpol_equal(policy
, vma_policy(next
)) &&
793 can_vma_merge_before(next
, vm_flags
,
794 anon_vma
, file
, pgoff
+pglen
) &&
795 is_mergeable_anon_vma(prev
->anon_vma
,
798 err
= vma_adjust(prev
, prev
->vm_start
,
799 next
->vm_end
, prev
->vm_pgoff
, NULL
);
800 } else /* cases 2, 5, 7 */
801 err
= vma_adjust(prev
, prev
->vm_start
,
802 end
, prev
->vm_pgoff
, NULL
);
809 * Can this new request be merged in front of next?
811 if (next
&& end
== next
->vm_start
&&
812 mpol_equal(policy
, vma_policy(next
)) &&
813 can_vma_merge_before(next
, vm_flags
,
814 anon_vma
, file
, pgoff
+pglen
)) {
815 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
816 err
= vma_adjust(prev
, prev
->vm_start
,
817 addr
, prev
->vm_pgoff
, NULL
);
818 else /* cases 3, 8 */
819 err
= vma_adjust(area
, addr
, next
->vm_end
,
820 next
->vm_pgoff
- pglen
, NULL
);
830 * Rough compatbility check to quickly see if it's even worth looking
831 * at sharing an anon_vma.
833 * They need to have the same vm_file, and the flags can only differ
834 * in things that mprotect may change.
836 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
837 * we can merge the two vma's. For example, we refuse to merge a vma if
838 * there is a vm_ops->close() function, because that indicates that the
839 * driver is doing some kind of reference counting. But that doesn't
840 * really matter for the anon_vma sharing case.
842 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
844 return a
->vm_end
== b
->vm_start
&&
845 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
846 a
->vm_file
== b
->vm_file
&&
847 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_READ
|VM_WRITE
|VM_EXEC
)) &&
848 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
852 * Do some basic sanity checking to see if we can re-use the anon_vma
853 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
854 * the same as 'old', the other will be the new one that is trying
855 * to share the anon_vma.
857 * NOTE! This runs with mm_sem held for reading, so it is possible that
858 * the anon_vma of 'old' is concurrently in the process of being set up
859 * by another page fault trying to merge _that_. But that's ok: if it
860 * is being set up, that automatically means that it will be a singleton
861 * acceptable for merging, so we can do all of this optimistically. But
862 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
864 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
865 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
866 * is to return an anon_vma that is "complex" due to having gone through
869 * We also make sure that the two vma's are compatible (adjacent,
870 * and with the same memory policies). That's all stable, even with just
871 * a read lock on the mm_sem.
873 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
875 if (anon_vma_compatible(a
, b
)) {
876 struct anon_vma
*anon_vma
= ACCESS_ONCE(old
->anon_vma
);
878 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
885 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
886 * neighbouring vmas for a suitable anon_vma, before it goes off
887 * to allocate a new anon_vma. It checks because a repetitive
888 * sequence of mprotects and faults may otherwise lead to distinct
889 * anon_vmas being allocated, preventing vma merge in subsequent
892 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
894 struct anon_vma
*anon_vma
;
895 struct vm_area_struct
*near
;
901 anon_vma
= reusable_anon_vma(near
, vma
, near
);
906 * It is potentially slow to have to call find_vma_prev here.
907 * But it's only on the first write fault on the vma, not
908 * every time, and we could devise a way to avoid it later
909 * (e.g. stash info in next's anon_vma_node when assigning
910 * an anon_vma, or when trying vma_merge). Another time.
912 BUG_ON(find_vma_prev(vma
->vm_mm
, vma
->vm_start
, &near
) != vma
);
916 anon_vma
= reusable_anon_vma(near
, near
, vma
);
921 * There's no absolute need to look only at touching neighbours:
922 * we could search further afield for "compatible" anon_vmas.
923 * But it would probably just be a waste of time searching,
924 * or lead to too many vmas hanging off the same anon_vma.
925 * We're trying to allow mprotect remerging later on,
926 * not trying to minimize memory used for anon_vmas.
931 #ifdef CONFIG_PROC_FS
932 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
933 struct file
*file
, long pages
)
935 const unsigned long stack_flags
936 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
939 mm
->shared_vm
+= pages
;
940 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
941 mm
->exec_vm
+= pages
;
942 } else if (flags
& stack_flags
)
943 mm
->stack_vm
+= pages
;
944 if (flags
& (VM_RESERVED
|VM_IO
))
945 mm
->reserved_vm
+= pages
;
947 #endif /* CONFIG_PROC_FS */
950 * The caller must hold down_write(¤t->mm->mmap_sem).
953 unsigned long do_mmap_pgoff(struct file
*file
, unsigned long addr
,
954 unsigned long len
, unsigned long prot
,
955 unsigned long flags
, unsigned long pgoff
)
957 struct mm_struct
* mm
= current
->mm
;
959 unsigned int vm_flags
;
961 unsigned long reqprot
= prot
;
964 * Does the application expect PROT_READ to imply PROT_EXEC?
966 * (the exception is when the underlying filesystem is noexec
967 * mounted, in which case we dont add PROT_EXEC.)
969 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
970 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
976 if (!(flags
& MAP_FIXED
))
977 addr
= round_hint_to_min(addr
);
979 /* Careful about overflows.. */
980 len
= PAGE_ALIGN(len
);
984 /* offset overflow? */
985 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
988 /* Too many mappings? */
989 if (mm
->map_count
> sysctl_max_map_count
)
992 /* Obtain the address to map to. we verify (or select) it and ensure
993 * that it represents a valid section of the address space.
995 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
996 if (addr
& ~PAGE_MASK
)
999 /* Do simple checking here so the lower-level routines won't have
1000 * to. we assume access permissions have been handled by the open
1001 * of the memory object, so we don't do any here.
1003 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
1004 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1006 if (flags
& MAP_LOCKED
)
1007 if (!can_do_mlock())
1010 /* mlock MCL_FUTURE? */
1011 if (vm_flags
& VM_LOCKED
) {
1012 unsigned long locked
, lock_limit
;
1013 locked
= len
>> PAGE_SHIFT
;
1014 locked
+= mm
->locked_vm
;
1015 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1016 lock_limit
>>= PAGE_SHIFT
;
1017 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1021 inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1024 switch (flags
& MAP_TYPE
) {
1026 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
1030 * Make sure we don't allow writing to an append-only
1033 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1037 * Make sure there are no mandatory locks on the file.
1039 if (locks_verify_locked(inode
))
1042 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1043 if (!(file
->f_mode
& FMODE_WRITE
))
1044 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1048 if (!(file
->f_mode
& FMODE_READ
))
1050 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1051 if (vm_flags
& VM_EXEC
)
1053 vm_flags
&= ~VM_MAYEXEC
;
1056 if (!file
->f_op
|| !file
->f_op
->mmap
)
1064 switch (flags
& MAP_TYPE
) {
1070 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1074 * Set pgoff according to addr for anon_vma.
1076 pgoff
= addr
>> PAGE_SHIFT
;
1083 error
= security_file_mmap(file
, reqprot
, prot
, flags
, addr
, 0);
1087 return mmap_region(file
, addr
, len
, flags
, vm_flags
, pgoff
);
1089 EXPORT_SYMBOL(do_mmap_pgoff
);
1091 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1092 unsigned long, prot
, unsigned long, flags
,
1093 unsigned long, fd
, unsigned long, pgoff
)
1095 struct file
*file
= NULL
;
1096 unsigned long retval
= -EBADF
;
1098 if (!(flags
& MAP_ANONYMOUS
)) {
1099 if (unlikely(flags
& MAP_HUGETLB
))
1104 } else if (flags
& MAP_HUGETLB
) {
1105 struct user_struct
*user
= NULL
;
1107 * VM_NORESERVE is used because the reservations will be
1108 * taken when vm_ops->mmap() is called
1109 * A dummy user value is used because we are not locking
1110 * memory so no accounting is necessary
1112 len
= ALIGN(len
, huge_page_size(&default_hstate
));
1113 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, len
, VM_NORESERVE
,
1114 &user
, HUGETLB_ANONHUGE_INODE
);
1116 return PTR_ERR(file
);
1119 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1121 down_write(¤t
->mm
->mmap_sem
);
1122 retval
= do_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1123 up_write(¤t
->mm
->mmap_sem
);
1131 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1132 struct mmap_arg_struct
{
1136 unsigned long flags
;
1138 unsigned long offset
;
1141 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1143 struct mmap_arg_struct a
;
1145 if (copy_from_user(&a
, arg
, sizeof(a
)))
1147 if (a
.offset
& ~PAGE_MASK
)
1150 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1151 a
.offset
>> PAGE_SHIFT
);
1153 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1156 * Some shared mappigns will want the pages marked read-only
1157 * to track write events. If so, we'll downgrade vm_page_prot
1158 * to the private version (using protection_map[] without the
1161 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1163 unsigned int vm_flags
= vma
->vm_flags
;
1165 /* If it was private or non-writable, the write bit is already clear */
1166 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1169 /* The backer wishes to know when pages are first written to? */
1170 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1173 /* The open routine did something to the protections already? */
1174 if (pgprot_val(vma
->vm_page_prot
) !=
1175 pgprot_val(vm_get_page_prot(vm_flags
)))
1178 /* Specialty mapping? */
1179 if (vm_flags
& (VM_PFNMAP
|VM_INSERTPAGE
))
1182 /* Can the mapping track the dirty pages? */
1183 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1184 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1188 * We account for memory if it's a private writeable mapping,
1189 * not hugepages and VM_NORESERVE wasn't set.
1191 static inline int accountable_mapping(struct file
*file
, unsigned int vm_flags
)
1194 * hugetlb has its own accounting separate from the core VM
1195 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1197 if (file
&& is_file_hugepages(file
))
1200 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1203 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1204 unsigned long len
, unsigned long flags
,
1205 unsigned int vm_flags
, unsigned long pgoff
)
1207 struct mm_struct
*mm
= current
->mm
;
1208 struct vm_area_struct
*vma
, *prev
;
1209 int correct_wcount
= 0;
1211 struct rb_node
**rb_link
, *rb_parent
;
1212 unsigned long charged
= 0;
1213 struct inode
*inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1215 /* Clear old maps */
1218 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1219 if (vma
&& vma
->vm_start
< addr
+ len
) {
1220 if (do_munmap(mm
, addr
, len
))
1225 /* Check against address space limit. */
1226 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1230 * Set 'VM_NORESERVE' if we should not account for the
1231 * memory use of this mapping.
1233 if ((flags
& MAP_NORESERVE
)) {
1234 /* We honor MAP_NORESERVE if allowed to overcommit */
1235 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1236 vm_flags
|= VM_NORESERVE
;
1238 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1239 if (file
&& is_file_hugepages(file
))
1240 vm_flags
|= VM_NORESERVE
;
1244 * Private writable mapping: check memory availability
1246 if (accountable_mapping(file
, vm_flags
)) {
1247 charged
= len
>> PAGE_SHIFT
;
1248 if (security_vm_enough_memory(charged
))
1250 vm_flags
|= VM_ACCOUNT
;
1254 * Can we just expand an old mapping?
1256 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
, NULL
, file
, pgoff
, NULL
);
1261 * Determine the object being mapped and call the appropriate
1262 * specific mapper. the address has already been validated, but
1263 * not unmapped, but the maps are removed from the list.
1265 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1272 vma
->vm_start
= addr
;
1273 vma
->vm_end
= addr
+ len
;
1274 vma
->vm_flags
= vm_flags
;
1275 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1276 vma
->vm_pgoff
= pgoff
;
1277 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1281 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1283 if (vm_flags
& VM_DENYWRITE
) {
1284 error
= deny_write_access(file
);
1289 vma
->vm_file
= file
;
1291 error
= file
->f_op
->mmap(file
, vma
);
1293 goto unmap_and_free_vma
;
1294 if (vm_flags
& VM_EXECUTABLE
)
1295 added_exe_file_vma(mm
);
1297 /* Can addr have changed??
1299 * Answer: Yes, several device drivers can do it in their
1300 * f_op->mmap method. -DaveM
1302 addr
= vma
->vm_start
;
1303 pgoff
= vma
->vm_pgoff
;
1304 vm_flags
= vma
->vm_flags
;
1305 } else if (vm_flags
& VM_SHARED
) {
1306 error
= shmem_zero_setup(vma
);
1311 if (vma_wants_writenotify(vma
)) {
1312 pgprot_t pprot
= vma
->vm_page_prot
;
1314 /* Can vma->vm_page_prot have changed??
1316 * Answer: Yes, drivers may have changed it in their
1317 * f_op->mmap method.
1319 * Ensures that vmas marked as uncached stay that way.
1321 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
& ~VM_SHARED
);
1322 if (pgprot_val(pprot
) == pgprot_val(pgprot_noncached(pprot
)))
1323 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
1326 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1327 file
= vma
->vm_file
;
1329 /* Once vma denies write, undo our temporary denial count */
1331 atomic_inc(&inode
->i_writecount
);
1333 perf_event_mmap(vma
);
1335 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1336 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1337 if (vm_flags
& VM_LOCKED
) {
1338 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
1339 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1340 } else if ((flags
& MAP_POPULATE
) && !(flags
& MAP_NONBLOCK
))
1341 make_pages_present(addr
, addr
+ len
);
1346 atomic_inc(&inode
->i_writecount
);
1347 vma
->vm_file
= NULL
;
1350 /* Undo any partial mapping done by a device driver. */
1351 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1354 kmem_cache_free(vm_area_cachep
, vma
);
1357 vm_unacct_memory(charged
);
1361 /* Get an address range which is currently unmapped.
1362 * For shmat() with addr=0.
1364 * Ugly calling convention alert:
1365 * Return value with the low bits set means error value,
1367 * if (ret & ~PAGE_MASK)
1370 * This function "knows" that -ENOMEM has the bits set.
1372 #ifndef HAVE_ARCH_UNMAPPED_AREA
1374 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1375 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1377 struct mm_struct
*mm
= current
->mm
;
1378 struct vm_area_struct
*vma
;
1379 unsigned long start_addr
;
1381 if (len
> TASK_SIZE
)
1384 if (flags
& MAP_FIXED
)
1388 addr
= PAGE_ALIGN(addr
);
1389 vma
= find_vma(mm
, addr
);
1390 if (TASK_SIZE
- len
>= addr
&&
1391 (!vma
|| addr
+ len
<= vma
->vm_start
))
1394 if (len
> mm
->cached_hole_size
) {
1395 start_addr
= addr
= mm
->free_area_cache
;
1397 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1398 mm
->cached_hole_size
= 0;
1402 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1403 /* At this point: (!vma || addr < vma->vm_end). */
1404 if (TASK_SIZE
- len
< addr
) {
1406 * Start a new search - just in case we missed
1409 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1410 addr
= TASK_UNMAPPED_BASE
;
1412 mm
->cached_hole_size
= 0;
1417 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1419 * Remember the place where we stopped the search:
1421 mm
->free_area_cache
= addr
+ len
;
1424 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1425 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1431 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1434 * Is this a new hole at the lowest possible address?
1436 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
) {
1437 mm
->free_area_cache
= addr
;
1438 mm
->cached_hole_size
= ~0UL;
1443 * This mmap-allocator allocates new areas top-down from below the
1444 * stack's low limit (the base):
1446 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1448 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1449 const unsigned long len
, const unsigned long pgoff
,
1450 const unsigned long flags
)
1452 struct vm_area_struct
*vma
;
1453 struct mm_struct
*mm
= current
->mm
;
1454 unsigned long addr
= addr0
;
1456 /* requested length too big for entire address space */
1457 if (len
> TASK_SIZE
)
1460 if (flags
& MAP_FIXED
)
1463 /* requesting a specific address */
1465 addr
= PAGE_ALIGN(addr
);
1466 vma
= find_vma(mm
, addr
);
1467 if (TASK_SIZE
- len
>= addr
&&
1468 (!vma
|| addr
+ len
<= vma
->vm_start
))
1472 /* check if free_area_cache is useful for us */
1473 if (len
<= mm
->cached_hole_size
) {
1474 mm
->cached_hole_size
= 0;
1475 mm
->free_area_cache
= mm
->mmap_base
;
1478 /* either no address requested or can't fit in requested address hole */
1479 addr
= mm
->free_area_cache
;
1481 /* make sure it can fit in the remaining address space */
1483 vma
= find_vma(mm
, addr
-len
);
1484 if (!vma
|| addr
<= vma
->vm_start
)
1485 /* remember the address as a hint for next time */
1486 return (mm
->free_area_cache
= addr
-len
);
1489 if (mm
->mmap_base
< len
)
1492 addr
= mm
->mmap_base
-len
;
1496 * Lookup failure means no vma is above this address,
1497 * else if new region fits below vma->vm_start,
1498 * return with success:
1500 vma
= find_vma(mm
, addr
);
1501 if (!vma
|| addr
+len
<= vma
->vm_start
)
1502 /* remember the address as a hint for next time */
1503 return (mm
->free_area_cache
= addr
);
1505 /* remember the largest hole we saw so far */
1506 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1507 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1509 /* try just below the current vma->vm_start */
1510 addr
= vma
->vm_start
-len
;
1511 } while (len
< vma
->vm_start
);
1515 * A failed mmap() very likely causes application failure,
1516 * so fall back to the bottom-up function here. This scenario
1517 * can happen with large stack limits and large mmap()
1520 mm
->cached_hole_size
= ~0UL;
1521 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1522 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1524 * Restore the topdown base:
1526 mm
->free_area_cache
= mm
->mmap_base
;
1527 mm
->cached_hole_size
= ~0UL;
1533 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1536 * Is this a new hole at the highest possible address?
1538 if (addr
> mm
->free_area_cache
)
1539 mm
->free_area_cache
= addr
;
1541 /* dont allow allocations above current base */
1542 if (mm
->free_area_cache
> mm
->mmap_base
)
1543 mm
->free_area_cache
= mm
->mmap_base
;
1547 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1548 unsigned long pgoff
, unsigned long flags
)
1550 unsigned long (*get_area
)(struct file
*, unsigned long,
1551 unsigned long, unsigned long, unsigned long);
1553 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
1557 /* Careful about overflows.. */
1558 if (len
> TASK_SIZE
)
1561 get_area
= current
->mm
->get_unmapped_area
;
1562 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1563 get_area
= file
->f_op
->get_unmapped_area
;
1564 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1565 if (IS_ERR_VALUE(addr
))
1568 if (addr
> TASK_SIZE
- len
)
1570 if (addr
& ~PAGE_MASK
)
1573 return arch_rebalance_pgtables(addr
, len
);
1576 EXPORT_SYMBOL(get_unmapped_area
);
1578 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1579 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
1581 struct vm_area_struct
*vma
= NULL
;
1584 /* Check the cache first. */
1585 /* (Cache hit rate is typically around 35%.) */
1586 vma
= mm
->mmap_cache
;
1587 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1588 struct rb_node
* rb_node
;
1590 rb_node
= mm
->mm_rb
.rb_node
;
1594 struct vm_area_struct
* vma_tmp
;
1596 vma_tmp
= rb_entry(rb_node
,
1597 struct vm_area_struct
, vm_rb
);
1599 if (vma_tmp
->vm_end
> addr
) {
1601 if (vma_tmp
->vm_start
<= addr
)
1603 rb_node
= rb_node
->rb_left
;
1605 rb_node
= rb_node
->rb_right
;
1608 mm
->mmap_cache
= vma
;
1614 EXPORT_SYMBOL(find_vma
);
1616 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1617 struct vm_area_struct
*
1618 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1619 struct vm_area_struct
**pprev
)
1621 struct vm_area_struct
*vma
= NULL
, *prev
= NULL
;
1622 struct rb_node
*rb_node
;
1626 /* Guard against addr being lower than the first VMA */
1629 /* Go through the RB tree quickly. */
1630 rb_node
= mm
->mm_rb
.rb_node
;
1633 struct vm_area_struct
*vma_tmp
;
1634 vma_tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1636 if (addr
< vma_tmp
->vm_end
) {
1637 rb_node
= rb_node
->rb_left
;
1640 if (!prev
->vm_next
|| (addr
< prev
->vm_next
->vm_end
))
1642 rb_node
= rb_node
->rb_right
;
1648 return prev
? prev
->vm_next
: vma
;
1652 * Verify that the stack growth is acceptable and
1653 * update accounting. This is shared with both the
1654 * grow-up and grow-down cases.
1656 static int acct_stack_growth(struct vm_area_struct
*vma
, unsigned long size
, unsigned long grow
)
1658 struct mm_struct
*mm
= vma
->vm_mm
;
1659 struct rlimit
*rlim
= current
->signal
->rlim
;
1660 unsigned long new_start
;
1662 /* address space limit tests */
1663 if (!may_expand_vm(mm
, grow
))
1666 /* Stack limit test */
1667 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
))
1670 /* mlock limit tests */
1671 if (vma
->vm_flags
& VM_LOCKED
) {
1672 unsigned long locked
;
1673 unsigned long limit
;
1674 locked
= mm
->locked_vm
+ grow
;
1675 limit
= ACCESS_ONCE(rlim
[RLIMIT_MEMLOCK
].rlim_cur
);
1676 limit
>>= PAGE_SHIFT
;
1677 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1681 /* Check to ensure the stack will not grow into a hugetlb-only region */
1682 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
1684 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
1688 * Overcommit.. This must be the final test, as it will
1689 * update security statistics.
1691 if (security_vm_enough_memory_mm(mm
, grow
))
1694 /* Ok, everything looks good - let it rip */
1695 mm
->total_vm
+= grow
;
1696 if (vma
->vm_flags
& VM_LOCKED
)
1697 mm
->locked_vm
+= grow
;
1698 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1702 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1704 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1705 * vma is the last one with address > vma->vm_end. Have to extend vma.
1707 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1711 if (!(vma
->vm_flags
& VM_GROWSUP
))
1715 * We must make sure the anon_vma is allocated
1716 * so that the anon_vma locking is not a noop.
1718 if (unlikely(anon_vma_prepare(vma
)))
1723 * vma->vm_start/vm_end cannot change under us because the caller
1724 * is required to hold the mmap_sem in read mode. We need the
1725 * anon_vma lock to serialize against concurrent expand_stacks.
1726 * Also guard against wrapping around to address 0.
1728 if (address
< PAGE_ALIGN(address
+4))
1729 address
= PAGE_ALIGN(address
+4);
1731 anon_vma_unlock(vma
);
1736 /* Somebody else might have raced and expanded it already */
1737 if (address
> vma
->vm_end
) {
1738 unsigned long size
, grow
;
1740 size
= address
- vma
->vm_start
;
1741 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1743 error
= acct_stack_growth(vma
, size
, grow
);
1745 vma
->vm_end
= address
;
1747 anon_vma_unlock(vma
);
1750 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1753 * vma is the first one with address < vma->vm_start. Have to extend vma.
1755 static int expand_downwards(struct vm_area_struct
*vma
,
1756 unsigned long address
)
1761 * We must make sure the anon_vma is allocated
1762 * so that the anon_vma locking is not a noop.
1764 if (unlikely(anon_vma_prepare(vma
)))
1767 address
&= PAGE_MASK
;
1768 error
= security_file_mmap(NULL
, 0, 0, 0, address
, 1);
1775 * vma->vm_start/vm_end cannot change under us because the caller
1776 * is required to hold the mmap_sem in read mode. We need the
1777 * anon_vma lock to serialize against concurrent expand_stacks.
1780 /* Somebody else might have raced and expanded it already */
1781 if (address
< vma
->vm_start
) {
1782 unsigned long size
, grow
;
1784 size
= vma
->vm_end
- address
;
1785 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1787 error
= acct_stack_growth(vma
, size
, grow
);
1789 vma
->vm_start
= address
;
1790 vma
->vm_pgoff
-= grow
;
1793 anon_vma_unlock(vma
);
1797 int expand_stack_downwards(struct vm_area_struct
*vma
, unsigned long address
)
1799 return expand_downwards(vma
, address
);
1802 #ifdef CONFIG_STACK_GROWSUP
1803 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1805 return expand_upwards(vma
, address
);
1808 struct vm_area_struct
*
1809 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1811 struct vm_area_struct
*vma
, *prev
;
1814 vma
= find_vma_prev(mm
, addr
, &prev
);
1815 if (vma
&& (vma
->vm_start
<= addr
))
1817 if (!prev
|| expand_stack(prev
, addr
))
1819 if (prev
->vm_flags
& VM_LOCKED
) {
1820 mlock_vma_pages_range(prev
, addr
, prev
->vm_end
);
1825 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1827 return expand_downwards(vma
, address
);
1830 struct vm_area_struct
*
1831 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1833 struct vm_area_struct
* vma
;
1834 unsigned long start
;
1837 vma
= find_vma(mm
,addr
);
1840 if (vma
->vm_start
<= addr
)
1842 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1844 start
= vma
->vm_start
;
1845 if (expand_stack(vma
, addr
))
1847 if (vma
->vm_flags
& VM_LOCKED
) {
1848 mlock_vma_pages_range(vma
, addr
, start
);
1855 * Ok - we have the memory areas we should free on the vma list,
1856 * so release them, and do the vma updates.
1858 * Called with the mm semaphore held.
1860 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1862 /* Update high watermark before we lower total_vm */
1863 update_hiwater_vm(mm
);
1865 long nrpages
= vma_pages(vma
);
1867 mm
->total_vm
-= nrpages
;
1868 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1869 vma
= remove_vma(vma
);
1875 * Get rid of page table information in the indicated region.
1877 * Called with the mm semaphore held.
1879 static void unmap_region(struct mm_struct
*mm
,
1880 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1881 unsigned long start
, unsigned long end
)
1883 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1884 struct mmu_gather
*tlb
;
1885 unsigned long nr_accounted
= 0;
1888 tlb
= tlb_gather_mmu(mm
, 0);
1889 update_hiwater_rss(mm
);
1890 unmap_vmas(&tlb
, vma
, start
, end
, &nr_accounted
, NULL
);
1891 vm_unacct_memory(nr_accounted
);
1892 free_pgtables(tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1893 next
? next
->vm_start
: 0);
1894 tlb_finish_mmu(tlb
, start
, end
);
1898 * Create a list of vma's touched by the unmap, removing them from the mm's
1899 * vma list as we go..
1902 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1903 struct vm_area_struct
*prev
, unsigned long end
)
1905 struct vm_area_struct
**insertion_point
;
1906 struct vm_area_struct
*tail_vma
= NULL
;
1909 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1910 vma
->vm_prev
= NULL
;
1912 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1916 } while (vma
&& vma
->vm_start
< end
);
1917 *insertion_point
= vma
;
1919 vma
->vm_prev
= prev
;
1920 tail_vma
->vm_next
= NULL
;
1921 if (mm
->unmap_area
== arch_unmap_area
)
1922 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1924 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1925 mm
->unmap_area(mm
, addr
);
1926 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1930 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1931 * munmap path where it doesn't make sense to fail.
1933 static int __split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1934 unsigned long addr
, int new_below
)
1936 struct mempolicy
*pol
;
1937 struct vm_area_struct
*new;
1940 if (is_vm_hugetlb_page(vma
) && (addr
&
1941 ~(huge_page_mask(hstate_vma(vma
)))))
1944 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1948 /* most fields are the same, copy all, and then fixup */
1951 INIT_LIST_HEAD(&new->anon_vma_chain
);
1956 new->vm_start
= addr
;
1957 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
1960 pol
= mpol_dup(vma_policy(vma
));
1965 vma_set_policy(new, pol
);
1967 if (anon_vma_clone(new, vma
))
1971 get_file(new->vm_file
);
1972 if (vma
->vm_flags
& VM_EXECUTABLE
)
1973 added_exe_file_vma(mm
);
1976 if (new->vm_ops
&& new->vm_ops
->open
)
1977 new->vm_ops
->open(new);
1980 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
1981 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
1983 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
1989 /* Clean everything up if vma_adjust failed. */
1990 if (new->vm_ops
&& new->vm_ops
->close
)
1991 new->vm_ops
->close(new);
1993 if (vma
->vm_flags
& VM_EXECUTABLE
)
1994 removed_exe_file_vma(mm
);
1997 unlink_anon_vmas(new);
2001 kmem_cache_free(vm_area_cachep
, new);
2007 * Split a vma into two pieces at address 'addr', a new vma is allocated
2008 * either for the first part or the tail.
2010 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2011 unsigned long addr
, int new_below
)
2013 if (mm
->map_count
>= sysctl_max_map_count
)
2016 return __split_vma(mm
, vma
, addr
, new_below
);
2019 /* Munmap is split into 2 main parts -- this part which finds
2020 * what needs doing, and the areas themselves, which do the
2021 * work. This now handles partial unmappings.
2022 * Jeremy Fitzhardinge <jeremy@goop.org>
2024 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
2027 struct vm_area_struct
*vma
, *prev
, *last
;
2029 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2032 if ((len
= PAGE_ALIGN(len
)) == 0)
2035 /* Find the first overlapping VMA */
2036 vma
= find_vma_prev(mm
, start
, &prev
);
2039 /* we have start < vma->vm_end */
2041 /* if it doesn't overlap, we have nothing.. */
2043 if (vma
->vm_start
>= end
)
2047 * If we need to split any vma, do it now to save pain later.
2049 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2050 * unmapped vm_area_struct will remain in use: so lower split_vma
2051 * places tmp vma above, and higher split_vma places tmp vma below.
2053 if (start
> vma
->vm_start
) {
2057 * Make sure that map_count on return from munmap() will
2058 * not exceed its limit; but let map_count go just above
2059 * its limit temporarily, to help free resources as expected.
2061 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2064 error
= __split_vma(mm
, vma
, start
, 0);
2070 /* Does it split the last one? */
2071 last
= find_vma(mm
, end
);
2072 if (last
&& end
> last
->vm_start
) {
2073 int error
= __split_vma(mm
, last
, end
, 1);
2077 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2080 * unlock any mlock()ed ranges before detaching vmas
2082 if (mm
->locked_vm
) {
2083 struct vm_area_struct
*tmp
= vma
;
2084 while (tmp
&& tmp
->vm_start
< end
) {
2085 if (tmp
->vm_flags
& VM_LOCKED
) {
2086 mm
->locked_vm
-= vma_pages(tmp
);
2087 munlock_vma_pages_all(tmp
);
2094 * Remove the vma's, and unmap the actual pages
2096 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2097 unmap_region(mm
, vma
, prev
, start
, end
);
2099 /* Fix up all other VM information */
2100 remove_vma_list(mm
, vma
);
2105 EXPORT_SYMBOL(do_munmap
);
2107 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2110 struct mm_struct
*mm
= current
->mm
;
2112 profile_munmap(addr
);
2114 down_write(&mm
->mmap_sem
);
2115 ret
= do_munmap(mm
, addr
, len
);
2116 up_write(&mm
->mmap_sem
);
2120 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
2122 #ifdef CONFIG_DEBUG_VM
2123 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
2125 up_read(&mm
->mmap_sem
);
2131 * this is really a simplified "do_mmap". it only handles
2132 * anonymous maps. eventually we may be able to do some
2133 * brk-specific accounting here.
2135 unsigned long do_brk(unsigned long addr
, unsigned long len
)
2137 struct mm_struct
* mm
= current
->mm
;
2138 struct vm_area_struct
* vma
, * prev
;
2139 unsigned long flags
;
2140 struct rb_node
** rb_link
, * rb_parent
;
2141 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2144 len
= PAGE_ALIGN(len
);
2148 error
= security_file_mmap(NULL
, 0, 0, 0, addr
, 1);
2152 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2154 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2155 if (error
& ~PAGE_MASK
)
2161 if (mm
->def_flags
& VM_LOCKED
) {
2162 unsigned long locked
, lock_limit
;
2163 locked
= len
>> PAGE_SHIFT
;
2164 locked
+= mm
->locked_vm
;
2165 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
2166 lock_limit
>>= PAGE_SHIFT
;
2167 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
2172 * mm->mmap_sem is required to protect against another thread
2173 * changing the mappings in case we sleep.
2175 verify_mm_writelocked(mm
);
2178 * Clear old maps. this also does some error checking for us
2181 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2182 if (vma
&& vma
->vm_start
< addr
+ len
) {
2183 if (do_munmap(mm
, addr
, len
))
2188 /* Check against address space limits *after* clearing old maps... */
2189 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2192 if (mm
->map_count
> sysctl_max_map_count
)
2195 if (security_vm_enough_memory(len
>> PAGE_SHIFT
))
2198 /* Can we just expand an old private anonymous mapping? */
2199 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2200 NULL
, NULL
, pgoff
, NULL
);
2205 * create a vma struct for an anonymous mapping
2207 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2209 vm_unacct_memory(len
>> PAGE_SHIFT
);
2213 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2215 vma
->vm_start
= addr
;
2216 vma
->vm_end
= addr
+ len
;
2217 vma
->vm_pgoff
= pgoff
;
2218 vma
->vm_flags
= flags
;
2219 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2220 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2222 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2223 if (flags
& VM_LOCKED
) {
2224 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
2225 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2230 EXPORT_SYMBOL(do_brk
);
2232 /* Release all mmaps. */
2233 void exit_mmap(struct mm_struct
*mm
)
2235 struct mmu_gather
*tlb
;
2236 struct vm_area_struct
*vma
;
2237 unsigned long nr_accounted
= 0;
2240 /* mm's last user has gone, and its about to be pulled down */
2241 mmu_notifier_release(mm
);
2243 if (mm
->locked_vm
) {
2246 if (vma
->vm_flags
& VM_LOCKED
)
2247 munlock_vma_pages_all(vma
);
2255 if (!vma
) /* Can happen if dup_mmap() received an OOM */
2260 tlb
= tlb_gather_mmu(mm
, 1);
2261 /* update_hiwater_rss(mm) here? but nobody should be looking */
2262 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2263 end
= unmap_vmas(&tlb
, vma
, 0, -1, &nr_accounted
, NULL
);
2264 vm_unacct_memory(nr_accounted
);
2266 free_pgtables(tlb
, vma
, FIRST_USER_ADDRESS
, 0);
2267 tlb_finish_mmu(tlb
, 0, end
);
2270 * Walk the list again, actually closing and freeing it,
2271 * with preemption enabled, without holding any MM locks.
2274 vma
= remove_vma(vma
);
2276 BUG_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2279 /* Insert vm structure into process list sorted by address
2280 * and into the inode's i_mmap tree. If vm_file is non-NULL
2281 * then i_mmap_lock is taken here.
2283 int insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
2285 struct vm_area_struct
* __vma
, * prev
;
2286 struct rb_node
** rb_link
, * rb_parent
;
2289 * The vm_pgoff of a purely anonymous vma should be irrelevant
2290 * until its first write fault, when page's anon_vma and index
2291 * are set. But now set the vm_pgoff it will almost certainly
2292 * end up with (unless mremap moves it elsewhere before that
2293 * first wfault), so /proc/pid/maps tells a consistent story.
2295 * By setting it to reflect the virtual start address of the
2296 * vma, merges and splits can happen in a seamless way, just
2297 * using the existing file pgoff checks and manipulations.
2298 * Similarly in do_mmap_pgoff and in do_brk.
2300 if (!vma
->vm_file
) {
2301 BUG_ON(vma
->anon_vma
);
2302 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2304 __vma
= find_vma_prepare(mm
,vma
->vm_start
,&prev
,&rb_link
,&rb_parent
);
2305 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
2307 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2308 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2310 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2315 * Copy the vma structure to a new location in the same mm,
2316 * prior to moving page table entries, to effect an mremap move.
2318 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2319 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
2321 struct vm_area_struct
*vma
= *vmap
;
2322 unsigned long vma_start
= vma
->vm_start
;
2323 struct mm_struct
*mm
= vma
->vm_mm
;
2324 struct vm_area_struct
*new_vma
, *prev
;
2325 struct rb_node
**rb_link
, *rb_parent
;
2326 struct mempolicy
*pol
;
2329 * If anonymous vma has not yet been faulted, update new pgoff
2330 * to match new location, to increase its chance of merging.
2332 if (!vma
->vm_file
&& !vma
->anon_vma
)
2333 pgoff
= addr
>> PAGE_SHIFT
;
2335 find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2336 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2337 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2340 * Source vma may have been merged into new_vma
2342 if (vma_start
>= new_vma
->vm_start
&&
2343 vma_start
< new_vma
->vm_end
)
2346 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2349 pol
= mpol_dup(vma_policy(vma
));
2352 INIT_LIST_HEAD(&new_vma
->anon_vma_chain
);
2353 if (anon_vma_clone(new_vma
, vma
))
2354 goto out_free_mempol
;
2355 vma_set_policy(new_vma
, pol
);
2356 new_vma
->vm_start
= addr
;
2357 new_vma
->vm_end
= addr
+ len
;
2358 new_vma
->vm_pgoff
= pgoff
;
2359 if (new_vma
->vm_file
) {
2360 get_file(new_vma
->vm_file
);
2361 if (vma
->vm_flags
& VM_EXECUTABLE
)
2362 added_exe_file_vma(mm
);
2364 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2365 new_vma
->vm_ops
->open(new_vma
);
2366 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2374 kmem_cache_free(vm_area_cachep
, new_vma
);
2379 * Return true if the calling process may expand its vm space by the passed
2382 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2384 unsigned long cur
= mm
->total_vm
; /* pages */
2387 lim
= rlimit(RLIMIT_AS
) >> PAGE_SHIFT
;
2389 if (cur
+ npages
> lim
)
2395 static int special_mapping_fault(struct vm_area_struct
*vma
,
2396 struct vm_fault
*vmf
)
2399 struct page
**pages
;
2402 * special mappings have no vm_file, and in that case, the mm
2403 * uses vm_pgoff internally. So we have to subtract it from here.
2404 * We are allowed to do this because we are the mm; do not copy
2405 * this code into drivers!
2407 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2409 for (pages
= vma
->vm_private_data
; pgoff
&& *pages
; ++pages
)
2413 struct page
*page
= *pages
;
2419 return VM_FAULT_SIGBUS
;
2423 * Having a close hook prevents vma merging regardless of flags.
2425 static void special_mapping_close(struct vm_area_struct
*vma
)
2429 static const struct vm_operations_struct special_mapping_vmops
= {
2430 .close
= special_mapping_close
,
2431 .fault
= special_mapping_fault
,
2435 * Called with mm->mmap_sem held for writing.
2436 * Insert a new vma covering the given region, with the given flags.
2437 * Its pages are supplied by the given array of struct page *.
2438 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2439 * The region past the last page supplied will always produce SIGBUS.
2440 * The array pointer and the pages it points to are assumed to stay alive
2441 * for as long as this mapping might exist.
2443 int install_special_mapping(struct mm_struct
*mm
,
2444 unsigned long addr
, unsigned long len
,
2445 unsigned long vm_flags
, struct page
**pages
)
2447 struct vm_area_struct
*vma
;
2449 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2450 if (unlikely(vma
== NULL
))
2453 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2455 vma
->vm_start
= addr
;
2456 vma
->vm_end
= addr
+ len
;
2458 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
;
2459 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
2461 vma
->vm_ops
= &special_mapping_vmops
;
2462 vma
->vm_private_data
= pages
;
2464 if (unlikely(insert_vm_struct(mm
, vma
))) {
2465 kmem_cache_free(vm_area_cachep
, vma
);
2469 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2471 perf_event_mmap(vma
);
2476 static DEFINE_MUTEX(mm_all_locks_mutex
);
2478 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
2480 if (!test_bit(0, (unsigned long *) &anon_vma
->head
.next
)) {
2482 * The LSB of head.next can't change from under us
2483 * because we hold the mm_all_locks_mutex.
2485 spin_lock_nest_lock(&anon_vma
->lock
, &mm
->mmap_sem
);
2487 * We can safely modify head.next after taking the
2488 * anon_vma->lock. If some other vma in this mm shares
2489 * the same anon_vma we won't take it again.
2491 * No need of atomic instructions here, head.next
2492 * can't change from under us thanks to the
2495 if (__test_and_set_bit(0, (unsigned long *)
2496 &anon_vma
->head
.next
))
2501 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
2503 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2505 * AS_MM_ALL_LOCKS can't change from under us because
2506 * we hold the mm_all_locks_mutex.
2508 * Operations on ->flags have to be atomic because
2509 * even if AS_MM_ALL_LOCKS is stable thanks to the
2510 * mm_all_locks_mutex, there may be other cpus
2511 * changing other bitflags in parallel to us.
2513 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
2515 spin_lock_nest_lock(&mapping
->i_mmap_lock
, &mm
->mmap_sem
);
2520 * This operation locks against the VM for all pte/vma/mm related
2521 * operations that could ever happen on a certain mm. This includes
2522 * vmtruncate, try_to_unmap, and all page faults.
2524 * The caller must take the mmap_sem in write mode before calling
2525 * mm_take_all_locks(). The caller isn't allowed to release the
2526 * mmap_sem until mm_drop_all_locks() returns.
2528 * mmap_sem in write mode is required in order to block all operations
2529 * that could modify pagetables and free pages without need of
2530 * altering the vma layout (for example populate_range() with
2531 * nonlinear vmas). It's also needed in write mode to avoid new
2532 * anon_vmas to be associated with existing vmas.
2534 * A single task can't take more than one mm_take_all_locks() in a row
2535 * or it would deadlock.
2537 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2538 * mapping->flags avoid to take the same lock twice, if more than one
2539 * vma in this mm is backed by the same anon_vma or address_space.
2541 * We can take all the locks in random order because the VM code
2542 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2543 * takes more than one of them in a row. Secondly we're protected
2544 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2546 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2547 * that may have to take thousand of locks.
2549 * mm_take_all_locks() can fail if it's interrupted by signals.
2551 int mm_take_all_locks(struct mm_struct
*mm
)
2553 struct vm_area_struct
*vma
;
2554 struct anon_vma_chain
*avc
;
2557 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2559 mutex_lock(&mm_all_locks_mutex
);
2561 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2562 if (signal_pending(current
))
2564 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2565 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
2568 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2569 if (signal_pending(current
))
2572 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2573 vm_lock_anon_vma(mm
, avc
->anon_vma
);
2580 mm_drop_all_locks(mm
);
2585 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
2587 if (test_bit(0, (unsigned long *) &anon_vma
->head
.next
)) {
2589 * The LSB of head.next can't change to 0 from under
2590 * us because we hold the mm_all_locks_mutex.
2592 * We must however clear the bitflag before unlocking
2593 * the vma so the users using the anon_vma->head will
2594 * never see our bitflag.
2596 * No need of atomic instructions here, head.next
2597 * can't change from under us until we release the
2600 if (!__test_and_clear_bit(0, (unsigned long *)
2601 &anon_vma
->head
.next
))
2603 spin_unlock(&anon_vma
->lock
);
2607 static void vm_unlock_mapping(struct address_space
*mapping
)
2609 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2611 * AS_MM_ALL_LOCKS can't change to 0 from under us
2612 * because we hold the mm_all_locks_mutex.
2614 spin_unlock(&mapping
->i_mmap_lock
);
2615 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
2622 * The mmap_sem cannot be released by the caller until
2623 * mm_drop_all_locks() returns.
2625 void mm_drop_all_locks(struct mm_struct
*mm
)
2627 struct vm_area_struct
*vma
;
2628 struct anon_vma_chain
*avc
;
2630 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2631 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
2633 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2635 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2636 vm_unlock_anon_vma(avc
->anon_vma
);
2637 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2638 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
2641 mutex_unlock(&mm_all_locks_mutex
);
2645 * initialise the VMA slab
2647 void __init
mmap_init(void)
2651 ret
= percpu_counter_init(&vm_committed_as
, 0);