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/export.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>
31 #include <linux/audit.h>
32 #include <linux/khugepaged.h>
33 #include <linux/uprobes.h>
35 #include <asm/uaccess.h>
36 #include <asm/cacheflush.h>
38 #include <asm/mmu_context.h>
42 #ifndef arch_mmap_check
43 #define arch_mmap_check(addr, len, flags) (0)
46 #ifndef arch_rebalance_pgtables
47 #define arch_rebalance_pgtables(addr, len) (addr)
50 static void unmap_region(struct mm_struct
*mm
,
51 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
52 unsigned long start
, unsigned long end
);
55 * WARNING: the debugging will use recursive algorithms so never enable this
56 * unless you know what you are doing.
60 /* description of effects of mapping type and prot in current implementation.
61 * this is due to the limited x86 page protection hardware. The expected
62 * behavior is in parens:
65 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
66 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
67 * w: (no) no w: (no) no w: (yes) yes w: (no) no
68 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
70 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
71 * w: (no) no w: (no) no w: (copy) copy w: (no) no
72 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
75 pgprot_t protection_map
[16] = {
76 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
77 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
80 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
82 return __pgprot(pgprot_val(protection_map
[vm_flags
&
83 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
84 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
86 EXPORT_SYMBOL(vm_get_page_prot
);
88 int sysctl_overcommit_memory __read_mostly
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
89 int sysctl_overcommit_ratio __read_mostly
= 50; /* default is 50% */
90 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
92 * Make sure vm_committed_as in one cacheline and not cacheline shared with
93 * other variables. It can be updated by several CPUs frequently.
95 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp
;
98 * Check that a process has enough memory to allocate a new virtual
99 * mapping. 0 means there is enough memory for the allocation to
100 * succeed and -ENOMEM implies there is not.
102 * We currently support three overcommit policies, which are set via the
103 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
105 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
106 * Additional code 2002 Jul 20 by Robert Love.
108 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
110 * Note this is a helper function intended to be used by LSMs which
111 * wish to use this logic.
113 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
115 unsigned long free
, allowed
;
117 vm_acct_memory(pages
);
120 * Sometimes we want to use more memory than we have
122 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
125 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
126 free
= global_page_state(NR_FREE_PAGES
);
127 free
+= global_page_state(NR_FILE_PAGES
);
130 * shmem pages shouldn't be counted as free in this
131 * case, they can't be purged, only swapped out, and
132 * that won't affect the overall amount of available
133 * memory in the system.
135 free
-= global_page_state(NR_SHMEM
);
137 free
+= nr_swap_pages
;
140 * Any slabs which are created with the
141 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
142 * which are reclaimable, under pressure. The dentry
143 * cache and most inode caches should fall into this
145 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
148 * Leave reserved pages. The pages are not for anonymous pages.
150 if (free
<= totalreserve_pages
)
153 free
-= totalreserve_pages
;
156 * Leave the last 3% for root
167 allowed
= (totalram_pages
- hugetlb_total_pages())
168 * sysctl_overcommit_ratio
/ 100;
170 * Leave the last 3% for root
173 allowed
-= allowed
/ 32;
174 allowed
+= total_swap_pages
;
176 /* Don't let a single process grow too big:
177 leave 3% of the size of this process for other processes */
179 allowed
-= mm
->total_vm
/ 32;
181 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
184 vm_unacct_memory(pages
);
190 * Requires inode->i_mapping->i_mmap_mutex
192 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
193 struct file
*file
, struct address_space
*mapping
)
195 if (vma
->vm_flags
& VM_DENYWRITE
)
196 atomic_inc(&file
->f_path
.dentry
->d_inode
->i_writecount
);
197 if (vma
->vm_flags
& VM_SHARED
)
198 mapping
->i_mmap_writable
--;
200 flush_dcache_mmap_lock(mapping
);
201 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
202 list_del_init(&vma
->shared
.vm_set
.list
);
204 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
205 flush_dcache_mmap_unlock(mapping
);
209 * Unlink a file-based vm structure from its prio_tree, to hide
210 * vma from rmap and vmtruncate before freeing its page tables.
212 void unlink_file_vma(struct vm_area_struct
*vma
)
214 struct file
*file
= vma
->vm_file
;
217 struct address_space
*mapping
= file
->f_mapping
;
218 mutex_lock(&mapping
->i_mmap_mutex
);
219 __remove_shared_vm_struct(vma
, file
, mapping
);
220 mutex_unlock(&mapping
->i_mmap_mutex
);
225 * Close a vm structure and free it, returning the next.
227 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
229 struct vm_area_struct
*next
= vma
->vm_next
;
232 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
233 vma
->vm_ops
->close(vma
);
236 mpol_put(vma_policy(vma
));
237 kmem_cache_free(vm_area_cachep
, vma
);
241 static unsigned long do_brk(unsigned long addr
, unsigned long len
);
243 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
245 unsigned long rlim
, retval
;
246 unsigned long newbrk
, oldbrk
;
247 struct mm_struct
*mm
= current
->mm
;
248 unsigned long min_brk
;
250 down_write(&mm
->mmap_sem
);
252 #ifdef CONFIG_COMPAT_BRK
254 * CONFIG_COMPAT_BRK can still be overridden by setting
255 * randomize_va_space to 2, which will still cause mm->start_brk
256 * to be arbitrarily shifted
258 if (current
->brk_randomized
)
259 min_brk
= mm
->start_brk
;
261 min_brk
= mm
->end_data
;
263 min_brk
= mm
->start_brk
;
269 * Check against rlimit here. If this check is done later after the test
270 * of oldbrk with newbrk then it can escape the test and let the data
271 * segment grow beyond its set limit the in case where the limit is
272 * not page aligned -Ram Gupta
274 rlim
= rlimit(RLIMIT_DATA
);
275 if (rlim
< RLIM_INFINITY
&& (brk
- mm
->start_brk
) +
276 (mm
->end_data
- mm
->start_data
) > rlim
)
279 newbrk
= PAGE_ALIGN(brk
);
280 oldbrk
= PAGE_ALIGN(mm
->brk
);
281 if (oldbrk
== newbrk
)
284 /* Always allow shrinking brk. */
285 if (brk
<= mm
->brk
) {
286 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
291 /* Check against existing mmap mappings. */
292 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
295 /* Ok, looks good - let it rip. */
296 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
302 up_write(&mm
->mmap_sem
);
307 static int browse_rb(struct rb_root
*root
)
310 struct rb_node
*nd
, *pn
= NULL
;
311 unsigned long prev
= 0, pend
= 0;
313 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
314 struct vm_area_struct
*vma
;
315 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
316 if (vma
->vm_start
< prev
)
317 printk("vm_start %lx prev %lx\n", vma
->vm_start
, prev
), i
= -1;
318 if (vma
->vm_start
< pend
)
319 printk("vm_start %lx pend %lx\n", vma
->vm_start
, pend
);
320 if (vma
->vm_start
> vma
->vm_end
)
321 printk("vm_end %lx < vm_start %lx\n", vma
->vm_end
, vma
->vm_start
);
324 prev
= vma
->vm_start
;
328 for (nd
= pn
; nd
; nd
= rb_prev(nd
)) {
332 printk("backwards %d, forwards %d\n", j
, i
), i
= 0;
336 void validate_mm(struct mm_struct
*mm
)
340 struct vm_area_struct
*tmp
= mm
->mmap
;
345 if (i
!= mm
->map_count
)
346 printk("map_count %d vm_next %d\n", mm
->map_count
, i
), bug
= 1;
347 i
= browse_rb(&mm
->mm_rb
);
348 if (i
!= mm
->map_count
)
349 printk("map_count %d rb %d\n", mm
->map_count
, i
), bug
= 1;
353 #define validate_mm(mm) do { } while (0)
356 static struct vm_area_struct
*
357 find_vma_prepare(struct mm_struct
*mm
, unsigned long addr
,
358 struct vm_area_struct
**pprev
, struct rb_node
***rb_link
,
359 struct rb_node
** rb_parent
)
361 struct vm_area_struct
* vma
;
362 struct rb_node
** __rb_link
, * __rb_parent
, * rb_prev
;
364 __rb_link
= &mm
->mm_rb
.rb_node
;
365 rb_prev
= __rb_parent
= NULL
;
369 struct vm_area_struct
*vma_tmp
;
371 __rb_parent
= *__rb_link
;
372 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
374 if (vma_tmp
->vm_end
> addr
) {
376 if (vma_tmp
->vm_start
<= addr
)
378 __rb_link
= &__rb_parent
->rb_left
;
380 rb_prev
= __rb_parent
;
381 __rb_link
= &__rb_parent
->rb_right
;
387 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
388 *rb_link
= __rb_link
;
389 *rb_parent
= __rb_parent
;
393 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
394 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
396 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
397 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
400 static void __vma_link_file(struct vm_area_struct
*vma
)
406 struct address_space
*mapping
= file
->f_mapping
;
408 if (vma
->vm_flags
& VM_DENYWRITE
)
409 atomic_dec(&file
->f_path
.dentry
->d_inode
->i_writecount
);
410 if (vma
->vm_flags
& VM_SHARED
)
411 mapping
->i_mmap_writable
++;
413 flush_dcache_mmap_lock(mapping
);
414 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
415 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
417 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
418 flush_dcache_mmap_unlock(mapping
);
423 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
424 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
425 struct rb_node
*rb_parent
)
427 __vma_link_list(mm
, vma
, prev
, rb_parent
);
428 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
431 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
432 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
433 struct rb_node
*rb_parent
)
435 struct address_space
*mapping
= NULL
;
438 mapping
= vma
->vm_file
->f_mapping
;
441 mutex_lock(&mapping
->i_mmap_mutex
);
443 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
444 __vma_link_file(vma
);
447 mutex_unlock(&mapping
->i_mmap_mutex
);
454 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
455 * mm's list and rbtree. It has already been inserted into the prio_tree.
457 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
459 struct vm_area_struct
*__vma
, *prev
;
460 struct rb_node
**rb_link
, *rb_parent
;
462 __vma
= find_vma_prepare(mm
, vma
->vm_start
,&prev
, &rb_link
, &rb_parent
);
463 BUG_ON(__vma
&& __vma
->vm_start
< vma
->vm_end
);
464 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
469 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
470 struct vm_area_struct
*prev
)
472 struct vm_area_struct
*next
= vma
->vm_next
;
474 prev
->vm_next
= next
;
476 next
->vm_prev
= prev
;
477 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
478 if (mm
->mmap_cache
== vma
)
479 mm
->mmap_cache
= prev
;
483 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
484 * is already present in an i_mmap tree without adjusting the tree.
485 * The following helper function should be used when such adjustments
486 * are necessary. The "insert" vma (if any) is to be inserted
487 * before we drop the necessary locks.
489 int vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
490 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
492 struct mm_struct
*mm
= vma
->vm_mm
;
493 struct vm_area_struct
*next
= vma
->vm_next
;
494 struct vm_area_struct
*importer
= NULL
;
495 struct address_space
*mapping
= NULL
;
496 struct prio_tree_root
*root
= NULL
;
497 struct anon_vma
*anon_vma
= NULL
;
498 struct file
*file
= vma
->vm_file
;
499 long adjust_next
= 0;
502 if (next
&& !insert
) {
503 struct vm_area_struct
*exporter
= NULL
;
505 if (end
>= next
->vm_end
) {
507 * vma expands, overlapping all the next, and
508 * perhaps the one after too (mprotect case 6).
510 again
: remove_next
= 1 + (end
> next
->vm_end
);
514 } else if (end
> next
->vm_start
) {
516 * vma expands, overlapping part of the next:
517 * mprotect case 5 shifting the boundary up.
519 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
522 } else if (end
< vma
->vm_end
) {
524 * vma shrinks, and !insert tells it's not
525 * split_vma inserting another: so it must be
526 * mprotect case 4 shifting the boundary down.
528 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
534 * Easily overlooked: when mprotect shifts the boundary,
535 * make sure the expanding vma has anon_vma set if the
536 * shrinking vma had, to cover any anon pages imported.
538 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
539 if (anon_vma_clone(importer
, exporter
))
541 importer
->anon_vma
= exporter
->anon_vma
;
546 mapping
= file
->f_mapping
;
547 if (!(vma
->vm_flags
& VM_NONLINEAR
)) {
548 root
= &mapping
->i_mmap
;
549 uprobe_munmap(vma
, vma
->vm_start
, vma
->vm_end
);
552 uprobe_munmap(next
, next
->vm_start
,
556 mutex_lock(&mapping
->i_mmap_mutex
);
559 * Put into prio_tree now, so instantiated pages
560 * are visible to arm/parisc __flush_dcache_page
561 * throughout; but we cannot insert into address
562 * space until vma start or end is updated.
564 __vma_link_file(insert
);
568 vma_adjust_trans_huge(vma
, start
, end
, adjust_next
);
571 * When changing only vma->vm_end, we don't really need anon_vma
572 * lock. This is a fairly rare case by itself, but the anon_vma
573 * lock may be shared between many sibling processes. Skipping
574 * the lock for brk adjustments makes a difference sometimes.
576 if (vma
->anon_vma
&& (importer
|| start
!= vma
->vm_start
)) {
577 anon_vma
= vma
->anon_vma
;
578 anon_vma_lock(anon_vma
);
582 flush_dcache_mmap_lock(mapping
);
583 vma_prio_tree_remove(vma
, root
);
585 vma_prio_tree_remove(next
, root
);
588 vma
->vm_start
= start
;
590 vma
->vm_pgoff
= pgoff
;
592 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
593 next
->vm_pgoff
+= adjust_next
;
598 vma_prio_tree_insert(next
, root
);
599 vma_prio_tree_insert(vma
, root
);
600 flush_dcache_mmap_unlock(mapping
);
605 * vma_merge has merged next into vma, and needs
606 * us to remove next before dropping the locks.
608 __vma_unlink(mm
, next
, vma
);
610 __remove_shared_vm_struct(next
, file
, mapping
);
613 * split_vma has split insert from vma, and needs
614 * us to insert it before dropping the locks
615 * (it may either follow vma or precede it).
617 __insert_vm_struct(mm
, insert
);
621 anon_vma_unlock(anon_vma
);
623 mutex_unlock(&mapping
->i_mmap_mutex
);
634 uprobe_munmap(next
, next
->vm_start
, next
->vm_end
);
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) {
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 if (vma
->vm_flags
^ vm_flags
)
669 if (vma
->vm_file
!= file
)
671 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
676 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
677 struct anon_vma
*anon_vma2
,
678 struct vm_area_struct
*vma
)
681 * The list_is_singular() test is to avoid merging VMA cloned from
682 * parents. This can improve scalability caused by anon_vma lock.
684 if ((!anon_vma1
|| !anon_vma2
) && (!vma
||
685 list_is_singular(&vma
->anon_vma_chain
)))
687 return anon_vma1
== anon_vma2
;
691 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
692 * in front of (at a lower virtual address and file offset than) the vma.
694 * We cannot merge two vmas if they have differently assigned (non-NULL)
695 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
697 * We don't check here for the merged mmap wrapping around the end of pagecache
698 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
699 * wrap, nor mmaps which cover the final page at index -1UL.
702 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
703 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
705 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
706 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
707 if (vma
->vm_pgoff
== vm_pgoff
)
714 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
715 * beyond (at a higher virtual address and file offset than) the vma.
717 * We cannot merge two vmas if they have differently assigned (non-NULL)
718 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
721 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
722 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
724 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
725 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
, vma
)) {
727 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
728 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
735 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
736 * whether that can be merged with its predecessor or its successor.
737 * Or both (it neatly fills a hole).
739 * In most cases - when called for mmap, brk or mremap - [addr,end) is
740 * certain not to be mapped by the time vma_merge is called; but when
741 * called for mprotect, it is certain to be already mapped (either at
742 * an offset within prev, or at the start of next), and the flags of
743 * this area are about to be changed to vm_flags - and the no-change
744 * case has already been eliminated.
746 * The following mprotect cases have to be considered, where AAAA is
747 * the area passed down from mprotect_fixup, never extending beyond one
748 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
750 * AAAA AAAA AAAA AAAA
751 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
752 * cannot merge might become might become might become
753 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
754 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
755 * mremap move: PPPPNNNNNNNN 8
757 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
758 * might become case 1 below case 2 below case 3 below
760 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
761 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
763 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
764 struct vm_area_struct
*prev
, unsigned long addr
,
765 unsigned long end
, unsigned long vm_flags
,
766 struct anon_vma
*anon_vma
, struct file
*file
,
767 pgoff_t pgoff
, struct mempolicy
*policy
)
769 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
770 struct vm_area_struct
*area
, *next
;
774 * We later require that vma->vm_flags == vm_flags,
775 * so this tests vma->vm_flags & VM_SPECIAL, too.
777 if (vm_flags
& VM_SPECIAL
)
781 next
= prev
->vm_next
;
785 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
786 next
= next
->vm_next
;
789 * Can it merge with the predecessor?
791 if (prev
&& prev
->vm_end
== addr
&&
792 mpol_equal(vma_policy(prev
), policy
) &&
793 can_vma_merge_after(prev
, vm_flags
,
794 anon_vma
, file
, pgoff
)) {
796 * OK, it can. Can we now merge in the successor as well?
798 if (next
&& end
== next
->vm_start
&&
799 mpol_equal(policy
, vma_policy(next
)) &&
800 can_vma_merge_before(next
, vm_flags
,
801 anon_vma
, file
, pgoff
+pglen
) &&
802 is_mergeable_anon_vma(prev
->anon_vma
,
803 next
->anon_vma
, NULL
)) {
805 err
= vma_adjust(prev
, prev
->vm_start
,
806 next
->vm_end
, prev
->vm_pgoff
, NULL
);
807 } else /* cases 2, 5, 7 */
808 err
= vma_adjust(prev
, prev
->vm_start
,
809 end
, prev
->vm_pgoff
, NULL
);
812 khugepaged_enter_vma_merge(prev
);
817 * Can this new request be merged in front of next?
819 if (next
&& end
== next
->vm_start
&&
820 mpol_equal(policy
, vma_policy(next
)) &&
821 can_vma_merge_before(next
, vm_flags
,
822 anon_vma
, file
, pgoff
+pglen
)) {
823 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
824 err
= vma_adjust(prev
, prev
->vm_start
,
825 addr
, prev
->vm_pgoff
, NULL
);
826 else /* cases 3, 8 */
827 err
= vma_adjust(area
, addr
, next
->vm_end
,
828 next
->vm_pgoff
- pglen
, NULL
);
831 khugepaged_enter_vma_merge(area
);
839 * Rough compatbility check to quickly see if it's even worth looking
840 * at sharing an anon_vma.
842 * They need to have the same vm_file, and the flags can only differ
843 * in things that mprotect may change.
845 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
846 * we can merge the two vma's. For example, we refuse to merge a vma if
847 * there is a vm_ops->close() function, because that indicates that the
848 * driver is doing some kind of reference counting. But that doesn't
849 * really matter for the anon_vma sharing case.
851 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
853 return a
->vm_end
== b
->vm_start
&&
854 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
855 a
->vm_file
== b
->vm_file
&&
856 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_READ
|VM_WRITE
|VM_EXEC
)) &&
857 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
861 * Do some basic sanity checking to see if we can re-use the anon_vma
862 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
863 * the same as 'old', the other will be the new one that is trying
864 * to share the anon_vma.
866 * NOTE! This runs with mm_sem held for reading, so it is possible that
867 * the anon_vma of 'old' is concurrently in the process of being set up
868 * by another page fault trying to merge _that_. But that's ok: if it
869 * is being set up, that automatically means that it will be a singleton
870 * acceptable for merging, so we can do all of this optimistically. But
871 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
873 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
874 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
875 * is to return an anon_vma that is "complex" due to having gone through
878 * We also make sure that the two vma's are compatible (adjacent,
879 * and with the same memory policies). That's all stable, even with just
880 * a read lock on the mm_sem.
882 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
884 if (anon_vma_compatible(a
, b
)) {
885 struct anon_vma
*anon_vma
= ACCESS_ONCE(old
->anon_vma
);
887 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
894 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
895 * neighbouring vmas for a suitable anon_vma, before it goes off
896 * to allocate a new anon_vma. It checks because a repetitive
897 * sequence of mprotects and faults may otherwise lead to distinct
898 * anon_vmas being allocated, preventing vma merge in subsequent
901 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
903 struct anon_vma
*anon_vma
;
904 struct vm_area_struct
*near
;
910 anon_vma
= reusable_anon_vma(near
, vma
, near
);
918 anon_vma
= reusable_anon_vma(near
, near
, vma
);
923 * There's no absolute need to look only at touching neighbours:
924 * we could search further afield for "compatible" anon_vmas.
925 * But it would probably just be a waste of time searching,
926 * or lead to too many vmas hanging off the same anon_vma.
927 * We're trying to allow mprotect remerging later on,
928 * not trying to minimize memory used for anon_vmas.
933 #ifdef CONFIG_PROC_FS
934 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
935 struct file
*file
, long pages
)
937 const unsigned long stack_flags
938 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
940 mm
->total_vm
+= pages
;
943 mm
->shared_vm
+= pages
;
944 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
945 mm
->exec_vm
+= pages
;
946 } else if (flags
& stack_flags
)
947 mm
->stack_vm
+= pages
;
949 #endif /* CONFIG_PROC_FS */
952 * If a hint addr is less than mmap_min_addr change hint to be as
953 * low as possible but still greater than mmap_min_addr
955 static inline unsigned long round_hint_to_min(unsigned long hint
)
958 if (((void *)hint
!= NULL
) &&
959 (hint
< mmap_min_addr
))
960 return PAGE_ALIGN(mmap_min_addr
);
965 * The caller must hold down_write(¤t->mm->mmap_sem).
968 unsigned long do_mmap_pgoff(struct file
*file
, unsigned long addr
,
969 unsigned long len
, unsigned long prot
,
970 unsigned long flags
, unsigned long pgoff
)
972 struct mm_struct
* mm
= current
->mm
;
977 * Does the application expect PROT_READ to imply PROT_EXEC?
979 * (the exception is when the underlying filesystem is noexec
980 * mounted, in which case we dont add PROT_EXEC.)
982 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
983 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
989 if (!(flags
& MAP_FIXED
))
990 addr
= round_hint_to_min(addr
);
992 /* Careful about overflows.. */
993 len
= PAGE_ALIGN(len
);
997 /* offset overflow? */
998 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1001 /* Too many mappings? */
1002 if (mm
->map_count
> sysctl_max_map_count
)
1005 /* Obtain the address to map to. we verify (or select) it and ensure
1006 * that it represents a valid section of the address space.
1008 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1009 if (addr
& ~PAGE_MASK
)
1012 /* Do simple checking here so the lower-level routines won't have
1013 * to. we assume access permissions have been handled by the open
1014 * of the memory object, so we don't do any here.
1016 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
1017 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1019 if (flags
& MAP_LOCKED
)
1020 if (!can_do_mlock())
1023 /* mlock MCL_FUTURE? */
1024 if (vm_flags
& VM_LOCKED
) {
1025 unsigned long locked
, lock_limit
;
1026 locked
= len
>> PAGE_SHIFT
;
1027 locked
+= mm
->locked_vm
;
1028 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1029 lock_limit
>>= PAGE_SHIFT
;
1030 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1034 inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1037 switch (flags
& MAP_TYPE
) {
1039 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
1043 * Make sure we don't allow writing to an append-only
1046 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1050 * Make sure there are no mandatory locks on the file.
1052 if (locks_verify_locked(inode
))
1055 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1056 if (!(file
->f_mode
& FMODE_WRITE
))
1057 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1061 if (!(file
->f_mode
& FMODE_READ
))
1063 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1064 if (vm_flags
& VM_EXEC
)
1066 vm_flags
&= ~VM_MAYEXEC
;
1069 if (!file
->f_op
|| !file
->f_op
->mmap
)
1077 switch (flags
& MAP_TYPE
) {
1083 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1087 * Set pgoff according to addr for anon_vma.
1089 pgoff
= addr
>> PAGE_SHIFT
;
1096 return mmap_region(file
, addr
, len
, flags
, vm_flags
, pgoff
);
1099 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1100 unsigned long, prot
, unsigned long, flags
,
1101 unsigned long, fd
, unsigned long, pgoff
)
1103 struct file
*file
= NULL
;
1104 unsigned long retval
= -EBADF
;
1106 if (!(flags
& MAP_ANONYMOUS
)) {
1107 audit_mmap_fd(fd
, flags
);
1108 if (unlikely(flags
& MAP_HUGETLB
))
1113 } else if (flags
& MAP_HUGETLB
) {
1114 struct user_struct
*user
= NULL
;
1116 * VM_NORESERVE is used because the reservations will be
1117 * taken when vm_ops->mmap() is called
1118 * A dummy user value is used because we are not locking
1119 * memory so no accounting is necessary
1121 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, addr
, len
,
1122 VM_NORESERVE
, &user
,
1123 HUGETLB_ANONHUGE_INODE
);
1125 return PTR_ERR(file
);
1128 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1130 retval
= vm_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1137 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1138 struct mmap_arg_struct
{
1142 unsigned long flags
;
1144 unsigned long offset
;
1147 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1149 struct mmap_arg_struct a
;
1151 if (copy_from_user(&a
, arg
, sizeof(a
)))
1153 if (a
.offset
& ~PAGE_MASK
)
1156 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1157 a
.offset
>> PAGE_SHIFT
);
1159 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1162 * Some shared mappigns will want the pages marked read-only
1163 * to track write events. If so, we'll downgrade vm_page_prot
1164 * to the private version (using protection_map[] without the
1167 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1169 vm_flags_t vm_flags
= vma
->vm_flags
;
1171 /* If it was private or non-writable, the write bit is already clear */
1172 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1175 /* The backer wishes to know when pages are first written to? */
1176 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1179 /* The open routine did something to the protections already? */
1180 if (pgprot_val(vma
->vm_page_prot
) !=
1181 pgprot_val(vm_get_page_prot(vm_flags
)))
1184 /* Specialty mapping? */
1185 if (vm_flags
& VM_PFNMAP
)
1188 /* Can the mapping track the dirty pages? */
1189 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1190 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1194 * We account for memory if it's a private writeable mapping,
1195 * not hugepages and VM_NORESERVE wasn't set.
1197 static inline int accountable_mapping(struct file
*file
, vm_flags_t vm_flags
)
1200 * hugetlb has its own accounting separate from the core VM
1201 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1203 if (file
&& is_file_hugepages(file
))
1206 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1209 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1210 unsigned long len
, unsigned long flags
,
1211 vm_flags_t vm_flags
, unsigned long pgoff
)
1213 struct mm_struct
*mm
= current
->mm
;
1214 struct vm_area_struct
*vma
, *prev
;
1215 int correct_wcount
= 0;
1217 struct rb_node
**rb_link
, *rb_parent
;
1218 unsigned long charged
= 0;
1219 struct inode
*inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1221 /* Clear old maps */
1224 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1225 if (vma
&& vma
->vm_start
< addr
+ len
) {
1226 if (do_munmap(mm
, addr
, len
))
1231 /* Check against address space limit. */
1232 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1236 * Set 'VM_NORESERVE' if we should not account for the
1237 * memory use of this mapping.
1239 if ((flags
& MAP_NORESERVE
)) {
1240 /* We honor MAP_NORESERVE if allowed to overcommit */
1241 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1242 vm_flags
|= VM_NORESERVE
;
1244 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1245 if (file
&& is_file_hugepages(file
))
1246 vm_flags
|= VM_NORESERVE
;
1250 * Private writable mapping: check memory availability
1252 if (accountable_mapping(file
, vm_flags
)) {
1253 charged
= len
>> PAGE_SHIFT
;
1254 if (security_vm_enough_memory_mm(mm
, charged
))
1256 vm_flags
|= VM_ACCOUNT
;
1260 * Can we just expand an old mapping?
1262 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
, NULL
, file
, pgoff
, NULL
);
1267 * Determine the object being mapped and call the appropriate
1268 * specific mapper. the address has already been validated, but
1269 * not unmapped, but the maps are removed from the list.
1271 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1278 vma
->vm_start
= addr
;
1279 vma
->vm_end
= addr
+ len
;
1280 vma
->vm_flags
= vm_flags
;
1281 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1282 vma
->vm_pgoff
= pgoff
;
1283 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1285 error
= -EINVAL
; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */
1288 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1290 if (vm_flags
& VM_DENYWRITE
) {
1291 error
= deny_write_access(file
);
1296 vma
->vm_file
= get_file(file
);
1297 error
= file
->f_op
->mmap(file
, vma
);
1299 goto unmap_and_free_vma
;
1301 /* Can addr have changed??
1303 * Answer: Yes, several device drivers can do it in their
1304 * f_op->mmap method. -DaveM
1306 addr
= vma
->vm_start
;
1307 pgoff
= vma
->vm_pgoff
;
1308 vm_flags
= vma
->vm_flags
;
1309 } else if (vm_flags
& VM_SHARED
) {
1310 if (unlikely(vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
)))
1312 error
= shmem_zero_setup(vma
);
1317 if (vma_wants_writenotify(vma
)) {
1318 pgprot_t pprot
= vma
->vm_page_prot
;
1320 /* Can vma->vm_page_prot have changed??
1322 * Answer: Yes, drivers may have changed it in their
1323 * f_op->mmap method.
1325 * Ensures that vmas marked as uncached stay that way.
1327 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
& ~VM_SHARED
);
1328 if (pgprot_val(pprot
) == pgprot_val(pgprot_noncached(pprot
)))
1329 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
1332 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1333 file
= vma
->vm_file
;
1335 /* Once vma denies write, undo our temporary denial count */
1337 atomic_inc(&inode
->i_writecount
);
1339 perf_event_mmap(vma
);
1341 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1342 if (vm_flags
& VM_LOCKED
) {
1343 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
1344 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1345 } else if ((flags
& MAP_POPULATE
) && !(flags
& MAP_NONBLOCK
))
1346 make_pages_present(addr
, addr
+ len
);
1355 atomic_inc(&inode
->i_writecount
);
1356 vma
->vm_file
= NULL
;
1359 /* Undo any partial mapping done by a device driver. */
1360 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1363 kmem_cache_free(vm_area_cachep
, vma
);
1366 vm_unacct_memory(charged
);
1370 /* Get an address range which is currently unmapped.
1371 * For shmat() with addr=0.
1373 * Ugly calling convention alert:
1374 * Return value with the low bits set means error value,
1376 * if (ret & ~PAGE_MASK)
1379 * This function "knows" that -ENOMEM has the bits set.
1381 #ifndef HAVE_ARCH_UNMAPPED_AREA
1383 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1384 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1386 struct mm_struct
*mm
= current
->mm
;
1387 struct vm_area_struct
*vma
;
1388 unsigned long start_addr
;
1390 if (len
> TASK_SIZE
)
1393 if (flags
& MAP_FIXED
)
1397 addr
= PAGE_ALIGN(addr
);
1398 vma
= find_vma(mm
, addr
);
1399 if (TASK_SIZE
- len
>= addr
&&
1400 (!vma
|| addr
+ len
<= vma
->vm_start
))
1403 if (len
> mm
->cached_hole_size
) {
1404 start_addr
= addr
= mm
->free_area_cache
;
1406 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1407 mm
->cached_hole_size
= 0;
1411 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1412 /* At this point: (!vma || addr < vma->vm_end). */
1413 if (TASK_SIZE
- len
< addr
) {
1415 * Start a new search - just in case we missed
1418 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1419 addr
= TASK_UNMAPPED_BASE
;
1421 mm
->cached_hole_size
= 0;
1426 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1428 * Remember the place where we stopped the search:
1430 mm
->free_area_cache
= addr
+ len
;
1433 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1434 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1440 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1443 * Is this a new hole at the lowest possible address?
1445 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
)
1446 mm
->free_area_cache
= addr
;
1450 * This mmap-allocator allocates new areas top-down from below the
1451 * stack's low limit (the base):
1453 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1455 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1456 const unsigned long len
, const unsigned long pgoff
,
1457 const unsigned long flags
)
1459 struct vm_area_struct
*vma
;
1460 struct mm_struct
*mm
= current
->mm
;
1461 unsigned long addr
= addr0
, start_addr
;
1463 /* requested length too big for entire address space */
1464 if (len
> TASK_SIZE
)
1467 if (flags
& MAP_FIXED
)
1470 /* requesting a specific address */
1472 addr
= PAGE_ALIGN(addr
);
1473 vma
= find_vma(mm
, addr
);
1474 if (TASK_SIZE
- len
>= addr
&&
1475 (!vma
|| addr
+ len
<= vma
->vm_start
))
1479 /* check if free_area_cache is useful for us */
1480 if (len
<= mm
->cached_hole_size
) {
1481 mm
->cached_hole_size
= 0;
1482 mm
->free_area_cache
= mm
->mmap_base
;
1486 /* either no address requested or can't fit in requested address hole */
1487 start_addr
= addr
= mm
->free_area_cache
;
1495 * Lookup failure means no vma is above this address,
1496 * else if new region fits below vma->vm_start,
1497 * return with success:
1499 vma
= find_vma(mm
, addr
);
1500 if (!vma
|| addr
+len
<= vma
->vm_start
)
1501 /* remember the address as a hint for next time */
1502 return (mm
->free_area_cache
= addr
);
1504 /* remember the largest hole we saw so far */
1505 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1506 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1508 /* try just below the current vma->vm_start */
1509 addr
= vma
->vm_start
-len
;
1510 } while (len
< vma
->vm_start
);
1514 * if hint left us with no space for the requested
1515 * mapping then try again:
1517 * Note: this is different with the case of bottomup
1518 * which does the fully line-search, but we use find_vma
1519 * here that causes some holes skipped.
1521 if (start_addr
!= mm
->mmap_base
) {
1522 mm
->free_area_cache
= mm
->mmap_base
;
1523 mm
->cached_hole_size
= 0;
1528 * A failed mmap() very likely causes application failure,
1529 * so fall back to the bottom-up function here. This scenario
1530 * can happen with large stack limits and large mmap()
1533 mm
->cached_hole_size
= ~0UL;
1534 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1535 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1537 * Restore the topdown base:
1539 mm
->free_area_cache
= mm
->mmap_base
;
1540 mm
->cached_hole_size
= ~0UL;
1546 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1549 * Is this a new hole at the highest possible address?
1551 if (addr
> mm
->free_area_cache
)
1552 mm
->free_area_cache
= addr
;
1554 /* dont allow allocations above current base */
1555 if (mm
->free_area_cache
> mm
->mmap_base
)
1556 mm
->free_area_cache
= mm
->mmap_base
;
1560 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1561 unsigned long pgoff
, unsigned long flags
)
1563 unsigned long (*get_area
)(struct file
*, unsigned long,
1564 unsigned long, unsigned long, unsigned long);
1566 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
1570 /* Careful about overflows.. */
1571 if (len
> TASK_SIZE
)
1574 get_area
= current
->mm
->get_unmapped_area
;
1575 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1576 get_area
= file
->f_op
->get_unmapped_area
;
1577 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1578 if (IS_ERR_VALUE(addr
))
1581 if (addr
> TASK_SIZE
- len
)
1583 if (addr
& ~PAGE_MASK
)
1586 addr
= arch_rebalance_pgtables(addr
, len
);
1587 error
= security_mmap_addr(addr
);
1588 return error
? error
: addr
;
1591 EXPORT_SYMBOL(get_unmapped_area
);
1593 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1594 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
1596 struct vm_area_struct
*vma
= NULL
;
1598 if (WARN_ON_ONCE(!mm
)) /* Remove this in linux-3.6 */
1601 /* Check the cache first. */
1602 /* (Cache hit rate is typically around 35%.) */
1603 vma
= mm
->mmap_cache
;
1604 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1605 struct rb_node
*rb_node
;
1607 rb_node
= mm
->mm_rb
.rb_node
;
1611 struct vm_area_struct
*vma_tmp
;
1613 vma_tmp
= rb_entry(rb_node
,
1614 struct vm_area_struct
, vm_rb
);
1616 if (vma_tmp
->vm_end
> addr
) {
1618 if (vma_tmp
->vm_start
<= addr
)
1620 rb_node
= rb_node
->rb_left
;
1622 rb_node
= rb_node
->rb_right
;
1625 mm
->mmap_cache
= vma
;
1630 EXPORT_SYMBOL(find_vma
);
1633 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
1635 struct vm_area_struct
*
1636 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1637 struct vm_area_struct
**pprev
)
1639 struct vm_area_struct
*vma
;
1641 vma
= find_vma(mm
, addr
);
1643 *pprev
= vma
->vm_prev
;
1645 struct rb_node
*rb_node
= mm
->mm_rb
.rb_node
;
1648 *pprev
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1649 rb_node
= rb_node
->rb_right
;
1656 * Verify that the stack growth is acceptable and
1657 * update accounting. This is shared with both the
1658 * grow-up and grow-down cases.
1660 static int acct_stack_growth(struct vm_area_struct
*vma
, unsigned long size
, unsigned long grow
)
1662 struct mm_struct
*mm
= vma
->vm_mm
;
1663 struct rlimit
*rlim
= current
->signal
->rlim
;
1664 unsigned long new_start
;
1666 /* address space limit tests */
1667 if (!may_expand_vm(mm
, grow
))
1670 /* Stack limit test */
1671 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
))
1674 /* mlock limit tests */
1675 if (vma
->vm_flags
& VM_LOCKED
) {
1676 unsigned long locked
;
1677 unsigned long limit
;
1678 locked
= mm
->locked_vm
+ grow
;
1679 limit
= ACCESS_ONCE(rlim
[RLIMIT_MEMLOCK
].rlim_cur
);
1680 limit
>>= PAGE_SHIFT
;
1681 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1685 /* Check to ensure the stack will not grow into a hugetlb-only region */
1686 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
1688 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
1692 * Overcommit.. This must be the final test, as it will
1693 * update security statistics.
1695 if (security_vm_enough_memory_mm(mm
, grow
))
1698 /* Ok, everything looks good - let it rip */
1699 if (vma
->vm_flags
& VM_LOCKED
)
1700 mm
->locked_vm
+= grow
;
1701 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1705 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1707 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1708 * vma is the last one with address > vma->vm_end. Have to extend vma.
1710 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1714 if (!(vma
->vm_flags
& VM_GROWSUP
))
1718 * We must make sure the anon_vma is allocated
1719 * so that the anon_vma locking is not a noop.
1721 if (unlikely(anon_vma_prepare(vma
)))
1723 vma_lock_anon_vma(vma
);
1726 * vma->vm_start/vm_end cannot change under us because the caller
1727 * is required to hold the mmap_sem in read mode. We need the
1728 * anon_vma lock to serialize against concurrent expand_stacks.
1729 * Also guard against wrapping around to address 0.
1731 if (address
< PAGE_ALIGN(address
+4))
1732 address
= PAGE_ALIGN(address
+4);
1734 vma_unlock_anon_vma(vma
);
1739 /* Somebody else might have raced and expanded it already */
1740 if (address
> vma
->vm_end
) {
1741 unsigned long size
, grow
;
1743 size
= address
- vma
->vm_start
;
1744 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1747 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
1748 error
= acct_stack_growth(vma
, size
, grow
);
1750 vma
->vm_end
= address
;
1751 perf_event_mmap(vma
);
1755 vma_unlock_anon_vma(vma
);
1756 khugepaged_enter_vma_merge(vma
);
1759 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1762 * vma is the first one with address < vma->vm_start. Have to extend vma.
1764 int expand_downwards(struct vm_area_struct
*vma
,
1765 unsigned long address
)
1770 * We must make sure the anon_vma is allocated
1771 * so that the anon_vma locking is not a noop.
1773 if (unlikely(anon_vma_prepare(vma
)))
1776 address
&= PAGE_MASK
;
1777 error
= security_mmap_addr(address
);
1781 vma_lock_anon_vma(vma
);
1784 * vma->vm_start/vm_end cannot change under us because the caller
1785 * is required to hold the mmap_sem in read mode. We need the
1786 * anon_vma lock to serialize against concurrent expand_stacks.
1789 /* Somebody else might have raced and expanded it already */
1790 if (address
< vma
->vm_start
) {
1791 unsigned long size
, grow
;
1793 size
= vma
->vm_end
- address
;
1794 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1797 if (grow
<= vma
->vm_pgoff
) {
1798 error
= acct_stack_growth(vma
, size
, grow
);
1800 vma
->vm_start
= address
;
1801 vma
->vm_pgoff
-= grow
;
1802 perf_event_mmap(vma
);
1806 vma_unlock_anon_vma(vma
);
1807 khugepaged_enter_vma_merge(vma
);
1811 #ifdef CONFIG_STACK_GROWSUP
1812 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1814 return expand_upwards(vma
, address
);
1817 struct vm_area_struct
*
1818 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1820 struct vm_area_struct
*vma
, *prev
;
1823 vma
= find_vma_prev(mm
, addr
, &prev
);
1824 if (vma
&& (vma
->vm_start
<= addr
))
1826 if (!prev
|| expand_stack(prev
, addr
))
1828 if (prev
->vm_flags
& VM_LOCKED
) {
1829 mlock_vma_pages_range(prev
, addr
, prev
->vm_end
);
1834 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1836 return expand_downwards(vma
, address
);
1839 struct vm_area_struct
*
1840 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1842 struct vm_area_struct
* vma
;
1843 unsigned long start
;
1846 vma
= find_vma(mm
,addr
);
1849 if (vma
->vm_start
<= addr
)
1851 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1853 start
= vma
->vm_start
;
1854 if (expand_stack(vma
, addr
))
1856 if (vma
->vm_flags
& VM_LOCKED
) {
1857 mlock_vma_pages_range(vma
, addr
, start
);
1864 * Ok - we have the memory areas we should free on the vma list,
1865 * so release them, and do the vma updates.
1867 * Called with the mm semaphore held.
1869 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1871 unsigned long nr_accounted
= 0;
1873 /* Update high watermark before we lower total_vm */
1874 update_hiwater_vm(mm
);
1876 long nrpages
= vma_pages(vma
);
1878 if (vma
->vm_flags
& VM_ACCOUNT
)
1879 nr_accounted
+= nrpages
;
1880 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1881 vma
= remove_vma(vma
);
1883 vm_unacct_memory(nr_accounted
);
1888 * Get rid of page table information in the indicated region.
1890 * Called with the mm semaphore held.
1892 static void unmap_region(struct mm_struct
*mm
,
1893 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1894 unsigned long start
, unsigned long end
)
1896 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1897 struct mmu_gather tlb
;
1900 tlb_gather_mmu(&tlb
, mm
, 0);
1901 update_hiwater_rss(mm
);
1902 unmap_vmas(&tlb
, vma
, start
, end
);
1903 free_pgtables(&tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1904 next
? next
->vm_start
: 0);
1905 tlb_finish_mmu(&tlb
, start
, end
);
1909 * Create a list of vma's touched by the unmap, removing them from the mm's
1910 * vma list as we go..
1913 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1914 struct vm_area_struct
*prev
, unsigned long end
)
1916 struct vm_area_struct
**insertion_point
;
1917 struct vm_area_struct
*tail_vma
= NULL
;
1920 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1921 vma
->vm_prev
= NULL
;
1923 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1927 } while (vma
&& vma
->vm_start
< end
);
1928 *insertion_point
= vma
;
1930 vma
->vm_prev
= prev
;
1931 tail_vma
->vm_next
= NULL
;
1932 if (mm
->unmap_area
== arch_unmap_area
)
1933 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1935 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1936 mm
->unmap_area(mm
, addr
);
1937 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1941 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1942 * munmap path where it doesn't make sense to fail.
1944 static int __split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1945 unsigned long addr
, int new_below
)
1947 struct mempolicy
*pol
;
1948 struct vm_area_struct
*new;
1951 if (is_vm_hugetlb_page(vma
) && (addr
&
1952 ~(huge_page_mask(hstate_vma(vma
)))))
1955 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1959 /* most fields are the same, copy all, and then fixup */
1962 INIT_LIST_HEAD(&new->anon_vma_chain
);
1967 new->vm_start
= addr
;
1968 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
1971 pol
= mpol_dup(vma_policy(vma
));
1976 vma_set_policy(new, pol
);
1978 if (anon_vma_clone(new, vma
))
1982 get_file(new->vm_file
);
1984 if (new->vm_ops
&& new->vm_ops
->open
)
1985 new->vm_ops
->open(new);
1988 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
1989 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
1991 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
1997 /* Clean everything up if vma_adjust failed. */
1998 if (new->vm_ops
&& new->vm_ops
->close
)
1999 new->vm_ops
->close(new);
2002 unlink_anon_vmas(new);
2006 kmem_cache_free(vm_area_cachep
, new);
2012 * Split a vma into two pieces at address 'addr', a new vma is allocated
2013 * either for the first part or the tail.
2015 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2016 unsigned long addr
, int new_below
)
2018 if (mm
->map_count
>= sysctl_max_map_count
)
2021 return __split_vma(mm
, vma
, addr
, new_below
);
2024 /* Munmap is split into 2 main parts -- this part which finds
2025 * what needs doing, and the areas themselves, which do the
2026 * work. This now handles partial unmappings.
2027 * Jeremy Fitzhardinge <jeremy@goop.org>
2029 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
2032 struct vm_area_struct
*vma
, *prev
, *last
;
2034 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2037 if ((len
= PAGE_ALIGN(len
)) == 0)
2040 /* Find the first overlapping VMA */
2041 vma
= find_vma(mm
, start
);
2044 prev
= vma
->vm_prev
;
2045 /* we have start < vma->vm_end */
2047 /* if it doesn't overlap, we have nothing.. */
2049 if (vma
->vm_start
>= end
)
2053 * If we need to split any vma, do it now to save pain later.
2055 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2056 * unmapped vm_area_struct will remain in use: so lower split_vma
2057 * places tmp vma above, and higher split_vma places tmp vma below.
2059 if (start
> vma
->vm_start
) {
2063 * Make sure that map_count on return from munmap() will
2064 * not exceed its limit; but let map_count go just above
2065 * its limit temporarily, to help free resources as expected.
2067 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2070 error
= __split_vma(mm
, vma
, start
, 0);
2076 /* Does it split the last one? */
2077 last
= find_vma(mm
, end
);
2078 if (last
&& end
> last
->vm_start
) {
2079 int error
= __split_vma(mm
, last
, end
, 1);
2083 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2086 * unlock any mlock()ed ranges before detaching vmas
2088 if (mm
->locked_vm
) {
2089 struct vm_area_struct
*tmp
= vma
;
2090 while (tmp
&& tmp
->vm_start
< end
) {
2091 if (tmp
->vm_flags
& VM_LOCKED
) {
2092 mm
->locked_vm
-= vma_pages(tmp
);
2093 munlock_vma_pages_all(tmp
);
2100 * Remove the vma's, and unmap the actual pages
2102 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2103 unmap_region(mm
, vma
, prev
, start
, end
);
2105 /* Fix up all other VM information */
2106 remove_vma_list(mm
, vma
);
2111 int vm_munmap(unsigned long start
, size_t len
)
2114 struct mm_struct
*mm
= current
->mm
;
2116 down_write(&mm
->mmap_sem
);
2117 ret
= do_munmap(mm
, start
, len
);
2118 up_write(&mm
->mmap_sem
);
2121 EXPORT_SYMBOL(vm_munmap
);
2123 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2125 profile_munmap(addr
);
2126 return vm_munmap(addr
, len
);
2129 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
2131 #ifdef CONFIG_DEBUG_VM
2132 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
2134 up_read(&mm
->mmap_sem
);
2140 * this is really a simplified "do_mmap". it only handles
2141 * anonymous maps. eventually we may be able to do some
2142 * brk-specific accounting here.
2144 static unsigned long do_brk(unsigned long addr
, unsigned long len
)
2146 struct mm_struct
* mm
= current
->mm
;
2147 struct vm_area_struct
* vma
, * prev
;
2148 unsigned long flags
;
2149 struct rb_node
** rb_link
, * rb_parent
;
2150 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2153 len
= PAGE_ALIGN(len
);
2157 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2159 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2160 if (error
& ~PAGE_MASK
)
2166 if (mm
->def_flags
& VM_LOCKED
) {
2167 unsigned long locked
, lock_limit
;
2168 locked
= len
>> PAGE_SHIFT
;
2169 locked
+= mm
->locked_vm
;
2170 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
2171 lock_limit
>>= PAGE_SHIFT
;
2172 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
2177 * mm->mmap_sem is required to protect against another thread
2178 * changing the mappings in case we sleep.
2180 verify_mm_writelocked(mm
);
2183 * Clear old maps. this also does some error checking for us
2186 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2187 if (vma
&& vma
->vm_start
< addr
+ len
) {
2188 if (do_munmap(mm
, addr
, len
))
2193 /* Check against address space limits *after* clearing old maps... */
2194 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2197 if (mm
->map_count
> sysctl_max_map_count
)
2200 if (security_vm_enough_memory_mm(mm
, len
>> PAGE_SHIFT
))
2203 /* Can we just expand an old private anonymous mapping? */
2204 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2205 NULL
, NULL
, pgoff
, NULL
);
2210 * create a vma struct for an anonymous mapping
2212 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2214 vm_unacct_memory(len
>> PAGE_SHIFT
);
2218 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2220 vma
->vm_start
= addr
;
2221 vma
->vm_end
= addr
+ len
;
2222 vma
->vm_pgoff
= pgoff
;
2223 vma
->vm_flags
= flags
;
2224 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2225 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2227 perf_event_mmap(vma
);
2228 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2229 if (flags
& VM_LOCKED
) {
2230 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
2231 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2236 unsigned long vm_brk(unsigned long addr
, unsigned long len
)
2238 struct mm_struct
*mm
= current
->mm
;
2241 down_write(&mm
->mmap_sem
);
2242 ret
= do_brk(addr
, len
);
2243 up_write(&mm
->mmap_sem
);
2246 EXPORT_SYMBOL(vm_brk
);
2248 /* Release all mmaps. */
2249 void exit_mmap(struct mm_struct
*mm
)
2251 struct mmu_gather tlb
;
2252 struct vm_area_struct
*vma
;
2253 unsigned long nr_accounted
= 0;
2255 /* mm's last user has gone, and its about to be pulled down */
2256 mmu_notifier_release(mm
);
2258 if (mm
->locked_vm
) {
2261 if (vma
->vm_flags
& VM_LOCKED
)
2262 munlock_vma_pages_all(vma
);
2270 if (!vma
) /* Can happen if dup_mmap() received an OOM */
2275 tlb_gather_mmu(&tlb
, mm
, 1);
2276 /* update_hiwater_rss(mm) here? but nobody should be looking */
2277 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2278 unmap_vmas(&tlb
, vma
, 0, -1);
2280 free_pgtables(&tlb
, vma
, FIRST_USER_ADDRESS
, 0);
2281 tlb_finish_mmu(&tlb
, 0, -1);
2284 * Walk the list again, actually closing and freeing it,
2285 * with preemption enabled, without holding any MM locks.
2288 if (vma
->vm_flags
& VM_ACCOUNT
)
2289 nr_accounted
+= vma_pages(vma
);
2290 vma
= remove_vma(vma
);
2292 vm_unacct_memory(nr_accounted
);
2294 WARN_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2297 /* Insert vm structure into process list sorted by address
2298 * and into the inode's i_mmap tree. If vm_file is non-NULL
2299 * then i_mmap_mutex is taken here.
2301 int insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
2303 struct vm_area_struct
* __vma
, * prev
;
2304 struct rb_node
** rb_link
, * rb_parent
;
2307 * The vm_pgoff of a purely anonymous vma should be irrelevant
2308 * until its first write fault, when page's anon_vma and index
2309 * are set. But now set the vm_pgoff it will almost certainly
2310 * end up with (unless mremap moves it elsewhere before that
2311 * first wfault), so /proc/pid/maps tells a consistent story.
2313 * By setting it to reflect the virtual start address of the
2314 * vma, merges and splits can happen in a seamless way, just
2315 * using the existing file pgoff checks and manipulations.
2316 * Similarly in do_mmap_pgoff and in do_brk.
2318 if (!vma
->vm_file
) {
2319 BUG_ON(vma
->anon_vma
);
2320 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2322 __vma
= find_vma_prepare(mm
,vma
->vm_start
,&prev
,&rb_link
,&rb_parent
);
2323 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
2325 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2326 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2329 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2334 * Copy the vma structure to a new location in the same mm,
2335 * prior to moving page table entries, to effect an mremap move.
2337 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2338 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
2340 struct vm_area_struct
*vma
= *vmap
;
2341 unsigned long vma_start
= vma
->vm_start
;
2342 struct mm_struct
*mm
= vma
->vm_mm
;
2343 struct vm_area_struct
*new_vma
, *prev
;
2344 struct rb_node
**rb_link
, *rb_parent
;
2345 struct mempolicy
*pol
;
2346 bool faulted_in_anon_vma
= true;
2349 * If anonymous vma has not yet been faulted, update new pgoff
2350 * to match new location, to increase its chance of merging.
2352 if (unlikely(!vma
->vm_file
&& !vma
->anon_vma
)) {
2353 pgoff
= addr
>> PAGE_SHIFT
;
2354 faulted_in_anon_vma
= false;
2357 find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2358 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2359 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2362 * Source vma may have been merged into new_vma
2364 if (unlikely(vma_start
>= new_vma
->vm_start
&&
2365 vma_start
< new_vma
->vm_end
)) {
2367 * The only way we can get a vma_merge with
2368 * self during an mremap is if the vma hasn't
2369 * been faulted in yet and we were allowed to
2370 * reset the dst vma->vm_pgoff to the
2371 * destination address of the mremap to allow
2372 * the merge to happen. mremap must change the
2373 * vm_pgoff linearity between src and dst vmas
2374 * (in turn preventing a vma_merge) to be
2375 * safe. It is only safe to keep the vm_pgoff
2376 * linear if there are no pages mapped yet.
2378 VM_BUG_ON(faulted_in_anon_vma
);
2381 anon_vma_moveto_tail(new_vma
);
2383 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2386 pol
= mpol_dup(vma_policy(vma
));
2389 INIT_LIST_HEAD(&new_vma
->anon_vma_chain
);
2390 if (anon_vma_clone(new_vma
, vma
))
2391 goto out_free_mempol
;
2392 vma_set_policy(new_vma
, pol
);
2393 new_vma
->vm_start
= addr
;
2394 new_vma
->vm_end
= addr
+ len
;
2395 new_vma
->vm_pgoff
= pgoff
;
2396 if (new_vma
->vm_file
)
2397 get_file(new_vma
->vm_file
);
2398 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2399 new_vma
->vm_ops
->open(new_vma
);
2400 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2408 kmem_cache_free(vm_area_cachep
, new_vma
);
2413 * Return true if the calling process may expand its vm space by the passed
2416 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2418 unsigned long cur
= mm
->total_vm
; /* pages */
2421 lim
= rlimit(RLIMIT_AS
) >> PAGE_SHIFT
;
2423 if (cur
+ npages
> lim
)
2429 static int special_mapping_fault(struct vm_area_struct
*vma
,
2430 struct vm_fault
*vmf
)
2433 struct page
**pages
;
2436 * special mappings have no vm_file, and in that case, the mm
2437 * uses vm_pgoff internally. So we have to subtract it from here.
2438 * We are allowed to do this because we are the mm; do not copy
2439 * this code into drivers!
2441 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2443 for (pages
= vma
->vm_private_data
; pgoff
&& *pages
; ++pages
)
2447 struct page
*page
= *pages
;
2453 return VM_FAULT_SIGBUS
;
2457 * Having a close hook prevents vma merging regardless of flags.
2459 static void special_mapping_close(struct vm_area_struct
*vma
)
2463 static const struct vm_operations_struct special_mapping_vmops
= {
2464 .close
= special_mapping_close
,
2465 .fault
= special_mapping_fault
,
2469 * Called with mm->mmap_sem held for writing.
2470 * Insert a new vma covering the given region, with the given flags.
2471 * Its pages are supplied by the given array of struct page *.
2472 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2473 * The region past the last page supplied will always produce SIGBUS.
2474 * The array pointer and the pages it points to are assumed to stay alive
2475 * for as long as this mapping might exist.
2477 int install_special_mapping(struct mm_struct
*mm
,
2478 unsigned long addr
, unsigned long len
,
2479 unsigned long vm_flags
, struct page
**pages
)
2482 struct vm_area_struct
*vma
;
2484 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2485 if (unlikely(vma
== NULL
))
2488 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2490 vma
->vm_start
= addr
;
2491 vma
->vm_end
= addr
+ len
;
2493 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
;
2494 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
2496 vma
->vm_ops
= &special_mapping_vmops
;
2497 vma
->vm_private_data
= pages
;
2499 ret
= insert_vm_struct(mm
, vma
);
2503 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2505 perf_event_mmap(vma
);
2510 kmem_cache_free(vm_area_cachep
, vma
);
2514 static DEFINE_MUTEX(mm_all_locks_mutex
);
2516 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
2518 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->head
.next
)) {
2520 * The LSB of head.next can't change from under us
2521 * because we hold the mm_all_locks_mutex.
2523 mutex_lock_nest_lock(&anon_vma
->root
->mutex
, &mm
->mmap_sem
);
2525 * We can safely modify head.next after taking the
2526 * anon_vma->root->mutex. If some other vma in this mm shares
2527 * the same anon_vma we won't take it again.
2529 * No need of atomic instructions here, head.next
2530 * can't change from under us thanks to the
2531 * anon_vma->root->mutex.
2533 if (__test_and_set_bit(0, (unsigned long *)
2534 &anon_vma
->root
->head
.next
))
2539 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
2541 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2543 * AS_MM_ALL_LOCKS can't change from under us because
2544 * we hold the mm_all_locks_mutex.
2546 * Operations on ->flags have to be atomic because
2547 * even if AS_MM_ALL_LOCKS is stable thanks to the
2548 * mm_all_locks_mutex, there may be other cpus
2549 * changing other bitflags in parallel to us.
2551 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
2553 mutex_lock_nest_lock(&mapping
->i_mmap_mutex
, &mm
->mmap_sem
);
2558 * This operation locks against the VM for all pte/vma/mm related
2559 * operations that could ever happen on a certain mm. This includes
2560 * vmtruncate, try_to_unmap, and all page faults.
2562 * The caller must take the mmap_sem in write mode before calling
2563 * mm_take_all_locks(). The caller isn't allowed to release the
2564 * mmap_sem until mm_drop_all_locks() returns.
2566 * mmap_sem in write mode is required in order to block all operations
2567 * that could modify pagetables and free pages without need of
2568 * altering the vma layout (for example populate_range() with
2569 * nonlinear vmas). It's also needed in write mode to avoid new
2570 * anon_vmas to be associated with existing vmas.
2572 * A single task can't take more than one mm_take_all_locks() in a row
2573 * or it would deadlock.
2575 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2576 * mapping->flags avoid to take the same lock twice, if more than one
2577 * vma in this mm is backed by the same anon_vma or address_space.
2579 * We can take all the locks in random order because the VM code
2580 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
2581 * takes more than one of them in a row. Secondly we're protected
2582 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2584 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2585 * that may have to take thousand of locks.
2587 * mm_take_all_locks() can fail if it's interrupted by signals.
2589 int mm_take_all_locks(struct mm_struct
*mm
)
2591 struct vm_area_struct
*vma
;
2592 struct anon_vma_chain
*avc
;
2594 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2596 mutex_lock(&mm_all_locks_mutex
);
2598 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2599 if (signal_pending(current
))
2601 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2602 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
2605 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2606 if (signal_pending(current
))
2609 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2610 vm_lock_anon_vma(mm
, avc
->anon_vma
);
2616 mm_drop_all_locks(mm
);
2620 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
2622 if (test_bit(0, (unsigned long *) &anon_vma
->root
->head
.next
)) {
2624 * The LSB of head.next can't change to 0 from under
2625 * us because we hold the mm_all_locks_mutex.
2627 * We must however clear the bitflag before unlocking
2628 * the vma so the users using the anon_vma->head will
2629 * never see our bitflag.
2631 * No need of atomic instructions here, head.next
2632 * can't change from under us until we release the
2633 * anon_vma->root->mutex.
2635 if (!__test_and_clear_bit(0, (unsigned long *)
2636 &anon_vma
->root
->head
.next
))
2638 anon_vma_unlock(anon_vma
);
2642 static void vm_unlock_mapping(struct address_space
*mapping
)
2644 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2646 * AS_MM_ALL_LOCKS can't change to 0 from under us
2647 * because we hold the mm_all_locks_mutex.
2649 mutex_unlock(&mapping
->i_mmap_mutex
);
2650 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
2657 * The mmap_sem cannot be released by the caller until
2658 * mm_drop_all_locks() returns.
2660 void mm_drop_all_locks(struct mm_struct
*mm
)
2662 struct vm_area_struct
*vma
;
2663 struct anon_vma_chain
*avc
;
2665 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2666 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
2668 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2670 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2671 vm_unlock_anon_vma(avc
->anon_vma
);
2672 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2673 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
2676 mutex_unlock(&mm_all_locks_mutex
);
2680 * initialise the VMA slab
2682 void __init
mmap_init(void)
2686 ret
= percpu_counter_init(&vm_committed_as
, 0);