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>
31 #include <linux/audit.h>
32 #include <linux/khugepaged.h>
34 #include <asm/uaccess.h>
35 #include <asm/cacheflush.h>
37 #include <asm/mmu_context.h>
41 #ifndef arch_mmap_check
42 #define arch_mmap_check(addr, len, flags) (0)
45 #ifndef arch_rebalance_pgtables
46 #define arch_rebalance_pgtables(addr, len) (addr)
49 static void unmap_region(struct mm_struct
*mm
,
50 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
51 unsigned long start
, unsigned long end
);
54 * WARNING: the debugging will use recursive algorithms so never enable this
55 * unless you know what you are doing.
59 /* description of effects of mapping type and prot in current implementation.
60 * this is due to the limited x86 page protection hardware. The expected
61 * behavior is in parens:
64 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
65 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
66 * w: (no) no w: (no) no w: (yes) yes w: (no) no
67 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
69 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
70 * w: (no) no w: (no) no w: (copy) copy w: (no) no
71 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
74 pgprot_t protection_map
[16] = {
75 __P000
, __P001
, __P010
, __P011
, __P100
, __P101
, __P110
, __P111
,
76 __S000
, __S001
, __S010
, __S011
, __S100
, __S101
, __S110
, __S111
79 pgprot_t
vm_get_page_prot(unsigned long vm_flags
)
81 return __pgprot(pgprot_val(protection_map
[vm_flags
&
82 (VM_READ
|VM_WRITE
|VM_EXEC
|VM_SHARED
)]) |
83 pgprot_val(arch_vm_get_page_prot(vm_flags
)));
85 EXPORT_SYMBOL(vm_get_page_prot
);
87 int sysctl_overcommit_memory __read_mostly
= OVERCOMMIT_GUESS
; /* heuristic overcommit */
88 int sysctl_overcommit_ratio __read_mostly
= 50; /* default is 50% */
89 int sysctl_max_map_count __read_mostly
= DEFAULT_MAX_MAP_COUNT
;
91 * Make sure vm_committed_as in one cacheline and not cacheline shared with
92 * other variables. It can be updated by several CPUs frequently.
94 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp
;
97 * Check that a process has enough memory to allocate a new virtual
98 * mapping. 0 means there is enough memory for the allocation to
99 * succeed and -ENOMEM implies there is not.
101 * We currently support three overcommit policies, which are set via the
102 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting
104 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
105 * Additional code 2002 Jul 20 by Robert Love.
107 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
109 * Note this is a helper function intended to be used by LSMs which
110 * wish to use this logic.
112 int __vm_enough_memory(struct mm_struct
*mm
, long pages
, int cap_sys_admin
)
114 unsigned long free
, allowed
;
116 vm_acct_memory(pages
);
119 * Sometimes we want to use more memory than we have
121 if (sysctl_overcommit_memory
== OVERCOMMIT_ALWAYS
)
124 if (sysctl_overcommit_memory
== OVERCOMMIT_GUESS
) {
127 free
= global_page_state(NR_FILE_PAGES
);
128 free
+= nr_swap_pages
;
131 * Any slabs which are created with the
132 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
133 * which are reclaimable, under pressure. The dentry
134 * cache and most inode caches should fall into this
136 free
+= global_page_state(NR_SLAB_RECLAIMABLE
);
139 * Leave the last 3% for root
148 * nr_free_pages() is very expensive on large systems,
149 * only call if we're about to fail.
154 * Leave reserved pages. The pages are not for anonymous pages.
156 if (n
<= totalreserve_pages
)
159 n
-= totalreserve_pages
;
162 * Leave the last 3% for root
174 allowed
= (totalram_pages
- hugetlb_total_pages())
175 * sysctl_overcommit_ratio
/ 100;
177 * Leave the last 3% for root
180 allowed
-= allowed
/ 32;
181 allowed
+= total_swap_pages
;
183 /* Don't let a single process grow too big:
184 leave 3% of the size of this process for other processes */
186 allowed
-= mm
->total_vm
/ 32;
188 if (percpu_counter_read_positive(&vm_committed_as
) < allowed
)
191 vm_unacct_memory(pages
);
197 * Requires inode->i_mapping->i_mmap_lock
199 static void __remove_shared_vm_struct(struct vm_area_struct
*vma
,
200 struct file
*file
, struct address_space
*mapping
)
202 if (vma
->vm_flags
& VM_DENYWRITE
)
203 atomic_inc(&file
->f_path
.dentry
->d_inode
->i_writecount
);
204 if (vma
->vm_flags
& VM_SHARED
)
205 mapping
->i_mmap_writable
--;
207 flush_dcache_mmap_lock(mapping
);
208 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
209 list_del_init(&vma
->shared
.vm_set
.list
);
211 vma_prio_tree_remove(vma
, &mapping
->i_mmap
);
212 flush_dcache_mmap_unlock(mapping
);
216 * Unlink a file-based vm structure from its prio_tree, to hide
217 * vma from rmap and vmtruncate before freeing its page tables.
219 void unlink_file_vma(struct vm_area_struct
*vma
)
221 struct file
*file
= vma
->vm_file
;
224 struct address_space
*mapping
= file
->f_mapping
;
225 spin_lock(&mapping
->i_mmap_lock
);
226 __remove_shared_vm_struct(vma
, file
, mapping
);
227 spin_unlock(&mapping
->i_mmap_lock
);
232 * Close a vm structure and free it, returning the next.
234 static struct vm_area_struct
*remove_vma(struct vm_area_struct
*vma
)
236 struct vm_area_struct
*next
= vma
->vm_next
;
239 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
240 vma
->vm_ops
->close(vma
);
243 if (vma
->vm_flags
& VM_EXECUTABLE
)
244 removed_exe_file_vma(vma
->vm_mm
);
246 mpol_put(vma_policy(vma
));
247 kmem_cache_free(vm_area_cachep
, vma
);
251 SYSCALL_DEFINE1(brk
, unsigned long, brk
)
253 unsigned long rlim
, retval
;
254 unsigned long newbrk
, oldbrk
;
255 struct mm_struct
*mm
= current
->mm
;
256 unsigned long min_brk
;
258 down_write(&mm
->mmap_sem
);
260 #ifdef CONFIG_COMPAT_BRK
262 * CONFIG_COMPAT_BRK can still be overridden by setting
263 * randomize_va_space to 2, which will still cause mm->start_brk
264 * to be arbitrarily shifted
266 if (current
->brk_randomized
)
267 min_brk
= mm
->start_brk
;
269 min_brk
= mm
->end_data
;
271 min_brk
= mm
->start_brk
;
277 * Check against rlimit here. If this check is done later after the test
278 * of oldbrk with newbrk then it can escape the test and let the data
279 * segment grow beyond its set limit the in case where the limit is
280 * not page aligned -Ram Gupta
282 rlim
= rlimit(RLIMIT_DATA
);
283 if (rlim
< RLIM_INFINITY
&& (brk
- mm
->start_brk
) +
284 (mm
->end_data
- mm
->start_data
) > rlim
)
287 newbrk
= PAGE_ALIGN(brk
);
288 oldbrk
= PAGE_ALIGN(mm
->brk
);
289 if (oldbrk
== newbrk
)
292 /* Always allow shrinking brk. */
293 if (brk
<= mm
->brk
) {
294 if (!do_munmap(mm
, newbrk
, oldbrk
-newbrk
))
299 /* Check against existing mmap mappings. */
300 if (find_vma_intersection(mm
, oldbrk
, newbrk
+PAGE_SIZE
))
303 /* Ok, looks good - let it rip. */
304 if (do_brk(oldbrk
, newbrk
-oldbrk
) != oldbrk
)
310 up_write(&mm
->mmap_sem
);
315 static int browse_rb(struct rb_root
*root
)
318 struct rb_node
*nd
, *pn
= NULL
;
319 unsigned long prev
= 0, pend
= 0;
321 for (nd
= rb_first(root
); nd
; nd
= rb_next(nd
)) {
322 struct vm_area_struct
*vma
;
323 vma
= rb_entry(nd
, struct vm_area_struct
, vm_rb
);
324 if (vma
->vm_start
< prev
)
325 printk("vm_start %lx prev %lx\n", vma
->vm_start
, prev
), i
= -1;
326 if (vma
->vm_start
< pend
)
327 printk("vm_start %lx pend %lx\n", vma
->vm_start
, pend
);
328 if (vma
->vm_start
> vma
->vm_end
)
329 printk("vm_end %lx < vm_start %lx\n", vma
->vm_end
, vma
->vm_start
);
332 prev
= vma
->vm_start
;
336 for (nd
= pn
; nd
; nd
= rb_prev(nd
)) {
340 printk("backwards %d, forwards %d\n", j
, i
), i
= 0;
344 void validate_mm(struct mm_struct
*mm
)
348 struct vm_area_struct
*tmp
= mm
->mmap
;
353 if (i
!= mm
->map_count
)
354 printk("map_count %d vm_next %d\n", mm
->map_count
, i
), bug
= 1;
355 i
= browse_rb(&mm
->mm_rb
);
356 if (i
!= mm
->map_count
)
357 printk("map_count %d rb %d\n", mm
->map_count
, i
), bug
= 1;
361 #define validate_mm(mm) do { } while (0)
364 static struct vm_area_struct
*
365 find_vma_prepare(struct mm_struct
*mm
, unsigned long addr
,
366 struct vm_area_struct
**pprev
, struct rb_node
***rb_link
,
367 struct rb_node
** rb_parent
)
369 struct vm_area_struct
* vma
;
370 struct rb_node
** __rb_link
, * __rb_parent
, * rb_prev
;
372 __rb_link
= &mm
->mm_rb
.rb_node
;
373 rb_prev
= __rb_parent
= NULL
;
377 struct vm_area_struct
*vma_tmp
;
379 __rb_parent
= *__rb_link
;
380 vma_tmp
= rb_entry(__rb_parent
, struct vm_area_struct
, vm_rb
);
382 if (vma_tmp
->vm_end
> addr
) {
384 if (vma_tmp
->vm_start
<= addr
)
386 __rb_link
= &__rb_parent
->rb_left
;
388 rb_prev
= __rb_parent
;
389 __rb_link
= &__rb_parent
->rb_right
;
395 *pprev
= rb_entry(rb_prev
, struct vm_area_struct
, vm_rb
);
396 *rb_link
= __rb_link
;
397 *rb_parent
= __rb_parent
;
402 __vma_link_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
403 struct vm_area_struct
*prev
, struct rb_node
*rb_parent
)
405 struct vm_area_struct
*next
;
409 next
= prev
->vm_next
;
414 next
= rb_entry(rb_parent
,
415 struct vm_area_struct
, vm_rb
);
424 void __vma_link_rb(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
425 struct rb_node
**rb_link
, struct rb_node
*rb_parent
)
427 rb_link_node(&vma
->vm_rb
, rb_parent
, rb_link
);
428 rb_insert_color(&vma
->vm_rb
, &mm
->mm_rb
);
431 static void __vma_link_file(struct vm_area_struct
*vma
)
437 struct address_space
*mapping
= file
->f_mapping
;
439 if (vma
->vm_flags
& VM_DENYWRITE
)
440 atomic_dec(&file
->f_path
.dentry
->d_inode
->i_writecount
);
441 if (vma
->vm_flags
& VM_SHARED
)
442 mapping
->i_mmap_writable
++;
444 flush_dcache_mmap_lock(mapping
);
445 if (unlikely(vma
->vm_flags
& VM_NONLINEAR
))
446 vma_nonlinear_insert(vma
, &mapping
->i_mmap_nonlinear
);
448 vma_prio_tree_insert(vma
, &mapping
->i_mmap
);
449 flush_dcache_mmap_unlock(mapping
);
454 __vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
455 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
456 struct rb_node
*rb_parent
)
458 __vma_link_list(mm
, vma
, prev
, rb_parent
);
459 __vma_link_rb(mm
, vma
, rb_link
, rb_parent
);
462 static void vma_link(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
463 struct vm_area_struct
*prev
, struct rb_node
**rb_link
,
464 struct rb_node
*rb_parent
)
466 struct address_space
*mapping
= NULL
;
469 mapping
= vma
->vm_file
->f_mapping
;
472 spin_lock(&mapping
->i_mmap_lock
);
473 vma
->vm_truncate_count
= mapping
->truncate_count
;
476 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
477 __vma_link_file(vma
);
480 spin_unlock(&mapping
->i_mmap_lock
);
487 * Helper for vma_adjust in the split_vma insert case:
488 * insert vm structure into list and rbtree and anon_vma,
489 * but it has already been inserted into prio_tree earlier.
491 static void __insert_vm_struct(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
493 struct vm_area_struct
*__vma
, *prev
;
494 struct rb_node
**rb_link
, *rb_parent
;
496 __vma
= find_vma_prepare(mm
, vma
->vm_start
,&prev
, &rb_link
, &rb_parent
);
497 BUG_ON(__vma
&& __vma
->vm_start
< vma
->vm_end
);
498 __vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
503 __vma_unlink(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
504 struct vm_area_struct
*prev
)
506 struct vm_area_struct
*next
= vma
->vm_next
;
508 prev
->vm_next
= next
;
510 next
->vm_prev
= prev
;
511 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
512 if (mm
->mmap_cache
== vma
)
513 mm
->mmap_cache
= prev
;
517 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
518 * is already present in an i_mmap tree without adjusting the tree.
519 * The following helper function should be used when such adjustments
520 * are necessary. The "insert" vma (if any) is to be inserted
521 * before we drop the necessary locks.
523 int vma_adjust(struct vm_area_struct
*vma
, unsigned long start
,
524 unsigned long end
, pgoff_t pgoff
, struct vm_area_struct
*insert
)
526 struct mm_struct
*mm
= vma
->vm_mm
;
527 struct vm_area_struct
*next
= vma
->vm_next
;
528 struct vm_area_struct
*importer
= NULL
;
529 struct address_space
*mapping
= NULL
;
530 struct prio_tree_root
*root
= NULL
;
531 struct anon_vma
*anon_vma
= NULL
;
532 struct file
*file
= vma
->vm_file
;
533 long adjust_next
= 0;
536 if (next
&& !insert
) {
537 struct vm_area_struct
*exporter
= NULL
;
539 if (end
>= next
->vm_end
) {
541 * vma expands, overlapping all the next, and
542 * perhaps the one after too (mprotect case 6).
544 again
: remove_next
= 1 + (end
> next
->vm_end
);
548 } else if (end
> next
->vm_start
) {
550 * vma expands, overlapping part of the next:
551 * mprotect case 5 shifting the boundary up.
553 adjust_next
= (end
- next
->vm_start
) >> PAGE_SHIFT
;
556 } else if (end
< vma
->vm_end
) {
558 * vma shrinks, and !insert tells it's not
559 * split_vma inserting another: so it must be
560 * mprotect case 4 shifting the boundary down.
562 adjust_next
= - ((vma
->vm_end
- end
) >> PAGE_SHIFT
);
568 * Easily overlooked: when mprotect shifts the boundary,
569 * make sure the expanding vma has anon_vma set if the
570 * shrinking vma had, to cover any anon pages imported.
572 if (exporter
&& exporter
->anon_vma
&& !importer
->anon_vma
) {
573 if (anon_vma_clone(importer
, exporter
))
575 importer
->anon_vma
= exporter
->anon_vma
;
580 mapping
= file
->f_mapping
;
581 if (!(vma
->vm_flags
& VM_NONLINEAR
))
582 root
= &mapping
->i_mmap
;
583 spin_lock(&mapping
->i_mmap_lock
);
585 vma
->vm_truncate_count
!= next
->vm_truncate_count
) {
587 * unmap_mapping_range might be in progress:
588 * ensure that the expanding vma is rescanned.
590 importer
->vm_truncate_count
= 0;
593 insert
->vm_truncate_count
= vma
->vm_truncate_count
;
595 * Put into prio_tree now, so instantiated pages
596 * are visible to arm/parisc __flush_dcache_page
597 * throughout; but we cannot insert into address
598 * space until vma start or end is updated.
600 __vma_link_file(insert
);
604 vma_adjust_trans_huge(vma
, start
, end
, adjust_next
);
607 * When changing only vma->vm_end, we don't really need anon_vma
608 * lock. This is a fairly rare case by itself, but the anon_vma
609 * lock may be shared between many sibling processes. Skipping
610 * the lock for brk adjustments makes a difference sometimes.
612 if (vma
->anon_vma
&& (insert
|| importer
|| start
!= vma
->vm_start
)) {
613 anon_vma
= vma
->anon_vma
;
614 anon_vma_lock(anon_vma
);
618 flush_dcache_mmap_lock(mapping
);
619 vma_prio_tree_remove(vma
, root
);
621 vma_prio_tree_remove(next
, root
);
624 vma
->vm_start
= start
;
626 vma
->vm_pgoff
= pgoff
;
628 next
->vm_start
+= adjust_next
<< PAGE_SHIFT
;
629 next
->vm_pgoff
+= adjust_next
;
634 vma_prio_tree_insert(next
, root
);
635 vma_prio_tree_insert(vma
, root
);
636 flush_dcache_mmap_unlock(mapping
);
641 * vma_merge has merged next into vma, and needs
642 * us to remove next before dropping the locks.
644 __vma_unlink(mm
, next
, vma
);
646 __remove_shared_vm_struct(next
, file
, mapping
);
649 * split_vma has split insert from vma, and needs
650 * us to insert it before dropping the locks
651 * (it may either follow vma or precede it).
653 __insert_vm_struct(mm
, insert
);
657 anon_vma_unlock(anon_vma
);
659 spin_unlock(&mapping
->i_mmap_lock
);
664 if (next
->vm_flags
& VM_EXECUTABLE
)
665 removed_exe_file_vma(mm
);
668 anon_vma_merge(vma
, next
);
670 mpol_put(vma_policy(next
));
671 kmem_cache_free(vm_area_cachep
, next
);
673 * In mprotect's case 6 (see comments on vma_merge),
674 * we must remove another next too. It would clutter
675 * up the code too much to do both in one go.
677 if (remove_next
== 2) {
689 * If the vma has a ->close operation then the driver probably needs to release
690 * per-vma resources, so we don't attempt to merge those.
692 static inline int is_mergeable_vma(struct vm_area_struct
*vma
,
693 struct file
*file
, unsigned long vm_flags
)
695 /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
696 if ((vma
->vm_flags
^ vm_flags
) & ~VM_CAN_NONLINEAR
)
698 if (vma
->vm_file
!= file
)
700 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
705 static inline int is_mergeable_anon_vma(struct anon_vma
*anon_vma1
,
706 struct anon_vma
*anon_vma2
)
708 return !anon_vma1
|| !anon_vma2
|| (anon_vma1
== anon_vma2
);
712 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
713 * in front of (at a lower virtual address and file offset than) the vma.
715 * We cannot merge two vmas if they have differently assigned (non-NULL)
716 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
718 * We don't check here for the merged mmap wrapping around the end of pagecache
719 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
720 * wrap, nor mmaps which cover the final page at index -1UL.
723 can_vma_merge_before(struct vm_area_struct
*vma
, unsigned long vm_flags
,
724 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
726 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
727 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
728 if (vma
->vm_pgoff
== vm_pgoff
)
735 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
736 * beyond (at a higher virtual address and file offset than) the vma.
738 * We cannot merge two vmas if they have differently assigned (non-NULL)
739 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
742 can_vma_merge_after(struct vm_area_struct
*vma
, unsigned long vm_flags
,
743 struct anon_vma
*anon_vma
, struct file
*file
, pgoff_t vm_pgoff
)
745 if (is_mergeable_vma(vma
, file
, vm_flags
) &&
746 is_mergeable_anon_vma(anon_vma
, vma
->anon_vma
)) {
748 vm_pglen
= (vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
;
749 if (vma
->vm_pgoff
+ vm_pglen
== vm_pgoff
)
756 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
757 * whether that can be merged with its predecessor or its successor.
758 * Or both (it neatly fills a hole).
760 * In most cases - when called for mmap, brk or mremap - [addr,end) is
761 * certain not to be mapped by the time vma_merge is called; but when
762 * called for mprotect, it is certain to be already mapped (either at
763 * an offset within prev, or at the start of next), and the flags of
764 * this area are about to be changed to vm_flags - and the no-change
765 * case has already been eliminated.
767 * The following mprotect cases have to be considered, where AAAA is
768 * the area passed down from mprotect_fixup, never extending beyond one
769 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
771 * AAAA AAAA AAAA AAAA
772 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
773 * cannot merge might become might become might become
774 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
775 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
776 * mremap move: PPPPNNNNNNNN 8
778 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
779 * might become case 1 below case 2 below case 3 below
781 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
782 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
784 struct vm_area_struct
*vma_merge(struct mm_struct
*mm
,
785 struct vm_area_struct
*prev
, unsigned long addr
,
786 unsigned long end
, unsigned long vm_flags
,
787 struct anon_vma
*anon_vma
, struct file
*file
,
788 pgoff_t pgoff
, struct mempolicy
*policy
)
790 pgoff_t pglen
= (end
- addr
) >> PAGE_SHIFT
;
791 struct vm_area_struct
*area
, *next
;
795 * We later require that vma->vm_flags == vm_flags,
796 * so this tests vma->vm_flags & VM_SPECIAL, too.
798 if (vm_flags
& VM_SPECIAL
)
802 next
= prev
->vm_next
;
806 if (next
&& next
->vm_end
== end
) /* cases 6, 7, 8 */
807 next
= next
->vm_next
;
810 * Can it merge with the predecessor?
812 if (prev
&& prev
->vm_end
== addr
&&
813 mpol_equal(vma_policy(prev
), policy
) &&
814 can_vma_merge_after(prev
, vm_flags
,
815 anon_vma
, file
, pgoff
)) {
817 * OK, it can. Can we now merge in the successor as well?
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 is_mergeable_anon_vma(prev
->anon_vma
,
826 err
= vma_adjust(prev
, prev
->vm_start
,
827 next
->vm_end
, prev
->vm_pgoff
, NULL
);
828 } else /* cases 2, 5, 7 */
829 err
= vma_adjust(prev
, prev
->vm_start
,
830 end
, prev
->vm_pgoff
, NULL
);
833 khugepaged_enter_vma_merge(prev
);
838 * Can this new request be merged in front of next?
840 if (next
&& end
== next
->vm_start
&&
841 mpol_equal(policy
, vma_policy(next
)) &&
842 can_vma_merge_before(next
, vm_flags
,
843 anon_vma
, file
, pgoff
+pglen
)) {
844 if (prev
&& addr
< prev
->vm_end
) /* case 4 */
845 err
= vma_adjust(prev
, prev
->vm_start
,
846 addr
, prev
->vm_pgoff
, NULL
);
847 else /* cases 3, 8 */
848 err
= vma_adjust(area
, addr
, next
->vm_end
,
849 next
->vm_pgoff
- pglen
, NULL
);
852 khugepaged_enter_vma_merge(area
);
860 * Rough compatbility check to quickly see if it's even worth looking
861 * at sharing an anon_vma.
863 * They need to have the same vm_file, and the flags can only differ
864 * in things that mprotect may change.
866 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
867 * we can merge the two vma's. For example, we refuse to merge a vma if
868 * there is a vm_ops->close() function, because that indicates that the
869 * driver is doing some kind of reference counting. But that doesn't
870 * really matter for the anon_vma sharing case.
872 static int anon_vma_compatible(struct vm_area_struct
*a
, struct vm_area_struct
*b
)
874 return a
->vm_end
== b
->vm_start
&&
875 mpol_equal(vma_policy(a
), vma_policy(b
)) &&
876 a
->vm_file
== b
->vm_file
&&
877 !((a
->vm_flags
^ b
->vm_flags
) & ~(VM_READ
|VM_WRITE
|VM_EXEC
)) &&
878 b
->vm_pgoff
== a
->vm_pgoff
+ ((b
->vm_start
- a
->vm_start
) >> PAGE_SHIFT
);
882 * Do some basic sanity checking to see if we can re-use the anon_vma
883 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
884 * the same as 'old', the other will be the new one that is trying
885 * to share the anon_vma.
887 * NOTE! This runs with mm_sem held for reading, so it is possible that
888 * the anon_vma of 'old' is concurrently in the process of being set up
889 * by another page fault trying to merge _that_. But that's ok: if it
890 * is being set up, that automatically means that it will be a singleton
891 * acceptable for merging, so we can do all of this optimistically. But
892 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
894 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
895 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
896 * is to return an anon_vma that is "complex" due to having gone through
899 * We also make sure that the two vma's are compatible (adjacent,
900 * and with the same memory policies). That's all stable, even with just
901 * a read lock on the mm_sem.
903 static struct anon_vma
*reusable_anon_vma(struct vm_area_struct
*old
, struct vm_area_struct
*a
, struct vm_area_struct
*b
)
905 if (anon_vma_compatible(a
, b
)) {
906 struct anon_vma
*anon_vma
= ACCESS_ONCE(old
->anon_vma
);
908 if (anon_vma
&& list_is_singular(&old
->anon_vma_chain
))
915 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
916 * neighbouring vmas for a suitable anon_vma, before it goes off
917 * to allocate a new anon_vma. It checks because a repetitive
918 * sequence of mprotects and faults may otherwise lead to distinct
919 * anon_vmas being allocated, preventing vma merge in subsequent
922 struct anon_vma
*find_mergeable_anon_vma(struct vm_area_struct
*vma
)
924 struct anon_vma
*anon_vma
;
925 struct vm_area_struct
*near
;
931 anon_vma
= reusable_anon_vma(near
, vma
, near
);
936 * It is potentially slow to have to call find_vma_prev here.
937 * But it's only on the first write fault on the vma, not
938 * every time, and we could devise a way to avoid it later
939 * (e.g. stash info in next's anon_vma_node when assigning
940 * an anon_vma, or when trying vma_merge). Another time.
942 BUG_ON(find_vma_prev(vma
->vm_mm
, vma
->vm_start
, &near
) != vma
);
946 anon_vma
= reusable_anon_vma(near
, near
, vma
);
951 * There's no absolute need to look only at touching neighbours:
952 * we could search further afield for "compatible" anon_vmas.
953 * But it would probably just be a waste of time searching,
954 * or lead to too many vmas hanging off the same anon_vma.
955 * We're trying to allow mprotect remerging later on,
956 * not trying to minimize memory used for anon_vmas.
961 #ifdef CONFIG_PROC_FS
962 void vm_stat_account(struct mm_struct
*mm
, unsigned long flags
,
963 struct file
*file
, long pages
)
965 const unsigned long stack_flags
966 = VM_STACK_FLAGS
& (VM_GROWSUP
|VM_GROWSDOWN
);
969 mm
->shared_vm
+= pages
;
970 if ((flags
& (VM_EXEC
|VM_WRITE
)) == VM_EXEC
)
971 mm
->exec_vm
+= pages
;
972 } else if (flags
& stack_flags
)
973 mm
->stack_vm
+= pages
;
974 if (flags
& (VM_RESERVED
|VM_IO
))
975 mm
->reserved_vm
+= pages
;
977 #endif /* CONFIG_PROC_FS */
980 * The caller must hold down_write(¤t->mm->mmap_sem).
983 unsigned long do_mmap_pgoff(struct file
*file
, unsigned long addr
,
984 unsigned long len
, unsigned long prot
,
985 unsigned long flags
, unsigned long pgoff
)
987 struct mm_struct
* mm
= current
->mm
;
989 unsigned int vm_flags
;
991 unsigned long reqprot
= prot
;
994 * Does the application expect PROT_READ to imply PROT_EXEC?
996 * (the exception is when the underlying filesystem is noexec
997 * mounted, in which case we dont add PROT_EXEC.)
999 if ((prot
& PROT_READ
) && (current
->personality
& READ_IMPLIES_EXEC
))
1000 if (!(file
&& (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)))
1006 if (!(flags
& MAP_FIXED
))
1007 addr
= round_hint_to_min(addr
);
1009 /* Careful about overflows.. */
1010 len
= PAGE_ALIGN(len
);
1014 /* offset overflow? */
1015 if ((pgoff
+ (len
>> PAGE_SHIFT
)) < pgoff
)
1018 /* Too many mappings? */
1019 if (mm
->map_count
> sysctl_max_map_count
)
1022 /* Obtain the address to map to. we verify (or select) it and ensure
1023 * that it represents a valid section of the address space.
1025 addr
= get_unmapped_area(file
, addr
, len
, pgoff
, flags
);
1026 if (addr
& ~PAGE_MASK
)
1029 /* Do simple checking here so the lower-level routines won't have
1030 * to. we assume access permissions have been handled by the open
1031 * of the memory object, so we don't do any here.
1033 vm_flags
= calc_vm_prot_bits(prot
) | calc_vm_flag_bits(flags
) |
1034 mm
->def_flags
| VM_MAYREAD
| VM_MAYWRITE
| VM_MAYEXEC
;
1036 if (flags
& MAP_LOCKED
)
1037 if (!can_do_mlock())
1040 /* mlock MCL_FUTURE? */
1041 if (vm_flags
& VM_LOCKED
) {
1042 unsigned long locked
, lock_limit
;
1043 locked
= len
>> PAGE_SHIFT
;
1044 locked
+= mm
->locked_vm
;
1045 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
1046 lock_limit
>>= PAGE_SHIFT
;
1047 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
1051 inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1054 switch (flags
& MAP_TYPE
) {
1056 if ((prot
&PROT_WRITE
) && !(file
->f_mode
&FMODE_WRITE
))
1060 * Make sure we don't allow writing to an append-only
1063 if (IS_APPEND(inode
) && (file
->f_mode
& FMODE_WRITE
))
1067 * Make sure there are no mandatory locks on the file.
1069 if (locks_verify_locked(inode
))
1072 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1073 if (!(file
->f_mode
& FMODE_WRITE
))
1074 vm_flags
&= ~(VM_MAYWRITE
| VM_SHARED
);
1078 if (!(file
->f_mode
& FMODE_READ
))
1080 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
) {
1081 if (vm_flags
& VM_EXEC
)
1083 vm_flags
&= ~VM_MAYEXEC
;
1086 if (!file
->f_op
|| !file
->f_op
->mmap
)
1094 switch (flags
& MAP_TYPE
) {
1100 vm_flags
|= VM_SHARED
| VM_MAYSHARE
;
1104 * Set pgoff according to addr for anon_vma.
1106 pgoff
= addr
>> PAGE_SHIFT
;
1113 error
= security_file_mmap(file
, reqprot
, prot
, flags
, addr
, 0);
1117 return mmap_region(file
, addr
, len
, flags
, vm_flags
, pgoff
);
1119 EXPORT_SYMBOL(do_mmap_pgoff
);
1121 SYSCALL_DEFINE6(mmap_pgoff
, unsigned long, addr
, unsigned long, len
,
1122 unsigned long, prot
, unsigned long, flags
,
1123 unsigned long, fd
, unsigned long, pgoff
)
1125 struct file
*file
= NULL
;
1126 unsigned long retval
= -EBADF
;
1128 if (!(flags
& MAP_ANONYMOUS
)) {
1129 audit_mmap_fd(fd
, flags
);
1130 if (unlikely(flags
& MAP_HUGETLB
))
1135 } else if (flags
& MAP_HUGETLB
) {
1136 struct user_struct
*user
= NULL
;
1138 * VM_NORESERVE is used because the reservations will be
1139 * taken when vm_ops->mmap() is called
1140 * A dummy user value is used because we are not locking
1141 * memory so no accounting is necessary
1143 len
= ALIGN(len
, huge_page_size(&default_hstate
));
1144 file
= hugetlb_file_setup(HUGETLB_ANON_FILE
, len
, VM_NORESERVE
,
1145 &user
, HUGETLB_ANONHUGE_INODE
);
1147 return PTR_ERR(file
);
1150 flags
&= ~(MAP_EXECUTABLE
| MAP_DENYWRITE
);
1152 down_write(¤t
->mm
->mmap_sem
);
1153 retval
= do_mmap_pgoff(file
, addr
, len
, prot
, flags
, pgoff
);
1154 up_write(¤t
->mm
->mmap_sem
);
1162 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1163 struct mmap_arg_struct
{
1167 unsigned long flags
;
1169 unsigned long offset
;
1172 SYSCALL_DEFINE1(old_mmap
, struct mmap_arg_struct __user
*, arg
)
1174 struct mmap_arg_struct a
;
1176 if (copy_from_user(&a
, arg
, sizeof(a
)))
1178 if (a
.offset
& ~PAGE_MASK
)
1181 return sys_mmap_pgoff(a
.addr
, a
.len
, a
.prot
, a
.flags
, a
.fd
,
1182 a
.offset
>> PAGE_SHIFT
);
1184 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1187 * Some shared mappigns will want the pages marked read-only
1188 * to track write events. If so, we'll downgrade vm_page_prot
1189 * to the private version (using protection_map[] without the
1192 int vma_wants_writenotify(struct vm_area_struct
*vma
)
1194 unsigned int vm_flags
= vma
->vm_flags
;
1196 /* If it was private or non-writable, the write bit is already clear */
1197 if ((vm_flags
& (VM_WRITE
|VM_SHARED
)) != ((VM_WRITE
|VM_SHARED
)))
1200 /* The backer wishes to know when pages are first written to? */
1201 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
)
1204 /* The open routine did something to the protections already? */
1205 if (pgprot_val(vma
->vm_page_prot
) !=
1206 pgprot_val(vm_get_page_prot(vm_flags
)))
1209 /* Specialty mapping? */
1210 if (vm_flags
& (VM_PFNMAP
|VM_INSERTPAGE
))
1213 /* Can the mapping track the dirty pages? */
1214 return vma
->vm_file
&& vma
->vm_file
->f_mapping
&&
1215 mapping_cap_account_dirty(vma
->vm_file
->f_mapping
);
1219 * We account for memory if it's a private writeable mapping,
1220 * not hugepages and VM_NORESERVE wasn't set.
1222 static inline int accountable_mapping(struct file
*file
, unsigned int vm_flags
)
1225 * hugetlb has its own accounting separate from the core VM
1226 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1228 if (file
&& is_file_hugepages(file
))
1231 return (vm_flags
& (VM_NORESERVE
| VM_SHARED
| VM_WRITE
)) == VM_WRITE
;
1234 unsigned long mmap_region(struct file
*file
, unsigned long addr
,
1235 unsigned long len
, unsigned long flags
,
1236 unsigned int vm_flags
, unsigned long pgoff
)
1238 struct mm_struct
*mm
= current
->mm
;
1239 struct vm_area_struct
*vma
, *prev
;
1240 int correct_wcount
= 0;
1242 struct rb_node
**rb_link
, *rb_parent
;
1243 unsigned long charged
= 0;
1244 struct inode
*inode
= file
? file
->f_path
.dentry
->d_inode
: NULL
;
1246 /* Clear old maps */
1249 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
1250 if (vma
&& vma
->vm_start
< addr
+ len
) {
1251 if (do_munmap(mm
, addr
, len
))
1256 /* Check against address space limit. */
1257 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
1261 * Set 'VM_NORESERVE' if we should not account for the
1262 * memory use of this mapping.
1264 if ((flags
& MAP_NORESERVE
)) {
1265 /* We honor MAP_NORESERVE if allowed to overcommit */
1266 if (sysctl_overcommit_memory
!= OVERCOMMIT_NEVER
)
1267 vm_flags
|= VM_NORESERVE
;
1269 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1270 if (file
&& is_file_hugepages(file
))
1271 vm_flags
|= VM_NORESERVE
;
1275 * Private writable mapping: check memory availability
1277 if (accountable_mapping(file
, vm_flags
)) {
1278 charged
= len
>> PAGE_SHIFT
;
1279 if (security_vm_enough_memory(charged
))
1281 vm_flags
|= VM_ACCOUNT
;
1285 * Can we just expand an old mapping?
1287 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vm_flags
, NULL
, file
, pgoff
, NULL
);
1292 * Determine the object being mapped and call the appropriate
1293 * specific mapper. the address has already been validated, but
1294 * not unmapped, but the maps are removed from the list.
1296 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
1303 vma
->vm_start
= addr
;
1304 vma
->vm_end
= addr
+ len
;
1305 vma
->vm_flags
= vm_flags
;
1306 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
);
1307 vma
->vm_pgoff
= pgoff
;
1308 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
1312 if (vm_flags
& (VM_GROWSDOWN
|VM_GROWSUP
))
1314 if (vm_flags
& VM_DENYWRITE
) {
1315 error
= deny_write_access(file
);
1320 vma
->vm_file
= file
;
1322 error
= file
->f_op
->mmap(file
, vma
);
1324 goto unmap_and_free_vma
;
1325 if (vm_flags
& VM_EXECUTABLE
)
1326 added_exe_file_vma(mm
);
1328 /* Can addr have changed??
1330 * Answer: Yes, several device drivers can do it in their
1331 * f_op->mmap method. -DaveM
1333 addr
= vma
->vm_start
;
1334 pgoff
= vma
->vm_pgoff
;
1335 vm_flags
= vma
->vm_flags
;
1336 } else if (vm_flags
& VM_SHARED
) {
1337 error
= shmem_zero_setup(vma
);
1342 if (vma_wants_writenotify(vma
)) {
1343 pgprot_t pprot
= vma
->vm_page_prot
;
1345 /* Can vma->vm_page_prot have changed??
1347 * Answer: Yes, drivers may have changed it in their
1348 * f_op->mmap method.
1350 * Ensures that vmas marked as uncached stay that way.
1352 vma
->vm_page_prot
= vm_get_page_prot(vm_flags
& ~VM_SHARED
);
1353 if (pgprot_val(pprot
) == pgprot_val(pgprot_noncached(pprot
)))
1354 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
1357 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
1358 file
= vma
->vm_file
;
1360 /* Once vma denies write, undo our temporary denial count */
1362 atomic_inc(&inode
->i_writecount
);
1364 perf_event_mmap(vma
);
1366 mm
->total_vm
+= len
>> PAGE_SHIFT
;
1367 vm_stat_account(mm
, vm_flags
, file
, len
>> PAGE_SHIFT
);
1368 if (vm_flags
& VM_LOCKED
) {
1369 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
1370 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
1371 } else if ((flags
& MAP_POPULATE
) && !(flags
& MAP_NONBLOCK
))
1372 make_pages_present(addr
, addr
+ len
);
1377 atomic_inc(&inode
->i_writecount
);
1378 vma
->vm_file
= NULL
;
1381 /* Undo any partial mapping done by a device driver. */
1382 unmap_region(mm
, vma
, prev
, vma
->vm_start
, vma
->vm_end
);
1385 kmem_cache_free(vm_area_cachep
, vma
);
1388 vm_unacct_memory(charged
);
1392 /* Get an address range which is currently unmapped.
1393 * For shmat() with addr=0.
1395 * Ugly calling convention alert:
1396 * Return value with the low bits set means error value,
1398 * if (ret & ~PAGE_MASK)
1401 * This function "knows" that -ENOMEM has the bits set.
1403 #ifndef HAVE_ARCH_UNMAPPED_AREA
1405 arch_get_unmapped_area(struct file
*filp
, unsigned long addr
,
1406 unsigned long len
, unsigned long pgoff
, unsigned long flags
)
1408 struct mm_struct
*mm
= current
->mm
;
1409 struct vm_area_struct
*vma
;
1410 unsigned long start_addr
;
1412 if (len
> TASK_SIZE
)
1415 if (flags
& MAP_FIXED
)
1419 addr
= PAGE_ALIGN(addr
);
1420 vma
= find_vma(mm
, addr
);
1421 if (TASK_SIZE
- len
>= addr
&&
1422 (!vma
|| addr
+ len
<= vma
->vm_start
))
1425 if (len
> mm
->cached_hole_size
) {
1426 start_addr
= addr
= mm
->free_area_cache
;
1428 start_addr
= addr
= TASK_UNMAPPED_BASE
;
1429 mm
->cached_hole_size
= 0;
1433 for (vma
= find_vma(mm
, addr
); ; vma
= vma
->vm_next
) {
1434 /* At this point: (!vma || addr < vma->vm_end). */
1435 if (TASK_SIZE
- len
< addr
) {
1437 * Start a new search - just in case we missed
1440 if (start_addr
!= TASK_UNMAPPED_BASE
) {
1441 addr
= TASK_UNMAPPED_BASE
;
1443 mm
->cached_hole_size
= 0;
1448 if (!vma
|| addr
+ len
<= vma
->vm_start
) {
1450 * Remember the place where we stopped the search:
1452 mm
->free_area_cache
= addr
+ len
;
1455 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1456 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1462 void arch_unmap_area(struct mm_struct
*mm
, unsigned long addr
)
1465 * Is this a new hole at the lowest possible address?
1467 if (addr
>= TASK_UNMAPPED_BASE
&& addr
< mm
->free_area_cache
) {
1468 mm
->free_area_cache
= addr
;
1469 mm
->cached_hole_size
= ~0UL;
1474 * This mmap-allocator allocates new areas top-down from below the
1475 * stack's low limit (the base):
1477 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1479 arch_get_unmapped_area_topdown(struct file
*filp
, const unsigned long addr0
,
1480 const unsigned long len
, const unsigned long pgoff
,
1481 const unsigned long flags
)
1483 struct vm_area_struct
*vma
;
1484 struct mm_struct
*mm
= current
->mm
;
1485 unsigned long addr
= addr0
;
1487 /* requested length too big for entire address space */
1488 if (len
> TASK_SIZE
)
1491 if (flags
& MAP_FIXED
)
1494 /* requesting a specific address */
1496 addr
= PAGE_ALIGN(addr
);
1497 vma
= find_vma(mm
, addr
);
1498 if (TASK_SIZE
- len
>= addr
&&
1499 (!vma
|| addr
+ len
<= vma
->vm_start
))
1503 /* check if free_area_cache is useful for us */
1504 if (len
<= mm
->cached_hole_size
) {
1505 mm
->cached_hole_size
= 0;
1506 mm
->free_area_cache
= mm
->mmap_base
;
1509 /* either no address requested or can't fit in requested address hole */
1510 addr
= mm
->free_area_cache
;
1512 /* make sure it can fit in the remaining address space */
1514 vma
= find_vma(mm
, addr
-len
);
1515 if (!vma
|| addr
<= vma
->vm_start
)
1516 /* remember the address as a hint for next time */
1517 return (mm
->free_area_cache
= addr
-len
);
1520 if (mm
->mmap_base
< len
)
1523 addr
= mm
->mmap_base
-len
;
1527 * Lookup failure means no vma is above this address,
1528 * else if new region fits below vma->vm_start,
1529 * return with success:
1531 vma
= find_vma(mm
, addr
);
1532 if (!vma
|| addr
+len
<= vma
->vm_start
)
1533 /* remember the address as a hint for next time */
1534 return (mm
->free_area_cache
= addr
);
1536 /* remember the largest hole we saw so far */
1537 if (addr
+ mm
->cached_hole_size
< vma
->vm_start
)
1538 mm
->cached_hole_size
= vma
->vm_start
- addr
;
1540 /* try just below the current vma->vm_start */
1541 addr
= vma
->vm_start
-len
;
1542 } while (len
< vma
->vm_start
);
1546 * A failed mmap() very likely causes application failure,
1547 * so fall back to the bottom-up function here. This scenario
1548 * can happen with large stack limits and large mmap()
1551 mm
->cached_hole_size
= ~0UL;
1552 mm
->free_area_cache
= TASK_UNMAPPED_BASE
;
1553 addr
= arch_get_unmapped_area(filp
, addr0
, len
, pgoff
, flags
);
1555 * Restore the topdown base:
1557 mm
->free_area_cache
= mm
->mmap_base
;
1558 mm
->cached_hole_size
= ~0UL;
1564 void arch_unmap_area_topdown(struct mm_struct
*mm
, unsigned long addr
)
1567 * Is this a new hole at the highest possible address?
1569 if (addr
> mm
->free_area_cache
)
1570 mm
->free_area_cache
= addr
;
1572 /* dont allow allocations above current base */
1573 if (mm
->free_area_cache
> mm
->mmap_base
)
1574 mm
->free_area_cache
= mm
->mmap_base
;
1578 get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
1579 unsigned long pgoff
, unsigned long flags
)
1581 unsigned long (*get_area
)(struct file
*, unsigned long,
1582 unsigned long, unsigned long, unsigned long);
1584 unsigned long error
= arch_mmap_check(addr
, len
, flags
);
1588 /* Careful about overflows.. */
1589 if (len
> TASK_SIZE
)
1592 get_area
= current
->mm
->get_unmapped_area
;
1593 if (file
&& file
->f_op
&& file
->f_op
->get_unmapped_area
)
1594 get_area
= file
->f_op
->get_unmapped_area
;
1595 addr
= get_area(file
, addr
, len
, pgoff
, flags
);
1596 if (IS_ERR_VALUE(addr
))
1599 if (addr
> TASK_SIZE
- len
)
1601 if (addr
& ~PAGE_MASK
)
1604 return arch_rebalance_pgtables(addr
, len
);
1607 EXPORT_SYMBOL(get_unmapped_area
);
1609 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1610 struct vm_area_struct
*find_vma(struct mm_struct
*mm
, unsigned long addr
)
1612 struct vm_area_struct
*vma
= NULL
;
1615 /* Check the cache first. */
1616 /* (Cache hit rate is typically around 35%.) */
1617 vma
= mm
->mmap_cache
;
1618 if (!(vma
&& vma
->vm_end
> addr
&& vma
->vm_start
<= addr
)) {
1619 struct rb_node
* rb_node
;
1621 rb_node
= mm
->mm_rb
.rb_node
;
1625 struct vm_area_struct
* vma_tmp
;
1627 vma_tmp
= rb_entry(rb_node
,
1628 struct vm_area_struct
, vm_rb
);
1630 if (vma_tmp
->vm_end
> addr
) {
1632 if (vma_tmp
->vm_start
<= addr
)
1634 rb_node
= rb_node
->rb_left
;
1636 rb_node
= rb_node
->rb_right
;
1639 mm
->mmap_cache
= vma
;
1645 EXPORT_SYMBOL(find_vma
);
1647 /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
1648 struct vm_area_struct
*
1649 find_vma_prev(struct mm_struct
*mm
, unsigned long addr
,
1650 struct vm_area_struct
**pprev
)
1652 struct vm_area_struct
*vma
= NULL
, *prev
= NULL
;
1653 struct rb_node
*rb_node
;
1657 /* Guard against addr being lower than the first VMA */
1660 /* Go through the RB tree quickly. */
1661 rb_node
= mm
->mm_rb
.rb_node
;
1664 struct vm_area_struct
*vma_tmp
;
1665 vma_tmp
= rb_entry(rb_node
, struct vm_area_struct
, vm_rb
);
1667 if (addr
< vma_tmp
->vm_end
) {
1668 rb_node
= rb_node
->rb_left
;
1671 if (!prev
->vm_next
|| (addr
< prev
->vm_next
->vm_end
))
1673 rb_node
= rb_node
->rb_right
;
1679 return prev
? prev
->vm_next
: vma
;
1683 * Verify that the stack growth is acceptable and
1684 * update accounting. This is shared with both the
1685 * grow-up and grow-down cases.
1687 static int acct_stack_growth(struct vm_area_struct
*vma
, unsigned long size
, unsigned long grow
)
1689 struct mm_struct
*mm
= vma
->vm_mm
;
1690 struct rlimit
*rlim
= current
->signal
->rlim
;
1691 unsigned long new_start
;
1693 /* address space limit tests */
1694 if (!may_expand_vm(mm
, grow
))
1697 /* Stack limit test */
1698 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
))
1701 /* mlock limit tests */
1702 if (vma
->vm_flags
& VM_LOCKED
) {
1703 unsigned long locked
;
1704 unsigned long limit
;
1705 locked
= mm
->locked_vm
+ grow
;
1706 limit
= ACCESS_ONCE(rlim
[RLIMIT_MEMLOCK
].rlim_cur
);
1707 limit
>>= PAGE_SHIFT
;
1708 if (locked
> limit
&& !capable(CAP_IPC_LOCK
))
1712 /* Check to ensure the stack will not grow into a hugetlb-only region */
1713 new_start
= (vma
->vm_flags
& VM_GROWSUP
) ? vma
->vm_start
:
1715 if (is_hugepage_only_range(vma
->vm_mm
, new_start
, size
))
1719 * Overcommit.. This must be the final test, as it will
1720 * update security statistics.
1722 if (security_vm_enough_memory_mm(mm
, grow
))
1725 /* Ok, everything looks good - let it rip */
1726 mm
->total_vm
+= grow
;
1727 if (vma
->vm_flags
& VM_LOCKED
)
1728 mm
->locked_vm
+= grow
;
1729 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, grow
);
1733 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1735 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1736 * vma is the last one with address > vma->vm_end. Have to extend vma.
1738 int expand_upwards(struct vm_area_struct
*vma
, unsigned long address
)
1742 if (!(vma
->vm_flags
& VM_GROWSUP
))
1746 * We must make sure the anon_vma is allocated
1747 * so that the anon_vma locking is not a noop.
1749 if (unlikely(anon_vma_prepare(vma
)))
1751 vma_lock_anon_vma(vma
);
1754 * vma->vm_start/vm_end cannot change under us because the caller
1755 * is required to hold the mmap_sem in read mode. We need the
1756 * anon_vma lock to serialize against concurrent expand_stacks.
1757 * Also guard against wrapping around to address 0.
1759 if (address
< PAGE_ALIGN(address
+4))
1760 address
= PAGE_ALIGN(address
+4);
1762 vma_unlock_anon_vma(vma
);
1767 /* Somebody else might have raced and expanded it already */
1768 if (address
> vma
->vm_end
) {
1769 unsigned long size
, grow
;
1771 size
= address
- vma
->vm_start
;
1772 grow
= (address
- vma
->vm_end
) >> PAGE_SHIFT
;
1775 if (vma
->vm_pgoff
+ (size
>> PAGE_SHIFT
) >= vma
->vm_pgoff
) {
1776 error
= acct_stack_growth(vma
, size
, grow
);
1778 vma
->vm_end
= address
;
1779 perf_event_mmap(vma
);
1783 vma_unlock_anon_vma(vma
);
1784 khugepaged_enter_vma_merge(vma
);
1787 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1790 * vma is the first one with address < vma->vm_start. Have to extend vma.
1792 static int expand_downwards(struct vm_area_struct
*vma
,
1793 unsigned long address
)
1798 * We must make sure the anon_vma is allocated
1799 * so that the anon_vma locking is not a noop.
1801 if (unlikely(anon_vma_prepare(vma
)))
1804 address
&= PAGE_MASK
;
1805 error
= security_file_mmap(NULL
, 0, 0, 0, address
, 1);
1809 vma_lock_anon_vma(vma
);
1812 * vma->vm_start/vm_end cannot change under us because the caller
1813 * is required to hold the mmap_sem in read mode. We need the
1814 * anon_vma lock to serialize against concurrent expand_stacks.
1817 /* Somebody else might have raced and expanded it already */
1818 if (address
< vma
->vm_start
) {
1819 unsigned long size
, grow
;
1821 size
= vma
->vm_end
- address
;
1822 grow
= (vma
->vm_start
- address
) >> PAGE_SHIFT
;
1825 if (grow
<= vma
->vm_pgoff
) {
1826 error
= acct_stack_growth(vma
, size
, grow
);
1828 vma
->vm_start
= address
;
1829 vma
->vm_pgoff
-= grow
;
1830 perf_event_mmap(vma
);
1834 vma_unlock_anon_vma(vma
);
1835 khugepaged_enter_vma_merge(vma
);
1839 int expand_stack_downwards(struct vm_area_struct
*vma
, unsigned long address
)
1841 return expand_downwards(vma
, address
);
1844 #ifdef CONFIG_STACK_GROWSUP
1845 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1847 return expand_upwards(vma
, address
);
1850 struct vm_area_struct
*
1851 find_extend_vma(struct mm_struct
*mm
, unsigned long addr
)
1853 struct vm_area_struct
*vma
, *prev
;
1856 vma
= find_vma_prev(mm
, addr
, &prev
);
1857 if (vma
&& (vma
->vm_start
<= addr
))
1859 if (!prev
|| expand_stack(prev
, addr
))
1861 if (prev
->vm_flags
& VM_LOCKED
) {
1862 mlock_vma_pages_range(prev
, addr
, prev
->vm_end
);
1867 int expand_stack(struct vm_area_struct
*vma
, unsigned long address
)
1869 return expand_downwards(vma
, address
);
1872 struct vm_area_struct
*
1873 find_extend_vma(struct mm_struct
* mm
, unsigned long addr
)
1875 struct vm_area_struct
* vma
;
1876 unsigned long start
;
1879 vma
= find_vma(mm
,addr
);
1882 if (vma
->vm_start
<= addr
)
1884 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
1886 start
= vma
->vm_start
;
1887 if (expand_stack(vma
, addr
))
1889 if (vma
->vm_flags
& VM_LOCKED
) {
1890 mlock_vma_pages_range(vma
, addr
, start
);
1897 * Ok - we have the memory areas we should free on the vma list,
1898 * so release them, and do the vma updates.
1900 * Called with the mm semaphore held.
1902 static void remove_vma_list(struct mm_struct
*mm
, struct vm_area_struct
*vma
)
1904 /* Update high watermark before we lower total_vm */
1905 update_hiwater_vm(mm
);
1907 long nrpages
= vma_pages(vma
);
1909 mm
->total_vm
-= nrpages
;
1910 vm_stat_account(mm
, vma
->vm_flags
, vma
->vm_file
, -nrpages
);
1911 vma
= remove_vma(vma
);
1917 * Get rid of page table information in the indicated region.
1919 * Called with the mm semaphore held.
1921 static void unmap_region(struct mm_struct
*mm
,
1922 struct vm_area_struct
*vma
, struct vm_area_struct
*prev
,
1923 unsigned long start
, unsigned long end
)
1925 struct vm_area_struct
*next
= prev
? prev
->vm_next
: mm
->mmap
;
1926 struct mmu_gather
*tlb
;
1927 unsigned long nr_accounted
= 0;
1930 tlb
= tlb_gather_mmu(mm
, 0);
1931 update_hiwater_rss(mm
);
1932 unmap_vmas(&tlb
, vma
, start
, end
, &nr_accounted
, NULL
);
1933 vm_unacct_memory(nr_accounted
);
1934 free_pgtables(tlb
, vma
, prev
? prev
->vm_end
: FIRST_USER_ADDRESS
,
1935 next
? next
->vm_start
: 0);
1936 tlb_finish_mmu(tlb
, start
, end
);
1940 * Create a list of vma's touched by the unmap, removing them from the mm's
1941 * vma list as we go..
1944 detach_vmas_to_be_unmapped(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1945 struct vm_area_struct
*prev
, unsigned long end
)
1947 struct vm_area_struct
**insertion_point
;
1948 struct vm_area_struct
*tail_vma
= NULL
;
1951 insertion_point
= (prev
? &prev
->vm_next
: &mm
->mmap
);
1952 vma
->vm_prev
= NULL
;
1954 rb_erase(&vma
->vm_rb
, &mm
->mm_rb
);
1958 } while (vma
&& vma
->vm_start
< end
);
1959 *insertion_point
= vma
;
1961 vma
->vm_prev
= prev
;
1962 tail_vma
->vm_next
= NULL
;
1963 if (mm
->unmap_area
== arch_unmap_area
)
1964 addr
= prev
? prev
->vm_end
: mm
->mmap_base
;
1966 addr
= vma
? vma
->vm_start
: mm
->mmap_base
;
1967 mm
->unmap_area(mm
, addr
);
1968 mm
->mmap_cache
= NULL
; /* Kill the cache. */
1972 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
1973 * munmap path where it doesn't make sense to fail.
1975 static int __split_vma(struct mm_struct
* mm
, struct vm_area_struct
* vma
,
1976 unsigned long addr
, int new_below
)
1978 struct mempolicy
*pol
;
1979 struct vm_area_struct
*new;
1982 if (is_vm_hugetlb_page(vma
) && (addr
&
1983 ~(huge_page_mask(hstate_vma(vma
)))))
1986 new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
1990 /* most fields are the same, copy all, and then fixup */
1993 INIT_LIST_HEAD(&new->anon_vma_chain
);
1998 new->vm_start
= addr
;
1999 new->vm_pgoff
+= ((addr
- vma
->vm_start
) >> PAGE_SHIFT
);
2002 pol
= mpol_dup(vma_policy(vma
));
2007 vma_set_policy(new, pol
);
2009 if (anon_vma_clone(new, vma
))
2013 get_file(new->vm_file
);
2014 if (vma
->vm_flags
& VM_EXECUTABLE
)
2015 added_exe_file_vma(mm
);
2018 if (new->vm_ops
&& new->vm_ops
->open
)
2019 new->vm_ops
->open(new);
2022 err
= vma_adjust(vma
, addr
, vma
->vm_end
, vma
->vm_pgoff
+
2023 ((addr
- new->vm_start
) >> PAGE_SHIFT
), new);
2025 err
= vma_adjust(vma
, vma
->vm_start
, addr
, vma
->vm_pgoff
, new);
2031 /* Clean everything up if vma_adjust failed. */
2032 if (new->vm_ops
&& new->vm_ops
->close
)
2033 new->vm_ops
->close(new);
2035 if (vma
->vm_flags
& VM_EXECUTABLE
)
2036 removed_exe_file_vma(mm
);
2039 unlink_anon_vmas(new);
2043 kmem_cache_free(vm_area_cachep
, new);
2049 * Split a vma into two pieces at address 'addr', a new vma is allocated
2050 * either for the first part or the tail.
2052 int split_vma(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2053 unsigned long addr
, int new_below
)
2055 if (mm
->map_count
>= sysctl_max_map_count
)
2058 return __split_vma(mm
, vma
, addr
, new_below
);
2061 /* Munmap is split into 2 main parts -- this part which finds
2062 * what needs doing, and the areas themselves, which do the
2063 * work. This now handles partial unmappings.
2064 * Jeremy Fitzhardinge <jeremy@goop.org>
2066 int do_munmap(struct mm_struct
*mm
, unsigned long start
, size_t len
)
2069 struct vm_area_struct
*vma
, *prev
, *last
;
2071 if ((start
& ~PAGE_MASK
) || start
> TASK_SIZE
|| len
> TASK_SIZE
-start
)
2074 if ((len
= PAGE_ALIGN(len
)) == 0)
2077 /* Find the first overlapping VMA */
2078 vma
= find_vma_prev(mm
, start
, &prev
);
2081 /* we have start < vma->vm_end */
2083 /* if it doesn't overlap, we have nothing.. */
2085 if (vma
->vm_start
>= end
)
2089 * If we need to split any vma, do it now to save pain later.
2091 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2092 * unmapped vm_area_struct will remain in use: so lower split_vma
2093 * places tmp vma above, and higher split_vma places tmp vma below.
2095 if (start
> vma
->vm_start
) {
2099 * Make sure that map_count on return from munmap() will
2100 * not exceed its limit; but let map_count go just above
2101 * its limit temporarily, to help free resources as expected.
2103 if (end
< vma
->vm_end
&& mm
->map_count
>= sysctl_max_map_count
)
2106 error
= __split_vma(mm
, vma
, start
, 0);
2112 /* Does it split the last one? */
2113 last
= find_vma(mm
, end
);
2114 if (last
&& end
> last
->vm_start
) {
2115 int error
= __split_vma(mm
, last
, end
, 1);
2119 vma
= prev
? prev
->vm_next
: mm
->mmap
;
2122 * unlock any mlock()ed ranges before detaching vmas
2124 if (mm
->locked_vm
) {
2125 struct vm_area_struct
*tmp
= vma
;
2126 while (tmp
&& tmp
->vm_start
< end
) {
2127 if (tmp
->vm_flags
& VM_LOCKED
) {
2128 mm
->locked_vm
-= vma_pages(tmp
);
2129 munlock_vma_pages_all(tmp
);
2136 * Remove the vma's, and unmap the actual pages
2138 detach_vmas_to_be_unmapped(mm
, vma
, prev
, end
);
2139 unmap_region(mm
, vma
, prev
, start
, end
);
2141 /* Fix up all other VM information */
2142 remove_vma_list(mm
, vma
);
2147 EXPORT_SYMBOL(do_munmap
);
2149 SYSCALL_DEFINE2(munmap
, unsigned long, addr
, size_t, len
)
2152 struct mm_struct
*mm
= current
->mm
;
2154 profile_munmap(addr
);
2156 down_write(&mm
->mmap_sem
);
2157 ret
= do_munmap(mm
, addr
, len
);
2158 up_write(&mm
->mmap_sem
);
2162 static inline void verify_mm_writelocked(struct mm_struct
*mm
)
2164 #ifdef CONFIG_DEBUG_VM
2165 if (unlikely(down_read_trylock(&mm
->mmap_sem
))) {
2167 up_read(&mm
->mmap_sem
);
2173 * this is really a simplified "do_mmap". it only handles
2174 * anonymous maps. eventually we may be able to do some
2175 * brk-specific accounting here.
2177 unsigned long do_brk(unsigned long addr
, unsigned long len
)
2179 struct mm_struct
* mm
= current
->mm
;
2180 struct vm_area_struct
* vma
, * prev
;
2181 unsigned long flags
;
2182 struct rb_node
** rb_link
, * rb_parent
;
2183 pgoff_t pgoff
= addr
>> PAGE_SHIFT
;
2186 len
= PAGE_ALIGN(len
);
2190 error
= security_file_mmap(NULL
, 0, 0, 0, addr
, 1);
2194 flags
= VM_DATA_DEFAULT_FLAGS
| VM_ACCOUNT
| mm
->def_flags
;
2196 error
= get_unmapped_area(NULL
, addr
, len
, 0, MAP_FIXED
);
2197 if (error
& ~PAGE_MASK
)
2203 if (mm
->def_flags
& VM_LOCKED
) {
2204 unsigned long locked
, lock_limit
;
2205 locked
= len
>> PAGE_SHIFT
;
2206 locked
+= mm
->locked_vm
;
2207 lock_limit
= rlimit(RLIMIT_MEMLOCK
);
2208 lock_limit
>>= PAGE_SHIFT
;
2209 if (locked
> lock_limit
&& !capable(CAP_IPC_LOCK
))
2214 * mm->mmap_sem is required to protect against another thread
2215 * changing the mappings in case we sleep.
2217 verify_mm_writelocked(mm
);
2220 * Clear old maps. this also does some error checking for us
2223 vma
= find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2224 if (vma
&& vma
->vm_start
< addr
+ len
) {
2225 if (do_munmap(mm
, addr
, len
))
2230 /* Check against address space limits *after* clearing old maps... */
2231 if (!may_expand_vm(mm
, len
>> PAGE_SHIFT
))
2234 if (mm
->map_count
> sysctl_max_map_count
)
2237 if (security_vm_enough_memory(len
>> PAGE_SHIFT
))
2240 /* Can we just expand an old private anonymous mapping? */
2241 vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, flags
,
2242 NULL
, NULL
, pgoff
, NULL
);
2247 * create a vma struct for an anonymous mapping
2249 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2251 vm_unacct_memory(len
>> PAGE_SHIFT
);
2255 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2257 vma
->vm_start
= addr
;
2258 vma
->vm_end
= addr
+ len
;
2259 vma
->vm_pgoff
= pgoff
;
2260 vma
->vm_flags
= flags
;
2261 vma
->vm_page_prot
= vm_get_page_prot(flags
);
2262 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2264 perf_event_mmap(vma
);
2265 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2266 if (flags
& VM_LOCKED
) {
2267 if (!mlock_vma_pages_range(vma
, addr
, addr
+ len
))
2268 mm
->locked_vm
+= (len
>> PAGE_SHIFT
);
2273 EXPORT_SYMBOL(do_brk
);
2275 /* Release all mmaps. */
2276 void exit_mmap(struct mm_struct
*mm
)
2278 struct mmu_gather
*tlb
;
2279 struct vm_area_struct
*vma
;
2280 unsigned long nr_accounted
= 0;
2283 /* mm's last user has gone, and its about to be pulled down */
2284 mmu_notifier_release(mm
);
2286 if (mm
->locked_vm
) {
2289 if (vma
->vm_flags
& VM_LOCKED
)
2290 munlock_vma_pages_all(vma
);
2298 if (!vma
) /* Can happen if dup_mmap() received an OOM */
2303 tlb
= tlb_gather_mmu(mm
, 1);
2304 /* update_hiwater_rss(mm) here? but nobody should be looking */
2305 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2306 end
= unmap_vmas(&tlb
, vma
, 0, -1, &nr_accounted
, NULL
);
2307 vm_unacct_memory(nr_accounted
);
2309 free_pgtables(tlb
, vma
, FIRST_USER_ADDRESS
, 0);
2310 tlb_finish_mmu(tlb
, 0, end
);
2313 * Walk the list again, actually closing and freeing it,
2314 * with preemption enabled, without holding any MM locks.
2317 vma
= remove_vma(vma
);
2319 BUG_ON(mm
->nr_ptes
> (FIRST_USER_ADDRESS
+PMD_SIZE
-1)>>PMD_SHIFT
);
2322 /* Insert vm structure into process list sorted by address
2323 * and into the inode's i_mmap tree. If vm_file is non-NULL
2324 * then i_mmap_lock is taken here.
2326 int insert_vm_struct(struct mm_struct
* mm
, struct vm_area_struct
* vma
)
2328 struct vm_area_struct
* __vma
, * prev
;
2329 struct rb_node
** rb_link
, * rb_parent
;
2332 * The vm_pgoff of a purely anonymous vma should be irrelevant
2333 * until its first write fault, when page's anon_vma and index
2334 * are set. But now set the vm_pgoff it will almost certainly
2335 * end up with (unless mremap moves it elsewhere before that
2336 * first wfault), so /proc/pid/maps tells a consistent story.
2338 * By setting it to reflect the virtual start address of the
2339 * vma, merges and splits can happen in a seamless way, just
2340 * using the existing file pgoff checks and manipulations.
2341 * Similarly in do_mmap_pgoff and in do_brk.
2343 if (!vma
->vm_file
) {
2344 BUG_ON(vma
->anon_vma
);
2345 vma
->vm_pgoff
= vma
->vm_start
>> PAGE_SHIFT
;
2347 __vma
= find_vma_prepare(mm
,vma
->vm_start
,&prev
,&rb_link
,&rb_parent
);
2348 if (__vma
&& __vma
->vm_start
< vma
->vm_end
)
2350 if ((vma
->vm_flags
& VM_ACCOUNT
) &&
2351 security_vm_enough_memory_mm(mm
, vma_pages(vma
)))
2353 vma_link(mm
, vma
, prev
, rb_link
, rb_parent
);
2358 * Copy the vma structure to a new location in the same mm,
2359 * prior to moving page table entries, to effect an mremap move.
2361 struct vm_area_struct
*copy_vma(struct vm_area_struct
**vmap
,
2362 unsigned long addr
, unsigned long len
, pgoff_t pgoff
)
2364 struct vm_area_struct
*vma
= *vmap
;
2365 unsigned long vma_start
= vma
->vm_start
;
2366 struct mm_struct
*mm
= vma
->vm_mm
;
2367 struct vm_area_struct
*new_vma
, *prev
;
2368 struct rb_node
**rb_link
, *rb_parent
;
2369 struct mempolicy
*pol
;
2372 * If anonymous vma has not yet been faulted, update new pgoff
2373 * to match new location, to increase its chance of merging.
2375 if (!vma
->vm_file
&& !vma
->anon_vma
)
2376 pgoff
= addr
>> PAGE_SHIFT
;
2378 find_vma_prepare(mm
, addr
, &prev
, &rb_link
, &rb_parent
);
2379 new_vma
= vma_merge(mm
, prev
, addr
, addr
+ len
, vma
->vm_flags
,
2380 vma
->anon_vma
, vma
->vm_file
, pgoff
, vma_policy(vma
));
2383 * Source vma may have been merged into new_vma
2385 if (vma_start
>= new_vma
->vm_start
&&
2386 vma_start
< new_vma
->vm_end
)
2389 new_vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
2392 pol
= mpol_dup(vma_policy(vma
));
2395 INIT_LIST_HEAD(&new_vma
->anon_vma_chain
);
2396 if (anon_vma_clone(new_vma
, vma
))
2397 goto out_free_mempol
;
2398 vma_set_policy(new_vma
, pol
);
2399 new_vma
->vm_start
= addr
;
2400 new_vma
->vm_end
= addr
+ len
;
2401 new_vma
->vm_pgoff
= pgoff
;
2402 if (new_vma
->vm_file
) {
2403 get_file(new_vma
->vm_file
);
2404 if (vma
->vm_flags
& VM_EXECUTABLE
)
2405 added_exe_file_vma(mm
);
2407 if (new_vma
->vm_ops
&& new_vma
->vm_ops
->open
)
2408 new_vma
->vm_ops
->open(new_vma
);
2409 vma_link(mm
, new_vma
, prev
, rb_link
, rb_parent
);
2417 kmem_cache_free(vm_area_cachep
, new_vma
);
2422 * Return true if the calling process may expand its vm space by the passed
2425 int may_expand_vm(struct mm_struct
*mm
, unsigned long npages
)
2427 unsigned long cur
= mm
->total_vm
; /* pages */
2430 lim
= rlimit(RLIMIT_AS
) >> PAGE_SHIFT
;
2432 if (cur
+ npages
> lim
)
2438 static int special_mapping_fault(struct vm_area_struct
*vma
,
2439 struct vm_fault
*vmf
)
2442 struct page
**pages
;
2445 * special mappings have no vm_file, and in that case, the mm
2446 * uses vm_pgoff internally. So we have to subtract it from here.
2447 * We are allowed to do this because we are the mm; do not copy
2448 * this code into drivers!
2450 pgoff
= vmf
->pgoff
- vma
->vm_pgoff
;
2452 for (pages
= vma
->vm_private_data
; pgoff
&& *pages
; ++pages
)
2456 struct page
*page
= *pages
;
2462 return VM_FAULT_SIGBUS
;
2466 * Having a close hook prevents vma merging regardless of flags.
2468 static void special_mapping_close(struct vm_area_struct
*vma
)
2472 static const struct vm_operations_struct special_mapping_vmops
= {
2473 .close
= special_mapping_close
,
2474 .fault
= special_mapping_fault
,
2478 * Called with mm->mmap_sem held for writing.
2479 * Insert a new vma covering the given region, with the given flags.
2480 * Its pages are supplied by the given array of struct page *.
2481 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2482 * The region past the last page supplied will always produce SIGBUS.
2483 * The array pointer and the pages it points to are assumed to stay alive
2484 * for as long as this mapping might exist.
2486 int install_special_mapping(struct mm_struct
*mm
,
2487 unsigned long addr
, unsigned long len
,
2488 unsigned long vm_flags
, struct page
**pages
)
2491 struct vm_area_struct
*vma
;
2493 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
2494 if (unlikely(vma
== NULL
))
2497 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
2499 vma
->vm_start
= addr
;
2500 vma
->vm_end
= addr
+ len
;
2502 vma
->vm_flags
= vm_flags
| mm
->def_flags
| VM_DONTEXPAND
;
2503 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
2505 vma
->vm_ops
= &special_mapping_vmops
;
2506 vma
->vm_private_data
= pages
;
2508 ret
= security_file_mmap(NULL
, 0, 0, 0, vma
->vm_start
, 1);
2512 ret
= insert_vm_struct(mm
, vma
);
2516 mm
->total_vm
+= len
>> PAGE_SHIFT
;
2518 perf_event_mmap(vma
);
2523 kmem_cache_free(vm_area_cachep
, vma
);
2527 static DEFINE_MUTEX(mm_all_locks_mutex
);
2529 static void vm_lock_anon_vma(struct mm_struct
*mm
, struct anon_vma
*anon_vma
)
2531 if (!test_bit(0, (unsigned long *) &anon_vma
->root
->head
.next
)) {
2533 * The LSB of head.next can't change from under us
2534 * because we hold the mm_all_locks_mutex.
2536 spin_lock_nest_lock(&anon_vma
->root
->lock
, &mm
->mmap_sem
);
2538 * We can safely modify head.next after taking the
2539 * anon_vma->root->lock. If some other vma in this mm shares
2540 * the same anon_vma we won't take it again.
2542 * No need of atomic instructions here, head.next
2543 * can't change from under us thanks to the
2544 * anon_vma->root->lock.
2546 if (__test_and_set_bit(0, (unsigned long *)
2547 &anon_vma
->root
->head
.next
))
2552 static void vm_lock_mapping(struct mm_struct
*mm
, struct address_space
*mapping
)
2554 if (!test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2556 * AS_MM_ALL_LOCKS can't change from under us because
2557 * we hold the mm_all_locks_mutex.
2559 * Operations on ->flags have to be atomic because
2560 * even if AS_MM_ALL_LOCKS is stable thanks to the
2561 * mm_all_locks_mutex, there may be other cpus
2562 * changing other bitflags in parallel to us.
2564 if (test_and_set_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
))
2566 spin_lock_nest_lock(&mapping
->i_mmap_lock
, &mm
->mmap_sem
);
2571 * This operation locks against the VM for all pte/vma/mm related
2572 * operations that could ever happen on a certain mm. This includes
2573 * vmtruncate, try_to_unmap, and all page faults.
2575 * The caller must take the mmap_sem in write mode before calling
2576 * mm_take_all_locks(). The caller isn't allowed to release the
2577 * mmap_sem until mm_drop_all_locks() returns.
2579 * mmap_sem in write mode is required in order to block all operations
2580 * that could modify pagetables and free pages without need of
2581 * altering the vma layout (for example populate_range() with
2582 * nonlinear vmas). It's also needed in write mode to avoid new
2583 * anon_vmas to be associated with existing vmas.
2585 * A single task can't take more than one mm_take_all_locks() in a row
2586 * or it would deadlock.
2588 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2589 * mapping->flags avoid to take the same lock twice, if more than one
2590 * vma in this mm is backed by the same anon_vma or address_space.
2592 * We can take all the locks in random order because the VM code
2593 * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never
2594 * takes more than one of them in a row. Secondly we're protected
2595 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
2597 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2598 * that may have to take thousand of locks.
2600 * mm_take_all_locks() can fail if it's interrupted by signals.
2602 int mm_take_all_locks(struct mm_struct
*mm
)
2604 struct vm_area_struct
*vma
;
2605 struct anon_vma_chain
*avc
;
2608 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2610 mutex_lock(&mm_all_locks_mutex
);
2612 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2613 if (signal_pending(current
))
2615 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2616 vm_lock_mapping(mm
, vma
->vm_file
->f_mapping
);
2619 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2620 if (signal_pending(current
))
2623 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2624 vm_lock_anon_vma(mm
, avc
->anon_vma
);
2631 mm_drop_all_locks(mm
);
2636 static void vm_unlock_anon_vma(struct anon_vma
*anon_vma
)
2638 if (test_bit(0, (unsigned long *) &anon_vma
->root
->head
.next
)) {
2640 * The LSB of head.next can't change to 0 from under
2641 * us because we hold the mm_all_locks_mutex.
2643 * We must however clear the bitflag before unlocking
2644 * the vma so the users using the anon_vma->head will
2645 * never see our bitflag.
2647 * No need of atomic instructions here, head.next
2648 * can't change from under us until we release the
2649 * anon_vma->root->lock.
2651 if (!__test_and_clear_bit(0, (unsigned long *)
2652 &anon_vma
->root
->head
.next
))
2654 anon_vma_unlock(anon_vma
);
2658 static void vm_unlock_mapping(struct address_space
*mapping
)
2660 if (test_bit(AS_MM_ALL_LOCKS
, &mapping
->flags
)) {
2662 * AS_MM_ALL_LOCKS can't change to 0 from under us
2663 * because we hold the mm_all_locks_mutex.
2665 spin_unlock(&mapping
->i_mmap_lock
);
2666 if (!test_and_clear_bit(AS_MM_ALL_LOCKS
,
2673 * The mmap_sem cannot be released by the caller until
2674 * mm_drop_all_locks() returns.
2676 void mm_drop_all_locks(struct mm_struct
*mm
)
2678 struct vm_area_struct
*vma
;
2679 struct anon_vma_chain
*avc
;
2681 BUG_ON(down_read_trylock(&mm
->mmap_sem
));
2682 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex
));
2684 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2686 list_for_each_entry(avc
, &vma
->anon_vma_chain
, same_vma
)
2687 vm_unlock_anon_vma(avc
->anon_vma
);
2688 if (vma
->vm_file
&& vma
->vm_file
->f_mapping
)
2689 vm_unlock_mapping(vma
->vm_file
->f_mapping
);
2692 mutex_unlock(&mm_all_locks_mutex
);
2696 * initialise the VMA slab
2698 void __init
mmap_init(void)
2702 ret
= percpu_counter_init(&vm_committed_as
, 0);